A multi-agent system for optimal sizing of a cooperative self-sustainable multi-carrier microgrid

Sustainable Cities and Society 38 (2018) 452–465

Contents lists available at ScienceDirect

Sustainable Cities and Society journal homepage: www.elsevier.com/locate/scs

A multi-agent system for optimal sizing of a cooperative self-sustainable multi-carrier microgrid Soheil Mohseni, Seyed Masoud Moghaddas-Tafreshi

T

⁎

Department of Electrical Engineering, Faculty of Engineering, University of Guilan, Rasht, Iran

A R T I C L E I N F O

A B S T R A C T

Keywords: Multi-agent system Optimal sizing Microgrid Distributed energy resources Demand-side management

In this paper, an interactive multi-agent system (MAS) is applied to the problem of optimal sizing of a cooperative self-sustainable multi-carrier microgrid that includes various privately-owned entities. The proposed microgrid includes photovoltaic (PV) arrays, batteries, an electrolyzer, a hydrogen tank, a fuel cell (FC), a reactor-reformer system, a hydrogen compressor-dispenser system, a converter, residential electrical loads, and a charging/refilling station. The proposed MAS enables information exchange required for the application of demand-side management (DSM) strategy and has five agents, namely generation agent (GA), electrical load agent (LA), charging/refilling station agent (SA), control agent (CA), and design agent (DA). The GA is responsible for managing the distributed energy resources of the microgrid. The LA aggregates the residential electrical loads. The SA is responsible for charging of plug-in hybrid electric vehicles (PHEVs) and refilling of fuel cell electric vehicles (FCEVs). The CA coordinates the interactions between the field level agents. The DA finds the optimal sizes of the system’s components by minimizing the total cost of the system through particle swarm optimization (PSO) algorithm. Simulation results demonstrate that the proposed system can reduce the overall cost of the microgrid in comparison with non-interactive methods.

1. Introduction According to the limitation of fossil fuels, climate change issues, significance of energy diversification, and potential for job creation, governments have been encouraged to increase the share of renewable energies in their energy portfolios (Chmutina, Wiersma, Goodier, & Devine-Wright, 2014). Therefore, using renewable energy technologies would be affordable in remote areas as self-sustainable hybrid energy systems due to the high costs associated with network expansion (Bernal-Agustín & Dufo-López, 2009). From the grid point of view, PV systems and electric vehicles (EVs) will bring a great impact on the grid (Van Roy et al., 2014). Photovoltaic systems are one of the important solar energy utilization technologies that must be integrated into the power grid (Maleki, Khajeh, & Rosen, 2017). Furthermore, it is possible to shift the charging load demand of the EVs to off-peak hours that helps in solving the integration problem of renewable energy sources (Brad & O’Mahony, 2017; Shakouri & Kazemi, 2017). A microgrid is a discrete and small power grid that provides a platform for the integration of distributed energy resources and loads that can be operated in both grid-connected and islanded modes and is able to open up new opportunities for the utilization of renewable energy sources (Hassanzadehfard, Moghaddas-Tafreshi, & Hakimi, 2015;

⁎

Lasseter et al., 2002). Generally, demand curves are desired to be as flat as possible for generation cost and environmental considerations (Eichman, Mueller, Tarroja, Schell, & Samuelsen, 2013). DSM techniques aim at the reduction of the peak demand thereby flattening the load profile. On the other hand, the smart and sustainable microgrids that are based on green energy generation are getting more attention for the charging requirement of EVs and the unscheduled EVs connectivity with power system may lead to unreliable and interrupted power supply (Ahmad, Alam, & Asaad, 2017). Charging scheduling of EVs offers an opportunity for managing fluctuations in electricity generation and consumption (Galus, Fauci, & Andersson, 2010). Therefore, charging coordination of EVs and implementing the DSM strategies in the decarbonized transport sector are very important; hence several coordination scales for the integration of the EVs have been proposed in the literature (Leemput, Van Roy, Geth, Tant, & Driesen, 2011): The vehicle, building, residential distribution, and transmission grid scale. Various studies are conducted for EV fleet management in smart grids (Aghaei, Esmaeel Nezhad, Rabiee, & Rahimi, 2016; Hu, Morais, Sousa, & Lind, 2016; Sabri, Danapalasingam, & Rahmat, 2016). An optimization model for the integration of EVs is presented in (Tanguy, Dubois, Lopez, & Gagné, 2016) to maximize the overall community benefits of using the

Corresponding author. E-mail address: [email protected] (S.M. Moghaddas-Tafreshi).

https://doi.org/10.1016/j.scs.2018.01.016 Received 22 September 2017; Received in revised form 11 December 2017; Accepted 10 January 2018 Available online 11 January 2018 2210-6707/ © 2018 Elsevier Ltd. All rights reserved.

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi

charging coordination for a fleet of EVs on the campus of Université Laval on a daily basis. The optimal locating and sizing of renewable energy sources and EV charging stations while considering the management of vehicle charging process is investigated in (Mozafar, Moradi, & Amini, 2017) and it is proved that using EVs along with renewable energy sources in the power system can reduce the cost of both the system operator and subscribers. A method of simultaneous allocation of EV parking lots and distributed renewable resources is proposed in (Amini, Moghaddam, & Karabasoglu, 2017) and it is shown that the EV parking lots and distributed renewable resources utilization should be obtained in a scheduled way. The real implementation of a fast-charging station for electric vehicles which is incorporated in a microgrid is described in (Sbordone et al., 2015) and the experimental tests have shown that the system can deliver an effective peak shaving in respect of the main grid. Optimal sizing of the components of a microgrid is necessary as this ensures the reliable power supply while retaining the economic viability of the system. The sizing problem of the microgrid components is a complicated issue since it needs to consider both reliability and cost. In a grid-connected system, the sizing of the microgrid components is less complicated as it only considers the area and time duration to supply energy from the grid. In comparison, the self-sustainable microgrid has a more complex sizing system since it needs to work without disturbances and ensure the quality of power supply within its own boundaries. Therefore, various studies are conducted for optimal sizing of self-sustainable microgrids and different computational optimization techniques are used to realize the sustainable development of them (Bhowmik, Bhowmik, Ray, & Pandey, 2017; Calvillo, Sánchez-Miralles, & Villar, 2016; Fathima & Palanisamy, 2015; Gamarra & Guerrero, 2015; Jung & Villaran, 2017; Saboori, Hemmati, Ghiasi, & Dehghan, 2017). A single-objective optimal sizing approach for an islanded microgrid is proposed in (Bhuiyan, Yazdani, & Primak, 2015). The approach determines the optimal sizes of microgrid components such that the life-cycle cost is minimized while a low loss of power supply probability is ensured. An energy storage sizing method that considers a reliability index and a bi-level control strategy for the isolated grids, where the energy demand is completely supplied by wind power is described in (Luo, Shi, & Tu, 2014). An optimal sizing and management strategy for a stand-alone microgrid including renewable energy sources such as wind turbines and PV generators, diesel generators, and battery storage system is proposed in (Ma, Yang, Wang, Zhao, & Zheng, 2014). The optimization problem is formulated to minimize the annualized system cost through the PSO algorithm. A stochastic optimal planning model based on chance-constrained programming algorithm for islanded microgrids is proposed in (Guo, Liu, Jiao, Hong, & Wang, 2014). Based on multi-objective non-dominated sorting genetic algorithm II, optimal planning for an islanded microgrid is carried out, which has verified the model. A stand-alone wind/PV/FC generation microgrid is designed in (Baghaee, Mirsalim, Gharehpetian, & Talebi, 2016) for a 20-year period of operation using a multi-objective optimization algorithm to minimize the three objective functions, namely annualized cost of the system, loss of energy expected, and loss of load expected. A stand-alone microgrid considering electricity, cooling/ heating, and hydrogen consumption is designed in (Li, Roche, Paire, & Miraoui, 2017). The authors have used a genetic algorithm to search for the optimal sizes of the microgrid components and discussed the influence of operation strategy, accuracy of load and renewable generation forecast, and degradation of the fuel cell, electrolyzer, and battery on sizing results. Few researchers have addressed the problem of optimal sizing of the components of the microgrids considering DSM techniques in their operating strategies. The effects of implementing a demand response program on reducing the optimal size of the renewable energy resources within a microgrid, which in turn reduces its overall cost and improves its reliability are analyzed in (Hakimi & Moghaddas-Tafreshi, 2014). A methodology which incorporates both the sizes of the

resources and the strategy by which they will be operated is presented in (Zhao et al., 2014) and it has been implemented on Dongfushan Island, China. The authors have used a genetic algorithm-based method to solve the optimal sizing problem such that it minimizes the life-cycle cost, maximizes the renewable energy source penetration, and minimizes the pollutant emissions. An optimal sizing methodology for finding the optimal size of a battery energy storage system in a microgrid using PSO algorithm is proposed in (Kerdphol, Qudaih, & Mitani, 2016) that incorporates a dynamic demand response program to reduce the capital, operating and maintenance costs of the battery energy storage system. The authors have claimed that their method not only reduces the overall cost but also improves the system’s stability and performance during an emergency situation. A bi-objective optimization model that aims at minimization of the total investment and operation costs, as well as minimization of the loss of load expectation for optimal sizing of an energy storage system in a microgrid is proposed in (Nojavan, Majidi, & Esfetanaj, 2017). The authors have employed a demand response program which flattens load curve by shifting some percentage of the load from peak periods to off-peak periods to reduce the total cost of the microgrid. An optimal sizing strategy of power sources and energy storage system in an autonomous microgrid is proposed in (Liu, Chen, Zhuo, & Jia, 2018) that considers DSM and EV scheduling. The authors have concluded that their method results in significant saving in sizing of power sources of the microgrid. A combined microgrid sizing and energy management methodology is proposed in (Li, Roche, & Miraoui, 2017). The authors have formulated their methodology as a leader-follower problem. In the leader problem, they have focused on optimal sizing of the components of the microgrid which is solved using a genetic algorithm. In the follower problem, they have formulated the energy management issue as a unit commitment problem which is solved using a mixed integer linear problem. Based on their results, they have concluded that their proposed bi-level optimization approach for the sizing of the microgrid components reduces the overall cost of the microgrid. A multi-objective optimization method of planning and operation of a microgrid with DSM capability is presented in (Chen et al., 2018) that aims at minimizing the total annual cost as well as maximizing the customer satisfaction. By jointly optimizing the planning design and operation, the authors have demonstrated that DSM is an effective way to achieve the goal of cost reduction while not compromising customer satisfaction. The above review of the state-of-the-art has shown that an efficient optimal sizing methodology requires an appropriate operating strategy and DSM-based energy management fits this requirement. One question that needs to be asked while optimally sizing the components of a microgrid is whether the microgrid is a single-owner microgrid or a multi-stakeholder microgrid. Current solutions to the optimal sizing of the stand-alone microgrids are inconsistent with the interactive nature of the multi-stakeholder microgrids, in which various privately-owned entities work cooperatively to provide the energy requirements of the customers. Furthermore, it is not yet known how to implement the DSM programs when multiple stakeholders are present in a microgrid while optimally sizing its components. This is an important issue for successfully managing the potential conflicts of interest that might arise between different stakeholders in a microgrid. Multi-agent systems are inherently distributed and they can be used to model the autonomous entities in solving complicated problems. They have intrinsic advantages such as flexibility, autonomy, and scalability. Therefore using them for the power system operation within the smart grid technology would be beneficial (Dimes & Hatziargyrious, 2004; Logenthiran, Srinivasan, Khambadkone, & Aung, 2012). In this regard, various studies have applied the multi-agent systems to the microgrid control and operation (Coelho, Cohen, Coelho, Liu, & Guimarães, 2017; Howell, Rezgui, Hippolyte, Jayan, & Li, 2017; Kantamneni, Brown, Parker, & Weaver, 2015; Khan & Wang, 2017). A multi-agent supervisory control for optimal economic dispatch in DC microgrids is proposed in (Hamad & El-Saadany, 2016) that can reduce 453

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi

Fig. 1. Schematic diagram of the proposed microgrid system.

enables implementing a DSM strategy in the proposed microgrid and finds the optimal sizes of its components by minimizing the total cost of the system through PSO algorithm. The considered DSM strategy in the proposed MAS controls the charging of PHEVs and refilling of FCEVs depending on the hour of the day to reduce cost and avoid overload during peak hours. It is assumed that the system uses municipal solid waste (MSW) as an available energy resource that can be collected from residential areas. Also, the sizes of anaerobic reactor-reformer and hydrogen compressor-dispenser systems are fixed and do not participate in the optimal sizing procedure. It is worth mentioning that in this paper, gasoline is considered as the second fuel of the PHEVs. The main novelty of this paper is applying the MAS concept to the optimal sizing problem of microgrids that enables modeling of the autonomous nature of different privately-owned entities in cooperative self-sustainable microgrids and considering a DSM strategy in the optimal sizing problem of them. Furthermore, we have analyzed the technical requirements for implementing the proposed microgrid. This study is performed for Hendurabi Island which is a remote island in the Persian Gulf. Located approximately 133 kilometers off the coast of Hormozgan Province, Iran, the island is approximately 22.8 square kilometers in area. The island ranges in elevation from sea level to 43 m above sea level. The island receives a good amount of solar radiation throughout the year and has a population of 300. As Hendurabi Island is a habitat for some bird species that are in danger of extinction, installing the wind turbines has been forbidden by the Iranian Department of Environment. It is important to note that in this paper, the selection of the components of the microgrid is performed based on the local potential of renewable energy sources to produce electricity and hydrogen to meet customer demands through a flexible, reliable, zero-emission system. The rest of this paper is organized as follows: In Section 2, configuration and power flow of the proposed microgrid are presented. Section 3 describes the MAS used for optimal sizing of the proposed microgrid along with the description of components belonging to each agent. Section 4 presents the simulation results of the MAS-based architecture used for optimal sizing of the proposed microgrid. Finally, the conclusion of this study is presented in Section 5.

the implementation cost considerably. The simplicity of this algorithm and the minimal implementation and communication requirements are the main characteristics that make it suitable for the practical implementation. A multi-agent based hierarchical control for the smart control of an autonomous microgrid is proposed in (Dou & Liu, 2013) that considers both the hybrid dynamic behaviors and hierarchical control requirements. The hierarchical control system maximizes the economic benefits for an autonomous microgrid. A MAS is proposed in (Fazal, Solanki, & Solanki, 2012) in order to optimize the demand response by updating the generation resources and controlling the customer load within a microgrid. The MAS controls the charging of PHEVs depending on their battery state of charge and reduces the overall cost of electricity by shifting loads to off-peak hours. A scalable MAS for optimal operation of a stand-alone microgrid is proposed in (Logenthiran, Srinivasan, Khambadkone, & Aung, 2010) that maximizes the power production of distributed generators while minimizing the operation and maintenance (O&M) cost of the microgrid. In order to reach these two goals, the agents in the scalable MAS interact among themselves cooperatively. Multi-agent systems are very helpful in the modeling of the cooperative microgrids, especially in the presence of independent owners. Besides, they can realize the concept of dynamic demand response. Therefore, using a MAS-based architecture to solve the optimal sizing problem of the microgrids enables applying DSM strategies and utilizing the available devices better, which in turn decreases the total cost of the system. This paper seeks to address how to solve the optimal sizing problem of a cooperative self-sustainable multi-carrier microgrid while there are various privately-owned entities in it that are able to interact and cooperate with each other to provide the energy requirements of the customers. The microgrid is equipped with PV arrays, battery packs, an electrolyzer, a hydrogen tank, a fuel cell, an anaerobic reactor-reformer system, a hydrogen compressor-dispenser system, a DC/AC converter, and a charging/refilling station. It is assumed that the owners of the generation/storage devices and charging/refilling station are different and the sizes of their components should be found such that the total cost of the micro-grid is minimized. This, in turn, provides social welfare benefits to all legal residents due to decreasing the electricity and charging/refilling bills. For this purpose, a multi-agent system including the GA, LA, SA, CA, and DA is developed in this paper that

454

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi

2. Configuration and power flow of the microgrid

the agents associated with the generation or consumption of electricity/ hydrogen belong to the field level, which is the lowest level of the considered architecture for the multi-agent system and includes the GA, LA, and SA. In this framework, the generation and station agents represent the privately-owned entities that have agreed to follow the suggestions of the microgrid designer about the sizes of their components but have some freedom of action to decide how to use their own components. Also, it is assumed that the GA should select its generation units based on the local potential of renewable energy sources and the restrictions imposed by the microgrid designer. In the coordination level, coordination of the generation and consumption of the electrical and hydrogen loads obtains. The CA belongs to this level and coordinates the interactions between the field level agents. In this level, the microgrid operator dictates which loads are served and aims at extending the survivability, maintaining the stability, and achieving the energy management goals of the system. Design level is the highest level of the proposed architecture that finds the optimal sizes of the components of the microgrid according to the interactions with coordination level such that the residential loads and the charging and hydrogen power demands of its station are met. In the proposed MAS, the DA belongs to this level in which the microgrid designer aims at optimal sizing of the components of the proposed microgrid considering the following constraints: (i) minimizing the total cost of the microgrid, (ii) zero air pollution, (iii) high reliability, (iv) multi-stakeholder microgrid ownership, (v) integrating the PHEV and FCEV fleets into the microgrid. As in describing the multi-agent systems, it is necessary to define the percepts, actions, goals, and the environment where the agents of the system act, they are defined for each agent while describing it. The communication channels for data exchange in the proposed MAS is shown in Fig. 3. In this figure, the numbers indicate the sequence of messages sent between the agents. The messages 1 and 6 are sent after every 8760 h and the messages 2, 3, 4, and 5 are sent after every hour; this point will be discussed in more detail later. 1: The DA sends the determined sizes by PSO to the CA, 2: The CA requests the GA for the next hour value of power produced by PV arrays, requests the LA for the next hour value of residential loads, and requests the SA for the next hour values of charging and hydrogen power demands of the station for charging of PHEVs and refilling of FCEVs, 3: The GA sends the next hour value of solar irradiance, the LA sends the next hour value of load demand, and the SA sends the next hour values of charging and hydrogen power demands to the CA, 4 and 5: The DSM strategy implements according to the requests and actions that will be presented in the following sections, 6: The CA sends the information on the

A schematic representation of the proposed microgrid is presented in Fig. 1. This microgrid consists of three groups of units. The first group includes the energy production units which are photovoltaic arrays and a typical proton exchange membrane fuel cell that produce electricity, and an anaerobic reactor-reformer system which is responsible for supplying some fraction of the hydrogen. The second group includes the energy consumption units which are residential electrical loads and charging and hydrogen power demands of the station for charging of PHEVs and refilling of FCEVs that arrive at the station. The last group includes the storage units which are a typical battery bank and a typical low-pressure hydrogen tank. This hydrogen tank stores the hydrogen produced by the electrolyzer unit and reactor-reformer system. Power flow diagram of the proposed microgrid is also shown in Fig. 1. According to this diagram, the output power of the photovoltaic unit is PPV [kW] that depends on the solar irradiance. When the output power of PV arrays is more than the residential electrical loads (Pload ) , the surplus power to the amount of Psta,delivered [kW] which can be less than required charging demand of the station (Psta, el ) , will be delivered to the station for charging of PHEVs. After charging of PHEVs, if there is more surplus power, it will be used to the amount of Pch [kW] for charging of battery packs and after exceeding the battery bank’s rated power, the second priority of using this surplus power is using it to the amount of Pele [kW] in the electrolyzer to electrolyze the water and store the produced hydrogen which has the power of Pel−hb [kW] that is equal to Phb−tank [kW], in the hydrogen tank. If the injected power to the electrolyzer exceeds its rated power, then the excess power to the amount of Pdr,el [kW] will circulate in a dump load. On the other hand, when the output power of the PV arrays is less than the residential loads, at first, battery bank discharges to the amount of Pdch [kW] and if the battery bank is not able to compensate the shortage of power, then the fuel cell consumes an amount of stored hydrogen in the hydrogen tank which has the power of Phb−fc [kW] and supplies the power of Pfc [kW] to compensate the shortage of the power production. If the shortage of power exceeds the fuel cell’s rated power or the stored hydrogen cannot afford that, some fraction of the residential loads must be shed which leads to loss of load. Also, charging/refilling station consumes the stored hydrogen in the tank to the amount of Phb−sta [kW] for refilling of FCEVs that arrive at the station. It is worth mentioning that at each time step t, Ptank−hb is equal to Phb−fc plus Phb−sta, and Phb−tank is equal to Pel−hb plus Pref−hb. Furthermore, when the stored hydrogen is more than the capacity of the hydrogen tank, the additional hydrogen which has the power of Pdr,hyd [kW] will be fed to a dump hydrogen load.

3. Multi-agent system for optimal sizing of the microgrid The considered MAS for optimal sizing of the proposed microgrid has five agents, namely generation agent, electrical load agent, charging/refilling station agent, control agent, and design agent which are organized in three levels. These three levels are presented in Fig. 2. All

Fig. 2. The organization of the MAS.

Fig. 3. Data exchange in the MAS.

455

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi

The mass of stored hydrogen in the tank can be calculated according to the following equation (Dehghan et al., 2009):

operation of the microgrid to the DA. 3.1. Generation agent (GA)

mstorage (t ) = This agent is responsible for managing the generation resources of the microgrid. Power generation resources of the microgrid include PV arrays and a fuel cell. Also, this agent has an electrolyzer, a hydrogen tank, a reactor-reformer system, a battery bank, and a DC/AC converter. The GA is a goal-based agent that perceives the amount of power produced by the PV arrays or the fuel cell, the amount of hydrogen produced by the reactor-reformer system or the electrolyzer, state of charge of the battery bank, and the amount of hydrogen stored in the hydrogen tank by using the sensors. Charging/discharging of the battery bank, producing hydrogen by adjusting the operation point of the electrolyzer, generation of electricity in the fuel cell by discharging of the hydrogen tank and adjusting the operation point of the fuel cell, and sending the information on the operation of its components to the CA are its actions. This agent has the goals of generation and storage of electric and hydrogen powers to satisfy the residential loads and the charging and hydrogen power demands of the station. Furthermore, PV arrays, electrolyzer, hydrogen tank, fuel cell, battery bank, reactor-reformer system, and the DC/AC converter can be considered as its environment.

Etank (t ) , HHVH2

(5)

where HHVH2 is the higher heating value of hydrogen that is equal to 39.7 kWh/kg (Dehghan et al., 2009). It should be pointed out, there are lower and upper limits for the amount of hydrogen that can be stored in the tank. The rated capacity of the tank determines the upper limit of it. Conversely, it is not possible to extract a small fraction of the hydrogen stored in the tank which is called the lower limit of it. In this study, the lower limit of the tank is taken as 5 percent of the rated capacity of it. Therefore:

Etank, min ≤ Etank (t ) ≤ Etank, max .

(6)

In this paper, the energy of the tank at each time step t is considered as one of the decision variables and after solving the optimal sizing problem and finding the optimal size of the tank based on the stored energy level, the capacity of the tank can be derived by Eq. (5). Charge quantity of the battery bank at time step t is obtained by the following equation:

Ebat (t ) = Ebat (t − 1) + Pch (t ) × ηch × Δt − (Pdch (t )/ηdch ) × Δt ,

(7)

3.1.1. Description of the components In order to estimate the output power of PV arrays, a mathematical model is used in this paper. The output power of the PV generator (PPV) can be calculated according to the following equation (Smaoui, Abdelkafi, & Krichen, 2015):

where Pch is the transferred power from the PV unit to the battery bank, Pdch is the transferred power from the battery bank to the converter, and ηch and ηdch are the charge and discharge efficiencies of the batteries, respectively which are assumed to be 85% in this study. In all of the cases, the storage battery capacity is subject to the following constraints:

PPV = ηg NPV Am Gt ,

Ebat , min ≤ Ebat (t ) ≤ Ebat , max ,

(1)

(8)

where ηg is the efficiency of PV arrays that is considered to be 15.4% (Smaoui et al., 2015), Am is the area of a single array of the system [m2] which is considered to be 1.9 m2 (Smaoui et al., 2015), Gt is the total solar radiation incident on the titled plane [W/m2], and NPV is the number of arrays that is found by solving the optimal sizing problem. The fuel cell’s output power can be defined by the following equation:

where Ebat,min and Ebat,max are the minimum and maximum allowable storage capacities of the battery bank. Ebat,max is taken as the nominal battery bank’s capacity which is determined by the optimum number of the battery packs (that is found in the optimal sizing procedure) multiplied by the nominal capacity of each battery pack. Moreover, Ebat,min is determined by the maximum allowable depth of discharge (DOD) of the battery, as follows:

Pfc = Phb − fc × ηfc ,

Ebat , min = (1 − DOD) Ebat , max ,

(2)

The maximum allowable depth of discharge of the battery is equal to 0.85 in this study. Furthermore, DC output of the PV generation system, fuel cell power system, and battery storage system are considered as the inputs of the DC/AC converter.

where ηfc is the efficiency of the fuel cell. The electrolyzer’s output power can be calculated by the following equation:

Pel − hb = Pele × ηele ,

(3)

where ηele is the efficiency of the electrolyzer. The anaerobic reactor-reformer system consists of an anaerobic reactor and a reformer. It is assumed that the hydrogen produced by the MSW is constant per hour and can be stored in the hydrogen tank. The size of the reactor-reformer system is fixed and equal to 20 kW H2 per hour. This system acts as a negative load in the proposed microgrid. The amount of the energy of hydrogen stored in the tank at time step t can be calculated by the following equation:

3.1.2. Functions of the GA This agent, in coordination with CA, interacts with the LA and SA in order to supply the electricity and hydrogen demands of the microgrid. In this regard, at first, the CA sends the sizes of PV arrays, battery packs, electrolyzer, hydrogen tank, fuel cell, and DC/AC converter that are determined by the DA to the GA, and at each time step t, asks it for the next hour value of the output power of PV arrays. Then, the GA sends the requested data to the CA. It should be noted that the GA always stores the hydrogen produced by the reactor-reformer system in the hydrogen tank and allows the CA to use all of the power generated by the PV arrays so that it can satisfy the requested residential loads and charging and hydrogen power demands of the station. First, we consider the electric power of the system. If the CA recognizes that the output power of PV arrays satisfies the residential loads and charging demand of the station, it will not send any requests to the GA and in this situation, the GA does not do any specific action. If the CA recognizes that the output power of PV arrays, not only satisfies the residential loads and charging demand of the station but also it is more than the amounts needed by the residential loads and charging demand of the station, it will ask the GA to store the

Etank (t ) = Etank (t − 1) + (Pel − hb (t ) + Pref − hb (t ) − Phb − fc (t ) × ηstorage − Phb − sta (t )) × Δt ,

(9)

(4)

where Pel−hb is the transferred power from the electrolyzer to the hydrogen bus, Pref−hb is the transferred power from the reformer to the hydrogen bus, Phb−fc is the transferred power from the hydrogen bus to the fuel cell, ηstorage is the efficiency of the hydrogen storage system which is assumed to be 95% (Dehghan, Kiani, Kazemi, & Parizad, 2009), Phb−sta is the transferred power from the hydrogen bus to the refilling station, and Δt is the duration of each time step which is equal to one hour in this study. 456

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi

shortage of power, it will compensate the shortage of power production by producing electricity in the fuel cell that uses the stored hydrogen in the tank; hence,

surplus power. In this situation, the GA has the freedom of action to store the surplus power in the battery bank or produce hydrogen by this power in the electrolyzer and store the produced hydrogen in the hydrogen tank. Also, if the CA recognizes that the output power of PV arrays does not satisfy the residential loads, it will ask the GA to compensate the shortage of power. In this situation, the GA has the freedom of action to compensate the shortage of power by using the stored hydrogen in the fuel cell or discharging of battery packs. It is worth mentioning that the aforementioned procedure ensures that the charging demand will be supplied only by the surplus power produced by the PV arrays after meeting the residential loads. In this paper, it is assumed that the GA acts such that when it receives a request for storing the surplus power from the CA, its first priority to store this surplus power is to charge the battery bank. If there is no battery capacity for energy storage, then it will use the excess energy in the electrolyzer and will store the produced hydrogen in the tank. On the other hand, when the GA receives a request for supplying the shortage of production from the CA, at first, it discharges the battery bank to supply the shortage of power. If the battery bank is not able to compensate the shortage of power, then the fuel cell will use the stored hydrogen in the tank to compensate the shortage of the power production. Therefore, five cases may happen which are presented below. Case A: in the time steps that the CA does not ask the GA to store the surplus/shortage of power production by the PV arrays, the GA does not do any specific action and just the produced hydrogen of the reactorreformer system will be added to the hydrogen tank; hence,

In this situation, if the amount of power generation by the fuel cell (Pfc ) is more than its optimum size which is determined by the DA, then the fuel cell cannot satisfy the required power and the GA maximizes the operation point of the fuel cell. Also, when the energy content of the hydrogen tank is less than the minimum storage capacity of it, the energy content of the tank will not be sufficient for supplying the residential loads and the GA sets the level of the stored hydrogen in the tank to the minimum value. In both of these conditions, some fraction of the residential loads must be shed. The CA is responsible for calculating the loss of residential loads and sends it to the DA, which will be discussed in the following. Also, at each time step t, the GA sends the information on the operation of its components to the CA. It is important to note that Pref − hb (t ) is equal to 20 kW in all the above cases. After investigating the requests of the CA for electricity, the generation agent investigates the requests of the CA for hydrogen load and in the cases that the CA asks it for supplying the hydrogen load, it updates Etank (t ) according to the following equation:

Etank (t ) = Etank (t − 1) + Pref − hb (t ),

Etank (t ) = Etank (t ) − Phb − sta,

Pfc = (Pload/ηconv ) − PPV , Phb − fc = (Pfc /ηfc ) Etank (t ) = Etank (t − 1) + Pref − hb (t ) − Phb − fc (t ) × ηstorage .

(10)

3.2. Electrical load agent (LA) This is a simple reflex agent that aggregates the residential electrical loads. The electrical loads that the LA is responsible for forecasting them are uninterruptable loads and should be supplied subject to a reliability constraint. The aggregated residential electrical loads are considered as the only percept of this agent and sending the aggregated electricity demand to the CA is the only action of it. Furthermore, as this agent is considered to be a simple reflex agent, no specific goal is considered for it and the residential loads form its environment. It should be noted that the LA after receiving a request for announcing the amount of load demand from the CA, sends the next hour value of Pload to the CA.

Pch = PPV − Pconv, dc − Psta, delivered, Etank (t ) = Etank (t − 1) + Pref − hb (t ), (11)

Case C: in the time steps that the GA recognizes that there is a surplus power production by the PV arrays that if it is used for charging of battery bank, the capacity of the battery bank will reach its upper limit, this agent decides to electrolyze the water by electrolyzer and store the produced hydrogen in the hydrogen tank; hence,

Pele = PPV − Pconv, dc − Psta, delivered, Pel − hb = Pele × ηele , Etank (t ) = Etank (t − 1) + Pref − hb (t ) + Pel − hb (t ).

3.3. Charging/refilling station agent (SA) (12) This agent is responsible for charging of PHEVs and refilling of FCEVs that arrive at the station and sends the forecasted charging and hydrogen power demands to the CA. This agent is a goal-based agent that the number of PHEVs and FCEVs that arrive at the station and the required charging and hydrogen demands of them are its percepts, charging of PHEVs and refilling of FCEVs to the desired level of the customers and sending the charging and hydrogen power demands and uncharged energy of PHEVs and FCEVs at a predetermined time to the CA are its actions. Appropriate usage of the surplus electric power of PV arrays and the hydrogen supplied by the GA in coordination with the CA is its only goal and the arrived PHEVs and FCEVs at the station and the charging and refilling infrastructures form its environment. In order to charge a PHEV’s battery, electric vehicle supply equipment (EVSE) has been considered in this paper that can communicate with the PHEV so that charging process occurs in a safe manner (Clean Cities, 2012). In this paper, SmartCharge-12000 is chosen as EVSE which is programmable and can work with both AC and DC input voltages. This EVSE is designed and manufactured by Electric Motor

In this situation, if Pele exceeds the electrolyzer’s rated power, then the excess energy will circulate in a dump load and if the stored hydrogen is more than the capacity of the tank, the additional hydrogen will be fed to a dump hydrogen load. Case D: in the time steps that the GA recognizes that there is a shortage of power production by the PV arrays for supplying the residential loads and the battery bank is able to compensate the shortage of power without reaching its capacity to the lower limit, it will compensate the shortage of power production by discharging of battery packs; hence,

Pdch = (Pload/ηconv ) − PPV , Etank (t ) = Etank (t − 1) + Pref − hb (t ), Ebat (t ) = Ebat (t − 1) − (Pdch (t )/ηdch ) × Δt .

(15)

where Phb-sta is determined by the CA. Then, the GA sends the updated amount of Etank (t ) to the CA.

It should be noted that the time steps are taken to be 1 h in this study. Case B: in the time steps that the GA recognizes that there is a surplus power production by the PV arrays which can be stored in the battery bank without reaching the capacity of the battery bank to its upper limit, it decides to store the surplus power production in the battery bank; hence,

Ebat (t ) = Ebat (t − 1) + Pch (t ) × ηch × Δt .

(14)

(13)

Case E: in the time steps that the GA recognizes that there is a shortage of power production by the PV arrays for supplying the residential loads and the battery bank is not able to compensate the 457

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi

It should be noted that these two load profiles can be updated during the operation of the station that will be described in the following subsection.

Werks company (Anonymous, 2018a). In this paper, first come first serve scheme is utilized for charging of PHEVs, therefore newly arrived PHEVs will join the end of a queue and when an EVSE becomes available, they can get a charge by connecting to that EVSE. The considered plug-in hybrid electric vehicle in this study is Chevrolet Volt 2016 that is manufactured by General Motors. The maximum storage capacity of the battery of this PHEV is equal to 18.4 kWh that 14 kWh of it is usable. It is possible to charge the Volt’s battery with 120 and 240-V circuits that allow charging from 0% to 100% in 13 and 4.5 h, respectively (Letendre, Denholm, & Lilienthal, 2006). In this paper, hydrogen pipeline transport is used to deliver hydrogen from the hydrogen tank of the GA to the station. Then, hydrogen must be compressed from 1 bar to 700 bar and the gas should be pumped to storage vessels for delivery to the fuel pump. The considered FCEV in this study accepts hydrogen at 700 bar. Therefore, a hydrogen compressor must be installed at the station. Also, in order to fuel our 700 bar FCEVs, a hydrogen dispenser is required. In this analysis, Pure Energy Centre’s hydrogen compressor and dispenser are used to achieve the 700 bar and fuel FCEVs. This compressor has the maximum power of 200 kW which is enough for supplying the FCEVs that arrive at the station. Therefore, it is assumed that the size of hydrogen compressor-dispenser system is fixed and does not participate in the optimal sizing procedure. In this paper, Toyota Mirai is chosen as a mid-size sedan FCEV which is manufactured by Toyota. Mirai is equipped with a 113 kW fuel cell-powered electric motor and two hydrogen tanks that store hydrogen at 700 bar. The combined hydrogen storage capacity of these two tanks is equal to 5 kg. It is notable that refilling the Mirai’s tanks takes between 3 and 5 min (Gordon-Bloomfield, 2014). It is worth noting that the efficiency of EVSEs and hydrogen compressor-dispenser system will be denoted by ηsta,el and ηsta,hyd, respectively.

3.3.2. Functions of the SA This agent, in coordination with the CA, interacts with the GA in order to charge the PHEVs by solar resources and refill the EVSEs by the stored hydrogen in the tank. In this regard, at first, the CA sends the optimal number of EVSEs which is determined by the DA, to the SA and at each time step t, asks it for the next hour values of charging and hydrogen power demands. After receiving the amounts of charging and hydrogen power demands of the station from the SA, based on the available power of the PV arrays and the amount of stored hydrogen in the tank, the CA allocates an electrical power, Psta,delivered and a hydrogen power, Phb−sta to the station. The SA can charge/refill the PHEVs/FCEVs according to the amount of allocated electrical/hydrogen power by the CA to the station but has the freedom of action to decide how to use the allocated powers. First, we consider the PHEV charging demand of the station. There will be two different situations in the case of charging of PHEVs: Situation A: When the allocated electrical power by the CA to the station (Psta,delivered) meets the requested charging load (Psta,el/ηsta,el), the SA will not receive any requests and it will only send the information on the operation of the EVSEs to the CA. Situation B: When the allocated electrical power by the CA to the station is less than the requested charging load or is equal to zero, this agent first checks that whether the predetermined time for charging of any PHEV has been passed or not. If this is the case, then it calculates the uncharged energy of PHEVs at the predetermined time according to the following equation:

Qsta, el = Psta, el/ ηsta, el − Psta, delivered .

(16)

On the other hand, if none of the constraints imposed by the customers is violated, then the SA will update the charging demand and will send the updated values of charging load to the CA in the next time steps. Also, there will be two similar situations in the case of refilling of FCEVs and only Psta,el, ηsta,el, Qsta,el and Psta,delivered should be substituted by Psta,hyd, ηsta,hyd, Qsta,hyd and Phb−sta, respectively. Furthermore, in all of the above cases, this agent at each time step t sends the information on the operation of its EVSEs and hydrogen refilling infrastructures as well as uncharged energy of PHEVs and EVSEs at the predetermined time to the CA.

3.3.1. Modeling of charging/refilling station In this study, a charging/refilling station is considered in the proposed microgrid. It is assumed that PHEVs can arrive at the station between 8:00 am and 4:00 pm because the charging power is based on the surplus power produced by PV arrays after meeting the residential loads. Also, the predetermined time at which the SA sends the uncharged energy of PHEVs and FCEVs to the CA is considered to be 6:00 pm in this analysis. The charging rate of PHEVs is constant and equal to 4 kW that is denoted by ψ. The SA finds the probability distribution of the Poisson process of arriving PHEVs at this station and the charging level required by each PHEV is uniformly distributed between 0 and 14 kW. Accordingly, the aggregated charging load at each time step t which is denoted by Psta, el (t ) , is equal to the number of active EVSEs multiplied by the rate ψ. In order to find the number of vehicles being charged at each time step t, queuing theory is utilized. Among various queuing models, M/M/c queuing model suits the charging process in a charging station, where the first M denotes the exponential inter-arrival times of vehicles, the second M denotes the exponential service times, and c denotes the parallel identical EVSEs that the optimum number of them is found in the optimal sizing procedure. The M/M/c queuing model is described in details in (Kulkarni, 1999). In order to obtain the hydrogen power demand of the station at each time step t, it is assumed that the SA finds the probability distribution of the Poisson process of arriving FCEVs. Similar to the scheduling strategy of the PHEVs, first come first serve scheduling strategy is employed for refilling of FCEVs. Then, the SA by assuming that hydrogen demand of each FCEV is uniformly distributed between 0 and 5 kg, finds the required hydrogen amounts in kg per hour. Then, the SA by multiplying these amounts by HHVH2 which is equal to 39.7 kWh/kg, forms the hydrogen power demand profile of the charging station, Psta, hyd (t ) .

3.4. Control agent (CA) The CA coordinates the interactions between the field level agents. This agent is an intelligent goal-based agent that perceives the amount of power produced by PV arrays, state of charge of the battery bank, the amount of hydrogen stored in the hydrogen tank, residential loads, the uncharged energy of PHEVs and FCEVs at the predetermined time, and the information on the operation of the components of the microgrid. The CA perceives this information by receiving them from the field level agents. Allowing the GA to store the surplus power produced by PV arrays or supply the shortage of power production for meeting the residential loads, allowing the SA to charge the PHEVs by the surplus power and refill the FCEVs by the stored hydrogen in the tank, and sending its percepts to the DA are its actions. Coordination of the interactions between the field level agents is the only goal of this agent and the entire microgrid can be considered as its environment. 3.4.1. Functions of the CA This agent, after receiving a request for operating the microgrid with determined sizes from the DA and sending them to the GA and SA; at each time step t, requests the GA for the next hour value of power 458

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi

finds that the current level of hydrogen in the tank is enough for satisfying the required hydrogen load, it will ask the GA to supply the hydrogen load of the station from its tank to the amount of Psta,hyd [kW]. On the other hand, if the CA finds that only a fraction of hydrogen load can be met, it will ask the GA to supply the hydrogen load until the hydrogen level of the tank is minimized. In these two cases, the CA after sending the aforementioned requests to the GA, receives the updated value of Etank (t ) from it. Unlike these two cases, if the CA finds that the current level of the stored hydrogen in the tank is the lowest possible amount, it will not ask the GA to supply the hydrogen load. In all of these cases, the CA informs the SA about the decision made and receives the information on the operation of the EVSEs and refilling infrastructures containing Qsta,el and Qsta,hyd from the SA. Finally, the CA following the receipt of a request for operating the microgrid with the determined sizes from the DA, finds the information on the operation of the microgrid containing Qload, Qsta,el, and Qsta,hyd by applying the above procedure and sends them to the DA.

produced by PV arrays, requests the LA for the next hour value of residential loads, and requests the SA for the next hour values of charging and hydrogen power demands of the station for charging of PHEVs and refilling of FCEVs. Then, the GA sends the next hour value of the output power of PV generator (PPV), the LA sends the next hour value of residential electrical loads (Pload), and the SA sends the next hour values of charging demand (Psta,el) and hydrogen demand (Psta,hyd) to the CA. After receiving the aforementioned data from the field level agents, the CA establishes the following equation to check whether the output power of the PV arrays can satisfy the residential loads and charging demand of the station or not:

ΔP = PPV − (Pload/ηconv ) − (Psta, el/ηsta, el ).

(17)

According to the value of ΔP, three situations may happen: Situation A: when ΔP is positive, the CA asks the GA to store the surplus power of ΔP and satisfies the charging demand of the station and informs the SA about the decision made; therefore, it establishes the following equation:

Psta, delivered (t ) = Psta, el/ηsta, el .

3.5. Design agent (DA)

(18)

This agent has the highest level of the architecture of the proposed MAS and is responsible for optimal sizing of the components of the microgrid. This agent is an intelligent goal-based agent that its percepts are same as the CA. Requesting the CA for sending its percepts at each time step and minimizing the net present cost (NPC) function of the microgrid through PSO algorithm with the goal of finding the optimal sizes of the components of the microgrid are its actions and the entire microgrid can be considered as its environment.

Situation B: when ΔP is zero, the CA does not send any request to the GA and satisfies the charging demand of the station and informs the SA about the decision made. Situation C: when ΔP is negative, in order to check whether the output power of the PV arrays can satisfy the residential loads or not, this agent establishes the following equation:

ΔP′ = PPV − (Pload/ηconv ).

(19)

In this situation, when ΔP′ is positive, it means that the output power of PV arrays can satisfy the residential loads but some fraction of the charging load must be shed and the allocated power for the charging of PHEVs is as follows:

Psta, delivered (t ) = ΔP′ (t ).

3.5.1. System cost In this paper, the NPC method is used to calculate the total cost of the system. The NPC of each component can be calculated by the following equation (Hakimi, Tafreshi, & Kashefi, 2007):

(20)

NPC = N × ⎛capital cost+replacement cost × K + O&M cost ⎝ 1 ⎞, × CRF(ir , R) ⎠

Therefore, the CA does not send any request to the GA and satisfies a fraction of charging load and informs the SA about the decision made. In this situation, when ΔP′ is equal to zero, it means that the output power of PV arrays can satisfy the residential loads but the entire charging load must be shed and no power should be allocated for the charging of PHEVs; hence,

Psta, delivered (t ) = 0.

⎜

⎟

where CRF and K are the capital recovery factor and single payment present worth, respectively that are defined by the following equations:

(21)

CRF (ir , R) =

Therefore, the CA does not send any request to the GA and does not allocate any power for the charging of PHEVs and informs the SA about the decision made. In this situation, when ΔP′ is negative, it means that the output power of PV arrays cannot satisfy the residential loads and some fraction of them must be shed. Moreover, no power should be allocated for the charging of PHEVs. Therefore, the CA requests the GA for compensating the shortage of power generation required for satisfying the residential loads and does not allocate any power for the charging of PHEVs and informs the SA about the decision made. Then, the CA receives the information on the operation of the generation and storage components of the microgrid from the GA and when it finds that the residential loads are not met, it calculates the loss of residential loads, Qload (t ) according to the following equation:

Qload (t ) = Pload (t ) − Pgen (t ),

Y

K=

ir (1 + ir ) R , (1 + ir ) R − 1

1 , (1 + ir ) L × n

(25)

(26)

R Y = ⎡ ⎤ − 1; ifR is dividable toL, ⎣L⎦

(27)

R Y = ⎡ ⎤; ifR is not dividable toL, ⎣L⎦

(28)

where L is the lifetime of each component [yr], N is the optimal number/capacity of each component, R is the lifetime of the project [yr] that is considered to be 20 years, and ir is the real interest rate which is considered to be 6%. The objective function can be defined as

NPC = NPCPV + NPCele + NPCtank + NPCfc + NPCbat + NPCref & reac

where Pgen (t ) is the total generation of the considered distributed energy resources that can be obtained as ⎜

∑ n=1

(22)

EBat (t ) − EBat , min ⎞ Pgen (t ) = ⎜⎛PPV (t ) + Pfc (t ) + ⎛ × ηdch⎞⎟ × ηconv . Δt ⎝ ⎠ ⎝ ⎠

(24)

+ NPCconv + NPCsta + NPCloss − load + NPCloss − sta, el + NPCloss − sta, hyd, (29) where NPCloss−load, NPCloss−sta,el, and NPCloss−sta,hyd are the NPCs for the loss of residential loads, the loss of electric power demand of the station, and the loss of hydrogen power demand of the station, respectively that can be described by the following equations:

⎟

(23)

Then, the CA checks whether the current level of the stored hydrogen in the tank satisfies the required hydrogen load or not. If the CA 459

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi 8760

NPCloss − load =

∑

Qload (t ) × Cpen1 ×

t=1

1 , CRF (ir , R)

8760

∑

NPCloss − sta, el =

Qsta, el (t ) × Cpen2 ×

t=1

1 , CRF (ir , R)

8760

NPCloss − sta, hyd =

∑

Qsta, hyd (t ) × Cpen3 ×

t=1

1 . CRF (ir , R)

in this figure, after the determination of the sizes of the components through PSO by this agent, the determined sizes of the components will be sent to the CA. Accordingly, the CA sends the determined sizes of the components to the GA and SA. Then, the CA coordinates the GA with the LA and SA and sends the information on the operation of the microgrid to the DA. After receiving the aforementioned information, the DA calculates the objective function and ELF indices for the residential loads (ELFload ) , the electric power demand of the station (ELFsta, el ) , and the hydrogen power demand of the station (ELFsta, hyd ) . At this stage, if ELFload is lower than 0.01, ELFsta,el and ELFsta,hyd are lower than 0.1, and the energy content of the hydrogen tank and battery packs at the end of a year are not less than their initial energy, the obtained sizes will be selected as optimum sizes.

(30)

(31)

(32)

where Qload (t ) , Qsta, el (t ) , and Qsta, hyd (t ) represent the loss of residential loads, the loss of electric power demand of the station, and the loss of hydrogen power demand of the station at time step t [kWh], respectively. Moreover, Cpen1 and Cpen2 are the costs of electricity interruptions for the residential loads and PHEV charging station [$/kWh], respectively. Furthermore, Cpen3 is the cost of the interruption of hydrogen piping for refilling of FCEVs that arrive at the station. In this paper, Cpen1 is considered to be 5.6 $/kWh, while Cpen2 and Cpen3 are considered to be 0.56 $/kWh. The objective function is minimized through PSO algorithm in the DA. It is worth mentioning that at each time step t, the storage capacity constraints of the battery bank and hydrogen tank are checked in the GA and the power balance constraint is checked in the CA.

4. Results and discussion The aforementioned MAS-based architecture for the optimal sizing of the components of the proposed cooperative self-sustainable multicarrier microgrid is simulated and the optimum combination of the components is calculated. The system is optimized using the PSO algorithm at the design level of the considered architecture for the proposed MAS. The simulation of the proposed MAS is performed using MATLAB software. The annual solar irradiance data belong to Hendurabi Island in the Persian Gulf captured by one sample per hour precision that is shown in Fig. 6. The GA receives this curve as an input. The annual load curve of the residential electricity demand is shown in Fig. 7 that belongs to the residential area in Hendurabi Island. The considered load curve has a peak value of 60 kW and the LA receives it as an input. The annual electric and hydrogen power demands of the station can be updated during the operation of the microgrid according to the proposed DSM strategy in the MAS, as discussed in subsection 3.3.2. In obtaining the charging and hydrogen load curves, it is assumed that there are 50 citizens owning PHEVs and 50 citizens owning FCEVs. In order to find the number of vehicles being charged or refilled at each time step, M/M/c queuing model is utilized. Furthermore, PHEV charging rate is considered to be 4 kW. Moreover, the specifications of the components of the microgrid are shown in Table 1 (Dehghan et al., 2009; DOE Hydrogen & Fuel Cells Program Record, 2009; Forward, Glitman, & Roberts, 2013; Hakimi & Moghaddas-Tafreshi, 2009; Hassanzadehfard et al., 2015; Smaoui et al., 2015). The DA receives these specifications as input data. It is worth reminding that the sizes of the anaerobic reactor-reformer and hydrogen compressor-dispenser systems are fixed and do not participate in the optimal sizing procedure. Simulations are performed in two cases. In the first case that resembles non-interactive methods, we consider the electric and hydrogen power demands of the station as fixed loads and simulate the MAS without controlling the charging of PHEVs and refilling of FCEVs. The electric and hydrogen power demands of the station that are considered as fixed loads are shown in Figs. 8 and 9, respectively. The SA receives these two curves as inputs. The simulation process implemented in this case is a non-interactive simulation. The results of solving the optimal sizing problem using the MAS considering the charging and hydrogen power demands of the station as fixed loads (not implementing the DSM strategy that is discussed in subsection 3.3.2) are shown in Table 2. The table includes the optimal number of PV arrays, battery packs, and EVSEs, and the optimal capacities of the electrolyzer, hydrogen tank, fuel cell, and DC/AC converter. In this case, the total cost of the microgrid considering the penalties is equal to 1.56931 $M. In the second case, we assume that the considered DSM strategy in the proposed MAS controls the charging of PHEVs and refilling of FCEVs depending on the hour of the day. In this case, the electric and hydrogen power demands of the station are considered as deferrable

3.5.2. Reliability Equivalent loss factor (ELF) is considered as the reliability index in this paper because it contains information about the number and magnitude of the outages that occur in the system and can be calculated by the following equation (Garcia & Weisser, 2006):

ELF =

1 N

N

∑ t=1

Q (t ) . D (t )

(33)

where Q (t ) is the amount of lost load at time step t [kWh], D (t ) is the amount of load demand at time step t [kWh], and N is the number of time steps (Garcia & Weisser, 2006). Accordingly, the ELF indices for the residential loads, the electric power demand of the station, and the hydrogen power demand of the station can be described by the following equations:

ELFload =

ELFsta, el =

1 8760 1 8760

ELFsta, hyd =

8760

∑ t=1 8760

∑ t=1

1 8760

8760

∑ t=1

Qload (t ) , Pload (t )

(34)

Qsta, el (t ) , Psta, el (t )

(35)

Qsta, hyd (t ) Psta, hyd (t )

. (36)

In this paper, we consider that ELFload should be lower than 0.01, while ELFsta,el and ELFsta,hyd should be lower than 0.1. 3.5.3. Functions of the DA In order to minimize the objective function, particle swarm optimization algorithm is used in this analysis that was first introduced by Kennedy and Eberhart in (Kennedy & Eberhart, 1995). Compared to other alternative methods, PSO is very simple and powerful because instead of using a survival of fittest approach, the members of PSO population interact and influence each other (Hussain, Arif, Aslam, & Shah, 2017). The general optimization process is presented in Fig. 4. This process is embedded in the design agent of the proposed multiagent system. It can be seen from Fig. 4 that the DA is aware of the specifications of the components and among the combinations of the components that satisfy the system constraints, it selects the combination of components with the lowest cost. Fig. 5 shows the flowchart of the algorithm describing how the DA acts in the MAS to find the optimal sizes of the components. As shown 460

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi

Fig. 4. Flowchart of a single-run optimization process embedded in the DA.

With respect to the results, it can be seen that the optimal solution in both cases completely avoids fuel cell, so the electrolyzer and hydrogen tank are only used to charge the FCEVs. We, therefore, performed a sensitivity analysis and evaluated how much cheaper the hydrogen sub-system would need to become in the future, relative to the batteries, to be cost-effective. In this regard, the capital, replacement, and O&M costs of battery packs and hydrogen sub-system are reduced within reasonable bounds and the model is applied to each cost combination to find the optimal system type. It is important to note that in obtaining the optimal system type for each cost combination, we have assumed that the considered DSM strategy in the proposed MAS controls the charging of PHEVs and refilling of FCEVs depending on the hour of the day. For simplicity, it is assumed that the capital, replacement, and O&M costs of battery packs and hydrogen sub-system will be reduced at the same rate. The effects of changes in the costs of battery packs and hydrogen sub-system on the structure of the proposed microgrid are shown in Fig. 11. This type of graphical representation provides information about the optimal system structure at certain costs of hydrogen sub-system and battery packs. The optimal system type for each cost combination can be noticed in Fig. 11. In this figure, the current costs of battery packs and hydrogen sub-system are denoted by number 1. According to the sensitivity analysis, if we suppose that the battery costs will not change in the future years, the hydrogen sub-system costs should reduce to the about 0.7 of current costs so that a fuel cell would be selected instead of the battery bank. As stated in the previous sections, we aim at designing this microgrid for Hendurabi Island in the Persian Gulf. The obtained results belong to a fictitious microgrid that can be implemented in Hendurabi Island by investors in the near future. Currently, local diesel generators supply the required electrical power of Hendurabi Island. Because of the environmental issues associated with diesel generators, two development plans are under review by Hendurabi Island Council. One of them is installing the submarine power cables and supplying the electricity from Chiruye Harbor which is located about 9 km from Hendurabi Island. The other solution is constructing a self-sustainable microgrid like the microgrid we have proposed.

loads and the DSM strategy that is discussed in subsection 3.3.2 is implemented. The results of solving the optimal sizing problem using the proposed MAS considering the electric and hydrogen power demands of the station as deferrable loads and implementing the proposed DSM strategy are shown in Table 2. In this case, the total cost of the microgrid considering the penalties is equal to 1.46833 $M that shows a circa 6% cost saving in comparison with case1. Fig. 10 shows the total cost of the microgrid in terms of the iterations for the two cases. The results show that using the proposed flexible charging method in our proposed MAS decreases the optimal size of PV arrays and hydrogen tank, and increases the optimal size of EVSEs, battery packs, and electrolyzer, which in turn decreases the total cost of the system in comparison with case1. This is mainly because of that the cost of the reduced number of PVs is more than the cost of added EVSEs and battery packs. In the second case, the reason behind the decreased number of PVs is that our proposed MAS uses a DSM strategy that avoids overload during peak hours. The reason behind the increased number of EVSEs is that according to the decreased number of PV arrays, the SA needs more EVSEs so that it can charge the arrived PHEVs in less time. The reason behind the increased number of battery packs is that the battery bank should supply more energy in comparison with case1 in the evening hours to satisfy the residential loads because of the decreased number of PV arrays. In addition, there is not much difference in the determined capacities of the converter in the two cases because the residential loads are considered to be non-deferrable in both cases. Furthermore, results show that deferring the refilling of FCEVs through the proposed MAS decreases the size of the hydrogen tank and increases the size of the electrolyzer that also happen due to the decreased number of PV arrays. Our findings confirm that implementing the DSM strategies in the optimal sizing procedure of the self-sustainable microgrids can enhance the results and decrease the total cost. This underlines how important it is to develop a MAS-based framework for solving the optimal sizing problem of the self-sustainable microgrids in which different privatelyowned entities can interact and cooperate with each other to provide the energy requirements of the customers. 461

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi

2. the lifetime of our project is 20 years which is 5 more years than the alternative project; 3. we have designed a charging/refilling station for charging of PHEVs and refilling of FCEVs that has not been considered in the alternative project; 4. the power generation units in our microgrid can be easily extended based on the load growth in the future which would be difficult and costly in the alternative project as it needs re-installation of submarine power cables. Furthermore, we have reduced the total cost of our microgrid by the proposed MAS that controls the charging of PHEVs and refilling of FCEVs depending on the hour of the day. It is shown that our proposed MAS for optimal sizing of the microgrid components can reduce the total cost of the microgrid by ∼6%. It is a very important result because it shows that implementation of our proposed microgrid would be a better and more cost-effective solution than constructing the submarine power cables. In order to further validate the results, the dynamic payback period (DPP) of our project will be calculated in the following. The DPP combines the basic approach of the static payback period method with the discounting cash flow used in the NPC model and it can be defined as a period after which the capital invested has been recovered by the discounted net cash flows from the project (Götze, Northcott, & Schuster, 2016). In this application, the DPP can be calculated by the following equation: DPP

∑

S (1 + ir )−t − NPC = 0, (37)

t=0

where S is the annual income generated from selling the electricity to the residential customers and the owners of PHEVs and selling the hydrogen to the owners of FCEVs, ir is the real interest rate which is considered to be 6%, and NPC is the total cost of the microgrid considering the penalties which is equal to 1.46833 $M for our proposed system. In this paper, it is assumed that the electricity price is equal to 0.4 $/kWh. Furthermore, it is assumed that the owners of PHEVs and FCEVs that arrive at the station should pay 0.4 $/kWh and 16 $/kg, respectively. It is notable that the cumulative annual residential loads, PHEV charging loads, and FCEV refilling loads are equal to 188671 kWh, 122752 kWh, and 7884 kg, respectively. Therefore, S can be calculated as follows:

Fig. 5. Flowchart of the algorithm simulating the design agent.

It is reported that supplying power from Chiruye Harbor by submarine power cables to the Hendurabi Island will cost about 2 $M and can satisfy the electrical power demands of the Hendurabi Island over the next 15 years (Zabandan, 2017). According to the obtained results, our proposed self-sustainable microgrid has the following benefits over constructing the submarine power cables:

$ ⎞ $ ⎞ × 188671 kWh⎞ + ⎛ ⎛0.4 × 122752 kWh⎞ S = ⎛ ⎛0.4 kWh ⎠ kWh ⎠ ⎠ ⎝⎝ ⎠ ⎝⎝ ⎜

⎟

⎜

$ ⎞ ⎞ + ⎜⎛ ⎜⎛16 ⎟ × 7884 kg⎟ = 250713.2 $, kg ⎠ ⎝⎝ ⎠

1. our plan is about 0.5 $M cheaper;

Fig. 6. Annual solar irradiance.

462

⎟

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi

Fig. 7. Annual load curve of the residential electricity demand.

Table 1 Specifications of the microgrid components (Anonymous, 2018b; Dehghan et al., 2009; DOE Hydrogen and Fuel Cells Program Record, 2009; Forward et al., 2013; Hakimi and Moghaddas-Tafreshi, 2009; Hassanzadehfard et al., 2015; Smaoui et al., 2015). Component

Capital cost

Replacement cost

O&M cost

Lifetime [yr]

Efficiency [%]

PV array Electrolyzer Hydrogen tank Fuel cell Battery pack DC/AC converter EVSE Reactor-reformer Compressor-dispenser

2000 $/unit 1500 $/kW 500 $/kg 2000 $/kW 264 $/unit 700 $/kW 2000 $/unit 875 $ 100 k $

1500 $/unit 1000 $/kW 450 $/kg 1500 $/kW 260 $/unit 650 $/kW 1800 $/unit 785 $ 80k $

0 $/unit/yr 15 $/kW/yr 5 $/kg/yr 100 $/kW/yr 2.64 $/unit/yr 7 $/kW/yr 20 $/unit/yr 5 $/yr 200 $/yr

20 20 20 5 3 15 20 20 20

15.4 90 95 50 85 90 86.5 – 49

Fig. 8. Annual electric power demand curve of the station.

Fig. 9. Annual hydrogen power demand curve of the station.

463

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi

Table 2 Optimal size of the components for each case. Case

PV arrays

Battery packs

Electrolyzer [kW]

Hydrogen tank [kg]

Fuel cell [kW]

Converter [kW]

EVSEs

Case1 Case2

419 353

108 124

91.71 101.98

48.92 20.03

0 0

49.93 49.81

11 19

charging and refilling load demands of the station. It is observed that according to the considered DSM strategy in the proposed MAS which controls the charging of PHEVs and refilling of FCEVs and shifts the charging and refilling demands to off-peak hours, overloading can be avoided. In addition, using the flexible refilling procedure of FCEVs could reduce the size of the hydrogen tank by more than half. For these two reasons, by using the proposed MAS for optimal sizing of the components of the microgrid, the components can be operated better and a lower NPC for a 20-year investment horizon achieves. Therefore, the key result is a saving of ∼6% in the target system. Accordingly, a sensitivity analysis is performed to find that based on the changes in the costs of battery storage and hydrogen sub-system technologies in the future years, which type of backup system would be more cost-effective. The proposed microgrid is suitable for Hendurabi Island due to the following reasons: 1. the amount of solar radiation is considerable throughout the year; 2. large quantities of municipal solid wastes are available; 3. financial penalties for air pollution are very high because it is a tourist island. In the proposed microgrid, the battery bank/fuel cell can be considered as a backup for the PV power generation system that increases the reliability of supplying the electricity demand. Indeed, the energy storage components overcame the intermittent nature of the PVs. Besides, the hydrogen tank is fed by both the electrolyzer and reactorreformer system that increases the reliability of supplying the hydrogen load of the station. It is worth noting that the proposed multi-agent system is a general framework that can be used in the hybrid renewable energy system planning and design programs that include various privately-owned entities. In order to address the problem of selecting battery packs or fuel cell with their technological progress in the future years, future work aims to develop a MAS based on a dynamic approach for optimal sizing of the multi-stakeholder microgrids.

Fig. 10. Total cost in terms of the iterations.

References Fig. 11. Sensitivity of the battery costs to the hydrogen sub-system costs.

Aghaei, J., Esmaeel Nezhad, A., Rabiee, A., & Rahimi, E. (2016). Contribution of plug-in hybrid electric vehicles in power system uncertainty management. Renewable and Sustainable Energy Reviews, 59, 450–458. Ahmad, F., Alam, M. S., & Asaad, M. (2017). Developments in xEVs charging infrastructure and energy management system for smart microgrids including xEVs. Sustainable Cities and Society, 35, 552–564. Amini, M. H., Moghaddam, M. P., & Karabasoglu, O. (2017). Simultaneous allocation of electric vehicles’ parking lots and distributed renewable resources in smart power distribution networks. Sustainable Cities and Society, 28, 332–342. Anonymous (2018a). Electric motor werks company. [Internet]. [cited 2017 Sep. 5]. Available from: https://emotorwerks.com/index.php/dc-charging-systems/product/ 10-smartcharge-12000-a-12kw-universal-voltage-ev-charger-fully-assembled-tested/ category_pathway-17. Anonymous (2018b). Pure energy centre company. [Internet]. [cited 2017 Sep. 5]. Available from: https://pureenergycentre.com/hydrogen-fueling-station. Baghaee, H. R., Mirsalim, M., Gharehpetian, G. B., & Talebi, H. A. (2016). Reliability/ cost-based multi-objective Pareto optimal design of stand-alone wind/PV/FC generation microgrid system. Energy, 115, 1022–1041. Bernal-Agustín, J. L., & Dufo-López, R. (2009). Simulation and optimization of standalone hybrid renewable energy systems. Renewable and Sustainable Energy Reviews, 13, 2111–2118. Bhowmik, C., Bhowmik, S., Ray, A., & Pandey, K. M. (2017). Optimal green energy planning for sustainable development: A review. Renewable and Sustainable Energy Reviews, 71, 796–813. Bhuiyan, F. A., Yazdani, A., & Primak, S. L. (2015). Optimal sizing approach for islanded

hence, by solving Eq. (37) containing these values, DPP will be equal to 6.31 years. Therefore, it can be concluded that our project starts getting profit after 6.31 years that is a logical period and it would be a profitable project for investors to invest in. 5. Conclusion and future work This paper developed a MAS to find the optimal sizes of the components of a cooperative self-sustainable multi-carrier microgrid. We have found a way to the optimal sizing of the components of a selfsustainable multi-carrier microgrid that includes multiple stakeholders. Unlike the non-interactive methods, the proposed system finds the optimal sizes of the components based on the interactions between different entities in the microgrid which enables application of a DSM strategy to the optimal sizing problem of the microgrid. Two comparative cases are simulated with different conditions for supplying the 464

Sustainable Cities and Society 38 (2018) 452–465

S. Mohseni, S.M. Moghaddas-Tafreshi

Kantamneni, A., Brown, L. E., Parker, G., & Weaver, W. W. (2015). Survey of multi-agent systems for microgrid control. Engineering Applications of Artificial Intelligence, 45, 192–203. Kennedy, J., & Eberhart, R. C. (1995). Particle swarm optimization. Proceedings of the 1995 IEEE international conference on neural networks (pp. 1942–1948). Vol. 4. Kerdphol, T., Qudaih, Y., & Mitani, Y. (2016). Optimum battery energy storage system using PSO considering dynamic demand response for microgrids. Electrical Power and Energy Systems, 83, 58–66. Khan, M. W., & Wang, J. (2017). The research on multi-agent system for microgrid control and optimization. Renewable and Sustainable Energy Reviews, 80, 1399–1411. Kulkarni, V. G. (1999). Modeling, analysis, design, and control of stochastic systems (1st ed.). New York: Springer. Lasseter, R., Akhil, A., Marnay, C., Stephens, J., Dagle, J., Guttromson, R., et al. (2002). White paper on integration of distributed energy resources: The CERTS microgrid concept (CERTS, CA,R ep. LBNL-50829). Leemput, N., Van Roy, J., Geth, F., Tant, P., & Driesen, J. (2011). Comparative analysis of coordination strategies for electric vehicles. Proceedings of IEEE PES innovative smart grid technologies (ISGT) (pp. 1–8). Letendre, S., Denholm, P., & Lilienthal, P. (2006). Electric & hybrid cars: New load or new resource? Reston, VA: Public Utilities Fortnightly. Li, B., Roche, R., Paire, D., & Miraoui, A. (2017). Sizing of a stand-alone microgrid considering electric power, cooling/heating, hydrogen loads and hydrogen storage degradation. Applied Energy, 205, 1244–1259. Li, B., Roche, R., & Miraoui, A. (2017). Microgrid sizing with combined evolutionary algorithm and MILP unit commitment. Applied Energy, 188, 547–562. Liu, Z., Chen, Y., Zhuo, R., & Jia, H. (2018). Energy storage capacity optimization for autonomy microgrid considering CHP and EV scheduling. Applied Energy, 210, 1113–1125. http://dx.doi.org/10.1016/j.apenergy.2017.07.002. Available from: http://www.sciencedirect.com/science/article/pii/S0306261917308656. Logenthiran, T., Srinivasan, D., Khambadkone, A. M., & Aung, H. N. (2010). Scalable Multi-Agent System (MAS) for operation of a microgrid in islanded mode. Proceedings of 2010 joint international conference on power electronics drives and energy systems (PEDES) & 2010 power India (pp. 1–6). Logenthiran, T., Srinivasan, D., Khambadkone, A., & Aung, H. (2012). Multiagent system for real-time operation of a microgrid in real-time digital simulator. IEEE Transaction on Smart Grid, 3, 925–933. Luo, Y., Shi, L., & Tu, G. (2014). Optimal sizing and control strategy of isolated grid with wind power and energy storage system. Energy Conversion and Management, 80, 407–415. Ma, Y., Yang, P., Wang, Y., Zhao, Z., & Zheng, Q. (2014). Optimal sizing and control strategy of islanded microgrid using PSO technique. Proceedings of 2014 IEEE PES Asia-Pacific power and energy engineering conference (APPEEC) (pp. 1–5). Maleki, A., Khajeh, M. G., & Rosen, M. A. (2017). Two heuristic approaches for the optimization of grid-connected hybrid solar–hydrogen systems to supply residential thermal and electrical loads. Sustainable Cities and Society, 34, 278–292. Mozafar, M. R., Moradi, M. H., & Amini, M. H. (2017). A simultaneous approach for optimal allocation of renewable energy sources and electric vehicle charging stations in smart grids based on improved GA-PSO algorithm. Sustainable Cities and Society, 32, 627–637. Nojavan, S., Majidi, M., & Esfetanaj, N. N. (2017). An efficient cost-reliability optimization model for optimal siting and sizing of energy storage system in a microgrid in the presence of responsible load management. Energy, 89–97. Saboori, H., Hemmati, R., Ghiasi, S. M. S., & Dehghan, S. (2017). Energy storage planning in electric power distribution networks – A-state-of-the-art review. Renewable and Sustainable Energy Reviews, 79, 1108–1121. Sabri, M. F. M., Danapalasingam, K. A., & Rahmat, M. F. (2016). A review on hybrid electric vehicles architecture and energy management strategies. Renewable and Sustainable Energy Reviews, 53, 1433–1442. Sbordone, D., Bertini, I., Di Pietra, B., Falvo, M. C., Genovese, A., & Martirano, L. (2015). EV fast charging stations and energy storage technologies: A real implementation in the smart micro grid paradigm. Electric Power System Research, 120, 96–108. Shakouri, G. H., & Kazemi, A. (2017). Multi-objective cost-load optimization for demand side management of a residential area in smart grids. Sustainable Cities and Society, 32, 171–180. Smaoui, M., Abdelkafi, A., & Krichen, L. (2015). Optimal sizing of a stand-alone photovoltaic/wind/hydrogen hybrid system supplying a desalination unit. Solar Energy, 120, 263–276. Tanguy, K., Dubois, M. R., Lopez, K. L., & Gagné, C. (2016). Optimization model and economic assessment of collaborative charging using vehicle-to-building. Sustainable Cities and Society, 26, 496–506. Van Roy, J., Leemput, N., Geth, F., Büscher, J., Salenbien, R., & Driesen, J. (2014). Electric vehicle charging in an office building microgrid with distributed energy resources. IEEE Transactions on Sustainable Energy, 5, 1389–1396. Zabandan, M. (2017). Supplying the electricity demands of Hendurabi Island from Chiruye Harbor [Internet]. [cited 2017 Sep. 29]. Available from: http://www.irna.ir/fa/ News/82488655. [In Persian]. Zhao, B., Zhang, X., Li, P., Wang, K., Xue, M., & Wang, C. (2014). Optimal sizing, operating strategy and operational experience of a stand-alone microgrid on Dongfushan Island. Applied Energy, 113, 1656–1666.

microgrids. IET Renewable Power Generation, 9(2), 166–175. Brad, J., & O’Mahony, M. (2017). Modelling charging profiles of electric vehicles based on real-world electric vehicle charging data. Sustainable Cities and Society, 26, 203–216. Calvillo, C. F., Sánchez-Miralles, A., & Villar, J. (2016). Energy management and planning in smart cities. Renewable and Sustainable Energy Reviews, 55, 273–287. Chen, J., Zhang, W., Li, J., Zhang, W., Liu, Y., Zhao, B., et al. (2018). Optimal sizing for grid-tied microgrids with consideration of joint optimization of planning and operation. IEEE Transactions on Sustainable Energy, 9(1), 237–248. http://dx.doi.org/10. 1109/TSTE.2017.2724583http://ieeexplore.ieee.org/document/7972908. Chmutina, K., Wiersma, B., Goodier, C. I., & Devine-Wright, P. (2014). Concern or compliance? Drivers of urban decentralised energy initiatives. Sustainable Cities and Society, 10, 122–129. Clean Cities (2012). Plug-in electric vehicle handbook for public charging station hosts. Washington DC: U.S. Department of Energy. Coelho, V. N., Cohen, M. W., Coelho, I. M., Liu, N., & Guimarães, F. G. (2017). Multi-agent systems applied for energy systems integration: State-of-the-art applications and trends in microgrids. Applied Energy, 187, 820–832. DOE Hydrogen and Fuel Cells Program Record (2009). Energy requirements for hydrogen gas compression and liquefaction as related to vehicle storage needs. Washington DC: U.S. Department of Energy. Dehghan, S., Kiani, B., Kazemi, A., & Parizad, A. (2009). Optimal sizing of a hybrid wind/PV plant considering reliability indices, Vol. 56, World Academy of Science, Engineering and Technology527–535. Dimes, A. L., & Hatziargyrious, N. D. (2004). A multi-agent system for microgrids. Proceedings of the IEEE power engineering society general meeting (pp. 55–58). Dou, C.-X., & Liu, B. (2013). Multi-agent based hierarchical hybrid control for smart microgrid. IEEE Transactions on Smart Grid, 4(2), 771–778. Eichman, J. D., Mueller, F., Tarroja, B., Schell, L. S., & Samuelsen, S. (2013). Exploration of the integration of renewable resources into California’s electric system using the Holistic Grid Resource Integration and Deployment (HiGRID) tool. Energy, 50, 353–363. Fathima, A. H., & Palanisamy, K. (2015). Optimization in microgrids with hybrid energy systems – A review. Renewable and Sustainable Energy Reviews, 45, 431–446. Fazal, R., Solanki, J., & Solanki, S. K. (2012). Demand response using multi-agent system. Proceedings of 2012 North American power symposium (NAPS) (pp. 1–6). Forward, E., Glitman, K., & Roberts, D. (2013). An assessment of level 1 and level 2 electric charging efficiency. Burlington, VT: Vermont Energy Investment CorporationTransportation Efficiency Group. Götze, U., Northcott, D., & Schuster, P. (2016). Investment appraisal: Methods and models (2nd ed.). Berlin: Springer 2015. Galus, M. D., Fauci, R. L., & Andersson, G. (2010). Investigating PHEV wind balancing capabilities using heuristics and model predictive control. Proceedings of 2010 IEEE power and energy society general meeting (pp. 1–8). Gamarra, C., & Guerrero, J. M. (2015). Computational optimization techniques applied to microgrids planning: A review. Renewable and Sustainable Energy Reviews, 48, 413–424. Garcia, R. S., & Weisser, D. (2006). A wind-diesel system with hydrogen storage: Joint optimization of design and dispatch. Renewable Energy, 31, 2296–2320. Gordon-Bloomfield, N. (2014). Just how energy efficient is the toyota mirai? Time to do some maths [Internet]. [cited 2017 Sep. 5]. Available from: https://transportevolved.com/ 2014/11/19/just-energy-efficient-toyota-mirai-time-maths. Guo, L., Liu, W., Jiao, B., Hong, B., & Wang, C. (2014). Multi-objective stochastic optimal planning method for stand-alone microgrid system. IET, Generation, Transmission & Distribution, 8(7), 1263–1273. Hakimi, S. M., & Moghaddas-Tafreshi, S. M. (2009). Optimal sizing of a stand-alone hybrid power system via particle swarm optimization for Kahnouj area in south-east of Iran. Renewable Energy, 34, 1855–1862. Hakimi, S. M., & Moghaddas-Tafreshi, S. M. (2014). Optimal planning of a smart microgrid including demand response and intermittent renewable energy resources. IEEE Transaction on Smart Grid, 5(6), 2889–2900. Hakimi, S. M., Tafreshi, S. M., & Kashefi, A. (2007). Unit sizing of a stand-alone hybrid power system using particle swarm optimization (PSO). Proceedings of the international conference on automation and logistics (pp. 3107–3112). Hamad, A. A., & El-Saadany, E. (2016). Multi-agent supervisory control for optimal economic dispatch in DC microgrids. Sustainable Cities and Society, 27, 129–136. Hassanzadehfard, H., Moghaddas-Tafreshi, S. M., & Hakimi, S. M. (2015). Optimization of grid connected micro-grid consisting of PV/FC/UC with considered frequency control. Turkish Journal of Electrical Engineering & Computer Sciences, 23, 1–16. Howell, S., Rezgui, Y., Hippolyte, J.-L., Jayan, B., & Li, H. (2017). Towards the next generation of smart grids: Semantic and holonic multi-agent management of distributed energy resources. Renewable and Sustainable Energy Reviews, 77, 193–214. Hu, J., Morais, H., Sousa, T., & Lind, M. (2016). Electric vehicle fleet management in smart grids: A review of services, optimization and control aspects. Renewable and Sustainable Energy Reviews, 56, 1207–1226. Hussain, A., Arif, S. Y., Aslam, M., & Shah, S. D. A. (2017). Optimal siting and sizing of trigeneration equipment for developing an autonomous community microgrid considering uncertainties. Sustainable Cities and Society, 32, 318–330. Jung, J., & Villaran, M. (2017). Optimal planning and design of hybrid renewable energy systems for microgrids. Renewable and Sustainable Energy Reviews, 75, 180–191.

465