Real-time coordination of distributed energy resources for frequency control in microgrids with unreliable communication

Electrical Power and Energy Systems 96 (2018) 86–105

Contents lists available at ScienceDirect

Electrical Power and Energy Systems journal homepage: www.elsevier.com/locate/ijepes

Real-time coordination of distributed energy resources for frequency control in microgrids with unreliable communication Huadong Mo, Giovanni Sansavini

MARK

⁎

Reliability and Risk Engineering Laboratory, Institute of Energy Technology, Department of Mechanical and Process Engineering, ETH Zurich, Switzerland

A R T I C L E I N F O

A B S T R A C T

Keywords: Distributed energy resources System frequency fluctuation Reliability Open communication network Cyber-physical systems

The management of distributed energy resources (DER) via control strategies mitigates frequency fluctuations stemming from the volatility of renewable resources and fluctuating power demand. Recently, open communication networks are integrated with the traditional control strategies to overcome the ubiquity of DER system and the lack of dedicated communication infrastructures. However, open networks are exposed to communication degradation and can reduce the control performance. This work investigates the reliability of the integrated DER system and open communication networks, i.e. the cyber-physical microgrid system, with reference to the frequency control in the face of communication degradation. Adequate control strategy is provided by a discrete PID controller tuned via multi-objective particle swarm optimization. The integrated system is tested on a real-time platform with different MAC protocols and open-communication-network architectures to investigate how the communication degradation reduces the frequency control performance. Simulation results demonstrate that transmission delays and packet dropouts jeopardize the ability of the integrated system to maintain the system frequency deviation within bounds. In particular, the use of Ethernet ensures higher reliability as compared to 802.11 b/g. Moreover, the impact of interfering traffic and of the percentage of used bandwidth on the PID controller performance reduction is assessed. The optimized PID controller can compensate for communication degradation and uncertainty conditions of the microgrid, and ensures robustness against unknown network configurations.

1. Introduction The power sector is experiencing a structural trend towards decentralization stemming from the integration of large shares of renewable energy resources (RERs) [1]. This is fostered by distributed energy resources (DERs), which require the integration of power generation means located at or near the end-user side [2,3]. However, the stochastic nature of RERs and of the load demand induces system frequency fluctuations [4,5]. An effective control strategy is needed to keep the system frequency to its nominal value by balancing power generation and demand in real time. To this aim, automatic generation control (AGC) schemes are developed for damping frequency oscillations in distributed generation systems (DGS) [5–8]. AGC is performed by computing control signals based on the system frequency and delivering balancing inputs to various energy storage systems (ESSs) to absorb (release) the surplus (deficit) power from (to) the grid [8–10]. However, the ubiquity of DERs across wide areas and the complex structure of DGS hinder the development of dedicated communication infrastructures for the DGS with massive DERs [11–14].

⁎

Corresponding author. E-mail address: [email protected] (G. Sansavini).

http://dx.doi.org/10.1016/j.ijepes.2017.09.029 Received 10 May 2017; Received in revised form 21 July 2017; Accepted 19 September 2017 0142-0615/ © 2017 Elsevier Ltd. All rights reserved.

Recently, the AGC has been integrated with the open communication network, due to low cost, high speed, simple structure and flexible access. Data exchanges among PMUs, generators and the control center are provided by the open communication network in the form of time stamped data packets [7,13–15]. Stable AGC depends heavily on the performance of the open communication network [7–9,15–20]. Cognitive radio networks, Cellular Networks, Local Area Networks (LAN), Wide Area Networks (WAN) and Wireless Local Area Networks (WLAN) are employed as open communication infrastructures in these networked control systems [10,11,14]. However, open communication networks are exposed to various types of degradation processes, i.e. network-induced time delays [8,9,18,19], packet dropouts [20,21], failures of communication infrastructure [22], uncertain communication links [23] and cyberattacks [24]. As a result, the measurement signals (control signals) received by the control center (ESSs or generators) degrade, effective AGC cannot be carried out and the system frequency response worsens [9–13]. Studying the performance of open communication networks is critical for understanding the occurrence of time delays and packet dropouts.

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

Nomenclature

sampling interval of the PMU generated from a discrete uniform distribution in (0,2 Nc−1), where is the number of detected consecutive collisions τsc , τca time delay in sensor-to-controller and controller-to-actuator channel U (s ) , Y (s ) transfer functions of the control signal and system output indicates the transmission of Δf (kTs ) at kth period is sucγk = 1 cessful time instant when the n th data packet is received by the tn control center τscn transmission time of packet n from the PMU to the control center Ld expected packet loss probability indicates the control signal u (tn ) is not dropped γn = 1 time at which the m th data packet is received by the DER i tm m τca transmission time packet m from the control center to the DER i period that the interference node sends traffic data to the Ti network UNi uniformly distributed random number sampled at time period Ti in the interval [0,1] BWShare expected ratio of network bandwidth used by the interference node KP , KI , KD proportional, integral and derivative gain of the PID controller NL filter’s coefficient indicating location of pole in the derivative filter R system reliability TI total amount of time in which the system frequency remains smaller than the maximum permissible instantaneous frequency deviation T the total operating time of the AGC J objective function for the optimization of the PID controller η1 indicates the relative importance of the two terms η2 normalizing constant to scale both terms in a uniform range N number of MCS samples expectation of the stochastic objective function obtained E[J ] from MCS x ) objective functions (Δf (t ))2 and (Δu (t ))2 J1 (→ x ) , J2 (→ NP , NI maximum number of particles and iterations MP number of dimensions x i , vi current position and velocity of particle i c1, c2 cognitive and social factors r1, r2 random numbers drawn from a uniformly distributed interval [0,1] pbest , gbest local best-known position and global best-known position mfi , MF linear membership function and aggregate membership function

Ts RD

Acronyms DER Distributed Energy Resource AGC Automation Generation Control RERs Renewable Energy Resources DGS Distributed Generation Systems ESS Energy Storage Systems LAN Local Area Networks WAN Wide Area Networks WLAN Wireless Local Area Networks MAC Media Access Control CSMA/CD Carrier Sense Multiple Access with Collision Detection CSMA/AMP Carrier Sense Multiple Access with Arbitration on Message Priority PSO Particle Swarm Optimization MOPSO Multi-objective Particle Swarm Optimization DEG Diesel Engine Generator WTG Wind Turbine Generator PV Photovoltaic Generator BESS Battery Energy Storage System FESS Flywheel Energy Storage System HPS Hybrid Power System PMU Phasor Measurement Unit RTU Remote Terminal Unit MCS Monte Carlo Simulation GWTG , TWTG transfer function and time constant of the WTG GPV , TPV transfer function and time constant of the PV GDEG , TDEG transfer function and time constant of the DEG PW , Psol wind power and solar power vW , vcutin , vr , vcutout real-time, cut-in, rated and cut-out wind speed Pr ,WTG rated power of the wind turbine number of wind turbines in the wind farm NWT conversion efficiency and nominal operation temperature η , Tr of the PV cells maximum power temperature coefficient kpv Ta ambient temperature Φ sun irradiance level S measured area of the PV array u (t ) control signal sent out by the PID controller GFESS , TFESS transfer function and time constant of the FESS GBESS , TBESS transfer function and time constant of the BESS PFESS , PBESS output power of the FESS and BESS PDEG , PFESS , PBESS maximum rated output power of the DEG, FESS and BESS GHPS , M , D transfer function, inertia constant and damping constant of the HPS Δf (t ) power system frequency deviation PL power demand Tpre , Twait , Ttx , Tpost preprocessing time, waiting time, time for traveling across the channel, postprocessing time Td total time delay

AGC performance and reduce the stability region [9,10]. Packet dropouts refer to lost messages, which occupy network bandwidth but cannot reach destination. They affect the operations of DERs and the reduction of frequency fluctuations, particularly in uncertain network environments. Optimal feedback AGC regulators for DERs are investigated in numerous works for perfect communication networks and the impact of transmission delays and packet dropouts on the controller cannot be captured [28]. Robust PID controllers against constant or uniformly distributed time delays [8–11] are designed to cope with perturbations of the control parameters. Yet, constant or uniformly

To this aim, medium access and packet transmission must be analyzed. The media access control (MAC) layer is the lower layer of the data link layer of the Open System Interconnection model, and it is responsible for moving data packets among network interface cards across the communication channels. Several MAC protocols, e.g. CSMA/CD (Carrier Sense Multiple Access with Collision Detection, Ethernet), CSMA/ AMP (Carrier Sense Multiple Access with Arbitration on Message Priority, CAN) and 802.11 b/g (WLAN), prevent the collision of packets sent from different nodes across the same channel [14,25–27]. Time delays are variable, challenging to predict, deteriorate the 87

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

[18] and PI controller [39,40]. On the other hand, novel approaches for mitigating the impact of random time delays quantify robust delay margins [41]. The delay-margin-based sparsity-promoting wide-area control strategy, which requires few system observations, can reduce communication requirements, and obtain nearly-optimal performance compared to the centralized control [41]. Nevertheless, packet dropout has still the potential to affect the performance of this strategy. In this work, we investigate (a) the operations of the integrated DER system and open communication network, and (b) the design of optimal AGC strategies in the face of communication degradation. This work focuses on microgrids in islanded operations with real power generation and demand from a systemic perspective. The total generation of DERs supplies the demand. Therefore, the initial conditions of DERs are determined to balance the stochastic RERs and the demand. As a result,

distributed time delays cannot be generally assumed in realistic communication networks. In addition, recently studies focusing on primary and secondary control levels are extended to the power management level by considering fuzzy controllers [29,30], decentralized power management and sliding mode control strategies [31], static synchronous compensators [32], and two-degree-of-freedom feedback-feedforward robust controllers [33,34]. The reactive power reference can be determined and controlled by a novel application of radial basis function neural networks [35–37], to improve the power sharing and stability of microgrid with multi-DERs. To provide high reliability and robustness against network failure or time delays, droop-based control schemes are designed to specify the frequency of each DER unit by using the complementary loop and fuzzy logic controller [38], robust H∞ controller

Fig. 1. Schematics of (a) cyber layer, (b) physical layer and (c) control block diagram representing the physical layer of the integrated system.

(a)

(b)

(c) 88

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

DGS model. The open network and the communication degradation models are described in Section 3. Section 4 defines the reliability of the integrated system, and introduces the PSO-based PID controller. Section 5 presents the simulation results. Section 6 concludes the work.

voltage profiles and reactive power are neglected. This simplified model is adopted by several works [5,7,10], and is also integrated with microgrid power management systems [29–34]. Because wind and solar sources have intermittent characteristic, their power generations have stochastic behaviors. The probabilistic power patterns can be well justified by PDFs, so that Weibull and normal distribution for wind speed and solar irradiation [42,43]. Alternatively, in this work we consider actual wind speed and solar irradiation as in [44,45]. The historical wind speed and solar radiation dataset are from the Elia Grid, Belgium. The ability of the integrated system to maintain system frequency deviation within tolerance margins quantifies system reliability, and is evaluated by Monte Carlo simulation (MCS). Stochastic time delays are modelled by generating random congestion based on MAC protocols. Congestion of network channels depends on the activity level of the interfering traffic, which is the root cause of network-induced delays and packet dropouts [9,20,26,46]. The open communication network model is implemented via Truetime simulator testing different MAC protocols [24,47]. Truetime is a Matlab/Simulink-based simulator for real-time cyber-physical systems, which facilitates co-simulation of controller task execution in real-time kernels, network transmissions, and continuous plant dynamics. It can simulate most realistic communication networks, e.g. Ethernet, CAN, TDMA, FDMA, Round Robin, Switched Ethernet, FlexRay, PROFINET, 802.11b WLAN and 802.15.4 ZigBee. Multiple activity levels of the interfering traffic are simulated via the Interference node, which sends random interfering packets over the network. Packet dropout is described by Bernoulli-distributed variables [48]. To stabilize system frequency against RER, demand variability and communication degradation, a discrete PID controller is used [10,49]. The heuristic algorithms such as particle swarm optimization (PSO) [9–11,28,51], genetic algorithm (GA) [5], flower pollination algorithm [6], quasi-oppositional harmony search algorithm [7], Cuckoo Search algorithm [19], and artificial bee colony algorithm [50], have been introduced to provide easy implementation, conceptual simplicity, more flexibility, and independency to the initialization of the controller optimization procedure. PSO is adopted to minimize the stochastic objective function and achieve the optimal PID controller for various architectures and conditions of the open communication network, because the movement of particles is influenced by the local best-known positions as well as by the global best-known positions in the search space, PSO can avoid being trapped into local minimum [9–11,28,51]. However, the optimization of PID usually involves simultaneous optimizing two objectives, i.e. control performance and control effort, which are generally non-commensurable and conflicting with each other [52,53]. Better control performance means more control efforts and vice versa. Therefore, the multi-objective PSO algorithm is applied to achieve a best tradeoff between the two objectives [42,54,55]. Finally, the robustness and effectiveness of the proposed MOPSO-basedPID-controlled AGC against communication degradation is assessed. In addition, we conduct various simulations to test the performance of the integrated system with the MOPSO-optimized control strategy for three communication network architectures, under different uncertainty conditions and operating scenarios of the DERs and load. The performance of the integrated system with the perfect communication is used as a benchmark. The other two network architectures are the Ethernet and the hybrid network (a mix of Ethernet and 802.11b/g). Different network configurations, i.e. traffic conditions of the open communication networks, are assessed. The performance of the integrated system using the three network architectures are compared under the uncertainty conditions and the effectiveness and robustness of the proposed control strategy is shown. The rest of the work is organized as follows. Section 2 describes the

2. Model of DGS with DERs The schematic of the integrated system, which is made of a cyber layer and a physical layer, is illustrated by Fig. 1. The structure of physical layer in Fig. 1(b) is general, representative of DGS, and widely adopted in the literature [5,7,10,56]. It models a hybrid microgrid, which consists of conventional generators (diesel engine generator, DEG), RERs (wind turbine generator, WTG, and photovoltaic generator, PV) and ESSs (battery and flywheel energy storage system, BESS and FESS). The cyber layer Fig. 1(a) is responsible for the real-time communication between the control center and subsystems, i.e. providing data exchanges between the control center and controllable BESS, FESS and DEG in the physical layer. In addition, PMU measurements and control signals are transmitted via the shared and open communication network. The cyber layer is detailed in Section 3. The cyber layer and the physical layers have the same topology. Power imbalance is the root cause of system frequency fluctuations. The control center remotely monitors the ESS and the diesel generator to reduce the imbalance and to ensure good AGC performance. Following [10,56], the PID controller in the control center is responsible for the system frequency stabilization, and therefore we use it to represent the control center in the control block diagram shown in Fig. 1(c). The small signal stability analysis of the hybrid microgrid in Fig. 1(c), is based on time-domain simulations, i.e. transfer function models. In the AGC, the WTG, PV, DEG, FESS and BESS are described by first order transfer functions with specified gain and time constant. A centralized controller is used [7], as opposed to multiple decentralized controllers for each controllable component [5,57–59]. It enables easier maintenance and reduces wiring cost, and makes the AGC design problem traceable by reducing the number of controller parameters [9]. On the other hand, the centralized controller impacts the AGC performance, because a unique control signal is used by all the components. Nevertheless, current studies show that the centralized controller can ensure acceptable time-domain AGC performance [7,10,28]. The transfer functions GWTG (s ) , GPV (s ) , and GDEG (s ) of the WTG, PV and DEG, respectively, are expressed as

GWTG (s ) =

GPV (s ) =

1 P = WTG 1 + sTWTG PW

1 P = PV 1 + sTPV Psol

GDEG (s ) =

1 P = ∼ DEG 1 + sTDEG uDEG (t )

(1)

(2)

(3)

where TWTG , TPV and TDEG are the time constant of the WTG, PV and DEG. PWTG and PPV are the electrical power produced from the RERs, i.e. wind power PW and solar power Psol , which are generated from the data sets of wind speed and solar radiation of the Elia Grid, Belgium, and their curves are provided in the Appendix A. The DEG is controlled by uDEG (t ) sent by the remote control center and it the control signal ∼ generates power only when the RERs cannot meet the demand. The ESSs are critical in eliminating frequency fluctuations due to their fast response to the control signal. Based on [8,10,56,57], the transfer functions GFESS (s ) and GBESS (s ) of the FESS and BESS are given as

89

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

GFESS (s ) =

1 P = ∼ FESS 1 + sTFESS uFESS (t )

(4)

GBESS (s ) =

1 P = ∼ BESS 1 + sTBESS uBESS (t )

(5)

firmware. Ttx depends on the physical bandwidth, propagation spend and transmission distance. Twait is the major source of the time delay and it is influenced by the MAC protocol. For instance, if a collision with packets sent by other nodes is detected by the CSMA/CD protocol (Ethernet), the sender will back off for a random time [47]

where TFESS and TBESS are the time constant, PFESS and PBESS are the uBESS (t ) are the control signal of the FESS uFESS (t ) and ∼ output power, ∼ and BESS.

Twait =

minimum frame size × RD data rate

(12)

Remark 1. The DEG, FESS and BESS have rate constraint, i.e. |PDEG | < PDEG , |PFESS | < PFESS and |PBESS | < PBESS , where PDEG , PFESS and PBESS are the maximum rated output power of the DEG, FESS and BESS, respectively.

where RD follows a discrete uniform distribution in (0,2 Nc−1) and is the number of detected consecutive collisions. The time delays τsc and τca are determined by (9), and their impact on the transfer function from the control signal U (s ) to the system output Y (s ) is described as

The transfer function GHPS (s ) of the hybrid power system (HPS) models the relationship between the power imbalance, i.e. ΔPS−ΔPL , and the system frequency Δf (t )

Y (s ) = G (s ) e−(τsc + τca) s U (s )

GHPS (s ) =

Δf (t ) 1 = D + Ms ΔPS−ΔPL

where G (s ) denotes the transfer functions of the DERs. A second source of disturbance in the source-destination communication is packet dropout, which occurs for three major reasons, i.e. the network disconnection, time-out transmission and time-out retransmission [47]. The PMU sensor is time triggered, i.e. it takes measurement at every sampling interval Ts . The relationship between the frequency measured by the PMU and the frequency received by the control center is ∼ Δf (tn ) = γk Δf (kTs ) (14)

(6)

where M and D are the inertia constant and damping constant of the HPS [9], and PS is the total power generated, denoted by PWTG + PPV + PDEG + PFESS + PBESS . PL is the power demand and its curve is provided in Appendix A. According to [62], the linearized state-space realization of the microgrid in the physical layer of Fig. 1 can be expressed as

x ̇ = Ax + B1 w + B2 u y = Cx

where γk = 1 indicates the transmission of Δf (kTs ) is successful and tn is the time instant when the n -th data packet is received by the control center. If γk = 1, τscn = (tn−kTs ) , where τscn is the transmission time of packet n from the PMU to the control center. In real implementations of the CSMA/CD MAC protocol, the source node discards the packet and reports an error if ten consecutive data collisions occur [47]. Therefore, if congestion is severe, time delays increase and so does the number of packet dropouts. However, network congestion and packet dropout are challenging to link, and, therefore, packet dropout is usually modelled as a stochastic process [47,48,60]. Binary switching sequences are usually applied [37], which specify the expected packet dropout probability. The stochastic parameter γk of the binary switching sequence is Bernoulli distributed [38,60], taking value of 0 or 1 with

(7)

where

xT = [ΔPWTG ΔPPV ΔPDEG ΔPBESS ΔPFESS Δf ]

(8)

wT = [ΔPW ΔPsol ΔPL]

(9)

y = Δf

(10)

An elaborated expression of the microgrid state-space model is given in the Appendix B. 3. Model and implementation of the open communication network

P{γk = 0} = Ld

The model of the data transmission across the open communication network accounts for the composition of network-induced delays and for the stochastic packet dropout. We assess two general architectures for the communication network used in power systems, i.e. the Ethernet [24] and a mix of Ethernet and 802.11 b/g (henceforth called hybrid network) [10], and implement them via Truetime simulator [47].

(15)

where 0 < Ld < 1 is the expected packet loss probability. ∼ The frequency measurement Δf (t ) is not updated until the next ∼ packet is received, thus the evolution of Δf (t ) is given as ∼ ∼ Δf (t ) = Δf (tn ),tn ⩽ t < tn + 1 (16) On the other hand, the controller is event-driven, i.e. it updates the control signal and sends it to the DERs as soon as the control center receives an updated frequency measurement. Therefore, if the n -th ui (t ) , stored in the buffer of the packet is received, the control signal Δ∼ DER i (i = FESS, BESS, DEG) is updated as Δ∼ u (t ) = Δ∼ u (t ), t ⩽ t < t (17)

3.1. Model of time delay and packet dropout The PMU measurements and control signals are transmitted via a shared and open communication network. Limited bandwidth and random data traffic may result in re-transmission, multipath transmission and congestion. Consequently, an uncertain amount of time delay in data transmission is expected. Time delays can be categorized into two types: time delay τca at the forward channel, i.e. from the control center (controller) to DERs (actuators), and time delay τsc at the feedback channel, i.e. from the PMU (sensor) to the control center (controller). The time delay in the communication channel consists of four components, i.e. the preprocessing time Tpre , the waiting time Twait , the time for traveling across the channel Ttx and the postprocessing time Tpost [26,27]. Therefore, the total time delay Td can be expressed as

Td = Tpre + Twait + Ttx + Tpost

(13)

i

i

m

m

m+1

Δ∼ ui (tm) = γn u (tn )

(18) ∼ where γn = 1 if the control signal u (tn ) computed based on Δf (tn ) is not dropped (and γn = 0 otherwise) and tm is the time at which the m -th m m = (tm−tn ) , where τca is data packet is received by the DER i . If γn = 1, τca the transmission time packet m from the control center to the DER i . The control center sends control signals to the FESS first, then to the BESS and finally to the DEG. 3.2. Network architecture and model implementation

(11)

The architecture for the AGC of DERs via Ethernet and hybrid network, is illustrated by the cyber layer in Fig. 1. The data exchange

where Tpre and Tpost depend on the processing speed of the device 90

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

employed for monitoring and controlling DERs, offshore wind farms and smart home energy management systems [11]. The interference node simulates the interference user in the open network, who sends disturbing traffic over the channels and cause congestion. Generally, the length of data traffic is constant (i.e. 80 bytes in this work, same as the length of PMU measurement [47,61]), and the interference node sends it to the network at every time period Ti if

among the control center, DERs, interference node and PMU is wired in the Ethernet architecture. In the hybrid architecture, the data exchange between the routers and the RTU (BESS, FESS, DEG and PMU) is wireless and provided by the 802.11b/g. The data exchange among routers is wired and provide by the Ethernet. Low product prices make 802.11b/g more convenient and cost-effective compared to the Ethernet. As such, 802.11b/g has become an efficient approach to provide flexible data communication between routers and RTU, and is

Fig. 2. Sending time and receiving time for the (a) interference node, (b) control center, (c) DGE, (d) BESS and (e) FESS in the Ethernet architecture, for BWShare = 0.3 , Ti = 0.005 s and Ld = 5% .

91

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

UNi < BWShare

(19)

∼ ⎛ u (tn) = −⎜KP Δf (tn) + KI ⎝

where UNi is a uniformly distributed random number sampled at time period Ti in the interval [0,1], and BWShare is the expected ratio of the network bandwidth used by the interference node [47].

Remark 3. BWShare indicates the expected percentage of bandwidth used by the interference user. If the node sends out a message, it occupies the entire channel bandwidth, and other nodes cannot send messages until this channel is free, so as to avoid packet collisions. For example, BWShare = 0.1 means that 10% of bandwidth is used by the interference user, and from a statistical viewpoint, the interference user sends out disturbing traffic with probability 0.1.

• •

(20)

NL ⎞ C (z ) = −⎛KP + KI a (z ) + KD 1 + N L b (z ) ⎠ ⎝ ⎜

⎟

(21)

where NL is the filter’s coefficient indicating the location of the pole in T the derivative filter and a (z ) = b (z ) = z −s 1 . 4.2. Reliability of the integrated system

The architectures of the real-time AGC of DERs via the Ethernet and hybrid network all consists of:

•

k=0

ψ⎞ ∼ Δf (tk ) ∗Ts + KD ⎟ Ts ⎠

where KP , KI and KD are the proportional, integral and derivative gain ∼ ∼ of the controller, and ψ = Δf (tn )−Δf (tn − 1) . For the implementation of the discrete-time PID controller, the role of a first-order pole filter on the derivative action should be considered [63]. Therefore, the transfer function of the discrete-time PID controller in Eq. (20) is

Remark 2. The time interval Ti determines the activity level of the interference user. For example, Ti = 0.01 s means that the interference user tries to send out disturbing traffic every 0.01 s.

•

n

∑

Frequency deviations larger than the maximum permissible instantaneous frequency deviation compromise the components in the physical layer, reduce their useful lifetime and deteriorate the AGC performance [64]. In this work, the reliability is defined as the ability of the integrated system to ensure system frequency derivations smaller than the maximum permissible frequency deviation, in the face of stochastic RERs, uncertain load demand, variable time delays and random packet dropouts [65]. The value of the reliability R measures the ability of the integrated system to provide adequate AGC performance, and is computed as

PMU node (time-driven): the PMU takes measurements of the system frequency at every sampling interval Ts = 0.01 s and sends Δf (kTs ) to the control center over the network. The size of data packets is 80 bytes and the phase delay caused by the filter of the PMU is 0.006 s [61,62]. Control center node (event-driven): when a data packet from the PMU reaches the control center, the controller takes 0.002 s to compute the control signal u (tn ) and sends it to the DERs. The length of control signal is 500 bytes. DERs nodes (event-driven): the DEG, BESS and FESS adjust their operations based on the control signal received. Interference node (time-driven): it sends disturbing traffic over the network with period Ti and causes congestion, to generate different scenarios of time delay and packet dropout.

R=

TI T

(22)

where TI is the total amount of time in which the system frequency remains smaller than the maximum permissible frequency deviation and T is the total operating time of the AGC.

The two architectures for the open communication network are simulated using the Truetime simulator [47], which is a Matlab toolbox that generates scenarios of realistic network with different MAC protocols [14,25,47]. Fig. 2 presents the dynamics of sending and receiving time for data packets at the interference node, PMU, control center and DERs under the Ethernet architecture, for BWShare = 0.3, Ti = 0.005 s and Ld = 5%. The continuous lines show steps when a message is received by the corresponding node. The x coordinate of the steps identifies the receiving time of data packet, and the y coordinate identifies the time at which that corresponding packet was sent. Dotted vertical lines indicate that one of the nodes, i.e. interference node, PMU and control center, respectively in Fig. 2, sends out a data packet. The sending time marked by the dotted lines is consistent with the y value of step. In Fig. 2(a), the step (55, 46) in the control center shows that the PMU sends out a system frequency measurement at 46 ms, which is consistent with the dotted line showing a sending pulse from the PMU at 46 ms, and the control center receives it at 55 ms. Furthermore, Fig. 2 allows identifying the dropped packets. In Fig. 2(d), the control center sends out a packet to FESS at 29 ms (dotted lines marked with cross), but the FESS cannot receive it (no corresponding step with y = 29 ms), indicating that a packet dropout occurs in this transmission. Above two examples are highlighted in the Fig. 2 to provide better descriptions.

Remark 4. The maximum permissible instantaneous frequency deviation describes a symmetric interval, e.g. ± 0.8 Hz, around the ideal frequency baseline, i.e. Δf = 0 . The AGC is stable, if and only if Δf is within the tolerance interval. Due to the uncertainties in the DGS and the random communication degradation effects, the AGC performance evolves through stochastic trajectories. Thus, the reliability in Eq. (22) and objective function in Eq. (23) are stochastic and generally evaluated as the expected values of a stochastic process [47]. Thus, the reliability and the objective function are estimated via the MCS method [10]. In order to achieve a statistically acceptable estimate, the reliability is computed via 200 MCS-based samples, i.e. the integrated system is evaluated for 200 trials to compute the expected reliability. Our works improves on the convergence of MCS greatly, as compared to previous works [10]. In particular, Table 1 presents the comparison study among the reliability values derived using 10, 30, 50, 100, 200 and 300 samples. The results show that a reasonable degree of convergence is reached beyond 100 MCS samples. By comparing mean Table 1 Reliability (%) of integrated system with arbitrary PID controller ([0.52, 2.21, 1.00]) and Ethernet ([0.2, 0.005, 5%]) subject to different numbers of MCS samples.

4. Reliability of the integrated systems and PSO-tuned PID controller 4.1. Discrete-time PID controller Due to periodic PMU measurements and packet dropouts, the AGC process is discrete. Therefore, a discrete-time PID controller is adopted to compute the control signal 92

Number of MCS samples

Mean

95% Confidence interval

Variance

95% Confidence interval

Runtime (s)

10 30 50 100 200 300

92.29 91.74 92.78 92.82 92.89 92.93

[90.00, 94.58] [90.50, 92.99] [90.00, 93.56] [92.31, 93.32] [92.64, 93.14] [92.73, 93.13]

4.89 2.67 1.66 1.08 0.89 0.78

[3.72, 7.15] [2.03, 3.90] [1.26, 2.42] [0.82, 1.57] [0.74, 1.10] [0.64, 0.92]

43.60 128.18 217.56 429.25 866.56 1305.86

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

→ in Eq. (24). The variable x are the controller parameters, i.e. KP , TI and TD , and the 3-dimensional search space G∈ R 3 is pre-specified as G= [0,10]3 to widely span the optimization range of the controller design [10]. The fitness value of the stochastic population-based algorithms is the expectation of the stochastic objective function obtained from MCS:

values and variances, we can conclude that 200 MCS-based samples can guarantee a reasonable trade-off between the runtime and estimation accuracy. Each sample simulates a time frame of 30 s and requires around 4.33 s on a 64 bit Windows desktop with 32 GB memory and an Intel(R) Xeon(R) E5-1650 v3 @ 3.50 GHz CPU. 4.3. Stochastic performance of the PID-controlled AGC

N

E[Ji] =

The AGC performance of the integrated system depends on the discrete-time PID controller. Therefore, the PID controller is optimized to mitigate the communication network disturbances, and offer optimum AGC performance by reducing system frequency fluctuations. As a result, system reliability is also optimized. In previous works, the objective function for the optimization of the PID controller is an integral performance index over the total operating time T , which quantifies frequency and control signal deviations [10,49]

J=

∫0

T

[η1 (Δf (t ))2 + (1−η1) ∗η2 (Δu (t ))2] dt

j=1

J (→ x ) = [E[J1 (→ x )],E[J2 (→ x )]]T

Ji,j / N , i = 1,2 (25)

where Ji is defined in Eq. (24) and N = 200 samples. In this work, we use the PSO rather than other heuristic algorithms, → i.e. GA [5], to solve the optimization problem for x ∈ R 3, because PSO can avoid being trapped into local minima [9–11]. The particle movement in the PSO can be interpreted as a form of path re-linking via the exploitation of the local best-known positions, i.e. pbest [28]. In this sense, both the PSO and GA can be regarded as generating new solutions in the neighborhood of two parents, i.e. using crossover in the GA and using attractions to two pbest positions in the PSO. This multiparent effect is a key advantage over single-point techniques such as the simulated annealing and tabu search. As compared to GA, the PSO utilizes three available information sets to facilitate the search process. These are the local best-known positions (pbest), the global best-known position (gbest), and the current positions [9–11]. This allows greater diversity and exploration over a single population (which with elitism would only be a population of pbests). In addition, the momentum effects on particle movement allows fast convergence (e.g. when a particle is moving in the direction of a gradient) and increased variety/diversity in search trajectories. The performance of PSO has been shown superior to GA in parameters optimization of PID controller [66,67]. The MOPSO is preferred to the multi-objective GA because no evolution operators, i.e. the crossover

(23)

where η1 indicates the relative importance of the two terms and η2 is the normalizing constant to scale both terms in a uniform range and is set to 0.002. The first term is directly related to the reliability of the integrated system and the second term measures the disturbance rejection ability of the controller, which is the total control effort to be minimized. The total process time T is set to 30 s [49]. Motivated by [52–56], the objective function in Eq. (23) is separated into two objectives and multi-objective optimization is applied. In fact, the weighting factors in the objective function in Eq. (23) can be changed dynamically using multi-objective PSO algorithm:

Minimize:

∑

(24)

→ → → where J1 ( x ) = (Δf (t ))2 and J2 ( x ) = (Δu (t ))2 . J ( x ): R 3 → R is defined

Fig. 3. Flowchart of the methodology and the simulation studies.

93

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

And the solution with minimum MF2k is chosen as the best compromise between the two objectives and named “Best Tradeoff II”. The convergence of optimization problem solved by the single PSO has been investigated in details in works [10,47]. These works investigated similar optimization problem of PID controller, where both the objective value and the controller parameters become almost constant toward the end of 50 iterations (the number of population is 30). As for the convergence of optimization problem of PID controller solved by the MOPSO, works [52–56] showed fast convergence to the Pareto front in 50 iterations (the number of population is 50). The steps for applying the proposed framework on the designed system with physical and cyber layer are sketched in Fig. 3.

and mutation, are required, and the information of the current optimum particles effectively identifies candidate solutions throughout the problem space. The detailed introduction to PSO and MOPSO are provided in [9–11] and [52–56], respectively. The core part of the MOPSO algorithm is detailed in the following. Notation: A : solution set which provides optimal values of the PID controller parameters KP , KI and KD NP : number of particles x i and vi : current position and velocity of particle i MP : number of dimensions (MP = 3) I and NI : current and maximum number of iterations ω : inertia coefficient (ω = 0.5 × 0.99 I which tunes the impact of the previous history velocities on the current velocity) c1 and c2 : cognitive and social factors (which can accelerate the search towards the local and global best directions, and equal to c1 = 1 and c2 = 2) r1 and r2 : random numbers drawn from the uniform distribution [0,1]. Algorithm: STEP 1: Randomly initialize the position x i of the particle i (i = 1,2,…,NP ) sampling from a uniform distribution within the solution space, i.e. G= [0,10]3. The coordinates of the particle position represent the PID controller parameters, i.e. KP , KI and KD . Initialize vi , pbesti and gbesti to zero. STEP 2: Compute the fitness value for each particle based on Eq. (24). STEP 3: Update the values of pbest and gbest based on calculated fitness values; determine the velocity and position values for each particle in the next iteration using following equations:

vid: =ωvid + c1 r1 (Pid−x id ) + c2 r2 (Gid−x id )

(26)

x id : =x id + vid

(27)

5. Results and discussion This Section investigates the reliability of the integrated energycommunication system equipped with optimum PID controller to enhance the AGC performance and reduce system frequency fluctuations. We assess the impact of the two architectures, i.e. Ethernet and hybrid presented in Section 3.2, and of various configurations of the open communication network on the system reliability and on the AGC performance. 5.1. Specifications for the integrated system The physical system operates in nominal conditions, i.e. the stochastic wind speed, the variable sun irradiance and uncertain load, affecting, respectively, PWTG , PPV and PL , given in [5,10,32]. The coupled algebraic and ordinary differential equations for the DGS and the open communication network in Fig. 1, are numerically integrated using the Dormand-Prince method implemented in Matlab ode45 function with a fixed step size of 0.005 s. The parameters of the transfer functions for the DERs in Fig. 1 are provided in [5,10,32]. Additionally, the physical layer has the following specifications:

• The base value for the apparent power is 150 kW. • The rated apparent power of the WTG (Nordtank

where Pi = [pbesti1,pbesti2 ,…,pbestiMP Gi = [gbesti1,gbesti2,…,gbestiMP ]T and d = 1,2,…,MP . The current best solution is put into the set A . STEP 4: If the maximum number of iterations NI is reached, go to STEP 5; otherwise, go to STEP 2. STEP 5: Decode the particle with the minimum fitness value and choose the result as the optimally designed PID controller. As indicated by [52–56], the multi-objective optimization generates the Pareto optimal set of the non-dominated solutions from which the global optimum solution is selected based on the specific application. In this study, fuzzy set theory is applied to model the trade-off between the two objectives [52–56]. First, a linear membership function is defined → for each objective function Ji ( x ) :

]T ,

mfi =

Ji−Jimin , max Ji −Jimin

i = 1,2

•

• • •

(28)

According to Eq. (28), the minimum of the objective function Ji is associated to the minimum of the membership function, therefore it has maximum degree of achievement of the fuzzy objective. Then, for every non-dominated solution k , the aggregate membership function for the PID controller optimization is computed as:

MF 1k = mf1k + mf2k

Based on Remark 1, 0 ⩽ ΔPDEG ⩽ 2 pu , |ΔPBESS | ⩽ 0.33 pu and |ΔPFESS | ⩽ 0.17 pu , provide the largest rated output of the DEG, FESS and BESS [10]. The data sets of wind speed and solar radiation for Table 2 Parameters of frequency response model (physical layer in Fig. 1(b)).

(29)

The solution which minimizes Eq. (29) is regarded as the best compromise between the two objectives and named “Best Tradeoff I”. For comparison, the distance of each non-dominated solution k from the origin in the fuzzified coordinate is also computed [52–56]:

MF 2k = [(mf1k )2 + (mf2k )2]1/2

NTK 150) is 150 kW. The cut-in wind speed, the rated wind speed and the cutout wind speed are 4.0 m/s, 13.5 m/s and 25 m/s, respectively. The wind farm consists of two WTGs. The structure of a 750 kW PV generation has 465 parallel strings and consists of 7 series-connected modules (SunPower SPR-230E-WHTD). The rated power of each cell is 230 W under the nominal operation temperature 45 °C. The nominal efficiency is 18.5%. The length and width of each cell are 1.599 m and 0.798 m. The maximum power temperature coefficient is 3.37 × 10−5 . The rated power of the DEG (MQP300IV) is 300 kW. The rated power of the BESS (one Tesla Powerpacks) is 50 kW, whose energy capacity is 210 kWh. The rated power of the FESS (Amber Kinetics) is 25 kW, whose energy capacity is 40 kWh.

(30)

94

Parameter

Value

Parameter

Value

D (pu/Hz) M (pu/s) TBESS (s) TFESS (s)

0.03 0.4 0.1 0.1

TDEG (s) TWTG (s) TPV (s)

2 1.5 1.8

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

Fig. 5 shows the system frequency curves of the PI controller [57] and the proposed MOPSO-based PID controller in the presence of 20 ms and 390 ms, respectively. Consistently with the eigenvalue analysis shown in Fig. 4(b), the time delay causes instability in the PI controller. By using the MOPSO-based PID controller, the system stability greatly improves against delays of the feedback frequency signal and of the control signal. As compared with the curve without time delays, the MOPSO-based PID controller shows reasonable delays in the response time as expected.

Table 3 Parameters of communication network (cyber layer in Fig. 1(a)). Parameter

Value

Descriptions Value

BWShare

0∼1

Ti Ld

0∼ + ∞ 0∼1

indicates the expected percentage of bandwidth used by the interference user determines the activity level of the interference user expected packet loss probability

computing real-time output of RERs are provided by the Elia Grid, Belgium, with a data frequency 1 Hz. The details can be referred to Appendix A. The relevant system parameters [10,56] for a typical microgrid in the physical layer of Fig. 1(b) are presented in Table 2. The open communication network is implemented in Truetime simulator [47]. The detailed procedures are provided in the Appendix C. The data rate of the communication architectures is 800 Kbits/s, and the 802.11b/g is further characterized by transmission power 20 dbm, receiver signal threshold −48 dbm, ACK timeout 0.04 ms and the retry limit 5. The descriptions of simulation parameters Ti , BWShare and Ld is discussed in Remarks 2 and 3 and Eq. (13) and presented in Table 3. In order to simulate diverse operating conditions of the islanded microgrid, four exemplary cases of the DERs management are considered. The power generated by the RERs and the microgrid load is categorized into low and high, and the excess generation is used to charge ESS. In particular, these four different cases are:

5.3. Sensitivity analysis on the reliability indices for two network architectures For illustrating purpose, the maximum permissible instantaneous frequency deviation is set to ± 0.8 Hz according to [65]. The parameters of the discrete PID controller are tuned as KP = 0.52 , KI = 2.21 and KD = 1.00 [5,10]. Five combinations of network configurations, i.e. BWShare, Ti and Ld , are taken into account to investigate multiple communication scenarios in the Ethernet and in the hybrid network. For comparison purpose, the reliability and objective values of the integrated system with perfect communication, i.e. no time delays and no → packet dropouts, are R = 94.48% and E[J ( x )] = [373.37,2.82 × 10 4]. Fig. 6 presents the impact of the length of data traffic of the interference node on the reliability of the integrated system with the Ethernet ([0.2, 0.005, 5%]). The length of data traffic is from 20 to 200 bytes. As expected, the system reliability decreases as the length increases. It is in line with our common knowledge that a larger data packet consumes more bandwidth of the communication channel, which deteriorates network conditions and results in larger time delays and more packet dropouts. Fig. 7 illustrates the relationship between the reliability of the integrated system with Ethernet, and the communication configurations [Ld ,BWShare,Ti ]. Fig. 7(a) shows that the AGC becomes unstable and the system reliability degrades significantly if Ld ⩾ 40%, the controller does not have sufficient observations of the system frequency measurement, and is unable to compute correct control signals. The case (BWShare = 0.1 & Ti = 0.01) in Fig. 7(a), has the largest reliability due to light congestion. In this case, data collisions between the PMU and the interference node are fewer than for other cases with smaller Ti . In the case (BWShare = 0.2 & Ti = 0.005), the interference node sends out disturbing traffic every 0.005 s and consumes 20% of the network bandwidth, therefore data collisions between the PMU and the interference node increase. As a result, increased time delays degrade system reliability as compared to the other two cases of Fig. 7(a). Increasing the activity level of the interference user and the used bandwidth leads to increasing data collisions, and therefore network congestion, which causes longer time delays. Thus, system reliability decreases, i.e. comparing case (BWShare = 0.1 & Ld = 0.1) and case (BWShare = 0.2 & Ld = 0.1) in Fig. 7(c), and comparing case (Ti = 0.01 & Ld = 0.1) and case (Ti = 0.005 & Ld = 0.1) in the Fig. 7(b), respectively. On the other hand, if Ti ⩾ 0.01, the competition for sending data packet between the interference node and the PMU reduces, network conditions improve and thus the system reliability increases in Fig. 7(b) and (c).

Case 1: The power generated by RERs is low. The load is low and consumes most of the power generated by RERs. As a result, the DEG is OFF. Stochastic RERs caused by the dynamic weather conditions lead to unbalance between supply and demand, reflected by system frequency fluctuations. These can be mitigated by the ESS via the controller whose input is the system frequency derivation, and therefore system frequency is stabilized. Case 2: The power generated by RERs is low and the load is high. It indicates that RERs cannot supply sufficient power, and therefore complementary power is provided by the DEG and the discharging ESS. Case 3: The power generated by RERs is high and the load is low. RERs can supply sufficient power, and therefore the excess power is used to charge ESS and the DEG is OFF. Case 4: The power generated by RERs is high and the load is high. The combination of RERs and ESS is sufficient to supply the total load and the DEG is disconnected. For the above cases, the initial conditions of the DERs are listed in Table 4. These conditions match the realistic situations in Appendix A. The reference system frequency is set to 50 Hz.

5.2. Eigenvalue and time delay analysis According to the state-space model of the microgrid given in Appendix B, the eigenvalue analysis is conducted following the procedure detailed in [18,38,39]. Fig. 4 compares the effect of the time delay in the proposed MOPSO-based PID controller with the case of a PI controller [57], when packet dropout is neglected and the period is 0.01 s. Fig. 4(a) shows the MOPSO-based PID controller eigenvalue spectrum when the time delay is as large as 390 ms (maximum allowable time delay). As demonstrated, the system with MOPSO-PID controller remains stable in the presence of large time delays. The advantage of MOPSO-PID controller over the PI controller [57] is better represented by Fig. 4(b); for the latter, the instability can occur when the time delay is as small as 20 ms. As indicated by [38,39], typical time delays in reality can be in the order of 100–300 ms. Therefore, the MOPSO-PID controller is effective and robust in the reliable management of RERs in the microgrid to ensure stable system frequency.

Table 4 State of DERs under different microgrid operating conditions. Load RERs Low

High

95

Low OFF • DEG: discharging or charging • BESS: • FESS: discharging or charging OFF • DEG: charging • BESS: • FESS: charging

High ON • DEG: discharging • BESS: • FESS: discharging OFF • DEG: discharging or charging • BESS: • FESS: discharging or charging

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

Fig. 4. Eigenvalue spectrum of the microgrid in the presence of time delays with (a) the proposed MOPSO-based PID controller when the time delay increases from 0 to 390 ms, and (b) the PI controller when the time delay increases from 0 to 24 ms.

Fig. 7. System reliability as a function of the variable of the network configuration (a) Ld , (b) BWShare and (c) Ti .

Table 5 provides the estimated reliability of the integrated system with Ethernet and with hybrid network. Such values of communication parameters are selected because they can generate 10 ms to 2 s time delays, and losses of 1–10% of the overall packet stream, which can occur at realistic open communication networks [25,47]. A negative x )] can be inferred correlation between system reliability and E[J (→ from Table 5. This is expected because when the system reliability is high, most of the system frequency measurements are within the tolerance interval, which thereby leads to a small value of the objective x )]. Table 5 also shows that the Ethernet architecture function E[J1 (→ ensures higher reliable to the integrated system as compared to the hybrid architecture. This is expected because the data exchange between the AP and the RTU in the hybrid architecture is provided by 802.11 b/g, which introduces additional time delays and packet dropouts, and results in lower system reliability.

Fig. 5. System frequency responses of (a) the PI controller with 20 ms time delay, (b) the MOPSO-based PID controller with 390 ms time delay, and (c) the MOPSO-based PID controller without time delay.

5.4. Optimal PID controller for the AGC of DERs The MOPSO is employed to design the optimal PID controller. In the MOPSO, the number of particles and the number of iterations are set to 50 and 50. For each particle, the stochastic objective function J (→ x ) = [E[J1 (→ x )],E[J2 (→ x )]]T is estimated from 200 MCS samples. Table 6 lists the optimal control parameters [KP , KI , KD ] for the

Fig. 6. System reliability as a function of the length of data traffic.

96

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

Table 5 Reliability of the integrated system for different network configurations [BWShare , Ti , Ld ] of the two architectures. Configurations

R (Ethernet)

E[J (→ x )] (Ethernet)

[0.2, 0.005, 5%] [0.4, 0.005, 5%] [0.05, 0.005, 5%] [0.2, 0.0025, 5%] [0.2, 0.02, 5%] [0.2, 0.005, 20%] [0.2, 0.005, 1%]

92.89% 90.43% 93.58% 92.62% 93.08% 90.60% 93.09%

[465.51, [593.96, [435.91, [473.08, [433.92, [574.40, [432.49,

4.55 × 104] 5.80 × 104] 3.39 × 104] 4.40 × 104] 4.93 × 104] 6.27 × 104] 3.55 × 104]

R (Hybrid)

E[J (→ x )] (Hybrid)

92.65% 90.23% 93.43% 92.35% 93.05% 90.09% 92.97%

[471.63, [559.57, [429.05, [510.06, [436.42, [623.37, [433.17,

5.01 × 104] 4.33 × 104] 5.25 × 104] 3.07 × 104] 5.00 × 104] 4.50 × 104] 4.89 × 104]

Table 6 Optimal PID controllers [KP , KI , KD ] for different network configurations. Network

Tradeoff solution

Reliability

E[J1 (→ x )]

E[J2 (→ x )]

Optimal PID

Ethernet [0.2, 0.005, 5%]

Best tradeoff I Best tradeoff II

97.27% 97.27% 97.78%

186.30 186.30 143.11

402.85 402.85 8.34 × 104

[1.12, 0.094, 0.35] [1.12, 0.094, 0.35] [7.6, 0, 0.51]

12.26%

8.25 × 104

0

[0, 0, 0]

97.10% 97.10% 97.60%

197.82 197.82 176.06

435.37 435.37 9.84 × 103

[1.07, 0.105, 0.41] [1.07, 0.105, 0.41] [5, 0.34, 2.02]

12.40%

8.04 × 104

0

[0, 0, 0]

97.79% 97.79% 97.99%

160.46 160.46 140.19

559.30 559.30 2.19 × 103

[1.45, 0.12, 0.4] [1.45, 0.12, 0.4] [3.8, 0.04, 0.45]

12.13%

8.23 × 104

0

[0, 0, 0]

→ Minimum E[J1 ( x )] → Minimum E[J2 ( x )] Hybrid [0.2, 0.005, 5%]

Best tradeoff I Best tradeoff II

→ Minimum E[J1 ( x )] → Minimum E[J2 ( x )] Perfect communication

Best tradeoff I Best tradeoff II

→ Minimum E[J1 ( x )] → Minimum E[J2 ( x )]

efficient than single-objective PSO in finding good combinations with higher reliability index [42]. The Pareto optimal set and best tradeoff solutions for the optimal PID controller under perfect communication are shown in Fig. 8. The best compromises with respect to system frequency deviations and control efforts, i.e. the Best Tradeoff I and Best Tradeoff II, coincide. Results indicate that the reduction in the system frequency deviations leads to the increase in the control efforts, and vice versa. The Pareto front gathers to a very small area toward the maximum iterations. 5.5. Control performance under different microgrid operating conditions In this Section, we show particular realizations of the real-time output of the DERs and of the performance of the PID controller indicated by the system frequency. The four representative operating conditions in Table 4 are applied at t = 0 s, and the transient response and the steady-state are investigated. At time t = 25 s, a load step of 0.1 pu is added to the initial load to test system stability. The three network architectures are equipped with the Best Tradeoff I PID controller optimized for the communication configuration [0.2, 0.005, 5%], given in Table 6.

Fig. 8. Pareto optimal sets and tradeoff of the optimization objectives for the optimal PID controller and perfect communication network.

integrated system with three network architectures, which minimize J (→ x ) under the configuration [0.2, 0.005, 5%]. The reason behind selecting this specific configuration is that the generated time delays match typical time delays in most network conditions, which can be in the order of 100–300 ms [27]. Table 6 quantifies the amount of system reliability and control effectiveness that is lost due to communication degradation with respect to the perfect communication case. These results also show that the two objectives are non-commensurable and conflicting with each other. Higher system reliability means more control efforts and vice versa. Therefore, the MOPSO algorithm is more suitable for optimizing the PID controller, because it considers the two objectives separately. Furthermore, the MOPSO algorithm is more

5.5.1. Case 1: Low RERs and low load operations Fig. 9 illustrates the real-time output of the DERs and system frequency for three communication architectures, i.e. the perfect communication network, Ethernet and hybrid network, when the microgrid is characterized by low load and low RERs (Case 1 in Table 4). In this case, the load slightly exceeds the output of RERs. For the optimum use of energy, the DEG is not operated (Fig. 9(c)), and the deficit power in the microgrid is supplied by the BESS and FESS (Fig. 9(a) and (b)).

97

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

Fig. 9. Real-time output of (a) FESS, (b) BESS, (c) DEG, and (d) system frequency Δf under normal operation and step load change at t = 25 s, given low RERs and low load (Case 1 in Table 4).

Fig. 10. Real-time output of (a) FESS, (b) BESS, (c) DEG, and (d) system frequency Δf under normal operation and step load change at t = 25 s, given low RERs and high load (Case 2 in Table 4).

98

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

in component response and missing control signals ultimately cause the FESS, BESS and DEG to perform untimely and inaccurately, causing large frequency deviations in Fig. 10(d). Due to the step load change at t = 25 s, the output of BESS and DEG is increased by 0.04 pu and 0.06 pu respectively, as illustrated in Fig. 10(b) and (c). As the hybrid network is affected by large time delays and many packet dropouts, it results in the slowest convergence rate to the steady-state frequency (Δf = 0 ), as compared to other two cases in Fig. 10(d).

Due to the load step at t = 25 s, the output of FESS and BESS is increased by 0.05 pu, illustrated by Fig. 9(a) and (b). Fig. 9(d) shows that the PID controller in the perfect communication case provides the fastest response against the sudden load increase, and achieves minimum system frequency deviations. For the integrated system with Ethernet and hybrid network, the FESS, BESS and DEG output and Δf converge to the same values as the AGC with perfect communication, with small convergence rates due to time delays and packet dropouts. The strong effect of communication degradation introduced by the 802.11b/g in the hybrid network yields the smallest convergence rate. Furthermore, Fig. 9(a), (b) and (d) show that the tradeoff PID controllers eliminate the oscillations in the output of FESS and BESS and system frequency, caused by random communication degradation. Finally, the designed PID controller can cope with the spike in the FESS and BESS outputs occurring at t = 43 s caused by communication degradation (Fig. 9(a) and (b)), and results in minor frequency fluctuations in Fig. 9(d).

5.5.3. Case 3: High RERs and low load operations In this case, the load is smaller than the output of RERs, the surplus power charges the BESS and FESS as shown in Fig. 11(a) and (b), and the DEG is not operational (Fig. 11(c)). Due to the step load change at t = 25 s the charging rates of FESS and BESS are reduced by 0.048 pu and 0.052 pu, respectively. Fig. 11(d) indicates that the microgrid with Ethernet and hybrid network is under serious congestions in the early stage, and therefore large system frequency deviations and slow convergence rate can be observed. However, these conditions cease after t = 35 s and the performances of the three architectures coincide.

5.5.2. Case 2: Low RERs and high load operations In this case, the load exceeds the output of RERs and the deficit power cannot be only supplied by the discharging BESS and FESS. The DEG is operated to provide complementary power as shown by Fig. 10(c). Fig. 10(a) highlights the impact of time delays in the response of the controllable FESS, for the integrated systems with imperfect communication. The output of FESS in the Ethernet and the hybrid architecture stays constant in the interval [2.40 s, 2.80 s], as compared to the perfect communication case. Therefore, the FESS keeps releasing the same power even in unbalanced conditions because the data packets containing the control signals cannot reach their destination. At the same time, the control center lacks enough frequency measurements from the PMU, and issues inaccurate commands. Delays

5.5.4. Case 4: High RERs and high load operations In this case, both the load and the output of RERs are large. The DEG is off-line to make most use of RERs, as shown in Fig. 12(c). In normal operations, the FESS and BESS are charging as shown by Fig. 12(a) and (b), because the output of RERs slightly exceeds the load. In this random realization, the communication in the hybrid network is outperformed by the Ethernet in the time frame [5 s, 11 s], and therefore the frequency response of the hybrid network is slower (Fig. 12(d)).

Fig. 11. Evolution of (a) FESS, (b) BESS, (c) DEG, and (d) system frequency Δf under normal operation and step load change at t = 25 s, given high RERs and low load (Case 3 in Table 4).

99

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

Fig. 12. Evolution of (a) FESS, (b) BESS, (c) DEG, and (d) system frequency Δf under normal operation and step load change at t = 25 s, given high RERs and high load (Case 4 in Table 4).

Fig. 13. Evolution of the system frequency deviation under normal operations and step load change at t = 25 s for different levels of communication degradation (a) [0.2, 0.005, 5%], (b) [0.4, 0.005, 5%], (c) [0.2, 0.002, 5%], and (d) [0.2, 0.005, 30%].

100

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

(Ethernet and 802.11b/g), and random time delay and Bernoulli distributed packet dropout are modelled via the Truetime simulator. The reliability of the integrated system is estimated in the face of network degradation effects by changing the network configurations to generate different scenarios of time delays and packet dropouts. The results show that the activity level of interference traffic and the expected percentage of bandwidth used by the interference user have a significant impact on system reliability. Low system reliability is observed for large BWShare and small Ti . The AGC based on a discretetime PID controller is developed to suppressed the frequency oscillations in the integrated system and enhance reliability. Parameters of the PID controller are tuned by the PSO algorithm to minimize both the frequency deviation and control signal deviation. The optimization results demonstrate that the PSO-based PID controller is capable of reducing system frequency fluctuations and improve AGC performance. The PSO algorithm achieves better results in terms of the quality of the optimal solution and convergence rate, if the effects of communication degradation are small. Furthermore, communication degradation delays the component response to the control signal and cause their output to remain unchanged even in the presence of power imbalance until an updated packet is received. Finally, the AGC with Ethernet shows the largest reliability and robustness against unknown communication configurations, and results in small frequency oscillations. This framework extends the constant and uniformly distributed time-delay modelling of the cyber layer by representing realistic open communication networks and packet dropout using the TrueTime simulator. The proposed model simulates the specific network protocol and, thus quantifies time delays and packet dropouts without assuming ad-hoc delay distributions. Furthermore, the two objective functions, i.e. system frequency deviation and control signal derivation, are noncommensurable and conflicting, and therefore we introduce the MOPSO algorithm to solve this multi-objective optimization problem. Future work will focus on the time-delay compensation by designing a real-time Smith predictor based on the estimated communication delay. As a future work, current studies limited to the primary and secondary control levels can be extended to consider the power management level by designing robust control strategies. Additionally, the effect of packet dropout can be mitigated by reconstructing missing the data set to form a complete one. Finally, analytical approaches to design a robust controller can be developed for the investigated stochastic integrated systems.

Due to the load step at t = 25 s, the FESS and BESS transition from the charging to the discharging state. After t = 33 s the conditions of the Ethernet and hybrid network improve, and the system frequency deviations and convergence rate are close to the ones of the perfect communication case as illustrated in Fig. 12(d). 5.6. Robustness of the MOPSO-based PID controller Finally, we investigate the robustness of the PID-controlled AGC operations optimized for the configuration [0.2, 0.005, 5%] in Table 6, subjected to increasing communication degradation. Fig. 13 demonstrates the system frequency deviation Δf in the microgrids with three different network architectures of one MC simulation. Four configurations are investigated, respectively, BWShare increased from 0.2 to 0.4, Ti decreased from 0.005 s to 0.002 s, and Ld increased from 5% to 30%. Similar to Sections 5.5.1–5.5.4, the microgrid is influenced by a sudden load stepping increase 0.1 pu at t = 25 s. Fig. 13 unveils the abilities of various controllers in providing satisfying AGC performance for unknown network configurations. The optimized controllers for Ethernet and hybrid network show robustness to degraded communication even though small frequency fluctuations appear. The reliability of the four Ethernet configurations is, respectively, 97.27%, 97.21%, 97.10% and 96.25%. The reliability of the four hybrid network configurations is, respectively, 97.10%, 97.05%, 96.81% and 95.96%. The analysis of the PID tuning indicates that the MOPSO produces different optimal PID controllers for different runs of the optimization algorithm [52–56]. Nonetheless, even if the PID controllers tuned using heuristics result in x )], they are capable of stabilizing the system different R and E[J (→ frequency quickly in the analyzed network configurations [3,5–7,9,10]. 6. Conclusion This work proposes a system-of-systems framework, in which two dedicated systems, i.e. the microgrid and open communication network, are integrated and pool their capabilities to create a complex system with extended functionality. In particular, this framework exemplifies a cyber-physical system, and allows investigating the operations of integrated DER systems and open communication networks, and the design of optimal AGC strategies in the face of communication degradation. The open communication network is adopted to enable the data exchange because of reduced wiring, easy installation, simple maintenance, flexible access and low operation costs. Two communication architectures are studied, i.e. Ethernet and hybrid network Appendix A

The approximated models for the wind turbine and PV are detailed in [5,10,42–45]:

NWT Pr ,WTG × ⎧ ⎪ PW = NWT Pr ,WTG, ⎨ ⎪ 0, ⎩

vW − vcutin , vr − vcutin

vW ⩽ vcutin or vW ⩾ vr vr ⩽ vW ⩽ vcutout otherwise

(31)

Psol = ηS Φ{1 + kpv (Ta−Tr )}

(32)

where vW , vcutin , vr and vcutout are the real-time wind speed, cut-in wind speed, rated wind speed and cut-out wind speed. NWT is the number of wind turbines in the wind farm and Pr ,WTG is the rated power of a wind turbine. η is the conversion efficiency of the PV cells, S is the measured area of the PV array, kpv is the maximum power temperature coefficient, Tr is the nominal operation temperature of PV cells, and Ta = 25 °C is the ambient temperature. Fig. 14(a) shows the dataset of wind speed and solar irradiance at the Elia Grid, Belgium. Based on the models in Eqs. (31) and (32), we estimate the total power output of wind farm and PV farm, i.e. RERs, in Fig. 14(b). When compared with the real-time load, we can see that the values of RERs output and load are consistent with the four cases described in Table 4.

101

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

Fig. 14. Real-time curves of (a) wind speed and solar irradiance and (b) comparison between the total output power of RERs and the load.

Appendix B The state-space model of the microgrid in the physical layer of Fig. 1, is given as following:

⎡− 1/ TWTG ⎢0 ⎢0 ẋ = ⎢ ⎢0 ⎢0 ⎢ ⎣0

0 − 1/ TPV 0 0 0 0

0 0 − 1/ TDEG 0 0 0

0 0 0 − 1/ TBESS 0 0

0 0 0 0 − 1/ TFESS 0

0 ⎤ ⎡ ΔPWTG ⎤ ⎡1/ TWTG ⎥ ⎢ ΔPPV ⎥ ⎢ 0 0 ⎥ ⎢ ΔP ⎥ ⎢ 0 ⎥ ⎢ DEG ⎥ + ⎢ 0 P Δ 0 ⎥ ⎢ BESS ⎥ ⎢ 0 ⎥ ⎢ ΔPFESS ⎥ ⎢ 0 0 ⎥⎢ ⎥ ⎢0 − 2D / M ⎦ ⎣ Δf ⎦ ⎣

y = [000001] x

0 1/ TPV 0 0 0 0

0 ⎤ ⎤ ⎡0 ⎥ ⎢0 0 ⎥ ΔPW ⎡ ⎤ 0 ⎥ ⎢ ΔP ⎥ + ⎢1/ TDEG ⎥ u sol ⎥ ⎢ ⎥ 1/ T 0 ⎥⎢ ⎢ BESS ⎥ ΔPL ⎥ ⎣ ⎦ 0 ⎥ ⎢1/ TFESS ⎥ ⎥ ⎢ 2/ M ⎥ ⎦ ⎦ ⎣0

(33) (34)

Appendix C To implement the microgrid with two different communication network architectures in the Matlab/Simulink environment, the Truetime simulator is deployed for simulating different types of communication protocols [68]. The procedure is detailed in the “TrueTime 2.0-Reference Manual”, which provides representative examples of cyber-physical systems [47]. Fig. 15(a) and (b) shows the Matlab/Simulink implementations of 102

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

Fig. 15. Simulink realization of the microgrid with (a) Ethernet and (b) Hybrid network.

103

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

the microgrid with Ethernet and of the microgrid with Hybrid network, respectively, derived from the system schematics presented in Fig. 1. The network blocks from the Truetime simulator are directly linked with physical components, i.e. the PID controller, PMU, DERs, to enable data exchanges

[28] Ghoshal SP. Optimizations of PID gains by particle swarm optimizations in fuzzy based automatic generation control. Electr Pow Syst Res 2004;72(3):203–12. [29] Baghaee HR, Mirsalim M, Gharehpetian GB. Performance improvement of multiDER microgrid for small and large-signal disturbances and nonlinear loads: novel complementary control loop and fuzzy controller in a hierarchical droop-based control scheme. IEEE Syst J 2016;99(1):1–8. [30] Baghaee HR, Mirsalim M, Gharehpetian GB. Real-time verification of new controller to improve small/large-signal stability and fault ride-through capability of multiDER microgrids. IET Gen Trans Dist 2016;10(12):3068–84. [31] Baghaee HR, Mirsalim M, Gharehpetian GB, Talebi HA. A decentralized power management and sliding mode control strategy for hybrid ac/dc microgrids including renewable energy resources. IEEE Trans Indust Inf 2017;PP(99):1–10. [32] Guerrero JM, Loh PC, Lee TL. Advanced control architectures for intelligent microgrids-Part II: power quality, energy storage, and AC/DC microgrids. IEEE Trans Indus Electron 2013;60:1263–70. [33] Etemadi AH, Davison EJ, Iravani R. A generalized decentralized robust control of islanded microgrids. IEEE Trans Power Syst 2014;29:3102–13. [34] Maryam B, Houshang K. A robust two-degree-of-freedom control strategy for an islanded microgrid. IEEE Trans Power Del 2013;28(3):1339–47. [35] Baghaee HR, Mirsalim M, Gharehpetian GB. Power calculation using RBF neural networks to improve power sharing of hierarchical control scheme in multi-DER microgrids. IEEE J Emerg Sel Top Power Electron 2016;4(4):1217–25. [36] Baghaee HR, Mirsalim M, Gharehpetian GB, Talebi HA. Nonlinear load sharing and voltage compensation of microgrids based on harmonic power-flow calculations using radial basis function neural networks. IEEE Syst J 2017;(99):1–11. [37] Baghaee HR, Mirsalim M, Gharehpetian GB, Talebi HA. Three phase AC/DC powerflow for balanced/unbalanced microgrids including wind/solar, droop controlled and electronically-coupled distributed energy resources using RBF neural networks. IET Power Electron 2017;10(3):313–28. [38] Baghaee HR, Mirsalim M, Gharehpetian GB, Talebi HA. Eigenvalue, robustness and time delay analysis of hierarchical control scheme in multi-der microgrid to enhance small/large-signal stability using complementary loop and fuzzy logic controller. J Circ, Syst Comp 2017;26(6):1–34. [39] Kahrobaeian A, Ibrahim Mohamed YA-R. Networked-based hybrid distributed power sharing and control for islanded microgrid systems. IEEE Trans Power Electron 2015;30(2):603–17. [40] Lai J, Zhou H, Lu X, Yu X, Hu W. Droop-based distributed cooperative control for microgrids with time-varying delays. IEEE Trans Smart Grid 2016;7:1775–89. [41] Dörfler F, Jovanović MR, Chertkov M, Bullo F. Sparsity-promoting optimal widearea control of power networks. IEEE Trans Power Syst 2014;29(5):2281–91. [42] Baghaee HR, Mirsalim M, Gharehpetian GB, Talebi HA. Reliability/cost-based multi-objective Pareto optimal design of stand-alone wind/PV/FC generation microgrid system. Energy 2016;115:1022–41. [43] Baghaee HR, Mirsalim M, Gharehpetian GB, Talebi HA. Application of RBF neural networks and unscented transformation in probabilistic power-flow of microgrids including correlated wind/PV units and plug-in hybrid electric vehicles. Simul Model Pract Theory 2017;72(C):51–68. [44] Kaviani AK, Baghaee HR, Riahy GH. Optimal sizing of a stand-alone wind/photovoltaic generation unit using particle swarm optimization. J Simul 2009;85:89–99. [45] Baghaee HR, Mirsalim M, Gharehpetian GB. Fuzzy unscented transform for uncertainty quantification of correlated Wind/PV microgrids: possibilistic-probabilistic power flow based on RBFNNs. IET Renew Power Gen 2017;11(6):867–77. [46] Galbusera L, Theodoridis G, Giannopoulos G. Intelligent energy systems: introducing power-ICT interdependency in modeling and control design. IEEE Trans Ind Electron 2015;62(4):2468–77. [47] Cervin A, Henriksson D, Ohlin M. TrueTime 2.0-reference manual < http://www. control.lth.se/truetime/ > . [48] Minero P, Franceschetti M, Dey S, Nair GN. Data rate theorem for stabilization over time-varying feedback channels. IEEE Trans Autom Control 2009;54(2):243–55. [49] Pan I, Das S. Fractional order fuzzy control of hybrid power system with renewable generation using chaotic PSO. ISA Trans 2016;62:19–29. [50] Baghaee HR, Mirsalim M, Gharehpetian GB, Talebi HA, Niknam-Kumle A. A hybrid ANFIS/ABC-based online selective harmonic elimination switching pattern for cascaded multi-level inverters of microgrids. IEEE Trans Ind Elect 2017;(99):1–10. [51] Sahu RK, Panda S, Sekhar GTC. A novel hybrid PSO-PS optimized fuzzy PI controller for AGC in multi area interconnected power systems. Int J Electr Power Energy Syst 2015;64:880–93. [52] Pan I, Das S. Frequency domain design of fractional order PID controller for AVR system using chaotic multi-objective optimization. Int J Electr Power Energy Syst 2013;51:106–18. [53] Zhao SZ, Iruthayarajan MW, Baskar S, Suganthan P. Multi-objective robust PID controller tuning using two lbests multi-objective particle swarm optimization. Inf Sci 2011;181(16):3323–35. [54] Baghaee HR, Mirsalim M, Gharehpetian GB, Kashefi Kaviani A. Security/cost based optimal allocation of multi-type FACTS devices using multi-objective particle swarm Optimization. Simul Int Trans Soc Model Simul 2012;88(8):999–1010. [55] Baghaee HR, Mirsalim M, Gharehpetian GB. Multi-objective optimal power management and sizing of a reliable wind/PV microgrid with hydrogen energy storage using MOPSO. J Intell Fuzzy Syst 2017;32(3):1753–73.

References [1] Driesen J, Katiraei K. Design for distributed energy resources. IEEE Power Energy Mag 2008;6(3):30–40. [2] Basak P, Chowdhury S, Halder nee Dey S, Chowdhury SP. A literature review on integration of distributed energy resources in the perspective of control, protection and stability of microgrid. Renew Sust Energy Rev 2010;16(8):5545–56. [3] Kumar LVS, Kumar GV, Madichetty S. Pattern search algorithm based automatic online parameter estimation for AGC with effects of wind power. Int J Electr Power Energy Syst 2017;84:135–42. [4] Shotorbani AM, Zadeh SG, Mohammadi-Ivatloo B, Hosseini SH. A distributed nonLipschitz control framework for self-organizing microgrids with uncooperative and renewable generations. Int J Electr Power Energy Syst 2017;90:267–79. [5] Das DC, Roy AK, Sinha N. GA based frequency controller for solar thermal–diesel–wind hybrid energy generation/energy storage system. Int J Electr Power Energy Syst 2012;43(1):262–79. [6] Dash P, Saikia LC, Sinha N. Flower pollination algorithm optimized PI-PD cascade controller in automatic generation control of a multi-area power system. Int J Electr Power Energy Syst 2016;82:19–28. [7] Shiva CK, Mukherjee V. Design and analysis of multi-source multi-area deregulated power system for automatic generation control using quasi-oppositional harmony search algorithm. Int J Electr Power Energy Syst 2016;80:382–95. [8] Lee DJ, Wang L. Small-signal stability analysis of an autonomous hybrid renewable energy power generation/energy storage system, part I: time-domain simulations. IEEE Trans Energy Convers 2008;23(1):311–20. [9] Khadanga RK, Satapath JK. Time delay approach for PSS and SSSC based coordinated controller design using hybrid PSO–GSA algorithm. Int J Electr Power Energy Syst 2015;71:262–73. [10] Pan I, Das S. Fractional order AGC for distributed energy resources using robust optimization. IEEE Trans Smart Grid 2016;7(5):2175–86. [11] Liu SC, Liu P, Saddik AE. Modeling and stability analysis of automatic generation control over cognitive radio networks in smart grids. IEEE Trans Syst, Man, Cybern, Syst 2015;45(2):223–34. [12] Borghetti A, Bottura R, Barbiroli M. Synchrophasors-based distributed secondary voltage/var control via cellular network. IEEE Trans Smart Grid 2017;8(1):262–74. [13] Timbus A, Larsson M, Yuen C. Active management of distributed energy resources using standardized communications and modern information technologies. IEEE Trans Ind Electron 2009;56(10):4029–37. [14] Rakhshani E, Remon D, Rodriguez P. Effects of PLL and frequency measurements on LFC problem in multi-area HVDC interconnected systems. Int J Electr Power Energy Syst 2016;81:140–52. [15] Kuzlu M, Pipattanasomporn M, Rahman S. Communication network requirements for major smart grid applications in HAN, NAN and WAN. Comput Netw 2014;67:74–88. [16] Han GZ, Xu BY, Fan KJ, Lv GX. An open communication architecture for distribution automation based on IEC 61850. Int J Electr Power Energy Syst 2014;54:315–24. [17] Ahmadi A, Aldeen M. Robust overlapping load frequency output feedback control of multi-area interconnected power systems. Int J Electr Power Energy Syst 2017;89:156–72. [18] Baghaee HR, Mirsalim M, Gharehpetian GB, Talebi HAA. Generalized descriptorsystem robust H∞ control of autonomous microgrids to improve small and large signal stability considering communication delays and load nonlinearities. Int J Electr Power Energy Syst 2017;92:63–82. [19] Abdelaziz AY, Ali ES. Cuckoo Search algorithm based load frequency controller design for nonlinear interconnected power system. Int J Electr Power Energy Syst 2015;73:632–43. [20] Peng C, Han QL. On designing a novel self-triggered sampling scheme for networked control systems with data losses and communication delays. IEEE Trans Ind Informat 2016;63(2):1239–48. [21] Mo HD, Li YF, Zio E. A system-of-systems framework for the reliability analysis of distributed generation systems accounting for the impact of degraded communication networks. Appl Energy 2016;183:805–22. [22] Gungor VC, Sahin D, Kocak T, Ergut S, Buccella C, Cecati C, et al. A survey on smart grid potential applications and communication requirements. IEEE Trans Ind Informat 2013;9(1):28–42. [23] Lu X, Yu X, Lai J, Guerrero J, Zhou H. Distributed secondary voltage and frequency control for islanded microgrids with uncertain communication links. IEEE Trans Ind Informat 2017;13(2):448–60. [24] Wang WT, Lu Z. Cyber security in the Smart Grid: Survey and challenges. Comput Netw 2013;57(5):1344–71. [25] Grenier M, Navet N. Fine-tuning MAC-level protocols for optimized real-time QoS. IEEE Trans Ind Informat 2008;4(1):6–15. [26] Padilla E, Agbossou K, Cardenas A. Towards smart integration of distributed energy resources using distributed network protocol over Ethernet. IEEE Trans Smart Grid 2014;5(4):1686–95. [27] Zurawski R. Industrial communication technology handbook. 2nd ed. ISA Group, CRC Press; 2015.

104

Electrical Power and Energy Systems 96 (2018) 86–105

H. Mo, G. Sansavini

[62] IEEE standard for synchrophasors for power systems. IEEE Std 1344-1995 (R2001); 2001. [63] Tan W. Unified tuning of PID load frequency controller for power systems via IMC. IEEE Trans Power Syst 2010;25(1):341–50. [64] Reciprocating internal combustion engine driven alternating current generating sets. ISO 8528–5, Jul. 2005. [65] P1 – Policy 1: Load-frequency control and performance [C] < https://www.entsoe. eu/fileadmin/user_upload/_library/publications/entsoe/Operation_Handbook/ Policy_1_final.pdf > . [66] Nagaraj B, Murugananth N. A comparative study of PID controller tuning using GA, EP, PSO and ACO. In: 2010 IEEE international conference on communication control and computing technologies (ICCCCT). IEEE; 2010. p. 305–13. [67] Kumar A, Gupta R. Compare the results of tuning of PID controller by using PSO and GA technique for AVR system. Int J Adv Res Comput Eng Technol 2013;2:2131–8. [68] http://www.control.lth.se/truetime/.

[56] Hassan B, Mohammad RF, Sirwan S. Robust frequency control in an islanded microgrid: h∞ and μ-synthesis approaches. IEEE Trans Smart Grid 2016;7(2):706–17. [57] Ray PK, Mohanty SR, Kishor N. Proportional-integral controller based small-signal analysis of hybrid distributed generation systems. Energy Convers Manage 2011;52(4):1943–54. [58] Sharma G, Nasiruddin I, Niazi KR, Bansal RC. Optimal AGC of a multi-area power system with parallel AC/DC tie lines using output vector feedback control strategy. Int J Electr Power Energy Syst 2016;81:22–31. [59] Hota PK, Mohanty B. Automatic generation control of multi source power generation under deregulated environment. Int J Electr Power Energy Syst 2016;75:205–14. [60] Liang Y, Chen TW, Pan Q. Optimal linear state estimator with multiple packet dropout. IEEE Trans Autom Control 2010;55(6):1428–33. [61] Gharavi H, Hu B. Scalable synchrophasors communication network design and implementation for real-time distributed generation grid. IEEE Trans Smart Grid 2015;6(5):2539–50.

105