BBO-based small autonomous hybrid power system optimization incorporating wind speed and solar radiation forecasting

Renewable and Sustainable Energy Reviews 41 (2015) 1366–1375

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews journal homepage: www.elsevier.com/locate/rser

BBO-based small autonomous hybrid power system optimization incorporating wind speed and solar radiation forecasting R.A. Gupta a, Rajesh Kumar a, Ajay Kumar Bansal b a b

Department of Electrical Engineering, Malaviya National Institute of Technology, Jaipur Department of Electrical Engineering, Poornima Group of Colleges, Jaipur

art ic l e i nf o

a b s t r a c t

Article history: Received 6 November 2013 Received in revised form 25 August 2014 Accepted 17 September 2014

Rising carbon emission or carbon footprint imposes grave concern over the earth's climatic condition, as it results in increasing average global temperature. Renewable energy sources seem to be the favorable solution in this regard. It can reduce the overall energy consumption rate globally. However, the renewable sources are intermittent in nature with very high initial installation price. Off-grid Small Autonomous Hybrid Power Systems (SAHPS) are good alternative for generating electricity locally in remote areas, where the transmission and distribution of electrical energy generated from conventional sources are otherwise complex, difficult and costly. In optimizing SAHPS, weather data over past several years are generally the main input, which include wind speed and solar radiation. The weather resources used in this optimization process have unsystematic variations based on the atmospheric and seasonal phenomenon and it also varies from year to year. While using past data in the analysis of SAHPS performance, it was assumed that the same pattern will be followed in the next year, which in reality is very unlikely to happen. In this paper, we use BBO optimization algorithm for SAHPS optimal component sizing by minimizing the cost of energy. We have also analysed the effect of using forecast weather data instead of past data on the SAHPS performance. ANNs, which are trained with back-propagation training algorithm, are used for wind speed and solar radiation forecasting. A case study was used for demonstrating the performance of BBO optimization algorithm along with forecasting effects. The simulation results clearly showed the advantages of utilizing wind speed and solar radiation forecasting in a SAHPS optimization problem. & 2014 Elsevier Ltd. All rights reserved.

Keywords: Wind speed forecasting Solar radiation forecasting Small autonomous hybrid power system Optimization Wind energy conversion system Solar PV system

Contents 1. 2.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Small autonomous hybrid power system (SAHPS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Wind turbine generators (WTG) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2. Photovoltaic generation (PV). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3. Storage batteries (SB) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. Optimization problem of SAHPS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1. Cost of SAHPS power sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2. Cost of other parameters in SAHPS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3. Constraints in the optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. Forecasting in hybrid energy system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1. Artificial neural network model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2. Wind speed forecasting model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3. Solar radiation forecasting model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4. Testing the performance of the models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. Biogeography-based optimization (BBO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. The BBO algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7. Evaluation of the effect of forecasted data on optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

http://dx.doi.org/10.1016/j.rser.2014.09.017 1364-0321/& 2014 Elsevier Ltd. All rights reserved.

1367 1368 1368 1368 1368 1368 1368 1368 1369 1369 1369 1369 1370 1370 1370 1371 1371 1373 1374

R.A. Gupta et al. / Renewable and Sustainable Energy Reviews 41 (2015) 1366–1375

1. Introduction Electrical energy, one of the essential requirements of the modern world, has significant contribution towards the overall development of a nation. The electrical power generation system based on natural resources such as oil, coal and natural gases is unsustainable as these natural resources depend on the finite fossil fuel reserves, which are depleting gradually. If the present trend of exploitation of natural resources continues, the worldwide carbon emission level will rise to eight to twelve Giga tonnes by 2030, 62 percent higher as compared to the year 2002. With the increasing demand for electricity along with significant depletion in natural resources and several environmental issues, there is a need for advancement in renewable energy sources. In this regard, research and development (R&D) in the area of renewable energy technology can offer solutions in above all areas. Several optimization algorithms have been developed [1] for SAHPS optimization using past years meteorological data of solar radiation, wind speed, etc. Protogeropoulos et al. [2] used data for average and worst months and proposed two optimization methods for stand-alone hybrid energy systems. Morgan [3] also proposed method based on the data for worst month – the month, which requires the maximum size of photovoltaic module and wind turbine is selected as the worst month. Borowy and Salameh [4] developed a solar-wind hybrid system optimization algorithm using hourly meteorological data and load demand. Zhou et al. [5] and Notton et al.[6] proposed another optimization algorithms based on time series meteorological data, while the predictive algorithm proposed by Celik [7] uses monthly average wind speed and solar radiations. Other new optimization models have also been developed incorporating some components and weather constraints [8,9]. Optimization techniques proposed by the researchers are based on different methods such as graphical, iterative technique, artificial intelligence and multi-objective methods. Borowy and Salameh [4] developed a solar-wind hybrid system based on graphical construction technique using hourly meteorological data and load demand for past 30 years. Bagul [10] proposed a sizing method using a three-event probability approximation, while Yang et al. [11] proposed an HSWSO iterative technique using LPSP model. Similarly, Kellogg et al. [12] proposed an iterative procedure for making demand gap near to zero. Puri [13] has given a linear programing based convex optimization solution for hybrid energy system using power reliability criterion based on FLNS (Fraction of Load that can Not be Served) calculation. Yadav et al. [14] has designed a wind-diesel hybrid energy system through Hybrid Optimization Model for Electric Renewable (HOMER) and has shown the advantage of wind-diesel system as compared to diesel system. Sureshkumar et al. [15] also has used HOMER software for hybrid renewable energy system optimization and economic analysis. Banos et al. [16] have given a review of current state of the art about hybrid energy system optimization using computational methods and latest advances and future directions for Research and Development in this field. Haidar et al. [17] simulated a PV-diesel hybrid energy system through Hybrid Optimization Model for Electric Renewable (HOMER). Belfkira et al. [18] presented a deterministic algorithm for optimization of hybrid wind/PV/diesel energy system. LunaRubio et al. [19], Erdinc et al. [20] and Saber et al. [21] provided a detailed analysis about hybrid energy system optimization methodologies and presented comparison between advantages and disadvantages of various sizing methodologies developed in the recent years. Mohammed et al. [22] discuss critical state-ofthe-art review of hybrid energy systems planning and indexes multi-objective methods for optimal design. Tan et al. [23] summarizes different modelling approaches for RAPS system

1367

architecture design, component modelling and size optimization. They also discuss the technical challenges associated with RAPS systems design and control. Methods based on artificial intelligence such as artificial neural networks (ANNs), genetic algorithms (GAs) and particle swarm optimization (PSO) algorithm are widely used for finding the optimal sizing solution by minimizing the system cost. Optimization algorithms based on GAs for optimum design have been reported in literature [24–26]. Dufo-Lpez [27] and Seeling [28] implemented GAs for optimization, and simulation time has been reduced. Gupta et al. [29] has proposed GA for finding the optimal sizing coefficient of wind/PV hybrid energy system in remote areas. For optimal sizing calculation, Kalogirou [30] has developed ANNs and GAs. Optimization methods based on fuzzy logic and GA for sizing, battery and diesel generator scheduling have also been presented [31,32]. Using evolutionary algorithm for optimization based on multi-objective optimization which includes resources, load, CO2 emission and costs have been analysed [33]. Strength pareto evolutionary algorithm (SPEA) for hybrid system optimization with multi-objective of emission as well as cost minimization has also been reported [34]. Dufo-Lpez and coworkers [35] were the first to present a triple multi-objective optimization algorithm for cost, CO2 emission and load. Later, Bansal et al. [36] and Kumar et al. [37] have proposed a biogeography based optimization (BBO) algorithm for finding the optimal size of standalone hybrid energy systems. Datta et al. [38] has shown that the solar radiation forecasting using soft computing technique (Multi-Layer Perceptron Neural Network) gives better and accurate results as compared to statistical method (Multiple-Linear Regression). Lei et al. [39] have presented a bibliographical survey on the state-of-the-art research and developments in wind speed and power forecasting. Catalao et al. [40] proposed a novel hybrid approach using wavelet transform, PSO algorithm and an adaptive-network-based fuzzy inference system for short-term wind forecasting and compared the performance with seven other approaches. Rahmati et al. [41] compared the performance of BBO algorithm with similar algorithm that is GA and algorithms and shown that the BBO performance is better than other algorithms. Khan et al. [42] analysed the performance of five neural network training algorithms namely two gradient descent algorithms – back propagation and Levenberg-Marquardt – and three population-based heuristic – Bat Algorithm, Genetic Algorithm, and Particle Swarm Optimization – and has shown that the population-based heuristic algorithms are better as compared to gradient descent algorithms. Hong et al. [43] applied the fuzzy-c-means (FCM) to cluster the operation states and the genetic algorithm with Markov model to determine the optimal size of hybrid energy system. George et al. [15] proposed Markov chain Monte Carlo (MCMC) method for wind power forecast and has shown that MCMC method is excellent fit for probability density as well as autocorrelation function of the generated wind power time series. Mellit et al. [44] applied artificial neural network for generating sequences of global solar radiation and then compared the result with traditional models AR, ARMA, Markov chain, MTM and measured data. The wind speed forecasting results was compared with autoregressive integrated moving average reference model (NRM) proposed by Nielsen et al. [45], neuro-fuzzy (NF) suggested by Pousinho et al. [46], Markov chain Monte Carlo (MCMC) proposed by George et al. [47] and wavelet-PSO-ANFIS approach (WPA) developed by Catalao et al. [48]. The solar radiation forecasting results was compared with Radial Basis Function Neural Network (RBFNN) proposed by Mohandes et al. [49], Recurrent Neural Network (RNN) proposed by A. Chaouachi et al. [50], Empirical method suggested by K. Spokas et al. [51] and Markov chain given by Mellit et al. [44].

1368

R.A. Gupta et al. / Renewable and Sustainable Energy Reviews 41 (2015) 1366–1375

The optimization methods reported in the literature are generally based on the wind speed and solar radiation data of past years. However, it is an established fact that the meteorological data of previous years have significant variations when compared to the present values. Addressing this issue, various models have been proposed incorporating uncertainties for enhancing energy efficiency. In this regard, Khalid and Savkin [52] and Tascikaraogluet al. [53] used wind power prediction for optimizing hybrid energy system, while Hocaoglu et al. [54] studied the effect of use of solar radiation forecasting in improving the efficiency of optimal system design. Wang and Chanan [55] proposed a PSO-based optimization algorithm incorporating uncertainties because of variations in generation, load and equipment performance. It can easily be understood from past practices that accurate wind speed and solar radiation forecasting are crucial for achieving energy efficiency for optimization. These forecasting models are expected to produce more accurate sizing results. As optimization of SHAPS is complex and nonlinear, in this paper we have implemented BBO algorithm. As the wind speed and solar radiation have random variations, forecasting methods were developed using artificial intelligence techniques. Our results showed that the performance of the SAHPS can be improved with better decision-making support by using the forecast values in optimization. This paper is organized as follows: the SAHPS components and their mathematical modelling are explained in section II. Section III describes the optimization problem of SAHPS. In section III, ANN technique for wind speed and solar radiation forecasting along with their performance analysis is discussed. Section V explains the BBO optimization algorithm and section VI describes the application of BBO algorithm for SAHPS optimization. In section VII, performance evaluation of optimization algorithm and effect of forecasting is explained by a numerical case study.

number of WTG to be connected in parallel is directly dependent on the current requirement of the system. The WTG power output at any time ‘t’ can be expressed as: 8 n n n 3 n n n > < ηW ηg 0:5 ρa C P A vr vci o vr o vc vc o vr ovco P WT ðtÞ ¼ P max ð1Þ > : 0 vci Z vr and vr Z vco where P WT is the WTG power output, ηW is the efficiency of WTG, ηg is the efficiency of generator, ρa is the density of air, C P is the power coefficient of WTG, A is the wind turbine swept area, vci is the cut-in wind speed, vc is the rated wind speed and vco is the cut-off wind speed. On the basis of wind speed at any height, the wind speed hub height of wind turbine is calculated as:  γ h vðtÞ ¼ vr ðtÞ: ð2Þ hr where v is the wind speed at the hub height h, vr is the wind speed at reference height hr and γ is the power-law exponent (~1/7 for open land). 2.2. Photovoltaic generation (PV) The PV array size consists of PV panel size and the number of strings in a PV array. The number of PV in series and parallel depend on bus operating voltage and current requirements respectively. The photovoltaic panel output P PV ðtÞ at time ‘t’ is: P PV ðtÞ ¼ ηpc n Am n ηm n P f n I

ð3Þ

where Am is the PV array area, ηpc is the power conditioning efficiency, ηm is the module reference efficiency, P f is the packing factor and I is hourly solar radiation in W m  2. 2.3. Storage batteries (SB)

2. Small autonomous hybrid power system (SAHPS) A typical small autonomous hybrid power system (SAHPS) is a combination of renewable energy sources such as wind turbine generators (WTG), PV panels (PV) with auxiliary supplies like storage batteries (SB) and diesel generator (DG). A typical SAHPS used in our present work is shown in Fig. 1. In the SAHPS, renewable sources are integrated in such a way that they complement each other. In SHAPS optimization, individual components modeling were done first, and then these were combined to analyse the performance of SAHPS to supply a load. 2.1. Wind turbine generators (WTG) The power output of the WTG is dependent on local wind speed and height at which the wind turbine is installed. The

The power input to battery during charging process due to excess energy is generated and is calculated as:

ΔPðtÞ ¼ P total ðtÞ þ P DG ðtÞ  P d ðtÞ=ηbi

ð4Þ

where P total ðtÞ is renewable power generated, P d ðtÞ is connected load, ηbi is inverter efficiency and P DG ðtÞ is the diesel generator power. During the charging process ΔPðtÞ is greater than 0 and in discharging process ΔPðtÞ is less than 0. The battery state of charge (SOC) at time t þ1 is given as:  n ð5Þ P B ðt þ 1Þ ¼ P B ðtÞ þ ΔPðtÞ=U bus ηbb n Δt where ηbb is round-trip efficiency, U bus is DC bus voltage and Δt is the time step. Total renewable power is the sum of WTG and PV output power at time t and is calculated as: Wn

Sn

W ¼1

PV ¼ 1

P total ðt Þ ¼ ∑ P W ðtÞ þ ∑ P PV ðtÞ

ð6Þ

where W n and Sn represent the number of wind generators and photovoltaic panels respectively.

3. Optimization problem of SAHPS In the optimization process of SAHPS, the main aim is to minimize the total cost with the constraint's given in Eqs. (11)– (13). The total cost of SAHPS is represented as: min C t ðP W ; P PV ; P B ; P DG Þ ¼ minðC W þ C PV þ C b þ C g þ C r Þ

Fig. 1. Small autonomous hybrid power system.

ð7Þ

where C t is total cost of the SAHPS, C W , C PV , C b , C g and C r are the cost of wind generator, photovoltaic panels, batteries, diesel generators and cost function which represent power-supply reliability respectively.

R.A. Gupta et al. / Renewable and Sustainable Energy Reviews 41 (2015) 1366–1375

where N PV;Pmax , N W;Pmax and NBAT;P max are the maximum capacities of WTG, PV and battery respectively.

3.1. Cost of SAHPS power sources The general equation of total cost of various components are given as:   Kn r 0 ð1 þ r 0 Þm C i ¼ ∑ ai P i Þ þ repðP Þ þ f uelðP Þ ð8Þ þ omðP i i i ð1 þ r 0 Þm  1 i¼1 where C i is the individual component cost; K n is number; ai is the unit cost (Rs /kW); P i is the power capacity; omðP i Þ is the maintenance and operating costs; repðP i Þ is the replacement costs; f uelðP i Þ is the cost of fuel used for ith element; m is the life span of the project and r 0 is the interest rate. 3.2. Cost of other parameters in SAHPS For providing the reliable supply to the customer, the equivalent cost for power supply reliability is: C r ¼ ccon EENS

ð9Þ

where cco is the Compensation Coefficient and EENSis the Expected Energy Not Served. Within the run-time T (8760 hours), the EENS (kWh/year) is calculated as: T

EENS ¼ ∑ ðP bmin þ P bSOC ðtÞ  P sup ðtÞÞU ðt Þ t¼1

ð10Þ

where P bSOC ðtÞ is the battery state of charge (SOC), P bmin ðtÞ is minimum permissible storage level of battery and P sup ðtÞ ¼ P total ðtÞ þP DG ðtÞ  P d ðtÞ is excess power during hour t. U(t) is a step function, which is 0 if generated energy is more than demand and 1 when generated energy is less than the demand during hour t. 3.3. Constraints in the optimization Depending on the type, size, manufacturer and material of the hybrid energy system components used in SAHPS, there is a set of physical and operating constraints, which must be satisfied during the optimization process for getting feasible solution. The main constraints considered in this paper are as follows: 1. In SAHPS, at any instant, the total energy generated must be greater than or equal to the total load demand along with energy required for satisfying the reliability criterion. Mathematically, this constraint is expressed as: P W ðtÞ þ P PV ðtÞ þ P b ðtÞ þ P g ðtÞ Z ð1  RÞ P d ðtÞ P W ðtÞ þ P PV ðtÞ þ P b ðtÞ þ P g ðtÞ  P dump ðtÞ r P d ðtÞ

ð11Þ

where P W ; P PV ; P b ; P g ; P dump ; P d are the WTG power, solar PV power, battery power, diesel generator power, unused power and total load demand, respectively, and R is the ratio depending on maximum permissible unmet power. In SAHPS design, transmission losses are not considered. 2. The battery state of charge (SOC) P bSOC must be between the capacity of storage batteries P bcap and minimum permissible storage level P bmin at any time instant. The capacity of storage batteries P bcap should not exceed the maximum storage capacity P bcap max . The battery input power P bt must be less than the hourly inverter capacity P bmax . These constraints are mathematically represented as: P bmin r P bSOC r P bmax ; 0 rP bcap rP bcap max ; P bt r P bmax

ð12Þ

3. The number of WTGs, PVs and battery must be within the maximum permissible limit. 0 r Sn r N PV;Pmax ; 0 r W n r N W;P max ; 0 r Bn r N BAT;P max

1369

ð13Þ

4. Forecasting in hybrid energy system Artificial Intelligence (AI) techniques use the learning ability of nature and humans to identify the input-output relationship. Due to advancements in ANNs, ANNs have become important alternative methods for modeling input-output relationship. Generally, ANNs are applied on the problems associated with pattern recognition, forecasting, ecological and atmospheric sciences. The ANN methods have several advantages over conventional methods because ANN accuracy depends on known input-output data set without any assumptions. Artificial Neural Network (ANN) identifies, learn and map pattern in input and output data. Using these mapping, ANN is used for forecast output based on the suitable input. In ANNs, neurons are connected to each other through weighted links and form a black-box representation. ANNs are trained for making them suitable to forecast and predict or classify new patterns. ANN also model human characteristics of problem solving. Unlike statistical methods, ANNs have simpler construction, high tolerance of noisy data, can forecast the pattern which is not used during training, does not require high knowledge about problem domain and not require any mathematical expressions. The use of ANN can be divided into three parts: first is identifying the correct inputs and output, second is selecting a proper network configuration and third is training algorithm modeling. The literature evidences that recently ANNs along with bioinspired techniques are applied for mid- and long-term forecasting in different sectors, areas and countries and established their supremacies in comparison with conventional forecasting methods. As the wind and solar forecasting method are becoming integral part of hybrid energy system due to ever-increasing renewable power integration into the grid, advancements and improvements are required in the forecasting methods. Solar radiation and wind speed data are the critical input parameters for SAHPS size optimization methods. Historical weather data provide key metric to predict power supply from renewable energy sources. Different sources of renewable energy require different levels of modeling and prediction. The viability and reliability of prediction models improve over time as more data are collected and analysed. As power-generation sources become more diverse and distributed, the forecasting and predicting demand and supply are gaining importance. 4.1. Artificial neural network model ANN is constructed by interconnecting number of neurons or nodes in a proper structure. At every node, input signals are received from several nodes. These input signals are processed locally using selected activation function and finally a processed output signal is generated. A network can easily learn and can be trained to perform many complex nonlinear tasks quite efficiently. An ANN can easily understand and learn any relationships between input and output data sets. In this work, time series forecasting was done by using multilayer feed-forward back propagation neural network. During the training, that is during the adjustment of weights and biases of the network, LevenbergMarquardt optimization algorithm was used. In the LevenbergMarquardt algorithm, weights are adjusted through iterative process for error minimization. Using trial and error method, ANN structure, that is, hidden layers number, input, output and

1370

R.A. Gupta et al. / Renewable and Sustainable Energy Reviews 41 (2015) 1366–1375

hidden layers neurons are determined. The procedure to set up a BPANN network is as follows: Step 1: Identify input layer and output layer nodes on the basis of the problem. Step 2: Determine the number of hidden layer(s) and neurons in each hidden layers. No standard rules are defined for selection of neurons and hidden layer number and generally determined by trial and error. Step 3: Learning (or training) using input and corresponding output. In the ANN, training and learning processes are done in the hidden and output layers. A neural network adjusts its weights on the basis of error between predicted outputs to target outputs. Step 4: Testing of the trained network in step 3. When ANN is trained properly, the ANN performance is checked via a test set containing input and output data. The ANN is considered as fully trained, if the error in the testing phase is in the acceptable range. Step 5: Using the trained ANN. Once the ANN is trained and tested, ANN is used to find the response for a given set of input.

for output layer. On the basis of the above observations, the structure of the BPANN was 169-102-64-24. During the training, training function was selected as TRAINGD, maximum number of iteration as 1000 and training goal as 0.001. 4.3. Solar radiation forecasting model ANN model used for solar radiation forecasting is a two-layer feed-forward neural network with various input parameters and back propagation as training method. The input layer neurons act as buffers and input signals from the input layer neurons are distributed to the hidden layer neurons. In the ANN network for solar radiation forecasting, latitude, longitude, elevation of the field site and seasons were selected as input parameters along with solar radiation data for previous 168 instants of seven days. The measured data of six years were used as original data to train the network. The data of next one year was used to validate trained network. In the BPANN for solar radiation forecasting, two hidden layers were selected after trying several combinations. The input layer had 172 nodes, while the output layer had 24 nodes. On the basis of complexity of the problem, the nodes in first hidden layer were selected as 116 and in second hidden layer as 78. The final structure of our BPANN model is 172-116-78-24.

4.2. Wind speed forecasting model 4.4. Testing the performance of the models For the forecasting of the wind speed time series values, back propagation trained ANN was used. For ANN training and testing, wind speed data for past seven years were measured. The first six years wind speed data were used to train the network, while data for the last one year were used to validate the trained network. We used circulation method during the training of the network in which wind speed data of the first week were selected as the input vector, while next day data were used as the target vector. In one day, that is, for 24 hours, hourly values were measured and recorded. In the proposed ANN, one additional input based on the season was selected, which was directly related to the months. Each season comprised of three months as follows: December to February as winter, March to May as spring, June to August as summer and September to November as fall season. As in the proposed ANN, seven day wind speed data and one season variable were selected as the inputs to the network, and the number of input nodes was selected as 169 ( 24n7 ¼168 for wind speed and one for the season). The ANN was supposed to forecast the wind speed for one day, nodes in the output layer was selected as 24. Due to complex and nonlinear nature of wind speed data, two hidden layers were selected in back propagation trained artificial neural network (BPANN). After trying many configurations, the best result was obtained for 102 and 64 nodes for hidden layers. As the activation function is the main process in ANN, TANSIG was selected for hidden layers, while LOGSIG was selected

Due to the uncertainty associated with renewable sources like PV panels and wind turbine generators, the accuracy of ANN input data becomes very important. Generally, in most of the literatures available on optimization of the SAHPS, the measured data of wind speed and solar radiation of past years are used as inputs for optimal size calculations. In this section, we present the comparison between the outputs of the forecasting models and actual measured data. The data measured and collected in Jaipur, Rajasthan, from 2004 to 2010, were used to check the performance of the BPANN algorithm and forecasting methods. Mean absolute error (MAPE), mean squared error (MSE) and maximum error (ME) were used as the measures of forecast performance for the neural network. We applied the BPANN approach for wind speed and solar radiation forecasting in Jaipur, Rajasthan, India, with geographical coordinates latitude: 261 92 N, longitude: 751 82 E and altitude: 431 m above sea level. The training and testing data were taken from past seven years (January 1, 2003 to December 31, 2010). Numerical results of the wind speed and solar radiation forecasting with the proposed ANN approach are shown in Figs. 2 and 3, respectively. Table 1 presents the BPANN performance evaluation measures in the forecasting of wind speed and solar radiation. The results showed nearly similar accuracy throughout the year in all seasons that reflected reality and accuracy of the network.

Fig. 2. Wind speed prediction: (a) actual wind speed, solid line, together with forecasted wind speed, dashed line, in m/s and (b) error in estimation, in m/s.

R.A. Gupta et al. / Renewable and Sustainable Energy Reviews 41 (2015) 1366–1375

1371

Fig. 3. Solar radiation forecasting: (a) hourly actual solar radiation together with forecasted solar radiation, in KW/m2 and (b) error in estimation, in KW/m2.

Table 1 Statistical analysis of the daily forecasting error. √SSE

MAPE

Winter Spring Summer Fall

SDE

Solar

Wind

Solar

Wind

Solar

Wind

7.3 6.5 5.5 6.0

4.9 6.0 4.2 5.0

69.0 139.8 156.9 98.5

123.4 285.1 170.1 139.2

4.7 7.4 6.1 5.8

7.99 19.76 22.67 18.5

The BPANN approach has better wind speed forecasting accuracy as the MAPE has an average value of 4.48% as compared to the other approaches which have MAPE of 19.03%, 6.65%, 5.98% and 5.00% for autoregressive integrated moving average reference model (NRM), neuro-fuzzy (NF), Markov chain Monte Carlo (MCMC) method and wavelet-PSO-ANFIS approach (WPA) respectively. The BPANN approach has better solar radiation forecasting accuracy as the MAPE has an average value of 3.64% as compared to the other approaches which have MAPE of 5.82%, 4.63% 5.44% and 6.81% for Recurrent Neural Network (RNN), Radial Basis Function Neural Network (RBFNN), Markov chain (MC) and Empirical method respectively.

Fig. 4. The model of immigration rate and emigration rate of biology.

each solution Hi, λi and mi can be evaluated as Eq. (14). Since the species in the high-HSI habitat tend to emigrate to low-HSI habitat, a high-HSI habitat has a high mi and low λi. The values of emigration and immigration rates are given as:     λi ¼ I 1  ki =n μi ¼ E ki =n ð14Þ where I is the maximum immigration rate, E is the maximum emigration rate, ki is the number of species of the ith individual and n is the maximum number of species.

5. Biogeography-based optimization (BBO) 6. The BBO algorithm BBO algorithm is developed by using the mathematical model of biogeography. Biogeography is the science of studying the behaviour of species in nature against time and space and species immigration and emigration between habitats. A habitat is a probable solution of the problem. Habitat Suitability Index (HSI) represents the fitness of each habitat. Good habitat has high HSI where large numbers of species are present and bad habitat has a small HSI where small numbers of species are present. The good or bad habitat is decided by the suitability of the area for species. HSI is a function of features values, which characterizes habitability features and are known as Suitability Index Variables (SIV). Good habitats are likely to send features with other. During the iterative process, if HIS of any habitat remains low then that habitat is not suitable and immigration rate increases. Fig. 4 shows the relationships between number of species, emigration rate μ and immigration rate λ. Number of species in a habitat is given by S and represents fitness of that habitat. The maximum possible species in a habitat is Smax and the condition where the emigration rate μ is equal to the immigration rate λ is S0. It can be easily seen from Fig. 4 that island having outstanding performance (S2) has a high m and a low λ and vice versa. After calculating HSI for

The computational procedure of the BBO optimization algorithm used in the paper can be seen in detail in Refs. Bansal et al. [36] and Kumar et al. [37] and is summarized as: 1) Representation of the SIV: Initialization of the algorithm parameters was done. The habitat set is a matrix and each position vector of the habitat represents the possible optimization solution, | a | b | c | d | e | f | g | h|, where a is the number of PV panels, b is the type of PV panel, c is the number of wind turbines, d is the type of wind turbine, e is the number of batteries, f is the type of battery, g is the number of DC generator and h is the size of inverter. 2) Initialization of the SIV: Initialized all habitats set randomly within sizing constraints. The SAHPS BBO algorithm consists of the following steps: Step 1: Set SAHPS constraints. Step 2: Initialize each habitat that represents a candidate solution of the problem. Step 3: On the basis of specified μ and λ, calculate the HIS (Cost) for each habitat set.

1372

R.A. Gupta et al. / Renewable and Sustainable Energy Reviews 41 (2015) 1366–1375

Step 4: On the basis of calculated HSI, remove the bottom habitats with low HIS (elite habitats). In top p habitat, no changes were made. Step 5: On each non-elite habitats having high HIS, migration operation of BBO algorithm was performed. The selection of non-elite habitats for migration operation was done based on the following: 1) The lower and upper values of immigration rate λlower and λupper are selected. 2) The λ and μ for each habitat set is calculated. 3) The habitats having λ between λlower and λupper were selected for migration operation. Step 6: For the new habitats generated after migration process, species count probability was recalculated for each habitat. Some non-elite habitats were selected for mutation operation, if the mutation rate was greater than a randomly generated number. During the mutation, selected habitat was replaced by randomly generated habitat that satisfies SAHPS constraints. Step 7: After the completion of the predefined number of iterations, the optimization process was completed, and habitat with HIS is the optimal size of SAHPS (else go to step 3 for the next iteration).

7. Evaluation of the effect of forecasted data on optimization The BBO algorithm is applied to the design of SAHPS system with wind/ PV source for supplying power to a colony located in the Jaipur, Rajasthan, with geographical coordinates defined as follows: latitude, 26◦ 92 N; longitude, 751 82 E; and altitude, 431 m above sea level. Jaipur, a city in western Rajasthan, is selected for SAHPS design due to the availability of high wind and solar resources during all seasons. For analysing the effect of input weather data on optimal sizing result of SAHPS design, hourly solar radiation and wind speed data were forecast using the BPANN technique as mentioned in section IV. For simplicity, all losses during the SAHPS operation were ignored because the system was considered as remotely located stand-alone system. For studying and designing a forecasting system, eight years data were recorded. The wind speed measurement was done at a height of 30 meters, and solar radiation was measured on a horizontal surface. The daily load profile of study area is shown in Fig. 5(c). At the study area, the average daily load was 2263 kWh/day with a 261 kW peak demand, and variation of 30 percent in load profile was considered. The total energy demand during a complete one year was 8,25,985 kWh/year. As Rajasthan is blessed with large natural resources, monthly solar radiation is between 4 and 7 kWh/m2/d (Fig. 5(a)) and wind speed is between 4 and 11 m/s

(Fig. 5(b)). Input parameters in optimization algorithm include technical characteristics, capital and maintenance costs, economical parameters and study assumptions which are shown in Table 2. In the first step, the BBO optimization algorithm was applied to find the optimal sizing solution of the SAHPS system. The optimal result of BBO algorithms for the SHAPS system is presented in Table 3. The results show that the BBO algorithm was able to find optimum design parameters of SAHPS quite effectively. To analyse the performance and hourly behaviour of the SAHPS, the simulation

Table 2 Technical data and study assumptions of SAHPS [26]. Description

Data

PV Capital cost Lifetime Operation maintenance cost and replacement cost

200,000 Rs/kW 25 years 1,000 Rs /kW/year and 200,000 Rs/kW

Wind turbine Rated power Capital cost Lifetime Operation maintenance cost and replacement cost Cut-in speed (m/s) Vci Cut-out speed (m/s) Vco Hub height

Variable (0–300 KW) AC 65,000 Rs/kW 25 years 1,000 Rs./kw/year and 65,000 Rs/kW 2–3 25 30 m

Diesel generator units Capital cost Rated power Minimum possible output power Operation maintenance cost and replacement cost Operating hours Batteries Type of batteries Nominal voltage (V) Nominal capacity Nominal energy capacity of each battery Operation maintenance cost and replacement cost Dispatch/operating strategy Capital cost Converter Capital cost Operation maintenance cost and replacement cost Lifetime Spinning reserve and minimum renewable fraction Annual interest rate and project life time

20,000 Rs/kW Variable (0–100 MW) 30% of rated power 1 Rs/h/kW and 20,000 Rs/kW 16,000 h Tarjan L16P 6V 360 Ah 2.16 kWh 100 Rs/year and 4000 Rs Multiple diesel load following 10000 Rs. 50,000 Rs/kW 100 Rs/year/kW and 50,000 Rs/ kW 10 years 10% and 60% 10% and 25 years

Fig. 5. (a) Daily average solar radiation data for one year. (b) Average daily wind speed (m/s). (c) Daily average load profile.

R.A. Gupta et al. / Renewable and Sustainable Energy Reviews 41 (2015) 1366–1375

was run for one complete year using the optimal size configuration obtained from BBO algorithm as shown in Table 3. Using the simulation run, the power supplies from the renewable sources, the power demand, battery bank power, diesel generator output, and diesel used, etc., were calculated for optimal size of SAHPS. Figs. 6 and 7 show the power profile without and with wind speedsolar radiation forecasting case, respectively. When energy supplied from renewable sources was less than the demand, the storage battery was used to supply the remaining load. When the battery was also unable to supply the load, diesel generator was used to supply the load as well as charge the battery. In second step, to study the effect of using the forecast data as the input, on the optimal sizing solution performance, the simulation was run for actual data of year 2010, and then instead of previous year data, forecast data of year 2011 is used. The WTG power and PV cell power generated in both the cases is shown in Figs. 8 and 9, respectively. From these figures, it is evident that there is a large variation in the power generated when past years data and forecast data were used as the input. In third step, for checking the use of forecasting model in sizing studies, four different optimization simulations were performed. In all four cases, the performance of SAHPS was evaluated by calculating the PV output, wind turbine output, unmet load, diesel used and Loss of Load Probability (LLP), etc. In the first case, simulation was run for measured solar radiation and wind speed data of the year 2010 as inputs in the optimization algorithm. In the second case, measured wind speed and BPANN model forecast

1373

solar radiation data (on the basis of the year 2003–2010 data) were used as inputs. In the third case, measured solar radiation and BPANN model forecast wind speed data (on the basis of the year 2003–2010 data) were used as inputs. In the fourth simulation, both solar radiation and wind speed forecast data were used as inputs. To analyse the performance and hourly behaviour of the SAHPS, the simulation was run for the year 2011 using the optimal size configuration obtained from BBO algorithm in all four cases. From Table 4, it is evident that due to the use of forecasting values of wind speed and solar radiation for the year 2011, instead of measured data of year 2010, has improved the performance of the SAHPS. When BPANN model forecast solar radiation and wind speed were used, the SHAPS showed less unmet load and LLP. We also observed that the wind speed forecasting had greater effect on SAHPS performance as compared to the solar radiation forecasting because of linear nature of solar radiation as compared to the random nature of wind speed forecasting. When both solar radiation and wind speed forecasting were used in the simulation, the LLP probability reduced to 0.0191 from 0.0205. In fourth step, for checking the effect of forecasting model in sizing studies, four different cases are run in optimization algorithm. In first case, optimization is run for measured solar radiation and wind speed data of the year 2010. In second case, optimization was run with measured solar radiation of year 2010 and BPANN model forecast wind speed data (approximate data of year 2011). In third case, optimization was run with measured wind speed and BPANN model forecast solar radiation data (approximate data of

Table 3 Optimal size result through proposed BBO algorithm. SN

PV (kW)

Wind (units)

Diesel (kW)

Battery (units)

Inverter (kW)

Net present costs (Rs)

Energy produced (kWh)

COE (Rs/kWh)

1

52

1

163

813

93

67,877,356

1,059,639

10.81

Fig. 6. Power profile of optimal result without wind speed and solar radiation forecasting.

Fig. 7. Power profile of optimal result with wind speed and solar radiation forecasting.

1374

R.A. Gupta et al. / Renewable and Sustainable Energy Reviews 41 (2015) 1366–1375

Fig. 8. Effect of wind speed forecasting on wind turbine generator power

Fig. 9. Effect of solar radiation forecasting on solar PV cell power. Table 4 Effect of forecasting on SAHPS performance. SN

Solar radiation

Wind speed

PV %

WTG %

Unmet load (Wh/yr)

Diesel used (littres/year)

Loss of load probability (LLP)

1 2 3 4

Measured Forecasted Measured Forecasted

Measured Measured Forecasted Forecasted

9 10 9 10

63 63 65 65

0.724 .637 .597 .589

124.67 116.32 103.43 102.23

0.0205 0.0214 0.0197 0.0191

Table 5 Effect of forecasted data on optimal size result through proposed BBO algorithm. SN

PV (kW)

Wind (units)

Diesel (kW)

Battery (units)

Inverter (kW)

Net present costs (Rs)

Energy produced (kWh)

COE (Rs/kWh)

1 2 3 4

52 50 51 48

1 1 1 1

163 157 161 152

813 794 810 779

93 90 91 85

67,877,356 66,577,649 67,511,060 66,877,356

1,059,639 982,232 1,012,234 1,059,639

10.81 10.61 10.74 10.54

year 2011). In fourth case, optimization was run with solar radiation and wind speed forecast data (approximate data of year 2011). The optimal result of BBO algorithms for the SHAPS system is presented in Table 5. From Table 5, it can be easily seen that when forecast data (approximate data of 2011) is used instead of actual data of 2010, the optimal size configuration of hybrid energy system is significantly changed and net present cost, total energy produced by the system significantly reduced.

8. Conclusion In the SAHPS systems, power generated through renewable sources directly depends on the weather resource, as WTG and PV panel power varies with wind speed and solar radiation. The

WTG and PV panels are combined with battery storage and diesel generator to completely supply the demand with good degree of reliability. In this paper, optimal design algorithm for SAHPS system is accomplished through a Biogeography Based Optimization (BBO) Algorithm and Back propagation trained Artificial Neural Network (BPANN) based time-series forecasting methods are implemented for wind speed and solar radiation forecasting. It is emphasized that in the SAHPS optimal sizing method, use of forecast data of solar radiation and wind speed affects the optimization sizing results. The effect of forecast weather data on sizing applications is simulated and analysed using MATLAB. To demonstrate the effect of forecast data on optimization results, a case study for Jaipur, Rajasthan, region is done. From the case study, it can be easily concluded that use of forecast data has high influence on optimal sizing algorithm performance. The accuracy of the optimal sizing result improves when

R.A. Gupta et al. / Renewable and Sustainable Energy Reviews 41 (2015) 1366–1375

forecast data are used instead of the previous year data. However, advancements in wind speed and solar radiation predictions can play a vital role in the future power systems and can be easily integrated into power system design and control. References [1] Nema P, Nema R, Rangnekar S. A current and future state of art development of hybrid energy system using wind and PV-solar: a review. Renew Sust Energ Rev 2009;13(8):2096–103. [2] Protogeropoulos C, Brinkworth B, Marshall R. Sizing and techno-economical optimization for hybrid solar photovoltaic/wind power systems with battery storage. Int J Energ Res 1997;21(6):465–79. [3] Morgan T. The performance and optimization of autonomous renewable energy systems, phd thesis, University of Wales, Division of Mechanical Engineering and Energy Studies. Cardiff 1996. [4] Borowy BS, Salameh ZM. Optimum photovoltaic array size for a hybrid wind/ pv system. IEEE T Energ Conver 1994;9(3):482–8. [5] Zhou W, Lou C, Li Z, Lu L, Yang H. Current status of research on optimum sizing of stand-alone hybrid solar–wind power generation systems. Appl Energ 2010;87(2):380–9. [6] Notton G, Muselli M, Poggi P, Louche A. Autonomous photovoltaic systems: influences of some parameters on the sizing: simulation time step, input and output power profile. Renew Energ 1996;7(4):353–69. [7] Celik A. A simplified model for estimating the monthly performance of autonomous wind energy systems with battery storage. Renew Energ 2003;28(4):561–72. [8] Muselli M, Notton G, Poggi P, Louche A. Pv-hybrid power systems sizing incorporating battery storage: an analysis via simulation calculations. Renew Energ 2000;20(1):1–7. [9] Kaye, R., A new approach to optimal sizing of components in stand-alone photovoltaic power systems. In: Photovoltaic energy conversion, 1994. Conference record of the twenty fourth. IEEE photovoltaic specialists conference1994, 1994 IEEE first world conference on, Vol. 1, IEEE, 1994, pp. 1192–195. [10] Bagul A, Salameh Z, Borowy B. Sizing of a stand-alone hybrid windphotovoltaic system using a three-event probability density approximation. Sol Energy 1996;56(4):323–35. [11] Yang H, Lu L, Zhou W. A novel optimization sizing model for hybrid solar-wind power generation system. Sol Energy 2007;81(1):76–84. [12] Kellogg W, Nehrir M, Venkataramanan G, Gerez V. Generation unit sizing and cost analysis for stand-alone wind, photovoltaic, and hybrid wind/pv systems. IEEE T 1998;13(1):70–5. [13] Puri, A., Optimally sizing battery storage and renewable energy sources on an off-grid facility. In: Power and energy society general meeting (PES), 2013 IEEE, IEEE, 2013, pp. 1–5. [14] Yadav, D., Girimaji, S., Bhatti, T., Optimal hybrid power system design for rural electrification. In: Power, control and embedded systems (ICPCES), 2012 2nd international conference on, IEEE, 2012, pp. 1–6. [15] Sureshkumar, U., Manoharan, P., Ramalakshmi, A., Economic cost analysis of hybrid renewable energy system using homer. In: Advances in engineering, science and management (ICAESM), 2012 international conference on, IEEE, 2012, pp. 94–99. [16] Banos R, Manzano-Agugliaro F, Montoya F, Gil C, Alcayde A, Go´ mez J. Optimization methods applied to renewable and sustainable energy: a review. Renew Sust Energ Rev 2011;15(4):1753–66. [17] Haidar A, John PN, Shawal M. Optimal configuration assessment of renewable energy in Malaysia. Renew Energ 2011;36(2):881–8. [18] Belfkira R, Zhang L, Barakat G. Optimal sizing study of hybrid wind/pv/diesel power generation unit. Sol Energy 2011;85(1):100–10. [19] Luna-Rubio R, Trejo-Perea M, Vargas-Va´zquez D, R´ıos-Moreno G. Optimal sizing of renewable hybrids energy systems: a review of methodologies. Sol Energy 2012;86(4):1077–88. [20] Erdinc O, Uzunoglu M. Optimum design of hybrid renewable energy systems: Overview of different approaches. Renew Sust Energ Rev 2012;16(3):1412–25. [21] Saber AY, Venayagamoorthy GK. Plug-in vehicles and renewable energy sources for cost and emission reductions. IEEE T Ind Electron 2011;58 (4):1229–38. [22] Mohammed, O.H., Amirat, Y., Benbouzid, M., Tang, T., Hybrid generation systems planning expansion forecast: a critical state of the art review. In: Industrial electronics society, IECON 2013-39th annual conference of the IEEE, IEEE, 2013, pp. 1668–673. [23] Tan Y, Meegahapola L, Muttaqi KM. A review of technical challenges in planning and operation of remote area power supply systems. Renew Sust Energ Rev 2014;38:876–89. [24] Koutroulis E, Kolokotsa D, Potirakis A, Kalaitzakis K. Methodology for optimal sizing of stand-alone photovoltaic/wind- generator systems using genetic algorithms. Sol Energy 2006;80(9):1072–88. [25] Yang H, Zhou W, Lu L, Fang Z. Optimal sizing method for stand-alone hybrid solar–wind system with lpsp technology by using genetic algorithm. Sol Energy 2008;82(4):354–67.

1375

[26] Yang H, Wei Z, Chengzhi L. Optimal design and techno-economic analysis of a hybrid solar–wind power generation system. Appl Energ 2009;86(2):163–9. [27] Dufo-Lo´ pez R, Bernal-Agust´ın JL. Design and control strategies of pv-diesel systems using genetic algorithms. Sol Energy 2005;79(1):33–46. [28] Seeling-Hochmuth G. A combined optimisation concet for the design and operation strategy of hybrid-pv energy systems. Sol Energy 1997;61(2):77–87. [29] Gupta, R., Kumar, R., Bansal, A.K., Economic analysis and design of stand-alone wind/photovoltaic hybrid energy system using genetic algorithm. In: Computing, communication and applications (ICCCA), 2012 International conference on, IEEE, 2012, pp. 1–6. [30] Kalogirou SA. Optimization of solar systems using artificial neural-networks and genetic algorithms. Appl Energ 2004;77(4):383–405. [31] Belfkira, R., Hajji, O., Nichita, C., Barakat, G., Optimal sizing of stand-alone hybrid wind/pv system with battery storage. In: Power electronics and applications, 2007 European conference on, IEEE, 2007, pp. 1–10. [32] Senjyu T, Hayashi D, Urasaki N, Funabashi T. Optimum configuration for renewable generating systems in residence using genetic algorithm. IEEE T Energy Conver 2006;21(2):459–66. [33] Pelet X, Favrat D, Leyland G. Multiobjective optimisation of integrated energy systems for remote communities considering economics and co¡ sub¿ 2¡/sub¿ emissions. Int J Therm Sci 2005;44(12):1180–9. [34] Bernal-Agust´ın JL, Dufo-Lo´ pez R, Rivas-Ascaso DM. Design of isolated hybrid systems minimizing costs and pollutant emissions. Renew Energ 2006;31 (14):2227–44. [35] Dufo-Lopez R, Bernal-Agustin JL. Multi-objective design of pv–wind–diesel– hydrogen–battery systems. Renew Energ 2008;33(12):2559–72. [36] Bansal AK, Kumar R, Gupta R. Economic analysis and power management of a small autonomous hybrid power system (sahps) using biogeography based optimization (BBO) algorithm. IEEE T Smart Grid 2013;4(1):638–48. [37] Kumar R, Gupta R, Bansal AK. Economic analysis and power management of a stand-alone wind/photovoltaic hybrid energy system using biogeography based optimization algorithm. Swarm Evol Comput 2013;8(0):33–43. [38] B. Datta, S. Pal, R.R. Chowdhury, S. Sarkar, A. Datta. Estimation of solar radiation at a particular place: comparative study between soft computing and statistical approach. Int J Comput Sci Eng. 2011;3(8):3027–36. [39] Lei M, Shiyan L, Chuanwen J, Hongling L, Yan Z. A review on the forecasting of wind speed and generated power. Renew Sust Energ Rev 2009;13(4):915–20. [40] Catalao J, Pousinho H, Mendes V. Hybrid wavelet-PSO-ANFIS approach for shortterm wind power forecasting in Portugal. IEEE T Sust Energ 2011;2(1):50–9. [41] Rahmati SHA, Zandieh M. A new biogeography-based optimization (bbo) algorithm for the flexible job shop scheduling problem. Int J Adv Manuf Tech 2012;58(9-12):1115–29. [42] Khan K, Sahai A. A comparison of ba, ga, pso, bp and lm for training feed forward neural networks in e-learning context. Int J Intell Syst Appl (IJISA) 2012;4(7):23. [43] Hong Y-Y, Lian R-C. Optimal sizing of hybrid wind/PV/diesel generation in a stand-alone power system using Markov-based genetic algorithm. IEEE T Power Deliver 2012;27(2):640–7. [44] Mellit A, Benghanem M, Arab AH, Guessoum A. A simplified model for generating sequences of global solar radiation data for isolated sites: using artificial neural network and a library of Markov transition matrices approach. Sol Energy 2005;79(5):469–82. [45] Nielsen TS, Joensen A, Madsen H, Landberg L, Giebel G. A new reference for wind power forecasting. Wind Energy 1998;1(1):29–34. [46] Pousinho, H., Mendes, V., Catala~o, J., Neuro-fuzzy approach to forecast wind power in Portugal. In: International conference on renewables energies and power quality-ICREPQ, Vol. 10, 2010, pp. 1–4. [47] Papaefthymiou G, Klockl B. Mcmc for wind power simulation. IEEE T Energy Conver 2008;23(1):234–40. [48] Catalao J, Pousinho H, Mendes V. Hybrid wavelet-PSO-ANFIS approach for shortterm wind power forecasting in portugal. IEEE T Sust Energ 2011;2(1):50–9. [49] Mohandes M, Balghonaim A, Kassas M, Rehman S, Halawani T. Use of radial basis functions for estimating monthly mean daily solar radiation. Sol Energ 2000;68(2):161–8. [50] Chaouachi A, Kamel RM, Nagasaka K. Neural network ensemble-based solar power generation short-term forecasting. JACIII 2010;14(1):69–75. [51] Spokas K, Forcella F. Estimating hourly incoming solar radiation from limited meteorological data. Weed Sci 2006;54(1):182–9. [52] Khalid M, Savkin A. A model predictive control approach to the problem of wind power smoothing with controlled battery storage. Renew Energ 2010;35 (7):1520–6. [53] Tascikaraoglu A, Uzunoglu M, Vural B. The assessment of the contribution of short-term wind power predictions to the effi- ciency of stand-alone hybrid systems. Appl Energ 2012;94:156–65. [54] Hocaoglu FO, Gerek ON, Kurban M. The effect of model generated solar radiation data usage in hybrid (wind–pv) sizing studies. Energ Conver Manage 2009;50(12):2956–63. [55] Wang, L., Singh, C., Pso-based hybrid generating system design incorporating reliability evaluation and generation/load forecasting. In: Power Tech, 2007 IEEE Lausanne, IEEE, 2007, 1392-1397.