Modelling of Daily Solar Energy System Prediction using Support Vector Machine for Oman

Transcription

1 Modelling of Daily Solar Energy System Prediction using Support Vector Machine for Oman Hussein A Kazem Associate Professor, Faculty of Engineering, Sohar University PO Box 44, Sohar, PCI 311 Oman. Jabar H. Yousif Assistant Professor, Faculty of Computer & Information Technology, Sohar University PO Box 44, Sohar, PCI 311 Oman. Miqdam T Chaichan Assistant Professor, Department of Mechanical Engineering University of Technology Baghdad, Iraq. Abstract Solar energy is the most important renewable energy source, which could be utilized through photovoltaic or thermal systems. This paper aims to design and implement a machine learning technique called Support Vector Machine (SVM) for the management of energy generation based on experimental work. The SVM model is consist of two inputs and one output. The inputs are solar radiation, and ambient temperature; the output is the photovoltaic current. The practical implementation of the proposed SVM model is achieved a final MSE of ( ) in training phase and ( ) in cross validation phase. The experiments achieved a value of (R squared) equal to which indicates the predicting model is very close to the regression line and a well data fitting to the statistical model. Besides, the proposed model achieved less MSE in comparison with other related work. Keywords: Solar energy prediction; machine learning; support vector machine; Oman. INTRODUCTION Currently the subject of generating energy from renewable sources is considering as one of the important topics that are growing promptly as a result of its many benefits. The current step of technological development makes the extraction of renewable energy from various sources such as the sun, wind, geothermal energies, and many other sources commercially viable process [1]. Renewable energy started competing with non-renewable energy sources due to many reasons (i.e., fluctuation of oil prices, independency, reduction in renewable energy technology prices and improvement in efficiency, etc.) [2]. Renewable energy generation capacity significantly affected by fluctuations in the atmosphere for example, solar energy generated in the morning for the presence of sunlight during the day and the energy generated from the wind depends on the wind speed and geographic location. These fluctuations in the generation rates cause instability in the power grid. Thus, to overcome this problem, the energy is generated at the time availability of natural resources is and the excess energy is stored on the need for later use. At this point, the problem of finding an efficient management of these sources of electric power generation is more needed [3]. Solar energy could be utilized through photovoltaic PV or thermal systems. Solar energy technology and especially solar cells and photovoltaic have been experimented for more than 70 years and it shows good success. The PV system prices get down and at the same time, the PV efficiency is improved [4]. Machine learning is a subdivision of artificial intelligence that is developed as a result of studies in pattern classification and recognition for finding mathematical models for various real life problems. Machine learning investigates the construction of algorithms that can learn from the previous data and help in finding a forecast on the data in the current and future time. The machine learning applied iterative and interactive statistical methods in the construction of computational models to obtain the desired results [5]. The factors like efficiency of learning algorithms, the complexity of the problem, methods of representation of data are the most important factors affect the accuracy of the results and future forecasting for data. Several machine learning methods are used for designing and implementing different phases of a renewable energy power systems based on the problem requirements and its characteristics [6, 7 & 8]. Hence, the adaptation of optimal location and structures of renewable power plants is one of the important implementation of learning machine methods. Recently, the Support Vector Machines (SVMs) have been widely implemented into several problems of renewable energy power systems [9, 10, 11, and 12]

2 There is number of studies in literature related to the use of machine learning techniques to predict PV performance, including Multilayer Perceptron (MLP) network, Probabilistic Neural Networks (PNN), General Regression Neural Networks (GRNN), Radial Basis Function networks (RBF), Cascade Correlation, functional link networks, Kohonen networks, Gram-Charlier networks, learning vector quantization, Hebb networks, Adaline Networks, hetero Associative Networks, Recurrent Networks, and hybrid networks [9]. Reference [10] proposed prediction model for solar power generation based on experimental work. Different machine learning techniques has been used. The authors included that SVM in the multiple regression techniques. In SVM model they tried polynomial, linear, RBF kernels. They claimed that SVM model accuracy increased up to 27%. Furthermore, principal component analysis has been used to improve the model. Ref. [11] proposed hybrid intelligent predictor. The proposed system used regression models namely, RBF, MLP, Linear Regression (LR), SVM, Simple Linear Regression (SLR), Pace Regression (PR), Additive Regression (AR), Median Square (LMS), IBk (an implementation of knn) and Locally Weighted Learning (LWL). They claimed that LMS, MLP, and SVM are the most accurate models in term of MAE and MAPE. Ref. [12] used SVM model to predict solar energy; he found that SVM accuracy is less than Gaussian Process Regression method. Ref. [5] implements a SVM model to estimate the daily solar radiation using air temperatures. The developed SVM model used a polynomial kernel function which performed better than other SVM models. He obtained a highest NSE of 0.999, and the R-square of 0.969, while the lowest RMSE is and RRMSE of In this study, SVM model will be proposed for solar PV system in Sohar, the second largest city in Oman, based on experimental data for solar irradiations and installed PV system in the Solar Cells and Photovoltaic Research Lab in Sohar University, Oman. This paper aims to discuss and implement machine learning methods for acceptable management of the energy generation of a PV system. STUDY LOCATION In Oman due to growth in industry and population the electrical demand increased and it will keep increasing in near future as shown in Figure 1. The maximum load increased from 4,634 MW in 2013 to 5,691 MW in In 2018 the forecasted power demand is 6.8 GW, which mean more power plants need to be installed. Figure 1. Oman peak power demand for and projection till 2018 Sohar is the second largest city in Oman after the capital, Muscat. It is the primary industrial center which is just 230 km from Muscat, the capital [13]. The solar radiation intensity in Solar (Direct normal (DNI), diffuse horizontal (DHI) and global horizontal (GHI)) was measured and recorded experimentally. The measurements revealed that higher solar irradiance in summer (July) where the maximum irradiation recorded value was 950 W/m 2 at 1 PM, and the minimum ones was 202 W/m 2 in January at the same hour [14]. As Sohar stranded between the Al-Arab Sea coast and Al Hajar Mountains series, this area subjected to few dust storms [15]. Sohar city climate is harsh, with maximum temperatures that sometimes reach 50 C in the summer season. Also, Sohar has a humid climate as it is a coastal area [16, 17 & 18]. The Authority of Electricity Regulation (AER) in Oman has explored the renewable energy sources in 2008 and they claimed that the priorities are solar than wind [19]. The Public Authority for Electricity and Water (PAEW) in 2010 investigated the best locations to install large scale solar system to generate electricity [20]. In 2011, Sohar University (SU) won a grant from the Omani Research Council to investigate, design and assess PV systems in Oman in term of technical and economic criteria. Refs. [21 & 22] presented feasibility study of PV systems in Oman standalone and grid connected, respectively. Authors of reference [21] proposed optimization of the PV system tilt angle, array size and storage battery capacity using MATLAB numerical method. Load demand and hourly meteorological data has been used. It is found that the PV system sizing ratio for PV array and battery are 1.33 and 1.6, respectively. Also, they claimed that the cost of energy for standalone PV system in Oman is USD/kWh. Ref. [22] investigated and assessed the grid connected PV system in Oman. It is found that the Crist Factor and Yield Factor, two technical criteria of the grid connected PV system are 21% and 1875 kwh/kwp/year respectively. Meanwhile, the cost of the energy and the payback period as an economic criteria are USD/kWh and 11 years, respectively. As a conclusion they claimed that the energy generated by PV systems is cheaper than energy generated by fossil fuel in Oman

3 This study proposed model predicts the output of grid connected PV system in Oman. The proposed model is based on implementation of artificial intelligent techniques. The SVM is used to classify and predict the future amount of production of the PV system. The system experimental data has been measured in Sohar, Oman. EXPERIMENTAL SETUP 1. Practical measurement In the current study, 24 PV modules have been installed in Sohar University in Oman. The rating of PV module is 140W as shown in Table 1. Three PV systems configurations; standalone, grid connected and tracking systems have been designed and evaluated as shown in Figure 2 (a). The meteorological data has been measured and recorded using weather station as shown in Figure 2 (b). The grid connected system data (voltage, current, power and energy) has been recorded and monitored. Also, the measured environmental parameters are linked to the productivity of the PV system. This research is a part of a research on grid connected PV system (see Figure 3) and consequently the production of the PV system here is needed. Table 1: Installed solar system specifications PV array PV panels (24 panels) 140 Wp (3.36 kwp) Maximum voltage 17.7 Maximum current 7.91 Open circuit voltage 22.1 Short circuit current 8.68 Efficiency 13.9% Temperature coefficient of Vo.c Temperature coefficient of Is.c %/k 0.06 %/k Figure. 2. (a) Grid connected PV system Figure. 2. (b) Meteorological station Figure. 3. Solar photovoltaic grid connected system 2. Machine Learning The Machine Learning is one of the applications of artificial intelligence to simulate the thinking capabilities of human [23]. It is concerning with the techniques and methods that help the machine to learn. Machine learning is developed as a result of studies in pattern classification and recognition for finding mathematical models for various problems. Machine learning investigates the construction of algorithms that can learn from the previous data and help in finding a forecast on the data in the current and future time. The machine learning applied iterative and interactive statistical methods in the construction of computational models to obtain the desired results [24]. Several techniques and methodologies are established for machine learning tasks. The new development of machine learning techniques is mainly utilizing Kernelbased methods such as support vector machines (SVM), Gaussian processes, etc. [6]. A SVM is considered as a supervised learning technique which is presented by Refs. [23 & 25] in COLT-92 as presented in Figure 4. It is considering as a classification and regression prediction techniques. SVM is deployed in many applications, such as face analysis and detection, hand writing recognition, pattern classification and regression etc. A classification technique comprises of training and testing data attributes. The main benefits of SVM are minimizing the classification error by implement an iterative training algorithm and maximizing the margin between two hyper-planes. Figure 5 show the classification Hyper-planes (wx+b>1 and wx+b<-1). The SVM separates the cases of different class labels of hyper-planes of multidimensional space into binary cases (Positive, Negative) as illustrated in Figure 5. So, the positive zone is defined by (w. x+ + b = +1), while the negative zone is defined by (w. x- + b = -1). If the negative zone is subtract from the positive zone, then w. (x+-x-) = 2. Now, the margin M is defined by equation 1. ( x x ) w M (1) w 10168

4 The maximum margin linear classifier is defined as in equation 2. 2 Maximize M (2) w Formulate a Quadratic Optimization Problem in order to solve for w and b as defined in equation 3. Minimize φ (W) = 1 2 wt w (3) Subjected to y i (wx i + b) 1 for i Then the classifying function is defined as in equation 4. f(x) = α i y i X T i X + b (4) training phase are used for the sake of maximizing the chance of classification and prediction of data. The Gaussian Axon is implemented as a transfer function. Figure 6. SVM Architecture Figure 4. SVM classification Hyper-planes Figure 5. Positive and negative classes 3- The SVM design and architeture The SVM is applied using the Neuro-Solutions software package. The SVM utilizes different types of kernels which include linear, polynomial, and Radial Basis Function (RBF). The non-linear information will convert into a linear form by mapping them into a high-dimensional space. Then, only the closet data to the decision zone is choosing. Moreover, in order to have a good generalization the Neuro-Solutions software package utilizes the large margin classifiers in the training phase. Figure 6 demonstrates the SVM architecture, which has only (2 synapses) PEs as input data sets (the time stamp & corresponding UPV-Solar), and has (1 synapse) PE as output data set (Photovoltaic temperature). The data undertaken are generated from 24-PV modules, which have been installed at Sohar University in Oman. The grid connected system data (voltage, current, power and energy) has been recorded and monitored. Also, the measured environmental parameters are linked to the productivity of the PV system. In the experiment undertaken, the maximum number of epochs is set to The step size is (0.01). The experimental implement a Radial Basis Function (RBF) as SVM kernel. Besides, the large margin classifiers in the RESULTS & DISCUSSION Estimating the classification error is a significant matter. The Neuro-Solutions software package is implement different type of methods to evaluate and estimate the errors in output data. The main method is Mean Square Error (MSE). MSE is a measure of the variation of the predicted values from the measured data, which give information on the short-term performance of the models. A great positive MSE implies a big deviation in the predicted value of the measured value. MSE 1 n n i 1 ( I pi I i ) Where I i is the measured value, I pi is the predicted value and n is the number of observations. Figure 7 shows the training and cross validation MSE of the SVM network, which is clearly demonstrating the graph line of training data set in line with cross validation data sets. The practical implementation of SVM is achieved a final MSE of ( ) in training phase and ( ) in cross validation phase as illustrated in Table 2. The experiment is giving evidence that there is a strong relation between the input and the output variables. It is got a correlation factor of ( ). Figure 8 depicts the graph of comparison of desired PV output current and the network output, which is predicate exactly the tested data. And then can be generalized to utilize any unseen data for the current and future time. Figure 9 presents the trend line of desired output y 1 and the network output y. The predicting model for network output is generated by a polynomial of forth orders is defined as in equation 6. (5) 10169

5 MSE versus Epoch Training MSE Cross Validation MSE MSE Epoch Figure 7. The graph of training and cross validation MSE Output Desired Output and Actual Network Output 0 O_Upv_soll Exemplar O_Upv_soll Output Figure 8. The graph of training and cross validation MSE y= x x x x (6) The predicting system is generating an output closely to the desired output y1. The desired output is defined as in equation 7. y 1= x x x x (7) The coefficient of determination (R) is used to measure how close the data from the regression line. Usually determine its value between 0 and 1, and the closer the value of 1 gave an indication that the data correspond to the regression line. The experiment got a value of (R squared) equal to which indicates the predicting model is very close to the regression line and a well data fitting to the statistical model. Figure 9. The graph of trend line of output and desired results. The future individual values prediction could be shown through the MSE efficiency of the developed model (SEE Table 2). A large positive MSE means a large deviation in the predicted value from the real value. The developed model found to be able to predict the individual values since its MSE is Also, Table 3 illustrates a comparison of MSE for the proposed model and models in literature, which shows that the proposed model is more accurate than other models in China, UAE and Malaysia. Also, it is found that SVM is more efficient in comparison with other ANN models. Table 2: The Minimum MSE & Final MSE of SVM network Best Networks Training Cross Validation Epoch # Minimum MSE Final MSE Table 3: MSE comparison of the proposed model and models in literature Reference Model type Site MSE Proposed model SVM Oman J. Zeng et al, SVM, SVM China [8] and Wavelet Tamer et al, FFMLP Malaysia [26] Tamer et al, FFMP Malaysia [27] Maitha et al, 2013 [28] MLP UAE

6 CONCLUSIONS AND FUTURE WORK A prediction model of solar photovoltaic system using SVM has been proposed for Oman. This prediction was based on collected data from installed system in location called Sohar. Two inputs solar radiation and ambient temperature has been used and one output, photovoltaic current. The developed model has been evaluated using MSE. Results show that the predicting model is very close to the regression line and a well data fitting to the statistical model. The MSE found to be 2.6%, which is more accurate in comparison with some studies in literature. For the sake of reducing the MSE for the proposed network, alternative machine learning techniques need to be implemented for discovering the possibility of reducing the MSE rate like MLP with different hidden layers, SOFM, etc. The sensitivity analysis need to be implemented to determine the effects of the input variables on the output of network. ACKNOWLEDGMENT The research leading to these results has received Research Project Grant Funding from the Research Council of the Sultanate of Oman, Research Grant Agreement No. ORG SU EI The authors would like to acknowledge support from the Research Council of Oman. REFERENCES [1] Chaichan M T, Mohammed B A and Kazem H A, Effect of pollution and cleaning on photovoltaic performance based on experimental study, International Journal of Scientific and Engineering Research, vol. 6, No. 4, pp , [2] Al-Maamary H M S, Kazem H A, Chaichan M T, Changing the energy profile of the GCC States: A review, International Journal of Applied Engineering Research (IJAER), vol. 11, No. 3, pp , [3] Chaichan M T, Kazem H A, Kazem A A, Abaas Kh I, Al-Asadi K A H, The effect of environmental conditions on concentrated solar system in desertec weathers, International Journal of Scientific and Engineering Research, vol. 6, No. 5, pp , [4] Kazem H A, Al-Waeli A H A, Al-Mamari A S A, Al-Kabi A H K, Chaichan M T, A photovoltaic application in car parking lights with recycled batteries: A techno-economic study, Australian Journal of Basic and Applied Science, vol. 9, No. 36, pp , [5] Chen J L, Liu H B and Wu W, Estimation of monthly solar radiation from measured temperatures using support vector machines. Renewable Energy, vol. 36, No.1, pp , [6] Schölkopf B, and Smola A J, Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond, MIT Press, Cambridge, MA, USA, [7] Zhao P, Xia J, Dai Y and He J, Wind speed prediction using support vector regression Industrial Electronics and Applications (ICIEA), pp , [8] Zeng J and Qiao W, Short-term solar power prediction using a support vector machine, Renewable Energy, vol. 52, pp , [9] Banda d, Peña R, Gutiérrez G, Juárez E, Visairo N, Núñez C, Feasibility assessment of the installation of a photovoltaic system as a battery charging center in a Mexican Mining Company, Power, Electronics and Computing (ROPEC), 2014 IEEE International Autumn Meeting on Power, Electronics and Computing (ROPEC, 2014), 5 pages, [10] Caudill M and Butler C, Understanding neural networks: Computer explorations volume 1: Basic networks. The MIT Press, Cambridge, Mass, USA, [11] Sharma N, Sharma P, Irwin D and Shenoy P, Predicting solar generation from weather forecasts using machine learning. Proceedings of the 2011 IEEE International Conference on Smart Grid Communications, pp , [12] Hossain M R, Amanullah M T O, Ali A B M S, Hybrid prediction method of solar power using different computational intelligence algorithms. Proceedings of the 22nd Australasian Universities Power Engineering Conference, pp. 1-6, [13] Ministry of Tourism, [14] Chaichan M T, Kazem H A, Experimental analysis of solar intensity on photovoltaic in hot and humid weather conditions, International Journal of Scientific & Engineering Research, vol. 7, No. 3, 91-96, [15] Kazem H A, ChaichanM T, Saif S A, Dawood A A, Salim S A, Rashid A A and Alwaeli A A, Experimental investigations of dust type effect on photovoltaic systems in north region, Oman, International Journal of Scientific & Engineering Research, vol. 6, No. 7, pp , [16] H.A. Kazem, Renewable energy in Oman: Status and future prospects, Renewable and Sustainable Energy Reviews, vol. 15, pp , [17] H.A. Kazem, T. Khatib, Techno-economical assessment of grid connected photovoltaic power systems productivity in Oman, Sustainable Energy Technologies and Assessments, vol. 3, pp , [18] Kazem H A and Chaichan M T, Effect of humidity on photovoltaic performance based on experimental study, International Journal of Applied Engineering Research (IJAER), vol. 10, No. 23, pp , [19] Authority for Electricity Regulation in Oman (2008). Study on Renewable Resources. Final Report, Oman. [20] Public Authority for Electricity and Water [21]. Kazem H A, KhatibT and Sopian K, Sizing of a standalone photovoltaic/ battery system at minimum 10171

7 cost for remote housing electrification in Sohar, Oman. Energy and Building, vol. 6C, pp , [22] Kazem H A, Khatib T, Sopian K and Elmenreich W, Performance and feasibility assessment of a 1.4kW roof top grid-connected photovoltaic power system under desertec weather conditions, Energy and Building, vol. 82, pp , [23] Vapnik V, Statistical learning theory, Wiley, New York, [24] Hastie T, Tibshirani R and Friedman G, The elements of statistical learning, Springer, 2nd edition, [25] Pasolli L, MelganiF and Blanzieri E, Gaussian process regression for estimating chlorophyll concentration in subsurface waters from remote sensing data. IEEE Geosci. Remote Sens. Lett., vol.7,no. 3, pp , [26] Khatib T, Mohamed A, Mahmoud M, Sopian K, Modeling of daily solar energy on a horizontal surface for five main sites in Malaysia, International Journal of Green Energy, vol. 8, pp , [27] Khatib T, Mohamed A, Sopian K, Mahmoud M, Solar energy prediction for Malaysia using artificial neural networks, International Journal of Energy, vol. 6, No. 1, pp. 1-16, [28] Al-Shamisi M H, Assi A H, Hejase H A N, Artificial neural networks for predicting global solar radiation in Al-Ain City, UAE International Journal of Green Energy, vol. 10, pp ,