Classification and Prediction Model of Renewable Energy Depending on Weather Factors Using Artificial Intelligence Techniques رسالة ماجستير

Abstract

For many decades, renewable energy resources are used as a sustainable and environmentally friendly resource to compensate for the shortage of conventional energy resources. Solar and wind energy systems are the most popular renewable energy resources that have been used to produce electricity. These systems have fluctuating nature due to the dependency on weather conditions. Accurate prediction of the potentially available power in a specific location is economically feasible for designing and installing renewable power systems. For this reason, an accurate prediction and classification model is necessary for the operation of this sector in Palestine and other countries. This thesis proposes a hybrid model of Neural networks (NNs) and Optimization Algorithms (OAs) that uses the historical data of solar cells to predict the potential production of different solar energy farms installed in the main cities of Palestine. Also, meteorological data is used to calculate and predict the potential wind energy that can be obtained from wind fields in Palestine. The data sets of historical power production were collected from different locations in Palestine. these locations include Hebron in the south, Bethlehem and Salfit in the middle, and Tubas and Jenin in the north of the West Bank. The metrological data are collected using the Palestinian Meteorological Department for the cities of Jenin, Nablus, Bethlehem, and Hebron. The proposed model combines different types of NNs and OAs to enhance prediction results with the aim of training and testing the model to enhance the prediction result of the electrical energy production; using historical data of solar cells and weather data, especially solar radiation, temperature, and wind speed. The performance of the proposed model shows that the hybrid model system VI that combines GAs with Multi-Layer Perception neural network (MLPNNs) outperforms the Particle Swarm Optimization (PSO) algorithm with MLPNNs model and Radial basis function Neural Network (RBFNNs) with GAs of predicting power production. GAs-MLPNNs achieved the lowest MSE among the other models. For instance, with 40 neurons, the MSE using the proposed model was 0.0048, where the MSE for GAs-RBFNNs was 0.0094 using a complex dataset of four years; while the PSO-MLPNNs model failed to handle the dataset. The GAs-MLPNNs model also demonstrate that better results can be obtained using the metrological dataset rather than the use of the historical dataset, where the MSE using the historical dataset was double that for results of the metrological dataset. The results of the model used for classifying the geographical regions depending on their potential power production. We employed a general classification algorithm; the self-organizing map (SOM) using the Kohonen neural network to classify the regions based on the predicted annual power production. The classification results demonstrate that Jenin, Tubas, and Bethlehem have similar potential solar power production, while Nablus, Salfit, and Hebron have diverse capacities of electricity production from the solar resource. The results also show a variety of wind power production for each city of Jenin, Nablus, Bethlehem, and Hebron.