Design of Smart Energy Generation and Demand Response System in Saudi Arabia

This thesis was submitted for the award of Doctor of Philosophy and was awarded by Brunel University LondonThough the promising benefits of renewable sources have already pushed many countries into implementing RE units, Saudi Arabia is still highly dependent on fossil fuel. However, the decrease in value of oil reserves has enforced Saudi Arabia to prioritize renewable energy sources in the next decades. Such energy sources are highly dependable on accurate forecasts, due to their intermittence and operability. The present research has the objective to develop models that can accurately forecast energy load for implementation in a decision-making system. The case investigated is the western region of Saudi Arabia. Two modelling approaches were evaluated, linear regression (LR) and artificial neural network (ANN). This last one was chosen because it is a mathematical model able to deal with non-linear relationship among input(s) and output(s) in the data. Time series (past load data) and multivariate data from 2010 until 2016 were investigated A hybrid model structure (combiner) was implemented to analyse the effects of combining outputs of two models in a single one. This hybrid model consisted of a regular average and weighted average of the time series and multivariate model, with calibration through Fuzzy and Particle Swarm Optimisation. These two were selected because, while Particle Swarm Optimization is an optimization algorithm, Fuzzy consists in a complete structured model. The forecasted load and the available input were used in the last chapter for power generation planning and decision-making support. The software used for the modelling and simulation is ETAP®. Different scenarios for replacement of fossil fuel power plants by renewable units were tested considering the network of western Saudi Arabia. The results show that Artificial Neural Network with time series input and 15 neurons in hidden layer shows superior performance (MSE 3.7*105 and R2 equals 99%) compared to other neural networks and linear regression. Though the application of combiner models did not significantly improve model performance, the Fuzzy Combiner shows the best one (MSE 5.8*105 and R2 equals 93%) since it incorporates information from time series and multivariate data. It is important to mention that all the modelling approaches evaluated have some limitations, such as the necessity of accurate input data and they are limited in capability of extrapolating over the training range. In the last section, it was observed that renewable energy sources can be integrated in the grid network without excessive risk regarding demand. This occurs because the current energy management policy of western Saudi Arabia enables the use of energy units with fast compensation (using gas units) in the case of demand increase or decrease in solar or wind power