The investigation of the effect of net head variation on active power output in hydroelectric power plants

In this thesis, a different approach using artificial neural networks (ANN) to reduce active power generation imbalances of hydroelectric power plants (HEPP) with narrow reservoir volume and small net head range is presented. In this study, firstly hydroelectric power plants and turbine types used in power plants and then the control methods used in speed regulators are explained briefly. In addition, some environmental impacts that negatively affect active power generation, such as net head fall in hydroelectric power plants, are mentioned. The Feed Forward Back Propagation ANN was used for the HEPP to be applied. Since the synchronous generators operating in parallel in HEPP will affect each other, the designed ANNs are trained separately for each synchronous generator. The mean square error rate (MSE) was used as a performance function in all network structures. Supervisory Control and Data Acquisition (SCADA) data was normalized using the minimum-maximum normalization method. After normalization, these data have been used as ANN training and test data. The best results were tested with different training functions, and it was aimed to obtain the best possible results by training with different ANN models. For this purpose, the first synchronous generator has been modelled as multi-layered network with the performance function of MSE, the Trainlm as training function, the logarithmic-sigmoid function in the hidden layers and the linear activation function in the output layer. The best results for the second synchronous generator topology have been attained with multilayer ANN, having performance function MSE, Trainlm as training function, tangent-sigmoid function in hidden layers and tangent-sigmoid activation function in the output layer. The best results which have been obtained by testing the different network structures are presented in tables and finally the results are discussed.