Load and locational marginal pricing prediction in competitive electrical power environment using computational intelligence

In a deregulated power systems environment, the establishment of electric markets has resulted in more attention to electric load and price forecasting. These two issues are considered most essential as the basis for decision making in the power system. Short term load forecasting is essential to short term scheduling of generating and operation of electric utilities. On the other hand, forecasting price such as locational marginal pricing (LMP) plays a dominant part in wholesale electric market by enhancing the efficient and reliable operation of the bidding process. As a consequence, a technique for solving Load and LMP forecasting problems are addressed in this thesis as real problems in the deregulated electricity environment.Through this thesis, a series of studies have been addressed in an attempt to assess the convergence behavior of the adopted model and to improve the efficiency of its performance. Many of these investigations are concerned with the design of the optimal network architecture, selecting suitable input data, extracting redundant information from the original data to obtain regularity of the data using the wavelet transform and finally, choosing the proper parameter values for the PSO algorithm.The generalized error estimation is done by using the reverse part of the data as a test set. The effectiveness of the proposed model is demonstrated through the forecasting of the load and LMP. Quantifying the prediction accuracy and validity in terms of meaningful measures has been done. The simulated and numerical calculation result of adopted model showed better result in comparison with the BP algorithm and radial basis function network (RBF).Artificial neural networks (ANNs) have high potential for solving nonlinear prediction problems where the network can extract the implicit nonlinear relationship among its input variables by learning from training data. A conventional training method such as the back propagation algorithm (BP) using gradient methods has the disadvantages of slow convergence and sensitivity to initial guess values which could possibly result in trapped into local minima. Therefore, to overcome the disadvantages of this technique, a heuristic technique called particle swarm optimization (PSO) is presented as the training phase for forecasting of one-day ahead of the load demand and LMP in New York Independent System Operator (NYISO) electric market environment. Thus, PSO is applied in this research as a global optimization technique and it is investigated as an alternative approach for neural network training so as to achieve our goal which is to create a trained neural network with higher generalization accuracy and faster convergence speed.