Abstract of Thesis presented at COPPE/UFRJ as a partial fulfillment of the requirements for the degree of Master of Science (M.Sc.)

Regularization Techniques of Neural Models Applied to Short-Term Load Forecasting

Vitor Hugo Ferreira

February/2005

Advisor:

Alexandre Pinto Alves da Silva

Department:

Eletrical Engineering

      The knowledge of loads' future behavior is very important for decision making in power system operation and planning. During the last years, many short term load forecasting models have been proposed, and feedforward neural networks have presented the best results. One of the disadvantages of the neural models is the possibility of excessive adjustment of the training data, named overfitting, degrading the generalization performance of the estimated models. This problem can be tackled by using regularization techniques. The present Thesis investigates the application of promIsmg procedures for complexity control of short term load forecasting neural models.