Several activities of planning and operation in power systems rely on knowledge of early and accurate demand of electric load. For this reason, power generation and distribution companies are increasingly using technologies for load forecasting. These estimative may have a very short, short, medium or long-term horizon. Numerous statistical methods have been used for the problem of prediction. All these methods work well under normal conditions, but fail in situations where unexpected changes in the parameters of the environment occur. Currently, techniques based on Computational Intelligence have been presented in the literature with satisfactory results for the problem of load forecasting.Considering then the importance of load forecasting for the electric power systems, in this thesis a new approach to the load forecasting problem is evaluated by Auto-Associative Neural Networks and Genetic Algorithms. Three models based on Computational Intelligence are also presented with their performance evaluated and compared with the proposed system. With the obtained results, it was found that the proposed model is satisfactory for the problem of forecasting, thereby strengthening the applicability of computational intelligence methodologies to the problem of load prediction.
|
