Short-term load forecasting using neural network with principal component analysis

A neural-network-based approach for short-term load forecasting of electrical power is proposed. The principal component analysis (PCA) technique is used to reduce the original electric load variables to several characteristic variables. A single parameter dynamic search algorithm (SPDS) is employed to train the neural network. Since the training sample sets can be chosen before forecasting, the interference of the non-correlative samples for the forecasting can be avoided. The effectiveness and the feasibility of on line forecasting of the proposed method are examined using simulated experiments.