Reliable power load forecaster by neural networks with a modified FBP learning algorithm

摘要

In this paper, a more reliable power load forecaster by using neural networks with fuzzy back-propagation (FBP) learning algorithm based on system's dynamic behaviors is developed. Since the parameters of neural model are adaptively regulated in accordance with the immediate system's dynamic behaviors, such that the forecaster we developed has an efficient learning capability and has a more accurate performance in real line power load forecasting. For a comparison, the neural network forecasting models by using conventional BP learning algorithm with different pairs of constant learning rates (α = 0.1 ～ 0.9) and momentums (ξ = 0.1 ～ 0.9) are also experimented.