Power Load Forecast Based on Fuzzy BP Neural Networks with Dynamical Estimation of Weights

Abstract

The short-term power load forecast is deeply studied by integrating fuzzy BP neural networks and composite adaptive filtering in this paper. Due to the difficulty on accurate modeling of complex factors in smart grid, the fuzzy technology is introduced to deal with the uncertain factors. Meanwhile, data-driven method, such as neural networks, is used as the basic frame of the power load forecast by combing the fuzzy technology. Thereby, to design a highly effective training scheme becomes the main problem on the weights of the BP-NNs. To realize better weights training, the combined filtering technology is fully adopted by fusing adaptive filtering methods based on observable degree (OD) analysis, Sage–Husa adaptive technology, and variational Bayesian method. Thereby, a novel power load forecast algorithm is proposed based on the fuzzy BP-NNs and the combined adaptive cubature Kalman filter. Experiment based on practical power load data is presented to validate the effectiveness of the proposed short-term power load algorithm.