基于集成BP神经网络的短期负荷预测

利用混沌相空间重构理论对负荷时间序列研究,用改进的C_C方法求得时间延迟τ和嵌入维数m,得到系统最大李雅普诺夫指数,证明其具有混沌特性.对样本数据相空间重构,构建多个BP神经网络的预测子模型,所有子模型同步预测的加权平均作为集成负荷预测值.在线采集负荷数据,利用增量式训练获取新的预测子模型,按“先入先出”顺序进行BP神经网络集成更新.将预测结果同普通BP神经网络预测结果进行对比,结果证明这种方法提高了预测精度.

Short-term power load prediction based on integrated BP neural networks

The chaotic phase space reconstruction theory is used to study the power load time series in this paper By using the improved C_C method, the time delay r and embedding dimension m are obtained, maximal Lyapunov exponent is obtained, and it is proved that the system haschaotic characteristics. Based on the time delay and the embedding dimensiom the phase space of sample data is reconstructed, multiple synchronous prediction sub-modes of BP neural networks are built, and the weighted average of all synchronous prediction value is used as the prediction value of integrated load. Load data are collected online and incremental training is implemented to obtain new prediction sub-modes. The integration update of BP neural networks are completed following the principle of ＂first-in, first-out＂. The prediction results are compared with those of ordinary BP neural network modes and the results prove that the method make the prediction accuracy increased