基于贝叶斯正则优化NARX神经网络的电力负荷预测

Power load forecasting based on Bayesian regularization optimized NARX neural network

随着智能电网建设阶段的不断推进,以及电力市场运营机制的逐渐完善,电力负荷预测工作的重要性与日俱增。针对传统的前馈型神经网络预测模型存在泛化能力不强、预测精度较差的问题,设计了一种基于贝叶斯正则优化NARX(Bayesian regularization nonlinear autoregressive with exogenous inputs,BR-NARX)神经网络的电力负荷预测方法。介绍了NARX神经网络以及贝叶斯正则优化算法的理论分析与模型搭建;基于某省电网公司的历史负荷数据,以及供电区域内相应的气象信息等数据,进行数据分析及预处理工作;以实际应用算例测试,分析所构建BR-NARX神经网络的模型性能,并与BP、BR-BP、NARX3种神经网络模型进行预测效果的对比。算例结果显示,BR-NARX模型具备显著提升电力负荷预测精度的性能,能够为实际电力负荷预测工作的进一步发展提供思路。

With the continuous development of smart grid construction and the gradual improvement of power market operation mechanism,the importance of power load forecasting is increasing day by day.In view of the problems of weak generalization ability and poor prediction accuracy of the traditional feedforward neural network prediction model,a power load forecasting method based on Bayesian regularization non-linear auto-regressive with exogenous inputs(BR-NARX)neural network was proposed.The theoretical analysis and model building of NARX neural network model and Bayesian regular optimization algorithm are introduced.Based on the historical load data of the actual power grid company in a province and the corresponding meteorological information data in the power supply area,data analysis and preprocessing are carried out.Through the actual example test,the model performance of the BR-NARX neural network is analyzed and compared with BP,BR-BP and NARX neural network models.The calculation results show that BR-NARX model has the performance of significantly improving the accuracy of power load prediction,and can provide ideas for the further development of actual power load prediction.