优化神经网络用电量预测性能的多元线性回归方法

电网电能难以储存,其准确预测可用于指导发电计划,是电力系统运行的基础。针对用电量预测中需求波动幅度大与需求变动趋势不确定的特点,基于已有的用电数据,运用多元线性回归预测与时间序号、月度、日期、小时等因素直接相关的用电量趋势,实现用电量的宏观量级控制。其次,采用不同神经网络预测实际值与趋势值之间的偏移量,推演用电量的微观波动规律,进一步推算预测用电量。算例证明,融合多元线性回归和神经网络的预测方法减小了总体预测误差,其预测精度高于没有多元线性回归处理的神经网络。

Multiple Linear Regression Method for Optimizing Performance Prediction of Neural Network

When operating power grid,it is extremely hard to store energy and thus an accurate electricity demand forecast method is required as the basis of power scheme.Considering high fluctuation and changing trends in electricity demands,multiple linear regression prediction on the ground of existing data is used to forecast the tendency and to master the macroscopic fluctuation pattern day by day of the electricity consumption,since this tendency is directly related to time series with different timescales.Furthermore,different neural network is used to predict the offset between the actual value and the tendency value and to understand the microscopic fluctuation rule in a day.In this way,the electricity consumption at every quarter is predicted.Experimental results prove that the prediction method combining multiple linear regression and neural network reduces the overall prediction error and its prediction accuracy is higher than neural network without multiple linear regression processes.