基于变量相关性分析的LSTM网络多步预测

Multi-step Prediction of LSTM Network Based on Variable Correlation Analysis

为了提高光伏发电的有效预测长度和精度,提出了一种基于变量相关性分析的改进LSTM网络多步预测方法。首先,利用R/S分析法计算各变量的赫斯特指数,剔除本身不具有相关性的变量,再采用灰色关联法计算各变量与发电量的关联度,进一步剔除与光伏发电量关联度小的变量;然后,对变量数据进行归一化预处理,构建改进LSTM网络对光伏发电量进行多步预测;最后,通过光伏发电量多步预测仿真图和均方误差结果,证明了基于变量相关性分析的改进LSTM网络多步预测的有效性。

In order to improve the effective prediction length and accuracy of PV power generation,an improved LSTM network multi-step prediction method based on variable correlation analysis method is proposed.First,the R/S analysis method is used to calculate the Hurst exponent of each variable,excluding variables that are not relevant with themselves.Next,the gray correlation method is adopted to calculate the correlation between each influencing variable and the amount of power generation,and further eliminate thevariables with a small degree of correlation with the photovoltaic power generation.Then,the normalized preprocessing of variable data is carried out and an improved LSTM network is constructed to predict the PV power generation in multi-step.Finally,through the multi-step prediction simulation of PV power generation and the mean square error results,the effectiveness of the improved LSTM network multi-step prediction based on the correlation analysis of variables is proved.