基于Stacking模型集成的LSTM网络短期负荷预测研究

Short-term load forecasting of LSTM network based on Stacking model integration

为解决传统负荷预测方法存在的预测精度偏低的问题,通过分析短期负荷影响因素确定训练集,创建Stacking模型,并结合包括输入门、输出门与遗忘门在内的LSTM网络创建Stacking-LSTM混合模型,通过时间滑动窗口建立影响因素数据特征图,将其作为Stacking-LSTM混合模型的输入,经数据转换后得到特征类别更强的降维二级特征数据,输入到LSTM网络层实现短期负荷预测。该方法利用Stacking模型的集成作用和LSTM网络的强挖掘能力,增强降维后的数据类别特征,达到提升电力系统负荷动态平衡性的效果。仿真结果表明,该方法的负荷预测结果与实际值非常接近,具有较高的预测精准度。

To solve the low accuracy in traditional load forecasting methods,this paper proposed a method through the analysis of short-term load factors determine the training set,create Stacking model,and combined with input,output,door and door left the door of the LSTM network create Stacking-LSTM hybrid model.Affecting data characteristic chart was established through moving time window,as a Stacking-LSTM mixture model input,getting feature category stronger data dimension reduction secondary characteristics after data transformation,and inputted to the LSTM network layer to achieve short-term load forecasting.This method utilized the Stacking function of the Stacking model and the strong mining ability of the LSTM network to enhance the data class characteristics after dimension reduction,improving the dynamic balance of power system load.The simulation results show that the load prediction results of this method are very close to the actual value and have high forecasting accuracy.