计及误差修正的变分模态分解-长短期记忆神经网络短期负荷预测

Short-term Load Forecasting with Error Correction and Variational Mode Decomposition-Long Short-term Memory

精确地短期负荷预测为电力系统经济调度和机组最优负荷分配交易奠定基础。因此,提出了一种将变分模态分解(variational mode decomposition,VMD)和长短期记忆神经网络(long short-term memory,LSTM)结合的短期负荷预测模型,并使用支持向量回归(support vector regression,SVR)构建修正后的误差序列对初始预测序列补偿。首先,运用VMD算法将非平稳的负荷序列分解为多个相对平稳的模态分量;然后,将每个模态分量输入LSTM模型进行预测,并将各分量预测结果合并得到VMD-LSTM的预测结果;最后将残差值输入SVR模型中构造误差序列,来修正后一日的VMD-LSTM预测结果。通过实际案例测试,实验结果对比其他模型结果有更低的预测误差,证明所提方法的有效性。

Short-term load forecasting lays the foundation for the economic dispatch of the power system and optimal load distribution of units.Therefore,a short-term load forecasting model combined variational modal decomposition(VMD)and long short-term memory neural network(LSTM)was proposed,and support vector regression(SVR)was used to construct a revised error sequence to compensate the initial prediction sequence.Firstly,the VMD algorithm was used to decompose the non-stationary original load sequence into multiple relatively stable modal components.Then,each modal component was input to the LSTM model for prediction,and the prediction results of each component were combined to obtain the prediction result of VMD-LSTM.Finally,the residual value of the model was input into the SVR model to construct an error sequence to correct the VMD-LSTM load forecast results of the next day.Through actual case tests,the experimental results have lower prediction errors compared with the results of other models,which proves the effectiveness of the proposed method.