基于DL的LSTM网络配电台区短期负荷预测模型研究

Research on short-term load forecasting model of LSTM network distribution area based on DL

传统基于神经网络的配电台区短期负荷预测研究对象往往是整个系统,缺少针对单个台区的建模预测研究,而且预测准确率不高。提出将基于深度学习(Deep Learning,DL)的长短期记忆神经网络(Long Short-Term Memory,LSTM)引入单配电台区的短期负荷预测中,根据配电台区的负荷特征,考虑工作日、月份和气象因素,以TensorFlow深度学习框架构建单配电台区的短期负荷预测模型,对其进行实验分析。实验结果表明,该预测模型在2层LSTM网络时训练效果最好,平均预测精度达到93.2%,表明该预测模型总体预测效果较好。

The traditional neural network-based short-term load forecasting research object of distribution area is often the entire system,which lacks the modeling and forecasting research for a single area,and the prediction accuracy rate is not high.Therefore,the paper proposes the Long Short-Term Memory(LSTM)based on Deep Learning(DL),and put it into the short-term load forecasting of a single distribution area.On the basis of the load characteristics of the distribution area,workdays,months and meteorologic factors are also taken into account,TensorFlow deep learning framework is used to construct a short-term load forecasting model for a single distribution area and conduct experimental analysis.Experiment results show that the prediction model has the best training effect when using a 2-layer LSTM network,with an average prediction accuracy of 93.2%,indicating that the overall prediction effect of the prediction model is better.