基于CEMD-CNN-LSTM的中长期电力负荷预测

Medium and Long Term Power Load Forecasting Based on CEMD-CNN-LSTM

针对诸多复杂因素影响电力负荷在中长期运行阶段中的预测准确度的问题,提出一种卷积神经网络(CNN)与长短期记忆网络(LSTM)混合的预测算法,从而达到优化负荷预测性能的目的。CNN-LSTM混合预测算法利用模态分解法将负荷数据进行分解,并将其转化为本征模态分量IMF及残差两个部分。同时,引入k均值聚类方法,确定最优聚类标签,搭建神经网络并完成数据输入。在形成特征向量的过程中,运用神经网络挖掘数据间的各类特征并进行预测。最后,采用线性相加的形式处理预测结果,获取预测负荷。仿真结果表明了CNN-LSTM混合预测算法在预测速度与精度上的性能更为优越。

There are many complex factors in load,which will threaten the prediction effect of power load in medium and long terms operation stage.In order to solve the practical bottleneck,this paper proposes a hybrid prediction algorithm of CNN-LSTM.The algorithm is based on improving clustering empirical mode decomposition(CEMD),so as to optimize the efficiency of load prediction.CNN-LSTM hybrid prediction algorithm uses empirical mode decomposition method to decompose the load data and convert it into two parts:intrinsic mode function IMF and residual.At the same time,the k-means clustering method is introduced to determine the optimal clustering label,build a neural network and complete the data input.In the process of forming feature vectors CNN is used to mine all kinds of features among data for prediction.Finally,the prediction results are processed in the form of linear addition to obtain the predicted load.The simulation results show that the CNN-LSTM hybrid prediction algorithm proposed in this paper has better performance in prediction accuracy.