基于鹈鹕优化CNN-BiLSTM的电力负荷预测

Power Load Prediction Based on Pelican Optimized CNN-BiLSTM

为了提高电力负荷的预测精度,基于卷积神经网络(CNN)的空间特征提取能力、双向长短时记忆(BiLSTM)网络的时序预测性能以及鹈鹕优化算法(POA)的寻优能力,提出了一种新的基于CNN、BiLSTM、POA的组合电力负荷预测模型(POA-CNN-BiLSTM)。首先利用CNN提取电力负荷时间序列的特征向量,然后输入到BiLSTM网络进行双向循环训练,构建CNN-BiLSTM预测模型,并采用POA优化BiLSTM网络的隐藏层单元数、学习率和正则化系数等参数,最后输出电力负荷预测结果。将提出的模型应用于某区域电力负荷预测,结果表明,BiLSTM、LSTM模型预测精度优于最小二乘支持向量机(LSSVM)模型;BiLSTM模型预测精度优于LSTM模型;POA的寻优精度优于粒子群优化算法(PSO);CNN-LSTM、CNN-BiLSTM组合预测模型预测精度优于LSTM、BiLSTM模型;POA-CNN-BiLSTM模型预测精度优于POA-LSSVM、PSO-LSTM、POA-LSTM、POA-BiLSTM和POA-CNN-LSTM模型,能更好地追踪电力负荷的变化趋势。

In order to improve the accuracy of power load prediction,this paper proposes a new combined power load prediction model(POA-CNN-BiLSTM)based on the spatial feature extraction ability of convolutional neural networks(CNN),the predictive performance of bidirectional long short term memory(BiLSTM)networks in time series,and the optimization ability of the Pelican Optimization Algorithm(POA).Firstly,the feature vectors of the power load time series are extracted using CNN,and then it is input into the BiLSTM network for bidirectional cyclic training to construct a CNN-BiLSTM prediction model.The POA is used to optimize the parameters of the BiLSTM network,such as the unit number of hidden layer,learning rate,and regularization coefficient.Finally,the power load prediction results are output.The proposed model is applied to forecast the power load in a certain arera.The results show that the prediction accuracy of BiLSTM and LSTM networks is better than that of LSSVM;The BiLSTM has higher prediction accuracy than LSTM networks;The optimization accuracy of POA is superior to particle swarm optimization algorithms(PSO);The prediction accuracy of CNN-LSTM and CNN-BiLSTM models is better than that of a single LSTM or BiLSTM models;The POA-CNN-BiLSTM model has the best prediction accuracy compared to the POA-LSSVM,PSO-LSTM,POALSTM,POA-BiLSTM and POA-CNN-LSTM models,which can better track the change trend of power load.