基于CNN-GRU混合神经网络的负荷预测方法

Load Forecasting Method Based on CNN-GRU Hybrid Neural Network

为充分挖掘负荷数据中时序性特征的联系,提高负荷预测的精度,提出了一种基于卷积神经网络(convolutional neural networks,CNN)和门控循环单元(gated recurrent unit,GRU)混合神经网络的负荷预测方法。以日期因素、气候因素、相似日负荷因素构建特征集作为输入,首先采用k-means聚类方法对地区内的样本数据集进行分组;再运用CNN网络提取特征与负荷在高维空间的联系,构造时序序列的高维特征向量,并将结果输入到GRU网络中;最后训练各组GRU网络模型的参数并输出负荷预测值。使用该方法对浙江省某地区电力负荷数据进行预测,结果表明,所提负荷预测方法与长短期记忆(long short-termmemory,LSTM)网络模型、GRU网络模型、CNN-LSTM网络模型、支持向量机回归模型及决策树模型相比,在预测精度与预测效率方面具有显著优势。

In order to fully exploit the relationship between temporal features in load data and improve the accuracy of load-forecasting results,this paper proposes a load-forecasting method based on a mixed neural network model of convolutional neural network(CNN)and gated recurrent unit(GRU).This method takes date factor,climatic factor and similar daily load factor as the input feature sets.Firstly,the k-means clustering method is used to divide the sample data set into groups.Then the CNN network is used to extract the relationship between the features and the load data in the high-dimensional space,constructing the feature vector of the time series.Then the result is put into the GRU network.Finally,the parameters of each group in the GRU network models are trained and the load prediction values are obtained.This method is used to predict the electric load data of a certain area in Zhejiang Province.The result shows that the proposed load prediction method has significant advantages on prediction accuracy and prediction efficiency,compared with the long and short term memory(LSTM)network model,the GRU network model,the CNN-LSTM network model,the support vector machine regression(SVR)and the decision tree model.