基于CNN-GRU神经网络的短期负荷预测

Short-term Load Forecasting Based on CNN-GRU Neural Network

提出了一种基于CNN-GRU(convolutional neural networks-gate recurrent unit)神经网络的电力系统短期负荷预测方法。首先使用卷积神经网络(CNN)对负荷及气象数据进行卷积处理,以更好地提取数据新特征,增强输入数据与输出数据间的相关性。然后使用门控循环单元(GRU)实现短期负荷预测。使用某地区的负荷数据结合当地的气象数据,对CNN-GRU方法进行了测试。结果表明:与单独的CNN网络或GRU网络相比,CNN-GRU网络对电力系统短期负荷的预测误差更小,预测精度更高。

A short-term load forecasting method based on CNN-GRU(convolutional neural networks-gate recurrent unit)neural network is presented in this paper.Firstly,convolutional neural networks(CNN)is used to process the load and weather data,so as to better extract new features of data and enhance the correlation between input data and output data.Then the short-term load forecasting is realized by using the gate recurrent unit(GRU).The CNN-GRU method is tested by using the real load data and local meteorological data.The results show that compared with the CNN network and GRU network,the prediction error of CNN-GRU network for short-term load of power system is smaller and the prediction accuracy is higher.