摘要
针对电力系统负荷受天气、季节、工作日、周末和节假日等多种不确定外部因素影响的问题,在考虑这些外部因素的情况下,建立一个高精度的时负荷预测模型。首先,利用卷积神经网络(convolutional neural network,CNN)将负载负荷数据作为输入,构建由一维卷积层、池化层和全连接层组成的卷积神经网络提取反应负荷的高维特征;其次,利用负载序列的自相关系数来确定卷积层的核大小,卷积层通过应用二维卷积核将一维特征转换为多维特征,将二维CNN提取的多维特征作为输入,输入到双向门控循环单元(gated recurrent unit,GRU)和双向长短期记忆网络(long short-term memory,LSTM)单元,并引入注意力机制赋予双向GRU和LSTM隐藏状态不同的权重输入到全连接层,进行逐小时电力负荷预测;最后,通过负荷数据集验证该模型。结果表明:模型具有一定的优越性,是一个可行的负荷预测。
Considering that the power system load is affected by a variety of uncertain external factors such as weather,seasons,working days,weekends,holidays,etc.,a high-precision hourly load forecasting model was established taking these external factors into consideration.The outline of this load forecasting method was as follows:firstly,use the convolutional neural network(CNN)to take load data as input to construct a CNN consisting of a one-dimensional convolutional layer,a pooling layer,and a fully connected layer,and extract high-dimensional features of reaction load;secondly,the autocorrelation coefficient of the load sequence was used to determine the kernel size of the convolutional layer,the convolutional layer converts one-dimensional features into multi-dimensional features by applying a two-dimensional convolution kernel,and take the multi-dimensional features extracted by the two-dimensional CNN as inputs to the two-way gated recurrent unit(GRU)and the bidirectional long short-term memory(LSTM)unit.And the attention mechanism was introduced to give different weights of the bidirectional GRU and LSTM hidden states input to the fully connected layer for forecasting hourly power load;finally,verify the model through the load data set.The results show that the model has certain advantages and is a feasible load forecasting.
作者
王季
李润清
刘屾
曹万水
王昊
陈勇
Wang Ji;Li Runqing;Liu Shen;Cao Wanshui;Wang Hao;Chen Yong(Gansu Academy of Mechanical Sciences Co.,Ltd.,Lanzhou Gansu 730030,China;School of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou Gansu 730050,China;Key Laboratory of Advanced Control of Industrial Processes in Gansu Province,Lanzhou Gansu 730050,China)
出处
《电气自动化》
2022年第4期61-63,共3页
Electrical Automation
基金
国网甘肃省电力公司科技项目(26271220003T)。
关键词
短期负荷预测
卷积神经网络
长短期记忆网络
门控循环单元
自相关系数
short-term load forecasting
convolutional neural network(CNN)
long short-term memory(LSTM)
gated recurrent unit(GRU)
autocorrelation coefficient