摘要
为了增强电力调度的智能化程度,文中基于深度学习理论对用电负荷的分析与预测方法进行研究。对玻尔兹曼机(RBM)中的能量传递机制进行研究,将低层次的RBM作为高层次的输入搭建深度置信网络(DBN),实现电力负荷数据深层次的特征提取。通过引入一种基于信息熵的DBN隐藏层节点数确定方法,使用对比散度法完成网络训练。为了验证算法的有效性,使用EUNITE的开放数据集进行对比仿真实验。结果表明,在相同的时间复杂度下,BP网络与DBN网络在训练集上的相对误差分别为4.36%与4.21%;从训练集与测试集误差的变化上看,BP网络在测试集上的相对误差分别为9.64%与4.88%,相对误差大幅提升,而DBN网络则保持了训练集与测试集上相对误差的一致性。这说明DBN网络在用电负荷的预测上,具有较强的泛化能力。
In order to enhance the intelligent degree of power dispatching,this paper studies the analysis and prediction method of power load based on the deep learning theory.The mechanism of energy transfer in the Boltzmann machine(RBM)is studied,and a deep belief network(DBN)is constructed by taking the low-level RBM as the high-level input to realize the deep feature extraction of power load data.By introducing a method to determine the number of nodes in the hidden layer of DBN based on information entropy,the network training is completed by using contrast divergence method.In order to verify the effectiveness of the algorithm,a comparative simulation experiment is carried out with EUNITE’s open data set.The results show that under the same time complexity,the relative errors of BP network and DBN network on the training set are 4.36%and 4.21%respectively;from the change of the training set and test set errors,the relative errors of BP network on the test set are 9.64%and 4.88%respectively,and the relative errors are drastically increased,and DBN network keeps the consistency of relative error between training set and test set,which shows that DBN network has stronger generalization ability in the prediction of power load.
作者
左松林
梁晓伟
张靖
ZUO Songlin;LIANG Xiaowei;ZHANG Jing(Marketing Service Center,State Grid Anhui Electric Power Co.,Ltd.,Hefei 230088,China;Information and Communication Branch,State Grid Anhui Electric Power Co.,Ltd.,Hefei 230009,China)
出处
《电子设计工程》
2021年第4期43-47,共5页
Electronic Design Engineering
基金
国家电网科技项目(2018ZR21137)。
