基于RNN-RBM模型的配网馈线长期负荷预测方法  

Long-Term Load Forecasting Method of Distribution Network Feeder based on RNN-RBM Model

作　　者：杨剑文 朱林[1] 林凌雪[1] 吴子龙 陈元榉 陈展纶 YANG Jianwen;ZHU Lin;LIN Lingxue;WU Zilong;CHEN Yuanju;CHEN Zhanlun(School of Electric Power,South China Univ.of Technology,Guangzhou 510640;Guangzhou Power Supply Bureau of Guangdong Power Grid Co.,Ltd.,Guangzhou 510600,China)

机构地区：[1]华南理工大学电力学院,广州510640 [2]广东电网有限责任公司广州供电局,广州510600

出　　处：《三峡大学学报（自然科学版）》2022年第3期67-73,共7页Journal of China Three Gorges University：Natural Sciences

基　　金：南方电网公司重点科技项目(YNKJXM20191215)。

摘　　要：本文提出了一种基于循环神经网络(recurrent neural network,RNN)和受限玻尔兹曼机(restricted boltzmann machine,RBM)混合模型的配网馈线长期负荷预测方法.所提方法首先提炼混合模式(自上而下和自下而上)下的馈线负荷特征,然后利用RNN网络处理具有时序特点的配网馈线负荷数据,识别历史负荷数据的变化规律,再利用RBM深度挖掘数据中的特征规则,最后以无监督训练的方式获得特征与负荷变化的内在联系.选取广州某地区配网馈线数据验证所提算法的有效性,并与随机森林模型以及LSTM模型进行了比对.结果表明,本文所提出的混合RNN-RBM模型提高了配网馈线长期负荷预测的准确率.A long-term load forecasting method of distribution network feeders based on a hybrid model of recurrent neural network(RNN)and restricted boltzmann machine(RBM)is proposed in this paper.The load features for feeder data under a combination of top-down and bottom-up designs are fully extracted.Next,the hybrid RNN-RBM model is build.The RNN is utilized to process the load data with time series characteristics in the distribution network and the characteristics of historical load data are identified.The RBM is applied to mine the rules and the inherent relationship between characteristics and load changes in unsupervised learning is analyzed.The effectiveness is verified with actual distribution network operation data in Guangzhou.The results show that the proposed RNN-RBM model is superior to long-short term memory(LSTM)and random forest(RF),and the accuracy of long-term load forecasting is improved.

关 键 词：长期负荷预测 数据驱动 无监督学习 配网馈线 

分 类 号：TM715.1[电气工程—电力系统及自动化]