基于深度神经网络与权值共享的工业园区负荷预测  

Load forecasting in industrial park based on deep neural network and weight sharing

作　　者：王刚[1] 杨晓静[1] 张志军 刘丽新 于美丽 Abinet Tesfaye Eseye Wang Gang;Yang Xiaojing;Zhang Zhijun;Liu Lixin;Yu Meili;Abinet Tesfaye Eseye(State Grid Tianjin Electric Power Company,Tianjin 300090,China.;State Grid Tianjin ChengnanPower Supply Branch,Tianjin 300201,China.;Beijing Tsingsoft Technology Co.,Ltd.,Beijing 100085,China.;North China Electric Power University,Beijing 102206,China)

机构地区：[1]国网天津市电力公司,天津300090 [2]国网天津城南供电分公司,天津300201 [3]北京清软创新科技股份有限公司,北京100085 [4]华北电力大学,北京102206

出　　处：《电测与仪表》2021年第1期137-141,共5页Electrical Measurement & Instrumentation

基　　金：国家自然科学基金资助项目(59577060)。

摘　　要：电力体制市场化的有序推进对工业园区负荷预测提出了新的要求。文章提出了基于深度学习与权值共享机理的负荷预测方法。在预测模型中,将深度神经网络设置为训练中的有监督学习方法,权值共享模式分析了多个目标之间的相关性,并使用各个目标的负荷变化率对相关度最高的任务聚合。算例中使用天津某高新区数据对算法有效性进行了验证,结果显示该算法有效提高了工业园区负荷预测的精度,有着较高的应用价值。The orderly development of marketization of power system brings new requirements for load forecasting in industrial park.This paper proposes a load forecasting method for industrial park based on deep learning and weight sharing.In forecasting models,the deep neural network is regarded as supervised-learning approach,weight sharing is deployed to analyze the correlation among various objectives,and the most related objective task is selected by load change rate.The validity of the algorithm is verified through the simulations performed by actual operating data from industrial park load system in Tianjin.The positive results demonstrate that the proposed algorithm can effectively improve the accuracy of load prediction and has high application value.

关 键 词：工业园区负荷预测 深度学习 权值共享 任务聚合 

分 类 号：TM714[电气工程—电力系统及自动化]