摘要
针对空间负荷预测的影响因素多样及历史数据匮乏的问题,提出了一种考虑多维特征和数据增强的空间负荷预测方法。该方法首先综合考虑多种影响电力负荷的因素,从开发强度、发展水平、气候条件建立地区多维度指标模型。然后构建生成对抗网络(generative adversarial networks,GAN)的数据生成模型,对训练集进行数据增强,生成数量充足且符合地区特点的训练样本。其次采用基于粒子群算法(particle swarm optimization,PSO)优化初始权重和阈值后的反向传播(back propagation,BP)神经网络建立空间负荷预测模型,并利用增强后的数据集实现空间负荷预测。最后,以东部某市4个区为例,对本文的方法进行验证,仿真结果表明本文提出的方法可以提高空间负荷预测精度,具有实用性和有效性。
Aiming at the problems of various influencing factors and lack of historical data in spatial load forecasting,a spatial load forecasting method considering multi-dimensional feature and data enhancement was proposed.Firstly,considering various factors affecting power load,a regional multi-dimensional index model was established from the development intensity,development level and climate conditions.Then the data generation model of generative adversarial networks(GAN)was constructed to enhance the training set and generate a sufficient number of training samples in line with regional characteristics.Secondly,particle swarm optimization(PSO)algorithm was used to optimize the initial weight and threshold back propagation(BP)neural network to establish the spatial load forecasting model,and the enhanced data set was used to realize the spatial load forecasting.Finally,taking four districts of an eastern city as an example,the method was verified.The simulation results show that the method proposed can improve the accuracy of spatial load forecasting,and has practicability and effectiveness.
作者
黄冬梅
张宁宁
胡安铎
孙园
孙锦中
陈岸青
HUANG Dong-mei;ZHANG Ning-ning;HU An-duo;SUN Yuan;SUN Jin-zhong;CHEN An-qing(School of Electronics and Information Engineering,Shanghai University of Electric Power,Shanghai 201306,China;College of Mathematics,Shanghai University of Electric Power,Shanghai 201306,China;State Grid Info-Telecom Great Power Science and Technology Co.,Ltd.,Fuzhou 350001,China)
出处
《科学技术与工程》
北大核心
2022年第30期13330-13337,共8页
Science Technology and Engineering
基金
上海市科委地方院校能力建设项目(20020500700)。
关键词
多维特征
GAN
BP神经网络
空间负荷预测(SLF)
multidimensional features
generative adversarial networks(GAN)
BP neural network
spatial load forecasting(SLF)
作者简介
第一作者:黄冬梅(1964—),女,汉族,河南郑州人,硕士,教授。研究方向:海洋与电力时空信息技术。E-mail:dmhuang_dl@163.com;通信作者:胡安铎(1983—),男,汉族,河南信阳人,博士,讲师。研究方向:电力时空信息技术。E-mail:huanduo2003@163.com。
