摘要
电力系统超短期负荷预测易受到气象、假日等多种因素共同作用的影响,因此,实现其精准预测较为困难。为提高预测精度,往往需要大量的历史数据进行训练。针对历史数据较少的新建初期电力系统,提出了一种基于随机分布式嵌入框架及BP神经网络的超短期电力负荷预测方法。首先,将电力系统中电力负荷变量、气象变量等各种状态变量的延迟变量视为独立的影响因素,采用BP神经网络算法针对不同组延迟变量分别进行训练和预测,得到多个预测值。然后,采用核密度估计法拟合多个预测值形成分布的概率密度函数。最后,通过期望估计法或聚合估计法计算得出电力负荷的最终预测值。选取实际负荷数据进行算例分析,结果表明,所提方法适用于训练数据较少的超短期负荷预测,且相较于几种常规预测算法具有更高的预测精度以及较强的稳定性。
Easily affected by the combined various factors such as weather or holidays,it is difficult for the ultra short-term load forecasting in power systems to achieve accurate results.In order to improve the prediction precision,a large amount of historical data is required for training.Aiming at the newly-built initial power systems with less historical data,an ultra short-term power load prediction method based on a randomly distributive embedded framework and BP neural network is presented.Firstly,the delay variables of different state variables such as power load variables and meteorological variables in the power system are considered as an independent factor.So different sets of delay variables are trained and predicted with BP neutral network algorithm,and multiple prediction values are obtained.Then,the kernel density estimation method is used to fit multiple predicted values to form the probability density function of the distribution.Finally,the final predicted value of power load is calculated by expectation estimation or aggregation estimation.The simulation results of case analysis done with the actual load data show that the proposed method is suitable for ultra short-term load forecasting with less training data,and that it has higher prediction accuracy and stability than several conventional forecasting algorithms.
作者
李国庆
刘钊
金国彬
权然
LI Guoqing;LIU Zhao;JIN Guobin;QUAN Ran(Key Laboratory of Modern Power System Simulation and Control&Renewable Energy Technology,(Northeast Electric Power University),Ministry of Education,Jilin 132012,Jilin Province,China)
出处
《电网技术》
EI
CSCD
北大核心
2020年第2期437-445,共9页
Power System Technology
基金
国家重点研发计划项目(2018YFB0904700).
关键词
超短期负荷预测
随机分布式嵌入框架
BP神经网络
非线性动力系统
短期数据
ultra short-term load forecasting
randomly distributive embedded framework
BP neural network
nonlinear dynamical systems
short-term data
作者简介
李国庆(1963),男,博士,教授,博士生导师,研究方向为电力系统的安全性与稳定性分析、控制与决策、配电系统自动化,E-mail:LGQ@neepu.edu.cn;通信作者:刘钊(1994),女,硕士研究生,研究方向为电力系统配电网源、荷功率预测,E-mail:514580226@qq.com;金国彬(1977),男,博士,副教授,硕士生导师,研究方向为新能源接入、电能质量分析及其控制,E-mail:jgbjgb2005@126.com;权然(1994),男,硕士研究生,研究方向为电力系统优化调度,E-mail:445333133@qq.com。
