摘要
为了提高电力负荷的预测精度,提出使用粒子群算法(PSO)优化长短期记忆(LSTM)神经网络超参数的电力负荷预测模型(PSO-LSTM)。针对LSTM超参数较难选取的问题,利用PSO算法能有效寻找全局最优解的特点进行LSTM模型超参数寻优,不断训练找到合适的超参数并进行验证。通过实际案例数据进行仿真分析,并与传统的LSTM神经网络预测模型以及反向传播(back propagation,BP)神经网络预测模型进行对比,其平均绝对百分比误差(MAPE)分别降低了0.64%和1.67%,验证了本方法的预测效果更佳。实验表明,本电力负荷预测模型具有较好的精度和稳定性。
In order to improve the accuracy of power load prediction,a power load prediction model(PSO-LSTM)using particle swarm optimization(PSO)was proposed to optimize the hyper-parameters of long and short memory(LSTM)neural network.Aiming at the problem that it is difficult to select the LSTM hyper-parameters,the LSTM model hyper-parameters are optimized by using the PSO algorithm which can effectively find the global optimal solution and the appropriate hyper-parameters are continuously trained and verified.Through the simulation analysis of actual case data,and compared with the traditional LSTM neural network prediction model and Back Propagation(BP)neural network prediction model,the Mean Absolute Percentage Error(MAPE)of the proposed method is reduced by 0.64%and 1.67%respectively,which verifies that the prediction effect of the proposed method is better.Experimental results show that the proposed power load prediction model has good accuracy and stability.
作者
王晓辉
邓威威
齐旺
WANG Xiaohui;DENG Weiwei;QI Wang(Beijing Architecture University Electrical and Information College;State Grid Liaoning Power Supply Co.,Ltd State Grid Jinzhou Power Supply Company)
出处
《上海节能》
2022年第2期164-169,共6页
Shanghai Energy Saving
基金
安徽建筑大学智能建筑与建筑节能安徽省重点实验室开放课题(IBES2020KF06)。
关键词
长短期记忆神经网络
粒子群算法
负荷预测
超参数
LSTM(Long Short Time Memory)
PSO(Particle Swarm Optimization)
Load Forecasting
Hyper-Parameter