摘要
光伏发电预测时间尺度的变长会导致预测误差的积累,此时基于经典长短期记忆网络模型的预测精度在长预测时间尺度的场景下将会显著下降。为此,提出一种采用多输入多输出策略和长短期记忆网络的光伏出力多元预测方法,在长短期记忆网络模型的基础上采用多输入多输出策略,有效提高了长时间尺度预测场景下光伏发电功率的预测精确度。最后通过实际光伏发电站近3年发电数据集进行验证。结果表明,当预测的时间尺度大于6h,所提方案比传统递归策略的长短期记忆网络预测模型具有更小的均方根误差和平均绝对误差。
The increase in the time scale of photovoltaic power generation prediction wil lead to the accumulation of prediction errors,and the prediction accuracy of models based on classical long short-term memory(LSTM)networks will significantly decrease in scenarios with long prediction time scale.To this end,a multi-input and multi-output strategy and LSTM network based photovoltaic output multivariate prediction method was proposed.Based on the LSTM network model,a multi-input&multi-output strategy was adopted to effectively improve the prediction accuracy of photovoltaic power generation in long-term prediction scenarios.Finally,validation was conducted using a dataset of power generation data from actual photovoltaic power plants over the past three years.The results indicate that when the predicted time scale is more than 6 h,the proposed scheme has smaller root mean square error and average absolute error compared to the prediction model of traditional recursive LSTM networks.
作者
李鸿奎
袁帅
李福建
李喜同
张新明
Li Hongkui;Yuan Shuai;Li Fujian;Li Xitong;Zhang Xinming(State Grid Heze Power Supply Co.,Ltd.,Heze Shandong 274002,China;State Grid Shandong Electric Power Co.,Ltd.,Jinan Shandong 250001,China)
出处
《电气自动化》
2025年第3期37-39,42,共4页
Electrical Automation
基金
国网菏泽供电公司重点项目(SGSDHZ00DKJS2400540)
关键词
光伏出力多元预测
多输入多输出策略
长短期记忆网络
预测时间尺度
预测精确度
photovoltaic output multivariate forecast
multi-input&multi-output strategy
long short-term memory(LSTM)network
prediction time scale
prediction accuracy
