基于长短期记忆网络的水力发电功率预测研究

Research on Hydroelectric Power Prediction Based on Long Short-Term Memory Network

为解决水电站发电功率预测中存在的多时间尺度关联分析少、发电功率预测准确率低的问题,提出了一种基于长短期记忆(LSTM)网络的水力发电功率预测方法。首先,对历史的水电站发电数据进行质量评估,对缺失、错误的数据进行修复,以消除错误数据对模型预测精度造成的影响。其次,对影响水电站发电功率的因素进行分析,并在此基础上采用LSTM网络实现了年、季、月、日、时多时间尺度的发电功率关联分析与预测。最后,在某水电站的应用表明,水力发电功率预测率为97.2%。所提方法能有效提高水电站发电功率预测精度,提升水电站精益化管理水平。

To solve the problems of little multi-timescale correlation analysis and low accuracy of power prediction in hydroelectric plant power prediction,a hydroclectric power prediction method based on long short-term memory(LSTM)network is proposed.Firstlly,the quality of historical hydroclectric plant power generation data is evaluated,and the missing and erroneous data are repaired to eliminate the influence of erroneous data on the prediction accuracy of the model.Secondly,the factors affecting hydroelectric plant power generation are analyzed,and on this basis,the LSTM network is used to realize the correlation analysis and prediction of power generation in multi-timescale of year,season,month,day,and hour.Finally,the application in a hydroelectric plant shows that the prediction rate of hydroelectric power is 97.2%.The proposed method can effectively improve the accuracy of hydroelectric plant generation power prediction and enhance the lean management level of hydroelectric plant.