基于LSTM和自注意力机制的电力负荷预测

Power load forecasting based on LSTM and self attention mechanism

电力负荷预测作为电网系统的关键部分,对保障其稳定运行起着至关重要的作用。为提升电力负荷预测精度,该文以真实场景下的电网数据为基础结合深度学习技术,提出了基于LSTM(Long Short Term Memory)和自注意力机制的负荷预测模型,该模型由输入层、任务共享层和预测层构成。数据在输入层完成清洗和预处理;并在任务共享层进行特征提取,进而由预测层采用Relu和Dropout方法实现负荷预测。最后,以2021年5月至2022年5月威州气象和电站水机的数据集为基础进行实验验证,本文提出方法的预测性能优于对比模型(LSTM、GBRT与BP等模型),具备一定的有效性和可行性。

As a key part of power grid system,power load forecasting plays a vital role in ensuring its stable operation.In order to improve the accuracy of power load forecasting,this paper proposes a load forecasting model based on LSTM(Long Short Term Memory)and self attention mechanism,which is composed of input layer,task sharing layer and forecasting layer,based on the power grid data in the real scene and combined with deep learning technology.The data is cleaned and preprocessed at the input layer;And feature extraction is carried out in the task sharing layer,and then the load forecasting is realized in the forecasting layer by using Relu and Dropout methods.Finally,based on the data set of Weizhou meteorological and hydropower station water turbine from May 2021 to May 2022,the experimental verification is carried out.The prediction performance of the method proposed in this paper is better than that of the comparison model(LSTM,GBRT and BP models),and has certain validity and feasibility.