基于深度神经网络的光伏发电时间序列多元预测

Multivariate Prediction of Photovoltaic Power Time Series Based on Deep Neural Networks

利用不同时间序列间的相关性和依赖性基于深度神经网络(DNNs)提出了两种不同的多元长短期记忆网络(LSTM)光伏输出功率预测方法,充分考虑了空气温度、风速等影响因素之间的相关性特征。以光伏发电站运行数据为例,通过对光伏发电预测模型进行训练和测试,并与单变量LSTM和Stacked-LSTM模型的结果进行比较,研究结果表明,所提的Conv-LSTM可以在减少30.76%训练时间的基础上提升0.71%~1.33%的准确度,Conv-LSTM和Multi-LSTM分别以高达93.12%和96.12%的准确度跟踪实际光伏发电。

In this paper,two different multivariate long-and short-term memory(LSTM)network PV output power prediction methods are proposed based on deep neural networks(DNNs)by utilizing the correlation and dependence between different time series,with full consideration of the correlation characteristics between air temperature,wind speed and other influencing fac-tors.Taking the PV power plant operation data as an example,by training and testing the PV power prediction model and com-paring the results with those of the univariate LSTM and Stacked-LSTM models,the results of the study show that the pro-posed Conv-LSTM can improve the accuracy by 0.71%~1.33%on the basis of reducing the training time by 30.76%,and the Conv-LSTM and Multi-LSTM track real PV with up to 93.12%and 96.12% accuracy.