考虑空间相关性采用LSTM神经网络的光伏出力短期预测方法

Short-term Photovoltaic Output Forecasting Method Using LSTM Neural Network with Consideration of Spatial Correlation

为了提高光伏出力短期预测精度,提出了一种考虑空间相关性采用长短期记忆LSTM(long short-term memory)神经网络的预测方法。该方法首先在周边光伏电站中依据光伏序列的延迟相关性选取参考电站,在此基础上,依据光伏出力随机性部分持续时间的概率分布,分时段对其进行空间相关性分析,选择与目标电站相关性较强的光伏序列;然后,结合目标电站气象数据的主成分分析结果和历史光伏数据,构建LSTM神经网络模型;最后,通过仿真实验分析验证了所提预测方法的有效性。

To improve the prediction accuracy of short-term PV output,a novel forecasting approach using long shortterm memory(LSTM)neural network is proposed with the consideration of spatial correlation. First,the reference PV stations are selected according to the lagged correlations between PV power series at the surrounding PV stations. On this basis,according to the probability distribution of the duration of random PV output,spatial correlation analysis is carried out in different periods,and the PV power series with a stronger correlation to the target PV station is selected.Then,an LSTM neural network model is constructed by combining the principal component analysis(PCA)results of meteorological data at the target station and the historical PV data. Finally,the effectiveness of the proposed forecasting method is analyzed and verified by the result of a simulation experiment.