Ultrashort-term power prediction researched by LSTM memory neural networks based on partial least squares dimensionality reduction

The output power of photovoltaic power plants is significantly affected by a variety of external environmental factors, and characterized by nonlinear and large fluctuations. In response to these issues, an ultra short term power prediction method for photovoltaic power generation is proposed, which combines Partial Least Squares (PLS), Genetic Algorithm (GA), and Long Short Term Memory (LSTM). Taking into full consideration of the six environmental factors constraining the PV output power. Firstly, partial least squares regression (Partial Least Squares PLS) is used to extract the key influencing factors of feature sequences. By fully utilizing sequence information, the data size and complexity are reduced, the correlation and redundancy of the original sequence are eliminated, and the dimensionality of the model input is reduced. Then Genetic Algorithm (GA) is used to select the optimal hyperparameters for the LSTM neural network. Ultimately, dynamic time modelling of multivariate feature sequences using LSTM networks is used to achieve the prediction of PV power. The reduction of prediction error of this method compared to single LSTM and CNN models is verified by simulation example analysis and is feasible.