Probabilistic Prediction of Photovoltaic Power Using Bayesian Neural Network - LSTM Model

Abstract

The deterministic prediction of photovoltaic power can support the long-term optimization of the dispatching system, but under complex weather conditions, the short-term fluctuation of photovoltaic power will be large, and the prediction accuracy of the deterministic prediction method will be significantly reduced, which will affect the safe and stable operation of the power grid. A prediction method of the photovoltaic power probability distribution based on the Bayesian Neural Network -Long Short-Term Memory (BNN-LSTM) model is proposed. Firstly, use path analysis to select the most related numerical weather forecast feature variables. Then, the LSTM unit is used to extract features of the numerical weather forecast and historical time series data to improve model prediction accuracy. Finally, the parameters in the Bayesian neural network are expressed in the form of a probability distribution, which can be used to fit the probability distribution of photovoltaic power. The case study shows that the proposed method is more capable of dealing with photovoltaic power fluctuations than the deterministic prediction method. Compared with the traditional photovoltaic power interval predicting method, the prediction interval width is narrower under the same forecast accuracy.