Short-term Photovoltaic Power Prediction of Bi-LSTM Based on MIC

Abstract

Photovoltaic power generation has certain volatility and randomness. In order to reduce the impact of photovoltaic power on the stable operation of power grid, this paper proposes a short-term Bi-LSTM photovoltaic power method based on MIC. Firstly, the correlation degree between the predicted PV power and meteorological factors was analyzed based on the maximum information coefficient (MIC), so as to screen the characteristics of high-dimensional meteorological data in the power grid. Then, the historical similar days were selected by the weighted sum of the maximum information coefficient. Finally, a bi-directional long neural network was built short-term memory (Bi-LSTM) was used to predict the short-term photovoltaic power. Case simulation shows that the proposed MIC-Bi- LSTM power prediction model has lower prediction bias and higher prediction accuracy.