A Hybrid VMD-Based ARIMA-LSTM Model for Day-ahead PV Prediction and Uncertainty Analysis

Abstract

With the rapid growth of PV power system, PV power prediction is important for optimal operation and control of power grid connection. A novel efficient two-stage prediction approach is proposed to quantify the fluctuant range and uncertainty of PV power output. A hybrid model integrated ARIMA with LSTM based on temporal attention mechanism is proposed for point prediction in the first stage. In order to get a more accurate forecasting result, variational mode decomposition (VMD) with sample entropy (SE) is developed to decompose the original PV power and decide which model is used to predict for the decomposed sub-series. In the second stage, nonparametric kernel density estimation method is applied to generate prediction intervals (PIs) of the deterministic forecasting error without assuming the probability distribution. A practical case in southwest China is simulated. The results show that the performance of proposed model is best under 3 weather types.