Short-Term Interval Forecasting of Photovoltaic Power Based on CEEMDAN-GSA-LSTM and SVR

Abstract

Aimed at the intermittency and fluctuation of photovoltaic output power, a short-term interval prediction model of photovoltaic power is proposed. First, the model uses the complete ensemble empirical mode decomposition of adaptive noise (CEEMDAN) to decompose the historical photovoltaic output data into different components and define them as time-series components and random components according to their correlation with time-series features such as declination and time angles. Then, the long short-term memory (LSTM) neural network and the support vector regression (SVR) model optimized by the gravitational search algorithm (GSA) are used to predict the time series components and the random components respectively, and the prediction results of the time series components and the random components are superimposed to obtain the point prediction result. After the error is subjected to Johnson transformation and normal distribution modeling, the photovoltaic power interval prediction result is obtained. Finally, the effectiveness of the method is verified by an example. The comparison of the proposed model with other existing prediction models under different weather conditions suggests that the proposed model has a higher accuracy and a better robustness, which can provide precise confidence intervals based on point prediction values.

Keywords

• photovoltaic power prediction
• interval prediction
• complete ensemble empirical mode decomposition with adaptive noise (ceemdan)
• gravitational search algorithm (gsa)
• long short-term memory (lstm)
• support vector regression (svr)
• johnson transformation