Interval prediction of PV power based on double sample entropy framework and integrated bidomain decomposition deep learning

Abstract

Abstract A method for photovoltaic (PV) power interval prediction based on double sample entropy framework and integrated bidomain decomposition deep learning is proposed to address the volatility, randomness, intermittency and timeliness of PV power prediction. By using the integrated bidomain decomposition method under the framework of double sample entropy, the PV power time series is decomposed into several components with similar complexity and the deep learning network is used to learn and predict the results of each component reconstruction to obtain point predictions. The quantile regression method is then used based on the point prediction to obtain interval predictions. The actual calculation examples show that the RMSE and CPRS of the proposed method are reduced by an average of 45.56% and 31.42%, respectively, compared with other integrated bidomain decomposition deep learning methods that are not in the double sample entropy framework and single decomposition deep learning methods.

Keywords

• deep learning
• integrated bidomain decomposition
• prediction
• PV power
• sample entropy