Short-term photovoltaic power prediction with CPO-BILSTM based on quadratic decomposition

Abstract

To address the challenges of volatility and unpredictability in photovoltaic (PV) power, a short-term combined prediction model named EMD-VMD-CPO-BILSTM is proposed. The process begins with the selection of a similar day using the K-means algorithm, followed by the decomposition of historical PV power data into several signal components. The Empirical Mode Decomposition (EMD) method is employed to denoise the signal, and the residual signal is further decomposed using Variational Mode Decomposition (VMD) to minimize mode aliasing and improve accuracy. Subsequently, the parameters of the Bidirectional Long Short-Term Memory (BILSTM) model are optimized using the Crested Porcupine Optimization (CPO) algorithm. The optimized BILSTM model is subsequently applied to power prediction. The experiment was conducted using observation data from the Australian Desert Knowledge (DKA) Solar Energy Centre, located in Australia. The numerical outcomes demonstrate that the proposed EMD-VMD-CPO-BILSTM model reduces mean absolute error (MAE) and root mean square error (RMSE) by 6.67 % and 3.76 %, respectively.