Photovoltaic Power Prediction Based on VMD-BRNN-TSP

Overfitting often occurs in neural network training, and networks with higher generalization ability are less prone to this phenomenon. Aiming at the problem that of photovoltaic (PV) power prediction model is insufficient, a PV time-sharing (TSP) combining variational mode decomposition (VMD) Bayesian regularization (BRNN) proposed. Firstly, meteorological sequences related output selected by mutual information (MI) analysis. Secondly, VMD processing performed on filtered sequences, which aimed reducing non-stationarity data; then, normalized cross-correlation (NCC) signal-to-noise ratio (SNR) between components obtained signal original data calculated, after key influencing factors screened out eliminate correlation redundancy data. Finally, divided into two datasets based whether irradiance day zero or not. Meanwhile, predictions using BRNN for each datasets. Then, results reordered chronological order, realized conclusively. It was experimentally verified mean absolute value error (MAE) method proposed paper 0.1281, reduced 40.28% compared back propagation (BPNN) same dataset, squared (MSE) 0.0962, coefficient determination (R2) 0.9907. Other indicators also confirm much significance TSP contributive.