An improved variational mode decomposition (VMD) algorithm based on modal correlation and reconstruction error
and an improved subtraction-average-based optimizer (CSABO) to optimize short- and medium-term photovoltaic (PV) power prediction model consisting of temporal convolutional network (TCN) and bidirectional gated recurrent unit (BiGRU) are proposed. Firstly
the historical PV data are decomposed into multiple components with different frequencies using the improved VMD. Then
the components are combined with key meteorological factors
and the PV power forecasts are reconstructed by the TCN-BiGRU model by forecasting each time series data separately. Finally
the parameters of the prediction model are optimized using CSABO to improve the model performance. The actual Australian PV data is used as an arithmetic example for experimental analysis
and the results demonstrate that the proposed model exhibits the best evaluation indexes and higher prediction accuracy compared with EMD-TCN-BiGRU