Short-term wind speed prediction based on variational modal decomposition and improved BP neural networks

Abstract

The wind power forecasting technology can be an important basis for wind power grid connection and the power system dispatching department to make wind power dispatching plans. Firstly, for the drawback that BP neural network is prone to fall into local optimum, the simulated annealing algorithm is introduced to optimize the initial weights and thresholds of the BPNN to construct SA_BP prediction model. Secondly, VMD was implemented to decompose the raw wind speed series into a number of sub-series and reconstruct the subsequence based on the sample entropy. Finally, the reconstructed components are predicted separately by SA_BP and the predictions are then stacked to get the ultimate forecast results. To prove the proposed prediction model's validity, wind speed prediction of a wind farm was simulated and three other prediction models were compared through four indicators: mean square error, root mean square error, mean absolute error and mean absolute percentage error. The experimental data indicate that compared to the BPNN prediction model, the VMD_SA_BP prediction model has a 2.576 lower mean square error, 0.9535 lower root mean square error, 0.6913 lower mean absolute error and 4.5101 percentage points lower mean absolute percentage error.