Enhancing wind power prediction accuracy using MVMD decomposition and QPSO-optimized LSTM

Aiming at the complex characteristics of wind power, such as nonlinearity, non-stationarity and being affected by multiple factors, as well as the problem that existing single prediction models are often difficult to achieve ideal accuracy, a hybrid prediction model integrating Multivariate Variational Mode Decomposition (MVMD), Quantum-behaved Particle Swarm Optimization (QPSO) and Long Short-Term Memory network (LSTM), namely MVMD-QPSO-LSTM, is proposed. Firstly, MVMD decomposes the original wind power sequence into several relatively stable sub-sequences, which effectively reduces data complexity and nonlinearity. Secondly, QPSO optimizes LSTM's key parameters, solving the problem of prediction performance fluctuation caused by the dependence on experience in traditional LSTM parameter setting. Finally, the optimized LSTM is applied to the prediction of each sub-sequence respectively, and the final wind power prediction result is obtained through integration. Experimental results show that compared with LSTM, PSO-LSTM and QPSO-LSTM, the proposed model's prediction accuracy is improved by 70%-85%, which significantly enhances wind power prediction accuracy and stability and provides reliable technical support for power system economic dispatching and safe operation.