Research on a Novel Wind Power Prediction Method Based on VMD-IMPA-BiLSTM

Abstract

Wind power, as a pivotal technology in the fight against climate change and the advancement of sustainable energy, occupies an essential role in the global shift towards a new energy paradigm. However, the inherent fluctuation and episodic nature of wind power output pose formidable challenges to achieving precise forecasts, thereby impeding the accuracy of traditional forecasting methodologies. To address this, the paper introduces a novel forecasting approach, the VMD-IMPA-BiLSTM method. This method begins by applying the VMD technique to decompose wind power output into several sub-sequences, significantly reducing data non-stationarity and thus mitigating the negative impact of data volatility on forecasting accuracy. The IMPA method, an improvement over the traditional MPA, incorporates a Tent chaotic mapping for population initialization and a priority selection strategy based on average fitness, enhancing the model’s ability to perform global searches and local optimizations. Moreover, the IMPA is leveraged to optimize critical parameters of the BiLSTM, such as maximum training iterations, hidden layer units, and learning rate, thereby curtailing the subjectivity inherent in manual tuning and bolstering the model’s forecasting robustness. A thorough comparative analysis with other models has been conducted to evaluate the performance of the developed VMD-IMPA-BiLSTM forecasting model comprehensively. Simulation results indicate that the proposed method significantly outperforms the benchmark models in terms of prediction accuracy across various application scenarios, showcasing not only exceptional stability in forecasts but also a robust generalization capability across diverse contexts.

Keywords

• Wind power prediction
• variational modal decomposition
• improved marine predator algorithm
• bidirectional long and short-term memory neural networks
• hyperparameter optimization