A New Design of an Optimized Informer Wind Power Prediction Model Utilizing Wind Turbine Health Assessment

Wind power prediction is of significant value to the stability of the power grid. Employing the Informer model for wind power prediction yields better results than traditional neural networks, yet issues such as slow speed and insufficient accuracy persist. By utilizing a health assessment algorithm to optimize the Informer model, both prediction accuracy and speed can be concurrently enhanced. Initially, a health matrix is obtained by performing a health assessment of wind turbines based on operational data. Subsequently, this health matrix is used to optimize the encoding method of the Informer, improving prediction speed. Simultaneously, the decoding method, embedding vectors and prediction process of the Informer are refined to increase prediction accuracy. Finally, conventional Informer models and optimized Informer models are tested and compared using four distinct wind power datasets. The results indicate that the optimized Informer model achieves an approximately 15% increase in prediction accuracy and about a 100% increase in prediction speed.