Efficient interpretable wind speed prediction system

Designing a highly interpretable system that ensures both predictive accuracy and efficiency is a pivotal step toward integrating large-scale green power into the main grid. To enhance wind power forecasting performance while maintaining interpretability, this study proposes a fuzzy time series forecasting approach based on infinite-dimensional deep learning rules. Unlike current mainstream wind speed prediction systems, the proposed system replaces the rule-learning step in fuzzy time series forecasting with an infinite-dimensional convolutional neural network, thereby improving the predictive performance of interpretable models. It establishes a multi-fuzzy rule matching mechanism under infinite-dimensional convolution kernels and introduces an innovative intelligent mode decomposition combined with fuzzy information granulation to address challenges such as high randomness and missing anomalies in wind speed data. Finally, our detailed comparative experiments and sensitivity analysis on the Shandong Penglai wind farm dataset, the SOTAVENTO dataset, and two public datasets show that the proposed system: (1) outperforms mainstream models in both prediction efficiency and effectiveness (10% better than traditional interpretable models, 6% better than improved attention models, and 3% better than similar interpretable deep learning models); (2) improves data quality and prediction performance through data repair methods (MSE increased by 12% and MAPE increased by 33%), while reducing the impact of hyperparameters on the results of deep learning models and reducing a lot of manual parameter tuning work.