Wind Forecasting based on CEL-LSTM Hybrid Model for German Farm

With global interest in renewable energy surging, wind power emerges as a pivotal sustainable energy source. The prediction of wind power efficiency, however, is impeded by factors like wind speed, direction, and meteorological conditions. This study introduces the CEL-LSTM hybrid model, which merges CEEMDAN, EWT, Lion optimizer, and LSTM to improve wind energy efficiency prediction accuracy. Initially, wind energy data is decomposed via CEEMDAN to extract eigenmode functions at various frequencies. The EWT method then reduces noise in high-frequency IMFs, enhancing data coherence. Feature extraction utilizes the Lion optimizer to convert raw data into trainable features, followed by LSTM's application in forecasting wind energy efficiency, with the Lion optimizer refining model parameters. An empirical analysis conducted on data from a 2050kW wind farm in Germany evaluates the model's performance. This includes ablation experiments and a sensitivity analysis, revealing the CEL-LSTM model's superiority by approximately 17% over the prevailing dual decomposition algorithm in wind power predictions. Furthermore, the sensitivity analysis underscores the model's sensitivity, stability, and effectiveness in diverse wind energy prediction contexts.