A Deep Learning Method for Wind Turbine Active Power Prediction Utilizing Specialized Decomposition

In the era of artificial intelligence, energy serves as a vital foundation for technological advancement. To address the environmental pollution and carbon emissions associated with traditional energy sources, clean energy has become central to global green development strategies. Among them, wind energy plays a key role but is inherently unstable and intermittent, posing challenges for efficient grid integration. Accurate wind power forecasting significantly enhances wind energy utilization and eases grid scheduling pressure. This study leverages extensive wind turbine operation data and meteorological measurements from wind towers to perform active power forecasting. The constructed dataset enriches the research foundation with high spatial and temporal resolution, supporting multidimensional analysis. In order to address the intricate nonlinear and longterm dependencies present in wind power output sequences, this study introduces an innovative approach that employs a combination of signal decomposition, clustering, and a subsequent decomposition of high-frequency components, thereby streamlining the model's learning process. Moreover, critical meteorological factors, most notably wind speed, are utilized as covariates to enhance the model's responsiveness to varying atmospheric conditions. Comprehensive experimental results verify that the proposed technique outperforms existing benchmarks across all assessment criteria, underscoring its robustness in forecasting wind power and its promising potential for promoting the wider use of sustainable energy.