Wavelet-Constrained Multi-Decomposition Autocorrelation model for wind farm SCADA data imputation

Abstract Although traditional missing data management in existing supervisory control and data acquisition (SCADA) systems addresses the data loss problem in wind power generation, they rely on static assumptions, which hinders their adaptability and effectiveness in dynamic environments. In addition, existing Transformer methods usually fail to capture complex spatiotemporal correlations and have limited robustness to various missing data patterns. To address these challenges, the Wavelet-Constrained Multi-Decomposition Autocorrelation (WCMDA) model is proposed as an innovative framework for SCADA data imputation. The model integrates a Sequence Feature Embedding Module to extract structured surface information, a Multi-Scale Sequence Decomposition Module to capture global and local multi-scale correlations, and a Wavelet Constraint Module to enhance frequency and sequential data representation through wavelet domain features. WCMDA is validated on a SCADA dataset from a wind turbine gearbox provided by a wind power plant, demonstrating significant improvements over traditional methods in data completion accuracy and robustness against various missing data scenarios. This dynamic framework not only addresses the critical issue of data loss in wind power systems but also advances intelligent management practices by enhancing the reliability and efficacy of SCADA data processing.