Multi-data Models Control of Grid-Connected Inverters for Enhanced Adaptive Compensation Capability

Data-driven regression compensation control effectively mitigates the inherent nonlinearities in grid-connected inverters. However, the structural limitations inherent to the single-data regression model challenge the balance between minimal data requirements and adaptive optimization, thereby constraining compensation precision and generalization capability. To overcome the generalization limitations of the single-data model, this paper proposes a collaborative compensation control strategy based on multi-data regression models, which employs adaptive weighting for dynamic model fusion. This method integrates techniques for operating interval and boundary feature extraction to construct multi-data regression models using lightweight offline regression. Real-time matching of operating features enables dynamic model selection and weight assignment, thereby accurately reconstructing the compensation signal in extrapolation regions beyond the training data boundaries. Furthermore, by integrating with a low-order controller, a hybrid control architecture merging the mechanism-based and data-driven paradigms is established, significantly enhancing the adaptive compensation performance. Experimental results validate the effectiveness and superiority of the proposed strategy.