Adaptive Data-Driven Frequency Control With Minimal Cost for Renewables Dominated Microgrids

This paper proposes an adaptive data-driven frequency control with minimal cost for renewables-dominated microgrids, which does not rely on its accurate model. The Koopman operator provides the theoretical basis for the global linearization of nonlinear systems, and the extended dynamic mode decomposition (EDMD) is the most computationally efficient method for its approximation. However, insufficient disturbance in normal cases cannot excite all system modes, causing EDMD to overfit the estimated model and deteriorate the performance of the data-driven method. To address this issue, this paper develops adaptive extended dynamic mode decomposition (AEDMD). AEDMD reduces data requirements by extracting dominant dynamic modes through the persistent excitation condition and principal component analysis. It solves the issue of insufficient input excitation by designing the control matrix with the assistance of the control-affine structure of grid-forming converters. This paper utilizes model predictive control (MPC) to achieve rapid frequency recovery and minimize generation cost by coordinating different types of distributed resources. In addition, the stability of the control system is theoretically clarified. The effectiveness of the proposed method is demonstrated on several systems.