Hybrid Data-Model Driven Hierarchical Multi-objective AGC for Power System Considering Parameter Disturbances

This paper proposes a multi-objective automatic generation control (AGC) method that integrates data and model driven approaches in a hierarchical framework to handle parameter disturbances and time-varying dynamics. At the generation company (GenCo) level, a reduced-order ultra-local model (ULM) dynamically couples system frequency deviations with GenCo dispatch commands. An improved fast terminal sliding mode observer (IFTSMO) is developed to estimate nonlinear disturbance terms within the ULM. Based on the ULM and IFTSMO observations, an adaptive model-free predictive controller (MFPC) considering parameter variations is designed to perform real-time rolling optimization of GenCo-level dispatch signals. At the unit level, independent ULMs and IFTSMOs are designed for each generator, enabling the unit-level MFPC to adapt to diverse dynamics and parameter variations of heterogeneous frequency regulation units. This enables fast and accurate tracking of GenCo-level commands while optimizing cost and carbon emissions. The proposed hierarchical framework is validated through theoretical analysis and diverse simulation scenarios. It demonstrates enhanced frequency regulation performance compared with traditional PI and SMC controllers, achieving a 12.51% improvement over PI control. In addition, multi-objective optimization results in a 44.69% reduction in total generation cost and a 20.7% decrease in carbon emissions.