Improving frequency stability in grid-forming inverters with adaptive model predictive control and novel COA-jDE optimized reinforcement learning

The increasing utilization of renewable energy sources in low-inertia power systems demands advanced control strategies for grid-forming inverters (GFMs). Conventional Model Predictive Control (MPC) methods, which depend on static models and predefined boundaries, often struggle to preserve frequency stability in dynamic grid conditions. This research presents an Adaptive Model Predictive Control (AMPC) framework to enhance GFM performance in Virtual Synchronous Machine (VSM) mode, ensuring robust frequency stability under uncertainties. The primary issue addressed is the inefficiency of traditional MPC in adapting to dynamic grid conditions. To resolve this, the AMPC framework combines offline reinforcement learning for parameter tuning with online MPC using soft constraints. The offline phase employs a novel Hybrid Crayfish Optimization and Self-Adaptive Differential Evolution Algorithm (COA-jDE) to minimize the cost function, deriving optimal control parameters (Q, R) before real-time deployment. This process, termed cost function minimization using COA-jDE in a reinforcement learning framework, enhances GFM performance by adaptively adjusting virtual inertia and damping. Simulations on a 16MW wind-powered DFIG microgrid demonstrate that AMPC outperforms traditional MPC and VSM methods during grid disturbances, symmetrical faults, islanding, and load shifts. Furthermore, AMPC is computationally efficient compared to conventional reinforcement learning techniques, as adaptation is restricted to offline tuning. The framework not only improves compliance with grid codes (e.g., GC0137, IEEE 1547) but also provides a flexible, resilient control strategy for modern low-inertia grids.