Coordinated Active-Reactive Self-Optimizing Control of Distributed Resources in Distribution Networks Based on an Improved Adam-MPC Algorithm

Abstract

To address issues such as active power tracking errors, voltage overshoot, and disturbance propagation among resources that commonly occur at the distribution network's edge following the integration of distributed PV (Photovoltaic), energy storage, and controllable loads, this paper proposes a method for active-reactive power coordinated self-optimizing control of distributed resources based on an improved Adam-MPC(Adaptive Moment Estimation-Model Predictive Control). This method establishes a linearized DistFlow voltage-sensitivity model within the feeder coordination controller, embedding node-voltage constraints into the MPC (model predictive control) rolling optimization using a "hard constraints first, relaxation variables as a fallback" approach. Simultaneously, it utilizes Adam's first-order momentum and second-order variance-estimated momentum to identify measurement errors, reconstruct state tracking weights online, and achieve flexible weight reduction for disturbed resources and coordinated support for undisturbed resources. Simulations on a 10 kV radial distribution network demonstrate that, under a 45% sudden drop in end-of-line PV generation, the proposed method can control the total active power response error of the feeder within 3.8%, maintain the minimum end-of-line voltage near 0.985 p.u. (per-unit), and meet the real-time requirements of 1-second rolling control.