Probabilistic deep learning control of hybrid FACTS for real-time voltage stability in renewable-rich power grids

Abstract Including renewable energy sources into contemporary power systems has greatly added unpredictability and presented problems including harmonic distortion, voltage instability, and degraded power quality. In response, we provide a new deep learning-enabled framework for the real-time adaptive control and optimal allocation of hybrid Flexible AC Transmission Systems (FACTS), specifically the Unified Power Flow Controller (UPFC) and Static VAR Compensator (SVC), in renewable-rich transmission networks. For multi-objective planning, the suggested method combines a Two-Point Estimation Method (2PEM)-based probabilistic load flow model with a hybrid Grey Wolf Optimization–Particle Swarm Optimization (GWO–PSO) algorithm. Trained offline on a scenario-rich dataset, a Long Short-Term Memory (LSTM) neural controller inferences control set-points in real time with sub-15 ms latency. Under operational contingency and changeable photovoltaic (PV) generation, this integration of offline probabilistic optimization with online artificial intelligence-driven control enables dynamic stabilization of voltage, harmonic mitigation, and power loss minimization. Application to the Southern Interconnected Grid of Cameroon shows compliance with IEEE-519 harmonic thresholds, a 47% increase in voltage stability index, and a 30.9% reduction in active power loss. These findings support the scalability and efficiency of the framework in improving power system resilience and quality by means of intelligent control, so fitting for both advanced and developing energy infrastructures. Graphical abstract