Support Vector Regression for Reactive Power Compensation: A Data-Driven Approach to Power Loss Reduction

The modernization of distribution networks and the growing penetration of distributed energy resources (DERs) demand innovative strategies to manage reactive power and minimize losses. This study proposes a machine learning-based approach using Support Vector Regression (SVR) for determining optimal capacitor placement in the IEEE 33-bus distribution system. The method replaces traditional heuristic algorithms with a predictive model trained on various simulated configurations to estimate capacitor positions and sizes that minimize active power loss. Two configurations—using two and three capacitors—are assessed and validated via Backward-Forward Sweep (BFS) load flow analysis. Results indicate that SVR offers substantial loss reduction and voltage profile improvements, outperforming Particle Swarm Optimization (PSO) and Bat Algorithm (BAT) in terms of efficiency, accuracy, and computation time. This work highlights the potential of SVR as a practical tool for real-time optimization in smart distribution networks.