Data-Driven Successive Linearization for Optimal Voltage Control

A team from Caltech and Tsinghua University has developed a new AI-driven method to stabilize modern power grids struggling with voltage fluctuations. The core problem is that today's grids, increasingly powered by intermittent solar generation and rapidly varying loads from electric vehicles and storage, are governed by complex, nonlinear physics. Traditional advanced controllers for voltage regulation use fixed linear approximations of these dynamics, which often fail, producing infeasible solutions when applied to the actual nonlinear system, especially under heavy power injection.

The proposed solution, detailed in the arXiv paper 'Data-Driven Successive Linearization for Optimal Voltage Control,' introduces a successive linearization approach that is both data-driven and adaptive. It leverages a key mathematical insight: the deviation between the true nonlinear power flow solution and its linear approximation is bounded by the distance from the current operating point. Therefore, by continuously performing data-driven linearization around the most recent operating point, the controller can accurately track the real system. The researchers established theoretical convergence guarantees and demonstrated through case studies that their method achieves fast convergence and rapidly adapts to changes in net load, outperforming static models.

This work represents a significant shift from model-based control with fixed parameters to a more adaptive, data-informed paradigm. It directly tackles the operational challenges posed by the renewable energy transition, where grid conditions change too quickly for outdated models to keep pace. The convergence to a neighborhood of KKT (Karush–Kuhn–Tucker) points is proven by exploiting the problem's convex objective and the structure of its nonlinear constraints, providing a solid mathematical foundation for the controller's reliability.