Initializing Successive Linear Programming Solver for ACOPF using Machine Learning
This work addresses efficiency in power grid optimization for energy systems, but it is incremental as it applies existing ML methods to a specific initialization step.
The paper tackles the computational performance of Successive Linear Programming (SLP) for solving Alternating Current Optimal Power Flow (ACOPF) problems by using machine learning algorithms to initialize the solver with near-optimal solutions, achieving results compared to DCOPF initialization on 3-bus systems.
A Successive linear programming (SLP) approach is one of the favorable approaches for solving large scale nonlinear optimization problems. Solving an alternating current optimal power flow (ACOPF) problem is no exception, particularly considering the large real-world transmission networks across the country. It is, however, essential to improve the computational performance of the SLP algorithm. One way to achieve this goal is through the efficient initialization of the algorithm with a near-optimal solution. This paper examines various machine learning (ML) algorithms available in the Scikit-Learn library to initialize an SLP-ACOPF solver, including examining linear and nonlinear ML algorithms. We evaluate the quality of each of these machine learning algorithms for predicting variables needed for a power flow solution. The solution is then used as an initialization for an SLP-ACOPF algorithm. The approach is tested on a congested and non-congested 3 bus systems. The results obtained from the best-performed ML algorithm in this work are compared with the results of a DCOPF solution for the initialization of an SLP-ACOPF solver.