A learning-augmented approach for AC optimal power flow. (September 2021)

• Record Type: Journal Article
• Authors: Rahman, Jubeyer; Feng, Cong; Zhang, Jie
• Is Part Of: International journal of electrical power & energy systems. Volume 130(2021)
• Publication Date: 2021-09
• Subjects: AC OPF -- Random forest -- Decision tree -- Extreme learning machine -- Optimality; Electrical engineering -- Periodicals; Electric power systems -- Periodicals; Électrotechnique -- Périodiques; Réseaux électriques (Énergie) -- Périodiques; Electric power systems; Electrical engineering; Periodicals; 621.3
• DOI: 10.1016/j.ijepes.2021.106908
• Languages: English
• ISSNs: 0142-0615

Abstract

Highlights: A learning-augmented method is developed for AC OPF. Results have shown minimal constraints violation and loss of optimality. Reduce AC OPF solution time by 15–100 times. Abstract: Due to the high nonlinearity of AC optimal power flow (OPF), numerous efforts have been made in recent decades to find efficient methods. Machine learning (ML) has proven to significantly reduce the computational costs in many real-world problems. Thus, this paper develops a learning-augmented method for solving AC OPF, which integrates both power network equations and ML to yield near-optimal solutions. More specifically, ML models are developed to first predict bus voltage magnitudes and angles. Then, physics-based network equations are employed to calculate the power injection at different buses. Three ML algorithms, i.e., random forest, multi-target decision tree, and extreme learning machine, are explored and compared. To evaluate the efficiency of the proposed learning-augmented AC OPF solver, the MATPOWER Interior Point Solver is adopted as a baseline. Case studies on both 500-bus and 4918-bus test networks show that the proposed learning-augmented method has reduced the computational time by 15–100 times depending on the network size with a minimal loss in optimality.