Learning-based decomposition and sub-problem acceleration for fast production cost minimization simulation

Abstract

Production cost minimization (PCM) simulation is crucial for long-term power system assessments, yet it presents significant computational challenges due to the large number of binary variables involved. The traditional time-domain decomposition (TDD) method aims to expedite PCM solving but frequently lead to substantial temporal constraint violations, thereby compromising accuracy. While machine learning (ML) techniques have been integrated with branch and bound (B&B) algorithms to enhance solving speed and maintain optimality, they have not achieved significant acceleration. To address these challenges, this paper introduces a two-pronged framework: (1) a learning-based TDD approach that employs multiple binary classification techniques to generate a high-quality set of initial decomposition segments (IDSs), which helps in reducing constraint violations across sub-problems; and (2) a sub-problem acceleration approach that utilizes relay learning to expedite the solving of sub-problems while preserving optimality. Simulation results show that our approach can solve yearly time horizon PCMs within tens of seconds with a more than 35% reduction on the number of constraint violations compared to traditional TDD method.