Fast and explainable warm-start point learning for AC Optimal Power Flow using decision tree

Abstract

The quality of starting point greatly influences the result and convergence efficiency of the optimization algorithm, especially for the non-convex and constrained Alternating Current Optimal Power Flow problem. Generally, speed and accuracy are the two main evaluation metrics for generating starting points. The data-driven methods learn the starting point through historical data and show good performance. However, most methods utilize “black-box” models, which lack interpretability. Therefore, this paper proposes a fast and explainable warm-start point learning method based on the multi-target binary decision tree with a post-pruning module. The calculated warm-start points can accelerate the solving process and the model inference time is extremely short. The post-pruning module is applied to fit different power system scenarios fairly and alleviate the overfitting problem by pruning the completely grown tree. Also, a set of detailed decision rules for selecting warm-start points are generated after the learning process. The generated rules assist the power system operators in identifying important loads and thereby provide the model interpretability. The experiment shows that the proposed framework can reduce the solving times for the Alternating Current Optimal Power Flow solvers with an extremely short calculation time for the explainable warm-start point.