MLPP-DCS: Machine Learning-Based Performance Prediction for Discrete Controller Synthesis

Discrete Controller Synthesis (DCS) is an automated technology designed to synthesize controllers that ensure safety within specified environments. It utilizes environment and monitor models described by Labelled Transition Systems. However, DCS faces a significant challenge: the exponential growth of computational space as the increase in the number of environment models. This paper introduces Machine Learning-based Performance Prediction for DCS (MLPP-DCS), aimed at predicting the computation time and maximum memory usage required for DCS. MLPP-DCS generates predictors using machine learning (ML)-based algorithms that take the characteristics of the environment and monitor models as inputs. Our predictor generation algorithms leverage four ML techniques. Evaluation results demonstrate its effectiveness, achieving a coefficient of determination of up to 0.814 for predicting computation time and 0.536 for predicting maximum memory usage.