Optimizing Parameters of Signal Temporal Logic Formulas with Local Search

Abstract

Signal temporal logic (STL) is a formal language for expressing temporal and real-time properties of real valued signals. In this paper, we study the problem of generating an STL formula from a labeled dataset. We propose a local search algorithm to synthesize parameters of a template formula. Starting from a random initial point, the parameter space is explored in the directions improving the formula evaluation. In addition, the local search method is integrated to the genetic algorithms developed for formula synthesis as the adaptation step. The findings of the paper are shown on a case study and compared with the previous results, which shows that the adaptation step improves the convergence.