Abstract: In order to reduce the computational costs in conflictdriven clause learning solvers， an improved adaptive heuristic restart strategy named 2WSAT was proposed， which was based on two questions including“when to restart”and“where to restart”. In this strategy， the conflict decision levels and variable’s restart times were taken as important solving state parameters， so that overall performance could be improved by escaping from wrong search branches and choosing better decision variables. Comparative experiments were conducted among 2WSAT and two modern solvers， using the practical application benchmarks. The results showed that 2WSAT has higher scores in many key indicators such as solving speed， memory usage， conflicts number， propagation number etc.

Key words: propositional satisfiability problem, restart strategy, heuristic, conflict decision level, variable’s restart times