强化学习在路测覆盖分析中的实践

Practice of Reinforcement Learning in Road Test Coverage Analysis

传统分析路测问题时存在耗时费力、效率低下的问题,文章针对该问题提出了一种基于值函数迭代的Q学习算法的路测覆盖智能分析方法。该算法建立在覆盖状态规则和弱覆盖现象识别规则的基础上,训练两个基于Q学习的规则学习器,同时引入经验池、状态合并和决策树探索方法。实验证明,能在覆盖状况识别和弱覆盖现象判定方面取得良好的效果,实现了道路弱覆盖识别和智能分析判定的功能。

The traditional analysis of road test problems is time-consuming and inefficient.In this paper,a Q-learning algorithm based on value function iteration is proposed.The algorithm is based on covering state rules and weak covering phenomena recognition rules.It trains two rules learners based on Q-learning,and introduces experience pool,state merging and decision tree exploration methods.Experiments show that it could achieve good results in the recognition of coverage state and the judgment of weak coverage phenomena.The function of weak coverage recognition and intelligent analysis and judgment are realized.