多种会遇状态下基于强化学习的船舶自动避碰路径仿真

为实现海域中多会遇局面的船舶自动避碰,依据船舶间状态信息的实时获取,构建了自适应启发评价的船舶避碰算法。算法中通过再励信号由强化学习训练得到船舶从会遇形成到驶过让清动态过程的优化转向决策,模拟给出了两船会遇局面、多船会遇局面下的避碰优化路径。结果表明算法能够满足避碰规则并完成安全避碰,与已有分布式避碰决策方法对比,显示了更好的实时性和经济性。

Ship automatic anti-collision path simulations based on Reinforcement learning in different encounter situations

On the basis of real-time state information between ships, an adaptive heuristic critic algorithm for ship automatic anti-collision was presented in order to deal with different encounter situations. In the algorithm, the Optimization decision was formed by reinforcement learning trained in the whole dynamic process. The simulations of shipScollision avoidance paths were presented in different encounter situations. Results show that give-way ships are complied with COLREGs and implement secure anti-collision. Compared with the existing distributed anti-collision decision, it shows better real-time operation and economy.