高速公路单点入口匝道RLRM控制方法

为缓解交通堵塞,基于人工智能的强化学习理论,提出了不完全信息下的强化学习单点入口匝道控制方法(RLRM)。基于6个仿真实例,分别计算了平均速度、平均密度、流出交通量与旅行时间,比较了无控制、定时控制与RLRM控制的控制效果。仿真结果表明:在交通量较小的实例1中,以旅行时间为评价指标,定时控制与RLRM控制的交通阻塞缓解率分别为-6.25%、-9.38%,几乎没有控制效果;在交通量变大的实例3中,以旅行时间为评价指标,定时控制与RLRM控制的交通阻塞缓解率分别为-8.19%、3.51%,匝道控制有一定效果,RLRM控制略优于定时控制;在交通量最大的实例6中,以平均速度、平均密度、流出交通量与旅行时间为评价指标,定时控制的交通阻塞缓解率分别为8.20%、0.39%、18.97%与23.99%,RLRM控制的交通阻塞缓解率分别为18.18%、3.42%、30.65%与44.41%,RLRM控制明显优于定时控制。可见,交通量越大,RLRM控制效果越明显。

RLRM control method of single entrance ramp for highway

In order to relieve freeway traffic congestion,reinforcement learning ramp metering(RLRM) control method for single entrance ramp of highway under the incomplete information was proposed based on the artificial intelligence theories of reinforcement learning.Average speeds,average densities,traffic outflows and travel times of numerical cases 1-6 were calculated,and the control effect of RLRM was compared with no control and fixed-time control.Simulation result shows that in case 1 with the lowest traffic inflow,the congestion relief rates of fixed-time control and RLRM control depending on travel time are-6.25% and-9.38% respectively,which indicates that the control effect is not significant.When the traffic inflow increases in case 3,the congestion relief rates of fixed-time control and RLRM control depending on travel time are-8.19% and 3.51% respectively,which indicates that the control has some effect,and RLRM control performs better than fixed-time control.In case 6 with the highest traffic inflow,the congestion relief rates of fixed-time control are 8.20%,0.39%,18.97% and 23.99% respectively,and those of RLRM control are 18.18%,3.42%,30.65% and 44.41% taking average speed,average density,traffic outflow and travel time as evaluating indexes respectively,which shows that RLRM control effect is more significant than fixed-time control.So the greater the traffic inflow is,the better the control effect of RLRM is.5 tabs,14 figs,16 refs.