基于马尔科夫蒙特卡罗的管制员警觉概率预测

Probability prediction of the controller’s status-in-situ alertness based on the Markov chain Monte Carlo

为分析管制员在工作期间的警觉性变化规律,构建了一种基于马尔科夫蒙特卡罗(MCMC)的警觉性概率预测方法。首先,通过模拟管制员工作环境,利用PVT(Psychomotor Vigilance Test,PVT)获取管制员不同时间段的警觉性值;其次,根据试验数据的结构特点,采用Logistic函数构建警觉性概率预测模型;然后,基于MCMC方法的理论基础,对比M-H抽样与NUTS抽样的结果,选取更适合该模型的抽样方法进行参数估计以获取模型中参数的最优值;最后,通过模型估算特定时刻管制员处于警觉状态的概率,以反映其警觉性状态。结果表明,采用NUTS抽样算法的模型效果更好,管制员警觉与非警觉状态转变的最可能出现时间分别在03:45、06:40、09:05、13:50和15:50。

The present paper is intended to propose a novel alertness prediction model in order to analyze the controller’s alertness status-in-situ during his or her working shift based on the Markov chain Monte Carlo(short for MCMC)theory for the flying safety management.To achieve the research purpose,we have first of all laid out an experiment to gain the controller’s alertness degree data on the mimic controller’s working condition through the PVT(short for Psychomotor Vigilance Test).And,then,the alertness forecast function has to be established via the Logistic function based on the structural features of the experimental data.And,next,we have compared the results of M-H sample with that of the NUTS via the traditional sampling methods in MCMC to choose a better sampling method suitable for the function on the MCMC theoretical basis so as to gain an optimal value of the parameters through the parameter inference.And,last of all,the purpose of the model has been turned up to estimate the result of probability whether the controller is in a vigilant state or not at a particular moment to reflect their status-insitu alertness.The results of our estimation show that the NUTS sampling algorithm enjoys a better effect for convergence and the estimation results of parameters ought to be enough confident.Besides,when the parameters of the model are confirmed through calculation,the most likely transition moments for the controller’s alertness and non-alertness turn out to be at 03:45,06:40,09:05,13:50,and 15:50,respectively,with the remaining moments being from the outcome curve.The curve can also be used as the basis of the controller’s schedule and evaluate if the working schedule is suitable for them based on their own alertness status.And,since the measuring method can be simulated to predict the controllers’alertness change during their work shift,and the relationship between the scheduled plan and alertness can also be analyzed.Thus,the simulation model and the experiment method the given paper has proposed can provide a novel idea for the alertness prediction and evaluation.