【摘要】 首先建立泊松分布参数多变点模型,给出该分布参数多变点的似然函数,探究变点位置参数和分布参数的满条件后验分布。利用可逆跳跃马尔科夫链蒙特卡洛(RJMCMC)算法确定该模型中变点的个数,在变点个数确定的基础上,进一步利用马尔科夫链蒙特卡洛(MCMC)方法中的Gibbs抽样和Metropolis-Hastings算法对参数满条件后验分布进行抽样,利用抽样均值和最大后验法对变点位置参数和分布参数进行估计。仿真结果和美国矿难实例均表明,结合RJMCMC算法和普通MCMC方法对泊松分布序列的变点检测很有效。更多还原

【Abstract】 For detection problem of poisson distribution parameter change point, the change-point model of Poisson distribution parameters is established and the likelihood function of the change points of poisson distribution parameters is given to explore the full conditional posterior distribution of the positional parameters and distribution parameters of the change-point.RJMCMC algorithm is used to determine the number of change points in the model.On the basis of determining the number of change points, Gibbs sampling and MH algorithm in MCMC method are further used to sample the posterior distribution of parameter with full condition.The sampling mean value and maximum posterior methods are used to estimate the position and distribution parameters of change points.The simulation results and the example of American mine accident show that RJMCMC algorithm combined with ordinary MCMC method is very effective for change point detection of Poisson distribution sequences.更多还原