基于多种群遗传算法的概率积分法参数反演

Parameters Inversion of Probability Integral Method Based on Multi-population Genetic Algorithm

为了弥补标准遗传算法(SGA)求取概率积分法预计参数的早熟收敛,后期易陷入局部最优解的缺点,提出了多种群遗传算法(MPGA)来反演概率积分法参数,研究了该算法的准确性与可靠性。模拟试验表明:基于MPGA的概率积分法参数反演模型不仅能够准确求取预计参数,而且对于观测站数据中的随机误差、粗差和监测点缺失都具有较强的抗干扰能力。试验表明:在MPGA只迭代了57次就收敛,然而SGA迭代了100次才收敛的情况下,MPGA得出的下沉值和水平移动值的拟合标准差是31 mm,SGA得出的下沉值和水平移动值的拟合标准差是32 mm。

In order to make up for the precocious convergence of the expected parameters of the standard genetic algorithm(SGA)for the probability integration method, it is easy to fall into the disadvantages of local optimal solution in the later stage. A variety of group genetic algorithms(MPGAs) were proposed to invert the parameters of probability integration, and the accuracy and reliability of the algorithm were studied. The simulation experiment shows that the inversion model of probability integral method based on MPGA can not only accurately obtain the expected parameters, but also have strong anti-jamming ability for random error, coarse difference and monitoring point loss in the station data. The probability integral method test shows that: in the case that MPGA only iterates for 57 times before convergence, while SGA iterates for 100 times before convergence, the fitting standard deviation of the subsidence value and horizontal movement value obtained by MPGA is 31 mm, and the fitting standard deviation of the subsidence value and horizontal movement value obtained by SGA is 32 mm.