基于模式-遗传-神经网络的流变参数反演

介绍了一种岩石流变多参数反演的智能方法。该方法把遗传算法和神经网络有机结合起来,并在遗传算法中嵌入模式搜索加速优化进程;该方法基于均匀设计获得的样本进行神经网络学习,用模式–遗传–神经网络进行岩体流变参数的最优辩识。该方法用经过最佳预测学习算法训练的神经网络来表达岩体流变参数和位移之间的映射关系,除具有一般遗传算法的优点外,还提高了参数反演的精度,节省了参数反演的计算时间,使得某些原来用传统优化方法在时间上几乎无法进行的参数反演如今变为可能,并用工程实例验证了此方法的可行性与优越性。

BACK ANALYSIS ON RHEOLOGICAL PARAMETERS BASED ON PATTERN-GENETIC-NEURAL NETWORK

An intelligent algorithm by which multiple rheological parameters of rock can be analyzed simultaneously is proposed. This method,namely the pattern-genetic-neural network algorithm (PGNNA), naturally combines pattern search (PS),genetic algorithm (GA),and neural network (NN). The samples produced by uniform design are used to train NN whose architecture is determined in global optimization by pattern-genetic algorithm(PGA). NN that has optimal architecture and has been trained by optimal prediction algorithm is used to describe relationship between the rock rheological parameters and displacement. Rheological parameters are searched in global space by PGNNA,instead of a certain numerical calculation. This method improves the precision of back analysis on parameters,shorts the time of calculation,which is almost impossible for some traditional methods because of the long time of calculation. The practical engineering example shows feasibility and advantages of this method.