人工神经网络结合遗传算法反演岩体初始地应力的研究

提出综合应用实数编码的遗传算法与改进的BP神经网络的优化反演分析方法,并通过数值分析,探讨了该方法在应用于位移反演岩体初始应力与材料参数方面的有效性.在算例中,以Burgers模型的计算数据作为改进神经网络的训练样本,用遗传算法搜索待反演参数解向量.计算结果表明利用遗传算法优化神经网络权值能提高神经网络迭代算法的效率与可靠性.该方法应用于岩体初始应力场的反演具有迭代过程平稳、收敛快、结果准确等特点,能够有效地求得岩体初始应力与材料参数.

The combination of artificial neural network and genetic algorithm applied to inversion analysis of initial stress fields in rock masses

Based on the integral application of genetic algorithm with real number code and improved BP artificial neural network, an optimum inversion method is proposed. By numerical analysis, the availability of this method for inversion analysis of the initial stress field in rock masses and the material parameters is discussed. In calculation example, the calculative data of Burgers model is used as training sample and the inversion parameter vectors are searched by genetic algorithm. The calculation result indicates that the optimization of artificial neural network weight can increase the efficiency and reliability of artificial neural network calculation .The application of this method to inversion analysis of the rock initial stress field and material parameters has the following advantages: calculation stabilization, prompt convergence and preferable precision. By all appearances this application is valuable.