基于L-M算法优化的GA-BP网络在岩石高边坡损伤识别中的应用

Damage Identification of High Rock Slopes Based on GA-BP Network Optimized by L-M Algorithm

边坡的失稳破坏往往是由岩体损伤导致的,研究边坡岩体的损伤识别对于工程实践具有重大意义,而基于BP神经网络的损伤识别是边坡损伤识别的有效方法之一.本文针对传统BP神经网络在边坡损伤识别中存在的问题,首先利用遗传算法(GA)克服BP网络易陷入局部极小的不足,然后利用Levenberg-Marquardt(L-M)算法在解空间里对BP神经网络进行精调,搜索出最优解.结合优化后的GA-BP模型以及有限元计算结果,提取边坡损伤前后的频率值来实现损伤位置和程度的识别.经验证,改进后的BP网络有效提升了识别边坡损伤的性能,对岩石高边坡损伤识别方法的研究有理论指导意义与参考价值.

The failure of slope is often caused by damage of rock mass;so it is very important to study the damage identification of rock mass in engineering practice.The method based on BP neural network is an effective approach to damage identification of rock slope.In light of the drawbacks of damage identification of rock slope based on traditional BP neural network,first,the BP neural network is optimized by genetic algorithm(GA)to avoid local minima in the scheme;and then the BP neural network is finely tuned with Levenberg-Marquardt(L-M)algorithm in the local solution space to search the optimum solution or approximate optimal solutions.Combining with the optimized GA-BP model and finite element calculation results,this paper extracted the frequency value of the slope before and after damage to recognize location and degree of the damage.It has proved that the improved BP network effectively improved the performance of recognizing slope damage;and it has theoretical guidance meaning and reference value to the study of damage identification method of high rock slope.