基于GA-ANN的苏锡常地裂缝危险性评价

Evaluation of earth fissures based on GA-ANN coupling model in the Suzhou-Wuxi-Changzhou area

文章以苏锡常地区地裂缝危险性评价为例,利用遗传算法(GA)对人工神经网络(ANN)进行改进,先用GA优化BP网络初始权重,再用BP算法修改网络权重,实现不同尺度的同步调整。选择30点的不同地质条件组成样本对所建模型进行训练,评价指标包括:基岩埋深、基岩起伏度、地下水位、地面沉降梯度、含水层导水系数和粘性土层厚度。经过500次GA迭代,得到苏锡常地区地裂缝ANN模型的最佳权重组合,该耦合模型能对全区地裂缝地质条件进行正确分类,精度接近1‰。

Although the ANN （artificial neural network） method has displayed some potential on the earth fissures assessment, the converging problem of the simulation procedure at local proximity is still existing. The genic arithmetic （GA） is more suitable for whole - scale search than for finding local exact key. Thus, they should be improved by each other. In this article, a model set up by coupling GA and ANN is used to evaluate the hazard of earth fissures in the Suzhou-Wuxi-Changzhou area. On the one hand, the initial weight series of ANN are modified by GA, on the other hand, they can also be adjusted by BP （Back Propagation ） algorithm. Thirty earth fissures of different geological conditions are selected for the ANN training. The assessment factors consist of depth of bedrock, undulation of bedrock, depth to groundwater level, grade of land subsidence, conductibility of aquifer, thickness of clay. After 500 times iterative calculations of GA, the best weight series for the ANN is found. The whole area is classified into four degrees according to the assessment results of earth fissures. Practice results show that the earth fissures can be simulated with an accuracy of about 1‰ with this improved model.