基于粗糙集-模糊评判-神经网络的隧道施工安全状态评估

Renovated security status evaluation system for tunnel construction based on the rough setsfuzzy evaluation-neural network

为了有效地对隧道施工进行安全状态评估,建立了以人-机-环-管理系统为基础的隧道施工安全状态评估指标体系,构造了基于粗糙集-模糊评判-神经网络的隧道施工安全状态评估模型。该模型通过粗糙集约简输入变量,提炼学习样本,再利用神经网络对其进行训练和评价,提出了使用层次分析法和模糊数学的方法对隧道施工安全整体综合评判,得到的评价值作为神经网络训练目标值的方法。实际结果表明,通过使用该模型方法,神经网络训练的条件属性由原来的17个变成6个,训练周期由原来的2992次减少为1637次,泛化能力、安全状态评估的结果都优于约减前,能够对隧道施工安全状态做出有效的评估结论。

This paper is aimed to set up an evaluation index system of the safety condition in tunnel construction based on the man-machine-medium-management and manipulation. The system has been established on the theoretical basis of rough sets, fuzzy evaluation and neural network for evaluating the safety conditions in tunnel construction. More specifically speaking, such a system involves a lot of factors that are likely to affect the tunnel safety, the model projects when the variables were input through the reduction of rough sets to refine a sample set. Such a set has to be trained and evaluated by using a neural net so that a comprehensive and integrated assessment of the safety conditions of tunnel construction can be confirmed via hierarchical analysis and fuzzy mathematics. The assessment has to be trained as the target value of neural network training. In order to heighten the reliability, it is necessary to use a great number of samples for testing the disadvantages of the speed of neural network training. The proposed model intends to avoid its disadvantages of being too complicated and the too-long training period. In addition, it is necessary to solve the complicated nonlinear relation between the safety condition evaluation results and the evaluation components. Besides, it is also necessary to reduce the condition attributes of the neural network from the original 16 to 6 while the training circle should be reduced from the current 2 992 to 1 637. At the same time, the generalizing power and the outcome of the safety condition evaluation are likely to prove to be better than that without reduction. As a result, we could make a more effective judgment of safety condition of tunnel construction. Thus, it can be said that the proposed model comes into conformity with the engineering practice, which can provide effective safety management basis for the safety condition evaluation in tunnel construction with complicated geo-environment.