基于AGA-BP系统的涵洞结构应力预测研究

Study About Culvert Structure Stress Forecast Based on AGA-BP System

为了解决神经网络过训练和经验公式确定网络结构盲目性的问题,更好地实现对高填土涵洞结构应力预测研究,结合遗传算法和BP网络两种智能方法各自优点开发出自适应遗传算法-神经网络(AGA-BP)系统.该系统引进自适应交叉、变异概率公式,改进传统的交叉、变异遗传操作,提高局部搜索能力;同时将BP网络累积预测误差标准差作为自适应遗传算法的适应度函数,并作为遗传终止的判定准则之一.以模型试验数据为样本,通过AGA-BP系统和经验公式两种方法确定网络结构,实现对涵洞应力的预测.预测结果对比验证了AGA-BP系统在高填土涵洞应力预测研究中的可行性及优越性.

To solve the problem of neural network overtraining and blind network structure due to empirical formulae so as to forecast the stress in highly filled culvert better, an AGA-BP network system is developed via combining the advantages of GA and BP network together. By introducing the adaptive cross and mutative probability formula into the system to improve the conventional cross and mutative genetic operation, the local search ability is enhanced. Meanwhile, AGA fitness function is defined as the standard deviation of BP network＇s accumulation forecast errors and also one of genetic termination criteria. Taking the model test data as sample, the network architecture is determined by two methods, i.e. AGA-BP system and empirical formulae, to forecast the culvert stress. The results of both methods are compared with each other, and the AGA-BP system is verified more feasible and excellent in forecasting the highly filled culvert stress.