基于时间序列ACO-BP神经网络在基坑变形预测中的应用研究

BP神经网络技术因其良好的非线性动力学特性、函数逼近能力、自组织和自适应能力,已广泛应用于基坑变形预测中。但实际应用过程中发现BP神经网络具有收敛速度慢、初始权阈值对计算结果影响较大,且易陷入局部最优等缺陷。采用引入具有启发式寻优、全局优化特点的蚁群算法优化BP神经网络,对基坑变形进行预测,并与BP神经网络进行比较。结果表明:ACO-BP神经网络模型预测基坑变形可行;预测精度高于BP模型,且结果稳定、速度较快、误差满足工程的要求。

Study on Application of Time Sequence Based ACO-BP Neural Network in Prediction of Foundation Pit Deformation

BP neural network technology, due to its excellent nonlinear dynamic characteristics and its good performance in function approximation, self-assembling and self-adaptivity, has been applied to prediction of foundation pit deformation widely; however, it is found in actual application to be imperfect with slow convergence, large impact of initial weight threshold on calculation result, and easy to be trapped into local optimization. Therefore, Ant Colony Optimization, which is featured by heuristic optimizing and global optimization, was introduced to optimize BP neural network and predict the deformation, then the results of the both methods were compared. The result shows that ACO-BP neural network model is feasible in prediction of foundation pit deformation with higher accuracy than BP model; and the prediction is of stable result, fast speed and the error meeting the project requirement.