一种改进遗传神经网络的建筑基坑沉降预测模型

目前常见的沉降预测方法有灰色系统模型、时间序列分析法、BP神经网络及其改进算法等。针对BP神经网络容易出现过拟合和局部最优的缺点,部分学者利用遗传算法进行神经网络初始权值和阈值优化。但是遗传算法对于因监测数据质量问题而造成变形预测结果不佳的优化效果有限。因此引入自适应增强算法对遗传神经网络预测模型进行改进。并利用某高层建筑基坑实测50期监测数据进行仿真预测。实验结果表明,利用自适应增强算法改进之后的遗传神经网络预测模型在满足工程监测精度要求的前提下,在MAPE、MAE、MSE三项精度指标上分别提高80.57%、81.04%、70.83%。

An improved GA-BP network settlement prediction model of building foundation pit

The deformation prediction of foundation pit is of great significance for the early warning of foundation pit deformation.Currently there are several prediction methods such as gray model system,time series analysis method,BP neural network and its improved algorithm.Aiming at the disadvantage that BP neural network is prone to over fitting and local optimization,some scholars use genetic algorithm to optimize the initial weights and thresholds of neural network.However,due to the poor quality of the monitoring data,the improvement effect of the genetic neural network is not of satisfication.An adaboost algorithm is introduced to improve the genetic neural network prediction model,with 50 times measured data of a high-rise building foundation pit.And the experiment result shows that the GA-BP prediction model optimized by adaboost algorithm can meet the requirements of monitoring precision,and three precision indexs MAPE,MAE and MSE increase by 80.57%,81.04% and 70.83%.