基于PSO的GM-BP神经网络在基坑预测中的研究

Research on GM-BP Neural Network Based on PSO in Foundation Pit Prediction

基坑监测对确保基坑工程的顺利实施和周边建筑物的安全尤为重要。由于传统的GM-BP神经网络基坑预测算法在预测过程中连接权值和阈值容易在局部产生最小值,导致无法准确预测基坑周边地面的沉降量。因此,该文提出基于PSO算法的全局搜索优化性,对传统的GM-BP神经网络预测算法中的连接权值和阈值进行不断迭代和优化,并结合基坑的实际监测时间序列验证改进前后预测算法的精度。研究结果表明:改进的基坑预测算法在离基坑5、10、15m三处的RMSE、MAE、MAPE都小于传统的基坑预测算法结果。该方法提高了预测精度,为基坑预测分析提供了技术参考。

The foundation pit monitoring is very important to ensure the smooth implementation of foundation pit engineering and the safety of surrounding buildings.The traditional GM-BP neural network prediction algorithm can not accurately predict the settlement of the ground around the foundation pit,because it is easy to generate the minimum value in the local area of connection weights and thresholds in the prediction process.Therefore,in this paper,the global search optimization based on PSO algorithm is proposed,the connection weights and thresholds in the traditional GM-BP neural network prediction algorithm are continuously iterated and optimized,and the accuracy of the prediction algorithm before and after the improvement is verified by combining with the actual monitoring time series of foundation pit.The results show that the RMSE,MAE and MAPE of the improved prediction algorithm are smaller than those of the traditional prediction algorithm at 5、10 and 15 m away from the foundation pit.This method improves the prediction accuracy and provides technical reference for the prediction and analysis of foundation pit.