基于LSTM-GM神经网络模型的深基坑沉降变形预测

Settlement deformation prediction of deep foundation pit based on LSTM-GM neural network model

针对基于少量样本的长短记忆时(LSTM)神经网络深基坑变形预测精度较低的问题,提出组合的LSTM-GM预测模型,并运用于地铁深基坑变形预测。将LSTM预测结果的波动项,采用灰色模型(GM)对波动项进行循环预测,满足阈值则完成循环。通过3组不同样本数据的实验,结果表明组合模型在少量样本情况下预测精度高于LSTM模型。此外,将该模型与适用于深基坑变形预测的BP神经网络预测模型和支持向量机(SVM)预测模型对比,分析发现组合模型结合了LSTM预测模型和GM预测模型的优势,拥有更好的预测效果,预测结果趋势符合实际。

Considering the problem that the precision of deep foundation pit deformation is low based on a small amount of samples for LSTM neural network, a combined LSTM-GM prediction model is proposed. The steps of cyclic prediction of the fluctuating items obtained by LSTM are designed, and the fluctuation terms are optimized by using gray prediction(GM) model, and then the LSTM-GM prediction model is applied to the deformation prediction of deep foundation pit of subway. Through the experiment of three groups of different sample data, the results show that the prediction accuracy of the combined model under the condition of a small amount of samples is higher than the LSTM. In addition, the model is compared with BP model and support vector machine model applicable to deep foundation pit deformation prediction. It is found that the combined model integrates the advantages of LSTM model and GM model, and has better prediction effect.