基于BP神经网络的MICP加固滨海软土强度仿真和预测

Simulation and Prediction of Strength of MICP Reinforced Coastal Soft Soil Based on BP Neural Network

摘要: MICP是一种环保且有效的滨海软土加固方法，它是利用微生物分解尿素生成碳酸根，然后与钙源反应生成碳酸钙沉淀，填充土颗粒间孔隙的同时胶结分散的土颗粒进而改善其工程力学性能。但加固后土体的强度受营养液浓度、菌液浓度的影响较大，需要通过大量的试验找寻其规律。为了减少成本，通过Matlab软件构建BP神经网络模型，使用624组数据对模型进行训练，再对加固后土体的156组强度数据进行预测。研究结果表明：经过大量的试验数据训练后，预测的剪应力与实际试验剪应力差异范围在−5.370%到4.238%之间，决定系数R²均大于0.9。验证了通过BP神经网络预测MICP加固滨海软土强度是可靠的，可减少试验工作量，为实际工程节约成本，为日后类似加固土的研究提供指导。

Abstract: MICP is an environmentally friendly and effective method for strengthening coastal soft soil. It uses microorganisms to decompose urea to generate carbonate ions, which then react with calcium sources to form calcium carbonate precipitates. It fills the pores between soil particles while cementing dispersed soil particles to improve their engineering mechanical properties. However, the strength of the reinforced soil is greatly affected by the concentration of nutrient solution and bacterial solution, and a large number of experiments are needed to find their patterns. In order to reduce costs, a BP neural network model was constructed using Matlab software, trained with 624 sets of data, and predicted with 156 sets of strength data of the reinforced soil. The research results indicate that after a large amount of experimental data training, the difference between the predicted shear stress and the actual experimental shear stress ranges from −5.370% to 4.238%, and the determination coefficient R² is greater than 0.9. Verified the reliability of using BP neural network to predict the strength of MICP reinforced coastal soft soil, which can reduce experimental workload, save costs for practical engineering, and provide guidance for future research on similar reinforced soil.