Research On Experiment And Prediction Model Of Soil Consolidation Coefficient

With the development of social economy,various types of large-scale construction projects are gradually rising.However,in many cases,the construction conditions of the proposed area are more complex,and the foundation soil may have low strength,easy deformation and other characteristics.Therefore,the stability of engineering construction is often the main concern of people.In order to improve the stability of the project construction,it is very important to predict the deformation of the foundation soil accurately.Therefore,as one of the key parameters that affect the calculation of foundation deformation and foundation settlement,consolidation coefficient is of great significance.So far,how to obtain consolidation coefficient is mainly divided into the following four methods: Indoor test method represented by time logarithm method and time square root method.The error of this method is greatly influenced by human factors;The field test method,represented by the spiral plate load test method and the hole static pressure contact method,can not be judged the disturbance of the spiral plate to the soil;The inversion analysis method,represented by pore water pressure inversion,has long monitoring cycle and large capital cost;The mathematical prediction method represented by neural network is difficult to avoid problems that the algorithm itself has.In order to avoid the shortcomings of the existing research methods,the consolidation coefficients were calculated by three different methods respectively,including seepage consolidation test method and the measured pore pressure calculation method based on the simulation test of formation subsidence and support vector machine prediction method.The prediction results of neural network algorithm were compared with the prediction results of support vector machines.Details are as follows:1)The compressibility coefficient and permeability coefficient of soil were measured by self-made seepage consolidation apparatus.According to the calculation formula of consolidation coefficient,the indoor test value of consolidation coefficient of soil sample was obtained.Through the analysis of the consolidation coefficient obtained from the test,it is found that the consolidation coefficient is a variable value with the change of the external load.And as the load increases,the consolidation coefficient increases gradually.2)The pore water pressure sensor was embedded in the soil of the settlement simulation test,and the settlement simulation test was carried out to obtain the measured values of pore water pressure during the settlement process,and the consolidation coefficient was calculated by the measured pore pressure.The results showed that the consolidation coefficient obtained by this method was in the same order of magnitude as that obtained by the permeation consolidation test.Comparing the consolidation coefficient in different layers,it can be concluded that the coefficient of consolidation increases gradually with the increase of load,which was consistent with the results obtained by the first method.3)Based on the soil indexes and consolidation coefficient determined by laboratory tests,the factors affecting consolidation coefficient were determined by repeated tests and reference data.The principal components of these factors were determined by principal component analysis,and the support vector machines optimized by particle swarm optimization algorithm were used to predict the coefficient of consolidation.Through the prediction of test samples,the feasibility of model prediction was proved,and then the model was used to predict test samples.The results show that the error of the consolidation coefficient predicted by the support vector machine model based on principal component analysis and particle swarm optimization is smaller,and it is an effective method to predict the consolidation coefficient.4)LM-BP algorithm was used to predict consolidation coefficient with the samples SVM predicting.The results show that the prediction error of the LM-BP is larger than that of the SVM.This is due to the unavoidable problems of neural networks including Local convergence,over learning and lack of learning.