Performance Settlement Prediction Of Composite Foundation Improved By Cement-soil Piles Under Cyclic Loading

China’s coastal areas and some inland cities are widely distributed in the deep soft clay layer, with the rapid economic development of the region and contiuous improvement level of urbanization, It is inevitable that building high-speed railway,the highway in soft soil area, However, due to the soft soil has a low intensity, large deformation and a long duration, high water content and permeability of poor characteristics, and long-term cyclic loading subsidence problems are still outstanding, all kinds of traffic load on the roadbed can be seen as a long-term cyclic loading. so to find a reasonable settlement prediction methods accurately predict the work of the composite foundation under cyclic loading settlement, so in order to control the subgrade settlement more effectively and more economical and ensure the road traffic safety and fast operation, finding a reasonable settlement prediction methods to accurately predict the composite foundation settlement under cyclic loading has important practical significance.At present, the research about performance settlement prediction of composite foundation improved by cement-soil piles under static loading has made some achievements. But the present study for composite foundation settlement prediction under cyclic loading is less. In this paper, The author have used the artificial neural network model and genetic program model two artificial intelligence methods to predict the performance settlement of composite foundation improved by cement-soil piles under cyclic loading ,making the indoor large model test results as research samples.Two different predicting methods have been proposed by the author. The first method is to make use of artificial neural network model, using powerful neural network nonlinear mapping and learning ability, based on MATLAB7.0 platform programming. Making a large number of test results that indoor composite foundation model test under cyclic loading has completed as a sample, through the training of the neural network structure, find the internal relationship between the settlement and its main influencing factors (such as: cement ratio, the cyclic stress ratio, the replacement law, loading, etc.). According to these, the composite foundation under cyclic loading settlement prediction method has been proposed based on artificial neural networks. The second is the genetic programming of this new forecast method for cement-pile composite foundation permanent settlement prediction in geotechnical engineering, Establishing corresponding theoretical model and calculating analyze the engineering parameters of the cement-pile composite foundation settlement processing, then find a specific mathematical model. That is, based on the collected a large number of experimental material, using genetic program method for the simulation test, through the genetic programming reproduction, exchange and mutation genetic operation set up a GP model. This model can be established the nonlinear relationship between the permanent settlement and influence factors, then gives the explicit expression of function. While the predictions of these two methods were compared with ordinary linear regression model, From three different settlement prediction model analysis and comparison can be seen, the maximum error of the BP model was 7.64%, with an average error of 5.58%, significantly better than traditional linear regression model, has a high prediction accuracy. And the prediction accuracy of GP is the highest, the maximum error is only 5.86%, the minimum error is 1.22%. The most important is to simulated an explicit mathematical expression, which is significantly better than the neural network model. It provide a new way for the performance settlement prediction of composite foundation improved by cement-soil piles under cyclic loading.