Constitutive Models Of Reinforced Soils Based On Artificial Intelligence Methods

As a kind of new geotechnical composite material, reinforced soil has many advantages: With a good stability, it is easy for construction and costs much lower than with other materials, thus it provides civil engineers with many economical means. reinforced soil technology has been widely applied in reinforced retaining structures, reinforced slope and reinforced soft ground in water conservancy, highway, railway, port and architecture building. With the application of civil engineering synthetic materials in modern geotechnical engineering, the application of reinforced soil technology will be more widespread, so the theoretical study of reinforced soil is even more necessary.Because reinforcement material is added to the earth, the characteristics of the earth is changed, and the former constitutive model based on soil is hardly enough to explain reinforced soil. This thesis is based upon artificial intelligence method, adopts grassroots-reinforced soil(GRS) triaxial test data training intelligence network,self-act generating network parameter, and gets reinforced soil BP neural network constitutive model, RBF neural network constitutive model,ANFIS constitutive model,thus avoiding the difficulties of mathematical modeling determine function parameter.(1) The GRS BP neural network constitutive model are established,and the training error and the checking error are all small, under the training and checking process the curves of model simulating are in accordance with experimental curves,and it also modifies the mutation point caused by experimental errors on maximum principle stress and grassroot content relation curves,and proves that this network has a good fault-tolerant ability and high accuracy, which enables it to be used in making GRS constitutive model. The reinforced soil stress-strain predicted curve of different grassroot content simulated by the model are also in accordance with maximum principle stress and grassroot content relation curves,so the model has good generalization ability.(2) The GRS RBF neural network constitutive model are established, and the influence on network precision by training sample scale and error-controlling is discussed. The larger the training sample scale is, the better it is for the study of the data inner regularity by the network, as when the training error is too small, too fitting may appear, which will result in the decrease of fault-tolerant ability; When too large, it may lead to the inadequacy of the study of the inner regularity of the training data.(3) The ANFIS constitutive model are established, The reinforced soil stress-strain predicted curves of different grassroot content simulated by the model are smooth, and also in accordance with maximum principle stress and grassroot content relation curves, so the model has good generalization ability.(4) The comparison between predicted results simulated by the three intelligence network constitutive methods, suggests that ANFIS model has a better simulative accuracy, fault- tolerant ability and generalization ability, and shows the strong reasoning ability of ANFIS can better extract internal regulatity of reinforced soil.