Preliminary Study On Nonlinear Seismic Response Calculation Based On Neural Network Constitutive Model

The prediction and evaluation of seismic response under structural earthquakes is the leading and prepared work of seismic research,which greatly affects the accuracy of seismic research.Studying how to quickly and accurately calculate the response under earthquake is of great practical significance to design of seismic resistance,effectively mitigate earthquake disasters,guarantee the safety of people's property and ensure the sustainable development of society and economy.Based on the research of nonlinear response calculation under earthquakes,the related researches are carried out from two aspects:step-by-step integration method of dynamic equation and recognition of resilience model.The main work and achievements are as follows:?1?In the aspect of solving the dynamic equation,the algorithm mechanism of the commonly used stepwise integration algorithm is introduced.Based on the precise integration method,the HHT-?coupling precise integration method?HHT-PIM?is proposed.By introducing HHT-?for the assumption of the dynamic equation and the acceleration term,the simplify and reduce order of state equation and exponential matrix in the traditional precise integration method are realized.The theory and examples are used to analyze the algorithmic characteristics of the HHT-?coupled precise integration method.The stability is conditionally stable but the stability conditions are easy to satisfy.The precision is high and the algorithm damping could be adjusted as well as the controllable characteristics of the high-frequency respons.The calculation efficiency is also greatly improved by the matrix reduction.?2?In the aspect of resilience model identification,the eight-path hysteresis model neural network recognition theory and the corresponding complete input variable group are proposed,which could cover and identify the linear and nonlinear working states in the structural hysteresis model and ensute the single-valued mapping.Based on the eight-path hysteresis model neural network recognition theory,a 11 input variable resilience recognition neural network method is proposed,which is input variable group[??_n,?_n,x_n,?,x_history+,x_{his tory-},R_{his tory+},R_{his tory-},E_n-1,x_n-1,R_n-1]can achieve effective prediction of single output variable resilience[R_n].And algorithm development with[11-25-25-1]layer neural network architecture is achieved.The resilience prediction results under the training sample of ground motion and quasi-static loading are well fitted with the OPENSEES simulation results.?3?The general flow of the nonlinear response of seismic structures based on the dynamic equation solving module and the identification of identification module is proposed.The response calculation method and its substructure method which use the HHT-PIM method as the dynamic equation solving module and use the 11 input variable neural network method as the resilience identification module are proposed.The accuracy and feasibility of the proposed method and its substructure method are verified by SDOF and MDOF examples respectively.The example shows that the method has good precision under both SDOF and MDOF conditions and could fit the structural nonlinear ground motion response well.The existing errors mainly come from structural nonlinearity and error accumulation of neural network input variables.The structural nonlinear response method and its substructure method of the resilience recognition module based on the variable neural network method proposed in this paper can obtain the real seismic hysteresis behavior of the structure,avoiding the need to presuppose the restoration of components or materials in the traditional numerical method.The force constitutive model prevents the risk of the distortion of the response calculation result caused by the resilience model selected by the numerical modeling from being mismatched with the actual working condition,effectively improves the calculation accuracy of the structural seismic response,and is unknown or impossible for the restorative force constitutive model.The structure or component explicitly expressed by the function relationship can also effectively realize the response solution and has high use value.