Global Surrogate Model With Temporal-spatial Basis And Its Application In Parameter Inversion Of Dam

The feedback analysis of dam safety monitoring data is a necessary means to understand its real working conditions.The research on this problem is of great significance to the safety of the dam and has drew sustained and extensive attention worldwide.The development of "digital dam" has produced a multi-point,multi-effect dynamic measurement sequence.It is worth focusing on using the increasing monitoring data to understand the real work conditions of the dam structure timely and efficiently.In this dissertation,the whole process of the feedback analysis of dam structure is studied from two aspects: parameter inversion based on physical model and data analysis based on statistical model,to improve the efficiency of inversion of structural parameters and to promote timely and effective use of the monitoring data.A rapid feedback analysis platform for dam safety monitoring data is established.Main achievements of the dissertation are as follows,(1)Based on the concept of basis vector,a global surrogate model with temporalspatial basis is established.The representative basis vector of the response field in time and space dimension is extracted by using the two-layer proper orthogonal decomposition,and the response field is expressed as the linear combination of the basis states expanded by the temporal and spatial basis vectors.The theoretical equations are derived to obtain the combined coefficients of the sample response field on the basis states.And then the radial basis function interpolation model is used to establish the nonlinear mapping relationship between arbitrary parameters and their corresponding basis state combination coefficients.This method can be used to model the response field of any space-time DOF,and keep the same number of outputs as the numerical model and output it in real time.(2)The fast parameter inversion algorithm based on the global surrogate model with temporal-spatial basis is developed.When the parameter vector is in low-dimension condition,a high precision surrogate model is established by fitting the parameter space with appropriate density to implement the finite element calculation.Then the error objective function combined with the particle swarm optimization algorithm is carried out to implement fast inversion.The accuracy of the inversion algorithm is further improved by presenting the surrogate-based optimization strategy,meanwhile to improve the inversion efficiency of multi-parameter.The inversion algorithm is more adaptable to engineering practical problems such as monitoring point change,dynamic measurement sequence and multi-effect measurement,and can manipulate in real time with continuous measurement.(3)A response prediction method of dam structure based on vector error correction model is proposed.In order to take into account the advantages of multiple regression model and time series model,the error correction model is introduced.To further consider the cross-correlation between multi-point measurements,a response prediction method based on vector error correction model is proposed.This method can be used to model the multi-point monitoring series at the same time,taking into account the crosscorrelation relationship between the multi-point measurements,and can be established at the early stage of dam impoundment.With the method the working behavior of the dam can be analyzed online and the main influencing factors can be separated,which provides references for the parameter inversion process.