Quick updating method for nonlinear structure model based on deep Gaussian process and Bayesian optimization algorithm

Through the combination of deep Gaussian process and Bayesian optimization algorithm, a probability proxy model is built and structural nonlinear model is optimized, which solves the calculation time-consuming problem in structural nonlinear model update, and realizes efficient model update and rapid convergence.

CN118940602APending Publication Date: 2024-11-12HARBIN INST OF TECH

Patent Information

Application Number
CN202410913258.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-09
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The calculation is time-consuming and requires a lot of computing resources during the update process of structural nonlinear models, resulting in inefficient solution.

Method used

Using the rapid update method of structural nonlinear model based on deep Gaussian process and Bayesian optimization algorithm, the probability proxy model is constructed through deep Gaussian process and combined with Bayesian optimization algorithm to gradually find the optimal value of structural nonlinear model updates to reduce the target The number of times the function is calculated and the time of analysis.

Benefits of technology

It significantly improves the solution efficiency, reduces the calculation cost and calculation time, avoids local optimal solutions, and improves the convergence speed of model updates.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940602A_ABST
    Figure CN118940602A_ABST
Patent Text Reader

Abstract

The invention provides a structure nonlinear model rapid updating method based on a depth Gaussian process and a Bayesian optimization algorithm. According to the method, a probabilistic agent model of an objective function in a structural nonlinear model updating problem is established by adopting a deep Gaussian process model, and an acquisition function is established in combination with prediction of a posterior deep Gaussian process model and uncertainty quantitative information so as to guide an optimization process of updating the objective function by the structural nonlinear model. The depth Gaussian process has a structure of a hierarchical Gaussian process, and prior correlation modeling can be better performed on an objective function updated by a structure nonlinear model, so that better prediction and uncertainty quantification performance is provided. According to the method, the problem of structural nonlinear model updating can be solved in a small number of times of numerical model analysis, so that the calculation efficiency of model updating is effectively improved, the calculation time is shortened, and the method has good engineering application value.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning and structural model updating, and in particular relates to a method for quickly updating a structural nonlinear model based on a deep Gaussian process and a Bayesian optimization algorithm. Background Art

[0002] The performance of civil engineering structures gradually declines with service time, so it is crucial to achieve structural status assessment and maintenance decisions through structural health monitoring technology to ensure the service safety of the structure. In the process of structural health monitoring, a reliable numerical model that reflects the structural performance is of guiding significance for realizing the status assessment of the structure. The structural model update technology uses the response data of the measured structure to correct the initial structural numerical model established based on the original drawings, so that the corrected numerical model can better reflect the current service status of the structure. In addition, civil engineering structures often inevitably enter a nonlinear state after experiencing loads such as earthquakes and strong winds. The traditional structural model update method only considers the linear performance changes of the structure, and often updates the initial numerical model of the structure with the difference of modal indicators as the goal. In this way, the updated model cannot be used for the prediction and status assessment of the structure experiencing strong loads. On the contrary, the structural model update method that considers the nonlinear characteristics of the structure can provide more comprehensive damage diagnosis indicators, not just considering the changes in the linear performance of the structure, but also provides a tool for predicting the remaining service life of the structure. After considering the nonlinearity of the structure, the linear theory is no longer applicable, so the initial numerical model is usually updated directly with the difference in structural response as the goal. However, since structural nonlinear analysis is very time-consuming, a lot of computing resources and time are usually required to ensure the convergence of the model update problem. Summary of the invention

[0003] The purpose of the present invention is to solve the challenge in the updating process of structural nonlinear models, that is, the problem of low solution efficiency due to the time-consuming calculation process of structural nonlinear analysis and the need for a large amount of computing resources and time. In response to this problem, the present invention proposes a method for fast updating of structural nonlinear models based on deep Gaussian process and Bayesian optimization algorithm. The method is applicable to relevant test objects in the field of structural nonlinear model updating and can significantly improve the solution efficiency.

[0004] The present invention is implemented by the following technical scheme. The present invention proposes a method for quickly updating a structural nonlinear model based on a deep Gaussian process and a Bayesian optimization algorithm. The method comprises the following steps:

[0005] Step 1: Establish an initial structural numerical model based on the existing structural drawings, and obtain the measured structural response data through the monitoring system installed on the structure;

[0006] Step 2: Select the structural parameters to be updated The objective function to be optimized is constructed based on the measured data of the structure and the predicted data of the structural numerical model. Its standard expression is:

[0007]

[0008] The objective function to be optimized middle, z i (θ) represent the measured response of the structure and the predicted response of the numerical model, respectively, N q Indicates the number of actual observation points;

[0009] Step 3: Use the deep Gaussian process and Bayesian optimization algorithm to optimize the objective function of the structural nonlinear model constructed in step 2 to obtain the optimal parameter combination θ opt , in order to update the structural numerical model;

[0010] Step 4: Combine θ based on the optimal parameters opt Numerical calculations are performed to obtain an updated numerical model of the benchmark structure.

[0011] Furthermore, in step three, the updating process is based on Bayesian theorem. First, the probabilistic proxy model of the objective function to be optimized is constructed using the deep Gaussian process. Then, the acquisition function is established by combining the prediction of the posterior deep Gaussian process and the uncertainty quantification information. Through the iterative process of the Bayesian optimization algorithm, the optimal value of the objective function of the structural nonlinear model is gradually found.

[0012] Furthermore, in the updating process, when there is observation noise, the observed value s(θ) of the structural nonlinear model updating objective function is The relationship is expressed as Where ε represents the observation noise, which is assumed to be an independent and identically distributed Gaussian distribution. The Latin hypercube sampling method is used to generate M initial structure parameter sample sets θ1=[θ1,θ2,…,θ M ] T , and perform structural numerical model operations to obtain response data, and then calculate the corresponding objective function observation value S1 = [s(θ1), s(θ2), … s(θ M )] T ; Let D1 = {Θ1, S1} to construct the initial training data set for the probabilistic proxy model.

[0013] Furthermore, in the updating process, according to the current structural nonlinear parameters and the objective function sample set D j ={Θ j ,S j}, j = 1…N, N is the number of steps executed by Bayesian optimization, and the structural nonlinear model is established to update the objective function The deep Gaussian process proxy model f(θ) of L layers is expressed as:

[0014] f(θ)=f L (…f2(f1(θ)))

[0015] in represents the Gaussian process of the lth layer, l = 1, 2, ... L; m l (·), They represent the mean function and kernel function of the Gaussian process of the lth layer respectively; the Gaussian process assumes that the function value on any finite point set obeys the joint Gaussian distribution; represents the hyperparameter of the kernel function in the l-th layer Gaussian process; when L = 1, the deep Gaussian process model degenerates into a single-layer Gaussian process model; let Represents the set of structural parameters Θ j The output of the Gaussian process node at the lth layer, Represents the set of structural parameters Θ j The output of all nodes at all layers; the deep Gaussian process modeling method directly specifies all the implicit functions of the intermediate layers And the last layer of structure nonlinearly updates the objective function f L Prior distribution over space.

[0016] Further, The prior distribution on is expressed as:

[0017]

[0018] in represents the hyperparameters in all Gaussian process layer kernel functions; since the observation noise follows a Gaussian distribution, the likelihood distribution is expressed as:

[0019]

[0020] Furthermore, in the updating process, the deep Gaussian process model is trained and a prediction model of the objective function to be optimized is established. The marginal likelihood function of the deep Gaussian process model is expressed as:

[0021]

[0022] Training and inferring deep Gaussian process models through variational inference, and the variational lower bound of deep Gaussian process models It is expressed as:

[0023]

[0024] Among them, q(F L,i ) represents θ i The variational distribution of the output at the Lth layer Gaussian process; p(U l |Z l-1 ) and q(U l |Z l-1 ) represent the Gaussian process induced point Z of the lth layer respectively. l-1 Output U l The prior distribution and variational distribution of ; The above formula is optimized by a gradient-based optimization algorithm to obtain the optimal deep Gaussian process model parameters; At the same time, a gridding method is used to determine Z l-1 The number and initial position of the updated structural parameter θ with dimension d, two values ​​are selected for each parameter within its variation range, resulting in a total of 2 d Induction points; after training the deep Gaussian process model, any structural parameter θ * The objective function value f(θ * )’s predicted distribution q(f(θ * )|θ * ,Φ p ,σ 2 ) is expressed as:

[0025]

[0026] in, Denotes any structural parameter θ * At the output of the Gaussian process node at the lth layer, Indicates that the Gaussian process at the lth layer The variational distribution at .

[0027] Furthermore, in the updating process, according to the posterior deep Gaussian process, the acquisition function value is calculated and the acquisition function is maximized to select the next structural parameter combination for calculation; the acquisition function based on the lifting strategy is adopted, which is expressed as follows:

[0028]

[0029] Where S j,min =min(S j ); Since the posterior distribution of the deep Gaussian process model cannot be expressed analytically, EI(θ) cannot be calculated directly, so it is approximated by sampling approximation:

[0030]

[0031] Among them, N s is the number of sampling approximations, and then the variational evolution algorithm is used to maximize EI(θ) and select the next evaluation point:

[0032] θj+1 =argmax(EI(θ)).

[0033] Furthermore, during the update process, θ is calculated j+1 The corresponding objective function observation value s(θ j+1 ), and update the data set D j+1 ={Θ j+1 ,S j+1}, where Θ j+1 =Θ j ∪θ j+1 ,S j+1 =S j ∪s(θ j+1 ).

[0034] The present invention proposes an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for quickly updating a structural nonlinear model based on a deep Gaussian process and a Bayesian optimization algorithm are implemented.

[0035] The present invention proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the method for quickly updating a structural nonlinear model based on a deep Gaussian process and a Bayesian optimization algorithm.

[0036] Beneficial effects of the present invention:

[0037] 1. The present invention adopts the Bayesian optimization method to solve the structural nonlinear model updating problem. Compared with the traditional optimization algorithm, the Bayesian optimization method requires fewer objective function calculations, that is, fewer numerical model analyses, so the computational cost is lower and the computational efficiency is higher.

[0038] 2. The present invention uses a deep Gaussian process as a probabilistic proxy model for the function to be optimized. Compared with the method based on a single Gaussian process, the deep Gaussian process can give better prediction and uncertainty quantification results for the target function to be optimized. Therefore, the convergence speed of the entire optimization process is faster.

[0039] 3. The present invention uses the prediction and uncertainty information of the posterior probability proxy model to guide the optimization process. Compared with the traditional gradient-based optimization algorithm, this method is not easy to fall into the local optimal solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0041] Figure 1 This is a flow chart of the method for quickly updating a structural nonlinear model based on a deep Gaussian process and a Bayesian optimization algorithm according to the present invention;

[0042] Figure 2 is a schematic diagram of a numerical model in an embodiment;

[0043] Figure 3 is a schematic diagram of convergence curves of different optimization algorithms in an embodiment;

[0044] Figure 4 A schematic diagram of a comparison between the inter-story displacement angle calculated by a benchmark model obtained by different optimization algorithms and the true value in an embodiment;

[0045] Figure 5 Schematic diagram of the comparison between the inter-layer bending moment curvature relationship calculated by the benchmark model obtained by different optimization algorithms and the true value in an embodiment. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] In order to improve the computational efficiency of updating the structural nonlinear model, the present invention adopts a method based on deep Gaussian process and Bayesian optimization algorithm. First, the objective function to be optimized for updating the structural nonlinear model is established based on the measured data of the structure and the output data of the numerical model. Then, a deep Gaussian process model is used to establish a probabilistic proxy model of the objective function to be optimized. The deep Gaussian process model is a model with a hierarchical Gaussian process structure. Its hierarchical structure can provide more flexible prior correlation modeling for the objective function to be fitted, thereby achieving better prediction results. Next, the deep Gaussian process model is trained by maximizing the variational lower bound and its posterior model is obtained. Based on the posterior model, an acquisition function for selecting the next evaluation point is constructed. It utilizes historical information and uncertainty information, and effectively balances the relationship between width search and depth search by maximizing the acquisition function constructed according to the posterior model, thereby reducing unnecessary objective function evaluations, that is, structural nonlinear numerical model analysis and calculations, and improving the computational efficiency of model updates. After obtaining the new objective function observation data, according to Bayes' theorem, the posterior deep Gaussian process probabilistic proxy model is updated, and it is continuously iterated until convergence to obtain the optimal parameter combination θ opt .

[0048] Combination Figure 1 The present invention proposes a method for quickly updating a structural nonlinear model based on a deep Gaussian process and a Bayesian optimization algorithm, the method comprising the following steps:

[0049] Step 1: Establish an initial structural numerical model based on the existing structural drawings, and obtain the measured structural response data through the monitoring system installed on the structure;

[0050] Step 2: Select the structural parameters to be updated The objective function to be optimized is constructed based on the measured data of the structure and the predicted data of the structural numerical model. The objective function is an objective function based on the structural dynamic response, and its standard expression is:

[0051]

[0052] The objective function to be optimized middle, z i (θ) represent the measured response of the structure and the predicted response of the numerical model, respectively, N q Indicates the number of actual observation points;

[0053] Step 3: Use the deep Gaussian process and Bayesian optimization algorithm to optimize the objective function of the structural nonlinear model constructed in step 2 to obtain the optimal parameter combination θ opt , in order to update the structural numerical model;

[0054] Step 4: Combine θ based on the optimal parameters opt Numerical calculations are performed to obtain an updated numerical model of the benchmark structure.

[0055] In step three, the update process is based on the Bayesian theorem. First, the deep Gaussian process is used to construct a probabilistic proxy model of the objective function to be optimized. Then, the acquisition function is established by combining the prediction of the posterior deep Gaussian process and the uncertainty quantification information. Through the iterative process of the Bayesian optimization algorithm, the optimal value of the structural nonlinear model to update the objective function is gradually found. The specific operation flow of the update process is as follows:

[0056] Step 3.1: When there is observation noise, the structural nonlinear model updates the observed value s(θ) of the objective function and The relationship is expressed as Where ε represents the observation noise, which is assumed to be an independent and identically distributed Gaussian distribution. The Latin hypercube sampling method is used to generate M initial structure parameter sample sets θ1=[θ1,θ2,…,θ M ] T , and perform structural numerical model operations to obtain response data, and then calculate the corresponding objective function observation value S1 = [s(θ1), s(θ2), … s(θ M )] T ; Let D1 = {Θ1, S1} to construct the initial training data set for the probabilistic proxy model.

[0057] Step 3.2: According to the current structural nonlinear parameters and the target function sample set D j ={Θ j ,S j}, j = 1…N, N is the number of steps executed by Bayesian optimization, and the structural nonlinear model is established to update the objective function The deep Gaussian process proxy model f(θ) of L layers is expressed as:

[0058] f(θ)=f L (…f2(f1(θ)))

[0059] in represents the Gaussian process of the lth layer, l = 1, 2, ... L; m l (·), They represent the mean function and kernel function of the Gaussian process of the lth layer respectively; the Gaussian process assumes that the function value on any finite point set obeys the joint Gaussian distribution; represents the hyperparameter of the kernel function in the l-th layer Gaussian process; when L = 1, the deep Gaussian process model degenerates into a single-layer Gaussian process model; let Represents the set of structural parameters Θ j The output of the Gaussian process node at the lth layer, Represents the set of structural parameters Θ jThe output of all nodes at all layers; the deep Gaussian process modeling method directly specifies all the implicit functions of the intermediate layers And the last layer of structure nonlinearly updates the objective function f L Prior distribution over space.

[0060] The prior distribution on is expressed as:

[0061]

[0062] in represents the hyperparameters in all Gaussian process layer kernel functions; since the observation noise follows a Gaussian distribution, the likelihood distribution is expressed as:

[0063]

[0064] Train the deep Gaussian process model and establish a prediction model for the objective function to be optimized. The marginal likelihood function of the deep Gaussian process model is expressed as:

[0065]

[0066] Due to the nonlinear relationship between kernel functions of any two layers in the deep Gaussian process model, it is difficult to express the marginal likelihood function analytically. Therefore, we use the variational inference method to train and infer the deep Gaussian process model. The variational lower bound of the deep Gaussian process model is It is expressed as:

[0067]

[0068] Among them, q(F L,i ) represents θ i The variational distribution of the output at the Lth layer Gaussian process; p(U l |Z l-1 ) and q(U l |Z l-1 ) represent the Gaussian process induced point Z of the lth layer respectively. l-1 Output U l The prior distribution and variational distribution of ; The above formula is optimized by a gradient-based optimization algorithm to obtain the optimal deep Gaussian process model parameters; The induction point Z of each layer of Gaussian process l-1 The number and location of Z are also important hyperparameters that affect the performance of the deep Gaussian process model. l-1 The number and initial position of the updated structural parameter θ with dimension d, two values ​​are selected for each parameter within its variation range, resulting in a total of 2 d Induction points; after training the deep Gaussian process model, any structural parameter θ *The objective function value f(θ * )’s predicted distribution q(f(θ * )|θ * ,Φ p ,σ 2 ) is expressed as:

[0069]

[0070] in, Denotes any structural parameter θ * At the output of the Gaussian process node at the lth layer, Indicates that the Gaussian process at the lth layer The variational distribution at .

[0071] Step 3.3, according to the posterior deep Gaussian process model, calculate the acquisition function value and maximize the acquisition function to select the next parameter combination for calculation; adopt the acquisition function based on the lifting strategy, which is expressed as:

[0072]

[0073] Where S j,min =min(S j ); Since the posterior distribution of the deep Gaussian process model cannot be expressed analytically, EI(θ) cannot be calculated directly, so it is approximated by sampling approximation:

[0074]

[0075] Among them, N s is the number of sampling approximations, here it is 5000. Then the variational evolution algorithm is used to maximize EI(θ) and select the next evaluation point:

[0076] θ j+1 =argmax(EI(θ)).

[0077] Step 3.4: Calculate θ j+1 The corresponding objective function observation value s(θ j+1 ), and update the data set D j+1 ={Θ j+1 ,S j+1}, where Θ j+1 =Θ j ∪θ j+1 ,S j+1 =S j ∪s(θ j+1 ). Repeat steps 3.2 and 3.3 until the specified number of steps is reached.

[0078] The method for fast updating of structural nonlinear models based on deep Gaussian process and Bayesian optimization algorithm proposed in the present invention establishes a probabilistic proxy model of the objective function to be optimized through a deep Gaussian model, and by maximizing the acquisition function constructed according to the posterior distribution of the model, it can effectively balance the relationship between width search (exploring the uncertainty area to obtain more unknown information) and depth search (using existing information to find the current optimum), thereby reducing unnecessary objective function evaluations and effectively improving the computational efficiency of model updating. In structural health monitoring, the numerical model analysis and calculation required for nonlinear model update problems are often very time-consuming. The present invention significantly improves the convergence speed of structural nonlinear model update problems, reduces the number of analyses, saves analysis time, and enables structural nonlinear model update technology to better serve the field of structural health monitoring.

[0079] Example

[0080] Combination Figure 2 For a three-story steel frame structure under earthquake excitation, the material parameters in the model are updated using the structural nonlinear model fast updating method based on deep Gaussian process and Bayesian optimization algorithm of the present invention.

[0081] Next, the numerical model of the steel frame structure is updated using the structural nonlinear model rapid updating method based on the deep Gaussian process and the Bayesian optimization algorithm in the present invention.

[0082] The step 1 is specifically as follows: using OpenSees software to establish a numerical analysis model of a three-story steel frame structure, such as Figure 2 As shown in Figure 1, acceleration sensors are installed on each floor of the frame structure to obtain acceleration response data of the steel frame structure under an earthquake excitation.

[0083] The second step is specifically: based on the numerical model established in step one and the measured structural acceleration data, construct the objective function to be updated Since the top floor is not easy to enter a nonlinear state under the action of earthquake excitation, the elastic modulus and yield strength of the first and second floor steel of the steel frame structure and the elastic model parameters of the top floor steel are selected as the parameters to be updated. Their nominal values, true values ​​and value ranges are shown in Table 1.

[0084] Table 1 Selected representative values ​​and their ranges

[0085]

[0086] The step three is as follows: generate 10 sample sets of the initial space by Latin hypercube sampling, and calculate the corresponding objective function values ​​to form the initial data set D1 = {θ1, S1}, where θ1 = [θ1, θ2, ... θ 10 ] T,S1=[s(θ1),s(θ2),…s(θ 10 )] T .

[0087] Since there are 5 parameters to be updated, 2 are set in each Gaussian process layer. 5 = 32 induction points for training and inference of the deep Gaussian process model. Set the number of execution steps of Bayesian optimization to 100 times, j = 1…100. Based on the current data set D1, maximize the variational lower bound to train the deep Gaussian process model, obtain the posterior deep Gaussian process model, and then maximize the acquisition function EI(θ) to obtain the next objective function evaluation point θ j+1 , and calculate the corresponding objective function value s(θ j+1 ) and update the data set D j+1 ={Θ j+1 ,S j+1}, where Θ j+1 =Θ j ∪θ j+1 ,S j+1 =S j ∪s(θ j+1 ), repeat the above steps until the set number of steps is reached to obtain the optimal parameter combination θ opt = agrmin(S 101 ), as shown in Table 2.

[0088] The step 4 is specifically as follows: according to the obtained optimal parameter combination θ opt , and obtain the updated benchmark numerical model.

[0089] The present invention proposes a method for fast updating of structural nonlinear models based on deep Gaussian process and Bayesian optimization algorithm. The method adopts deep Gaussian process model to establish a probabilistic proxy model of the objective function in the structural nonlinear model updating problem, and establishes an acquisition function in combination with the prediction of the posterior deep Gaussian process model and uncertainty quantification information to guide the optimization process of the structural nonlinear model updating objective function. The deep Gaussian process has the structure of a hierarchical Gaussian process, and can better model the prior correlation of the objective function of the structural nonlinear model update, thereby providing better prediction and uncertainty quantification performance. The acquisition function uses historical information and uncertainty information, and can effectively balance the relationship between width search (exploring uncertainty areas to obtain more unknown information) and depth search (using existing information to find the current optimal solution), thereby reducing the number of unnecessary objective function evaluations. The present invention can solve the structural nonlinear model updating problem in a small number of numerical model analyses, thereby effectively improving the computational efficiency of model updating, reducing the computational time, and having good engineering application value.

[0090] Table 2 Parameter values ​​and errors after optimization of each algorithm

[0091]

[0092] At the same time, the algorithm proposed in the present invention is compared with the traditional algorithm based on Gaussian process and Bayesian optimization algorithm (BOA-GP) and genetic algorithm (GA, Genetic Algorithm). Genetic algorithm is an optimization algorithm based on natural selection and genetic mechanism, which can effectively find the global optimal solution in a complex search space. The optimization process convergence curves of the three algorithms are plotted, as shown in Figure 1. Figure 3 The responses of the benchmark numerical model output obtained after the three algorithms are updated and the responses of the real output of the structure are plotted, as shown in Figure 4 and 5 This result can be used for further analysis.

[0093] In the above examples, the algorithm proposed in the present invention based on the method of deep Gaussian process and Bayesian optimization algorithm (BOA-DGP) shows the fastest convergence speed. BOA-DGP reaches the result of the last step of BOA-GP at the 70th iteration, and the corresponding parameter update error is the smallest. The total parameter error of the proposed algorithm reaches 12.67% after only 100 iterations (numerical model analysis), and the average error of each parameter is 2.5%, which meets the requirements of engineering applications. At the same time, the response of the obtained benchmark model output is closest to the response of the real output of the structure.

[0094] The present invention proposes an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for quickly updating a structural nonlinear model based on a deep Gaussian process and a Bayesian optimization algorithm are implemented.

[0095] The present invention proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the method for quickly updating a structural nonlinear model based on a deep Gaussian process and a Bayesian optimization algorithm.

[0096] The memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus RAM (DRRAM). It should be noted that the memory of the method described in the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0097] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a high-density digital video disc (DVD)), or a semiconductor medium (eg, a solid state disc (SSD)).

[0098] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software. The steps of the method disclosed in conjunction with the embodiment of the present application can be directly embodied as a hardware processor for execution, or a combination of hardware and software modules in a processor for execution. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in conjunction with its hardware. To avoid repetition, it is not described in detail here.

[0099] It should be noted that the processor in the embodiment of the present application can be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor can be combined and performed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0100] The above is a detailed introduction to the method for fast updating of structural nonlinear models based on deep Gaussian process and Bayesian optimization algorithm proposed in the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A fast updating method for structural nonlinear models based on deep Gaussian process and Bayesian optimization algorithm, characterized in that: The method comprises the following steps: Step 1: Establish an initial structural numerical model based on the existing structural drawings, and obtain the measured structural response data through the monitoring system installed on the structure; Step 2: Select the structural parameters to be updated The objective function to be optimized is constructed based on the measured data of the structure and the predicted data of the structural numerical model. Its standard expression is: The objective function to be optimized middle, z i (θ) represent the measured response of the structure and the predicted response of the numerical model, respectively, N q Indicates the number of actual observation points; Step 3: Use the deep Gaussian process and Bayesian optimization algorithm to optimize the objective function of the structural nonlinear model constructed in step 2 to obtain the optimal parameter combination θ opt , in order to update the structural numerical model; Step 4: Combine θ based on the optimal parameters opt Numerical calculations are performed to obtain an updated numerical model of the benchmark structure.

2. The method according to claim 1, characterized in that In step three, the updating process is based on Bayesian theorem. First, the deep Gaussian process is used to construct a probabilistic proxy model of the objective function to be optimized. Then, the acquisition function is established by combining the prediction of the posterior deep Gaussian process and the uncertainty quantification information. Through the iterative process of the Bayesian optimization algorithm, the optimal value of the objective function of the structural nonlinear model is gradually found.

3. The method according to claim 2, characterized in that In the updating process, when there is observation noise, the structural nonlinear model updates the observed value s(θ) of the objective function with The relationship is expressed as Where ε represents the observation noise, which is assumed to be an independent and identically distributed Gaussian distribution. The Latin hypercube sampling method is used to generate M initial structure parameter sample sets θ1=[θ1,θ2,…,θ M ] T , and perform structural numerical model operations to obtain response data, and then calculate the corresponding objective function observation value S1 = [s(θ1), s(θ2), … s(θ M )] T ; Let D1 = {Θ1, S1} to construct the initial training data set for the probabilistic proxy model.

4. The method according to claim 3, characterized in that In the updating process, according to the current structural nonlinear parameters and the objective function sample set D j ={Θ j ,S j }, j = 1…N, N is the number of steps executed by Bayesian optimization, and the structural nonlinear model is established to update the objective function The deep Gaussian process proxy model f(θ) of L layers is expressed as: f(θ)=f L (…f2(f1(θ))) in represents the Gaussian process of the lth layer, l = 1, 2, ... L; They represent the mean function and kernel function of the Gaussian process of the lth layer respectively; the Gaussian process model assumes that the function values ​​on any finite set of points obey the joint Gaussian distribution; represents the hyperparameter of the kernel function in the l-th layer Gaussian process; when L = 1, the deep Gaussian process model degenerates into a single-layer Gaussian process model; let Represents the set of structural parameters Θ j The output of the Gaussian process node at the lth layer, Represents the set of structural parameters Θ j The output of all layer nodes at the deep Gaussian process modeling method directly specifies all intermediate layer implicit functions And the last layer of structure nonlinearly updates the objective function f L Prior distribution over space.

5. The method according to claim 4, characterized in that The prior distribution on is expressed as: in represents the hyperparameters in all Gaussian process layer kernel functions; since the observation noise follows a Gaussian distribution, the likelihood distribution is expressed as:

6. The method according to claim 5, characterized in that During the updating process, the deep Gaussian process model is trained and a prediction model of the objective function to be optimized is established. The marginal likelihood function of the deep Gaussian process model is expressed as: Training and inferring deep Gaussian process models through variational inference, and the variational lower bound of deep Gaussian process models It is expressed as: Among them, q(F L,i ) represents θ i The variational distribution of the output at the Lth layer Gaussian process; p(U l |Z l-1 ) and q(U l |Z l-1 ) represent the Gaussian process at the induction point Z l-1 Output U l The prior distribution and variational distribution of ; The above formula is optimized by a gradient-based optimization algorithm to obtain the optimal deep Gaussian process model parameters; At the same time, a gridding method is used to determine Z l-1 The number and initial position of the updated structural parameter θ with dimension d, two values ​​are selected for each parameter within its variation range, resulting in a total of 2 d Induction points; after training the deep Gaussian process model, any structural parameter θ * The objective function value f(θ * )’s predicted distribution q(f(θ * )|θ * ,Φ p ,σ 2 ) is expressed as: in, Denotes any structural parameter θ * At the output of the Gaussian process node at the lth layer, Indicates that the Gaussian process at the lth layer The variational distribution at .

7. The method according to claim 6, characterized in that During the updating process, according to the posterior deep Gaussian process model, the acquisition function value is calculated and maximized to select the next structural parameter combination for calculation; the acquisition function based on the lifting strategy is adopted, which is expressed as follows: Where S j,min =min(S j ); Since the posterior distribution of the deep Gaussian process model cannot be expressed analytically, EI(θ) cannot be calculated directly, so it is approximated by sampling approximation: Among them, N s is the number of sampling approximations, and then the variational evolution algorithm is used to maximize EI(θ) and select the next evaluation point: i j+1 = argmax(EI(θ)).

8. The method according to claim 7, characterized in that During the update process, we calculate θ j+1 The corresponding objective function observation value s(θ j+1 ), and update the data set D j+1 ={Θ j+1 ,S j+1 }, where Θ j+1 =Θ j ∪θ j+1 ,S j+1 =S j ∪s(θ j+1 ).

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium for storing computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Citation Information

Patent Citations

  • Mechanism optimization design method based on Bayesian kriging model

    CN110008499A

  • PCE_BO-based structural performance parameter rapid inversion method

    CN113033054A

  • Finite element model correction method based on Bayesian optimization and GP proxy model

    CN117057194A

  • Learning parameters of a probabilistic model comprising gaussian processes

    EP3716160A1

  • Computational inference system

    US20210056352A1

Cited By

  • Bridge structure risk early warning method based on Bayesian decision and information value

    CN120258513A

  • Seismic oscillation time-varying phase-dryness estimation method based on Bayesian collaborative optimization

    CN122151183A

  • A method for estimating time-varying coherence of ground motion based on bayesian co-optimization

    CN122151183B

  • Wind-induced vibration prediction method based on clustering and stage-adaptive multi-output surrogate model

    CN123013230A