Cycle-Transform network-based main shock and aftershock shock resistance toughness rapid evaluation method and system

Abstract

The invention discloses a rapid assessment method and system for shock resistance toughness of main and aftershocks based on a cycle-Transform network, and the method comprises the steps: building a finite element model of a building structure, inputting a plurality of groups of earthquake accelerations of main and aftershocks into the model, and obtaining a database comprising a plurality of groups of accelerations and corresponding structure displacement responses; building a loop-Transform neural network, and carrying out the training and optimization of the loop-Transform neural network; inputting the main shock and aftershock sequence into the optimized cycle-Transform neural network, predicting the displacement response of the structure under the double amplitude modulation main shock and aftershock sequence, then obtaining the building damage probability corresponding to different earthquake intensities under the main shock and aftershock situation by using maximum likelihood estimation, evaluating the corresponding damage index, calculating the recovery time, and finally obtaining the building damage probability corresponding to the different earthquake intensities under the main shock and aftershock situation. A function recovery curve of the structure is predicted, and shock resistance toughness evaluation is completed. According to the method, the precision of structure accumulated damage prediction under the aftershock effect is improved, and the calculation complexity and time consumption of toughness evaluation under the main and aftershock earthquake scene are remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of civil engineering and computer science, and in particular to a method and system for quickly evaluating the seismic toughness of main shocks and aftershocks based on a cyclic-Transformer network. Background Art

[0002] When conducting seismic resilience assessment, it is necessary to evaluate the damage state of the structure under different earthquake magnitudes, especially focusing on moderate to severe damage, or even collapse under strong earthquakes. Traditionally, numerical simulations represented by the finite element method have been widely used in earthquake response analysis. However, it usually requires a lot of calculations to calibrate the model to achieve sufficient accuracy. When a large number of simulations are required to consider the random uncertainty of external loads, the computational burden may increase significantly. In addition, when considering aftershocks, it is necessary to evaluate the long-term nonlinear time history response, which will further increase the computational cost of the finite element method. Therefore, computational efficiency severely limits the application of the finite element method in seismic resilience assessment, especially for actual structures at the community scale. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for rapid assessment of mainshock and aftershock seismic resilience based on a recurrent-Transformer network, to construct a new type of recurrent-Transformer neural network, to achieve nonlinear time-history response prediction of building structures under mainshock and aftershock sequences, to ensure accuracy while achieving high efficiency, and to achieve rapid resilience assessment under mainshock and aftershock scenarios.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0005] A rapid assessment method for main shock and aftershock seismic resilience based on a recurrent-Transformer network includes the following steps:

[0006] S1. Establish a finite element model of the building structure, input multiple sets of main shock and aftershock earthquake accelerations into the finite element model, obtain corresponding structural displacement response data, and establish a database including multiple sets of accelerations and their corresponding structural displacement responses.

[0007] S2. Construct a recurrent-Transformer neural network, and use the database in step S1 to train and optimize the recurrent-Transformer neural network to determine the optimal parameters.

[0008] S3. After double amplitude modulation processing, the main shock and aftershock sequence is input into the optimized recurrent-Transformer neural network to predict the displacement response of the structure under the double amplitude modulation main shock and aftershock sequence.

[0009] S4. Using the displacement response obtained in step S3, maximum likelihood estimation is used to obtain the building damage probability corresponding to different earthquake intensities under the main shock and aftershock scenario.

[0010] S5. According to the building damage probability obtained in step S4, the corresponding damage index is evaluated, the recovery time is calculated, and the functional recovery curve of the structure is predicted to complete the seismic toughness assessment.

[0011] Furthermore, in step S1, the construction of the mainshock and aftershock earthquake acceleration sequence includes:

[0012] (1) The main shock acceleration record and aftershock acceleration record come from the same station of the same earthquake event;

[0013] (2) The horizontal average PGA (Peak Ground Acceleration) of the selected earthquake is greater than 0.04g;

[0014] (3) The sampling interval of ground acceleration should be less than or equal to 0.02 s. A zero-value sequence should be filled between the main shock and aftershocks, and the length of the zero-value sequence should be greater than 30 s.

[0015] Further, in step S2, the recurrent-Transformer neural network includes an input layer, a GRU (Gated Recurrent Unit) layer, a Transformer layer, a fully connected layer and an output layer. The acceleration data of the earthquake is X. X is input into the input layer of the recurrent-Transformer neural network, passes through the GRU layer to capture the positional relationship in the acceleration data, passes through the Transformer layer to learn the global dependency and nonlinear characteristics between the main shock and aftershock sequences, passes through the fully connected layer, and is mapped to the displacement response Y of the structure by the output layer.

[0016] Among them, x n represents the acceleration data of the nth time step, represents the real number domain, y n represents the displacement response at the nth time step, where n represents the total number of time steps.

[0017] During the training and optimization process, the mean square error is used as the loss function and the gradient is calculated through the back-propagation algorithm.

[0018] Furthermore, the Transformer layer includes a multi-head attention module, which is used to calculate the query vector, key vector and value vector of the earthquake acceleration data, calculate the attention scores of multiple heads respectively, and concatenate them to obtain the total multi-head attention score. The calculation formula is:

[0019] Q=HWQ ,K=HW K ,V=HW V

[0020]

[0021] MultiHead(Q,K,V)=Concat(head1,…,head h )W o

[0022] Among them, Q represents the query vector, H represents the output result of the GRU layer, K represents the key vector, V represents the value vector, and W Q represents the weight matrix corresponding to Q, W K represents the weight matrix corresponding to K, W V represents the weight matrix corresponding to V, softmax(·) represents the normalized exponential function, d k represents the dimension of the key vector K, head i represents the attention score of the p-th attention head, represents the weight matrix corresponding to Q of the p-th attention head, represents the weight matrix corresponding to K of the p-th attention head, represents the weight matrix V corresponding to the pth attention head, Concat(·) represents vector concatenation, head h represents the attention score of the hth attention head, p = 1, 2, ... h, h represents the total number of attention heads, W o Represents the total weight of each attention head.

[0023] Furthermore, in step S3, the dual amplitude modulation process includes the following contents:

[0024] The main shock and aftershock seismic acceleration can be expressed as the combination of the main shock acceleration, interval, and aftershock acceleration. The specific expression is:

[0025] X={x1,x2,…,x n} T =[X main ,X0,X after ]

[0026] Among them, X main represents the main shock acceleration sequence, X0 represents the zero vector, X after Represents the aftershock acceleration sequence.

[0027] X main and X after For amplitude modulation, the specific formula is:

[0028]

[0029] in, represents the main shock acceleration sequence after amplitude modulation, represents the aftershock acceleration sequence after amplitude modulation, α represents the amplitude modulation coefficient, and max() represents the maximum value of the orientation quantity.

[0030] The amplitude-modulated main shock acceleration sequence and aftershock acceleration sequence are combined to obtain the acceleration sequence after double amplitude modulation.

[0031]

[0032] Furthermore, in step S4, a vulnerability function is used to represent the building damage probability corresponding to different earthquake intensities. The specific expression is:

[0033]

[0034] Among them, P(EDP a >C a |IM a =x,DS mi ) indicates that the initial state of the main epicenter is DS mi , and the aftershock intensity index IM a = x, the probability of reaching a certain aftershock damage state; EDP a represents the engineering demand parameter under aftershock; C a Indicates the structural bearing capacity under aftershocks; DS mi It represents the damage state under the main shock; Φ represents the standard normal cumulative distribution function; θ represents the mean; β represents the logarithmic standard deviation; P() represents the probability; x represents the value of the aftershock intensity index.

[0035] Furthermore, in step S5, the corresponding damage index is evaluated according to the damage probability of the building under different earthquakes. The specific formula is:

[0036] Furthermore, where L represents the main shock damage state DS mi The damage index at represents the damage index at the jth aftershock damage state; P(DS aj |IM a ,DS mi ) indicates that the aftershock intensity is IM a , the initial state of the main shock is DS mi When the structure reaches the destruction state DS in the aftershock aj The probability of time.

[0037] P(DS aj |IM a ,DS mi) is calculated as:

[0038]

[0039] Among them, EDP a Represents the engineering demand parameter under aftershock action; LS ai+1 The damage state index of the i+1th damage state under aftershock; LS ai The damage state index represents the i-th damage state under the aftershock.

[0040] Estimated recovery time:

[0041]

[0042] Among them, T REj represents the repair time required for the jth damage state; T RE Indicates the estimated time required for recovery.

[0043] The functional recovery curve of the structure is:

[0044]

[0045] Among them, f(t,t0,T RE ) represents the expected functional recovery curve, t0 represents the time when the earthquake occurs, and t represents the time.

[0046] Furthermore, the present invention also proposes a rapid assessment system for main shock and aftershock seismic resilience based on a cyclic-Transformer network, comprising:

[0047] The database construction module is used to establish a finite element model of the building structure, input multiple sets of main shock and aftershock earthquake accelerations into the finite element model, obtain corresponding structural displacement response data, and establish a database including multiple sets of accelerations and their corresponding structural displacement responses.

[0048] The model training module is used to build a recurrent-Transformer neural network. It uses the database in the database construction module to train and optimize the recurrent-Transformer neural network and determine the optimal parameters.

[0049] The prediction module is used to perform double amplitude modulation processing on the main shock and aftershock sequence and input it into the optimized recurrent-Transformer neural network to predict the displacement response of the structure under the double amplitude modulation main shock and aftershock sequence.

[0050] The building damage probability acquisition module is used to use the displacement response obtained by the prediction module and use maximum likelihood estimation to obtain the building damage probability corresponding to different earthquake intensities under the main aftershock scenario.

[0051] The seismic toughness assessment module is used to evaluate the corresponding damage index, calculate the recovery time, and predict the functional recovery curve of the structure based on the building damage probability obtained by the building damage probability acquisition module to complete the seismic toughness assessment.

[0052] Furthermore, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the method for rapid assessment of mainshock and aftershock seismic resilience based on the recurrent-Transformer network described above are implemented.

[0053] Furthermore, the present invention also proposes a computer-readable storage medium, which stores a computer program. When the computer program is run by a processor, it executes the method for rapid assessment of mainshock and aftershock seismic resilience based on a cyclic-Transformer network.

[0054] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0055] 1. The method proposed in the present invention shows excellent accuracy in long-term prediction, especially in the prediction of main shock and aftershock sequences, and effectively improves the accuracy of structural response prediction.

[0056] 2. The method proposed in the present invention significantly reduces the computational complexity and time consumption of resilience assessment in the main aftershock earthquake scenario. It can use a small amount of data and adopt a data-driven method to deduce the structural response in various earthquake scenarios, and achieve rapid and accurate resilience evaluation. In addition, due to the rapidity and low computational cost of the method, it is possible to expand the object of resilience assessment to urban buildings, which promotes the industrial application of resilience assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is an overall implementation structure diagram of the present invention.

[0058] Figure 2 It is a displacement response prediction effect diagram of an embodiment of the present invention.

[0059] Figure 3 It is a diagram showing the effect of vulnerability function prediction according to an embodiment of the present invention.

[0060] Figure 4 It is a diagram showing the prediction effect of the structural recovery curve of an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0062] To achieve the above objectives, the present invention proposes a rapid assessment method for the seismic toughness of main shocks and aftershocks based on a recurrent-Transformer network. Taking a five-story reinforced concrete frame as an example, Figure 1 As shown, the specific steps are as follows:

[0063] S1. Establish a finite element model of the building structure, input multiple sets of main shock and aftershock earthquake accelerations into the finite element model, obtain corresponding structural displacement response data, and establish a database including multiple sets of accelerations and their corresponding structural displacement responses; the specific contents are as follows:

[0064] The following measures are taken to construct the mainshock and aftershock earthquake acceleration sequence:

[0065] (1) The main shock acceleration record and aftershock acceleration record come from the same station of the same earthquake event.

[0066] (2) The horizontal average PGA (Peak Ground Acceleration) of the selected earthquake is greater than 0.04g.

[0067] (3) The sampling interval of ground acceleration should be less than or equal to 0.02 s. A zero-value sequence should be filled between the main shock and aftershocks, and the length of the zero-value sequence should be greater than 30 s.

[0068] In this embodiment, the selected mainshock and aftershock earthquake acceleration data are from the NGA-West2 public seismic motion data set. A 100-s mainshock sequence consisting of 30s mainshock, 40s free vibration and 30s aftershock is used as the earthquake acceleration and input into a five-story reinforced concrete frame finite element model to obtain the corresponding displacement response data.

[0069] S2. Construct a recurrent-Transformer neural network, and use the database in step S1 to train and optimize the recurrent-Transformer neural network to determine the optimal parameters; the specific contents are as follows:

[0070] The recurrent-Transformer neural network includes an input layer, a GRU (Gated Recurrent Unit) layer, a Transformer layer, a fully connected layer, and an output layer. The acceleration data of the earthquake is X. The input is fed into the input layer of the recurrent-Transformer neural network, passes through the GRU layer, and captures the positional relationship in the acceleration data; passes through the Transformer layer, learns the global dependency and nonlinear characteristics between the main shock and aftershock sequences, passes through the fully connected layer, and is mapped to the displacement response of the structure by the output layer. During the training and optimization process, the mean square error is used as the loss function, and the gradient is calculated through the back propagation algorithm to continuously reduce the loss and optimize the model performance.

[0071] Among them, x n represents the acceleration data of the nth time step, represents the real number domain, y n represents the displacement response at the nth time step, where n represents the total number of time steps.

[0072] The Transformer layer includes a multi-head attention module, which calculates the query vector, key vector, and value vector of the earthquake acceleration data, calculates the attention scores of multiple heads respectively, and concatenates them to obtain the total multi-head attention score. The calculation formula is:

[0073] Q=HW Q ,K=HW K ,V=HW V

[0074]

[0075] MultiHead(Q,K,V)=Concat(head1,…,head h )W o

[0076] Among them, Q represents the query vector, H represents the output result of the GRU layer, K represents the key vector, V represents the value vector, and W Q represents the weight matrix corresponding to Q, W K represents the weight matrix corresponding to K, W V represents the weight matrix corresponding to V, softmax(·) represents the normalized exponential function, d k represents the dimension of the key vector K, head i represents the attention score of the p-th attention head, represents the weight matrix corresponding to Q of the p-th attention head, represents the weight matrix corresponding to K of the p-th attention head, represents the weight matrix V corresponding to the pth attention head, Concat(·) represents vector concatenation, head h represents the attention score of the hth attention head, p = 1, 2, ... h, W o Represents the total weight of each attention head.

[0077] S3, after double amplitude modulation processing of the main shock and aftershock sequence, input it into the optimized recurrent-Transformer neural network to predict the displacement response of the structure under the double amplitude modulation main shock and aftershock sequence; the specific content is:

[0078] The main shock and aftershock earthquake acceleration can also be expressed as the splicing of the main shock acceleration, interval, and aftershock acceleration. The specific expression is:

[0079] X={x1,x2,…,x n} T =[X main ,X0,X after ]

[0080] Among them, X main represents the main shock acceleration sequence, X0 represents the zero vector, X after Represents the aftershock acceleration sequence.

[0081] X main and X after For amplitude modulation, the specific formula is:

[0082]

[0083] in, represents the main shock acceleration sequence after amplitude modulation; It represents the aftershock acceleration sequence after amplitude modulation; α represents the amplitude modulation coefficient, which takes a value of 0.1 to 1.0g; max() represents the maximum value of the orientation quantity.

[0084] The amplitude-modulated main shock acceleration sequence and aftershock acceleration sequence are combined to obtain the acceleration sequence after double amplitude modulation.

[0085]

[0086] S4. Using the structural response obtained in step S3, the maximum likelihood estimation is used to obtain the building damage probability corresponding to different earthquake intensities under the main shock and aftershock scenarios; the specific contents are:

[0087] The probability of building damage corresponding to different earthquake intensities can be expressed using a vulnerability function, which refers to the probability of exceeding the aftershock damage state after experiencing an aftershock under a certain initial damage state caused by the main shock, and is calculated by the following formula:

[0088]

[0089] Among them, P(EDP a >C a |IM a =x,DS mi ) indicates that the initial state of the main epicenter is DS mi , and the aftershock intensity index IM a = x, the probability of reaching a certain aftershock damage state, EDP a represents the engineering demand parameter under aftershock, C aDenotes the structural bearing capacity under aftershock, DS mi It represents the damage state under the main shock, Φ represents the standard normal cumulative distribution function, θ represents the mean, β represents the logarithmic standard deviation, which is calculated by maximum likelihood estimation, P() represents the probability, and x represents the value of the aftershock intensity index.

[0090] In this embodiment, the four damage states under the main shock are: DS m1 DS m2 DS m3 DS m4 ; The four damage states under aftershocks are: DS a1 DS a2 DS a3 DS a4 In this case, the damage state is defined by the interlayer displacement angle, as shown in Table 1.

[0091] Table 1 Definition of damage status

[0092]

[0093]

[0094] S5. According to the damage probability of the building under different earthquakes, evaluate the corresponding damage index, calculate the expected recovery time, and predict the functional recovery curve of the structure to complete the seismic resilience assessment; the specific contents are:

[0095] According to the probability of damage to the building under different earthquakes, the corresponding damage index is evaluated. The damage index represents the proportion of structural function loss, with a value of 0 to 1, where 0 represents that the structural function is intact. The specific calculation method is as follows:

[0096]

[0097] Among them, L indicates that the main shock damage state is DS mi The damage index at represents the damage index at the jth aftershock damage state; P(DS aj |IM a ,DS mi ) indicates that the aftershock intensity is IM a , the initial state of the main shock is DS mi When the structure reaches the destruction state DS in the aftershock aj The probability of time.

[0098] P(DS aj |IM a ,DS mi ) is calculated by the following formula:

[0099]

[0100] Where LS represents the damage state index, the subscript a represents the aftershock situation, m represents the main shock situation, and i represents the i-th damage state.

[0101] The estimated recovery time is calculated using the following formula:

[0102]

[0103] Among them, T REj represents the repair time required for the jth damage state, T RE Indicates the estimated time required for recovery.

[0104] The functional recovery curve of the structure can provide the possible functional recovery route, which is calculated using the following formula:

[0105]

[0106] Among them, f(t,t0,T RE ) represents the expected functional recovery curve, t0 represents the time when the earthquake occurs, and t represents the time.

[0107] In this embodiment, the damage index R at each aftershock damage state is DSj The values ​​are shown in Table 2. The recovery time T required for each damage state is REj The values ​​are shown in Table 3.

[0108] Table 2 Damage index corresponding to each damage state

[0109] Damage Status Injury Index <![CDATA[DS a1 ]]> 0.12 <![CDATA[DS a2 ]]> 0.25 <![CDATA[DS a3 ]]> 0.6 <![CDATA[DS a4 ]]> 1.0

[0110] Table 3 Recovery time corresponding to each damage state

[0111] Damage Status Recovery time (days) <![CDATA[DS a1 ]]> 10 <![CDATA[DS a2 ]]> 36 <![CDATA[DS a3 ]]> 160 <![CDATA[DS a4 ]]> 309

[0112] Use the recurrent-Transformer neural network to predict the structural response on the unknown test set and verify the network accuracy. The prediction results are as follows Figure 2 As shown, it can be seen that the predicted results are very close to the true values.

[0113] Figure 3 This is a diagram of the vulnerability function prediction effect obtained based on the recurrent-Transformer neural network. Figure 3 (a) is the aftershock vulnerability curve when the main shock does not cause damage. Figure 3 (b) is the aftershock vulnerability curve when the main shock causes slight damage. Figure 3 (c) is the aftershock vulnerability curve when the main shock causes moderate damage. Figure 3(d) is the aftershock vulnerability curve when the main shock causes serious damage. It can be seen from the figure that as the degree of damage caused by the main shock increases, the probability of the structure reaching the same damage state under the action of aftershocks increases.

[0114] Figure 4 is the functional recovery curve predicted by the recurrent-Transformer network. Assuming that the main shock occurs at time t0, and the aftershock occurs 50 days after the main shock, Figure 4 (a) is the functional recovery curve obtained when the main earthquake center is in moderate damage, taking the aftershock PGA as an example of 0.5g, 0.8g, 1.0g and 1.2g. Figure 4 (b) is the functional recovery curve obtained when the main earthquake is in serious damage, taking the aftershock PGA as 0.5g, 0.8g, 1.0g and 1.2g as examples. It can be seen from the figure that aftershocks will further aggravate the damage to the structural toughness and prolong the time required for functional recovery.

[0115] The embodiment of the present invention also proposes a rapid assessment system for the seismic resilience of main and aftershocks based on a cyclic-Transformer network, including a database construction module, a model training module, a prediction module, a building damage probability acquisition module, a seismic resilience assessment module, and a computer program that can be run on a processor. It should be noted that each module in the above system corresponds to the specific steps of the method provided in the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided in the embodiment of the present invention.

[0116] The embodiment of the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. It should be noted that when the processor executes the computer program, the specific steps of the method provided in the embodiment of the present invention are corresponding to the specific steps of the method provided in the embodiment of the present invention, and the processor has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided in the embodiment of the present invention.

[0117] The embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program. It should be noted that when the computer program is executed by the processor, it corresponds to the specific steps of the method provided in the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided in the embodiment of the present invention.

[0118] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A rapid assessment method for mainshock and aftershock seismic resilience based on a recurrent-Transformer network, characterized in that: include: S1. Establish a finite element model of the building structure, input multiple sets of main shock and aftershock earthquake accelerations into the finite element model, obtain corresponding structural displacement response data, and establish a database including multiple sets of accelerations and their corresponding structural displacement responses; S2, construct a recurrent-Transformer neural network, use the database in step S1 to train and optimize the recurrent-Transformer neural network, and determine the optimal parameters; S3, after double amplitude modulation processing of the main shock and aftershock sequence, the processing is input into the optimized recurrent-Transformer neural network to predict the displacement response of the structure under the double amplitude modulation main shock and aftershock sequence; S4, using the displacement response obtained in step S3, using maximum likelihood estimation to obtain the building damage probability corresponding to different earthquake intensities under the main shock and aftershock scenario; S5. According to the building damage probability obtained in step S4, the corresponding damage index is evaluated, the recovery time is calculated, and the functional recovery curve of the structure is predicted to complete the seismic toughness assessment.

2. The method for rapid assessment of mainshock and aftershock seismic resilience based on a cyclic-Transformer network according to claim 1 is characterized in that: In step S1, the construction of the mainshock and aftershock earthquake acceleration sequence includes: (1) The main shock acceleration record and aftershock acceleration record are from the same station of the same earthquake event; (2) The horizontal average peak ground acceleration of the selected earthquake is greater than 0.04g; (3) The sampling interval of ground acceleration should be less than or equal to 0.02 s. A zero-value sequence should be filled between the main shock and aftershocks, and the length of the zero-value sequence should be greater than 30 s.

3. The method for rapid assessment of mainshock and aftershock seismic resilience based on a recurrent-Transformer network according to claim 1 is characterized in that: In step S2, the recurrent-Transformer neural network includes an input layer, a GRU layer, a Transformer layer, a fully connected layer, and an output layer. The acceleration data of the earthquake is X. X is input into the input layer of the recurrent-Transformer neural network, passes through the GRU layer to capture the positional relationship in the acceleration data, passes through the Transformer layer to learn the global dependency and nonlinear characteristics between the main shock and aftershock sequences, passes through the fully connected layer, and is mapped to the displacement response Y of the structure by the output layer. Among them, x n represents the acceleration data of the nth time step, represents the real number domain, y n represents the displacement response at the nth time step, where n represents the total number of time steps; During the training and optimization process, the mean square error is used as the loss function and the gradient is calculated through the back propagation algorithm.

4. The method for rapid assessment of mainshock and aftershock seismic resilience based on a cyclic-Transformer network according to claim 3 is characterized in that: The Transformer layer includes a multi-head attention module, which is used to calculate the query vector, key vector, and value vector of the earthquake acceleration data. The attention scores of multiple heads are calculated separately and the total score of the multi-head attention is obtained by concatenation. The calculation formula is: Q=HW Q ,K=HW K ,V=HW V MultiHead(Q,K,V)=Concat(head1,…,head h )W o Among them, Q represents the query vector, H represents the output result of the GRU layer, K represents the key vector, V represents the value vector, and W Q represents the weight matrix corresponding to Q, W K represents the weight matrix corresponding to K, W V represents the weight matrix corresponding to V, softmax(·) represents the normalized exponential function, d k Indicates the dimension of the key vector K, head i represents the attention score of the p-th attention head, represents the weight matrix corresponding to Q of the p-th attention head, represents the weight matrix corresponding to K of the p-th attention head, represents the weight matrix V corresponding to the pth attention head, Concat(·) represents vector concatenation, head h represents the attention score of the hth attention head, p = 1, 2, ... h, h represents the total number of attention heads, W o Represents the total weight of each attention head.

5. The method for rapid assessment of mainshock and aftershock seismic resilience based on a recurrent-Transformer network according to claim 1 is characterized in that: In step S3, the dual amplitude modulation process includes the following contents: The main shock and aftershock seismic acceleration can be expressed as the combination of the main shock acceleration, interval, and aftershock acceleration. The specific expression is: X={x1,x2,…,x n } T =[X main ,X0,X after ] Among them, X main represents the main shock acceleration sequence, X0 represents the zero vector, X after represents the aftershock acceleration sequence; X main and X after For amplitude modulation, the specific formula is: in, represents the main shock acceleration sequence after amplitude modulation, represents the aftershock acceleration sequence after amplitude modulation, α represents the amplitude modulation coefficient, and max() represents the maximum value of the orientation quantity; The amplitude-modulated main shock acceleration sequence and aftershock acceleration sequence are combined to obtain the acceleration sequence after double amplitude modulation.

6. The method for rapid assessment of mainshock and aftershock seismic resilience based on a recurrent-Transformer network according to claim 1 is characterized in that: In step S4, the fragility function is used to represent the building damage probability corresponding to different earthquake intensities. The specific expression is: F(x)=P(EDP a >C a |IM a =x,DS mi ) Among them, P(EDP a >C a |IM a =x,DS mi ) indicates that the initial state of the main epicenter is DS mi , and the aftershock intensity index IM a = x, the probability of reaching a certain aftershock damage state; EDP a represents the engineering demand parameter under aftershock; C a Indicates the structural bearing capacity under aftershocks; DS mi It represents the damage state under the main shock; P() represents the probability; x represents the value of the aftershock intensity index.

7. The method for rapid assessment of mainshock and aftershock seismic resilience based on a cyclic-Transformer network according to claim 1 is characterized in that: In step S5, the corresponding damage index is evaluated according to the damage probability of the building under different earthquakes. The specific formula is: Among them, L indicates that the main shock damage state is DS mi The damage index at the time of R DSj represents the damage index at the jth aftershock damage state; P(DS aj |IM a ,DS mi ) indicates that the aftershock intensity is IM a , the initial state of the main shock is DS mi When the structure reaches the destruction state DS in the aftershock aj The probability when P(DS aj |IM a ,DS mi ) is calculated as: Among them, EDP a Represents the engineering demand parameter under aftershock action; LS ai+1 The damage state index of the i+1th damage state under aftershock; LS ai represents the damage state index of the ith damage state under aftershock; Estimated recovery time: Among them, T REj represents the repair time required for the jth damage state; T RE Indicates the estimated time required for recovery; The functional recovery curve of the structure is: Among them, f(t,t0,T RE ) represents the expected functional recovery curve, t0 represents the time when the earthquake occurs, and t represents the time.

8. A system for the rapid assessment method of mainshock and aftershock seismic resilience based on a cyclic-Transformer network as claimed in claim 1, characterized in that: include: A database construction module is used to establish a finite element model of the building structure, input multiple sets of main shock and aftershock earthquake accelerations into the finite element model, obtain corresponding structural displacement response data, and establish a database including multiple sets of accelerations and their corresponding structural displacement responses; The model training module is used to build a recurrent-Transformer neural network, and uses the database in the database construction module to train and optimize the recurrent-Transformer neural network and determine the optimal parameters; The prediction module is used to perform dual amplitude modulation processing on the main shock and aftershock sequence and then input it into the optimized recurrent-Transformer neural network to predict the displacement response of the structure under the dual amplitude modulation main shock and aftershock sequence; The building damage probability acquisition module is used to use the displacement response obtained by the prediction module to obtain the building damage probability corresponding to different earthquake intensities under the main shock and aftershock scenarios using maximum likelihood estimation; The seismic toughness assessment module is used to evaluate the corresponding damage index, calculate the recovery time, and predict the functional recovery curve of the structure based on the building damage probability obtained by the building damage probability acquisition module to complete the seismic toughness assessment.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for rapid assessment of mainshock and aftershock seismic resilience based on a recurrent-Transformer network as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for rapid assessment of mainshock and aftershock seismic resilience based on a recurrent-Transformer network described in any one of claims 1 to 7 is executed.

Images