Building structure health assessment method, system, equipment and medium

Abstract

The invention provides a building structure health assessment method, system and device and a medium. The method comprises the steps of obtaining building state data of a preset part of a building structure; inputting the building state data into a pre-trained fusion prediction model to obtain building state prediction data; wherein the fusion prediction model comprises a multi-model prediction layer and a fusion layer; inputting the building state prediction data into a pre-generated digital twin model to correct model parameters; inputting the building state data into the corrected digital twinborn model to obtain building state simulation data; and according to the building state simulation data, a weighted comprehensive score is calculated through a PCA method, and the weighted comprehensive score is converted into a health probability. According to the method, the advantages of multiple models are fused, closed-loop updating is realized in combination with digital twinning, the actual demand is considered while the evaluation accuracy and reliability are improved, and the method has wider applicability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of structural health monitoring, and particularly relates to a building structure health assessment method, system, device and medium. BACKGROUND

[0002] The health assessment of key building structures such as the containment of nuclear power plants currently relies on manual inspection or single-point sensors, which have sparse data, delayed response and high misjudgment rate. In recent years, although AI prediction models have been introduced, the mainstream solutions mainly use single LSTM (Long Short-Term Memory) or GNN (Graph Neural Network), which are difficult to simultaneously depict the spatial topology and long-term coupling evolution of the structure, and the prediction accuracy is limited. Moreover, the model is highly black-boxed and lacks explainability, and cannot intuitively express the damage correlation between nodes. At the same time, the existing system lacks the ability to close loop with the digital twin, the parameter updating is slow, and it is difficult to support predictive maintenance and emergency decision-making. SUMMARY

[0003] The present application provides a building structure health assessment method, system, device and medium to solve the technical problem that the single prediction model in the prior art is difficult to simultaneously depict the spatial topology and long-term coupling evolution relationship of the structure, and lacks the ability to close loop when combined with the digital twin model, making it difficult to ensure the accuracy of the judgment result.

[0004] The building structure health assessment method, system, device and medium provided by the present application, the building structure health assessment method comprises: acquiring building state data of a preset part of a building structure; inputting the building state data into a pre-trained fusion prediction model to obtain building state prediction data; wherein the fusion prediction model comprises a multi-model prediction layer and a fusion layer; inputting the building state prediction data into a pre-generated digital twin model to correct model parameters; and inputting the building state data into the corrected digital twin model to obtain building state simulation data; calculating a weighted comprehensive score according to the building state simulation data by a PCA method, and converting the weighted comprehensive score into a health probability.

[0005] In an embodiment of the present application, the multi-model prediction layer comprises an LSTM sub-model, a Transformer sub-model and a GNN sub-model; the step of inputting the building state data into the fusion prediction model to obtain prediction data comprises: inputting the building state data into the LSTM sub-model to obtain building state short-term sub-prediction data; inputting the building state data into the Transformer sub-model to obtain building state global sub-prediction data; inputting the building state data into the GNN sub-model to obtain building state spatial sub-prediction data; inputting the building state short-term sub-prediction data, the building state global sub-prediction data and the building state spatial sub-prediction data into the fusion layer, and performing weighted operation on the prediction data sequence output by each sub-model according to the attention weight corresponding to each sub-model to obtain prediction data; wherein the attention weight is calculated based on the prediction data sequence of each sub-model through an attention mechanism.

[0006] In an embodiment of the present application, the building state data and the building state simulation data output by the digital twin model are used as new training samples to update the LSTM sub-model, the Transformer sub-model and / or the GNN sub-model.

[0007] In an embodiment of the present application, the step of inputting the building state data into the Transformer sub-model to obtain building state global sub-prediction data comprises: performing linear transformation on the building state data to obtain a Q (Query) matrix, a K (Key) matrix and a V (Value) matrix; performing dot product operation according to the transpose of the Q matrix and the K matrix to obtain a correlation score; after scaling processing of the correlation score, combining the V matrix through a softmax function to obtain a weight corresponding to each feature; and performing weighted summation on the V matrix by using the weight to obtain the building state global sub-prediction data.

[0008] In an embodiment of the present application, the step of inputting the building state data into the GNN sub-model to obtain building state spatial sub-prediction data comprises: constructing an adjacency matrix according to the building state data; inputting the adjacency matrix into the GNN sub-model to perform multi-layer graph convolution to obtain the building state spatial sub-prediction data.

[0009] In an embodiment of the present application, the step of calculating the health score from the building state simulation data by a PCA (Principal Component Analysis) method and converting the health score into a health probability comprises: constructing a data matrix and a covariance matrix of the building state simulation data, and performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues and corresponding eigenvectors; calculating the contribution rate corresponding to each eigenvalue, and selecting the principal components from the eigenvectors according to the cumulative contribution rate of each eigenvalue, and constructing an eigenvector matrix according to the eigenvectors corresponding to the principal components; calculating the score of each principal component according to the data matrix and the eigenvector matrix; performing weighted operation on the score of the principal component and the contribution rate corresponding to the eigenvalue to obtain the weighted comprehensive score, and converting the weighted comprehensive score into the health probability.

[0010] In an embodiment of the present application, the step of calculating the health probability from the building state simulation data by a PCA method comprises: calculating the health probability from the building state simulation data by a PCA method; matching the building state short-term sub-prediction data and the building state global sub-prediction data to the node dimension of the building state spatial sub-prediction data, calculating the residual between each two sub-models node by node, and taking the maximum value of the residual as a residual consistency index; introducing a graph disturbance into the GNN sub-model, calculating the change amount of the health probability, and taking it as a first judgment index; judging the residual consistency index and the first judgment index: when the residual consistency index is greater than or equal to a preset residual index threshold and / or the first judgment index is greater than or equal to a preset change threshold, it is determined that the accuracy of the health probability is low, and a re-evaluation is prompted; otherwise, it is determined that the accuracy of the health probability is high, and the health probability is taken as the health probability of the building structure.

[0011] The present application also provides a building structure health evaluation system, comprising: a data acquisition module for acquiring building state data of a preset part of a building structure; a fusion prediction module for inputting the building state data into a fusion prediction model to obtain building state prediction data; wherein the fusion prediction model comprises a multi-model prediction layer and a fusion layer; a twin simulation module for inputting the building state prediction data into a digital twin model to correct model parameters; and inputting the building state data into the corrected digital twin model to obtain building state simulation data; wherein the digital twin model is constructed according to the building state data of the preset part of the building structure; a probability acquisition module for calculating a weighted comprehensive score from the building state simulation data by a PCA method, and converting the weighted comprehensive score into a health probability.

[0012] The application further provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, which, when executed by the one or more processors, cause the electronic device to implement the building structure health assessment method as described above.

[0013] The application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor of a computer, causes the computer to perform the building structure health assessment method as described above.

[0014] The application has the following beneficial effects: the multi-model prediction layer of the fused prediction model is used to predict the building state data in the short term, the long term and the space of the building state, and meanwhile, the coupling evolution relationship between the structure space topology and the long time sequence is depicted. The prediction data is obtained through the fusion layer, so as to realize the fused prediction of the multi-model and avoid the uncertainty of the prediction result of a single model; the building state simulation data is obtained after the prediction data is input into the digital twin model, and the health probability is calculated according to the building state simulation data, wherein the data simulated by the digital twin model is used as a new training sample to update the fused prediction model in actual application, so as to realize the feedback loop and greatly reduce the misjudgment rate. BRIEF DESCRIPTION OF DRAWINGS

[0015] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and serve to explain the principles of the application. It is apparent that the accompanying drawings described below are only some embodiments of the application, and other drawings can be obtained according to the drawings without creative labor.

[0016] In the drawings:

[0017] Figure 1 A building structure health assessment method flowchart provided in an embodiment of the application;

[0018] Figure 2 A building structure health assessment system schematic diagram provided in an embodiment of the application;

[0019] Figure 3 A structure schematic diagram of an electronic device provided in an embodiment of the application. DETAILED DESCRIPTION

[0020] Following make the specific concrete example explain the embodiment of the present application, the person skilled in the art can be easily understood from the disclosure of the present application other advantages and efficacy.The present application can also be implemented or applied by another different specific embodiment, the details in the specification can be based on different views and applications, various modifications or changes are made without departing from the spirit of the present application.It should be noted that the following examples and features in the embodiment can be combined with each other without conflict.

[0021] It should be noted that the drawings provided in the following examples only illustrate the basic concept of the present application in a schematic manner, and only the components related to the present application are shown in the drawings, not the number, shape and size of the components when actually implemented, the actual implementation of each component type, quantity and proportion can be a random change, and the component layout type may be more complex.

[0022] In the following description, a large number of details are discussed to provide a more thorough explanation of the embodiments of the present application, however, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details, and in other embodiments, the known structures and devices are shown in the form of block diagrams rather than in the form of details to avoid making the embodiments of the present application difficult to understand.

[0023] As Figure 1 shown, the building structure health assessment method provided by the embodiment of the present application comprises the following steps:

[0024] S100, obtaining the building state data of the preset part of the building structure.

[0025] Physical field sensors such as stress, strain, acceleration, temperature and humidity are arranged at the more critical preset parts of the building structure, which are used to obtain the physical field data of the preset parts, and are uniformly represented as building state data.

[0026] S200, inputting the building state data into the pre-trained fusion prediction model to obtain building state prediction data; wherein the fusion prediction model comprises a multi-model prediction layer and a fusion layer.

[0027] Since the purpose of constructing the fusion prediction model is to simultaneously obtain short-term, global and spatial relationship information in the building state data through multiple models, the final building state prediction data of the fusion prediction model can be obtained by uniformly fusing the data predicted by multiple sub-models. In order to obtain the short-term, global and spatial relationship information in the building state data, in a preferred embodiment, the building state data is input into the LSTM sub-model to obtain building state short-term sub-prediction data, the attention mechanism is used to complete the extraction of the global information of the building state through the Transformer sub-model, and the GNN sub-model is used to obtain the building state spatial sub-prediction data. In addition to the above models, those skilled in the art can also select models capable of achieving the same function to obtain the short-term, global and spatial relationship information in the building state data according to actual needs.

[0028] Specifically, in an embodiment of the present application, the multi-model prediction layer includes an LSTM sub-model, a Transformer sub-model and a GNN sub-model; and the step of inputting the building state data into the fusion prediction model to obtain prediction data includes:

[0029] S210, inputting the building state data into the LSTM sub-model to obtain building state short-term sub-prediction data;

[0030] The obtained building state data is input into the LSTM sub-model, preprocessed and feature extracted, and then the LSTM sub-model is used to realize short-term prediction of the building state data, and time series dependent features are extracted to obtain building state short-term sub-prediction data.

[0031] S220, inputting the building state data into the Transformer sub-model to obtain building state global sub-prediction data;

[0032] Specifically, in an embodiment of the present application, the step of inputting the building state data into the Transformer sub-model to obtain building state global sub-prediction data includes:

[0033] The building state data is linearly transformed to obtain a Q matrix, a K matrix and a V matrix;

[0034] The correlation score is obtained by performing dot product operation according to the transpose of the Q matrix and the K matrix;

[0035] After scaling the correlation score, the weight corresponding to each feature is obtained by normalizing the V matrix through the softmax function;

[0036] The expression for calculating the weight corresponding to each feature through the softmax function in combination with the V matrix is:

[0037]

[0038] In the expression, Q represents a Q matrix, K represents a K matrix, V represents a V matrix, Attention(Q, K, V) represents a weighted feature representation calculated according to the Q matrix, the K matrix and the V matrix, from which a weight corresponding to each feature can be obtained. softmax(·) is used to normalize the attention weight, indicating the importance of different positions, d K is the dimension of each vector in the K matrix.

[0039] After obtaining the weight corresponding to the feature, the V matrix is weighted and summed using the weight, to obtain the building state global sub-prediction data.

[0040] S230, input the building state data into the GNN sub-model to obtain building state space sub-prediction data;

[0041] Specifically, in an embodiment of the present application, the step of inputting the building state data into the GNN sub-model to obtain building state space sub-prediction data includes:

[0042] An adjacency matrix is constructed according to the building state data;

[0043] An adjacency matrix A∈{0,1} N×N is established using a graph construction method, to represent the topological relationship between sensor nodes, where A is an adjacency matrix, and N is the number of sensor nodes; a preferred method of constructing the adjacency matrix is to construct it by thresholding: if the physical distance between two nodes is less than a set threshold, then the element A ij of the corresponding position is 1, otherwise it is 0, where i is the i-th row of the matrix, and j is the j-th column of the matrix.

[0044] The adjacency matrix is input into the GNN sub-model, and multi-layer graph convolution is performed to obtain the building state space sub-prediction data.

[0045] where the expression of the building state space sub-prediction data is:

[0046]

[0047] In the expression, H l is the node feature matrix of the l-th layer, each row corresponding to a feature vector of a node, H 1+l represents the building state space sub-prediction data. is a reinforced adjacency matrix, representing the adjacency matrix A plus the identity matrix I, i.e. represents adding a self-connection to the node, to avoid ignoring its own features, and D~ is a diagonal matrix, so that represents taking each diagonal element to the power of -0.5, W lis a trainable weight matrix for the l-th layer, and σ(·) represents a nonlinear activation function.

[0048] S240, input the building state short-term sub-prediction data, the building state global sub-prediction data and the building state spatial sub-prediction data into the fusion layer, and according to the attention weight corresponding to each sub-model, perform weighted operation on the prediction data sequence output by each sub-model to obtain building state prediction data; wherein the attention weight is calculated based on the prediction data sequence of each sub-model through the attention mechanism.

[0049] Since the data dimensions of the building state short-term sub-prediction data, the building state global sub-prediction data and the building state spatial sub-prediction data are not the same, it is necessary to unify the dimensions of the above data through the fusion layer, and further perform weighted operation on the prediction data output by each sub-module after the dimension unification to obtain the fused building state prediction data according to the attention weight preselected through the attention mechanism.

[0050] S300, input the prediction data into the digital twin model generated in advance to correct the model parameters; and input the building state data into the corrected digital twin model to obtain building state simulation data.

[0051] In actual application, the simulation value of the digital twin model is often used as a direct basis for judging the health of the building structure in the prior art, so the digital twin model is usually constructed in combination with the sensor data arranged at each part of the building structure in actual evaluation of the building;

[0052] In a preferred embodiment of the present application, the construction method of the digital twin model is: based on the structure data obtained by the sensor, a physical driving model is constructed according to the physical law, an LSTM model is trained after feature extraction to obtain a data-driven model. Then the physical driving model and the data-driven model are integrated to obtain the digital twin model.

[0053] In actual use, the prediction data is input into the digital twin model which has been constructed in advance to correct the parameters of the digital twin model, and the building state data is input into the corrected digital twin model to obtain more accurate building state simulation data.

[0054] Further, in an embodiment of the present application, the obtained building state data and the building state simulation data of the digital twin model are used as new training samples to update the LSTM sub-model, the Transformer sub-model and / or the GNN sub-model in the fusion prediction model, so as to not only ensure that the error of the fusion prediction model will not be increased, but also further improve the simulation accuracy of the digital twin model to make it more consistent with the actual situation.

[0055] S400, calculating a weighted comprehensive score according to the building state simulation data by a PCA method, and converting the weighted comprehensive score into a health probability.

[0056] Specifically, in an embodiment of the present application, the step of converting the health score into the health probability according to the building state simulation data by the PCA method comprises:

[0057] S410, constructing a data matrix and a covariance matrix of the building state simulation data, and performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues and corresponding eigenvectors;

[0058] The building state simulation data is a matrix composed of multiple groups of sample data obtained by multiple simulations according to parameters corresponding to sensors, so as to improve the accuracy of the result.

[0059] The building state simulation data is standardized to obtain a data matrix, and a covariance matrix is constructed according to the following expression:

[0060]

[0061] Wherein, Cov is the covariance matrix constructed based on the building state simulation data, X is the data matrix obtained by standardizing the building state simulation data, n is the number of samples, X T is the transpose of the matrix X;

[0062] The covariance matrix is eigenvalue decomposed to obtain eigenvalues and corresponding eigenvectors.

[0063] S420, calculating a contribution rate corresponding to each eigenvalue, and selecting principal components from the eigenvectors according to the cumulative contribution rate of each eigenvalue, and constructing an eigenvector matrix according to the eigenvectors corresponding to the principal components;

[0064] The contribution rate corresponding to each eigenvalue is calculated by the following formula:

[0065]

[0066] Wherein, ω i is the contribution rate corresponding to the i th eigenvalue, λ i is the i th eigenvalue, and p is the number of all eigenvalues. The eigenvalues are sorted from large to small, and the first k eigenvectors are selected as the principal components according to the cumulative contribution rate of the eigenvalues. An optimal way is: according to the cumulative contribution rate calculated from large to small, when the cumulative contribution rate is greater than or equal to 80% of the eigenvalues corresponding to the eigenvectors, the eigenvectors are selected as the principal components. The calculation formula of the cumulative contribution rate is:

[0067]

[0068] wherein ω is the cumulative contribution rate of the first k characteristic values, and the first k eigenvectors with a cumulative contribution rate greater than 80% are taken as principal components.

[0069] S430, calculating a score of each principal component according to the data matrix and the eigenvector matrix;

[0070] The eigenvectors corresponding to the characteristic values of the principal components are arranged in descending order of the characteristic values to construct an eigenvector matrix, and the score of the principal component is calculated according to the eigenvector matrix and the data matrix. The score of the principal component is calculated by the following formula:

[0071] PC = v · X,

[0072] wherein PC is the score matrix of the principal component, from which the score corresponding to each principal component can be obtained, X is the data matrix obtained after standardization of the building state simulation data, and v is the eigenvector matrix.

[0073] S440, performing weighted operation and averaging to obtain a weighted comprehensive score according to the score of the principal component and the contribution rate corresponding to the characteristic value, and converting the weighted comprehensive score into a health probability.

[0074] Since the cumulative contribution rate is calculated by the contribution rate of the characteristic value when the principal component is screened, multiple characteristic values may be screened, that is, the principal component may be multiple eigenvectors. Therefore, the contribution rate of each characteristic value is taken as the weight of the score of the principal component, and because the PCA method is a dimension reduction of sample data, the mean value is selected to obtain the final weighted comprehensive score. The calculation expression of the weighted comprehensive score is:

[0075]

[0076] wherein H is the weighted comprehensive score of the principal component, PC i is the score of the i-th principal component, and n is the sample number. After obtaining the weighted comprehensive score, the score value is converted into a probability value by a probability density function, and the expression is:

[0077]

[0078] wherein S is the health probability, and it is taken as the health probability for judging the health condition of the building structure, so as to intuitively evaluate the health condition of the building structure. The higher the probability value is, the healthier the corresponding building structure is.

[0079] In actual application, model deployment will realize more flexible high-precision analysis combined with the Internet, that is, a lightweight LSTM sub-model in the fusion prediction model is deployed to the local to realize preliminary real-time prediction, and the cloud runs the complete fusion prediction model to obtain high-precision analysis.

[0080] To ensure the credibility of the analysis results, the output results of the sub-models and the building state simulation data output by the digital twin model need to be judged. Further, the health probability is calculated according to the building state simulation data by a PCA method, including:

[0081] The health probability is calculated according to the building state simulation data by a PCA method.

[0082] The building state short-term sub-prediction data and the building state global sub-prediction data are matched to the node dimension of the building state space sub-prediction data, the residual error between each two sub-models is calculated node by node, and the maximum residual error is taken as the residual error consistency index.

[0083] A graph perturbation is introduced into the GNN sub-model, the change amount of the health probability is calculated, and it is taken as the first judgment index.

[0084] The residual error consistency index and the first judgment index are judged.

[0085] When the residual error consistency index is greater than or equal to a preset residual index threshold and / or the first judgment index is greater than or equal to a preset change threshold, it is determined that the accuracy of the health probability is low, and a re-evaluation is prompted. The residual error consistency index can be used to judge whether the health probability anomaly is caused by the parameter anomaly of the fusion prediction model. After introducing the graph perturbation, if the first judgment index is greater than or equal to the preset change threshold, it indicates that the importance of the data obtained by the corresponding sensor is high, and the influence on the health probability is large. At the same time, it also indicates that the layout relationship of the actual sensor or the hardware has a problem and needs to be repaired.

[0086] Otherwise, it is determined that the accuracy of the health probability is high, and it is taken as the health probability of the building structure.

[0087] The health probability judged by the above verification method has higher credibility and is suitable for realizing more stable building structure health evaluation in combination with alarm triggering and repair prompting.

[0088] For example, Figure 2As shown, the building structure health assessment system 200 comprises a data acquisition module 210, a fusion prediction module 220, a twin simulation module 230, and a probability acquisition module 240. The data acquisition module 210 is configured to acquire building state data of a preset part of a building structure. The fusion prediction module 220 is configured to input the building state data into a fusion prediction model to obtain building state prediction data. The fusion prediction model comprises a multi-model prediction layer and a fusion layer. The twin simulation module 230 is configured to input the prediction data into a digital twin model to correct model parameters, and input the building state data into the corrected digital twin model to obtain building state simulation data. The digital twin model is constructed according to the building state data of the preset part of the building structure. The probability acquisition module 240 is configured to calculate a health score from the building state simulation data by a PCA method, and convert the health score into a health probability.

[0089] The above modules can be embedded in a processor in a computer device in a hardware format or independent of the processor in the computer device, or stored in a memory in the computer device in a software format, so as to be called by the processor to perform operations corresponding to the modules.

[0090] It should be noted that, in order to highlight the innovative part of the present application, modules not closely related to solving the technical problems proposed in the present application are not introduced in the present embodiment, but this does not mean that there are no other modules in the present embodiment.

[0091] As shown in Figure 3 The electronic device 1 can comprise a memory 11, a processor 12, and a bus, and can further comprise a computer program stored in the memory 11 and executable on the processor 12, such as a residual calculation program between sub-models.

[0092] The memory 11 includes at least one type of readable storage medium, such as a flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 can also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the memory 11 can include both an internal storage unit and an external storage device of the electronic device 1. The memory 11 can be used to store application software and various data installed in the electronic device 1, such as parameters of a digital twin model, and can also be used to temporarily store data that has been output or will be output.

[0093] The processor 12 can be composed of an integrated circuit in some embodiments, such as a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more combinations of a central processing unit (CPU), a microprocessor, a digital processing chip, a graphics processor, and various control chips, etc. The processor 12 is a control unit of the electronic device 1, which connects various components of the entire electronic device 1 through various interfaces and lines, and executes programs or modules stored in the memory 11 and calls data stored in the memory 11 to perform various functions and process data of the electronic device 1.

[0094] The processor 12 executes an operating system and various application programs installed in the electronic device 1. The processor 12 executes the application programs to implement the steps in the building structure health assessment method described above.

[0095] For example, the computer program can be divided into one or more modules, which are stored in the memory 11 and executed by the processor 12 to complete the present application. The one or more modules can be a series of computer program instruction segments that can complete a specific function, which are used to describe the execution process of the computer program in the electronic device 1.

[0096] The above embodiments are only illustrative of the principles of the present application and its efficacy, and are not intended to limit the present application. Any modification or change made by any person skilled in the art without departing from the spirit and scope of the present application shall be covered by the claims of the present application.

Claims

1. A method of building structure health assessment, characterized by, The method comprises: obtaining building state data of a preset part of a building structure; inputting the building state data into a pre-trained fusion prediction model to obtain building state prediction data; wherein the fusion prediction model comprises a multi-model prediction layer and a fusion layer; inputting the building state prediction data into a pre-generated digital twin model to correct model parameters; and inputting the building state data into the corrected digital twin model to obtain building state simulation data; calculating a weighted comprehensive score from the building state simulation data by a PCA method, and converting the weighted comprehensive score into a health probability.

2. The building structural health assessment method according to claim 1, wherein, The multi-model prediction layer comprises an LSTM sub-model, a Transformer sub-model and a GNN sub-model; and the step of inputting the building state data into the fusion prediction model to obtain prediction data comprises: inputting the building state data into the LSTM sub-model to obtain building state short-term sub-prediction data; inputting the building state data into the Transformer sub-model to obtain building state global sub-prediction data; inputting the building state data into the GNN sub-model to obtain building state spatial sub-prediction data; inputting the building state short-term sub-prediction data, the building state global sub-prediction data and the building state spatial sub-prediction data into the fusion layer, and performing weighted operation on the prediction data sequence output by each sub-model according to the attention weight corresponding to each sub-model to obtain prediction data; wherein the attention weight is calculated based on the prediction data sequence of each sub-model by an attention mechanism.

3. The building structural health assessment method according to claim 2, wherein, The building state data and the building state simulation data output by the digital twin model are used as new training samples to update the LSTM sub-model, the Transformer sub-model and / or the GNN sub-model.

4. The building structural health assessment method according to claim 2, wherein, The step of inputting the building state data into the Transformer sub-model to obtain building state global sub-prediction data comprises: performing linear transformation on the building state data to obtain Q matrix, K matrix and V matrix; performing dot product operation on the transpose of Q matrix and K matrix to obtain correlation score; performing scaling processing on the correlation score, combining the V matrix and performing normalization by a softmax function to obtain the weight corresponding to each feature; performing weighted summation on the V matrix by using the weight to obtain the building state global sub-prediction data.

5. The building structural health assessment method according to claim 2, wherein, The step of inputting the building state data into the GNN sub-model to obtain building state spatial sub-prediction data comprises: constructing an adjacency matrix according to the building state data; inputting the adjacency matrix into the GNN sub-model to perform multi-layer graph convolution to obtain the building state spatial sub-prediction data.

6. The building structural health assessment method according to claim 1, wherein, The step of calculating a health score from the building state simulation data by a PCA method, and converting the health score into a health probability comprises: constructing a data matrix and a covariance matrix of the building state simulation data, and performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues and corresponding eigenvectors; Calculate the contribution rate corresponding to each eigenvalue, and select the principal component from the feature vector according to the contribution rate of each eigenvalue, and construct a feature vector matrix according to the principal component; Calculate the score of the principal component according to the data matrix and the feature vector matrix; According to the score of the principal component and the contribution rate corresponding to the eigenvalue, the weighted comprehensive score is obtained by weighted operation and averaging, and the weighted comprehensive score is converted into health probability.

7. The building structural health assessment method according to claim 2, wherein, According to the building state simulation data, the health probability is calculated by the PCA method, which includes: According to the building state simulation data, the health probability is calculated by the PCA method; Match the building state short-term sub-prediction data and the building state global sub-prediction data to the node dimension of the building state space sub-prediction data, calculate the residual between each two sub-models node by node, and take the maximum value of the residual as the residual consistency index; Introduce graph perturbation into the GNN sub-model, calculate the change amount of the health probability, and take it as the first judgment index; Determine the residual consistency index and the first judgment index: When the residual consistency index is greater than or equal to the preset residual index threshold and / or the first judgment index is greater than or equal to the preset change threshold, it is determined that the accuracy of the health probability is low, and re-evaluation is prompted; Otherwise, it is determined that the accuracy of the health probability is high, and it is taken as the health probability of the building structure.

8. A building structure health assessment system, characterized by, The system includes: A data acquisition module for acquiring building state data of a predetermined part of a building structure; A fusion prediction module for inputting the building state data into a fusion prediction model to obtain building state prediction data; wherein the fusion prediction model includes a multi-model prediction layer and a fusion layer; A twin simulation module for inputting the building state prediction data into a digital twin model to correct the model parameters; and inputting the building state data into the corrected digital twin model to obtain building state simulation data; wherein the digital twin model is constructed according to the building state data of the predetermined part of the building structure; A probability acquisition module for calculating a weighted comprehensive score from the building state simulation data by a PCA method, and converting the weighted comprehensive score into a health probability.

9. An electronic device, comprising: The electronic device includes: One or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, cause the electronic device to implement the building structure health evaluation method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored thereon, which, when executed by a processor of a computer, causes the computer to execute the building structure health evaluation method according to any one of claims 1 to 7.

Images