Method, device, equipment, medium and product for predicting quality risk of building structure

Abstract

The present application discloses a method, device, equipment, medium and product for predicting the quality risk of a building structure. The method includes: extracting the structure from a pre-constructed target building information model to obtain a plurality of target building structure data, where the target building structure data at least includes building foundation structure data, building main body structure data, and building floor structure data; inputting the target building structure data into a pre-trained target prediction model to obtain a corresponding prediction result of the building structure quality risk, and the target prediction model is trained according to a plurality of first building structure data samples and second building structure data samples obtained by transforming the feature space of the first building structure data samples. According to the embodiments of the present application, the target prediction model can maintain high prediction accuracy and robustness when processing building structure data of different types and complexities.

Description

Technical Field

[0001] This application belongs to the field of building safety technology, and particularly relates to a method, device, equipment, medium and product for predicting the quality risk of building structures. Background Art

[0002] In the field of construction engineering, predicting the quality risk of building structures is crucial, which involves the evaluation of aspects such as the stability, safety and durability of building structures. With the rapid development of the construction industry, the complexity and diversity of building structures are increasing continuously, posing higher requirements for the accuracy of predicting the quality risk of building structures.

[0003] Existing building structure quality risk prediction models are mainly based on traditional statistical methods and machine learning algorithms. These methods perform well in dealing with simple and standardized building structure data, but when faced with complex and diverse building structure data, it is often difficult to accurately capture the building structure features and the complex relationships between them, resulting in deviations in prediction results. Therefore, for the models in the related technology when faced with building structure data of new types or higher structural complexity, their prediction performance often significantly decreases, showing poor robustness. Summary of the Invention

[0004] Embodiments of this application provide a method, device, equipment, medium and product for predicting the quality risk of building structures, so as to at least solve the problems of insufficient prediction accuracy and poor robustness existing in dealing with complex and diverse building structure data in the related technology.

[0005] In the first aspect, embodiments of this application provide a method for predicting the quality risk of building structures, including:

[0006] Performing structure extraction on a pre-constructed target building information model to obtain a number of target building structure data, where the target building structure data at least includes building foundation structure data, building main structure data, and building floor structure data;

[0007] Inputting the target building structure data into a pre-trained target prediction model to obtain a corresponding prediction result of the building structure quality risk, where the target prediction model is trained according to a plurality of first building structure data samples and second building structure data samples obtained by transforming the feature space of the first building structure data samples.

[0008] In the second aspect, embodiments of this application provide a device for predicting the quality risk of building structures, where the device includes:

[0009] An extraction module for extracting the structure of a pre-built target building information model to obtain a number of target building structure data, where the target building structure data at least includes building foundation structure data, building main body structure data, and building floor structure data;

[0010] An input module for inputting the target building structure data into a pre-trained target prediction model to obtain a corresponding building structure quality risk prediction result, where the target prediction model is trained based on a number of first building structure data samples and second building structure data samples obtained by transforming the feature space of the first building structure data samples.

[0011] In a third aspect, an embodiment of the present application provides an electronic device, which includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the steps of the building structure quality risk prediction method described in any one of the embodiments of the first aspect are implemented.

[0012] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the steps of the building structure quality risk prediction method described in any one of the embodiments of the first aspect are implemented.

[0013] In a fifth aspect, an embodiment of the present application provides a computer program product, the program product is stored in a storage medium, and the program product is executed by at least one processor to implement the steps of the building structure quality risk prediction method provided in the first aspect of the embodiments of the present application.

[0014] The building structure quality risk prediction method, device, equipment, medium and product of the embodiments of the present application are based on the sample building structure data after feature space transformation, iteratively optimize the preset model, and by continuously adjusting the model parameters, the trained target prediction model can maintain high prediction accuracy and robustness when processing building structure data of different types and complexities. Description of the Drawings

[0015] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required to be used in the embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0016] Figure 1 is a flowchart of a building structure quality risk prediction method provided by an embodiment of the present application;

[0017] Figure 2 is a flowchart of a training method of a target prediction model provided by an embodiment of the present application;

[0018] Figure 3 It is a schematic structural diagram of a building structure quality risk prediction device provided by an embodiment of the present application;

[0019] Figure 4 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

[0020] Reference numerals:

[0021] Building structure quality risk prediction device 300, extraction module 301, input module 302,

[0022] Electronic device 400, processor 401, memory 402, communication interface 403, bus 410. Detailed implementation manners

[0023] The features and exemplary embodiments of various aspects of the present application will be described in detail below. To make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without some of these specific details. The following description of the embodiments is only intended to provide a better understanding of the present application by showing examples of the present application.

[0024] It should be noted that in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprises", "comprising" or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0025] In the field of construction engineering, the prediction of building structure quality risks is crucial, which involves the evaluation of aspects such as the stability, safety and durability of building structures. With the rapid development of the construction industry, the complexity and diversity of building structures have been increasing continuously, posing higher requirements for the accuracy of building structure quality risk prediction.

[0026] Existing building structure quality risk prediction models mainly rely on traditional statistical methods and machine learning algorithms. These methods perform well when dealing with simple and standardized building structure data. However, when faced with complex and diverse building structure data, it is often difficult to accurately capture the building structure characteristics and the complex relationships between them, resulting in biased prediction results.

[0027] Specifically, when the building structure quality risk prediction models in the related technologies deal with complex and diverse building structure data, it is often difficult to accurately capture the structural characteristics and the complex relationships between them. This is because the representation forms of data with different building types and structural complexities in the feature space vary greatly, and it is difficult for existing models to be effectively adapted. Due to the difficulty in accurately capturing complex structural characteristics, there are often large errors in predicting the building structure quality risk in the related technologies, leading to inaccurate prediction results. Therefore, when the models in the related technologies face new types or building structure data with higher structural complexity, their prediction performance often significantly decreases, showing poor robustness.

[0028] To solve the problems of the related technologies, the embodiments of the present application provide a method, device, equipment, medium, and product for predicting the quality risk of building structures.

[0029] Next, in combination with the accompanying drawings, through specific embodiments and their application scenarios, the method for predicting the quality risk of building structures provided by the embodiments of the present application will be described in detail.

[0030] Figure 1 The flowchart of a method for predicting the quality risk of building structures according to an embodiment of the present application is shown. As Figure 1 shown, the method for predicting the quality risk of building structures may specifically include the following steps:

[0031] S101. Perform structural extraction on a pre-constructed target building information model to obtain a number of target building structure data, where the target building structure data at least includes building foundation structure data, building main structure data, and building floor structure data;

[0032] S102. Input the target building structure data into a pre-trained target prediction model to obtain corresponding building structure quality risk prediction results, where the target prediction model is trained according to a plurality of first building structure data samples and second building structure data samples obtained by transforming the feature space of the first building structure data samples.

[0033] Thus, based on the example building structure data after feature space transformation, the preset model is iteratively optimized. By continuously adjusting the model parameters, the trained target prediction model can maintain high prediction accuracy and robustness when dealing with different types and complexities of building structure data.

[0034] It should be noted that the embodiments of this application include multiple target building information models corresponding to multiple different construction projects. Taking a new high-rise residential building project as an example, its target building information model may include overall building planning information, building function layout information, building structure design parameters, etc. Then, for the target building structure data in S101, it includes at least one building structure data generated by extracting the structure from the target building information model.

[0035] In some embodiments, in S101, during the structure extraction process, information directly related to the building structure is screened out from the target building information model. Specifically, for a high-rise residential building, building foundation structure information is extracted, including but not limited to the type of foundation (such as pile foundation), the size of the foundation (such as pile diameter, pile length, etc.), and the design value of the bearing capacity of the foundation; for the building's main structure, the distribution of columns, the size of columns (such as diameter, height, etc.), the layout of beams, the size of beams (such as width, height, etc.), and the building materials used for beams and columns (such as the strength grade of concrete, the model of steel, etc.) are extracted; and, building floor structure information, such as the thickness of the floor slab, the structural form of the stairs (such as slab stairs or beam stairs), etc.

[0036] Optionally, the target building information model can also be the target building information model corresponding to a commercial complex building project. For a commercial complex, due to its complex structure, during the structure extraction process, the large-span structure requirements of the commercial space need to be considered. Therefore, the target building structure data not only includes building foundation structure data, building main structure data, and building floor structure data, but also includes information related to special structures such as large-span beams and arches, such as their structural forms, force characteristics, and material properties. In addition, since there may be a large change in the human flow load in the commercial complex, the target building structure data also includes information related to the structure of the evacuation passage for people, such as but not limited to: the width of the evacuation stairs, the structural stability of the evacuation passage, etc.

[0037] In this way, based on the building structure data extracted from multiple different target building information models, it is used as the target building structure data and input into the target prediction model for predicting the quality risk of the building structure.

[0038] It should be noted that before S102, a target prediction model is constructed and trained to use the target prediction model to output the prediction result of the building structure quality risk, so as to provide a reliable guarantee for the safety and durability of the building project.

[0039] Specifically, the training process of the target prediction model provided by the embodiments of this application can refer to Figure 2 . Such as Figure 2As shown in the figure, a method for training a target prediction model provided by an embodiment of the present application may include the following steps: S201 to S204.

[0040] S201. Obtain a plurality of first building structure data samples;

[0041] S202. Use a transformation iterative learning sub-instruction to perform feature space transformation on the first building structure data samples to obtain second building structure data samples after the transformation of the feature space;

[0042] S203. Generate a training cost result when generating a model iteration instruction for predicting the quality risk of the operating building structure corresponding to the preset model according to the first building structure data samples and the second building structure data samples;

[0043] S204. Adjust the weight parameters of the preset model according to the training cost result, and train the adjusted preset model until a preset training stop condition is met to obtain a trained target prediction model.

[0044] It should be noted that the preset model in the embodiment of the present application can be applied to the prediction of the quality risk of the building structure of various types of buildings, including but not limited to residential buildings, commercial buildings, industrial factories, etc. Then, the sample data for model training includes a large number of building structure data samples corresponding to various types of buildings respectively, and this embodiment does not make specific limitations on this.

[0045] Optionally, the first building structure data sample at least includes structural mechanics parameters.

[0046] In some embodiments, in S202, when the initial state of the first building structure data sample is the first feature space, establish an association relationship between the structural mechanics parameters and the preset non-structural factor parameters to obtain a second building structure data sample after the transformation of the feature space. The state of the second building structure data sample is the second feature space, and the second building structure data sample includes structural mechanics parameters, non-structural factor parameters, and features of the association relationship between the building structure and non-structural factors; when the initial state of the first building structure data sample is the second feature space, simplify the association relationship between the building structure and non-structural factors to obtain a second building structure data sample after the transformation of the feature space, and the state of the second building structure data sample is the first feature space.

[0047] That is to say, for each first building structure data sample, the model iteration instruction for feature space transformation needs to be executed. Taking the first building structure data sample of a multi-story residential building as an example, it at least includes building foundation structure data (such as foundation type, depth, etc.), building main structure data (such as the size and material of beams and columns, etc.), and building floor structure data (such as floor slab thickness, staircase structure, etc.). It should be noted that these data are in a specific feature space in the initial state, which is defined as the first feature space or the second feature space in this embodiment.

[0048] Specifically, hereinafter, the first building structure data sample with the initial state of the first feature space is called the first sample building structure data, and the first building structure data sample with the initial state of the second feature space is called the second sample building structure data. The specific implementation methods for transforming the feature space of these two data samples will be described in detail respectively.

[0049] For the first sample building structure data, for example, in the sample data of a multi-story residential building, the foundation structure data may be represented in a specific format and numerical range, which is an embodiment of the first feature space. Run the model iteration instruction for the first feature space transformation from the first feature space to the second feature space. Among them, the second feature space is determined through research and analysis. For the parameter iteration optimization of the building structure quality risk prediction model, it is the ideal feature space for the model to adapt to when performing risk prediction.

[0050] In this process, it is necessary to convert the foundation type in the foundation structure data. In the first feature space, the foundation type is usually represented according to the traditional classification method, such as simple classifications like shallow foundation and deep foundation. When converting to the second feature space, it needs to be refined into more precise types, including but not limited to raft foundation, pile foundation, etc., and establish an associated relationship with relevant geological parameters. At the same time, for the size data of beams and columns, when in the first feature space, it is represented by actual length and width values. When converting to the second feature space, it is converted into relative values related to the overall structure ratio of the building and establish new connections with factors such as the load borne by the building.

[0051] For the second sample of building structure data, taking a commercial building as an example, its initial feature space is the second feature space, and it is necessary to run the model iteration instruction for the second feature space transformation from the second feature space to the first feature space. The structure data of the commercial building is relatively complex. In the second feature space, its structure data is in a highly optimized and complexly correlated form. For example, its column grid structure data may be associated with various factors such as the functional zoning of the building and the distribution of pedestrian flow. When converting to the first feature space, it is necessary to simplify these complex associations, convert the column grid structure data into a more basic representation form. Specifically, it can be represented according to simple geometric shapes and dimensions, while weakening the association with other non-structural factors.

[0052] It can be understood that during the execution of the model iteration instruction for the entire feature space transformation, for different samples of building structure data, whether it is the transformation from the first feature space to the second feature space or from the second feature space to the first feature space, specific algorithms and rules need to be followed, that is, the transformation iteration learning sub-instruction. Specifically, based on the principles of building structure engineering, the preset quality risk assessment requirements, and the model optimization target information, the feature space transformation algorithm is determined.

[0053] During the specific implementation, during the conversion process, factors such as the mechanical properties of building materials and the stability of the structure need to be considered. For different parts of the building structure, such as beams and columns in the frame structure, during the feature space transformation, conversion operations are performed according to their force characteristics and mutual relationships in the entire building structure.

[0054] In this way, the feature space transformation is performed on the sample of building structure data, converting it from the original feature space to a feature space more suitable for model prediction. This transformation process can capture the structural features and their complex relationships, enabling the model to better understand the data structure. That is, it can dynamically adjust the feature space according to the type and complexity of the building structure data, enabling the building structure quality risk prediction model to better adapt to different types and complexities of data. The key innovation point lies in the dynamic and self-adaptive nature of the feature space transformation, significantly improving the prediction accuracy and robustness of the model.

[0055] Furthermore, in some embodiments, in S203, a feature space inverse transformation is performed on the second building structure data sample after the transformed feature space to obtain a third building structure data sample; the deviation between the first building structure data sample and the third building structure data sample is determined; in the case where the deviation is greater than or equal to a preset threshold, adjustment information for the transformation algorithm corresponding to the transformation iteration learning sub-instruction is determined according to the deviation; the transformation iteration learning sub-instruction is optimized through the adjustment information; and a training cost result is generated according to the deviation and the adjustment information.

[0056] In addition, in some embodiments, a plurality of second building structure data samples are input into the encoding module of the preset model; the encoding module encodes the plurality of second building structure data samples respectively to obtain a plurality of sample feature vectors; and the encoding module cross-fuses the plurality of sample feature vectors in the spatial domain dimension.

[0057] During specific implementation, set the internal parameters of the encoding module, including but not limited to the weights, bias terms, activation function types, and learning rates of the neural network layers, etc., to initialize the encoding module. In addition, prepare appropriate data structures for the intermediate data and final results during the encoding process, such as tensors, lists, or matrices, to efficiently store and access the data.

[0058] During specific implementation, the input data of the encoding module is a plurality of example building structure data (i.e., second building structure data samples) after feature space transformation. These data have been in the second feature space through feature space transformation, denoted as {S1', S2',..., S n '}, where n is the number of example building structure data.

[0059] During specific implementation, perform normalization processing on each example building structure data S i ', which can include normalization, standardization, and data augmentation, to ensure the stability and consistency of the data during the encoding process, and output the normalized data {S1'', S2'',..., S n ''}; then, extract key features from the normalized data {S1'', S2'',..., S n ''}, which specifically include structural dimensions, material types, load-bearing capacities, etc., and output the feature set {F1, F2,..., F n}, completing the preprocessing of the example building structure data.

[0060] During specific implementation, input each feature F i into the first layer (fully connected layer or convolutional layer) of the encoding module, and perform linear transformation and non-linear activation on the feature through the weights and biases of this layer, and output the preliminarily encoded features {E1, E2,..., E n}; perform feature fusion on the preliminarily encoded features {E1, E2,..., E n}, and output the fused feature E fused , where the specific implementation method of feature fusion can include concatenation, weighted summation, and attention fusion mechanism; further, input the fused feature E fusedInput into the deep network structure of the encoding module. These deep structures include multiple convolutional layers, pooling layers, fully connected layers, or recurrent neural network layers, etc. Through layer-by-layer processing, higher-level feature representations are gradually refined; after each layer or a specific layer, the features are non-linearly transformed through an activation function to enhance the model's expressive power. At the same time, techniques such as dropout and batch normalization are used to prevent overfitting and accelerate convergence, and the deeply encoded feature E is output. deep , where the activation function can include ReLU, Sigmoid, or Tanh, and this embodiment does not make specific limitations in this regard; in the spatial domain dimension, the features in E deep are cross-fused. Specifically, it can be achieved through a preset network structure (cross-attention network or graph convolutional network) or by using a specific fusion algorithm to capture the complex relationships and interactions between features, and the cross-fused feature E cross is output; the spatial dimension of E cross is transformed. Specifically, it can be through operations such as transpose, reshape, or unfold to convert it into a form suitable for subsequent processing or decoding, and the transformed feature E transformed is output; E transformed is input into the last layer of the encoding module (i.e., the fully connected layer or output layer). Through the processing of this layer, the final encoding representation {C1, C2, ..., C n} is generated, where C i is the encoding representation of the sample building structure data S i '. It should be understood that the encoding representation {C1, C2, ..., C n} is the input of the preset model for the subsequent building structure quality risk prediction task.

[0061] The above encoding process will be further described in detail through a scenario example below.

[0062] Use the building structure quality risk prediction model to encode and represent the sample building structure data after multiple transformed feature spaces generated by the model iteration instructions for feature space transformation. Taking the building structure data such as large commercial centers and high-rise office buildings mentioned before as an example, these data are in the second feature space after feature space transformation and serve as the input data of the model.

[0063] First, initialize the parameters and data structures of the encoding module. Multiple sample building structure data after the input feature space transformation, such as for a large commercial center, the raft foundation data (thickness, reinforcement, etc.) of the foundation structure, the mega-column data (structural form, combination method, etc.) of the main structure, and the large-span steel structure roof data (structural parameters, etc.) of the atrium, as well as the pile foundation data of the foundation structure of a high-rise office building, the frame column and beam data of the main structure, the floor slab and staircase data of the floor structure, etc., are denoted as {S1', S2',..., S n '}. Set internal parameters for the encoding module. For example, for the weights of the neural network layer, they can be initialized to specific values according to the scale and complexity of the building structure data. The bias term will also be set to an appropriate value. The activation function type can choose the ReLU function, and the learning rate is set to a small value (such as 0.001) to ensure the stability of training. At the same time, prepare data structures for intermediate data and final results, such as storing the foundation structure data in tensor form, the main structure data in matrix form, etc., for efficient storage and access of data.

[0064] Furthermore, preprocess the sample building structure data. For the input {S1', S2',..., S n '}, perform data normalization operations. Taking the raft foundation thickness data of a large commercial center as an example, its original value may be in a relatively large range, and it is converted to the [0, 1] interval through normalization. For the concrete strength grade data of the frame columns of a high-rise office building, perform standardization processing to make it conform to the standard normal distribution. At the same time, data augmentation operations can be carried out, such as slightly randomly perturbing the structural parameter data of the large-span steel structure roof of the atrium in the commercial center to increase the diversity of data, and output the normalized data {S1'', S2'',..., S n ''}. Then extract key features from {S1'', S2'',..., S n ''}. For the raft foundation of the commercial center, extract features such as its bearing capacity and connection characteristics with the surrounding structures; for the frame columns of the office building, extract features such as their cross-sectional dimensions and steel strength, and output the feature set {F1, F2,..., F n}.

[0065] Secondly, perform feature encoding and preliminary fusion. Input each feature F i into the first layer of the encoding module. For example, for the bearing capacity feature of the raft foundation of a commercial center, perform a linear transformation through the weights and biases of the fully connected layer, and then perform non-linear activation through the ReLU activation function to obtain the preliminarily encoded feature. The same operation is also performed on the cross-sectional dimension feature of the frame column of the office building, and output the preliminarily encoded features {E1, E2,..., E n}. Then, fuse these preliminarily encoded features. For the structural data of commercial centers and office buildings, the weighted summation method can be adopted. Different weights are assigned according to the importance of different structural parts to the overall structural risk, and the preliminarily encoded features are fused into E fused .

[0066] Then, perform deep encoding and feature extraction. Input E fused into the deep network structure of the encoding module. Among them, the deep structure includes a convolutional layer. For the structural data of commercial centers, the convolutional layer will extract the structural features of different local regions, and the pooling layer will downsample these features to reduce the data volume. When the structural data of office buildings passes through the fully connected layer, the global features are further integrated. After each layer, the ReLU activation function can be used to perform non-linear transformation on the features. For example, for the structural features of the large-span steel structure roof of the atrium in a commercial center, the complex structural relationship can be better represented after passing through the activation function. At the same time, the dropout technique is used to prevent overfitting. For example, in the fully connected layer, neuron connections are randomly disconnected with a certain probability (such as 0.5), and the features after deep encoding, E deep , are output.

[0067] Next, perform feature cross-fusion and spatial dimension transformation. In the spatial domain dimension, cross-fuse the features in E deep . For the structural data of commercial centers and office buildings, through a special network structure such as a cross-attention network, the features of different structural parts such as raft foundations and frame columns, and large-span roofs and floor slabs are correlated with each other. For example, the structural stability features of the large-span roof are cross-fused with the support capacity features of the frame columns to capture the complex relationship between them, and the features after cross-fusion, E cross , are output. Then, perform spatial dimension transformation on E cross , such as converting its three-dimensional structural feature representation into a two-dimensional form suitable for subsequent processing, and output the transformed features E transformed .

[0068] Finally, realize the final generation of the encoded representation. Input E transformed into the last layer of the encoding module. For the structural data of commercial centers and office buildings, the last layer is a fully connected layer. Through the processing of this layer, the final encoded representation {C1, C2, ..., C n} is generated. Among them, C1 is used to represent the encoded representation of the raft foundation structural data of the commercial center, and C2 is used to represent the encoded representation of the frame column structural data of the office building.

[0069] As an optional embodiment, the encoded representation can be verified and optimized, and then the optimized encoded representation can be stored in an appropriate storage medium (such as memory, database, or file system) so that the subsequent model can quickly access and call it. That is, in the building structure quality risk prediction task, the stored encoded representation is used as input data and input into the trained target prediction model. Through the prediction and decision-making of the target prediction model, the quality risk prediction result of the building structure is output. For example, when predicting the structural quality risk of a commercial center, the model will output the quality risk prediction result of the building structure based on the encoded representation of structural data such as raft foundation and mega-columns, combined with the pre-trained model parameters.

[0070] During specific implementation, the quality of the encoded representation is verified by calculating metrics such as the similarity, distance, or reconstruction error between the encoded representations to ensure that they can accurately reflect the characteristics of the original sample building structure data. According to the verification results, the parameters of the encoding module are adjusted and optimized. Specifically, the weights and bias terms can be updated through the backpropagation algorithm, or regularization techniques can be used to reduce the complexity of the model to improve the accuracy and robustness of the encoded representation.

[0071] Taking the encoded representations of the structural data of a commercial center and an office building as an example, the cosine similarity between them can be calculated to ensure that the encoded representations of different structural data can accurately reflect the characteristics of the original data. According to the verification results, the parameters of the encoding module are adjusted through the backpropagation algorithm. If it is found that the accuracy of the encoded representation is insufficient, such as there is a deviation in the structural association between the encoded representations of the raft foundation and the frame column, the weights and bias terms of the neural network layer are adjusted, or the L2 regularization technique is used to reduce the complexity of the model and improve the accuracy and robustness of the encoded representation.

[0072] In this way, by designing the encoding module, the encoded representation of the sample building structure data after transforming the feature space is carried out by the encoding module, and cross-fusion is performed in the spatial domain dimension. This feature encoding and fusion mechanism can capture the complex relationships between structural features, further enhancing the model's understanding ability of complex structural features, thereby improving the prediction performance of the model.

[0073] Furthermore, in some embodiments, in S204, multiple sample feature vectors are used as the input of the preset model, and the adjusted preset model is trained until the preset training stop condition is met to obtain the trained target prediction model.

[0074] In some embodiments, based on the example building structure features after encoding representation, the training results of the building structure quality risks for the example building structure data after being transformed in multiple feature spaces are determined. Taking commercial centers and office buildings as examples, for the foundation structure of a commercial center, according to the information related to bearing capacity in the raft foundation encoding representation and combining with the risk assessment criteria in the building structure quality risk prediction model, if the encoding representation of the raft foundation's bearing capacity shows that it may face overloading risks in the future (possibly due to increased loads caused by the functional expansion of the commercial center), the corresponding risk level is marked in the risk training results of the foundation structure. For the main structure of an office building, according to the structural strength information in the encoding representations of the frame columns and beams, if the encoding representations of the frame columns and beams show insufficient structural strength considering factors such as seismic loads, the risk situation is marked in the risk training results of the main structure.

[0075] In some embodiments, based on the training results of the building structure quality risks for the example building structure data after being transformed in multiple feature spaces, the training cost results when generating model iteration instructions for running the building structure quality risk prediction by the building structure quality risk prediction model are obtained. If in the risk training results of a commercial center, it is found that the risk assessment errors of the raft foundation and the large-span steel structure roof of the atrium are relatively large, such as there is actually a risk but the model fails to accurately identify it or vice versa, then the training cost results will be relatively high, indicating that the model requires more optimization in this regard. For an office building, inaccurate risk assessment of the frame columns and floor slabs will also lead to an increase in the training cost results.

[0076] In another embodiment, the training cost results corresponding to the model iteration instructions for feature space transformation are determined and compared with the training cost results corresponding to the model iteration instructions for building structure quality risk prediction; the training cost results corresponding to the model iteration instructions for feature space transformation are weighted and calculated with the training cost results of the model iteration instructions for building structure quality risk prediction, and based on the calculation results, the parameters of the preset model are iteratively optimized until the training termination conditions are met. In this way, based on the model iteration instructions for feature space transformation and the model iteration instructions for building structure quality risk prediction, the parameter iterative optimization of the building structure quality risk prediction model is achieved.

[0077] In specific implementation, first determine the training cost results corresponding to the model iteration instructions for feature space transformation and the training cost results corresponding to the model iteration instructions for building structure quality risk prediction. For feature space transformation, for example, when previously performing feature space transformation on the structural data of commercial centers and office buildings, if there is a lot of data distortion or the structural relationship cannot be accurately transformed during the conversion from the first feature space to the second feature space or vice versa, the training cost result will be relatively high. For building structure quality risk prediction, if there is a large deviation between the prediction result and the actual risk situation when predicting the structural data of commercial centers and office buildings, the training cost result will also be relatively high.

[0078] In specific implementation, perform weighted calculation on the training cost result corresponding to the model iteration instruction for feature space transformation and the training cost result of the model iteration instruction for building structure quality risk prediction. For example, according to the characteristics of the building structure data and the key focus direction of the model, assign a weight of 0.4 to the training cost result of feature space transformation and a weight of 0.6 to the training cost result of building structure quality risk prediction. Then, based on the calculation result, perform parameter iteration optimization on the building structure quality risk prediction model. For the weights of the neural network layers in the model, if the calculation result indicates that the inaccurate risk prediction is mainly due to unreasonable weight allocation for structural features, such as too low a weight for the long-span roof structure feature of a commercial center, resulting in a deviation in risk assessment, then increase its weight. Continuously repeat this process until the building structure quality risk prediction model meets the training termination conditions. For example, when the accuracy rate of the prediction result reaches more than 90% and the training cost result is lower than the preset threshold, it is considered that the model has been trained well and can be used for actual building structure quality risk prediction tasks.

[0079] In this way, by continuously adjusting the model parameters, realize the model iteration optimization based on feature space transformation and building structure quality risk prediction, so that the model can maintain high prediction accuracy and robustness when processing building structure data of different types and complexities, and can ensure the effectiveness and reliability of the model in actual applications.

[0080] Thus, based on the above S201 to S204, the target prediction model is trained. Then, use the target prediction model to process the target building structure data and output the building structure quality risk prediction result.

[0081] In addition, this application also provides another method for predicting building structure quality risk.

[0082] Specifically, another method for predicting the quality risk of a building structure according to an embodiment of the present application may include the following steps: running a model iteration instruction for feature space transformation based on the sample building structure data scheduled during the parameter iteration optimization process of the building structure quality risk prediction model; calling the building structure quality risk prediction model to run a model iteration instruction for predicting the quality risk of the building structure based on the sample building structure data after the feature space transformation generated by the model iteration instruction for feature space transformation, where the model iteration instruction for predicting the quality risk of the building structure is used to iteratively enhance the prediction performance of the building structure quality risk prediction model for the building structure quality risk structure; performing parameter iteration optimization on the building structure quality risk prediction model based on the model iteration instruction for feature space transformation and the model iteration instruction for predicting the quality risk of the building structure to generate a trained building structure quality risk prediction model; determining any input target building structure data, where the target building structure data includes at least one building structure data generated by extracting the structure from the target building information model; and inputting the target building structure data into the building structure quality risk prediction model to generate a prediction result of the building structure quality risk of the target building structure data.

[0083] The following introduces the specific implementation manners of the above steps.

[0084] After the feature space transformation is completed, the sample building structure data after the feature space transformation is obtained. Taking a multi-story residential building as an example, after the conversion from the first feature space to the second feature space, the foundation structure data, the main structure data, and the floor structure data are all in the new feature space. At this time, the building structure quality risk prediction model is called to process these transformed sample building structure data.

[0085] For the transformed residential building structure data, the building structure quality risk prediction model first encodes it. In this process, the model will convert the data of each structural part of the residential building into a form suitable for model calculation according to its own algorithm structure. For example, for the foundation bearing capacity data in the foundation structure, it can be encoded as a specific numerical vector, which contains the relationship information between the foundation bearing capacity and other relevant structural parameters. For the beam-column strength data in the main structure, it will also be encoded into a numerical representation form related to the overall stability of the building.

[0086] Based on the example building structure features after this encoding representation, determine the training results of the building structure quality risks for these example building structure data after transformation in the feature space. The model will conduct risk assessments on each structural part of the residential building according to the pre-set risk assessment criteria. For example, for the foundation part, if there are potential risks in the encoded foundation bearing capacity data compared with the overall weight and expected load of the building, the model will mark the corresponding risk level. For the beam-column part, if the encoded strength values show risk hazards after considering factors such as the service life of the building and the extreme loads it may bear, they will also be accurately evaluated.

[0087] Then, based on the training results of the building structure quality risks for these example building structure data after transformation in the feature space, generate the training cost results when the building structure quality risk prediction model runs the model iteration instructions for building structure quality risk prediction. This training cost result reflects the performance of the model in the current iteration process. If in the risk assessment of the residential building, the model accurately identifies the quality risks that may be caused by uneven settlement of the foundation in the basic structure, but there are certain deviations in the long-term durability risk assessment of the beam-column structure, then this training cost result will reflect the direction and degree of the model's need for further improvement in this regard.

[0088] Similarly, for the example building structure data after transformation in the feature space of the commercial building, the building structure quality risk prediction model will also perform similar operations. The structure of the commercial building is more complex and its risk factors are more diverse. During the encoding representation process, the model needs to consider the interactive effects of more structural and non-structural factors. For example, when encoding the frame structure of the commercial building, not only the structural strength and stability of the beam-columns need to be considered, but also the dynamic effects of commercial activities on the building structure, such as the change in load distribution when the crowd is dense. When determining the training results of the building structure quality risks, the model needs to comprehensively evaluate the structural risks of the commercial building under different operating scenarios, such as the fire resistance performance of the structure in a fire scenario and the seismic performance of the structure in an earthquake scenario. The training cost results generated based on these evaluation results can provide accurate basis for the further optimization of the model.

[0089] Furthermore, after obtaining the training cost results related to the model iteration instructions for feature space transformation and the model iteration instructions for building structure quality risk prediction, perform parameter iteration optimization on the building structure quality risk prediction model.

[0090] Take the processing results of the sample building structure data of multi-story residential buildings and commercial buildings as an example. If, in the processing of the building structure data of multi-story residential buildings, it is found that in the risk prediction of the foundation structure by the model, due to the unreasonable setting of a certain parameter, there is a deviation in the assessment of the foundation settlement risk, then adjust the relevant parameters in the building structure quality risk prediction model according to this feedback. For example, if the weight parameter related to the geotechnical parameters of the foundation soil is set improperly, then adjust this weight parameter to improve the accuracy of the model's assessment of the foundation settlement risk. For the case of commercial buildings, assume that in the building structure quality risk prediction model, the parameters related to the assessment of the structural fire resistance performance did not fully consider the influence of the interior decoration materials of the building in the previous iteration. In the process of this iteration optimization, adjust the parameters related to the assessment of the structural fire resistance performance according to the training cost results obtained from the model iteration instructions of the building structure quality risk prediction. For example but not limited to: adding parameter items related to the flammability of decoration materials, or adjusting the weights of the existing parameters related to the fire spread speed.

[0091] In the whole process of parameter iteration optimization, comprehensively consider the processing results of multiple sample building structure data. In addition to residential buildings and commercial buildings, it can also include sample data of other types of buildings such as industrial plants. For example, there may be the influence of special equipment loads on the building structure in industrial plants. If this special load distribution and its influence on the structural risk were not well considered in the previous model, then adjust the model parameters according to the relevant training cost results to ensure that the model can accurately evaluate the structural quality risk of industrial plants under the equipment operation state.

[0092] With continuous iteration optimization, continuously adjust the parameters of the building structure quality risk prediction model until the model can achieve satisfactory prediction performance when processing various sample building structure data. When the prediction performance of the model reaches the set standard, for example, in the risk prediction of a large number of different types of building structure data, the accuracy rate reaches a relatively high level and the error rate is controlled within an acceptable range, it is considered that a trained building structure quality risk prediction model has been generated. This trained model can accurately predict the building structure quality risk, whether for the building structure data corresponding to new residential buildings, commercial buildings, or industrial plants and other various types of buildings.

[0093] Furthermore, input the pre-determined target building structure data into the already trained building structure quality risk prediction model. Take a new high-rise residential building as an example. After inputting its target building structure data (including building foundation structure data, building main structure data, and building floor structure data) into the model, the model first processes these data.

[0094] For building foundation structure data, the model evaluates whether the foundation type and size can meet the load requirements of high-rise residential buildings based on the knowledge and algorithms accumulated during the previous training process. If there are risks in the design value of the bearing capacity of the foundation after considering the self-weight of the high-rise residential building, the load of the residents, and the possible wind load and seismic load, the model will mark the corresponding risk level. For example, if under certain geological conditions, the design of the pile length and pile diameter of the pile foundation may not be able to effectively resist the horizontal load during an earthquake, the model will mark the seismic risk of the foundation structure as a higher level.

[0095] For building main structure data, the model analyzes whether the structural dimensions and material properties of columns and beams can ensure the overall stability of the building. For example, if in a high-rise residential building, the diameter of some columns is too small, there may be a risk of buckling under compression considering the building height and the upper load, and the model will evaluate the structural stability risk of these columns and give corresponding risk warnings. At the same time, for the structure of beams, if the design of the beam height and width is unreasonable, resulting in insufficient bending resistance, the model will also identify this risk and conduct a quantitative assessment.

[0096] For building floor structure data, the model checks whether the floor slab thickness meets the residential use requirements and the safety of the staircase structure. If the floor slab thickness is too thin, it may not be able to bear the load generated by furniture placement and human activities, and the model will mark the structural safety risk of the floor slab. For the staircase structure, if there may be a risk of crowd congestion in its structural form during emergency evacuation, the model will also give the corresponding risk assessment result.

[0097] Similarly, for the target building structure data of commercial complexes, the model will also conduct a comprehensive risk assessment. When evaluating the long-span structure of a commercial complex, the model will judge its structural safety risk under normal use and special load conditions (such as the concentrated load of people when holding large-scale activities) based on the structural form and material properties. For the personnel evacuation passage structure, the model will comprehensively consider factors such as the width of the evacuation staircase, the length and layout of the evacuation passage, and evaluate the personnel evacuation risk in case of fire or other emergencies.

[0098] Thus, the building structure quality risk prediction model generates building structure quality risk prediction results for different types of target building structure data, and these results can provide valuable references for the design, construction, and operation of building projects, helping relevant personnel to timely discover and solve quality risk problems in building structures.

[0099] Among them, when dealing with the parameter iteration optimization of the building structure quality risk prediction model, it involves model iteration instructions for running feature space transformation on sample building structure data. The following is a detailed scenario example.

[0100] For the sample building structure data scheduled during the parameter iterative optimization process of the building structure quality risk prediction model, take the data of various building types provided by a large construction company as an example. First, look at the operations related to the first sample building structure data. Take the building structure data of a high-rise office building as the first sample building structure data, which is derived from the sample building information model data with the first feature space. In the sample building information model data of this office building, there are many detailed information about the building structure, such as the layout, pile diameter, pile length, etc. of the pile foundation in the foundation structure, the spacing, cross-sectional dimensions, and concrete strength grade of the frame columns in the main structure, the span, cross-sectional shape, and reinforcement of the frame beams, as well as the floor thickness, floor height, and staircase structure form of the floor structure. These data are all in the first feature space, and this first feature space has its own feature representation form. For example, the foundation structure data may exist in the form of a simple numerical combination and has no complex connection with other structural or non-structural factors.

[0101] For the first sample building structure data of this high-rise office building, the present invention runs the model iteration instruction of the first feature space transformation from the first feature space to the second feature space. This second feature space is an ideal feature space set to better adapt the building structure quality risk prediction model during risk prediction. During this transformation process, for the foundation structure data, the pile foundation data that was simply represented in the first feature space will be deeply integrated with the geological exploration data of the site where the office building is located when converting to the second feature space. For example, the pile diameter and pile length of the pile foundation will establish a correlation relationship with geological factors such as the bearing capacity of the stratum and the groundwater level, forming a new representation form characterized by the interaction between geology and foundation structure. For the frame column data in the main structure, in the first feature space, its cross-sectional dimensions and concrete strength grade are independently represented, while when converting to the second feature space, they will be combined with external load factors such as the wind load and seismic fortification intensity of the office building. For example, the cross-sectional dimensions of the frame columns will be re-quantified according to the wind load calculation results and seismic requirements, and the concrete strength grade will also be associated with the mechanical properties of the structure under different load combinations. For the floor thickness data in the floor structure, which is just a simple value in the first feature space, when converting to the second feature space, it will be combined with factors such as the internal functional partition and personnel load distribution of the office building. For example, if a certain floor is an office area with a large and uneven personnel load, the representation of the floor thickness will be adjusted in combination with this load distribution to better reflect its impact on the building structure quality risk.

[0102] Next, look at the operations related to the second sample building structure data. Take the building structure data of a large commercial center as the second sample building structure data, which comes from the sample building information model data with the second feature space. The building structure information model data of this commercial center covers complex structure information, such as the raft foundation thickness, reinforcement configuration of the foundation structure, and the associated relationship with the underground parking lot structure, the special structural form of the mega-columns in the main structure, the combination method of steel structure and concrete structure, and the structural parameters of the large-span steel structure roof of the atrium, etc. These data are initially in the second feature space, and the representation form of this second feature space is already relatively complex, including the associations of many structural and non-structural factors.

[0103] For the second sample building structure data of this large commercial center, the present invention runs the model iteration instruction of the second feature space transformation from the second feature space to the first feature space. In this process, for the raft foundation data of the foundation structure, when converting from the second feature space to the first feature space, it is necessary to simplify the representation of its complex associated relationships with structures such as the underground parking lot structure. For example, the original representation form of the complex relationships between the raft foundation thickness and the column grid structure of the underground parking lot, the vehicle load transfer path, etc. in the second feature space will be simplified to a representation form that only considers the basic relationships of the raft foundation's own thickness, reinforcement, and foundation bearing capacity when converting to the first feature space. For the structural parameters of the mega-columns and the large-span steel structure roof of the atrium in the main structure, when converting from the second feature space to the first feature space, the associations with non-structural factors such as the internal commercial layout and the evacuation of people flow in the commercial center will be weakened. For example, the structural form of the mega-columns may be associated with factors such as the atrium space layout of the commercial center and the distribution of surrounding stores in the second feature space, and when converting to the first feature space, it will focus on its own structural mechanical properties, such as the basic structural properties of compressive and flexural resistance, etc.

[0104] The model iteration instruction of the first feature space transformation here is of great significance. It can enhance the performance when performing feature space transformation from the first feature space to the second feature space. For example, when converting a high-rise office building from the first feature space to the second feature space, this instruction can ensure the accuracy and effectiveness of the foundation structure, main structure, and floor structure data during the conversion process. Through a series of algorithms and rules, it enables each structural data to better match the requirements of the target feature space, thereby improving the overall feature space transformation performance.

[0105] The model iteration instruction for the second feature space transformation synergistically enhances the performance during the feature space transformation from the first feature space to the second feature space by performing a feature space transformation from the second feature space to the first feature space. Taking the sample building structure data of large commercial centers and high-rise office buildings as an example, when converting the structure data of a commercial center from the second feature space to the first feature space, the algorithms and rules followed in this process can provide reference and supplementation for the conversion of the office building from the first feature space to the second feature space. For example, in dealing with the relationship between structural data and external loads, some processing methods of the commercial center in simplifying the association between structural and non-structural factors can provide optimization ideas for the office building in establishing the relationship between structural data and external loads, thus synergistically enhancing the performance during the feature space transformation from the first feature space to the second feature space.

[0106] Finally, the sample building structure data after transforming the feature space contains the sample building structure data after the transformation of the first sample building structure data through the model iteration instruction for the first feature space transformation from the first feature space to the second feature space. For example, after such a transformation for a high-rise office building, its foundation structure, main structure, and floor structure data are all in the new feature space, and the representation forms and correlation relationships of these data have changed, forming a new form of building structure data. This transformed sample building structure data will provide an important data basis for the operation of the subsequent building structure quality risk prediction model.

[0107] For the first sample building structure data (such as the building structure data of the high-rise office building mentioned above) scheduled during the parameter iteration optimization process of the building structure quality risk prediction model, the process of the model iteration instruction for the first feature space transformation from the first feature space to the second feature space is as follows.

[0108] First, perform a feature space transformation on the first sample building structure data from the first feature space to the second feature space. For high-rise office buildings, in the first feature space, the basic structure data (such as the layout of pile foundations, pile diameters, pile lengths, etc.) exists in a relatively simple and independent form, with weak correlations with other factors. When transforming to the second feature space, more factors need to be combined. For example, the pile diameter and pile length of the pile foundation not only need to consider their own values but also be correlated with the geological conditions of the site where the office building is located (such as the lithology of the strata, groundwater level, soil bearing capacity, etc.). Data such as the column spacing, cross-sectional dimensions, and concrete strength grade in the main structure are independently characterized in the first feature space. When transforming to the second feature space, they need to be combined with factors such as the wind load, seismic load, and overall building height faced by the office building. Data such as the floor slab thickness, floor height, and staircase structure form of the floor structure need to be related to the functional layout inside the office building (such as the distribution of office areas, equipment rooms, etc.) and the personnel load distribution.

[0109] Run the first feature space discrimination iteration sub-instruction corresponding to the model iteration instruction of the first feature space transformation based on the first sample building structure data after transforming the feature space, and generate the training cost result corresponding to the first feature space discrimination iteration sub-instruction. In this process, the first feature space discrimination iteration sub-instruction aims to iteratively enhance the discrimination performance of the building structure data generated by the feature space transformation from the first feature space to the second feature space. For the transformed basic structure data of high-rise office buildings, the discrimination iteration sub-instruction will check the rationality of the pile foundation after being correlated with the geological conditions. For example, if the pile diameter and pile length of the pile foundation can meet the bearing requirements of the office building under different working conditions (such as normal use, seismic action, etc.) after being adjusted according to the geological conditions. If, under a certain geological condition, the pile length is too short or the pile diameter is too small, resulting in insufficient bearing capacity, the discrimination iteration sub-instruction should be able to identify this situation. For the main structure, judge whether the structural performance of the frame columns is reasonable after being correlated with factors such as wind load, seismic load, and building height. If the cross-sectional dimensions of the frame columns do not conform to the principles of structural mechanics after considering these factors, the discrimination iteration sub-instruction should be able to detect it. For the floor structure, judge the safety of the floor slab thickness after being correlated with the functional layout and personnel load distribution. If the floor slab thickness does not meet the requirements after considering the load in the area where people are concentrated, the discrimination iteration sub-instruction should be able to determine it. Generate the training cost result based on these discrimination results. If the discrimination performance is good, the training cost result is low; if there are many discrimination errors, the training cost result is high.

[0110] Next, perform a feature space inverse transformation on the first sample building structure data after transforming the feature space from the second feature space to the first feature space. For the basic structure data of high-rise office buildings, return from the state deeply associated with geological conditions to a relatively independent representation form. For example, restore the pile foundation representation considering factors such as formation lithology and groundwater level to a form that only focuses on its own basic parameters. In the main structure, restore the frame column data associated with wind load, seismic load, and building height to the original relatively independent dimension and strength grade representation. In terms of floor structure, restore data such as floor slab thickness related to functional layout and personnel load distribution to the initial simple representation form.

[0111] Run the first feature space transformation iteration learning sub-instruction corresponding to the model iteration instruction of the first feature space transformation on the first sample building structure data after the feature space inverse transformation and the first sample building structure data before the feature space transformation to generate the training cost result of the first feature space transformation iteration learning sub-instruction. The first feature space transformation iteration learning sub-instruction is used to enhance the effectiveness of the building structure data obtained by the feature space transformation from the first feature space to the second feature space so as to improve the error discrimination probability when judging the first feature space discrimination iteration sub-instruction. For high-rise office buildings, compare the basic structure data after the inverse transformation with the original basic structure data. If during the feature space transformation process, some adjustments to the pile foundation data are found to have a large deviation from the original data after the inverse transformation, and this deviation may lead to structural safety risks, then the transformation iteration learning sub-instruction needs to adjust the relevant transformation algorithm. For the main structure, if the frame column data after the inverse transformation is compared with the original data and it is found that the previous transformation has caused a large error in the evaluation of structural performance, the transformation iteration learning sub-instruction needs to be optimized. For the floor structure, if there are unreasonable differences between the data such as floor slab thickness and the original data after the inverse transformation, affecting the judgment of structural safety, the transformation algorithm also needs to be improved. Generate the training cost result based on the results of these comparisons and adjustments. If a large number of adjustments to the transformation algorithm are required, the training cost result is high; if only a small amount of optimization is needed, the training cost result is low. This training cost result will provide an important basis for subsequent model optimization to continuously improve the accuracy and effectiveness of the feature space transformation from the first feature space to the second feature space, thereby enhancing the performance of the building structure quality risk prediction model.

[0112] For the second sample building structure data (such as the building structure data of the large commercial center mentioned before) scheduled during the parameter iteration optimization process based on the building structure quality risk prediction model, run the model iteration instruction of the second feature space transformation from the second feature space to the second feature space. The following is the detailed process.

[0113] First, perform a feature space transformation on the second sample building structure data from the second feature space to the first feature space. For a large commercial center, data such as the thickness of the raft foundation, reinforcement configuration, and the correlation with the underground parking lot structure in its basic structure have complex representation forms in the second feature space, including correlations with numerous structural and non-structural factors. When transforming to the first feature space, these data need to be simplified. For example, the representation form of the complex relationships between the thickness of the raft foundation and the column grid structure of the underground parking lot, vehicle load transfer paths, etc. in the second feature space should be converted to a representation form that only considers the basic relationships of the thickness, reinforcement, and bearing capacity of the raft foundation itself. Data such as the special structural form of the mega-columns in the main structure and the combination method of steel structure and concrete structure are correlated with non-structural factors such as the internal commercial layout and pedestrian evacuation in the commercial center in the second feature space. When converting to the first feature space, focus should be placed on its own structural mechanical properties, such as the representation of basic structural properties like the compressive and flexural capacities of the mega-columns, and weaken the correlation with non-structural factors. The structural parameters of the large-span steel structure roof in the atrium are correlated with factors such as the overall spatial layout and daylighting design of the commercial center in the second feature space. When converting to the first feature space, it should be simplified to only focus on its own structural characteristics, such as the representation form of basic parameters like the structural stability of the roof and the internal forces of the members.

[0114] Run the second feature space discrimination iteration sub-instruction corresponding to the model iteration instruction of the second feature space transformation on the second sample building structure data after the transformation feature space to generate the training cost result corresponding to the second feature space discrimination iteration sub-instruction. In this process, the second feature space discrimination iteration sub-instruction is used to iteratively strengthen the discrimination performance of the building structure data obtained by the feature space transformation from the second feature space to the first feature space. For the transformed basic structure data of a large commercial center, the discrimination iteration sub-instruction will check the rationality of the simplified representation of the raft foundation. For example, when only considering the basic relationship between the thickness, reinforcement, and bearing capacity of the raft foundation itself, it is necessary to determine whether this simplified representation can accurately reflect the performance of the raft foundation under different working conditions (such as normal use, uneven settlement, etc.). If the reinforcement of the raft foundation does not meet the bearing capacity requirements after simplified calculation, the discrimination iteration sub-instruction should be able to identify this situation. For the main structure, determine whether the structural performance of the giant column is reasonable after considering its own structural mechanical properties. If the calculation of the compressive or flexural capacity of the giant column does not conform to the principles of structural mechanics after simplified association, the discrimination iteration sub-instruction should be able to detect it. For the large-span steel structure roof of the atrium, determine whether its structural stability meets the requirements after being simplified to its own structural characteristics. If the internal force of the roof members leads to insufficient structural stability after simplified calculation, the discrimination iteration sub-instruction should be able to determine it. Generate the training cost result based on these discrimination results. If the discrimination performance is good, the training cost result is low; if there are many discrimination errors, the training cost result is high.

[0115] Next, perform a feature space inverse transformation on the second sample building structure data after the transformation feature space from the first feature space to the second feature space. For the basic structure data of a large commercial center, restore from the representation form that only considers its own basic relationship to the representation form that is related to the column grid structure of the underground parking lot, the vehicle load transfer path, etc. In the main structure, restore the giant column data that focuses on its own structural mechanical properties to the representation form that is related to non-structural factors such as the internal commercial layout and the evacuation of people flow in the commercial center. The structural parameters of the large-span steel structure roof of the atrium are restored from only focusing on its own structural characteristics to the representation form that is related to factors such as the overall spatial layout and lighting design of the commercial center.

[0116] Run the second feature space transformation iteration learning sub-instruction corresponding to the model iteration instruction of the second feature space transformation on the second sample building structure data based on the inverse transformation of the feature space and the second sample building structure data before the feature space transformation to generate the training cost result of the second feature space transformation iteration learning sub-instruction. The second feature space transformation iteration learning sub-instruction is used to enhance the effectiveness of the building structure data obtained by the feature space transformation from the second feature space to the first feature space so as to improve the error discrimination probability when discriminating the iteration sub-instruction in the second feature space. For large commercial centers, compare the foundation structure data after the inverse transformation with the original foundation structure data. If during the feature space transformation process, some adjustments to the raft foundation data are found to have a large deviation from the original data after the inverse transformation, and this deviation may lead to structural safety risks, then the transformation iteration learning sub-instruction needs to adjust the relevant transformation algorithm. For the main structure, if the comparison between the giant column data after the inverse transformation and the original data shows that the previous transformation has caused a large error in the evaluation of the structural performance, the transformation iteration learning sub-instruction needs to be optimized. For the atrium long-span steel structure roof, if its structural parameters have unreasonable differences from the original data after the inverse transformation, affecting the judgment of structural safety, the transformation algorithm also needs to be improved. Generate the training cost result based on the results of these comparisons and adjustments. If a large number of adjustments to the transformation algorithm are required, the training cost result is high; if only a small amount of optimization is needed, the training cost result is low. This training cost result will provide an important basis for subsequent model optimization to continuously improve the accuracy and effectiveness of the feature space transformation from the second feature space to the first feature space, thereby enhancing the performance of the building structure quality risk prediction model. Through such operations, the present invention can effectively process the building structure data of large commercial centers and other types through feature space transformation, laying a foundation for the accurate operation of the building structure quality risk prediction model. In actual operation, the present invention needs to perform precise calculations and discriminations on each data point and each structural element to ensure that the processing of the entire building structure data meets the requirements of building structure quality risk prediction and can continuously optimize the performance of the model so that it can accurately predict the building structure quality risk when processing different types of building structure data.

[0117] Thus, by introducing the model iteration instruction of feature space transformation, effective feature space transformation is performed on the sample building structure data, enabling the building structure quality risk prediction model to better adapt to different types and complexities of building structure data, thereby improving the accuracy and robustness of the prediction.

[0118] In some alternative embodiments, for the example building structure data after the transformed feature space generated based on the model iteration instruction of the feature space transformation, invoking the model iteration instruction for predicting the building structure quality risk by running the building structure quality risk prediction model includes: encoding and representing the multiple pieces of example building structure data after the transformed feature space generated by the model iteration instruction of the feature space transformation by using the building structure quality risk prediction model; wherein, the multiple pieces of example building structure data after the transformed feature space are the building structure data in the second feature space; determining the training result of the building structure quality risk for the multiple pieces of example building structure data after the transformed feature space based on the encoded example building structure features; and generating the training cost result when running the model iteration instruction for predicting the building structure quality risk by the building structure quality risk prediction model based on the training result of the building structure quality risk for the multiple pieces of example building structure data after the transformed feature space.

[0119] In some alternative embodiments, encoding and representing the multiple pieces of example building structure data after the transformed feature space generated by the model iteration instruction of the feature space transformation by using the building structure quality risk prediction model includes: inputting the multiple pieces of example building structure data after the transformed feature space generated by the model iteration instruction of the feature space transformation into the encoding module of the building structure quality risk prediction model; encoding and representing the multiple pieces of example building structure data after the transformed feature space by using the encoding module; and cross-fusing the building structure features corresponding to the multiple pieces of example building structure data after the transformed feature space in the spatial domain dimension by using the encoding module.

[0120] In summary, the advantages of the present invention are mainly reflected in the following aspects:

[0121] First of all, by introducing the model iteration instruction of the feature space transformation, the present invention can dynamically adjust the feature space, enabling the building structure quality risk prediction model to better adapt to different types and complexities of data. This advantage solves the problem in the prior art that the model is difficult to process complex and diverse data structures, and improves the accuracy and robustness of the prediction. In practical applications, this means that the present invention can more accurately evaluate the quality risks of various building structures, providing a more reliable guarantee for the safety and durability of building projects.

[0122] Secondly, the feature encoding and fusion mechanism of the present invention can capture the complex relationships between structural features, further enhancing the model's ability to understand complex structural features. This advantage enables the model to better understand the physical meaning and structural relationships behind the data when processing complex building structure data, thereby improving the accuracy and reliability of the prediction. In practical applications, this means that the present invention can more accurately predict the quality risks of complex building structures, providing a more scientific basis for the design and construction of building projects.

[0123] Finally, by proposing a model iterative optimization method based on feature space transformation and building structure quality risk prediction, the present invention can ensure that the model maintains high prediction accuracy and robustness when dealing with different types and complexities of data structures. This advantage makes the model more adaptable and reliable in practical applications and can cope with various complex and changeable situations. In practical applications, this means that the present invention can provide a more stable and reliable solution for the quality risk prediction of building projects, helping to improve the safety level and engineering quality of the entire construction industry.

[0124] It should be noted that some embodiments of the present application have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the above embodiments and still achieve the desired results. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0125] Based on the same inventive concept, corresponding to the method of any of the above embodiments, the present application also provides a building structure quality risk prediction device 300.

[0126] As Figure 3 shown, the building structure quality risk prediction device 300 may include:

[0127] An extraction module 301, configured to perform structure extraction on a pre-constructed target building information model to obtain a plurality of target building structure data, where the target building structure data at least includes building foundation structure data, building main structure data, and building floor structure data;

[0128] An input module 302, configured to input the target building structure data into a pre-trained target prediction model to obtain a corresponding building structure quality risk prediction result, where the target prediction model is trained according to a plurality of first building structure data samples and second building structure data samples obtained by transforming the feature space of the first building structure data samples.

[0129] In some embodiments, the building structure quality risk prediction device 300 further includes a training module ( Figure 3 not shown in the figure). Specifically, the training module includes the following units:

[0130] An acquisition unit, configured to acquire a plurality of first building structure data samples;

[0131] A transformation unit for performing feature space transformation on the first building structure data sample by using transformation iterative learning sub-instructions to obtain a second building structure data sample after the transformation feature space;

[0132] A generation unit for generating a training cost result when generating a model iteration instruction for predicting the quality risk of the operating building structure corresponding to a preset model according to the first building structure data sample and the second building structure data sample;

[0133] An adjustment unit for adjusting the weight parameters of the preset model according to the training cost result and training the adjusted preset model until a preset training stop condition is met to obtain a trained target prediction model.

[0134] In some alternative embodiments, the generation unit is specifically configured to: perform inverse feature space transformation on the second building structure data sample after the transformation feature space to obtain a third building structure data sample; determine the deviation between the first building structure data sample and the third building structure data sample; in the case where the deviation is greater than or equal to a preset threshold, determine adjustment information of the transformation algorithm corresponding to the transformation iterative learning sub-instruction according to the deviation; optimize the transformation iterative learning sub-instruction through the adjustment information; and generate a training cost result according to the deviation and the adjustment information.

[0135] Optionally, the first building structure data sample includes at least structural mechanics parameters.

[0136] In some alternative embodiments, the transformation unit is specifically configured to: establish an association relationship between the structural mechanics parameters and pre-set non-structural factor parameters in the case where the initial state of the first building structure data sample is the first feature space to obtain a second building structure data sample after the transformation feature space, the state of the second building structure data sample is the second feature space, and the second building structure data sample includes structural mechanics parameters, non-structural factor parameters, and features of the association relationship between the building structure and non-structural factors; simplify the association relationship between the building structure and non-structural factors in the case where the initial state of the first building structure data sample is the second feature space to obtain a second building structure data sample after the transformation feature space, and the state of the second building structure data sample is the first feature space.

[0137] In some alternative embodiments, the training module further includes an encoding unit for inputting a plurality of second building structure data samples into the encoding module of the preset model; respectively encoding the plurality of second building structure data samples by using the encoding module to obtain a plurality of sample feature vectors; and cross-fusing the plurality of sample feature vectors in the spatial domain dimension by using the encoding module.

[0138] In some alternative embodiments, the adjustment unit is specifically configured to: use a plurality of sample feature vectors as the input of a preset model, train the adjusted preset model until a preset training stop condition is met, and obtain a trained target prediction model.

[0139] It should be noted that for the convenience of description, when describing the above device, various modules are described separately according to their functions. Of course, when implementing the present application, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0140] The device in the above embodiment is used to implement the corresponding building structure quality risk prediction method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be elaborated here.

[0141] Based on the same technical concept, corresponding to the method in any of the foregoing embodiments, the present application also provides an electronic device.

[0142] Figure 4 FIG. shows a more specific schematic diagram of the hardware structure of the electronic device provided in this embodiment.

[0143] The electronic device 400 may include a processor 401 and a memory 402 storing computer program instructions.

[0144] Specifically, the above processor 401 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0145] The memory 402 may include a mass storage for data or instructions. By way of example and not limitation, the memory 402 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a tape, or a universal serial bus (USB) drive, or a combination of two or more of these. In a suitable case, the memory 402 may include a removable or non-removable (or fixed) medium. In a suitable case, the memory 402 may be internal or external to the integrated gateway disaster recovery device. In a specific embodiment, the memory 402 is a non-volatile solid state memory.

[0146] In certain embodiments, the memory may include read-only memory (ROM), random access memory (RAM), magnetic disk storage media devices, optical storage media devices, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. Thus, in general, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described in reference to the method according to one aspect of the present application.

[0147] The processor 401 reads and executes the computer program instructions stored in the memory 402 to implement any one of the building structure quality risk prediction methods in the above embodiments.

[0148] In some examples, the electronic device 400 may further include a communication interface 403 and a bus 410. Among them, as Figure 4 shown, the processor 401, the memory 402, and the communication interface 403 are connected through the bus 410 and complete communication with each other.

[0149] The communication interface 403 is mainly used to implement communication between various modules, devices, units, and / or devices in the embodiments of the present application.

[0150] The bus 410 includes hardware, software, or both, and couples the components of the online data flow metering device to each other. By way of example and not limitation, the bus 410 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a MicroChannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses or a combination of two or more of these. In a suitable case, the bus 410 may include one or more buses. Although the embodiments of the present application describe and illustrate specific buses, the present application contemplates any suitable bus or interconnect.

[0151] Exemplarily, the electronic device 400 may be a mobile phone, a tablet computer, a laptop computer, a handheld computer, an in-vehicle electronic device, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA), etc.

[0152] Based on the same inventive concept, corresponding to the method of any of the above embodiments, the present application further provides a non-transitory computer-readable storage medium. Computer program instructions are stored on the computer-readable storage medium; when the computer program instructions are executed by a processor, any one of the building structure quality risk prediction methods in the above embodiments is implemented. Examples of computer-readable storage media include non-transitory computer-readable storage media such as portable disks, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, and the like.

[0153] Based on the same inventive concept, corresponding to the method of any of the above embodiments, the present application further provides a computer program product, which includes computer program instructions. In some embodiments, the computer program instructions can be executed by one or more processors of a computer so that the computer and / or the processor execute the building structure quality risk prediction method. Corresponding to the execution subject of each step in each embodiment of the building structure quality risk prediction method, the processor that executes the corresponding step can belong to the corresponding execution subject.

[0154] It should be clear that the present application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order between steps after understanding the spirit of the present application.

[0155] The functional blocks shown in the above structure block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a functional card, and the like. When implemented in software, the elements of the present application are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium or transmitted via a data signal carried in a carrier wave on a transmission medium or a communication link. "Machine-readable medium" can include any medium that can store or transmit information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROMs, flash memories, erasable ROMs (EROMs), floppy disks, CD-ROMs, optical discs, hard disks, optical fiber media, radio frequency (RF) links, and the like. The code segment can be downloaded via a computer network such as the Internet, an intranet, and the like.

[0156] It should also be noted that in the exemplary embodiments mentioned in this application, some methods or systems are described based on a series of steps or devices. However, this application is not limited to the order of the above steps. That is to say, the steps can be executed in the order mentioned in the embodiments, can be different from the order in the embodiments, or several steps can be executed simultaneously.

[0157] The above has described various aspects of this application with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of this application. It should be understood that each block in the flowchart and / or block diagram, and the combination of each block in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing devices to generate a machine, such that these instructions executed by the processor of the computer or other programmable data processing devices enable the implementation of the functions / actions specified in one or more blocks of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It can also be understood that each block in the block diagram and / or flowchart, and the combination of the blocks in the block diagram and / or flowchart, can also be implemented by dedicated hardware that performs the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0158] As described above, the above is only the specific implementation manner of this application. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, modules, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for predicting the quality risk of a building structure, characterized in that, Including: Performing structure extraction on a pre - constructed target building information model to obtain a number of target building structure data, where the target building structure data at least includes building foundation structure data, building main body structure data, and building floor structure data; Inputting the target building structure data into a pre - trained target prediction model, and based on the risk assessment criteria in the target prediction model, obtaining a corresponding building structure quality risk prediction result, where the target prediction model is trained according to a number of first building structure data samples and second building structure data samples after transforming the feature space of the first building structure data samples; Before inputting the target building structure data into the pre - trained target prediction model, the method further includes: Obtaining a number of first building structure data samples; Using a transformation iterative learning sub - instruction to perform feature space transformation on the first building structure data samples to obtain second building structure data samples after transforming the feature space, where the feature space transformation includes dynamically adjusting the feature space according to the type and complexity of the building structure data; Generating a training cost result when generating a model iteration instruction for predicting the quality risk of the running building structure corresponding to the preset model according to the first building structure data samples and the second building structure data samples; Adjusting the weight parameters of the preset model according to the training cost result, and training the adjusted preset model until a preset training stop condition is met to obtain a trained target prediction model; The training cost result when generating a model iteration instruction for predicting the quality risk of the running building structure corresponding to the preset model according to the first building structure data samples and the second building structure data samples includes: Performing feature space inverse transformation on the second building structure data samples after transforming the feature space to obtain third building structure data samples; Determining the deviation between the first building structure data samples and the third building structure data samples; In the case where the deviation is greater than or equal to a preset threshold, determining adjustment information of the transformation algorithm corresponding to the transformation iterative learning sub - instruction according to the deviation; Optimizing the transformation iterative learning sub - instruction through the adjustment information; Generating a training cost result according to the deviation and the adjustment information.

2. The method according to claim 1, wherein The first building structure data samples at least include structural mechanics parameters; The performing feature space transformation on the first building structure data samples to obtain second building structure data samples after transforming the feature space includes: In the case where the initial state of the first building structure data samples is the first feature space, establishing an association relationship between the structural mechanics parameters and pre - set non - structural factor parameters to obtain second building structure data samples after transforming the feature space, where the state of the second building structure data samples is the second feature space, and the second building structure data samples include structural mechanics parameters, non - structural factor parameters, and features of the association relationship between the building structure and non - structural factors; In the case where the initial state of the first building structure data samples is the second feature space, simplifying the association relationship between the building structure and non - structural factors to obtain second building structure data samples after transforming the feature space, where the state of the second building structure data samples is the first feature space.

3. The method according to claim 1, characterized in that The method further includes: Inputting a plurality of second building structure data samples into the encoding module of the preset model; Using the encoding module to encode the plurality of second building structure data samples respectively to obtain a plurality of sample feature vectors; Using the encoding module to cross-fuse the plurality of sample feature vectors in the spatial domain dimension.

4. The method according to claim 1, characterized in that The adjusting the weight parameters of the preset model according to the training cost result and training the adjusted preset model until a preset training stop condition is met to obtain a trained target prediction model includes: Using the plurality of sample feature vectors as the input of the preset model, training the adjusted preset model until a preset training stop condition is met to obtain a trained target prediction model.

5. A device for predicting the quality risk of a building structure, characterized in that, The apparatus includes: An extraction module, configured to perform structure extraction on a pre-constructed target building information model to obtain a plurality of target building structure data, where the target building structure data at least includes building foundation structure data, building main body structure data, and building floor structure data; An input module, configured to input the target building structure data into a pre-trained target prediction model to obtain a corresponding building structure quality risk prediction result, where the target prediction model is trained according to a plurality of first building structure data samples and second building structure data samples obtained by transforming the feature space of the first building structure data samples; A training module, configured to obtain a plurality of first building structure data samples before inputting the target building structure data into the pre-trained target prediction model; using a transformation iteration learning sub-instruction to perform feature space transformation on the first building structure data samples to obtain second building structure data samples after transforming the feature space, where the feature space transformation includes dynamically adjusting the feature space according to the type and complexity of the building structure data; generating a training cost result when generating a model iteration instruction for predicting the building structure quality risk corresponding to the preset model according to the first building structure data samples and the second building structure data samples; adjusting the weight parameters of the preset model according to the training cost result, and training the adjusted preset model until a preset training stop condition is met to obtain a trained target prediction model; The generating a training cost result when generating a model iteration instruction for predicting the building structure quality risk corresponding to the preset model according to the first building structure data samples and the second building structure data samples includes: performing feature space inverse transformation on the second building structure data samples after transforming the feature space to obtain third building structure data samples; determining the deviation between the first building structure data samples and the third building structure data samples; in the case where the deviation is greater than or equal to a preset threshold, determining adjustment information of a transformation algorithm corresponding to the transformation iteration learning sub-instruction according to the deviation; optimizing the transformation iteration learning sub-instruction through the adjustment information; generating a training cost result according to the deviation and the adjustment information.

6. An electronic device, characterized in that, The device includes: a processor and a memory storing computer program instructions; when the processor calls the computer program instructions, the building structure quality risk prediction method according to any one of claims 1-4 is implemented.

7. A computer-readable storage medium, characterized in that, Computer program instructions are stored on the computer-readable storage medium, and when the computer program instructions are called by a processor, the building structure quality risk prediction method described in any one of claims 1-4 is implemented.

8. A computer program product, characterized in that, When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device is caused to execute the building structure quality risk prediction method described in any one of claims 1-4.

Images