Dynamic registration method, system and equipment for capital construction digital twin based on deep neural network, medium and product

Abstract

The invention discloses an infrastructure digital twinborn dynamic registration method, system and device based on a deep neural network, a medium and a product, and relates to the field of digital twinborn registration, and the method comprises the steps: building a digital twinborn model based on the project design of an infrastructure project; predicting growth data of the digital twin model; based on the predicted growth data, using the trained deep neural network to predict a compensation error; and compensating the digital twin model based on the compensation error to realize dynamic registration of the digital twin model. According to the method, the error between the digital twinborn model and the physical entity can be reduced, and the fitting precision of the digital twinborn model and the actual construction state is improved.

Description

Technical Field

[0001] The present application relates to the field of digital twin registration, and in particular to a method, system, device, medium and product for dynamic registration of infrastructure digital twins based on deep neural networks. Background Art

[0002] Early digital twin registration technologies mainly relied on static registration or geometric matching methods, especially in the field of infrastructure. These methods often used geometric models (such as CAD or BIM models) generated in the initial design phase to build digital twins. Through these models, engineers can simulate the design, construction, and later operation of buildings and infrastructure to a certain extent.

[0003] Traditional static registration methods have obvious limitations in practical applications. First, such methods lack the ability to cope with dynamic changes. During the infrastructure construction process, fluctuations in construction progress, material properties, and environmental factors can cause deviations between the physical structure and the design model. Since static registration relies only on the initial model, it cannot capture changes that occur during construction in real time, nor can it dynamically update the model. As a result, as the construction progresses, the errors between the physical object and the digital twin will continue to accumulate, eventually leading to the failure of the monitoring and early warning functions of the twin model. In addition, traditional methods react slowly to long-term changing factors such as equipment aging, wear and tear during use, and external shocks, resulting in the inability to meet the needs of high-precision dynamic monitoring during long-term operation and maintenance.

[0004] Geometric matching methods mostly rely on geometric features in the design for comparison, while ignoring the influence of non-geometric factors such as material properties and construction errors. During the construction process, changes in the mechanical properties of building materials and local structural deviations introduced by construction errors are all things that geometric models cannot cover. In addition, under changing external environments, such as extreme weather, vibration, geological changes and other factors, the actual performance of the physical structure is often significantly different from the design model. Since traditional geometric matching methods cannot effectively model these complex nonlinear factors, their application effect in complex scenarios is poor. Summary of the invention

[0005] The purpose of this application is to provide a method, system, equipment, medium and product for dynamic registration of infrastructure digital twins based on deep neural networks, which can reduce the error between the digital twin model and the physical entity through regular data collection and error compensation, and improve the fitting accuracy of the digital twin model and the actual construction status.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] In a first aspect, the present application provides a method for dynamic registration of infrastructure digital twins based on a deep neural network, comprising:

[0008] Build digital twin models based on project design of infrastructure projects;

[0009] Predicting growth data of the digital twin model;

[0010] Based on the predicted growth data, the trained deep neural network is used to predict the compensation error;

[0011] The digital twin model is compensated based on the compensation error to achieve dynamic alignment of the digital twin model.

[0012] In the second aspect, the present application provides an infrastructure digital twin dynamic registration system based on a deep neural network, comprising:

[0013] Digital twin model construction module, used to build digital twin models based on project design of infrastructure projects;

[0014] A growth data prediction module, used to predict the growth data of the digital twin model;

[0015] A compensation error prediction module is used to predict the compensation error based on the predicted growth data using a trained deep neural network;

[0016] A dynamic registration module is used to compensate the digital twin model based on the compensation error to achieve dynamic registration of the digital twin model.

[0017] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned deep neural network-based dynamic registration method for infrastructure digital twins.

[0018] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned deep neural network-based infrastructure digital twin dynamic registration method.

[0019] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned deep neural network-based infrastructure digital twin dynamic registration method.

[0020] According to the specific embodiments provided in this application, this application has the following technical effects:

[0021] The present application provides a method, system, device, medium and product for dynamic alignment of infrastructure digital twins based on deep neural networks. The compensation error of the growth data of the digital twin model is predicted by a trained deep neural network, and the digital twin model is compensated by the predicted compensation error, ensuring that the digital twin model can maintain a high degree of consistency with the actual physical object for a long time. This dynamic adjustment greatly improves the accuracy and response speed of the digital twin model, enabling it to adapt to complex and changeable construction and operation environments, making decision-making, monitoring and management during the construction process more intelligent. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 A schematic diagram of a flow chart of a method for dynamic registration of infrastructure digital twins based on a deep neural network provided in one embodiment of the present application;

[0024] Figure 2 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0026] In the field of infrastructure digital twins, traditional digital twin models are usually built based on rough models in the early stages of design, which often reflect the initial state of the project. However, as the project progresses and the environment changes dynamically, the deviation between the digital twin model and the actual physical object will gradually increase. This deviation may come from changes in material properties, uncertainties in the construction process, the influence of environmental conditions, etc., which in turn causes the digital twin model to lose its accurate representation of the physical object. If this error cannot be compensated in a timely manner, it will directly affect the effectiveness of the digital twin in real-time monitoring, fault warning and decision support. Especially in large-scale infrastructure projects, accurate model updates and error correction are particularly important.

[0027] When building a digital twin model, data collection is not done in real time, but rather at regular intervals to obtain actual physical object data. This discontinuous data collection method means that there may be a gradual deviation between the self-growing data in the digital twin model (i.e., the predicted data that the model self-derives over time and the environment) and the actual state of the physical object. These differences are inevitable because the model relies on the initial prediction algorithm to update in the gaps between discontinuous data collection. However, it is these differences that provide a key source of training data for the dynamic alignment of the neural network.

[0028] In order to cope with the deviation caused by this discontinuous acquisition, this application proposes to train the neural network through actual data collected regularly to improve the accuracy of the digital twin model. After each acquisition cycle, the actual collected data is compared and analyzed with the self-growing data of the digital twin. Through this process, the prediction errors of the model at different time points can be identified, and these errors can be used to perform back propagation training on the neural network, optimize the model parameters, and make the digital twin more accurately fit the actual infrastructure process.

[0029] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0030] In an exemplary embodiment, Figure 1 As shown, a method for dynamic alignment of infrastructure digital twins based on deep neural networks is provided. The method is executed by a computer device, and can be specifically executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In an embodiment of the present application, the method is applied to a server as an example for illustration, including the following steps 201 to 208.

[0031] in:

[0032] S1: Build a digital twin model based on the project design of infrastructure projects.

[0033] S2: Predicting growth data of the digital twin model.

[0034] The digital twin model usually self-grows or self-updates according to the initial conditions of the project design (such as the geometric model of the building, material properties, construction plan, etc.). The growth data of the digital twin model is predicted by the following formula:

[0035]

[0036] in, is the predicted growth data (such as stress, vibration, temperature, etc. of the bridge), xinit is the initial condition of the project design, t is the time, and fpredict (·) is a function of predicted growth data, which is used to estimate the expected state of the digital twin model at a certain point in the future. However, the predicted data will deviate over time, so dynamic alignment is performed.

[0037] S3: Based on the predicted growth data, the trained deep neural network is used to predict the compensation error.

[0038] The training process of a deep neural network specifically includes:

[0039] (1) Constructing a training sample set; the training samples include historical physical entity data, growth data of historical digital twin models, and errors between historical physical entity data and growth data of historical digital twin models.

[0040] The construction site of the infrastructure project is regularly photographed by drones to obtain high-resolution drone images. Drone images include the panoramic view of the infrastructure project site, the location and details of the structures. The construction management system collects text data in real time, including construction progress, construction material status, meteorological data, construction parameters, design standards and environmental changes.

[0041] The collected data (drone images and text data) are cleaned, formatted, and standardized to ensure data consistency. This includes image denoising, image enhancement, and text data cleaning to improve data quality and usability. The collected data is in the form of a time series, which is used to reflect the state of a physical entity at a specific point in time.

[0042] Drone images are acquired by the drone’s visual camera, and text data are collected through sensors, manual input, or other automated devices, and then uploaded to the construction management system.

[0043] D img (t) = {I 1 (t),I 2 (t),...,I n (t)}

[0044] Where D img (t) is the set of drone images collected at time t, I i (t) is the i-th drone image.

[0045] D text (t) = {T 1 (t),T 2 (t),...,T m (t)}

[0046] Where D text (t) is the text data at time t, Tj (t) is the text record related to construction in Article j.

[0047] Historical physical entity data is obtained through image dataset D img (t) and text dataset D text (t) obtained.

[0048] 1. Image dataset processing: Use the object detection model to locate the key components in the construction scene, use image segmentation technology to extract the boundaries of the construction components, extract key features as part of the physical data, and derive the geometric parameters of the construction components from the detection results.

[0049] 2. Text dataset processing: Use natural language processing technology (NLP) to extract structure-related information from text data.

[0050] Then combine the text information with the image data, align the image data and text data with timestamps or geographic coordinates, fuse the information, and form physical entity data y real .

[0051] Take the construction of bridge piers in bridge construction as an example:

[0052] Use the target detection model (automatically identify key construction components in the image (such as bridge piers, bridge main beams, steel cables, etc.), extract the position coordinates of the components in the image, use image segmentation technology to extract the contours of key components, and calculate geometric parameters based on the segmentation results: bridge pier height, main beam width, length, surface features (such as crack area, corrosion area); use natural language processing (NLP) technology to extract structured information such as construction progress, material properties, and environmental impact of bridge components from text, and analyze the impact of material properties and environment on construction components; use timestamps to match image data and text data to ensure that the image and text data describe the construction status at the same time point, use geographic coordinates or object identifiers (such as "bridge pier number") to align to ensure that the same object is described in the image and text data, and combine image and text data to generate complete physical entity data. After fusion, physical entity data containing comprehensive information such as geometry (bridge pier location, height, width, main beam length, assembly status), materials (such as concrete strength, steel beam toughness), and environment (temperature and humidity, rainfall, and the impact of the environment on construction progress) is formed.

[0053] Since the prediction frequency of the digital twin model is usually high, but the actual collected data is discrete, it is necessary to ensure that the data is aligned in time. Through the aligned data, the difference between the physical entity data and the growth data of the digital twin model is calculated to obtain the error Δx(t):

[0054]

[0055] Among them, y real (t) is the physical entity data.

[0056] (2) The principal component analysis method is used to reduce the dimensionality of the error between the historical physical entity data and the historical digital twin model data.

[0057] In infrastructure projects, the physical entity data collected is usually high-dimensional and contains a variety of information (such as stress, vibration, temperature, etc.), and there may be correlations between these information, so the error data obtained is also high-dimensional. In order to reduce the dimensionality of the data and extract the most important features, principal component analysis (PCA) can be used. PCA projects high-dimensional data into a low-dimensional space through linear transformation and extracts key factors from it. These key factors can be used for training deep neural networks. The following is a detailed process of feature extraction of error data using PCA.

[0058] Assume that physical entity data at m time points are collected, and each time point contains n variables (such as stress, vibration, temperature, etc.). The errors between these variables and the growth data of the historical digital twin model constitute an error data matrix △X, whose dimension is m×n.

[0059]

[0060] Among them, x ij is the error data of the jth variable obtained at the ith time point. m is the number of time points collected, and n is the number of variables at each time point.

[0061] Before using PCA, it is usually necessary to standardize the data so that each variable has a mean of 0 and a variance of 1. This is because different variables may have different dimensions (for example, stress and temperature have different units), and standardization can prevent a variable from having an excessively large dimension and having an unnecessary impact on the PCA results.

[0062] Standardization: Scaling the data to have zero mean and unit variance:

[0063]

[0064] Among them, x is the original data, μ is the mean, and σ is the standard deviation.

[0065] Normalization: Scale the data to the interval [0,1]:

[0066]

[0067] Among them, min(x) and max(x) are the minimum and maximum values ​​of the data respectively.

[0068] The core idea of ​​principal component analysis is to map high-dimensional data to low-dimensional space by finding the principal components in the data. The principal components are the directions with the largest variance in the data, and through these directions, the main sources of variation in the data can be found.

[0069] Specific steps:

[0070] Step 1: Calculate the covariance matrix: The covariance matrix is ​​used to describe the correlation between variables in the data. For the standardized data matrix ΔX', the calculation formula of the covariance matrix C is:

[0071]

[0072] Among them, ΔX' is the standardized data matrix, X' T is the transposed matrix, and C is the n×n covariance matrix.

[0073] Step 2: Calculate eigenvalues ​​and eigenvectors: Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalue λ 1 ,λ 2 , ..., λ n and the corresponding eigenvector v 1 , v 2 , ..., v n The larger the eigenvalue, the larger the variance of the data in the direction corresponding to the eigenvector, that is, the more important the change in that direction.

[0074] Step 3: Select principal components: Sort by eigenvalues ​​and select the first k eigenvectors as principal components for constructing low-dimensional data. Generally, eigenvectors whose cumulative explained variance reaches a certain threshold (such as 90%) are selected. The formula for cumulative explained variance is:

[0075]

[0076] Principal component transformation: Matrix V consisting of k selected eigenvectors k Perform a linear transformation on ΔX' to obtain the reduced-dimensional data matrix X PCA :

[0077] X PCA =ΔX'*V k .

[0078] The data matrix X after dimensionality reduction PCAContains the most important k principal components in infrastructure projects, which represent the most important sources of change in the data. Each principal component may correspond to one or more physical meanings, for example: the first principal component may reflect the dominant factor of stress change, and the second principal component may be related to periodic fluctuations in temperature and humidity. By analyzing these principal components, engineers can understand which factors dominate the project and use these factors to optimize construction plans or perform maintenance.

[0079] (3) Input the training sample set into the deep neural network and calculate the loss function.

[0080] Use the reduced dimension data X PCA To train a deep neural network. Since PCA has extracted the most important features, the input data dimension of the deep neural network is greatly reduced, thereby reducing the computational complexity and avoiding the risk of overfitting. Design a deep neural network architecture, including input layers, hidden layers, and output layers, to process multi-source heterogeneous data.

[0081] Deep neural network architecture: The input layer receives the fused feature vector, which contains image, sensor and text features. The design of the hidden layer realizes deep processing of infrastructure data through fully connected layers and activation functions. The neurons in each layer learn the state changes and potential errors of the infrastructure objects through the feature relationship of the input data. The infrastructure data (such as the shape of the building in the image, stress sensor data) is processed by multiple layers of neurons, and each layer can be understood as the complex dynamic relationship modeling of the infrastructure structure, such as the deformation of the building and the stress-strain relationship of the material. 3-4 hidden layers are used, each containing 128, 256 or 512 neurons, which are responsible for processing the complex relationship between the structural changes of the infrastructure objects and environmental factors. For example, sensor data can reflect the changes in the stress distribution of a bridge or building during construction. The ReLU activation function is used to process nonlinear features and capture complex error patterns in the infrastructure process:

[0082] hi=RELU(W i *x+b i )

[0083] Among them, W i and b i are the weight and bias of the i-th layer, respectively, and x is the input feature. The compensation value generated by the output layer is used to adjust the parameters of the twin model to better reflect the actual state of the infrastructure object. The output compensation error can be used to correct the stress, displacement or temperature and humidity simulation data in the twin.

[0084] In infrastructure scenarios, the difference between sensor data (such as stress, displacement, temperature, etc.) and the growth data of the twin model is usually measured by mean square error, which is suitable for numerical compensation tasks. The loss function is defined as follows:

[0085]

[0086] Among them, y i Indicates the actual value of the error, y i ′ represents the output of the deep neural network.

[0087] (4) Based on the loss function, the GPSO algorithm is used to update the parameters of the deep neural network to complete the training of the deep neural network. The specific process is as follows:

[0088] ①Particle initialization.

[0089] Initialize particle swarm P = {p 1 ,p 2 ,···,p m}, where each particle represents a set of parameters (weights and biases) of a deep neural network, and the particle velocity v is initialized i is a random value.

[0090] ②Calculate fitness.

[0091] For each particle pi, update its corresponding weight and bias value to the deep neural network, calculate the output of the network, calculate the fitness value of the current particle according to the loss function, and update the individual optimal value p i best and the global optimal value p g best .

[0092] ③Update particle position and velocity.

[0093] The velocity of each particle is updated according to the following formula:

[0094]

[0095] Among them, w is the inertia weight, which controls the particle's dependence on the current speed, c 1 and c 2 is the learning factor, which controls the particle's dependence on individual optimality and global optimality, r 1 and r 2 is a random number in the range of [0,1].

[0096] The particle's position p i Update according to the following formula:

[0097] p i =pi +v i .

[0099] ④Neural network training.

[0100] The updated particle p i As the weights and bias values ​​of the deep neural network, the updated output and loss values ​​are recalculated, and the fitness evaluation and particle update process are repeated until the loss function converges or the maximum number of iterations is reached.

[0101] The network weights are adjusted through the back-propagation algorithm, and the GPSO algorithm is used to update the network weights and biases according to the fitness of the particles to minimize the prediction error. Cross-validation, regularization and other technologies are used to prevent overfitting and improve the generalization ability of the model.

[0102] The deep neural network provided in this application learns the deviation pattern between the growth data of the digital twin model and the physical entity data under different conditions by training historical error data. Its core task is to use the trained deep neural network to predict the compensation error when the real-time data has not yet been collected, so as to correct the digital twin model.

[0103] S4: Compensating the digital twin model based on the compensation error to achieve dynamic alignment of the digital twin model.

[0104] The compensation error between the growth data of the digital twin model and the physical entity data is predicted according to the trained deep neural network, and the growth data of the digital twin model is adjusted according to the compensation error.

[0105]

[0106] Where Δx i is the predicted compensation error, It is the growth data of the digital twin model after compensation.

[0107] During the period of non-continuous data collection, the growth data of the digital twin model is continuously input into the deep neural network, and the deep neural network generates a series of compensation errors based on the growth data at different time points. This process is carried out continuously to ensure that the digital twin model can still maintain a high level of accuracy without the input of new field data.

[0108] Through the compensation errors generated by the deep neural network, the digital twin model will continue to adjust during the non-continuous acquisition period, which can reduce the cumulative error of the model. When the next field data collection arrives, the error between the digital twin model and the physical entity data will be smaller, which can ensure that the digital twin model does not need to be significantly revised each time new data is input. For example, during the construction of a large bridge, the digital twin model mainly relies on stress and vibration data to monitor the health of the structure. Since the field data is collected regularly, the system will rely on the prediction algorithm based on the initial geometric model in the gap between data collection. However, the actual stress and vibration state of the bridge may be affected by factors such as wind speed, temperature, and traffic load, resulting in errors between the state of the physical object and the predicted value. During this gap, the deep neural network will identify these potential errors based on the previously trained pattern and generate corresponding compensation errors. By compensating for the errors, the system can continue to adjust the digital twin model before the next data collection arrives to ensure that it maintains a high fitting accuracy with the actual bridge state.

[0109] As construction progresses, the digital twin model will continue to receive new physical entity data, and these error data will be saved to form a historical error data set:

[0110]

[0111] After each new data collection, the deep neural network will use the latest error data and historical data to retrain to optimize its parameters:

[0112]

[0113] Among them, θ is the optimized parameter of the deep neural network after retraining, and f NN is a deep neural network, x init is the initial condition of the digital twin model, is the new compensation error.

[0114] As more and more historical data is accumulated and new data is continuously collected and used, deep neural networks can learn more complex error patterns and better identify various complex factors that affect the accuracy of digital twins (such as material fatigue, temperature fluctuations, subtle differences in the construction process, etc.). These new patterns will be integrated into the deep neural network model to generate more accurate compensation errors.

[0115] The infrastructure digital twin dynamic registration method based on deep neural network provided in this application has the following advantages:

[0116] High-precision registration: Through regular data collection and error compensation, the error between the digital twin model and the physical entity can be significantly reduced, and the fitting accuracy between the model and the actual construction status can be improved.

[0117] Intelligent early warning: Use digital twin models to simulate the construction process, identify potential problems in a timely manner and issue early warnings, thereby improving construction safety and reliability.

[0118] Construction optimization: Provide data-driven construction optimization suggestions to improve construction efficiency and quality and reduce costs. Through error compensation, it can provide optimization references for the construction process, such as: discovering resource waste that may be caused by construction deviations in advance; providing accurate status feedback at critical stages to avoid potential construction defects. The main sources of error extracted by principal component analysis can help identify bottlenecks in the construction process.

[0119] Progressive optimization and adaptive capability: Continuous optimization through historical data accumulation ensures its long-term accurate fitting and robustness in complex construction environments, and enhances its adaptive capability to environmental changes.

[0120] Reduce the frequency of data collection: By predicting and compensating errors, dynamic calibration can still be performed during non-continuous data collection, and the collection frequency can be appropriately reduced to reduce costs.

[0121] Based on the same inventive concept, the embodiment of the present application also provides a system for implementing the above-mentioned method for dynamic registration of infrastructure digital twins based on deep neural networks. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more embodiments of the dynamic registration system for infrastructure digital twins based on deep neural networks provided below can be referred to the limitations of the dynamic registration method for infrastructure digital twins based on deep neural networks in the above text, and will not be repeated here.

[0122] In an exemplary embodiment, a deep neural network-based infrastructure digital twin dynamic registration system is provided, comprising:

[0123] The digital twin model construction module is used to construct a digital twin model based on the project design of infrastructure projects.

[0124] A growth data prediction module is used to predict the growth data of the digital twin model.

[0125] The compensation error prediction module is used to predict the compensation error based on the predicted growth data using a trained deep neural network.

[0126] A dynamic registration module is used to compensate the digital twin model based on the compensation error to achieve dynamic registration of the digital twin model.

[0127] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program. The computer device may be a server or a terminal, and its internal structure diagram may be as shown in FIG. Figure 2 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data to be processed. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a dynamic alignment method of infrastructure digital twins based on a deep neural network is implemented.

[0128] Those skilled in the art will understand that Figure 2 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0129] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0130] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0131] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0132] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0133] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0134] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0135] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A dynamic registration method for infrastructure digital twins based on deep neural networks, characterized in that: include: Build digital twin models based on project design of infrastructure projects; Predicting growth data of the digital twin model; Based on the predicted growth data, the trained deep neural network is used to predict the compensation error; The digital twin model is compensated based on the compensation error to achieve dynamic alignment of the digital twin model.

2. The method for dynamic registration of infrastructure digital twins based on deep neural network according to claim 1, characterized in that: Using formula Calculate growth data for digital twin models Among them, x init is the initial condition of the project design, t is the time, f predict (·) is the function for predicting growth data.

3. The method for dynamic registration of infrastructure digital twins based on deep neural network according to claim 1, characterized in that: The training process of a deep neural network specifically includes: Constructing a training sample set; the training sample includes historical physical entity data, growth data of the historical digital twin model, and the error between the historical physical entity data and the growth data of the historical digital twin model; Inputting the training sample set into a deep neural network and calculating a loss function; Based on the loss function, the GPSO algorithm is used to update the parameters of the deep neural network to complete the training of the deep neural network.

4. The method for dynamic registration of infrastructure digital twins based on deep neural network according to claim 1, characterized in that: The acquisition process of the historical physical entity data is as follows: The construction site of the infrastructure project is photographed regularly by drones to obtain drone-photographed images; Obtain text data from construction sites of infrastructure projects; The historical physical entity data of the infrastructure project is obtained based on the images taken by the drone and the text data.

5. The method for dynamic registration of infrastructure digital twins based on deep neural network according to claim 4 is characterized in that: After constructing the training sample set, it also includes: The principal component analysis method is used to reduce the dimensionality of the error between the historical physical entity data and the historical digital twin model data.

6. A dynamic registration system for infrastructure digital twins based on deep neural networks, characterized in that: include: Digital twin model construction module, used to build digital twin models based on project design of infrastructure projects; A growth data prediction module, used to predict the growth data of the digital twin model; A compensation error prediction module is used to predict the compensation error based on the predicted growth data using a trained deep neural network; A dynamic registration module is used to compensate the digital twin model based on the compensation error to achieve dynamic registration of the digital twin model.

7. The infrastructure digital twin dynamic registration system based on deep neural network according to claim 6 is characterized in that: It also includes a model training module; the model training module includes: A training sample set construction unit, used to construct a training sample set; the training sample includes historical physical entity data, growth data of the historical digital twin model, and an error between the historical physical entity data and the growth data of the historical digital twin model; A loss function calculation unit, used to input the training sample set into the deep neural network and calculate the loss function; A parameter updating unit is used to update the parameters of the deep neural network based on the loss function and adopt the GPSO algorithm to complete the training of the deep neural network.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the deep neural network-based dynamic registration method for infrastructure digital twins as described in any one of claims 1 to 5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the infrastructure digital twin dynamic registration method based on deep neural network as described in any one of claims 1 to 5.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the infrastructure digital twin dynamic registration method based on deep neural network as described in any one of claims 1 to 5.

Images