Rural infrastructure intelligent operation and maintenance method and system based on digital twinning

Abstract

The application provides a kind of rural infrastructure intelligent operation and maintenance method and system based on digital twinning, it is related to intelligent operation and maintenance technical field, wherein, the method comprises: obtaining the time series infrared data, visible light image and environmental parameter of target area;Infrared data is calibrated to obtain initial temperature field data, and is fused with visible light image to generate fusion image data;Then the fusion data is input into the generative adversarial network model to extract thermal anomaly features, and the digital twin is updated synchronously;Then the environmental parameter is input into the updated digital twin, and thermodynamic simulation is carried out, and simulation temperature field data is output;Then the simulation data and the fusion image are registered and difference calculated, and the abnormal heat zone is determined;Finally, the shape of the heat zone, the time domain change and the temperature gradient are analyzed, the fault type is determined combined with the facility type, and the operation and maintenance work order is automatically generated and sent.The application improves the intelligent diagnosis and active operation and maintenance level of rural concealed facility latent fault.

Description

Technical Field

[0001] This application relates to the field of intelligent operation and maintenance technology, and in particular to an intelligent operation and maintenance method and system for rural infrastructure based on digital twins. Background Technology

[0002] Intelligent operation and maintenance technology for rural infrastructure aims to improve the management efficiency and reliability of facilities through digital means, and has broad application prospects. This technology integrates multiple methods such as the Internet of Things, data analysis, and simulation modeling, providing new possibilities for the long-term health monitoring and maintenance of rural infrastructure.

[0003] Existing digital twin-based operation and maintenance methods typically rely on pre-defined static models or rules to monitor facility status. Their data sources are mainly conventional sensors, and they primarily focus on reactive responses to visible faults. When building digital models, these methods often use historical experience data for initialization, and then update model parameters through periodic data collection to simulate and evaluate the facility's operational status.

[0004] Existing methods, when dealing with the early and subtle anomaly detection of concealed facilities such as buried pipelines and cable trenches, suffer from insufficient sensitivity and accuracy in identifying latent faults due to the limited dimensionality of the sensor data they rely on and the limited coupling depth between the model update mechanism and real-time physical field changes. Therefore, existing technologies face the technical challenge of improving the diagnostic accuracy for early latent faults in rural concealed facilities. Summary of the Invention

[0005] This application provides a digital twin-based intelligent operation and maintenance method and system for rural infrastructure, which aims to solve the problem of low level of intelligent diagnosis and proactive operation and maintenance of hidden faults in rural facilities in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a method for intelligent operation and maintenance of rural infrastructure based on digital twins, comprising:

[0007] Acquire time-series infrared data, visible light image data, and environmental parameter data for areas where rural infrastructure is located;

[0008] Temperature calibration and radiation correction are performed on the time-series infrared data to obtain initial temperature field data, and the initial temperature field data is fused with the visible light image data to generate fused image data;

[0009] The fused image data is input into a generative adversarial network model to identify and extract thermal anomaly features of rural infrastructure, and the thermal anomaly features are used as state update parameters to be synchronized to the corresponding digital twin.

[0010] The environmental parameter data is input into the updated digital twin, which drives the physical information neural network integrated in the updated digital twin to perform thermodynamic simulation and outputs the simulated temperature field data of the rural infrastructure surface.

[0011] Spatial registration and pixel-by-pixel difference calculation are performed on the simulated temperature field data and the fused image data to obtain temperature difference data. Adaptive threshold segmentation is then performed on the temperature difference data to determine abnormal hot areas.

[0012] The morphological characteristics, temporal variation trend, and temperature gradient with the surrounding area of ​​the abnormal hot zone are analyzed. Combined with the type of rural infrastructure, the final fault type is determined. Based on the final fault type and the geographical location of the abnormal hot zone, an operation and maintenance work order is automatically generated and sent to the operation and maintenance terminal.

[0013] Secondly, this application provides a smart operation and maintenance system for rural infrastructure based on digital twins, comprising:

[0014] The acquisition module is used to acquire time-series infrared data, visible light image data, and environmental parameter data collected in areas where rural infrastructure is located.

[0015] The correction module is used to perform temperature calibration and radiation correction on the time-series infrared data to obtain initial temperature field data, and to fuse the initial temperature field data with the visible light image data to generate fused image data.

[0016] The identification module is used to input the fused image data into the generative adversarial network model, identify and extract the thermal anomaly features of rural infrastructure, and synchronize the thermal anomaly features as state update parameters to the corresponding digital twin.

[0017] The simulation module is used to input the environmental parameter data into the updated digital twin, drive the physical information neural network integrated in the updated digital twin to perform thermodynamic simulation, and output the simulated temperature field data of the rural infrastructure surface.

[0018] The calculation module is used to perform spatial registration and pixel-by-pixel difference calculation on the simulated temperature field data and the fused image data to obtain temperature difference data, and to perform adaptive threshold segmentation on the temperature difference data to determine abnormal hot areas.

[0019] The analysis module is used to analyze the morphological characteristics, temporal variation trend, and temperature gradient with the surrounding area of ​​the abnormal hot zone. Combined with the type of rural infrastructure, it determines the final fault type and automatically generates an operation and maintenance work order based on the final fault type and the geographical location of the abnormal hot zone, and sends it to the operation and maintenance terminal.

[0020] Thirdly, this application provides an electronic device, comprising:

[0021] Memory, used to store computer programs;

[0022] A processor, used to execute the computer program to implement the steps of the intelligent operation and maintenance method for rural infrastructure based on digital twins as described in the first aspect above.

[0023] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the intelligent operation and maintenance method for rural infrastructure based on digital twins as described in the first aspect above.

[0024] The technical solution provided in this application has the following beneficial effects:

[0025] This application also enables the construction of a comprehensive information view including temperature and texture by acquiring and fusing multi-source data, thereby providing a more comprehensive data foundation for subsequent analysis. Then, by extracting thermal anomaly features through generative adversarial networks and synchronizing them to the digital twin, the state of the digital twin can reflect the dynamic changes of the on-site facilities in real time, improving the consistency between the model and the physical world. Next, by using the updated digital twin in combination with a physical information neural network for simulation, a theoretical temperature field that is more in line with the current environmental conditions can be generated, providing a reliable comparison benchmark for anomaly detection.

[0026] Subsequently, by calculating the difference between simulation data and measured data and segmenting the threshold, subtle, initial temperature anomaly areas can be effectively identified. Finally, by conducting in-depth analysis of the anomaly areas and combining them with facility types, the fault type can be accurately determined and maintenance work orders can be automatically generated, thereby directly converting the detection results into executable maintenance instructions and improving maintenance response efficiency.

[0027] Furthermore, by embedding physical laws and evaluating the uncertainty of local heat exchange during the simulation process, this application ensures that the generated simulated temperature field data not only follows basic thermodynamic principles but also considers the randomness of material properties and environmental fluctuations in actual working conditions. This improves the reliability and accuracy of the simulation results in complex real-world environments and provides a more solid and valuable theoretical basis for subsequent comparisons.

[0028] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 A flowchart illustrating a digital twin-based intelligent operation and maintenance method for rural infrastructure, provided as an embodiment of this application;

[0031] Figure 2 A schematic diagram illustrating a specific implementation of a digital twin-based intelligent operation and maintenance method for rural infrastructure, as provided in this application embodiment;

[0032] Figure 3 This is a schematic diagram of the structure of a smart operation and maintenance system for rural infrastructure based on digital twins, provided as an embodiment of this application. Detailed Implementation

[0033] To address the problems existing in the prior art, this application proposes a smart operation and maintenance method for rural infrastructure based on digital twins. The core idea of ​​this method is to enable the digital twin to keenly capture and accurately characterize the early thermal anomalies and their evolution of hidden facilities by strengthening data fusion, dynamic driving and physical law constraints. This effectively solves the problem of insufficient accuracy and timeliness in the diagnosis of early hidden faults in the prior art, and improves the initiative and accuracy of operation and maintenance.

[0034] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] The core of this application is to provide a smart operation and maintenance method for rural infrastructure based on digital twins, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0036] Step 101: Acquire time-series infrared data, visible light image data, and environmental parameter data for the areas where rural infrastructure is located.

[0037] In step 101, rural infrastructure refers to physical facilities in the rural environment that require operation and maintenance, such as buried water pipes, heating pipes or power cable trenches, and time-series infrared data refers to measurement data that reflects the temperature changes of surface and shallow facilities, which are continuously collected by infrared sensors in time sequence.

[0038] Visible light image data refers to optical image data taken in the visible light band that reflects the visual texture of the earth's surface. Environmental parameter data includes external environmental variables that may affect the heat exchange process of the facility, such as air temperature, soil moisture, and solar radiation intensity.

[0039] In this embodiment, an infrared sensor array and a visible light camera are first deployed in the area where rural infrastructure is located, such as under roads or inside utility tunnels, to collect data periodically. The infrared sensor array is used to acquire time-series infrared data, and the visible light camera is used to acquire visible light image data. Simultaneously, real-time environmental parameter data is acquired through an environmental monitoring station. The above collection process is carried out synchronously, thereby obtaining a multi-source data set that is correlated in time.

[0040] Step 102: Perform temperature calibration and radiation correction on the time-series infrared data to obtain initial temperature field data, and fuse the initial temperature field data with the visible light image data to generate fused image data.

[0041] In step 102, temperature calibration refers to the process of converting the raw signal value output by the infrared sensor into a temperature value with physical meaning based on the characteristics of the infrared sensor; radiation correction refers to the compensation processing of the temperature value to eliminate measurement errors caused by inconsistent sensor response and environmental radiation interference such as atmospheric and ground background.

[0042] Initial temperature field data refers to two-dimensional distribution data that reflects the accurate temperature values ​​of each point in the measurement area after temperature calibration and radiation correction; fused image data refers to a single image formed by combining initial temperature field data with visible light image data, which contains both temperature information and visual texture information.

[0043] In this embodiment, the raw infrared signal of each frame in the time-series infrared data is first calibrated for temperature and corrected for radiometrics to obtain initial temperature field data that accurately reflects the temperature distribution of the surface and shallow facilities. Then, the visible light image data is spatially scaled to align with the resolution of the initial temperature field data. Finally, the adjusted visible light image data and the initial temperature field data are combined using a multi-scale information fusion method, superimposing temperature information while preserving visible light texture details, thereby generating fused image data that simultaneously includes heat distribution and surface texture features.

[0044] Step 103: Input the fused image data into the generative adversarial network model, identify and extract the thermal anomaly features of rural infrastructure, and synchronize the thermal anomaly features as state update parameters to the corresponding digital twin.

[0045] Among them, thermal anomaly features refer to comprehensive information extracted from fused image data that can characterize the deviation of rural infrastructure temperature distribution from the normal pattern. This information includes both spatial morphological attributes and temporal variation trends.

[0046] A digital twin is a digital mapping model of rural infrastructure in virtual space, used to simulate and reflect the real state of physical facilities; state update parameters are control instructions or sets of data used to drive the adjustment of internal state variables of the digital twin.

[0047] Furthermore, the structure of the generative adversarial network model consists of two core substructures: a generator and a discriminator. The generator adopts a convolutional neural network with an encoder-decoder structure. The encoder part consists of multiple downsampled convolutional layers to extract abstract features of the input noise, and the decoder part consists of multiple upsampled transposed convolutional layers to reconstruct and generate thermal images that simulate normal working conditions.

[0048] The discriminator specifically employs a design containing three sequentially connected feature extraction network layers: the first network layer is an initial convolutional layer used to extract low-level features; the second network layer is a deeper convolutional module with a larger receptive field, used to identify abnormal regions and extract their contours and statistical features by combining spatial context; the third network layer is a temporal feature extraction module, which can typically be designed with a structure containing a convolutional long short-term memory network or a three-dimensional convolution, used to correlate and analyze the current frame with historical frames to extract temperature change trends.

[0049] The training process of the model is still the alternating iterative training of generative adversarial training. During the training phase, the three network layers of the discriminator are optimized as a whole. The goal is to maximize the ability to distinguish between real normal images and images forged by the generator. The training goal of the generator is to make the generated images "deceive" the discriminator, causing each layer of the discriminator to misclassify them as real images.

[0050] It should be noted that the above structure is exemplary. This application does not impose specific limitations on the structural design of layers and other components used in the internal structure of the generative adversarial network model. The structure can be set according to the actual situation.

[0051] In this embodiment, step 103 includes the following process:

[0052] Step 1031: Input the fused image data into the discriminator of the generative adversarial network model, and extract low-level features from the fused image data through the first network layer of the discriminator.

[0053] In step 1031, the low-level features refer to the set of basic visual elements obtained from the fused image data through initial convolution, such as points, edges, and color blocks.

[0054] In this embodiment, the fused image data is first input into the discriminator of the generative adversarial network model. The first network layer of the discriminator is typically composed of convolutional kernels. These kernels perform sliding calculations on the image to capture the most basic texture and temperature distribution patterns in the image, forming low-level feature maps describing these basic patterns. This embodiment does not limit the size of the convolutional kernels and can set them according to actual conditions.

[0055] Step 1032: The discriminator processes the underlying features through its second network layer and, in conjunction with the spatial relationships of the fused image data, identifies abnormal hot regions and extracts the contour features and temperature statistical features of the abnormal hot regions as the first features.

[0056] In step 1032, the abnormal hot region refers to a set of continuous pixels identified in the underlying features whose temperature distribution pattern differs from the surrounding environment; the contour features refer to parameters describing the geometry of the outer boundary of the abnormal hot region, such as boundary length, area, and shape factor; the temperature statistical features refer to quantitative indicators obtained by statistically calculating the temperature values ​​of all pixels in the abnormal hot region, such as average temperature, maximum temperature, and temperature standard deviation.

[0057] In this embodiment, the second network layer of the discriminator performs deeper processing on the received low-level features. This layer has a larger receptive field and can combine the spatial proximity relationship between pixels to locate local regions with abnormal temperature patterns, i.e., abnormal hot regions, from the low-level features. Then, for each identified abnormal hot region, the geometric parameters of its outer boundary are calculated as contour features, and the temperature statistics of the pixels inside the region are calculated as temperature statistical features. Finally, the extracted contour features and temperature statistical features are combined to form the first feature.

[0058] Step 1033: By combining the fused image data of the current frame and historical frames through the third network layer of the discriminator, the temperature change trend of the abnormal hot area is analyzed as a second feature.

[0059] In step 1033, the second feature refers to the dynamic properties that describe the evolution of the abnormal thermal region over time, such as the rate of temperature change, the direction of diffusion, or the trend of area growth.

[0060] In this embodiment, the third network layer of the discriminator is designed to process time series information. The third network layer performs correlation analysis between the information of the abnormal hot area in the current frame image and the information of the same area in the previous consecutive historical images. By comparing the state at different time points, the temperature change rate and spatial evolution direction of the area are calculated, and these dynamic information are summarized as the second feature.

[0061] Step 1034: Combine the first feature with the second feature to generate thermal anomaly features.

[0062] Step 1035: Convert the thermal anomaly features into a parameter format recognizable by the digital twin to obtain the state update parameters.

[0063] In step 1035, the identifiable parameter format refers to the data form that matches the data structure, physical meaning, and numerical range of the thermodynamic state variables within the digital twin.

[0064] In this embodiment of the application, thermal anomaly features are fed into a parameter mapping network. This network has learned the correspondence between feature data and state variables of the digital twin. It can encode the feature data and convert it into a series of values. The physical meaning and dimension of these values ​​are consistent with the state variables such as the heat source intensity and thermal conductivity coefficient to be adjusted in the digital twin, thereby obtaining state update parameters.

[0065] Step 1036: Send the state update parameters to the digital twin, adjust the thermodynamic state parameters at the corresponding positions in the digital twin, and obtain the updated digital twin.

[0066] In this embodiment of the application, the state update parameters are transmitted to the digital twin system. The digital twin system modifies the thermodynamic state parameters such as heat source intensity and thermal conductivity coefficient at the corresponding geographical coordinates in its internal model according to the specified location and value in the parameters. After all the specified parameters are adjusted, an updated digital twin with the state synchronized with the latest field observation characteristics is obtained.

[0067] This application automatically extracts accurate spatiotemporal features of thermal anomalies from fused data through a generative adversarial network model, and efficiently transforms them into parameters that drive the dynamic updating of the digital twin, thereby realizing the real-time and accurate perception and status synchronization of the digital model for the hidden thermal anomalies of concealed facilities.

[0068] Step 104: Input the environmental parameter data into the updated digital twin, drive the physical information neural network integrated in the updated digital twin to perform thermodynamic simulation, and output the simulated temperature field data of the rural infrastructure surface.

[0069] Among them, the physical information neural network can adopt a hybrid structure that integrates traditional neural networks with physical equation solvers. Its core substructures include boundary condition encoding layer, spatial discretization layer, equation embedding layer, iterative solution layer and surface mapping layer.

[0070] The boundary condition encoding layer is a fully connected neural network, for example, containing two hidden layers, each with 64 neurons, using the ReLU activation function, responsible for mapping environmental parameters to boundary constraint vectors; the spatial discretization layer is essentially a parameterized mesh generator, which can adopt a structure based on graph neural networks or coordinate transformation networks. Its input is the facility structure parameters and boundary constraint vectors, and its output is the coordinates of finite element mesh nodes and the element connection relationship.

[0071] Furthermore, the equation embedding layer is not an independent layer, but rather a physical loss term introduced during the network's forward propagation. Its structural design incorporates the residual of the heat conduction partial differential equation as an additional loss function into the total loss, constraining the network output to satisfy the equation.

[0072] The iterative solution layer is designed as a differentiable solver module, such as a differentiable finite element solver or an iterative optimization layer simulated by a neural network. Its structure includes a trainable parameter matrix to simulate the stiffness matrix assembly and solution process. This layer receives mesh data and boundary conditions and simulates iterative solution through internal multi-step loop expansion operations until the convergence criterion is met. The surface mapping layer is a lightweight convolutional neural network or interpolation network, such as a three-layer deconvolutional network, which is responsible for interpolating and mapping the volume data to the target surface.

[0073] The training process of this model is a joint training of end-to-end supervision and physical constraints: First, a dataset containing the real temperature field and corresponding environmental parameters and facility structure is used. By inputting the environmental parameters and facility structure into the network, the predicted temperature field is obtained through forward propagation. The mean square error between the predicted value and the real value is calculated as the data loss.

[0074] Meanwhile, the physical residual of the temperature field predicted by the network in the equation embedding layer to satisfy the heat conduction equation on a given grid is used as the physical loss; the final total loss is the weighted sum of the data loss and the physical loss. All trainable parameters in the network are optimized through the backpropagation algorithm, including the parameters in the boundary condition encoding layer, the spatial discretization layer, the simulation matrix parameters in the iterative solution layer, and the weights of the surface mapping layer, so that the network can simultaneously learn to fit the real data and obey the physical laws.

[0075] It should be noted that the above structure is exemplary. This application does not impose specific limitations on the layer and other structural designs used in the internal structure of the physical information neural network, and corresponding settings can be made according to the actual situation.

[0076] Simulated temperature field data refers to the distribution data of temperature values ​​at various points on the surface of rural infrastructure obtained through simulation calculations.

[0077] In this embodiment, step 104 includes the following process, such as... Figure 2 As shown:

[0078] Step 1041: Input the environmental parameter data into the physical information neural network of the updated digital twin, process the environmental parameter data through the boundary condition encoding layer of the physical information neural network, and generate a boundary constraint vector that interacts with the environment.

[0079] In step 1041, the boundary condition encoding layer is a specific structural layer in the physical information neural network, used to convert environmental parameter data into boundary condition data required for model calculation; the boundary constraint vector refers to the set of constraints expressed in vector form to define the external environmental effects on the simulation model.

[0080] In this embodiment, environmental parameter data is first input into the physical information neural network integrated in the updated digital twin. The boundary condition encoding layer of the network receives this data. This layer is composed of a fully connected neural network, and its internal weights are trained and learned to map environmental parameters such as temperature and humidity into a series of values. These values ​​represent the constraints such as temperature and heat flow applied to the boundary of the simulation model, thereby generating a boundary constraint vector.

[0081] In practical applications, for example, the input environmental parameter data at a certain moment might include an air temperature of 25 degrees Celsius, a surface humidity of 0.3, and a solar radiation intensity of 800 watts per square meter. The boundary condition encoding layer calculates these parameters through its internal fully connected network and outputs a boundary constraint vector, which may contain values ​​used to define boundary conditions such as soil surface temperature and the heat exchange coefficient between soil and air.

[0082] Step 1042: Through the spatial discrete layer in the physical information neural network, meshing is performed based on the boundary constraint vector and facility structure parameters to obtain a discrete computational mesh. The finite element analysis method is then used to assign material properties and apply boundary conditions to the computational mesh to obtain a finite element mesh.

[0083] In step 1042, the spatial discrete layer is a structural layer in the physical information neural network that transforms the continuous spatial structure into discrete computing units, and the computing grid refers to the set formed by dividing the simulation region into numerous small geometric units;

[0084] Facility structure parameters are data describing the geometry, size, and spatial location of rural infrastructure. Finite element mesh refers to a computational mesh that has been processed by the finite element analysis method and includes material properties and boundary conditions.

[0085] In this embodiment of the application, the spatial discrete layer of the physical information neural network automatically divides the three-dimensional spatial region to be simulated into a large number of small tetrahedral or hexahedral units based on the digital geometric model of rural infrastructure, i.e., the facility structural parameters, to form an initial computational grid.

[0086] Then, using the finite element analysis method, each element in the computational mesh is assigned its corresponding material thermal property parameters, such as the thermal conductivity and specific heat capacity of soil. At the same time, the constraint conditions expressed by the boundary constraint vector generated in step 1041 are applied to the corresponding boundary element nodes in the computational mesh, thereby obtaining a complete finite element mesh containing all the necessary physical information.

[0087] In practical applications, taking a buried water pipe as an example, the facility structure parameters define its length, diameter and burial depth. Based on this, the spatial discrete layer generates a computational mesh containing tens of thousands of tetrahedral elements. The finite element analysis method assigns a thermal conductivity of 1.5 W / m / degree Celsius to the soil elements, assigns steel material properties to the water pipe elements, and applies the corresponding surface temperature values ​​in the boundary constraint vector to the nodes of the surface soil elements.

[0088] Step 1043: The heat conduction law is applied as a constraint term to the data stream of the finite element mesh through the equation embedding layer in the physical information neural network to obtain constrained mesh state data. The Monte Carlo method is introduced to randomly sample the constrained mesh state data to evaluate the uncertainty of local heat exchange and generate temperature fluctuation parameters.

[0089] Among them, the equation embedding layer is a structural layer in the physical information neural network responsible for integrating the mathematical expressions of physical laws into the forward propagation process of the neural network; the heat conduction law is a physical law describing the law of heat transfer in a medium; constrained mesh state data refers to finite element mesh data that already contains constraints of physical laws; the uncertainty of local heat exchange refers to the fluctuation of heat flow and temperature calculation results caused by random changes in material properties or boundary conditions; and the temperature fluctuation parameter is a parameter used to quantify the degree of influence of this uncertainty on the temperature field.

[0090] In this embodiment, the equation embedding layer of the physical information neural network adds the discrete form of the partial differential equation of heat conduction as an additional loss term or constraint term to the data stream processing associated with the finite element mesh. This forces the neural network to follow this physical law in subsequent calculations, thereby obtaining constrained mesh state data that conforms to the law of heat conduction.

[0091] Step 1043 may specifically include the following steps:

[0092] A1: Based on the constrained grid state data, determine the parameter variation range and distribution information of each grid cell.

[0093] In step A1, the parameter variation range refers to the possible numerical range of a certain thermal property parameter of a grid cell, and the distribution information refers to the probability distribution type and characteristic parameters of the parameter within the range.

[0094] In this embodiment, the thermal property parameters associated with each grid cell are read from the constrained grid state data. These parameters are not absolutely constant in the real world, so it is necessary to set a reasonable range of numerical variation for them. For example, the thermal conductivity of soil may vary between 1.3 and 1.7 watts per meter Celsius. At the same time, based on prior knowledge or measured data, it is determined that they follow a specific probability distribution within this range, such as a normal distribution or a uniform distribution, so as to obtain the parameter variation range and distribution information of each cell.

[0095] In practical applications, for the soil unit surrounding the aforementioned buried water pipe, according to the geological survey report, its thermal conductivity ranges from 1.35 to 1.65 watts per meter per degree Celsius, and approximately follows a normal distribution with a mean of 1.5 and a standard deviation of 0.075.

[0096] A2: Based on the parameter variation range and the distribution information, the Monte Carlo method is used to randomly sample the thermal property parameters of each grid cell multiple times to generate multiple sets of parameter samples.

[0097] In this embodiment, based on the parameter variation range and distribution information of each grid cell determined in step A1, a random sampling technique from the Monte Carlo method is used to independently generate a large number of random numbers for each cell. These random numbers conform to the probability distribution specified for that cell, and each random number represents a possible value of the thermal property parameter of that cell. Combining the parameter values ​​sampled from all cells in the same batch constitutes a set of parameter samples covering the entire simulation area. Repeating this sampling process multiple times, for example, one thousand times, generates one thousand sets of distinct parameter samples.

[0098] In practical applications, for the above soil unit, based on its normal distribution information, a random number generator is used to generate 1000 random numbers that conform to the distribution. For example, the first sample value is 1.52 watts per meter per degree Celsius, the second is 1.48 watts per meter per degree Celsius, and so on. These 1000 values ​​constitute the parameter sample sequence of the unit. After performing the same operation on all units, the parameter values ​​of all units at the i-th sampling are combined to obtain the i-th set of global parameter samples.

[0099] A3: For each set of parameter samples, calculate the heat flow data and temperature change data of the corresponding grid cell based on the heat conduction constraint.

[0100] In step A3, heat flow data refers to a calculated value that characterizes the amount of heat transferred per unit area per unit time, usually in watts per square meter. This data reflects the intensity and direction of heat transfer in the medium. Temperature change data refers to a calculated value that characterizes the rate at which the temperature value of a specific location or area changes over time, usually in degrees Celsius per second. This data reflects the rate at which the temperature rises or falls at that location or area.

[0101] In this embodiment, for each set of parameter samples generated in step A2, the parameter values ​​of each unit are substituted into the heat conduction constraint relationship defined by the equation embedding layer for calculation. This calculation process is completed in the forward propagation of the physical information neural network, using the finite element method to solve the steady-state or transient heat conduction problem under the given set of parameters, thereby obtaining the heat flux density data and temperature change rate data of each grid unit under the specific set of parameters.

[0102] In practical applications, taking the first set of parameters mentioned above, the thermal conductivity of a certain soil unit is 1.52 W / m / °C. Using this value along with other boundary conditions, the heat flux density of this unit under steady state is calculated to be 10.5 W / m², and the temperature change rate is 0°C / s.

[0103] A4: Based on the calculation results of all parameter samples, determine the heat flow fluctuation data and temperature fluctuation data for each grid cell.

[0104] In step A4, heat flux fluctuation data refers to an indicator used to quantify the degree of fluctuation of heat flux values ​​due to the randomness of input parameters after statistical analysis of heat flux data calculated from multiple sets of random parameter samples. It is usually expressed by variance or standard deviation. Temperature fluctuation data refers to an indicator used to quantify the degree of fluctuation of temperature change rate due to the randomness of input parameters after statistical analysis of temperature change data calculated from multiple sets of random parameter samples. It is usually expressed by variance or standard deviation.

[0105] In this embodiment, after the heat flux and temperature change data corresponding to all parameter samples have been calculated, the calculation results for each grid cell under all sample groups are collected. Statistical analysis is performed on these results, such as calculating their variance or standard deviation, to quantify the degree of fluctuation in the heat flux and temperature change of the cell due to parameter randomness, thereby obtaining the heat flux fluctuation data and temperature fluctuation data of the cell.

[0106] In practical applications, for the aforementioned soil unit, 1000 heat flux density calculation results were collected from 1000 sets of samples. The variance of these 1000 values ​​was calculated, and the heat flux fluctuation data was obtained as 0.021 squared. Similarly, the variance of the temperature change rate was calculated to obtain the temperature fluctuation data.

[0107] A5: Generate temperature fluctuation parameters based on the heat flow fluctuation data and temperature fluctuation data of all grid cells.

[0108] In this embodiment of the application, the heat flow fluctuation data and temperature fluctuation data of all grid cells in the simulation area obtained in step A4 are summarized and integrated to form a field data set corresponding to the grid structure. This set comprehensively describes the possible fluctuations in the entire simulation temperature field due to the uncertainty of material properties. This field data set serves as the final temperature fluctuation parameter.

[0109] In practical applications, temperature fluctuation parameters can be represented as a list containing tens of thousands of elements. Each element in the list corresponds to a grid cell, and each element itself contains two sub-items, which record the heat flow fluctuation data and temperature fluctuation data of that cell, respectively.

[0110] Step 1044: Through the iterative solution layer in the physical information neural network, combined with the temperature fluctuation parameters, the constrained grid state data is iteratively solved until the preset convergence condition is met, and stable temperature field data is obtained.

[0111] In step 1044, the iterative solution layer is a structural layer in the physical information neural network that performs cyclic numerical calculations to approximate the solution of the equation. The convergence condition refers to the criterion used to determine whether the iterative calculation can be terminated. The stable temperature field data refers to the temperature distribution data that no longer changes after the iterative calculation has converged.

[0112] In this embodiment, the iterative solution layer of the physical information neural network receives constrained grid state data from the equation embedding layer and temperature fluctuation parameters generated in step 1043. This layer uses numerical algorithms such as the conjugate gradient method or Newton's iteration method for iterative solution. In each iteration, not only is the residual calculated based on the current temperature field guess, but the temperature fluctuation parameters are also used to compensate for or correct the fluctuations in local heat exchange, thereby updating the temperature field.

[0113] This process is repeated until the temperature field change calculated in two adjacent iterations is less than a pre-set minimum threshold, which satisfies the convergence condition. At this point, stable temperature field data that takes into account both physical laws and parameter uncertainties is obtained.

[0114] In practical applications, the iterative solution layer starts with an initial temperature guess. Each iteration calculates a new temperature distribution and introduces the temperature fluctuation parameter as an additional term into the update formula. After, for example, 50 iterations, the maximum difference between the two consecutive full-field temperatures is less than 0.01 degrees Celsius, which satisfies convergence. The temperature field at this time is then output as stable temperature field data.

[0115] Step 1045: Extract the temperature data of the rural infrastructure surface from the stable temperature field data through the surface mapping layer in the physical information neural network, and use the temperature data as simulated temperature field data.

[0116] In step 1045, the surface mapping layer is a structural layer in the physical information neural network specifically used to extract specified surface data from three-dimensional volume data.

[0117] In this embodiment, the surface mapping layer of the physical information neural network receives stable temperature field data output by the iterative solution layer. This data is defined on all grid nodes. Based on the surface geometry information defined in the rural infrastructure digital model, the surface mapping layer selects the temperature values ​​of those grid nodes that are exactly located on the outer surface of the facility from the stable temperature field data, and reorganizes these temperature values ​​according to their spatial coordinates to form a two-dimensional temperature distribution map or a corresponding dataset. This result is the required simulation temperature field data.

[0118] In practical applications, stable temperature field data contains temperature values ​​from 100,000 nodes. The surface mapping layer extracts the temperature values ​​of 2,000 nodes that constitute the outer surface of the water pipe based on the model, and maps these values ​​to the coordinates of the nodes to generate simulated temperature field data describing the temperature distribution on the water pipe surface.

[0119] This application achieves a refinement and robustness improvement in thermodynamic simulation based on strict adherence to physical laws by combining physical information neural networks with Monte Carlo uncertainty analysis. This makes the output simulated temperature field data not only conform to theoretical laws, but also reflect the random fluctuations of material properties in the real world, providing a more reliable and realistic benchmark for subsequent anomaly detection.

[0120] Step 105: Spatial registration and pixel-by-pixel difference calculation are performed on the simulated temperature field data and the fused image data to obtain temperature difference data, and adaptive threshold segmentation is performed on the temperature difference data to determine abnormal hot areas.

[0121] Among them, temperature difference data is a set of data that reflects the distribution of differences between measured temperature and simulated temperature, obtained by difference calculation; adaptive threshold segmentation refers to a method that automatically determines the segmentation boundary based on the statistical characteristics of the input data itself; and abnormal hot zone refers to the spatially continuous region with differences in temperature difference data.

[0122] In this embodiment, step 105 includes the following process:

[0123] Step 1051: Map the simulated temperature field data and the fused image data to the same coordinate system.

[0124] In the embodiments of this application, firstly, since the simulated temperature field data and the fused image data may initially be based on different spatial references, it is necessary to map them to the same unified geographic coordinate system or image pixel coordinate system through coordinate transformation operations. This coordinate transformation process is completed using known control point coordinate transformation parameters or image affine transformation parameters, thereby ensuring that the two data can be accurately aligned in spatial location.

[0125] Step 1052: For each location point in the coordinate system, calculate the difference between the temperature value in the fused image data and the temperature value in the simulated temperature field data to obtain the temperature difference value of the location point.

[0126] In step 1052, the temperature difference refers to the result obtained by subtracting the temperature simulation value read from the simulated temperature field data from the temperature measurement value read from the fused image data at the same location point after registration.

[0127] In this embodiment of the application, after spatial registration is completed, for each pixel or grid point in a unified coordinate system, the measured temperature value of that point is extracted from the fused image data, and the simulated temperature value of the same point is extracted from the simulated temperature field data. The measured value is subtracted from the simulated value, and the calculated value is the temperature difference at that location. A positive value indicates that the measured temperature is higher than the simulated temperature, and a negative value indicates the opposite.

[0128] Step 1053: Generate temperature difference data based on the temperature difference values ​​at all locations.

[0129] In this embodiment of the application, the temperature difference values ​​calculated from all location points are arranged and organized according to their original spatial positional relationship to form a new data field with the same spatial dimension as the original data. This complete new data field is the temperature difference data, which intuitively shows the distribution of the inconsistency between the measured temperature and the simulated temperature in the entire area.

[0130] Step 1054: Perform feature transformation on the temperature difference data using principal component analysis to extract the main changing components and obtain feature-enhanced difference data.

[0131] In step 1054, the main changing component refers to the new feature direction with the largest variance, and the feature-enhanced difference data refers to the data representation obtained after principal component transformation, which can more prominently display the main abnormal patterns.

[0132] In this embodiment, to more effectively separate anomalies caused by real faults from general random noise in temperature difference data, principal component analysis is used to process the data. This method first expands the two-dimensional temperature difference data matrix into a set of one-dimensional vectors and calculates its covariance matrix. Then, it solves for the eigenvalues ​​and eigenvectors of the covariance matrix, selects the eigenvectors corresponding to the first few largest eigenvalues ​​as principal component directions, and finally projects the original temperature difference data onto these principal component directions to obtain feature-enhanced difference data that strengthens the main anomaly patterns while suppressing secondary noise.

[0133] Step 1055: Based on the statistical distribution of the difference data after feature enhancement, use a clustering algorithm to automatically determine multiple candidate segmentation thresholds.

[0134] In step 1055, statistical distribution refers to the distribution pattern of all values ​​in the feature-enhanced differential data, such as a histogram; candidate segmentation threshold refers to the possible boundary value initially proposed based on data distribution or clustering results to distinguish between background and anomalies.

[0135] In this embodiment, the statistical distribution of the enhanced difference data is analyzed, for example, by plotting its numerical histogram. A clustering algorithm, such as K-means clustering, is used to automatically group the data values. This algorithm divides all values ​​into, for example, three clusters based on the similarity between data points: one cluster may represent background noise, one cluster represents moderate difference, and one cluster represents high difference. The center value or boundary value of each cluster, such as the midpoint between adjacent clusters, can be initially determined as candidate segmentation thresholds for distinguishing different regions.

[0136] Step 1056: Use morphological methods to screen and optimize multiple candidate segmentation thresholds to determine the final segmentation threshold.

[0137] In step 1056, the final segmentation threshold refers to the unique threshold determined after optimization and screening and used for the actual segmentation operation.

[0138] In this embodiment of the application, for the multiple candidate segmentation thresholds obtained in step 1055, they are used to perform preliminary segmentation on the feature-enhanced difference data to obtain multiple binary images. Then, morphological methods are applied to these binary images, such as opening operations to remove small noise or closing operations to fill small holes, and the average area, roundness, and other shape features of the connected regions segmented under each threshold are calculated. By comparing these shape features and combining the spatial continuity characteristics that rural infrastructure anomalies usually have, for example, selecting candidate thresholds that can produce connected regions of moderate size and relatively regular shape, or performing a weighted average of several candidate thresholds, the final segmentation threshold that best matches the actual situation is screened and determined.

[0139] Step 1057: Identify the regions in the temperature difference data where the temperature difference value is greater than the final segmentation threshold as abnormal hot zones.

[0140] In this embodiment, the final segmentation threshold determined in step 1056 is applied to the original temperature difference data, rather than the feature-enhanced data. Each value in the temperature difference data is traversed, and all pixels with temperature differences greater than the final segmentation threshold are marked. Then, these marked points are spatially adjacent and grouped together to form connected regions. These connected regions with temperature differences higher than the threshold are identified as the abnormal hot areas to be found.

[0141] This application integrates principal component analysis, clustering, and morphological methods to achieve adaptive and intelligent threshold segmentation of temperature difference data. It can effectively distinguish between real fault thermal anomalies and background noise, thereby accurately locating abnormal hot areas of rural infrastructure.

[0142] Step 106: Analyze the morphological characteristics, temporal variation trend, and temperature gradient with the surrounding area of ​​the abnormal hot zone. Combined with the type of rural infrastructure, determine the final fault type. Based on the final fault type and the geographical location of the abnormal hot zone, automatically generate an operation and maintenance work order and send it to the operation and maintenance terminal.

[0143] Among them, morphological characteristics refer to parameters describing the spatial attributes such as the geometry and size of the anomalous hot zone; temporal variation trend refers to parameters describing the evolution of the area and center location of the anomalous hot zone over time; temperature gradient refers to parameters quantifying the temperature difference between the anomalous hot zone and the surrounding area; final fault type refers to the specific classification result of hidden faults in rural infrastructure, such as minor pipe leaks or overheating of cable joints; maintenance work order refers to a digital work order containing information such as fault handling tasks, location, type, and priority; and maintenance terminal refers to a mobile computing device held by maintenance personnel for receiving and processing work order information.

[0144] In this embodiment, step 106 includes the following process:

[0145] Step 1061: Extract the contour shape and area size from the abnormal hot zone as morphological features.

[0146] In step 1061, the contour shape refers to the external morphology of the anomalous hot zone described by the sequence of boundary points, and the area size refers to the size of the two-dimensional planar region covered by the anomalous hot zone.

[0147] In this embodiment, image processing technology is first used to operate on the binary image of the identified abnormal hot area. Specifically, the boundary point sequence of each abnormal hot area is found through the contour tracking algorithm. These sequences are used to calculate parameters describing the shape, such as the aspect ratio of the circumscribed rectangle or the shape compactness. These parameters, together with the area size, serve as morphological features.

[0148] Step 1062: Obtain the area change rate and center position movement trajectory of the abnormal hot zone in the fused image data as a temporal change trend.

[0149] In step 1062, the area change rate refers to how quickly the area of ​​the abnormal hot zone changes over time, and the center position movement trajectory refers to the position change path of the geometric center point of the abnormal hot zone at a series of consecutive time points.

[0150] In the embodiments of this application, the same abnormal hot area is tracked in a time series composed of continuous multi-frame fused image data. The area growth or decrease value per unit time is calculated by comparing the area of ​​the region at different time points to obtain the area change rate. At the same time, the geometric center point coordinates of the region at each time point are calculated and these coordinates are connected in time order to form the movement trajectory of its center position. These two together constitute the temporal change trend.

[0151] Step 1063: Calculate the difference between the average temperature inside the abnormal hot zone and the average temperature of the outer annular region, as the temperature gradient.

[0152] In step 1063, the internal average temperature refers to the result of arithmetically averaging the temperature values ​​of all pixels within the abnormal hot zone, the outer ring region refers to the ring-shaped area with a certain width adjacent to the boundary of the abnormal hot zone, and the temperature gradient is the difference between the internal average temperature and the average temperature of the outer ring region.

[0153] In this embodiment, based on the fused image data of the current frame, the average temperature of all pixels inside the abnormal hot zone is first calculated. Then, a region of a preset width is extended outward from the boundary of the hot zone to form an outer ring region, and the average temperature of all pixels in the region is calculated. Finally, the average temperature of the inner ring region is subtracted from the average temperature of the outer ring region to obtain the temperature gradient.

[0154] Step 1064: Input the morphological features, the temporal variation trend, and the temperature gradient into the support vector machine classifier. The support vector machine classifier matches the corresponding fault determination rule according to the type of rural infrastructure, and inputs the morphological features, the temporal variation trend, and the temperature gradient into the fault determination rule corresponding to the facility type for matching, and outputs the initial fault type.

[0155] In step 1064, the fault determination rule is a decision boundary that the support vector machine classifier has pre-learned based on different facility types to distinguish different fault categories, and the initial fault type is the fault classification result initially given by the support vector machine classifier.

[0156] In this embodiment, the morphological features, area change rate, center movement trajectory, and temperature gradient extracted in steps 1061 to 1063 are combined into a feature vector. This feature vector is input into a pre-trained support vector machine classifier. The classifier has independent models trained for different types of rural infrastructure. The corresponding model is selected according to the type of facility being analyzed. The model determines the category of the input feature vector in its multidimensional feature space based on the value of the input feature vector, and outputs an initial fault type, such as "minor pipe leak" or "overheating of cable joint".

[0157] Step 1065: Verify and refine the initial fault type using a convolutional neural network to obtain the final fault type.

[0158] In this embodiment of the application, in order to improve the accuracy of diagnosis, the original fused image data block containing the abnormal hot zone is input into a pre-trained convolutional neural network. The network structure includes multiple convolutional layers and pooling layers, which can extract richer spatial pattern features. The network receives the image block and the initial fault type output by the support vector machine as part of the input. Through its internal calculation, it determines whether the actual pattern of the image block supports the initial classification or further refines it based on the initial classification. For example, it refines "cable joint overheating" into "joint A phase overheating", thereby outputting the final fault type.

[0159] Step 1066: Associate the final fault type with the geographic coordinates of the center of the abnormal hot zone, select the corresponding template from the preset work order template library, fill in the fault type, geographic coordinates and urgency level derived from temperature gradient and rate of change into the template, generate an operation and maintenance work order, and send the operation and maintenance work order to the mobile terminal of the designated operation and maintenance personnel.

[0160] In this embodiment, the final fault type determined in step 1065 is associated with the geographical coordinates of the center of the abnormal hot zone calculated in step 1062. Based on the final fault type, the most matching work order template is automatically selected from a pre-set work order template library. Then, key information such as the fault type, geographical coordinates, and an urgency level calculated based on the temperature gradient magnitude and area change rate are automatically filled into the corresponding fields of the template to generate a complete digital operation and maintenance work order. This work order is immediately pushed to the mobile terminal of the designated personnel responsible for the operation and maintenance of the area via wireless network.

[0161] In this embodiment, after step 106, the method further includes the following steps:

[0162] B1: Based on the location information and fault type of the abnormal hot zone, relevant historical maintenance solutions are searched through the knowledge graph of historical maintenance.

[0163] In step B1, the historical maintenance knowledge graph is a knowledge base that organizes and stores historical maintenance data and experience in a graph structure, where nodes represent entities such as fault types and maintenance parts, and edges represent relationships between entities such as "adoptable solutions".

[0164] In this embodiment of the application, based on the geographical location of the abnormal hot zone and the final fault type obtained in step 106, the historical maintenance knowledge graph is queried. The query process is to find entity nodes in the graph that are the same or similar to the current fault type and are geographically close, and retrieve the associated historical maintenance cases and their specific solutions along the relationship edges such as "maintenance plan".

[0165] B2: Merge the historical maintenance plan with the current detection data of the abnormal hot zone to generate operation and maintenance guidance information.

[0166] In step B2, the maintenance guidance information is a comprehensive information document that integrates historical experience with the current specific situation to guide on-site maintenance operations.

[0167] In this embodiment of the application, the historical maintenance plan retrieved in step B1 is integrated with the specific detection data of the current abnormal hot zone and added to the operation and maintenance work order to form an operation and maintenance guidance information containing historical reference plans and current detailed data.

[0168] B3: Input the operation and maintenance work order and operation and maintenance guidance information into the multi-agent reinforcement learning model, which includes a path planning agent, a resource scheduling agent, and a risk assessment agent.

[0169] In this embodiment, the generated operation and maintenance work orders and operation and maintenance guidance information are input as environmental states into a multi-agent reinforcement learning model. The model contains three specialized agents: a path planning agent is responsible for calculating the optimal path; a resource scheduling agent is responsible for managing materials; and a risk assessment agent is responsible for assessing operational risks. These agents have been trained through reinforcement learning using historical task data and have learned the optimal action strategy in a given state.

[0170] B4: The path planning agent dynamically calculates the optimal arrival path based on the real-time location of the maintenance personnel, road conditions, and fault location in the maintenance work order; the resource scheduling agent generates material preparation information based on the maintenance guidance information; and the risk assessment agent assesses maintenance risks and generates risk assessment information based on current environmental data and fault change trends.

[0171] In step B4, the optimal arrival path refers to a recommended route calculated by a path planning algorithm under multiple constraints such as a given starting point, target point, and current road conditions, which comprehensively considers optimization objectives such as the shortest travel distance, the least time, or the lowest cost; the material preparation information refers to a list and status description automatically generated by the system based on the specific fault type and repair plan, which includes the names, specifications, and recommended quantities of various repair materials and special tools required to complete this repair operation.

[0172] Current environmental data refers to external condition information that may affect the safety of maintenance operations at the time of risk assessment, mainly including real-time meteorological data such as temperature, humidity, wind speed, and precipitation probability; fault change trend refers to the recent evolution of the fault in terms of heat, area, or location, determined by the analysis of multiple frames of monitoring data of abnormal hot areas, such as continued spread or stabilization.

[0173] Risk assessment information refers to the quantitative or qualitative assessment conclusions and response recommendations made by analyzing current environmental data and failure change trends, combined with a historical accident case database, on the potential risks that may be encountered during the execution of this maintenance task, such as personal safety, equipment damage, or operation failure, as well as their probability and severity.

[0174] In this embodiment, three intelligent agents work in parallel. The path planning agent receives the real-time GPS location, real-time traffic information, and fault location of the maintenance personnel and uses its internal path planning algorithm to dynamically calculate the optimal route from the personnel's current location to the fault location. The resource scheduling agent analyzes the fault type and maintenance plan sections in the maintenance guidance information and generates a list of required maintenance materials and tools and their inventory status based on a pre-set material-fault association library to form material preparation information. The risk assessment agent integrates information such as the current weather conditions and the temperature change trend of the fault, uses its internal risk assessment model to predict the risks that may be encountered in the maintenance operation, and generates corresponding risk assessment information.

[0175] B5: Package the operation and maintenance guidance information, the optimal arrival path, the material preparation information, and the risk assessment information into an enhanced work order data package and send it to the operation and maintenance terminal.

[0176] In this embodiment, the original operation and maintenance guidance information, the optimal arrival path calculated by the path planning agent, the material preparation information generated by the resource scheduling agent, and the risk assessment information generated by the risk assessment agent are packaged together into an enhanced work order data package with more comprehensive information and higher decision support. This data package is sent and updated to the mobile terminal of the operation and maintenance personnel through the mobile network, providing them with comprehensive auxiliary decision-making information from fault diagnosis to on-site execution.

[0177] This application integrates support vector machines and convolutional neural networks for two-level fault diagnosis and introduces a multi-agent collaborative mechanism for operation and maintenance decision support, realizing a complete closed loop from accurate fault identification to efficient and intelligent operation and maintenance response, thereby improving the intelligence level and decision-making efficiency of rural infrastructure operation and maintenance.

[0178] Figure 3 A schematic diagram of the structure of a digital twin-based intelligent operation and maintenance system for rural infrastructure is provided in this application embodiment, as shown below. Figure 3 As shown, the system includes:

[0179] The acquisition module 31 is used to acquire time-series infrared data, visible light image data and environmental parameter data collected for the area where rural infrastructure is located.

[0180] The correction module 32 is used to perform temperature calibration and radiation correction on the time-series infrared data to obtain initial temperature field data, and to fuse the initial temperature field data with the visible light image data to generate fused image data.

[0181] The identification module 33 is used to input the fused image data into the generative adversarial network model, identify and extract the thermal anomaly features of rural infrastructure, and synchronize the thermal anomaly features as state update parameters to the corresponding digital twin.

[0182] The simulation module 34 is used to input the environmental parameter data into the updated digital twin, drive the physical information neural network integrated in the updated digital twin to perform thermodynamic simulation, and output the simulated temperature field data of the rural infrastructure surface.

[0183] The calculation module 35 is used to perform spatial registration and pixel-by-pixel difference calculation on the simulated temperature field data and the fused image data to obtain temperature difference data, and to perform adaptive threshold segmentation on the temperature difference data to determine abnormal hot areas.

[0184] Analysis module 36 is used to analyze the morphological characteristics, temporal variation trend and temperature gradient with the surrounding area of ​​the abnormal hot zone, and determine the final fault type in combination with the type of rural infrastructure. Based on the final fault type and the geographical location of the abnormal hot zone, it automatically generates an operation and maintenance work order and sends it to the operation and maintenance terminal.

[0185] The intelligent operation and maintenance system for rural infrastructure based on digital twins in this application is used to implement the aforementioned intelligent operation and maintenance method for rural infrastructure based on digital twins. Therefore, the specific implementation of the intelligent operation and maintenance system for rural infrastructure based on digital twins can be found in the embodiment section of the intelligent operation and maintenance method for rural infrastructure based on digital twins above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0186] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described intelligent operation and maintenance methods for rural infrastructure based on digital twins.

[0187] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described intelligent operation and maintenance methods for rural infrastructure based on digital twins.

[0188] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0189] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the above-described intelligent operation and maintenance method for rural infrastructure based on digital twins.

[0190] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0191] 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, data stored, data displayed, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0192] The above provides a detailed description of a digital twin-based intelligent operation and maintenance method and system for rural infrastructure. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for intelligent operation and maintenance of rural infrastructure based on digital twins, characterized in that, include: Acquire time-series infrared data, visible light image data, and environmental parameter data for areas where rural infrastructure is located. Rural infrastructure refers to physical facilities in the rural environment that require maintenance, such as buried water pipes, heating pipes, or power cable trenches. Time-series infrared data refers to measurement data that reflects the temperature changes of the surface and shallow facilities, which are continuously collected by infrared sensors in chronological order. Visible light image data refers to optical image data that reflects the visual texture of the surface, which are taken in the visible light band. Environmental parameter data includes air temperature, soil moisture, and solar radiation intensity, which affect the heat exchange of the facilities. Temperature calibration and radiation correction are performed on the time-series infrared data to obtain initial temperature field data, and the initial temperature field data is fused with the visible light image data to generate fused image data; The fused image data is input into a generative adversarial network model to identify and extract thermal anomaly features of rural infrastructure, and the thermal anomaly features are used as state update parameters to be synchronized to the corresponding digital twin. The environmental parameter data is input into the updated digital twin's physical information neural network. The boundary condition encoding layer of the physical information neural network processes the environmental parameter data to generate boundary constraint vectors that interact with the environment. The spatial discretization layer of the physical information neural network performs mesh generation based on the boundary constraint vectors and facility structure parameters to obtain a discrete computational mesh. The finite element analysis method is then used to assign material properties and apply boundary conditions to the computational mesh to obtain a finite element mesh. The heat conduction law is applied as a constraint term to the data stream of the finite element mesh through the equation embedding layer in the physical information neural network, resulting in constrained mesh state data. The Monte Carlo method is then introduced to randomly sample the constrained mesh state data to evaluate the uncertainty of local heat exchange and generate temperature fluctuation parameters. The iterative solution layer in the physical information neural network, combined with the temperature fluctuation parameters, performs iterative calculations on the constrained mesh state data until a preset convergence condition is met, resulting in stable temperature field data. Temperature data of the rural infrastructure surface is extracted from the stable temperature field data through the surface mapping layer in the physical information neural network, and the temperature data is used as simulated temperature field data. Spatial registration and pixel-by-pixel difference calculation are performed on the simulated temperature field data and the fused image data to obtain temperature difference data. Adaptive threshold segmentation is then performed on the temperature difference data to determine abnormal hot areas. The morphological characteristics, temporal variation trend, and temperature gradient with the surrounding area of ​​the abnormal hot zone are analyzed. Combined with the type of rural infrastructure, the final fault type is determined. Based on the final fault type and the geographical location of the abnormal hot zone, an operation and maintenance work order is automatically generated and sent to the operation and maintenance terminal.

2. The method according to claim 1, characterized in that, The Monte Carlo method is introduced to randomly sample the constrained grid state data, evaluate the uncertainty of local heat exchange, and generate temperature fluctuation parameters, including: Based on the constrained grid state data, determine the parameter variation range and distribution information of each grid cell; Based on the parameter variation range and the distribution information, the Monte Carlo method is used to randomly sample the thermal property parameters of each grid cell multiple times to generate multiple sets of parameter samples. For each set of parameter samples, calculate the heat flow data and temperature change data of the corresponding grid cell based on the heat conduction constraint; Based on the calculation results of all parameter samples, determine the heat flow fluctuation data and temperature fluctuation data for each grid cell; Temperature fluctuation parameters are generated based on the heat flow fluctuation data and temperature fluctuation data of all grid cells.

3. The method according to claim 1, characterized in that, The step of inputting the fused image data into a generative adversarial network model to identify and extract thermal anomaly features of rural infrastructure, and synchronizing these thermal anomaly features as state update parameters to the corresponding digital twin, includes: The fused image data is input into the discriminator of the generative adversarial network model, and low-level features are extracted from the fused image data through the first network layer of the discriminator. The discriminator processes the underlying features through its second network layer and combines them with the spatial relationships of the fused image data to identify abnormal hot regions. It then extracts the contour features and temperature statistical features of the abnormal hot regions as the first features. The third network layer of the discriminator combines the fused image data of the current frame and historical frames to analyze the temperature change trend of the abnormal hot area as a second feature. The first feature is combined with the second feature to generate thermal anomaly features; The thermal anomaly features are converted into a parameter format recognizable by the digital twin to obtain the state update parameters; The state update parameters are sent to the digital twin, and the thermodynamic state parameters at the corresponding positions in the digital twin are adjusted to obtain the updated digital twin.

4. The method according to claim 1, characterized in that, The step of adaptive threshold segmentation of the temperature difference data to determine abnormal hot zones includes: The temperature difference data is subjected to feature transformation by principal component analysis to extract the main changing components and obtain feature-enhanced difference data. Based on the statistical distribution of the difference data after feature enhancement, a clustering algorithm is used to automatically determine multiple candidate segmentation thresholds; Morphological methods were used to screen and optimize multiple candidate segmentation thresholds to determine the final segmentation threshold; The regions in the temperature difference data where the temperature difference value is greater than the final segmentation threshold are identified as abnormal hot zones.

5. The method according to claim 1, characterized in that, The analysis of the morphological characteristics, temporal variation trends, and temperature gradient with the surrounding areas of the abnormal hot zone, combined with the type of rural infrastructure, determines the final failure type, including: The contour shape and area size of the abnormal hot zone are extracted as morphological features; The area change rate and center position movement trajectory of the abnormal hot zone in the fused image data are obtained as the temporal variation trend. The difference between the average temperature inside the abnormal hot zone and the average temperature of the outer annular region is calculated as the temperature gradient. The morphological features, the temporal variation trend, and the temperature gradient are input into a support vector machine classifier. The support vector machine classifier matches the corresponding fault determination rule according to the type of rural infrastructure, and matches the morphological features, the temporal variation trend, and the temperature gradient with the fault determination rule corresponding to the facility type, and outputs the initial fault type. The initial fault type is verified and refined using a convolutional neural network to obtain the final fault type.

6. The method according to claim 1, characterized in that, After automatically generating a maintenance work order and sending it to the maintenance terminal based on the final fault type and the geographical location of the abnormal hotspot, the process also includes: The operation and maintenance work order and operation and maintenance guidance information are input into the multi-agent reinforcement learning model, which includes a path planning agent, a resource scheduling agent, and a risk assessment agent. The path planning agent dynamically calculates the optimal arrival path based on the real-time location of the maintenance personnel, road conditions, and fault location in the maintenance work order; the resource scheduling agent generates material preparation information based on the maintenance guidance information; and the risk assessment agent assesses maintenance risks and generates risk assessment information based on current environmental data and fault change trends. The operation and maintenance guidance information, the optimal arrival path, the material preparation information, and the risk assessment information are packaged into an enhanced work order data package and sent to the operation and maintenance terminal.

7. A smart operation and maintenance system for rural infrastructure based on digital twins, characterized in that, include: The acquisition module is used to acquire time-series infrared data, visible light image data, and environmental parameter data collected for the areas where rural infrastructure is located. Rural infrastructure refers to physical facilities in the rural environment that require maintenance, such as buried water pipes, heating pipes, or power cable trenches. Time-series infrared data refers to measurement data that reflects the temperature changes of the surface and shallow facilities, which are continuously collected by infrared sensors in chronological order. Visible light image data refers to optical image data that reflects the visual texture of the surface, which are taken in the visible light band. Environmental parameter data includes air temperature, soil moisture, and solar radiation intensity, which affect the heat exchange of the facilities. The correction module is used to perform temperature calibration and radiation correction on the time-series infrared data to obtain initial temperature field data, and to fuse the initial temperature field data with the visible light image data to generate fused image data. The identification module is used to input the fused image data into the generative adversarial network model, identify and extract the thermal anomaly features of rural infrastructure, and synchronize the thermal anomaly features as state update parameters to the corresponding digital twin. The simulation module is used to input the environmental parameter data into the physical information neural network of the updated digital twin. Through the boundary condition encoding layer of the physical information neural network, the environmental parameter data is processed to generate boundary constraint vectors that interact with the environment. Through the spatial discretization layer in the physical information neural network, the boundary constraint vectors and facility structure parameters are used to perform mesh generation to obtain a discrete computational mesh. The finite element analysis method is then used to assign material properties and apply boundary conditions to the computational mesh to obtain a finite element mesh. The heat conduction law is applied as a constraint term to the data stream of the finite element mesh through the equation embedding layer in the physical information neural network, resulting in constrained mesh state data. The Monte Carlo method is then introduced to randomly sample the constrained mesh state data to evaluate the uncertainty of local heat exchange and generate temperature fluctuation parameters. The iterative solution layer in the physical information neural network, combined with the temperature fluctuation parameters, performs iterative calculations on the constrained mesh state data until a preset convergence condition is met, resulting in stable temperature field data. Temperature data of the rural infrastructure surface is extracted from the stable temperature field data through the surface mapping layer in the physical information neural network, and the temperature data is used as simulated temperature field data. The calculation module is used to perform spatial registration and pixel-by-pixel difference calculation on the simulated temperature field data and the fused image data to obtain temperature difference data, and to perform adaptive threshold segmentation on the temperature difference data to determine abnormal hot areas. The analysis module is used to analyze the morphological characteristics, temporal variation trend, and temperature gradient with the surrounding area of ​​the abnormal hot zone. Combined with the type of rural infrastructure, it determines the final fault type and automatically generates an operation and maintenance work order based on the final fault type and the geographical location of the abnormal hot zone, and sends it to the operation and maintenance terminal.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the intelligent operation and maintenance method for rural infrastructure based on digital twins as described in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the intelligent operation and maintenance method for rural infrastructure based on digital twins as described in any one of claims 1 to 6.

Images