An infrastructure intelligent monitoring method of skin-like structure

Abstract

The application discloses a kind of infrastructure intelligent monitoring methods of skin-like structure, it is related to civil engineering internet of things technical field.Method comprising: obtaining the infrastructure data of multiple sensor nodes in sensor network;According to the infrastructure data of multiple sensor nodes, construct space-time correlation model, according to space-time correlation model, distinguish the nature of each sensor node in sensor network;Time feature and spatial feature of space-time correlation model are input into conditional variation auto-encoder CVAE model, determine the risk coefficient of each sensor node;According to the risk coefficient of each sensor node, the structural health state of infrastructure is visualized, and according to the risk coefficient of each sensor node and the nature of each sensor node, the infrastructure monitoring work of sensor network is adjusted.The application can guarantee monitoring accuracy, effectively reduce system energy consumption, and prolong the service life of sensor network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of civil engineering Internet of Things, in particular to a kind of infrastructure intelligent monitoring method of skin structure. BACKGROUND

[0002] With the acceleration of urbanization, the safety and durability of infrastructure have become the focus of society. Wireless sensor networks have become one of the important tools for intelligent monitoring of skin structure infrastructure due to their low cost, wide range of detection and easy deployment. However, traditional structure health monitoring methods still have deficiencies in data processing, feature extraction and health state evaluation, which makes it difficult to meet the needs of comprehensive understanding and accurate prediction of complex structure data. In addition, infrastructure structure health monitoring faces problems such as difficulty in sensor installation, high energy consumption in long-term operation, and strong data heterogeneity, which limits the widespread application of wireless sensor networks in structure health monitoring. SUMMARY

[0003] In order to solve the technical problems that the prior art cannot meet the needs of comprehensive understanding and accurate prediction of complex structure data, the embodiments of the present application provide a kind of infrastructure intelligent monitoring method and system of skin structure. The technical solution is as follows:

[0004] On the one hand, a kind of infrastructure intelligent monitoring method of skin structure is provided, which is realized by infrastructure intelligent monitoring equipment of skin structure. The method comprises:

[0005] S1, obtaining infrastructure data of a plurality of sensor nodes in a sensor network;

[0006] S2, constructing a spatio-temporal correlation model according to the infrastructure data of the plurality of sensor nodes, and distinguishing the properties of each sensor node in the sensor network according to the spatio-temporal correlation model;

[0007] S3, inputting the time feature and space feature of the spatio-temporal correlation model into a conditional variational autoencoder (CVAE) model to determine the risk coefficient of each sensor node;

[0008] S4, visualizing the structure health state of the infrastructure according to the risk coefficient of each sensor node, and adjusting the infrastructure monitoring work of the sensor network according to the risk coefficient of each sensor node and the properties of each sensor node.

[0009] On the other hand, a kind of infrastructure intelligent monitoring system of skin structure is provided, which is applied to the infrastructure intelligent monitoring method of skin structure. The system comprises:

[0010] The data substrate layer is configured to obtain infrastructure data of a plurality of sensor nodes in a sensor network.

[0011] The connection structure layer is configured to construct a spatio-temporal correlation model according to the infrastructure data of the plurality of sensor nodes, and distinguish properties of the sensor nodes in the sensor network according to the spatio-temporal correlation model.

[0012] The pathological manifestation layer is configured to input time characteristics and space characteristics of the spatio-temporal correlation model into a conditional variational autoencoder (CVAE) model, determine a risk coefficient of each sensor node, visually process a structural health state of the infrastructure according to the risk coefficient of each sensor node, and adjust infrastructure monitoring work of the sensor network according to the risk coefficient of each sensor node and the properties of the sensor nodes.

[0013] In another aspect, an infrastructure intelligent monitoring device of a skin-like structure is provided, which includes a processor and a memory having computer readable instructions stored thereon. The computer readable instructions are executed by the processor to implement any one of the above-mentioned methods for intelligent monitoring of infrastructure of a skin-like structure.

[0014] In another aspect, a computer readable storage medium is provided, which stores at least one instruction. The at least one instruction is loaded and executed by a processor to implement any one of the above-mentioned methods for intelligent monitoring of infrastructure of a skin-like structure.

[0015] The technical solutions provided by the embodiments of the present application have at least the following beneficial effects:

[0016] The present application adopts a time series expansion convolution network and a dimension isolation graph attention network to construct spatio-temporal correlation, adopts a conditional variational autoencoder for data fusion, and simultaneously proposes a multi-granularity perception map adjustment strategy to adapt to monitoring requirements under different safety conditions, thereby achieving energy saving. The present application can effectively perform structural health assessment, has the characteristics of high precision, intuitiveness and low energy consumption, and prolongs the service life of the sensor network. In addition, the present application also proposes a multi-granularity perception map adjustment strategy, which can dynamically adjust the monitoring granularity according to actual monitoring requirements, further improving the flexibility and adaptability of monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Figure 1 is a structural schematic diagram of a skin-like structure infrastructure intelligent monitoring system provided by an embodiment of the present application;

[0019] Figure 2 is a flow chart of a skin-like structure infrastructure intelligent monitoring method provided by an embodiment of the present application;

[0020] Figure 3 (a)-(d) are effect schematic diagrams presented by pathological performance layers of an application example provided by an embodiment of the present application;

[0021] Figure 4 is a block diagram of a skin-like structure infrastructure intelligent monitoring system provided by an embodiment of the present application;

[0022] Figure 5 is a structural schematic diagram of a skin-like structure infrastructure intelligent monitoring device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0023] The technical solutions in the present application will be described below with reference to the drawings.

[0024] In the embodiments of the present application, the words such as "example", "for example" are used to represent as an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0025] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times. It should be pointed out that the meanings expressed are consistent when the distinction is not emphasized. "Of", "corresponding" and "corresponding" can be used interchangeably at times. It should be pointed out that the meanings expressed are consistent when the distinction is not emphasized.

[0026] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1. The meanings expressed are consistent when the distinction is not emphasized.

[0027] In order to make the technical problems, technical solutions and advantages to be solved by the present application more clear, specific embodiments will be described in detail below with reference to the drawings.

[0028] The embodiment of the present invention provides a method for intelligent monitoring of infrastructure with skin-like structures. The method can be implemented by an intelligent monitoring system for infrastructure with skin-like structures. The intelligent monitoring system for infrastructure with skin-like structures improves the accuracy and intuitiveness of wireless sensor networks (WSNs) in structural health monitoring by imitating the multi-layered structure of biological skin. Figure 1 As shown in the figure, the infrastructure intelligent monitoring system of this type of skin structure consists of three main layers: data base layer, connection structure layer and pathological manifestation layer. The data base layer is equivalent to the subcutaneous tissue layer, which is used for signal or data collection; the connection structure layer is equivalent to the dermis layer, which is used for material or information transmission; the pathological manifestation layer is equivalent to the epidermis layer, which is used to characterize the health status. This method is applicable to the structural health monitoring of various infrastructures, including but not limited to buildings, bridges and tunnels. Figure 2 The flowchart of the method for intelligent monitoring of infrastructure with a skin-like structure is shown. The processing flow of the method may include the following steps:

[0029] S1. Obtain infrastructure data of multiple sensor nodes in the sensor network.

[0030] Optionally, the specific operation of S1 may include the following steps S11-S12:

[0031] S11. Collect data from different types of sensor nodes in the sensor network.

[0032] In one feasible implementation, the data base layer collects data from various sensors evenly distributed across key infrastructure locations, such as load-bearing columns and beams. These sensors include, but are not limited to, force sensors, torsion sensors, displacement sensors, and water seepage sensors. These sensors transmit this data to a central processing unit via wireless communication technologies such as WiFi and ZigBee.

[0033] S12. Preprocess the collected data to obtain infrastructure data.

[0034] In one feasible implementation, the collected data first undergoes a preprocessing step, including data cleaning, format unification, and preliminary filtering, to remove invalid or erroneous data points and ensure the accuracy and reliability of subsequent processing.

[0035] S2. Construct a spatiotemporal correlation model based on the infrastructure data of multiple sensor nodes, and distinguish the properties of each sensor node in the sensor network based on the spatiotemporal correlation model.

[0036] Optionally, the specific operation of S2 may include the following steps S21-S22:

[0037] S21, according to the infrastructure data of a plurality of sensor nodes, a method combining a time series expansion convolutional network and a dimension-isolated graph attention network is adopted to construct a spatio-temporal correlation model.

[0038] Optionally, the specific operation of S21 can include the following steps:

[0039] S211, time series data is generated according to the data of a plurality of sensor nodes, the time series data is input to a time series expansion convolutional network, and a time feature vector on different time scales is captured through a plurality of layers of causal convolutional layers in the time series expansion convolutional network;

[0040] S212, the time feature vector is combined with original spatial data in the data of a plurality of sensor nodes to obtain a new feature matrix, the new feature matrix is input to a dimension-isolated graph attention network, a weight between nodes is calculated according to a multi-head attention mechanism in the graph attention network, and an attention coefficient between each node and all adjacent nodes is calculated to obtain a spatio-temporal correlation between nodes.

[0041] In a feasible implementation, the connection structure layer extracts the spatio-temporal relationship between sensor nodes by constructing a spatio-temporal correlation model. The present application adopts a method combining a time series expansion convolutional network (TCN) and a dimension-isolated graph attention network (GAT), the TCN is used to capture long-term dependence in time series data, and the GAT is used to capture node correlation in spatial data. Through this combination, spatio-temporal features can be extracted more effectively, providing a solid foundation for health assessment. The specific construction process is introduced as follows:

[0042] First, the time series data collected from the data base layer is input to the TCN module. In the TCN module, a plurality of layers of causal convolutional layers are designed, and the size of the convolution kernel increases layer by layer to capture features on different time scales. At the same time, a residual block is added after each layer;

[0043] After the TCN module is processed, the time feature vector obtained is combined with the original spatial data to form a new feature matrix. The feature matrix is transmitted as input to a graph attention network (GAT) module. In the GAT module, an undirected graph is defined, in which each sensor node corresponds to a vertex, and the weight of the edge reflects the spatial correlation between the nodes. Each node (i.e., sensor) not only considers its own features, but also performs weighted aggregation according to the features of its neighbor nodes, and the weight is automatically learned by the attention mechanism. The GAT calculates the weight between nodes through a multi-head attention mechanism. Specifically, for each node i, the attention coefficient between it and all adjacent nodes j is calculated , which reflects the influence degree of node j on node i. The calculation formula is as follows:

[0044]

[0045] where is the learning parameter of the attention mechanism, W is a linear transformation matrix, , and are the feature vectors of nodes i, j and v, respectively, Ni represents the adjacent node set of node i, represents the vector splicing operation.

[0046] Finally, the spatio-temporal correlation between nodes can be obtained after GAT processing.

[0047] S22, according to the attention coefficient between nodes in the spatio-temporal correlation model and the information of the adjacent nodes of each sensor node, determine the membership value and the dominant value of each sensor node, and distinguish the dominant node, the membership node and the independent node in the sensor network.

[0048] Optionally, the specific operation of S22 can include the following steps S221-S222:

[0049] S221, according to the following formula (1), calculate the first-order membership value of the i-th sensor node in the sensor network, and according to the following formula (2), calculate the first-order dominant value of the i-th sensor node:

[0050] (1)

[0051] (2)

[0052] wherein, represents the first-order membership value of the i-th sensor node in the sensor network, represents the attention coefficient between the i-th sensor node and the j-th sensor node in the sensor network, i is not equal to j, and N represents the total number of first-order adjacent nodes of the i-th sensor node, a first-order dominance value of the i-th sensor node in the sensor network, an attention coefficient between the j-th sensor node and the i-th sensor node.

[0053] In an implementable embodiment, the attention coefficient between nodes is calculated to determine which nodes play a key role in structural health monitoring. The selection of the dominant nodes is based on the degree of influence of the surrounding affiliated nodes, and these dominant nodes are usually located at key positions of the structure, such as the main beam connection or the deformation sensitive area of the structure.

[0054] The dominant node distribution matrix is calculated according to the attention coefficient between nodes. First, two parameter values are calculated for each node, namely the affiliation value and the dominance value, which represent the influence of the node on the adjacent nodes. If the node is affected by the changes of the surrounding nodes and causes fluctuations in its own characteristics, it has a larger affiliation; if the changes in the characteristics of a node easily cause fluctuations in the characteristics of the surrounding nodes, it has a larger dominance; in addition to the above two cases, if the changes in the characteristics of a node do not easily cause fluctuations in the characteristics of the surrounding nodes, and the node is not affected by the changes of the surrounding nodes to cause fluctuations in its own characteristics, it has a larger independence. For the first-order neighbors of the i-th node, the calculation methods of the two attributes are as follows.

[0055] It should be noted that the above steps calculate the affiliation value and the dominance value of the i-th node for the first-order neighbors. In order to further expand the adjustable range of the node network granularity, when the affiliation value or the dominance value of the i-th node exceeds a certain threshold , the calculation order of the neighbors of the i-th node is expanded to M order until its affiliation value and dominance value are both lower than the threshold , and the calculation methods are as follows:

[0056] (3)

[0057] (4)

[0058] After the above calculation, the affiliation value, the dominance value and the expandable neighbor level of each node can be obtained. The expansion neighbor level M of the dominance is the maximum granularity that the node can cover, and the nodes within the coverage range can be represented by the characteristics of the node. Correspondingly, the level M of the affiliation represents the maximum granularity that the node can be represented by the nearby nodes. It should be noted that by changing the threshold The coverage of the node granularity can be artificially intervened, that is, the model allows prior knowledge to intervene in the adjustment of the size of the granularity. The invocation of the dominance in different granularity levels can achieve that the nodes of the granularity can approximately represent the neighboring nodes with the granularity membership in the coverage of the nodes of the granularity, and the nodes that cannot reach the granularity remain to be characterized by the original value, that is, the construction of the multi-granularity perception map relevance layer can be realized.

[0059] S222, if the first membership value of the i-th sensor node is greater than or equal to the first attribute threshold, the i-th sensor node is determined as a membership node; if the first dominance value of the i-th sensor node is greater than or equal to the second attribute threshold, the i-th sensor node is determined as a dominant node; if the first membership value of the i-th sensor node is less than the first attribute threshold, and the first dominance value is less than the second attribute threshold, the i-th sensor node is determined as an independent node.

[0060] S3, input the time feature and the space feature of the space-time relevance model into the conditional variational autoencoder CVAE model to determine the risk coefficient of each sensor node.

[0061] Optionally, the training process of the conditional variational autoencoder CVAE model comprises:

[0062] Obtaining a training sample, the training sample comprising sample input data related to the infrastructure and label information corresponding to the sample input data;

[0063] Inputting the sample input data into the CVAE model to be trained to obtain the mean and variance corresponding to the sample input data, determining a predicted latent variable according to the mean and variance, comparing the predicted latent variable with the label information corresponding to the sample input data and calculating a loss function, adjusting the parameters in the CVAE model to be trained, iteratively executing until the loss function converges, stopping training, and obtaining the trained conditional variational autoencoder CVAE model.

[0064] In a feasible implementation, the loss function of the CVAE consists of two parts: reconstruction error and KL divergence. These two parts together constitute the objective function of the CVAE, aiming to minimize the difference between the input data and its reconstructed version, while ensuring that the distribution of the latent variable is close to the prior distribution.

[0065] is the probability distribution of reconstructing the input data from the latent variable z and the conditional label y, then the reconstruction error can be expressed as:

[0066]

[0067] wherein, is the encoder distribution.

[0068] The KL divergence is used to measure the difference between the posterior distribution of the encoder output and the prior distribution. It is generally desirable for the posterior distribution to be as close as possible to the prior distribution, which helps to maintain good properties of the latent space, such as smoothness and continuity. The KL divergence can be expressed as:

[0069]

[0070] where, and are the mean and standard deviation of the latent variable z output by the encoder, respectively.

[0071] Optionally, the specific operation of S3 can include steps S31-S32 as follows:

[0072] S31, obtaining time series data according to the time feature and the space feature of the space-time correlation model, and generating input information in a preset format according to the time series data; the preset format is (sensor id, time, data type);

[0073] S32, inputting the input information into the trained conditional variational autoencoder CVAE model, and determining the risk coefficient of the corresponding sensor node according to the output latent variable; wherein the latent variable is used to represent the mechanical properties, material strength properties and dynamic characteristics of the infrastructure in different dimensions.

[0074] Optionally, the specific operation of S32 can include steps S321-S324 as follows:

[0075] S321, using the conditional variational autoencoder to extract features and reduce dimensions of the multi-source heterogeneous sensor data, to obtain a node risk index for representing the health status of the sensor node;

[0076] S322, using the latent variable calculation method of the conditional variational autoencoder, in combination with the prior knowledge of the physical properties of the infrastructure, to extract main structure health state influencing factors;

[0077] S323, taking the output mean and variance of the variational autoencoder as a quantitative indicator of the node health state, and determining the spatial distribution of the node risk index by constructing a normal distribution model;

[0078] S324, calculating the risk coefficient z of each node, and using the Hellinger distance to measure the difference between the node risk index and the risk index probability distribution under ideal conditions, to evaluate the health status of the node.

[0079] In an implementable embodiment, the pathological manifestation layer converts the spatiotemporal features extracted by the connection structure layer into a visual representation of the structural health state using a health assessment algorithm. The embodiment of the present application adopts a conditional variational autoencoder (CVAE) to fuse heterogeneous data, which can map different types of sensor data to a common low-dimensional space to extract key features reflecting the structural health state. By training the CVAE model, a risk index for each node can be obtained to quantify the health state of the node.

[0080] The time series data received by the sensor is input in the format (sensor id, time, data type), and then the data is segmented for processing as needed to obtain a data sequence input as a three-dimensional tensor, where n is the number of sampling points contained in each time segment, d represents different sensor nodes, and represents different sensors. The health state assessment part only needs to determine a three-dimensional risk index based on the input data x to represent the overall safety state of the entire infrastructure at this moment, which is obtained by averaging the safety state of each node.

[0081] The method of conditional variational autoencoder is used to extract and reduce the dimension of multi-source heterogeneous sensor information, and the node risk index obtained by dimension reduction is used to represent the health state of the sensor node. The conditional variational autoencoder is improved from the autoencoder, and the principle of the autoencoder is briefly introduced as follows:

[0082] The autoencoder is a generative model that converts high-dimensional original information into low-dimensional latent variables through an encoder to save key information, and then restores the latent variables to the original information through a decoder; in the conversion process, in order to ensure the continuity of the generated latent variables, the model adds additional constraints to the latent variables, such as the Gaussian distribution of the latent variables in the generation. From the perspective of probability theory, the generation process of VAE is as follows: a series of latent variables z is generated from the prior distribution , and the generated result is generated from the generation distribution , where z is subject to , is subject to . Generally, the parameter estimation of the directed graph model is difficult, and the VAE uses stochastic gradient variational Bayes to effectively evaluate the accuracy of the parameters. Specifically, the variational lower bound is used as a substitute objective function, and the variational lower bound is as follows:

[0083]

[0084] In VAE, the encoder model adopts an approximate distribution to estimate the accurate actual posterior probability In both the encoder and the decoder, a multi-layer perceptron is used as the network structure for training. Since the first term KL divergence of the variational lower bound can be marginalized, but the second term cannot, the second term is approximated by the approximate distribution The way of sampling the latent variable z is obtained, and the actual VAE objective function is transformed into the following formula:

[0085]

[0086] where, The encoding distribution is reparameterized by a deterministic differentiable function g, whose parameters are input x and noise variable This makes the error can be backpropagated through the Gaussian latent variable, so that the VAE can be effectively trained using the stochastic gradient descent method.

[0087] VAE is an unsupervised clustering method with certain feature extraction and dimension reduction capability, but the distinction between different categories is not obvious enough when dimension reduction, so some label information is added to assist the model to distinguish. The implementation process of conditional variational autoencoder CVAE is as follows: for a given input x, the latent variable z is obtained by the prior distribution , and the output y is obtained by the distribution The training objective of CVAE is to maximize the conditional probability, and the treatment of the variational lower bound is the same as VAE, and its variational lower bound is:

[0088]

[0089] The empirical lower bound is:

[0090]

[0091] where, L is the number of samples.

[0092] Conditional variational autoencoder is usually used to process image classification, image target extraction and other tasks, while in this paper, the goal is to extract the features of the input time period data and map them to a new feature dimension with smooth transition. Therefore, on the basis of the original data , the feature label information y is added, which is obtained from the prior knowledge of the physical characteristics of the infrastructure, and y and the latent variable z have the same dimension, representing the mechanical properties, material strength properties and dynamic characteristics of the structure in different dimensions.

[0093] The mechanical property is mainly related to the supporting force, the tensile force and the torsional force, and the mutation of the mechanical property can reflect the physical properties such as structure cracking and internal mechanical damage; the material strength property is mainly related to the information of water leakage, temperature and humidity, and can reflect the slow change of the material itself under long-term environmental change; the dynamic characteristic is mainly related to the information such as the speed and acceleration of the structure displacement and torsion, and the large-scale change of the dynamic characteristic can easily cause damage to the structure safety of the infrastructure. The characteristic label information y acts as a variation condition on the CVAE model, guides the mapping of the model input x to the label direction, and achieves the purpose of multi-source data fusion feature extraction.

[0094] The quantification of the risk coefficient includes the health index and the confidence information of the sensor node in the three dimensions of the mechanical property, the material strength property and the dynamic characteristic.

[0095] The hidden variable is a three-row one-column feature vector, and the values of the first, second and third rows represent the health index in the three dimensions of the mechanical property, the material strength property and the dynamic characteristic.

[0096] It should be noted that the hidden variable z is obtained by sampling the mean and the variance of the model input x obtained by the encoder, and the hidden feature variable reflecting the input data x should be the mean output by the encoder, and the same as the model label y has the same dimension. After each sensor unit collects data, the mean and the variance are calculated locally by the encoder of the CVAE, and only the mean and variance information need to be transmitted to the processing unit, so that the original data can be restored in the processing unit through the decoder. This method can significantly reduce the frequency and total amount of data transmission, and improve the security of information transmission.

[0097] S4, according to the risk coefficient of each sensor node, visualizing the structure health state of the infrastructure, and adjusting the infrastructure monitoring work of the sensor network according to the risk coefficient of each sensor node and the properties of each sensor node.

[0098] Optionally, the specific operation of S4 can be as follows:

[0099] When the risk coefficient of each sensor node in the sensor network is less than the preset risk threshold, it is judged that all regions of the infrastructure at the current time are in a safe state, the activation state of the dominant node and the independent node is maintained, and the subordinate node is in a dormant state;

[0100] When the danger coefficient of some sensor nodes in the sensor network is greater than or equal to the preset danger threshold, it is judged that part of the infrastructure area is in a dangerous state at the current moment, and the subordinate nodes dominated by the dominant node in the dangerous state are activated, and the independent nodes in the dangerous state are kept activated.

[0101] In one feasible implementation, this embodiment of the present invention proposes a multi-granularity perception map adjustment strategy to adapt to monitoring needs under varying security conditions. In a good security situation, most nodes (referred to as slave nodes) remain dormant, with only the master and independent nodes remaining active, thereby reducing system energy consumption. When structural health deterioration is detected, the system automatically awakens more nodes in the affected area, increasing monitoring frequency and density to provide more detailed health status information. This approach not only improves monitoring flexibility and adaptability but also effectively extends the lifecycle of the sensor network.

[0102] In order to further illustrate the effect of the method of the present invention, Figure 3 (a)-(d) are schematic diagrams of the effects presented by the pathological manifestation layer of an application example. In this example, 20 consecutive point data in the tunnel are selected. Since the sensors are arranged in a ring in the tunnel, the sensors in the same ring interval can be expanded into a plane for analysis. There are six sensor points in each ring interval. 120 sensor points are selected to analyze the perception map granularity adjustment strategy. First, the data collected by the tunnel structure monitoring wireless sensor network (WSN) is input into the connection structure layer to construct a dimensionally independent spatial feature model to obtain the spatial correlation information between sensor nodes, and the nodes with first-order and second-order dominance are obtained by calculation to form a multi-granularity key node map, as shown in FIG. Figure 3(a) shows that there are three nodes with second-order dominance, whose coordinates are [3, 2], [2, 13], [4, 18], these nodes have high dominance to the nodes in the two adjacent grids near them, that is, the data changes of the sensor points in the two grids will be reflected in the data changes collected by the nodes, such nodes are usually located at key positions such as main beams and inter-segment joints, and are also the focus of observation in the traditional monitoring field, which shows the consistency of the method proposed in the embodiment of the application with the actual situation; there are six nodes with first-order dominance, whose coordinates are [1, 6], [5, 7], [5, 9], [1, 11], [0, 19], [3, 17], such nodes only have high dominance to the nodes in the adjacent grids near them, and correspond to some relatively marginal but also important points in the actual scene, such as the joint between the tunnel wall and the ground; except for the nodes with second-order dominance and the nodes in the two adjacent grids near them, and the nodes with first-order dominance and the nodes in the adjacent grid near them, all the other nodes are independent nodes; on the structure layer, the health state of the structure is evaluated, and the danger coefficients of the nodes are calculated, Figure 3 (b), Figure 3 (c) and Figure 3 (d) respectively show that the entire tunnel section is in a relatively safe condition, and the danger coefficient gradually increases on the right side of the section, the color of the grid point reflects the danger coefficient, and the color gradually changes from green to dark red as the danger coefficient increases. In the safe condition, all nodes with second-order dominance can cover a large range of adjacent nodes near them, and nodes with first-order dominance can cover nodes in a small adjacent range near them, at this time, the covered nodes are in a dormant state, the health state of the dominant node represents the entire covered area, and the energy consumption under normal conditions can be reduced; when the danger coefficient increases to a certain extent, the second-order dominant nodes in the area where the danger coefficient increases reduce to cover only their first-order adjacent range, the nodes far away from the dominant nodes are awakened, and the conditions of the positions are independently monitored, and the monitoring of the area is strengthened; when some regional construction or geological changes near the tunnel and the like occur, the danger coefficient continuously increases to a high level, all the dormant nodes on the right side are awakened, and the monitoring of the area is strengthened, such as Figure 3 (d), all the grid points on the right side calculate their respective danger coefficients, and the danger coefficients between the points are different, while the left side of the relatively safe area still maintains a high dominant node coverage range and maintains a low-energy-consumption working state.

[0103] The embodiment of the present application adopts a time series expansion convolution network and a dimension isolated graph attention network to construct spatio-temporal correlation, adopts a conditional variational autoencoder for data fusion, and simultaneously proposes a multi-granularity perception map adjustment strategy to adapt to monitoring requirements in different safety conditions, so that the energy saving effect is achieved.

[0104] Figure 4 is a kind of infrastructure intelligent monitoring system block diagram of skin structure according to an exemplary embodiment, which is used for infrastructure intelligent monitoring method of skin structure.The system includes data base 410, connecting structure layer 420 and pathological performance layer 430. Figure 4 , wherein:

[0105] The data base layer 410 is used to obtain infrastructure data of a plurality of sensor nodes in a sensor network;

[0106] The connecting structure layer 420 is used to construct a spatio-temporal correlation model according to the infrastructure data of the plurality of sensor nodes, and distinguish the properties of each sensor node in the sensor network according to the spatio-temporal correlation model;

[0107] The pathological performance layer 430 is used to input the time characteristics and spatial characteristics of the spatio-temporal correlation model into a conditional variational autoencoder CVAE model, determine the risk coefficient of each sensor node, visualize the structural health state of the infrastructure according to the risk coefficient of each sensor node, and adjust the infrastructure monitoring work of the sensor network according to the risk coefficient of each sensor node and the properties of each sensor node.

[0108] Figure 5 is a structural schematic diagram of an infrastructure intelligent monitoring device of skin structure provided by the embodiment of the present application, as Figure 5 shown, the infrastructure intelligent monitoring device of skin structure can include the infrastructure intelligent monitoring system of skin structure shown in the above Figure 4 Optionally, the infrastructure intelligent monitoring device 510 of skin structure can include the first processor 2001.

[0109] Optionally, the infrastructure intelligent monitoring device 510 of skin structure can further include the memory 2002 and the transceiver 2003.

[0110] The first processor 2001, the memory 2002 and the transceiver 2003 may be connected via a communication bus.

[0111] The following combination Figure 5 The components of the skin-like infrastructure intelligent monitoring device 510 are described in detail:

[0112] The first processor 2001 is the control center of the skin-like infrastructure intelligent monitoring device 510 and can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs) or one or more field programmable gate arrays (FPGAs).

[0113] Optionally, the first processor 2001 can execute various functions of the skin-like infrastructure intelligent monitoring device 510 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0114] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 5 CPU0 and CPU1 are shown in FIG.

[0115] In a specific implementation, as an embodiment, the skin-like structure infrastructure intelligent monitoring device 510 may also include multiple processors, such as Figure 5 1 and 2. The first processor 2001 and the second processor 2004 are shown in FIG. Each of these processors can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). A processor herein can refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0116] The memory 2002 is used to store the software program for executing the solution of the present invention, and is controlled by the first processor 2001 for execution. The specific implementation method can refer to the above method embodiment and will not be repeated here.

[0117] Optionally, the memory 2002 can be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disk storage, a magnetic disk storage or other magnetic storage devices, or any other medium capable of storing desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to this. The memory 2002 can be integrated with the first processor 2001 or exist independently and be coupled to the first processor 2001 through the interface circuit (not shown in the figure) of the skin-structure infrastructure intelligent monitoring device 510, and the embodiments of the present application do not make specific limitations here. Figure 5

[0118] The transceiver 2003 is configured to communicate with a network device or a terminal device.

[0119] Optionally, the transceiver 2003 can include a receiver and a transmitter (not shown separately in the figure). The receiver is configured to implement the receiving function, and the transmitter is configured to implement the transmitting function. Figure 5

[0120] Optionally, the transceiver 2003 can be integrated with the first processor 2001 or exist independently and be coupled to the first processor 2001 through the interface circuit (not shown in the figure) of the skin-structure infrastructure intelligent monitoring device 510, and the embodiments of the present application do not make specific limitations here. Figure 5

[0121] It should be noted that the structure of the skin-structure infrastructure intelligent monitoring device 510 shown in the figure does not constitute a limitation on the router, and the actual knowledge structure recognition device can include more or fewer components than those shown in the figure, or combine certain components, or different component arrangements. Figure 5 In addition, the technical effects of the skin-structure infrastructure intelligent monitoring device 510 can refer to the technical effects of the skin-structure infrastructure intelligent monitoring method described in the above method embodiments, which will not be repeated here.

[0122]

[0123] ​​​​It is to be understood that the first processor 2001 in the embodiments of the present application can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic components, discrete hardware components, etc. The general-purpose processor can be a microprocessor or can also be any conventional processor.

[0124] It is also to be understood that the memory in the embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memory. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically EPROM (EEPROM) or a flash memory. The volatile memory can be a random access memory (RAM) used as an external cache. By way of example but not limitation, many forms of random access memory (RAM) are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchlink dynamic random access memory (SLDRAM) and direct rambus RAM (DR RAM).

[0125] The above-described embodiments can be implemented in whole or in part by software, hardware (such as a circuit), firmware, or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable systems. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center through a wired (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0126] It should be understood that the term "and / or" herein merely describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the following three cases: A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. In addition, the character " / " herein generally represents that the associated objects before and after it are in an "or" relationship, but it can also represent an "and / or" relationship, which can be understood according to the context before and after it.

[0127] In the present application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following" or the like means any combination of the items, including any combination of single or multiple items. For example, at least one of a, b, or c can represent a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

[0128] It should be understood that in various embodiments of the present application, the size of the sequence number of the above-described processes does not mean the order of execution, and the execution order of the processes should be determined according to their functions and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0129] Those skilled in the art can clearly understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed 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 the present application.

[0130] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the devices, systems and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0131] In several embodiments provided by the present application, it should be understood that the disclosed devices, systems and methods can be implemented in other ways. For example, the system embodiments described above are merely schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, systems or units, which can be electrical, mechanical or other forms.

[0132] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0133] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically independently, or two or more units can be integrated into one unit.

[0134] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the present application that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of software products. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0135] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for intelligent monitoring of infrastructure with skin-like structures, characterized in that: The method comprises: S1, obtain infrastructure data of multiple sensor nodes in the sensor network; S2. constructing a spatiotemporal correlation model based on the infrastructure data of the plurality of sensor nodes, and distinguishing properties of each sensor node in the sensor network based on the spatiotemporal correlation model; S3, input the temporal features and spatial features of the spatiotemporal correlation model into the conditional variational autoencoder (CVAE) model to determine the risk coefficient of each sensor node; S4. Visualize the structural health status of the infrastructure based on the risk factor of each sensor node, and adjust the infrastructure monitoring work of the sensor network based on the risk factor of each sensor node and the properties of each sensor node; The step S2 constructs a spatiotemporal correlation model based on the infrastructure data of the plurality of sensor nodes, and distinguishes the properties of each sensor node in the sensor network based on the spatiotemporal correlation model, including: S21. Constructing a spatiotemporal correlation model based on the infrastructure data of the multiple sensor nodes by combining a time series dilated convolutional network and a dimensionally isolated graph attention network; S22, determining a membership value and a dominance value of each sensor node based on the attention coefficient between nodes in the spatiotemporal correlation model and information about adjacent nodes of each sensor node, and distinguishing dominant nodes, subordinate nodes, and independent nodes in the sensor network; The step S21 constructs a spatiotemporal correlation model based on the data of the multiple sensor nodes by combining a time series dilated convolutional network with a dimensionally isolated graph attention network, including: generating time series data according to the data of the plurality of sensor nodes, inputting the time series data into a time series dilated convolutional network, and capturing time feature vectors at different time scales through multiple causal convolutional layers in the time series dilated convolutional network; Combining the temporal feature vector with the original spatial data in the data of the multiple sensor nodes to obtain a new feature matrix, inputting the new feature matrix into a dimensionally isolated graph attention network, calculating the weights between nodes based on the multi-head attention mechanism in the graph attention network, and calculating the attention coefficient between each node and all adjacent nodes, thereby obtaining the spatiotemporal correlation between the nodes; The step S3 inputs the temporal features and spatial features of the spatiotemporal correlation model into the conditional variational autoencoder (CVAE) model to determine the risk coefficient of each sensor node, including: S31. Obtain time series data based on the temporal and spatial features of the spatiotemporal correlation model, and generate input information in a preset format based on the time series data; the preset format is (sensor ID, time, data type); S32. Input the input information into the trained conditional variational autoencoder (CVAE) model, and determine the risk coefficient of the corresponding sensor node based on the output latent variables; wherein the latent variables are used to characterize the mechanical properties, material strength properties, and dynamic characteristics of the infrastructure in different dimensions; The step S4 adjusts the infrastructure monitoring work of the sensor network according to the risk factor of each sensor node and the properties of each sensor node, including: When the risk factor of each sensor node in the sensor network is less than the preset risk threshold, it is determined that all areas of the infrastructure at the current moment are in a safe state, the dominant node and independent nodes are kept activated, and the subordinate nodes are put into a dormant state; When the danger coefficient of some sensor nodes in the sensor network is greater than or equal to the preset danger threshold, it is judged that part of the infrastructure area is in a dangerous state at the current moment, and the subordinate nodes dominated by the dominant node in the dangerous state are activated, and the independent nodes in the dangerous state are kept activated.

2. The method for intelligent monitoring of infrastructure of skin-like structure according to claim 1, characterized in that: The step S22, determining the membership value and the dominance value of each sensor node based on the attention coefficient between nodes in the spatiotemporal correlation model and the information of the adjacent nodes of each sensor node, and distinguishing dominant nodes, subordinate nodes, and independent nodes in the sensor network, includes: S221. Calculate the first-order membership value of the i-th sensor node in the sensor network according to the following formula (1), and calculate the first-order dominance value of the i-th sensor node according to the following formula (2): （1） （2） in, represents the first-order membership value of the i-th sensor node in the sensor network, represents the attention coefficient between the i-th sensor node and the j-th sensor node in the sensor network, i is not equal to j, N represents the total number of first-order neighboring nodes of the i-th sensor node, represents the first-order dominance value of the i-th sensor node in the sensor network, represents the attention coefficient between the j-th sensor node and the i-th sensor node; S222. If the first-order membership value of the i-th sensor node is greater than or equal to the first attribute threshold, the i-th sensor node is determined as a member node; if the first-order dominance value of the i-th sensor node is greater than or equal to the second attribute threshold, the i-th sensor node is determined as a dominant node; if the first-order membership value of the i-th sensor node is less than the first attribute threshold and the first-order dominance value is less than the second attribute threshold, the i-th sensor node is determined as an independent node.

3. The method for intelligent monitoring of infrastructure of skin-like structure according to claim 1, characterized in that: The training process of the conditional variational autoencoder CVAE model includes: Acquire a training sample, where the training sample includes sample input data related to the infrastructure and label information corresponding to the sample input data; The sample input data is input into the CVAE model to be trained, and the mean and variance corresponding to the sample input data are obtained. The latent variables are determined according to the mean and variance. The latent variables are compared with the label information corresponding to the sample input data and the loss function is calculated. The parameters in the CVAE model to be trained are adjusted, and the iterative execution is performed until the loss function converges. The training is stopped to obtain a trained conditional variational autoencoder CVAE model.

4. An intelligent monitoring system for infrastructure with a skin-like structure, wherein the intelligent monitoring system for infrastructure with a skin-like structure is used to implement the intelligent monitoring method for infrastructure with a skin-like structure as claimed in any one of claims 1 to 3, characterized in that: The system includes a data base layer, a connection structure layer, and a pathology presentation layer, wherein: The data base layer is used to obtain infrastructure data of multiple sensor nodes in the sensor network; The connection structure layer is used to construct a spatiotemporal correlation model based on the infrastructure data of the plurality of sensor nodes, and distinguish the properties of each sensor node in the sensor network based on the spatiotemporal correlation model; The pathology representation layer is used to input the temporal and spatial features of the spatiotemporal correlation model into the conditional variational autoencoder (CVAE) model to determine the risk coefficient of each sensor node; based on the risk coefficient of each sensor node, the structural health status of the infrastructure is visualized, and the infrastructure monitoring work of the sensor network is adjusted based on the risk coefficient of each sensor node and the properties of each sensor node.

5. An intelligent monitoring device for infrastructure with a skin-like structure, characterized in that: The skin-like structure infrastructure intelligent monitoring equipment includes: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 3 is implemented.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, which can be called by a processor to execute the method according to any one of claims 1 to 3.

Images