Ground bearing capacity intelligent identification method and system based on sensor network

Abstract

The application discloses a ground bearing capacity intelligent identification method and system based on a sensing network, relates to the technical field of geotechnical engineering detection, and comprises the following steps: a mixed perception network is constructed, and multi-source heterogeneous monitoring data of a foundation is acquired; a unified space-time datum of the multi-source heterogeneous monitoring data is established, and a space-time database is constructed; data in the space-time database is subjected to homogeneous fusion and heterogeneous fusion, comprehensive judgment of a ground bearing capacity state is carried out based on a fusion result, and a bearing capacity distribution field of the ground bearing capacity is generated. The application solves the technical problems in traditional ground bearing capacity identification, such as difficulty in giving consideration to coverage range and precision during data acquisition, lack of effective integration of multi-source data, and lagged response to abnormal conditions, achieves accurate identification of the ground bearing capacity, global coverage and long-term reliable optimization, and meets the technical effect of engineering safety monitoring and control of the foundation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geotechnical engineering detection, and in particular to a foundation bearing capacity intelligent identification method and system based on a sensor network. BACKGROUND

[0002] The foundation bearing capacity is a core index of engineering safety, and its accurate identification is crucial to the stability of buildings. The existing technologies rely on static load tests, single sensor monitoring and simple data statistics, which play a role in conventional foundation detection. However, with the increasing demand for foundation identification accuracy, these technologies have limitations in complex foundation scenarios: multi-source heterogeneous monitoring data lack a unified spatio-temporal reference, and homogeneous and heterogeneous fusion data processing has not been carried out, resulting in fragmented data, incomplete coverage, and inaccurate bearing capacity identification results, which cannot meet the needs of accurate foundation bearing capacity identification and engineering control. SUMMARY

[0003] The present application provides a foundation bearing capacity intelligent identification method and system based on a sensor network, which solves the technical problems of traditional foundation bearing capacity identification, such as the difficulty in balancing coverage and accuracy of data collection, the lack of effective integration of multi-source data, and the lag in response to abnormal situations.

[0004] In a first aspect, the present application provides a foundation bearing capacity intelligent identification method based on a sensor network, which includes: constructing a hybrid perception network to obtain multi-source heterogeneous monitoring data of the foundation; establishing a unified spatio-temporal reference for the multi-source heterogeneous monitoring data and constructing a spatio-temporal database; performing homogeneous fusion and heterogeneous fusion on the data in the spatio-temporal database, making a comprehensive judgment on the foundation bearing capacity state based on the fusion results, and generating a bearing capacity distribution field of the foundation bearing capacity.

[0005] In a second aspect, the present application provides a foundation bearing capacity intelligent identification system based on a sensor network, which includes: a monitoring data acquisition module for constructing a hybrid perception network to obtain multi-source heterogeneous monitoring data of the foundation; a database construction module for establishing a unified spatio-temporal reference for the multi-source heterogeneous monitoring data and constructing a spatio-temporal database; and a distribution field construction module for performing homogeneous fusion and heterogeneous fusion on the data in the spatio-temporal database, making a comprehensive judgment on the foundation bearing capacity state based on the fusion results, and generating a bearing capacity distribution field of the foundation bearing capacity.

[0006] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0007] The application obtains multi-source monitoring data by constructing a perception network suitable for the foundation scene, generates a bearing capacity distribution field through unified space-time reference and multi-dimensional data fusion processing, and optimizes the identification model through on-site verification of abnormal areas and data feedback, so that the global precise identification and dynamic iterative optimization of the foundation bearing capacity are realized, the foundation bearing capacity identification result is more comprehensive and reliable, the needs of engineering for foundation safety monitoring and control are met, the technical effects of precise identification, global coverage and long-term reliable optimization of the foundation bearing capacity are achieved, and the needs of engineering for foundation safety monitoring and control are met. BRIEF DESCRIPTION OF DRAWINGS

[0008] 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.

[0009] Figure 1 is a flowchart of the foundation bearing capacity intelligent identification method provided by the embodiment of the present application based on a sensing network.

[0010] Figure 2 is a structural schematic diagram of the foundation bearing capacity intelligent identification system provided by the embodiment of the present application based on a sensing network.

[0011] The drawings are described as follows: monitoring data acquisition module 1, database construction module 2, and distribution field construction module 3. DETAILED DESCRIPTION

[0012] The present application provides a foundation bearing capacity intelligent identification method and system based on a sensing network, which solves the technical problems of traditional foundation bearing capacity identification, such as difficulty in balancing coverage range and accuracy in data collection, lack of effective integration of multi-source data, and slow response to abnormal situations.

[0013] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0014] It should be noted that the terms "first", "second", etc. in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or modules not clearly listed or inherent to these processes, methods, products or devices.

[0015] Embodiment one, as shown in the figure, the method for identifying the bearing capacity of the foundation based on the sensing network, wherein the method comprises: Figure 1

[0016] Constructing a hybrid sensing network to obtain multi-source heterogeneous monitoring data of the foundation.

[0017] Specifically, the hybrid sensing network comprises sparse fixed first sensor nodes and second distributed sensing units covering key areas, wherein the accuracy of the first sensor nodes is higher than that of the second distributed sensing units.

[0018] First, based on the prior geological model, finite element numerical simulation is carried out, and through model parameter sensitivity analysis, key areas with sensitivity higher than the preset sensitivity to bearing capacity inversion are screened out. Then in the key areas, the optimal layout position is determined according to the preset number of first sensor nodes and the deployment is completed. Finally, in the area between any two adjacent first sensor nodes, the second distributed sensing unit is deployed in a planar grid covering manner. This step is described in detail in the subsequent content.

[0019] Then, according to the foundation monitoring requirements, data acquisition parameters are set for the first sensor nodes and the second distributed sensing units in the hybrid sensing network. Among them, the sampling frequency of the first sensor nodes is set to 1 time per minute, and the sampling frequency of the second distributed sensing units is consistent with that of the first sensor nodes. The acquisition time of both is set to 24 hours continuously. After the parameter setting is completed, the network control module sends a start instruction to the two types of sensors to ensure that the first sensor nodes stably collect high-precision physical quantity data in the key areas of the foundation, and the second distributed sensing units stably collect physical quantity data in the whole domain. Both types of sensors continuously output raw collected data according to the set parameters.

[0020] ​Subsequently, the raw data of the two types of sensors are synchronously collected. During the operation of the sensors, a unified clock synchronization module is configured for the hybrid perception network, and the same format of time stamp is added to each piece of collected data of the first sensor node and the second distributed sensing unit, so as to ensure the consistency of the two types of data in the time dimension. At the same time, the deployment position information of each sensor is recorded in real time during the data collection process, and the position information is associated with the collected data of the corresponding sensor, so that each piece of raw data can be clearly corresponded to the specific monitoring area of the ground station, so as to obtain the raw collected data of the two types of sensors which are time-synchronized and position-identified.

[0021] Then, for the synchronized raw collected data, the mean filtering method is used to process the continuously collected numerical data. Specifically, for each sensor, the numerical average value of the five data points is calculated, and the average value is used to replace the middle data point of the five data points, so as to reduce the fluctuation of the data caused by instantaneous interference. Subsequently, the data processed by the mean filtering method is compared with the preset reasonable numerical range. If the data exceeds the range, it is determined as abnormal data and is rejected, and finally the effective collected data without obvious fluctuation and abnormal value is obtained.

[0022] Finally, the preprocessed effective collected data is classified according to the sensor type. The effective collected data of the first sensor node is classified as high-precision monitoring data, and the effective collected data of the second distributed sensing unit is classified as global monitoring data. Within each type of data, the data is sorted in chronological order, and the corresponding position information of each data is retained, forming a data set with the core structure of sensor type-collected time-monitoring position-effective data. Through this sorting process, the multi-source heterogeneous monitoring data of the ground station containing different precision, different monitoring range and different physical quantity are finally obtained.

[0023] A unified space-time reference of the multi-source heterogeneous monitoring data is established, and a space-time database is constructed.

[0024] Optionally, first, a unified time reference source is determined, and the network time protocol (NTP) is selected as the standard time reference. The clock synchronization module of all sensors in the hybrid perception network is connected with the NTP server, and time synchronization is performed every 30 seconds, so as to ensure that the time stamps generated by the collected data of each sensor are based on the same time standard. For the historical data that has been collected and has a time deviation, the average offset of the time stamp of each sensor and the NTP standard time is calculated, and the offset is used to correct the time stamp of the historical data, so that the time information of all multi-source heterogeneous monitoring data is consistent.

[0025] Then a unified spatial reference is established, and the CGCS2000 geodetic coordinate system is selected as the target spatial coordinate system. First, the original coordinate data of the deployment position of each sensor is obtained. If the original coordinate is a local coordinate system, for example, a self-defined coordinate system of the construction site, the conversion parameters between the local coordinate system and the geodetic coordinate system are obtained through measurement, including 3 translation parameters, 3 rotation parameters and 1 scale parameter. The original position coordinates of each sensor are converted to coordinate values in the geodetic coordinate system by using the seven-parameter coordinate conversion formula. Then the converted spatial coordinates are associated with the data collected by the corresponding sensors, so that each monitoring data carries spatial position information in the unified coordinate system, and the spatial information of the multi-source data is unified.

[0026] Finally, a space-time database is constructed. First, the table structure of the database is designed. The table includes a time field, which stores the unified timestamp; a space field, which stores the coordinate values in the unified coordinate system; a data type field, which indicates whether the data comes from the first sensor node or the second distributed sensor unit; a monitoring value field, which stores the preprocessed monitoring data; and a device identification field, which is associated with the corresponding sensor number. A database management system that supports space-time data management is selected, and a space-time data table is created according to the designed table structure. The multi-source heterogeneous monitoring data with unified space-time reference is imported into the data table according to the table structure fields one by one, and indexes are created for the time field and the space field to improve data query efficiency. After the import is completed, 10% of the data samples are randomly extracted to verify the time consistency, spatial accuracy and data-device identification association of the samples. After confirming that there is no error, the construction of the space-time database is completed.

[0027] The data in the space-time database is subjected to homogeneous fusion and heterogeneous fusion, and based on the fusion results, a comprehensive judgment of the state of the foundation bearing capacity is made to generate a bearing capacity distribution field of the foundation bearing capacity.

[0028] In the embodiments of the present application, the foundation bearing capacity refers to the maximum capacity of the foundation soil or rock layer to withstand the load of the upper building, structure and the like without excessive deformation or damage.

[0029] In one embodiment of the present application, first, for sensor data collecting the same physical quantity, homogeneous fusion is completed through message passing and aggregation operation to generate homogeneous node fusion features representing the local area pressure state. Then, a heterogeneous information graph is constructed based on the features, and a node-level fusion of the heterogeneous information graph is performed by using a heterogeneous graph attention network to obtain a site state embedding representation. Subsequently, the state embedding representation is input into a fully connected neural network, and the bearing capacity safety factor of each position is calculated by the network. Finally, in combination with the coordinates of all positions and the corresponding bearing capacity safety factors, a spatial interpolation algorithm is used to generate a foundation bearing capacity distribution field covering the entire monitoring area in a continuous grid form. This step is described in detail in the subsequent content.

[0030] Further, the method provided by the embodiments of the present application comprises:

[0031] The mixed perception network comprises sparse fixed first sensor nodes and second distributed sensing units covering key areas, and the accuracy of the first sensor nodes is higher than that of the second distributed sensing units.

[0032] Specifically, first, a priori geological data of a ground area is collected by a person skilled in the art to establish a priori geological model, and a finite element numerical simulation method is used to perform mechanical analysis on the model, and then parameter sensitivity analysis is carried out to identify key positions with high sensitivity to bearing capacity inversion, such as stress concentration areas, potential sliding surfaces and the like.

[0033] Then, in these key positions, high-precision point sensors are selected as first sensor nodes for sparse deployment according to a preset number of deployments. Such sensors have high accuracy and can provide accurate reference data at key positions as a basis for subsequent data calibration, which is a conventional deployment method of high-precision sensors at key measuring points in the prior art.

[0034] After that, in the area between any two adjacent first sensor nodes, low-cost distributed sensors are deployed as second distributed sensing units in a planar grid layout. For example, flexible pressure films based on conductive rubber, simple strain gauges based on string vibration or surface acoustic wave sensors are selected, and by laying a large number of these low-cost sensors, wide-area coverage of the key area is achieved, and the accuracy of the first sensor nodes is higher than that of the second distributed sensing units.

[0035] After the deployment of the first sensor nodes and the second distributed sensing units, a linear regression algorithm is used to calibrate and correct the data of the low-cost sensor network. The specific process is as follows: first, the monitoring data of the high-precision first sensor nodes and the surrounding adjacent second distributed sensing units in the same time period are collected synchronously to ensure that the two types of data correspond to the same ground monitoring conditions, such as the same load action and the same environmental conditions. Then, the collected raw data is preprocessed, and mean filtering is used to remove outliers in the data of the two types of sensors, such as sudden data caused by transient interference of the equipment, and valid monitoring data sets are retained.

[0036] Subsequently, the high-precision data of the first sensor node is taken as the dependent variable and as the true value reference, and the low-cost data of the corresponding second distributed sensing unit is taken as the independent variable to construct a linear regression model. The slope and intercept parameters of the model are calculated by the least square method to establish a mathematical mapping relationship between the low-cost data and the high-precision data. Finally, the real-time low-cost data collected by the second distributed sensing unit is substituted into the linear regression model to calculate the corrected monitoring data, realizing the calibration and correction of the low-cost sensor network data. At the same time, the same period data of the two types of sensors is re-collected every preset period, such as 7 days, to update the linear regression model parameters, ensuring the accuracy of the long-term correction results.

[0037] By determining the key area to deploy high-precision sensors and deploying low-cost sensors on a large scale between adjacent key areas, a hybrid sensing network is constructed, which achieves the effect of ensuring the data accuracy of key areas and realizing wide-area monitoring of ground bearing capacity, and provides accurate and comprehensive sensing data basis for intelligent identification of ground bearing capacity.

[0038] Further, the method provided by the embodiment of the application comprises:

[0039] Based on the prior geological model, finite element numerical simulation is performed, and through model parameter sensitivity analysis, a key area with a sensitivity greater than a preset sensitivity to bearing capacity inversion is identified. In the key area, the optimal deployment position of the first sensor node is determined according to the preset first sensor deployment quantity, and the first sensor node is deployed. A second distributed sensing unit is deployed by covering the area between any two adjacent first sensor nodes with a planar mesh.

[0040] Optionally, at the initial stage of constructing the hybrid sensing network, a prior geological model needs to be constructed first. Specifically, the prior geological model can be constructed by collecting geological survey data of the target ground area, such as obtaining drilling data, stratum distribution report, and basic information of rock and soil physical and mechanical parameters such as cohesion, internal friction angle, and elastic modulus. Then, the information is integrated by using a geological modeling software such as PLAXIS and FLAC3D to generate a prior geological model that can reflect the real geological conditions of the area.

[0041] Then, the prior geological model is analyzed by using the finite element numerical simulation method. The prior geological model is first discretized into a finite number of calculation units, and reasonable boundary conditions are set, such as fixed foundation bottom displacement, constrained lateral horizontal displacement, and load conditions for simulating the upper structure load that the foundation may bear in actual engineering. The stress, displacement and other mechanical response data of the model under different working conditions are obtained by solving the finite element calculation software. Subsequently, based on these mechanical response data, model parameter sensitivity analysis is carried out. The control variable method is used to achieve this, that is, keeping other parameters unchanged, changing the key parameters of the foundation rock and soil body one by one, such as cohesion and elastic modulus, and observing the influence degree of each parameter change on the bearing capacity inversion result. Then, a preset sensitivity threshold is set, which can be determined by referring to the experience value of similar projects in the industry. The areas with an influence degree greater than the threshold are marked as key areas with high sensitivity to bearing capacity inversion, such as stress concentration areas and potential slip surfaces.

[0042] After that, the first sensor node is deployed, and the grid priority method is used to determine the optimal deployment position. First, the key area is divided into several equal-sized monitoring grids, such as 0.5 meters x 0.5 meters. The maximum stress in each grid is calculated by finite element simulation results, and the higher the stress value, the higher the monitoring priority of the grid. Then, according to the preset number of first sensor nodes, the corresponding number of grid centers from high-priority grids are selected as candidate positions for the first sensor nodes, and combined with the actual situation on site, such as avoiding underground pipelines, obstacles, etc., the optimal deployment position is finally determined. After that, high-precision point sensors such as vibrating wire pressure sensors and fiber Bragg displacement sensors are selected and installed according to the determined position to ensure that the sensors are in close contact with the foundation rock and soil body to accurately collect mechanical data from the key area.

[0043] After the deployment of the first sensor node is completed, the second distributed sensing unit is deployed in the area between the two adjacent nodes. First, the orthogonal surface grid is used for division, and the grid size is determined according to the distance between the two adjacent first sensor nodes, for example, when the distance is 5 meters, the grid size can be set to 1 meter x 1 meter to ensure coverage without blind spots and generate uniform surface grids in the area between the two first sensor nodes. Then, low-cost distributed sensors such as conductive rubber flexible pressure film, string vibration type simple strain gauge and surface acoustic wave sensor are selected and laid according to the node position of each surface grid. During the laying process, the sensors are ensured to be in close contact with the foundation surface, and waterproof tape can be used to fix the edges to ensure that each grid node has a corresponding sensor, thereby achieving comprehensive coverage of the area between the adjacent first sensor nodes and completing the deployment of the second distributed sensing unit.

[0044] By constructing a prior geological model, simulating and analyzing to identify key areas, and deploying different precision sensors in steps, high-precision monitoring data of key areas are ensured, and the whole ground monitoring area is covered, thereby providing comprehensive and reliable sensor data support for subsequent accurate inversion of ground bearing capacity.

[0045] Further, the method provided in the embodiments of the application comprises:

[0046] For sensors of the same physical quantity, message passing and aggregation are performed to generate homogeneous node fusion features representing the pressure state of a local area after homogeneous fusion; a heterogeneous information graph is constructed based on the homogeneous node fusion features, and node-level fusion of the heterogeneous information graph is performed using a heterogeneous graph attention network to generate a site state embedding representation; the state embedding representation is input into a fully connected neural network, and the fully connected neural network calculates a bearing capacity safety factor for each position; and a continuous gridded bearing capacity distribution field covering the entire monitoring area is generated by a spatial interpolation algorithm based on the coordinates and bearing capacity safety factors of all positions.

[0047] Specifically, first, a homogeneous sensor graph is constructed for a sensor group collecting the same physical quantity, and each sensor is taken as a node of the graph, and the spatial distance between nodes is taken as the weight of the edge in the graph. Then, a neighborhood aggregation operation is performed on the homogeneous sensor graph using a graph convolution network, and finally, homogeneous node fusion features fused with spatial context information are output. This step is described in detail in the subsequent content.

[0048] Then, the homogeneous node fusion features representing different physical quantities are defined as different node types in the heterogeneous graph to construct a heterogeneous information graph. Then, a heterogeneous graph attention network is used to calculate the mutual influence weight between different physical quantity node types in the heterogeneous information graph. Finally, according to the obtained mutual influence weight, the features of different physical quantities at the same spatial position are fused into a unified comprehensive state feature vector, and then a site state embedding representation is obtained. This step is described in detail in the subsequent content.

[0049] Subsequently, a fully connected neural network is constructed and trained, and a three-layer network structure of input layer-hidden layer-output layer is adopted. The dimension of the input layer is set to the feature dimension of the site state embedding representation to ensure matching with the input data dimension; two hidden layers are set, the first hidden layer has 64 units, and the second hidden layer has 32 units, and ReLU is selected as the hidden layer activation function. The specific process is that the weighted input of the hidden layer neuron, i.e., the feature vector multiplied by the weight matrix plus the bias term, is substituted into the ReLU function, when the input value is greater than 0, the value is directly output, and when the input value is less than 0, 0 is output, thereby enhancing the nonlinear expression ability of the network; the output layer has only one unit, and the Linear activation function is selected to directly output a continuous numerical value to correspond to the bearing capacity safety factor.

[0050] The parameter initialization adopts the Xavier method, the initial value range of the weight matrix is determined by calculating the reciprocal of the square root of the input and output dimensions, the variance of the output signal of each layer is kept consistent, and the bias term is initialized to 0. The loss function selects the mean square error loss function, calculates the square difference average of the predicted safety factor of the network and the measured foundation bearing capacity safety factor (sample label); the optimizer selects Adam, sets the learning rate to 0.001, and iteratively updates the weights and bias terms through back propagation. In each iteration, first calculate the gradient of the loss function with respect to each parameter, then adjust the parameters according to the gradient and learning rate, until the loss function value does not decrease for 100 consecutive iterations and is less than the preset threshold 0.001, and the fully connected neural network training is completed.

[0051] After training, the site state embedding representation of each monitoring position is input into the fully connected neural network as a feature vector one by one. The site state embedding representation is first transmitted to the input layer, then transmitted to the first hidden layer through the input layer, multiplied by the trained weight matrix of the layer and added with the bias term, and then the weighted input is input into the ReLU function to generate the output feature of the first hidden layer. The output feature continues to be transmitted to the second hidden layer, and the process of weighting operation-bias addition- ReLU activation is repeated to obtain the output feature of the second hidden layer. Finally, the output feature of the second hidden layer is transmitted to the output layer, multiplied by the output layer weight and added with the bias, and then directly outputs a specific numerical value through the Linear activation function. The numerical value is the bearing capacity safety factor of the corresponding monitoring position. At the same time, the coordinates of each monitoring position under the unified spatial reference are retrieved, and the coordinates-bearing capacity safety factor are corresponded one by one to form a basic data pair set containing all monitoring positions.

[0052] Next, the inverse distance weighting method is used for spatial interpolation. Specifically, first, the coordinate boundary of the monitoring area is determined, the maximum and minimum values of the X and Y axes are obtained by traversing all monitoring position coordinates, and a uniform grid is divided according to the grid accuracy of 1 meter x 1 meter. The specific process is to start from the minimum value of the X axis, take a point every 1 meter, until the maximum value of the X axis, and the Y axis is the same. All intersection points of X and Y axes are grid nodes, which ensure that the grid completely covers the entire monitoring area. Then, using the above data pair set of coordinates-bearing capacity safety factor as known samples, interpolation calculation is performed for each grid node: first, calculate the Euclidean distance between the grid node and all known sample positions; second, take the reciprocal of each distance value as the weight of the corresponding sample safety factor, that is, the closer the distance, the larger the reciprocal, and the higher the weight; third, sum the results of the bearing capacity safety factor of all samples x weight, and then divide by the sum of all weights to obtain the bearing capacity safety factor of the grid node. Traverse all grid nodes and repeat the above interpolation steps to obtain the bearing capacity safety factor of each grid node. After obtaining the bearing capacity safety factor of each grid node, the coordinates of all grid nodes and the corresponding bearing capacity safety factors are integrated to form a continuous gridded data covering the entire monitoring area, i.e. the foundation bearing capacity distribution field.

[0053] By constructing a training fully connected neural network to calculate the safety factor of bearing capacity of each monitoring position, and combining the inverse distance weighted method for spatial interpolation, the conversion from the discrete site state embedding representation to the global continuous foundation bearing capacity distribution field is realized, and the effect of accurately presenting the foundation bearing capacity state of the entire monitoring area is achieved.

[0054] Further, the method provided in the embodiment of the application comprises:

[0055] For a sensor group of the same physical quantity, a homogeneous sensor graph is constructed, wherein the sensors are nodes, and the spatial distance between the nodes is the weight of the edge; a neighborhood aggregation operation is performed on the homogeneous sensor graph using a graph convolution network, and a homogeneous node fusion feature that fuses spatial context information is output.

[0056] Specifically, first, a sensor group collecting the same physical quantity is selected to form a sensor group, and the coordinate values of each sensor in the group under a unified spatial reference are obtained. For any two sensors, the spatial distance between the two sensors is calculated using the Euclidean distance formula, each sensor is taken as a node of a homogeneous sensor graph, and the calculated spatial distance is taken as the weight of the edge between the corresponding two nodes, and the construction of the homogeneous sensor graph is completed.

[0057] Then, a graph convolution network is constructed, and the specific steps are as follows: first, the network structure is determined, and 1 input layer, 2 hidden layers and 1 output layer are set, wherein the input layer dimension corresponds to the number of original features of the sensor, such as the pressure value of the pressure sensor, the feature dimension of the collection time difference, etc., 64 hidden units are set in the first hidden layer, 32 hidden units are set in the second hidden layer, and the output layer dimension matches the dimension requirement of the final homogeneous node fusion feature. Then, the network parameters are initialized, the convolution kernel weight and the bias term of each layer are initialized using the Xavier initialization method, so that the mean of the initialized parameters is close to 0 and the variance is uniform, avoiding gradient disappearance or explosion in subsequent training. Then, the activation function is determined, the ReLU activation function is selected for the hidden layer to enhance the non-linear expression ability of the network, and the Linear activation function is selected for the output layer to directly output the fusion feature vector. Finally, the loss function and the optimizer are configured, the mean square error loss function is used as the loss function to measure the difference between the fusion feature output by the network and the sample label (such as the local area pressure feature manually labeled), and the Adam optimizer is selected as the optimizer, the learning rate is set to 0.001, the network parameters are updated through back propagation iteration until the loss function value converges to a preset threshold, and the construction of the graph convolution network is completed.

[0058] For the constructed homogeneous sensor graph, a neighborhood aggregation operation is performed using the constructed graph convolution network. First, a neighborhood range threshold is set for each node, and other nodes with a spatial distance less than the threshold from the node are selected as neighborhood nodes. In setting the neighborhood range threshold, the average of the spatial distances between all adjacent sensors in the same physical quantity sensor group is calculated as a basis reference. Then, combined with the monitoring accuracy requirement, the neighborhood range threshold is set to 1.5-2 times the average distance, and the multiple selection can ensure that each node can cover a sufficient number of adjacent sensors to obtain rich spatial context information, while avoiding the inclusion of nodes that are too far away and have low relevance to the target node. After setting, 3-5 sensor nodes are randomly selected for verification to check whether the number of neighborhood nodes selected by each node is within a reasonable range of 5-10. If the number is too large, i.e. more than 10, the threshold multiple is appropriately reduced; if the number is too small, i.e. less than 5, the threshold multiple is appropriately increased, and finally the neighborhood range threshold that meets the sensor group layout density and monitoring requirement is determined.

[0059] Finally, the contribution weight of the neighborhood node is calculated according to the edge weight, i.e. the spatial distance, between the neighborhood node and the target node. Specifically, the inverse of the edge weight is taken and normalized, so that the neighborhood nodes with closer distances have larger contribution weights. Then the original features of the target node and the original features of each neighborhood node are weighted and summed according to the corresponding contribution weights, and the sum result is nonlinearly transformed through the ReLU activation function, and finally the homogeneous node fusion feature fused with the spatial context information is output.

[0060] Further, the method provided by the embodiments of the present application comprises:

[0061] The homogeneous node fusion features representing different physical quantities are defined as different node types in the heterogeneous graph, and the heterogeneous information graph is constructed; the mutual influence weight between different physical quantity node types in the heterogeneous information graph is calculated using the heterogeneous graph attention network; and different physical quantity features at the same spatial position are fused into a unified comprehensive state feature vector according to the mutual influence weight, to obtain the site state embedding representation.

[0062] Specifically, first, all generated homogeneous node fusion features are sorted and classified according to their corresponding physical quantity types, for example, homogeneous node fusion features reflecting pressure state are classified into one category, and homogeneous node fusion features reflecting displacement state are classified into another category. The homogeneous node fusion features corresponding to each type of physical quantity are defined as a node type in the heterogeneous information graph. Then, the spatial position information of each node under each node type is obtained, and an association relationship between nodes of different physical quantity types at the same spatial position is established, i.e. there is a corresponding relationship between the pressure class nodes and the displacement class nodes at the same monitoring position, so as to construct a heterogeneous information graph containing multiple node types and associated spatial positions.

[0063] In the heterogeneous graph attention network, first, linear transformation is performed on the node types of different physical quantities in the heterogeneous information graph respectively, and the projected feature vectors are obtained. Then, for the projected feature vectors of any two different physical quantity types at the same spatial position, the original attention coefficient is calculated through the shared attention vector. Finally, the original attention coefficient is normalized to obtain the normalized mutual influence weight, which is described in detail in the subsequent content.

[0064] Then, for each monitoring spatial position, the homogeneous node fusion features corresponding to all different physical quantity node types at the position are extracted, each feature is multiplied by the corresponding mutual influence weight to obtain a weighted vector of each feature. Then, all weighted vectors are added element by element to obtain a vector integrating all physical quantity information at the position, which is the comprehensive state feature vector of the spatial position. After traversing all monitoring spatial positions, the above weighted summation operation is repeated to finally obtain a comprehensive state feature vector set covering all monitoring positions, which is the site state embedding representation.

[0065] Further, the method provided in the embodiments of the application comprises:

[0066] The different physical quantity node types in the heterogeneous information graph are respectively linearly transformed and projected into a unified feature space to obtain projected feature vectors. For the projected feature vectors of any two different physical quantity types at the same spatial position, the original attention coefficient is calculated through the shared attention vector. The original attention coefficient is normalized to obtain the normalized mutual influence weight.

[0067] In one embodiment, first, a heterogeneous graph attention network is constructed, and the network structure includes an input layer, an attention calculation layer and an output layer: the input layer dimension is set to the unified dimension of the projected feature vector, such as 64 dimensions, for receiving the projected features of different physical quantity node types; the attention calculation layer is set to have one attention head, which is used to simplify the calculation and meet the basic fusion requirements, and each attention head includes a linear transformation unit and an attention coefficient calculation unit; the output layer dimension is consistent with the input layer, which is used to output the normalized mutual influence weight.

[0068] Then the network parameters are initialized, the linear transformation weight matrix of the attention calculation layer is initialized by using the Xavier initialization method, the bias term is initialized to 0, and the shared attention vector is also generated by Xavier initialization to ensure uniform parameter distribution to avoid training bias. Then the activation function and loss function of the network are configured, the tanh activation function is selected for the attention calculation unit to process the intermediate calculation value, and the Softmax activation function is used before the output layer to realize normalization; the cross-entropy loss function is used as the loss function to measure the difference between the network output weight and the sample label, for example, the physical quantity influence weight manually labeled, and the Adam optimizer is selected, the learning rate is set to 0.001, and the loss function value is converged to a preset threshold, such as 0.001, through iterative training, and finally the construction of the heterogeneous graph attention network is completed.

[0069] After the construction is completed, the homogeneous node fusion features of different physical quantity node types in the heterogeneous information graph are input into the input layer of the heterogeneous graph attention network, the linear transformation unit of the attention calculation layer is used to perform linear transformation on the features of each physical quantity node, that is, through the operation of multiplying the feature vector by the weight matrix and adding the bias term, all features are projected into a unified 64-dimensional feature space to obtain the projected feature vector.

[0070] Subsequently, for the projected feature vectors of any two different physical quantity types at the same spatial position, the shared attention vector is called by the attention coefficient calculation unit, and the dot product operation is performed on the two projected feature vectors and the shared attention vector respectively, and the operation results are added and input into the tanh activation function. The output value is the original attention coefficient calculated by the network, which describes the influence degree of one physical quantity on another physical quantity. When the calculation conditions are uniform, such as the same linear transformation rule, shared attention vector and no abnormal data interference, the larger the original attention coefficient is, the deeper the influence of one physical quantity on another physical quantity is.

[0071] Finally, all original attention coefficients at the same spatial position are transmitted to the output layer, and the ratio of each coefficient to the exponential sum of all coefficients is calculated by the Softmax activation function to obtain a normalized result with a total sum of 1. The result is the mutual influence weight between different physical quantity node types.

[0072] Further, the method provided in the embodiments of the present application comprises:

[0073] When the bearing capacity distribution field displays that the bearing capacity safety factor of the first area is lower than the preset threshold, a diagnostic task containing the coordinates of the first area is automatically created; the diagnostic task is issued to the movable inspection test unit; the movable inspection test unit is controlled to navigate to the first area and perform a touch test, and a foundation bearing capacity warning is performed according to diagnostic data.

[0074] In this embodiment, the mobile inspection and testing unit is a field device that integrates positioning and navigation, foundation penetration testing and data transmission functions, and can be controlled to move to the target area to perform bearing capacity testing.

[0075] Optionally, firstly, all grid node data in the bearing capacity distribution field are traversed, and the bearing capacity safety factor and corresponding coordinates of each grid node are extracted. The bearing capacity safety factor of each grid node is then compared one by one with a preset threshold, which is determined by those skilled in the art according to foundation design specifications. When it is found that the bearing capacity safety factors of multiple consecutive grid nodes are all lower than the preset threshold, the coordinates of these consecutive grid nodes are integrated, and the maximum and minimum values ​​of the X and Y axes of that region are taken to determine the coordinate range of the first region. Subsequently, a diagnostic task is generated, containing the coordinate range of the first region, the type of test to be performed, and the data acquisition requirements. A unique identifier is assigned to the task, completing the automatic creation of the diagnostic task.

[0076] Next, the MQTT communication protocol is used to distribute diagnostic tasks. First, a network connection is established between the server and the mobile inspection test unit. After starting, the mobile inspection test unit automatically connects to the designated wireless network and subscribes to the server's diagnostic task topic. The server encapsulates the created diagnostic task into JSON format data, which includes a unique task identifier, the coordinate range of the first area, test parameters, and other information. This JSON data is then published to the diagnostic task topic. The mobile inspection test unit listens to this topic in real time. Upon receiving the JSON data, it extracts the task information using its built-in parser, verifies the validity of the task identifier, and, if everything is correct, completes the receipt of the diagnostic task, thus distributing the task.

[0077] The mobile inspection test unit then navigated to the first area using GPS positioning combined with the A* path planning algorithm. First, it obtained its current coordinates via the built-in GPS module. These current coordinates, along with the target coordinates of the first area, were input into the A* algorithm. The target coordinates were taken as the center coordinates of the first area. The algorithm aimed to minimize the travel distance, avoiding known obstacles such as construction equipment and ditches. A pre-set obstacle coordinate library was used to calculate the optimal path from the current location to the target area. During navigation, the mobile inspection test unit updated its GPS coordinates every 10 seconds, comparing them with the coordinates of key nodes on the optimal path and adjusting its direction of movement until it reached the first area.

[0078] After the movable inspection test unit reaches the first area, a built-in light power sounding test device is started, a sounding probe is vertically aligned with the ground, the probe is pressed into the foundation soil with constant hammering energy, the number of hammering is recorded every 30 cm, after continuous testing of 3 test points, the actual bearing capacity safety factor is calculated according to the corresponding formula of the number of hammering and the bearing capacity of the foundation soil determined by engineering experience formula. The actual bearing capacity safety factor calculated is compared with the preset threshold value, if it is still lower than the threshold value, the audible and visual early warning device of the inspection test unit is triggered to flash red light, buzzer alarm and the like, at the same time, the number of hammering, the actual bearing capacity safety factor, the test coordinates and the test time of the diagnostic data are uploaded to the server end through the MQTT communication protocol, and the foundation bearing capacity early warning is completed.

[0079] Through automatic creation of a diagnostic task, task distribution and control of the movable inspection test unit to execute positioning and navigation, sounding test and early warning, accurate positioning and timely early warning of the low safety factor area of the foundation bearing capacity are realized.

[0080] Further, the method provided in the embodiment of the application comprises:

[0081] The movable inspection test unit returns the diagnostic data collected in real time, compares the diagnostic data with the corresponding data in the bearing capacity distribution field, calculates the confidence error, and if the confidence error exceeds the allowable error range, an incremental learning process is started, and the homogenous fusion and heterogeneous fusion fusion parameters are optimized by using new diagnostic data.

[0082] In one embodiment, the movable inspection test unit returns the diagnostic data collected in real time to the server end, and the returned data includes the number of hammering in the sounding test, the specific coordinates of each test point, the test time and the actual bearing capacity safety factor. After receiving the diagnostic data, the server end calculates the confidence error by using a relative error formula, the formula is: confidence error = | (actual bearing capacity safety factor-predicted bearing capacity safety factor) / actual bearing capacity safety factor | x 100%, and the confidence error value of the position is obtained, wherein the predicted bearing capacity safety factor is the bearing capacity safety factor in the bearing capacity distribution field.

[0083] Then the calculated confidence error is compared with the preset allowable error range (such as 5%, set according to the foundation monitoring accuracy requirement), if the confidence error exceeds 5%, the incremental learning process is started. First, the new diagnostic data is preprocessed, and the abnormal values are removed, for example, the data of the number of hammering is 0 or exceeds the reasonable range, the 3σ criterion is used to judge the abnormality, and the actual bearing capacity safety factor after preprocessing is associated with the site state embedding at the corresponding position, and a new sample set for incremental learning is formed. Then load the graph convolution network parameters used for homogeneous fusion in the foregoing step and the heterogeneous graph attention network parameters used for heterogeneous fusion, input the new sample set into the two networks, and use the mini-batch gradient descent method to optimize the parameters, set the batch size to 16, the learning rate to 0.0001, and only update the network part parameters such as the neighborhood aggregation weight of the graph convolution network and the shared attention vector of the heterogeneous graph attention network with the new sample each time. After 15 iterations, the matching error between the fusion features output by the network and the actual bearing capacity safety factor in the new sample is calculated, and if the error is reduced to the allowable error range, the optimized fusion parameters are saved, and the optimization of the homogeneous fusion and heterogeneous fusion parameters is completed.

[0084] By feeding back the diagnostic data in real time and calculating the confidence error, the incremental learning is started when the allowable error is exceeded, which dynamically improves the accuracy of homogeneous fusion and heterogeneous fusion, and further improves the accuracy of foundation bearing capacity evaluation.

[0085] In summary, the foundation bearing capacity intelligent identification method based on a sensing network provided by the embodiments of the application has the following technical effects:

[0086] The application collects the physical quantity data of the foundation through the sensing network, obtains the site state embedding representation through the graph convolution network homogeneous fusion and the heterogeneous graph attention network heterogeneous fusion, inputs the full connection neural network to calculate the bearing capacity safety factor of each position, interpolates to generate the bearing capacity distribution field, and optimizes the parameters combined with the retest data of the movable inspection test unit, so as to accurately identify the foundation bearing capacity, make the foundation bearing capacity detection result more accurate and reliable, and achieve the technical effects of accurate identification of the foundation bearing capacity, global coverage and long-term reliable optimization, and meet the technical effects of the engineering safety monitoring and control of the foundation.

[0087] Embodiment two, as shown in Figure 2 The system based on the same inventive concept as the foregoing embodiment one, comprises:

[0088] The monitoring data acquisition module 1 is used to construct a hybrid perception network and acquire multi-source heterogeneous monitoring data of the foundation.

[0089] A database construction module 2 is configured to establish a unified space-time reference of the multi-source heterogeneous monitoring data and construct a space-time database.

[0090] A distribution field construction module 3 is configured to homogeneously fuse and heterogeneously fuse the data in the space-time database, comprehensively judge the state of the foundation bearing capacity based on the fusion result, and generate a bearing capacity distribution field of the foundation bearing capacity.

[0091] Further, the monitoring data acquisition module 1 is configured to perform the following steps:

[0092] The mixed perception network includes sparsely arranged fixed first sensor nodes and second distributed sensing units covering key areas, and the accuracy of the first sensor nodes is higher than that of the second distributed sensing units.

[0093] Further, the monitoring data acquisition module 1 is configured to perform the following steps:

[0094] Based on the prior geological model, finite element numerical simulation is performed, and through model parameter sensitivity analysis, a key area with a sensitivity greater than a preset sensitivity to bearing capacity inversion is identified; in the key area, the optimal arrangement position of the first sensor nodes is determined according to the preset first sensor arrangement quantity and is deployed; the area between any two adjacent first sensor nodes is covered by a planar mesh to deploy and establish a second distributed sensing unit.

[0095] Further, the distribution field construction module 3 is configured to perform the following steps:

[0096] For sensors of the same physical quantity, message passing and aggregation are performed to generate a homogenous node fusion feature representing the local area pressure state after homogenous fusion; a heterogeneous information graph is constructed based on the homogenous node fusion feature, a node-level fusion of the heterogeneous information graph is performed using a heterogeneous graph attention network to generate a site state embedding representation; the state embedding representation is input into a fully connected neural network, which calculates a bearing capacity safety factor for each position; based on the coordinates and bearing capacity safety factors of all positions, a continuous meshed bearing capacity distribution field covering the entire monitoring area is generated by a spatial interpolation algorithm.

[0097] Further, the distribution field construction module 3 is configured to perform the following steps:

[0098] For a sensor group of the same physical quantity, a homogenous sensor graph is constructed, in which sensors are nodes and the spatial distance between nodes is the weight of the edge; a neighborhood aggregation operation is performed on the homogenous sensor graph using a graph convolution network to output a homogenous node fusion feature that fuses spatial context information.

[0099] Furthermore, the distribution field construction module 3 is used to perform the following steps:

[0100] The fusion features of homogeneous nodes representing different physical quantities are defined as different node types in the heterogeneous graph, and the heterogeneous information graph is constructed. The mutual influence weights between different physical quantity node types in the heterogeneous information graph are calculated using a heterogeneous graph attention network. Based on the mutual influence weights, the features of different physical quantities at the same spatial location are fused into a unified comprehensive state feature vector to obtain the site state embedding representation.

[0101] Furthermore, the distribution field construction module 3 is used to perform the following steps:

[0102] For different physical quantity node types in the heterogeneous information graph, a linear transformation is performed respectively, and the vectors are projected onto a unified feature space to obtain projected feature vectors. For any two different physical quantity types of projected feature vectors at the same spatial location, the original attention coefficients are calculated through a shared attention vector. The original attention coefficients are normalized to obtain the normalized mutual influence weights.

[0103] Furthermore, the distribution field construction module 3 is used to perform the following steps:

[0104] When the bearing capacity distribution field shows that the bearing capacity safety factor of the first area is lower than a preset threshold, a diagnostic task containing the coordinates of the first area is automatically created; the diagnostic task is sent to the mobile inspection and testing unit; the mobile inspection and testing unit is controlled to navigate to the first area, perform a penetration test, and issue a foundation bearing capacity warning based on the diagnostic data.

[0105] Furthermore, the distribution field construction module 3 is used to perform the following steps:

[0106] The mobile inspection and testing unit transmits the diagnostic data collected on-site in real time, compares the diagnostic data with the corresponding data in the bearing capacity distribution field, and calculates the confidence error. If the confidence error exceeds the allowable error range, the incremental learning process is started to optimize the fusion parameters of homogeneous fusion and heterogeneous fusion using the new diagnostic data.

[0107] The intelligent foundation bearing capacity identification system based on sensor network provided in this embodiment of the invention can execute the intelligent foundation bearing capacity identification method based on sensor network provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0108] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server, the various units and modules are only divided according to the functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific name of each functional unit is only for the convenience of mutual differentiation, and does not serve to limit the protection scope of the present application.

[0109] The above detailed description does not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application. In some cases, the actions or steps described in the present application can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

Claims

1. A method for intelligent identification of foundation bearing capacity based on sensor networks, characterized in that, include: Construct a hybrid sensing network to acquire multi-source heterogeneous monitoring data from the ground; Establish a unified spatiotemporal benchmark for the multi-source heterogeneous monitoring data and construct a spatiotemporal database; The data in the spatiotemporal database are subjected to homogeneous fusion and heterogeneous fusion. Based on the fusion results, a comprehensive judgment is made on the bearing capacity state of the foundation, and a bearing capacity distribution field of the foundation is generated. The process of homogeneous and heterogeneous fusion of data in the spatiotemporal database, and comprehensive judgment of the foundation bearing capacity state based on the fusion results to generate a bearing capacity distribution field of the foundation bearing capacity, includes: For sensors that measure the same physical quantity, message passing and aggregation are performed to generate homogeneous node fusion features that represent the local pressure state after homogeneous fusion. Based on the homogeneous node fusion features, a heterogeneous information graph is constructed. Using a heterogeneous graph attention network, node-level fusion of the heterogeneous information graph is performed to generate a site state embedding representation. The state embedding representation is input into a fully connected neural network, which calculates a load-bearing capacity safety factor for each location. Based on the coordinates of all locations and the bearing capacity safety factor, a continuous gridded bearing capacity distribution field covering the entire monitoring area is generated using a spatial interpolation algorithm.

2. The intelligent identification method for foundation bearing capacity based on sensor networks as described in claim 1, characterized in that, in, The hybrid sensing network includes sparsely deployed fixed first sensor nodes and a second distributed sensing unit covering key areas; wherein the accuracy of the first sensor node is higher than that of the second distributed sensing unit.

3. The intelligent identification method for foundation bearing capacity based on sensor networks as described in claim 2, characterized in that, Constructing a hybrid sensing network includes: Finite element numerical simulations were performed based on a priori geological models. Through model parameter sensitivity analysis, key areas with a sensitivity to bearing capacity inversion greater than the preset sensitivity were identified. In the critical area, the optimal deployment location of the first sensor nodes is determined and deployed according to the preset number of first sensors; A second distributed sensing unit is deployed by covering the area between any two adjacent first sensor nodes with a planar grid.

4. The intelligent identification method for foundation bearing capacity based on sensor networks as described in claim 1, characterized in that, For sensors measuring the same physical quantity, message passing and aggregation are performed to generate homogeneous node fusion features representing the local pressure state after homogeneous fusion, including: For a group of sensors that measure the same physical quantity, a homogeneous sensor graph is constructed, where the sensors are nodes and the spatial distance between nodes is the weight of the edges. A graph convolutional network is used to perform neighborhood aggregation on the homogeneous sensor graph, and output homogeneous node fusion features that fuse spatial context information.

5. The intelligent identification method for foundation bearing capacity based on sensor networks as described in claim 1, characterized in that, Based on the homogeneous node fusion features, a heterogeneous information graph is constructed. A heterogeneous graph attention network is then used to perform node-level fusion of the heterogeneous information graph, generating a site state embedding representation, including: The fusion features of homogeneous nodes representing different physical quantities are defined as different node types in the heterogeneous graph, and the heterogeneous information graph is constructed accordingly. Using a heterogeneous graph attention network, the mutual influence weights between different physical quantity node types in the heterogeneous information graph are calculated; Based on the mutual influence weights, the characteristics of different physical quantities at the same spatial location are fused into a unified comprehensive state feature vector to obtain the site state embedding representation.

6. The intelligent identification method for foundation bearing capacity based on sensor networks as described in claim 5, characterized in that, Using a heterogeneous graph attention network, the mutual influence weights between different physical quantity node types in the heterogeneous information graph are calculated, including: For different physical quantity node types in the heterogeneous information graph, perform linear transformations respectively, project them onto a unified feature space, and obtain the projected feature vectors; For any two different types of physical quantities projected at the same spatial location, the original attention coefficients are calculated using a shared attention vector. The original attention coefficients are normalized to obtain the normalized mutual influence weights.

7. The intelligent identification method for foundation bearing capacity based on sensor networks as described in claim 1, characterized in that, After generating the bearing capacity distribution field of the foundation bearing capacity, the following is also included: When the bearing capacity distribution field shows that the bearing capacity safety factor of the first region is lower than a preset threshold, a diagnostic task containing the coordinates of the first region is automatically created; The diagnostic task is then sent to the mobile inspection and testing unit. The mobile inspection and testing unit is controlled to navigate to the first area, perform a probing test, and provide an early warning of the foundation bearing capacity based on the diagnostic data.

8. The intelligent identification method for foundation bearing capacity based on sensor networks as described in claim 7, characterized in that, After performing the touch test, it also includes: The mobile inspection and testing unit transmits the diagnostic data collected on-site back in real time, compares the diagnostic data with the corresponding data in the bearing capacity distribution field, and calculates the confidence error. If the confidence error exceeds the allowable error range, an incremental learning process is initiated to optimize the fusion parameters of homogeneous and heterogeneous fusions using new diagnostic data.

9. A foundation bearing capacity intelligent identification system based on sensor networks, characterized in that, The system is used to implement the intelligent identification method for foundation bearing capacity based on sensor networks according to any one of claims 1-8, the system comprising: The monitoring data acquisition module is used to construct a hybrid sensing network and acquire multi-source heterogeneous monitoring data from the ground. The database construction module is used to establish a unified spatiotemporal reference for the multi-source heterogeneous monitoring data and construct a spatiotemporal database. The distribution field construction module is used to perform homogeneous fusion and heterogeneous fusion of data in the spatiotemporal database, make a comprehensive judgment on the bearing capacity state of the foundation based on the fusion results, and generate a bearing capacity distribution field of the foundation bearing capacity.

Images