A method, system and medium for determining bridge load limit based on actual traffic load

Abstract

The present invention discloses a method, system and storage medium for determining the bridge load limit based on actual traffic load, which relates to the technical field of bridge bearing capacity assessment, and includes the following steps: collecting passing vehicle characteristic information; using passing vehicle characteristic information and a dynamic weighing model to obtain assumed bridge strain data to determine estimated vehicle driving information; obtaining actual passing vehicle information; establishing a digital model of real traffic load; generating a test simulation traffic flow, and obtaining the overloading violation rate of overloaded vehicles in the test simulation traffic flow; using an influence surface loading method to obtain the test strain history of the bridge; calculating the equivalent stress amplitude and average stress value; and performing fatigue life calculation. In the method of the present invention, the computer recognition device is only used to obtain the vehicle model and the number of axles for photographing and identifying, which avoids the shortcomings of the existing vehicle position information collected by the computer vision system, which leads to inaccuracy and the existence of false axles in the axle information, and can accurately assess the bridge load.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge bearing capacity assessment, and in particular to a method, system and medium for determining a bridge load limit based on actual traffic load. Background Art

[0002] There are differences between the design loads of my country's in-service highway bridges and the growing actual traffic loads in terms of size, frequency, spatial distribution, etc., resulting in the current prescribed bridge load limits may not be able to ensure the long-term safe operation of in-service bridges.

[0003] The current technology uses cameras to collect vehicle types, lanes, and longitudinal positions, and then uses lane and longitudinal position information with dynamic weighing equipment to calculate vehicle axle weight information; based on this information, a load model is established, and then the influence line of the bridge control section is loaded to obtain the bridge response history and fatigue limit load verification. The main defects include: the inability to accurately identify vehicle position information in complex environments (such as occlusion, weak light, etc.) based on the computer vision system; the number of vehicle axles identified by computer vision may not match the actual number of effective axles (such as false axles identified by large vehicles); when using influence line loading, since the influence line is a two-dimensional curve, only different influence lines of different lanes can be considered, and the precise lateral position of the vehicle cannot be considered. When the bridge response is sensitive to the lateral position (such as orthotropic steel bridge decks), the response history obtained may have a large deviation from the actual response history.

[0004] In view of this, it is necessary to propose a method, system and medium for determining bridge load limits based on actual traffic load to solve or at least alleviate some of the above-mentioned defects. Summary of the invention

[0005] The main purpose of the present invention is to provide a method and system for determining bridge load limits based on actual traffic loads, aiming to solve the technical problem in the prior art of completely relying on visual technology and obtaining the response history of the bridge and fatigue load limit verification based on influence line loading technology to perform bridge load verification, resulting in inaccurate bridge load determination and poor bridge safety.

[0006] To achieve the above object, the present invention provides a method for determining a bridge load limit based on actual traffic load, S10, collecting passing vehicle feature information through a computer vision device, the passing vehicle feature information including the number of photographed passing vehicles, the photographed vehicle model of each vehicle, and the photographed and identified axle number;

[0007] S20, using the assumed vehicle driving information to obtain assumed bridge strain data based on a dynamic weighing model, and determining the assumed vehicle driving information corresponding to the difference between the assumed bridge strain data and the actual bridge strain data within a first threshold range as the estimated vehicle driving information; the estimated vehicle driving information includes the estimated single wheel weight, estimated vehicle gross weight, estimated lateral position, estimated longitudinal position, and estimated wheel-to-head distance of each vehicle passing through;

[0008] S30, fusing the passing vehicle characteristic information and the estimated vehicle travel information to obtain actual passing vehicle information, the actual passing vehicle information including the actual number of passing vehicles and the actual number of effective axles of each vehicle, the actual axle weight of a single axle, the actual gross vehicle weight, the actual lateral position, the actual longitudinal position, the actual axle spacing and the actual travel speed;

[0009] S40, establishing a real traffic load digital model using actual passing vehicle information within the target time period;

[0010] S50, generating a test simulated traffic flow using a real traffic load digital model, and obtaining an overload violation rate of overloaded vehicles in the test simulated traffic flow based on a current load limit value of the bridge;

[0011] S60, using test simulation of traffic flow and using influence surface loading to obtain the test strain history of the bridge;

[0012] S70, calculate the equivalent stress amplitude and mean stress value based on the test strain history using the rain flow counting method;

[0013] S80, fatigue life calculation using equivalent stress amplitude and mean stress value;

[0014] S91, if the fatigue life is not less than the expected service life, the current load limit value meets the requirements;

[0015] S92, if the fatigue life is less than the expected service life, the current load limit value is adjusted with a preset step size to correct the overloading violation rate of the test simulated traffic flow, and enter step S60 until the fatigue life is not less than the expected service life.

[0016] Furthermore, in step S80, the fatigue life of the bridge is verified by using the Palmgren-Miner linear cumulative damage theory to calculate the stress amplitude spectrum of the bridge.

[0017] Furthermore, if the actual axle weight of the single axle of the photographed and identified axle is not less than the preset axle weight threshold, the photographed and identified axle is determined to be a true axle; if the actual axle weight of the single axle of the photographed and identified axle is less than the preset axle weight threshold, the photographed and identified axle is determined to be a false axle.

[0018] Furthermore, the steps of establishing the dynamic weighing model include:

[0019] S201, obtaining training traffic information of each training vehicle within the training duration, and evenly dividing the training duration into K training time nodes;

[0020] S202, obtaining J training longitudinal time travel curves of each training vehicle within the training duration;

[0021] S203, based on J training longitudinal time travel sub-curves, obtaining a training longitudinal time travel total curve corresponding to each training vehicle within the training duration, and obtaining a training ordinate matrix consisting of training time nodes for each training vehicle;

[0022] S204, obtaining the training transverse coordinates of each training vehicle at each training time node, and obtaining a training transverse coordinate matrix of each training vehicle formed by the training time nodes;

[0023] S205, fusing the training ordinate matrix and the training transverse matrix of each training vehicle to obtain a training transverse and longitudinal matrix of each training vehicle;

[0024] S206, obtaining a wheel transverse and longitudinal coordinate matrix corresponding to each training wheel group based on the training transverse spacing and the training longitudinal spacing of the training wheel group relative to the vehicle head in each training vehicle;

[0025] S207, obtaining the sub-coordinates of the training influence surface of each training wheel group of the training vehicle in the Q-order mode;

[0026] S208, obtaining the total coordinates of the training influence surface of each training wheel set of each training vehicle based on the sub-coordinates of the training influence surface;

[0027] S209, obtaining a training bridge score response corresponding to each training wheel group in each training vehicle based on the training wheel load corresponding to the training wheel group;

[0028] S210: Obtaining a total response of a training bridge composed of training wheel groups corresponding to all training vehicles based on all training bridge partial responses.

[0029] Further, S201, obtain the training pass information within the training duration, and evenly divide the training duration into K training time nodes to form a K×1 training time sequence matrix [ t 1 ,…t k ,…t K ] T ,in, t k yes k Ordinal training time nodes;

[0030] S202, training time series matrix [ t1 ,…t k ,…t K ] T Substitute into the sigmoid layer with J cores to generate a K×J training transformation matrix;

[0031] S203, performing a convolution operation on the training transformation matrix to obtain a K×1 training longitudinal coordinate matrix [ x 1 ,…x k ,… x K ] T , where the convolution kernel is a J×1 matrix of all 1s, x k is the training longitudinal coordinate of the training vehicle at the kth training time node;

[0032] S204, training time series matrix [ t 1 ,…t k ,…t K ] T Connect a fully connected layer to get the training horizontal coordinate matrix [ y 1 ,…y k ,…y K ] T ,in, y k is the training lateral coordinate of the training vehicle at the kth training time node;

[0033] S205, merging the training longitudinal coordinates and the training transverse coordinates to obtain a K×2 training transverse and longitudinal coordinate matrix , where one training vehicle corresponds to one training horizontal and vertical coordinate matrix.

[0034] Further, in S206, assuming that the number of training wheels of the training vehicle is I, a convolution calculation of a 2×2 convolution kernel is performed on the horizontal and vertical coordinate matrix of the training vehicle to obtain I channels, each channel is a K×2 matrix, and the convolution kernel is , D i,x and D i,y They are i The training lateral spacing and training longitudinal spacing of each tire from the front of the vehicle, i∈I;

[0035] S207, substituting the K×2 matrix of each channel into a Gaussian layer with Q cores to generate a K×Q influence surface matrix, where Q represents the selected Gaussian mode order;

[0036] S208, for each channel i, the influence surface matrix is ​​convolved to obtain a K×1 influence surface coordinate matrix [ IS i 1 ,…IS i k ,…IS i K ] T , where the convolution kernel is a J×1 matrix of all 1s;

[0037] S209, the influence surface coordinate matrix of each channel i is calculated with a convolution kernel of 1×1 to obtain a K×1 bridge response time history matrix [ R i 1 ,…R i k ,…R i K ] T , where the convolution kernel of the i-th channel is the training wheel load of the i-th training wheel axle;

[0038] S210, for all channels, the total response time history matrix of the bridge is obtained by accumulation [ R 1 ,…R k ,…R K ] T .

[0039] Further, in step S207, the matrix element in the kth row and qth column of the influence surface matrix of the i-th channel is

[0040] ,in, represents the axle of the i-th training wheel, represents the axle of the i-th training wheel, express, express.

[0041] The present invention also provides a method for determining a bridge load limit based on actual traffic load.

[0042] Furthermore, a multi-Gaussian model is used to establish a digital model of real traffic load.

[0043] The present invention also provides a system for determining a bridge load limit based on actual traffic load, comprising a processing device, wherein the processing device is used to implement the steps of the above-mentioned method for determining a bridge load limit based on actual traffic load.

[0044] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned method for determining the bridge load limit based on the actual traffic load are implemented.

[0045] Compared with the prior art, the method for determining the bridge load limit based on actual traffic load provided by the present invention has the following beneficial effects:

[0046] The present invention provides a method for determining the bridge load limit based on actual traffic flow load, which collects the characteristic information of passing vehicles through computer vision equipment, the characteristic information of passing vehicles includes the number of passing vehicles photographed, the vehicle model photographed and the number of axles photographed and identified for each vehicle; then, based on the assumed vehicle driving information, the dynamic weighing model and the actual bridge strain data, the estimated vehicle driving information is determined, the estimated vehicle driving information includes the estimated single-wheel weight, the estimated total vehicle weight, the estimated lateral position, the estimated longitudinal position and the estimated wheel-to-head distance of each passing vehicle; the actual passing vehicle information is obtained by integrating the passing vehicle characteristic information and the estimated vehicle driving information; it is realized that the actual passing vehicle information can be determined based on the photographed vehicle model, the photographed and identified axle number and the dynamic weighing model, and a real traffic flow load digital model can be established based on the actual passing vehicle information, and the computer recognition equipment is only used to obtain the photographed vehicle model and the photographed and identified axle number, which avoids the shortcomings of the existing vehicle position information collected by the computer vision system, which leads to inaccuracy and the existence of false axles in the axle information; at the same time, in the scheme of the present invention, A digital model of real traffic load is established based on the coordinates of the bridge influence surface of the actual lateral position and the actual longitudinal position, and a test simulated traffic is generated using the digital model of real traffic load. The overloading violation rate of overloaded vehicles in the test simulated traffic is obtained based on the current load limit value of the bridge; the test strain history of the bridge is obtained by using the test simulated traffic to load the influence surface; the equivalent stress amplitude and the average stress value are calculated based on the test strain history; the fatigue life is calculated using the equivalent stress amplitude and the average stress value; if the fatigue life is not less than the expected service life, the current load limit meets the requirements; if the fatigue life is less than the expected service life, the current load limit is adjusted with a preset step size to correct the overloading violation rate of the test simulated traffic, and the steps are repeated until the fatigue life is not less than the expected service life; if the fatigue life is not less than the expected service life, the current load limit is determined to be the load limit value of the in-service bridge. In determining the load limit value of the in-service bridge, the lateral position and the longitudinal position of the vehicle are considered for loading to obtain the stress amplitude, so as to maximize the usability of the material while ensuring the safety of the bridge. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0048] Figure 1 It is a flow chart of a method for determining a bridge load limit based on actual traffic load in one embodiment of the present invention;

[0049] Figure 2 It is a schematic diagram of the flow chart of the dynamic weighing model in the scheme of the present invention;

[0050] Figure 3 It is a schematic diagram of the model principle of the dynamic weighing model in the solution of the present invention.

[0051] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0052] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0055] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0056] Please refer to the attached Figure 1 , Figure 2 as well as Figure 3 The present invention provides a method for determining a bridge load limit based on actual traffic load, comprising the following steps:

[0057] S10, collecting passing vehicle feature information through a computer vision device, the passing vehicle feature information including the number of photographed passing vehicles, the photographed vehicle model of each vehicle, and the photographed and identified axle number;

[0058] S20, using the passing vehicle characteristic information and the assumed vehicle driving information to obtain assumed bridge strain data based on a dynamic weighing model, and determining the assumed vehicle driving information corresponding to the difference between the assumed bridge strain data and the actual bridge strain data within a first threshold range as the estimated vehicle driving information; the estimated vehicle driving information includes the estimated single wheel weight, estimated vehicle gross weight, estimated lateral position, estimated longitudinal position and estimated wheel-to-head distance of each passing vehicle;

[0059] S30, fusing the passing vehicle characteristic information and the estimated vehicle driving information to obtain the actual passing vehicle information, the actual passing vehicle information including the actual number of passing vehicles and the actual number of effective axles of each vehicle, the actual axle weight of a single axle, the actual gross vehicle weight, the actual lateral position, the actual longitudinal position, the actual axle spacing and the actual driving speed; optionally, the actual passing vehicle information may also include the actual load eccentricity;

[0060] S40, establishing a real traffic load digital model using actual passing vehicle information within the target time period;

[0061] S50, generating a test simulated traffic flow using a real traffic load digital model, and obtaining an overload violation rate of overloaded vehicles in the test simulated traffic flow based on a current load limit value of the bridge;

[0062] S60, using test simulation of traffic flow and using influence surface loading to obtain the test strain history of the bridge;

[0063] S70, calculate the equivalent stress amplitude and mean stress value based on the test strain history using the rain flow counting method;

[0064] S80, fatigue life calculation using equivalent stress amplitude and mean stress value;

[0065] S91, if the fatigue life is not less than the expected service life, the current load limit value meets the requirements;

[0066] S92, if the fatigue life is less than the expected service life, the current load limit value is adjusted with a preset step size to correct the overloading violation rate of the test simulated traffic flow, and enter step S60 until the fatigue life is not less than the expected service life.

[0067] The method for determining the bridge load limit based on the actual traffic flow load provided by the present invention collects the characteristic information of passing vehicles through computer vision equipment, and the characteristic information of passing vehicles includes the number of passing vehicles photographed, the vehicle model photographed and the number of axles photographed and identified for each vehicle; then, based on the assumed vehicle driving information, the dynamic weighing model and the actual bridge strain data, the estimated vehicle driving information is determined, and the estimated vehicle driving information includes the estimated single-wheel weight, the estimated total vehicle weight, the estimated lateral position, the estimated longitudinal position and the estimated wheel-to-head distance of each passing vehicle; the actual passing vehicle information is obtained by integrating the passing vehicle characteristic information and the estimated vehicle driving information; it is realized that the actual passing vehicle information can be determined based on the photographed vehicle model, the photographed and identified axle number and the dynamic weighing model, and a real traffic flow load digital model can be established based on the actual passing vehicle information, and the computer recognition equipment is only used to obtain the photographed vehicle model and the photographed and identified axle number, thereby avoiding the shortcomings of the existing vehicle position information collected by the computer vision system, which leads to inaccuracy and the existence of false axles in the axle information; at the same time, in the scheme of the present invention, based on A digital model of real traffic load is established based on the coordinates of the bridge influence surface at the actual lateral position and the actual longitudinal position, and a test simulated traffic is generated using the digital model of real traffic load. The overloading violation rate of overloaded vehicles in the test simulated traffic is obtained based on the current load limit of the bridge; the test strain history of the bridge is obtained by using the test simulated traffic to load the influence surface; the equivalent stress amplitude and the average stress value are calculated based on the test strain history using the rain flow counting method; the fatigue life is calculated using the equivalent stress amplitude and the average stress value; if the fatigue life is not less than the expected service life, the current load limit meets the requirements; if the fatigue life is less than the expected service life, the current load limit is adjusted with a preset step size to correct the overloading violation rate of the test simulated traffic, and the steps are repeated until the fatigue life is not less than the expected service life; if the fatigue life is not less than the expected service life, the current load limit is determined to be the load limit of the in-service bridge, and the vehicle lateral position and the vehicle longitudinal position are considered in the determination of the load limit of the in-service bridge to obtain the stress amplitude, so as to maximize the material usability while ensuring the safety of the bridge.

[0068] It can be understood that the photographed identification axis is not necessarily the real actual identification axis. Only when the photographed identification axis is determined to have a certain weight through the dynamic weighing model can it be determined as the real actual identification axis, which can facilitate the elimination of false axes collected by computer vision equipment; the determination of bridge load limit is mainly to obtain the bridge load limit. The bridge load limit refers to the restriction on the mass of vehicles passing through the bridge to ensure the safety and service life of the bridge structure, so as to ensure that the bridge will not be damaged by overload; the dynamic weighing model is mainly used for dynamic weighing. Bridge dynamic weighing refers to the technology that relies on the relationship between bridge response and load and calculates the vehicle load on the bridge from the dynamic response of the bridge, involving signal processing, vehicle-bridge coupling vibration analysis, etc.; computer vision equipment is mainly used for target detection. Target detection refers to video-based vehicle detection technology, which uses cameras to monitor the feature changes of video images to identify vehicles without physical contact. Through image processing and other technologies, it detects the existence of vehicles and obtains information such as traffic flow, involving computer vision, image processing, pattern recognition and signal processing, etc.

[0069] It can be understood that in step S20, the assumed vehicle driving information corresponding to the minimum difference between the assumed bridge strain data and the actual bridge strain data is determined as the estimated vehicle driving information.

[0070] Optionally, the characteristic information of the passing vehicles is collected by computer vision equipment. When the vehicle passes through the bridge, the real-time video image of the vehicle passing through is captured by the camera system installed on the bridge, and then the category of the vehicle is identified by computer vision technology, which mainly includes vehicle detection and vehicle category identification. Specifically, vehicle category identification can rely on deep learning algorithms. First, the video of the vehicle passing through the bridge captured by the camera will be framed, and each frame in the video will be extracted into a static image. The static image constitutes a basic data set for identifying the vehicle category. The static image can clearly show the appearance characteristics of the vehicle, which is convenient for subsequent model training; next, the extracted vehicle image data set will be annotated, that is, each vehicle image will be labeled with a corresponding label (2-axle sedan, 2-axle truck, 3-axle truck, etc.); after the data set is constructed, the entire data set is divided into a training set and a test set, usually in a ratio of 7:3 or 8:2, 70% to 80% of the data is used for model training, and the remaining 20% ​​to 30% of the data is used to test the performance of the model. In order to identify the vehicle category, the convolutional neural network (CNN) model is used in the scheme of the present invention, which can automatically extract the visual features of the vehicle and identify different types of vehicles.

[0071] Preferably, after obtaining the stress history (test strain history), the rain flow counting method is used to calculate the equivalent stress amplitude and the average stress; then the bridge stress amplitude spectrum is used to verify the fatigue life through the Palmgren-Miner linear cumulative damage theory; if the fatigue life is greater than the expected service life, the current load limit value meets the requirements; if not, the load limit value is adjusted to correct the simulated traffic flow, and the fatigue verification is performed again, and finally the load limit value of the in-service bridge that meets the specification requirements is obtained.

[0072] Furthermore, in the solution of the present invention, whether the photographed and identified axle is a real axle is determined based on the actual axle weight of the single axle. Specifically, if the actual axle weight of the photographed and identified axle is not less than the preset axle weight threshold, the photographed and identified axle is determined to be a real axle; if the actual axle weight of the photographed and identified axle is less than the preset axle weight threshold, the photographed and identified axle is determined to be a false axle. Optionally, the preset axle weight threshold can be set according to actual conditions, and can be 50 kg, 500 kg, or other values.

[0073] Furthermore, the steps of establishing the dynamic weighing model include:

[0074] S201, obtaining training traffic information of each training vehicle within the training duration, and evenly dividing the training duration into K training time nodes;

[0075] S202, obtaining J training longitudinal time travel curves of each training vehicle within the training duration;

[0076] S203, based on J training longitudinal time travel sub-curves, obtaining a training longitudinal time travel total curve corresponding to each training vehicle within the training duration, and obtaining a training ordinate matrix consisting of training time nodes for each training vehicle;

[0077] S204, obtaining the training transverse coordinates of each training vehicle at each training time node, and obtaining a training transverse coordinate matrix of each training vehicle formed by the training time nodes;

[0078] S205, fusing the training ordinate matrix and the training transverse matrix of each training vehicle to obtain a training transverse and longitudinal matrix of each training vehicle;

[0079] S206, obtaining a wheel transverse and longitudinal coordinate matrix corresponding to each training wheel group based on the training transverse spacing and the training longitudinal spacing of the training wheel group relative to the vehicle head in each training vehicle;

[0080] S207, obtaining the sub-coordinates of the training influence surface of each training wheel group of the training vehicle in the Q-order mode;

[0081] S208, obtaining the total coordinates of the training influence surface of each training wheel set of each training vehicle based on the sub-coordinates of the training influence surface;

[0082] S209, obtaining a training bridge score response corresponding to each training wheel group in each training vehicle based on the training wheel load corresponding to the training wheel group;

[0083] S210: Obtaining a total response of a training bridge composed of training wheel groups corresponding to all training vehicles based on all training bridge partial responses.

[0084] Further, S201, obtain the training pass information within the training duration, and evenly divide the training duration into K training time nodes to form a K×1 training time sequence matrix [ t 1 ,…t k ,…t K ] T ,in, t k yes k Ordinal training time nodes;

[0085] S202, training time series matrix [ t 1 ,…t k ,…t K ] T Substitute into the sigmoid layer with J cores to generate a K×J training transformation matrix;

[0086] S203, performing a convolution operation on the training transformation matrix to obtain a K×1 training longitudinal coordinate matrix [ x 1 ,…x k ,… x K ] T , where the convolution kernel is a J×1 matrix of all 1s, x k is the training longitudinal coordinate of the training vehicle at the kth training time node;

[0087] S204, training time series matrix [ t 1 ,…t k ,…t K ] T Connect a fully connected layer to get the training horizontal coordinate matrix [ y 1 ,…y k ,…y K ] T ,in,y k is the training lateral coordinate of the training vehicle at the kth training time node;

[0088] S205, merging the training longitudinal coordinates and the training transverse coordinates to obtain a K×2 training transverse and longitudinal coordinate matrix , where one training vehicle corresponds to one training horizontal and vertical coordinate matrix.

[0089] Furthermore, the calculation formula for the k-th row and j-th column element of the training transformation matrix is , where a, b, and c are all unknown coefficients of the sigmoid layer.

[0090] Further, in S206, assuming that the number of training wheels of the training vehicle is I, a convolution calculation of a 2×2 convolution kernel is performed on the horizontal and vertical coordinate matrix of the training vehicle to obtain I channels, each channel is a K×2 matrix, and the convolution kernel is , D i,x and D i,y They are i The training lateral spacing and training longitudinal spacing of each tire from the front of the vehicle, i∈I;

[0091] S207, substituting the K×2 matrix of each channel into a Gaussian layer with Q cores to generate a K×Q influence surface matrix, where Q represents the selected Gaussian mode order;

[0092] S208, for each channel i, the influence surface matrix is ​​convolved to obtain a K×1 influence surface coordinate matrix [ IS i 1 ,…IS i k ,…IS i K ] T , where the convolution kernel is a J×1 matrix of all 1s;

[0093] S209, the influence surface coordinate matrix of each channel i is convolved with a kernel of 1×1 to obtain a K×1 bridge sub-response time history matrix [ R i 1 ,…R i k ,…R i K ] T , where the convolution kernel of the i-th channel is the training wheel load w of the i-th training wheel axle i ;

[0094] S210, for all channels, the total response time history matrix of the bridge is obtained by accumulation [ R 1 ,…R k ,…R K ] T .

[0095] Further, in step S207, the matrix element in the kth row and qth column of the influence surface matrix of the i-th channel is

[0096] .

[0097] Furthermore, in step S40, a multi-Gaussian model is used to establish a real traffic load digital model.

[0098] In the specific implementation, the steps of obtaining the dynamic weighing model include:

[0099] Obtain the traffic information of each training vehicle within the target duration, divide the target duration evenly into K time nodes, and form a K×1 time series matrix [ t 1 ,…t k ,…t K ] T , t k yes k Ordinal time nodes, for example, dividing the target duration into [0.01s, 0.02s, 0.03s, …, 2s];

[0100] Due to the complex situations such as acceleration and deceleration in the longitudinal direction, it is considered that the time relationship curve between the training ordinate and the target duration is composed of 5 (J equals 5) time-course small curves (training longitudinal time-course sub-curves), and each training longitudinal time-course sub-curve has different parameters. At this time, each training longitudinal time-course sub-curve is a branch of the time-course large curve (training longitudinal time-course total curve) under different modes. The general expression of the time-course small curve is: ,in , , For modal j The undetermined coefficients of the sigmoid layer; taking the time information matrix as input, each modality will obtain a training longitudinal position matrix [ x 1,m ,… x k,m ,…x K,m ]T , a total of 5 modes are output as a K×5 matrix; then each mode is added, that is, the convolution of the all-1 matrix with a product kernel of J×1 is performed to obtain the training ordinate matrix of each training vehicle [ x 1 ,…x k ,…x K ] T ;

[0101] Since the lateral movement of the vehicle usually changes little, it is assumed that the time relationship between the training lateral coordinates of the training vehicle and the target duration is a linear relationship. When the time information matrix is ​​used as input, a training horizontal coordinate matrix will be obtained; specifically, a fully connected layer is used for connection, and the time series matrix [ t 1 ,…t k ,…t K ] T As input, we get the training lateral coordinate matrix [ y 1 ,…y k ,…y K ] T ,in, y k The training lateral coordinates of each training vehicle at the kth time node;

[0102] The training horizontal and vertical coordinate information (training horizontal coordinates and training vertical coordinates) is combined to obtain a K×2 matrix (training horizontal and vertical coordinate matrix); a training vehicle has a training horizontal and vertical coordinate matrix; assuming that the vehicle model of the training vehicle is the number of training wheels is I, the training horizontal and vertical coordinate matrix is ​​convolved with a 2×2 convolution kernel to obtain I channels, each channel is a K×2 matrix, and the elements of each kernel are, where, D i,x and D i,x It is i The lateral and longitudinal distances of each wheel load (training wheel set / training axle) from the front of the vehicle, and finally the matrix of each channel corresponds to the lateral and longitudinal coordinate matrix of each training wheel set;

[0103] For each channel K Substitute the matrix of ×2 into the Gaussian layer with Q kernels to generate K × Q The influence surface matrix, Q Represents the total number of selected Gaussian mode orders. For the matrix of the i-th channel,k Row, No. q The matrix elements of the columns are, where, , represents the ordinate and abscissa of the i-th training wheel group, , represents the core area position under the q-order mode, q∈Q; innovatively proposed that the influence surface is composed of the superposition of sub-surfaces under several modes, each sub-surface corresponds to a Gaussian layer, and then the influence surface coordinates of the corresponding training wheel group under Q modes are obtained; for each channel i, the influence surface matrix is ​​convolved to obtain a K×1 influence surface coordinate matrix [ IS i 1 ,…IS i k ,…IS i K ] T , where the convolution kernel is a J×1 matrix of all 1s; finally, I channels are obtained, each of which is a K×1 matrix; then, the influence surface coordinates of all Q-order modes are accumulated to obtain the final influence surface coordinate matrix;

[0104] The influence surface coordinate matrix of each channel is convolved with a kernel of 1×1 to obtain a K×1 bridge sub-response time history matrix [ R i 1 ,…R i k ,…R i K ] T , and finally we get I channels, each channel is a K×1 matrix, and the convolution kernel of the i-th channel is the wheel load w of the i-th training wheel i ; Then calculate the bridge sub-response corresponding to each training wheel set;

[0105] For all channels, the total bridge response is the superposition of each sub-response, and the K×1 total bridge response time history matrix is ​​obtained by accumulation. R 1 ,…R k ,…R K ] T ;

[0106] The error between the calculated total response time-history matrix of the bridge and the time-history matrix actually measured in the training data is minimized by taking the training vehicle wheel load weight and training position information parameters as variable parameters, and finally obtaining the weight, axle position and wheelbase information of each axle load in a given response.

[0107] In the present invention, the training of the model is to train the parameters of the Gaussian layer, and the Gaussian layer parameters of the model are trained by knowing the training wheel load weighing weight, training vehicle total weight, training lateral position, training longitudinal position, training head spacing and training driving speed of each vehicle in the training set. During the weighing calculation, the Gaussian layer parameters are not updated.

[0108] It is better to use multi-Gaussian model to establish the digital model of real traffic load. , assuming that the probability of a certain characteristic of traffic load (such as speed) at a certain value x is f(x) , using a multi-Gaussian model for fitting, where j, σ and μ are the coefficients of the Gaussian function under the corresponding mode.

[0109] The present invention also provides a system for determining a bridge load limit based on actual traffic load, comprising a processing device, wherein the processing device is used to implement the steps of the method for determining a bridge load limit based on actual traffic load as described above.

[0110] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for determining the bridge load limit based on the actual traffic load are implemented.

[0111] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for determining bridge load limit based on actual traffic load, characterized in that: The steps include: S10, collecting passing vehicle feature information through a computer vision device, wherein the passing vehicle feature information includes the number of photographed passing vehicles, the photographed vehicle model of each vehicle, and the photographed and identified axle number; If the actual axle weight of the axle identified by the photographing is not less than the preset axle weight threshold, the photographing and identifying axle is determined to be a true axle; if the actual axle weight of the axle identified by the photographing is less than the preset axle weight threshold, the photographing and identifying axle is determined to be a false axle; S20, using the assumed vehicle driving information to obtain assumed bridge strain data based on a dynamic weighing model, and determining the assumed vehicle driving information corresponding to the difference between the assumed bridge strain data and the actual bridge strain data within a first threshold range as estimated vehicle driving information; the estimated vehicle driving information includes an estimated single wheel weight, an estimated vehicle gross weight, an estimated lateral position, an estimated longitudinal position, and an estimated wheel-to-head distance of each vehicle passing through; The Gaussian layer parameters of the training model are trained by knowing the training wheel load weight, training vehicle total weight, training lateral position, training longitudinal position, training head spacing and training driving speed of each vehicle in the training set; Using the time information matrix as input, the training longitudinal position matrix is ​​obtained; Using the time information matrix as input, the training horizontal coordinate matrix is ​​obtained; S30, fusing the passing vehicle characteristic information and the estimated vehicle travel information to obtain actual passing vehicle information, the actual passing vehicle information including the actual number of passing vehicles and the actual number of effective axles of each vehicle, the actual axle weight of a single axle, the actual gross vehicle weight, the actual lateral position, the actual longitudinal position, the actual axle spacing and the actual travel speed; S40, using actual passing vehicle information within the target time period to establish a real traffic load digital model; S50, generating a test simulated traffic flow using a real traffic flow load digital model, and obtaining an overload violation rate of overloaded vehicles in the test simulated traffic flow based on a current load limit value of the bridge; S60, using test simulation of traffic flow and using influence surface loading to obtain the test strain history of the bridge; S70, calculate the equivalent stress amplitude and mean stress value based on the test strain history using the rain flow counting method; S80, fatigue life calculation using equivalent stress amplitude and mean stress value; S91, if the fatigue life is not less than the expected service life, the current load limit value meets the requirement; S92, if the fatigue life is less than the expected service life, the current load limit value is adjusted with a preset step size to correct the overloading violation rate of the test simulated traffic flow, and enter step S60 until the fatigue life is not less than the expected service life.

2. The method for determining bridge load limit based on actual traffic load according to claim 1, characterized in that: In step S80, the fatigue life of the bridge is verified by using the Palmgren-Miner linear cumulative damage theory to calculate the stress amplitude spectrum.

3. The method for determining bridge load limit based on actual traffic load according to any one of claims 1 or 2, characterized in that: The steps to establish the dynamic weighing model include: S201, obtaining training traffic information of each training vehicle within a training duration, and evenly dividing the training duration into K training time nodes; S202, obtaining J training longitudinal time travel curves of each training vehicle within the training duration; S203, based on the training longitudinal time travel sub-curves, obtaining a training longitudinal time travel total curve corresponding to each training vehicle within the training duration, and obtaining a training ordinate matrix consisting of training time nodes for each training vehicle; S204, obtaining the training transverse coordinates of each training vehicle at each training time node, and obtaining a training transverse coordinate matrix of each training vehicle formed by the training time nodes; S205, fusing the training ordinate matrix and the training transverse matrix of each training vehicle to obtain a training transverse and longitudinal matrix of each training vehicle; S206, obtaining a wheel transverse and longitudinal coordinate matrix corresponding to each training wheel group based on the training transverse spacing and the training longitudinal spacing of the training wheel group relative to the front of each training vehicle; S207, obtaining the sub-coordinates of the training influence surface of each training wheel group of the training vehicle in the Q-order mode; S208, obtaining the total coordinates of the training influence surface of each training wheel set of each training vehicle based on the sub-coordinates of the training influence surface; S209, obtaining a training bridge score response corresponding to each training wheel group in each training vehicle based on the training wheel load corresponding to the training wheel group; S210: Obtaining a total response of a training bridge composed of training wheel groups corresponding to all training vehicles based on all the training bridge sub-responses.

4. The method for determining bridge load limit based on actual traffic load according to claim 3 is characterized in that: S201, obtaining training pass information within the training duration, and evenly dividing the training duration into K training time nodes to form a K×1 training time sequence matrix [ t 1 ,…t k ,…t K ] T ,in, t k yes k Ordinal training time nodes; S202, training time series matrix [ t 1 ,…t k ,…t K ] T Substitute into the sigmoid layer with J cores to generate a K×J training transformation matrix; S203, performing a convolution operation on the training transformation matrix to obtain a K×1 training longitudinal coordinate matrix [ x 1 ,…x k ,… x K ] T , where the convolution kernel is a J×1 matrix of all 1s, x k is the training longitudinal coordinate of the training vehicle at the kth training time node; S204, training time series matrix [ t 1 ,…t k ,…t K ] T Connect a fully connected layer to get the training horizontal coordinate matrix [ y 1 ,…y k ,…y K ] T ,in, y k is the training lateral coordinate of the training vehicle at the kth training time node; S205: Merge the training longitudinal coordinates and the training transverse coordinates to obtain a K×2 training transverse and longitudinal coordinate matrix. , where one training vehicle corresponds to one training horizontal and vertical coordinate matrix.

5. The method for determining bridge load limit based on actual traffic load according to claim 4, characterized in that: The calculation formula for the k-th row and j-th column element of the training transformation matrix is ,in , , They are the unknown coefficients of the j-th sigmoid layer, k∈K, j∈J.

6. The method for determining bridge load limit based on actual traffic load according to claim 4, characterized in that: S206, assuming that the number of training wheels of the training vehicle is I, convolution calculation of the horizontal and vertical coordinate matrix of the training vehicle with a 2×2 convolution kernel is performed to obtain I channels, each channel is a K×2 matrix, and the convolution kernel is , D i,x and D i,y They are i The training lateral spacing and training longitudinal spacing of each tire from the front of the vehicle, i∈I; S207, substituting the K×2 matrix of each channel into a Gaussian layer with Q cores to generate a K×Q influence surface matrix, where Q represents the selected Gaussian mode order; S208, for each channel i, the influence surface matrix is ​​convolved to obtain a K×1 influence surface coordinate matrix [ IS i 1 ,…IS i k ,…IS i K ] T , where the convolution kernel is a J×1 matrix of all 1s; S209, the influence surface coordinate matrix of each channel i is calculated with a convolution kernel of 1×1 to obtain a K×1 bridge response time history matrix [ R i 1 ,…R i k ,…R i K ] T , where the convolution kernel of the i-th channel is the training wheel load of the i-th training wheel axle; S210, for all channels, the total response time history matrix of the bridge is obtained by accumulation [ R 1 ,…R k ,…R K ] T .

7. The method for determining bridge load limit based on actual traffic load according to claim 2, characterized in that: In step S40, a multi-Gaussian model is used to establish a real traffic load digital model.

8. A bridge load limit determination system based on actual traffic load, characterized in that: It comprises a processing device, which is used to implement the steps of the method for determining the bridge load limit based on actual traffic load as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for determining the bridge load limit based on actual traffic load as described in any one of claims 1 to 7 are implemented.

Images