Steel structure construction component tracking and tracing method based on internet of things

Abstract

The present application relates to steel structure construction traceability technical field, disclose a steel structure construction component tracking traceability method based on internet of things. The method collects the internet of things sensing information and construction process information of component and features correlation, constructs component tracking feature set; Based on the feature set, define tracking node, associated edge and edge weight, build multi-dimensional tracking network, combined with time sequence tracking network model learning node evolution law generates component tracking feature vector, time sequence tracking network model contains node coding layer, time sequence memory layer, attention module and the like; Train the traceability prediction model, input the feature vector and real-time construction data, output the current construction stage traceability state index of component; The construction state is divided into state group, and the group traceability threshold value and dynamic traceability threshold value are calculated; Compare the traceability state index with the dynamic traceability threshold value, and if it exceeds, generate adjustment instruction and trigger internet of things execution terminal action, regularly update the dynamic traceability threshold value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of steel structure construction traceability, in particular to a steel structure construction component tracking and traceability method based on the Internet of Things. BACKGROUND

[0002] Steel structures are widely used in modern construction engineering, and the tracking and traceability of their construction components are crucial for engineering quality, progress, and safety management. Traditional methods of tracking and traceability of steel structure construction components have many defects. Information collection mainly relies on manual recording, which is not only inefficient but also prone to human error, resulting in incomplete and timely construction component perception information and construction process information, which cannot accurately reflect the actual state of the components.

[0003] Existing tracking methods mostly lack deep correlation and multidimensional analysis of information, making it difficult to build a comprehensive and accurate tracking model and effectively capture the evolution of components during the construction process. In addition, traditional methods often lack dynamic adjustment mechanisms and cannot update tracking thresholds in a timely manner based on actual conditions during the construction process, resulting in inaccurate judgments of component traceability status and difficulties in timely identifying and solving problems. With the development of the Internet of Things technology, although some tracking methods based on the Internet of Things have been proposed, these methods mostly focus on single-dimensional information collection and processing, and cannot fully utilize the advantages of the Internet of Things technology to achieve multidimensional and full-process tracking and traceability of components.

[0004] There is an urgent need for a steel structure construction component tracking and traceability method that can comprehensively collect information, deeply analyze and correlate, and dynamically adjust thresholds to improve the efficiency and quality of construction management and ensure the smooth progress of the project. SUMMARY

[0005] The purpose of the present application is to provide a steel structure construction component tracking and traceability method based on the Internet of Things to solve the problems raised in the background.

[0006] To achieve the above purpose, the present application provides a steel structure construction component tracking and traceability method based on the Internet of Things, which comprises:

[0007] Collecting Internet of Things perception information and construction process information of steel structure construction components, correlating the perception information with the process information, and constructing a component tracking feature set;

[0008] Based on the component tracking feature set, defining tracking nodes, defining associated edges and edge weights based on construction process information, constructing a multidimensional tracking network, learning the evolution of tracking nodes in the multidimensional tracking network based on a time sequence tracking network model, and generating a component tracking feature vector;

[0009] The trace prediction model for predicting the real-time trace state of the component is trained, the component tracking feature vector and the real-time construction data are taken as the input data, and the trace state index of the component is taken as the output data, so as to predict the trace state index of the current construction stage of the steel structure construction component;

[0010] All construction states are divided into several state groups based on the component tracking feature vector, the group trace threshold of each state group is calculated, and the dynamic trace threshold of the construction state is calculated based on the group trace threshold and the matching degree between the component tracking feature vector of each construction state and the typical vector of the state group to which it belongs;

[0011] The trace state index output by the trace prediction model is compared with the dynamic trace threshold of the construction state, if the trace state index exceeds the dynamic trace threshold, the corresponding tracking adjustment instruction is generated, and the Internet of Things execution terminal is triggered to act, and the dynamic trace threshold of each construction state is updated regularly.

[0012] Preferably, the Internet of Things sensing information and construction process information of the steel structure construction component are collected, the sensing information is associated with the process information, and the specific method for constructing the component tracking feature set is as follows:

[0013] The sensing information and construction process information are obtained from the Internet of Things sensing terminal of the steel structure construction component, the sensing information includes: component unique identification code, installation coordinate positioning, material detection data; the construction process information includes: transportation timestamp, hoisting sequence record, welding quality log;

[0014] The feature attributes of the sensing information and the construction process information are extracted and associated to obtain the component tracking feature set of each construction state.

[0015] Preferably, based on the component tracking feature set, a tracking node is defined, an associated edge and an edge weight are defined according to the construction process information, and the specific method for constructing a multi-dimensional tracking network is as follows:

[0016] The sensing information is taken as the basic attribute of the tracking node, and the tracking node is composed of the component tracking feature set of each tracking node and the corresponding basic attribute;

[0017] The associated edge between the tracking nodes is constructed according to the construction process information, the time sequence correlation degree between the tracking nodes is calculated based on the time sequence record data in the construction process information; the time sequence record data of the tracking node p is the information sequence in the continuous construction period, the time sequence record data of the tracking node q is the information sequence in the corresponding period, the time sequence alignment algorithm is adopted, the adjustment weight is given to the time sequence data correlation degree of different periods based on the time attenuation parameter, and the time sequence correlation degree between the tracking node p and the tracking node q is calculated;

[0018] The time sequence correlation degree is converted into a tracking correlation degree;

[0019] According to the historical collaborative construction records of the components, the adjacent node set of the tracking nodes p and q in the collaborative construction network is obtained, according to the adjacent node set, the collaborative matching index between the tracking nodes is calculated, different weights are given to the nodes with different collaborative distances to obtain a weighted collaborative matching index, and the weighted collaborative matching index is normalized to the interval [0, 1] to obtain the collaborative correlation degree between the tracking nodes p and q;

[0020] Based on the tracking correlation degree and the collaborative correlation degree, the edge weight of the associated edge between the tracking nodes p and q is calculated, and the tracking nodes, the associated edge and the edge weight thereof form a multi-dimensional tracking network.

[0021] Preferably, the training method of the time sequence tracking network model is:

[0022] Step B1: constructing a time sequence tracking network model, including a node encoding layer, a time sequence memory layer, an attention module, an output layer and a parameter optimization module, taking the multi-dimensional tracking network at each time point as input and taking the evolution features of the tracking nodes at each time point as output;

[0023] The node encoding layer is used to map the features of the tracking nodes and the associated edges to a unified feature space; on the basis of the node encoding layer, the time sequence memory layer is introduced to capture the dynamic evolution mode of the tracking features, taking the multi-dimensional tracking network sequence at each time point as input and outputting the evolution features of the tracking nodes at each time point to depict the trajectory of the evolution features of the tracking nodes changing with time; the attention module is used to adaptively assign the influence weight of different adjacent nodes to highlight the role of key adjacent nodes; the output layer outputs the evolution features of the tracking nodes at each time point; and the parameter optimization module is used to calculate the gradient of the model parameters with respect to the loss function using the back propagation algorithm;

[0024] Step B2: for each tracking node p, constructing a positive sample set and a negative sample set based on the tracking correlation degree and the collaborative correlation degree, setting a correlation degree threshold and a matching degree threshold, taking the tracking nodes with a tracking correlation degree greater than the correlation degree threshold or a collaborative correlation degree greater than the matching degree threshold as positive samples, taking the tracking nodes not belonging to the positive sample set into the negative sample set, taking the positive samples and the negative samples as training data, using an unsupervised training method, taking the nodes in the multi-dimensional tracking network as input and taking the evolution features of the nodes as output, applying the time sequence tracking network model on the multi-dimensional tracking network, and forward propagating to generate the evolution features of all tracking nodes at each time point t;

[0025] Step B3: using the evolution characteristics of the tracking node p at time point t, calculate the feature similarity of the tracking node p with the tracking nodes in the positive sample set and the negative sample set, based on the feature similarity, calculate the contrast loss function at the current time point t, average the contrast loss functions of all tracking nodes and all time points to obtain the final loss function, and minimize the loss function as the training target;

[0026] Step B4: calculate the gradient of the loss function with respect to the model parameters by the back propagation algorithm, and update the model parameters using the optimizer;

[0027] Step B5: repeat steps B2 to B4 until the loss function converges.

[0028] Preferably, the specific method for learning the evolution law of the tracking nodes in the multi-dimensional tracking network based on the time series tracking network model to generate the component tracking feature vector is:

[0029] On the multi-dimensional tracking network, the evolution characteristics of the tracking nodes are learned by using the time series tracking network model, the evolution characteristics of the tracking node p at the nth layer are obtained by forward propagation, the attention module is introduced, and the attention weights of the adjacent tracking nodes with different collaborative relationships to the tracking node p are calculated;

[0030] The gating memory unit aggregation function is defined in the time series memory layer, which is used to fuse the evolution characteristics of the adjacent tracking node q at the current time point and the evolution characteristics of the tracking node p at the previous time point to obtain the evolution characteristics of the tracking node p at the current time point t;

[0031] The evolution characteristics of the tracking node p at time point t are spliced with the perception information of the tracking node to obtain the component tracking feature vector of the tracking node p.

[0032] Preferably, the specific method for training the trace prediction model for predicting the real-time trace state of the component is as follows: taking the component tracking feature vector and the real-time construction data as input data, and taking the component trace state index as output data, the trace prediction model is trained to predict the trace state index of the current construction stage of the steel structure construction component.

[0033] A batch of historical construction records of steel structure construction components are obtained, the component tracking feature vectors are extracted, the state index of abnormal trace is marked as an abnormal trace label, and the training samples are constructed based on the component tracking feature vectors and the abnormal trace label. The component tracking feature vector and the real-time construction data are used as input data, and the abnormal trace label is used as output data to train the trace prediction model;

[0034] For each training sample, the accurately predicted abnormal trace label is taken as the prediction target, the cross-entropy loss function is used as the loss function of the training model, the value of the loss function is minimized as the training target, and the training is completed when the loss function converges.

[0035] Using the trained traceability prediction model, for the current real-time construction state, the component tracking feature vector corresponding to the tracking node and the real-time construction data are extracted as input data, and the traceability state index of the real-time construction state is output.

[0036] Preferably, the component tracking feature vector divides all construction states into several state groups, calculates the group traceability threshold of each state group, and calculates the dynamic traceability threshold of each construction state based on the group traceability threshold and the matching degree between the component tracking feature vector of each construction state and the typical vector of the state group to which it belongs. The specific method is:

[0037] Based on the component tracking feature vector, all construction states are divided into several state groups by a density clustering algorithm;

[0038] For each state group, the component tracking feature vector and the historical construction record of the construction state in the state group are extracted, the component tracking feature vector of the construction state and each historical construction data are input into the traceability prediction model as input data, and the traceability state index of each historical construction is obtained;

[0039] The traceability state indexes of each historical construction of the construction state are counted to obtain the traceability state index distribution of the construction state, and the statistical characteristics of the traceability state index distribution of the construction state are calculated, including the average value and the dispersion;

[0040] The statistical characteristics of the traceability state index distribution of each construction state in the state group are calculated to obtain the group traceability threshold of the state group;

[0041] For each construction state, the matching degree between the component tracking feature vector and the typical vector of the state group to which it belongs is calculated, wherein the component tracking feature vector represents the characteristic set of the construction state, and the typical vector represents the typical characteristic set of the state group;

[0042] According to the matching degree between the component tracking feature vector of the construction state and the typical vector of the state group to which it belongs and the group traceability threshold, the dynamic traceability threshold of the construction state is calculated.

[0043] Preferably, the specific method for periodically updating the dynamic traceability threshold of each construction state is:

[0044] A fixed update period is set, and in each update period, the deviation data of the actual traceability state index and the dynamic traceability threshold of all construction states in the period are collected;

[0045] The number and amplitude of the deviation data exceeding the threshold are counted, and the dynamic traceability threshold of the construction state is adjusted in combination with the change trend of the component tracking feature vector in the current state group.

[0046] If the typical vector of the state group deviates significantly due to component loss or construction process adjustment, recalculate the dynamic traceability threshold of all construction states in the group.

[0047] Preferably, the specific method of generating corresponding tracking adjustment instructions is:

[0048] According to the type and degree of the traceability state index exceeding the dynamic traceability threshold, call the preset adjustment strategy table;

[0049] If the exceeding type is missing identification, generate component identification supplement adjustment instructions; if it is positioning deviation, generate installation coordinate correction adjustment instructions; if it is material abnormality, generate component batch replacement adjustment instructions.

[0050] Preferably, the tracking adjustment instruction contains specific parameter adjustment value and execution time node.

[0051] Compared with the prior art, the beneficial effects of the present application are:

[0052] By collecting Internet of Things sensing information and construction process information of steel structure construction components, and performing feature correlation on the two, a component tracking feature set is constructed, which can comprehensively and accurately obtain various information of the components in the construction process, and provides a rich and reliable data basis for subsequent tracking and tracing. Based on the component tracking feature set, tracking nodes are defined, associated edges and edge weights are defined according to the construction process information, and a multi-dimensional tracking network is constructed, so that the tracking network can comprehensively reflect the time sequence correlation and cooperative correlation between nodes in the construction process of the components, thereby more accurately depicting the evolution law of the components.

[0053] Learning the evolution law of the tracking nodes in the multi-dimensional tracking network by using the time sequence tracking network model, generating the component tracking feature vector, can deeply mine the dynamic change characteristics of the components in the construction process, and provides strong support for predicting the traceability state of the components. Training a traceability prediction model for predicting the real-time traceability state of the components, taking the component tracking feature vector and real-time construction data as input data, can predict the traceability state index of the components in the current construction stage in real time and accurately, and provides timely and accurate decision basis for construction management.

[0054] All construction states are divided into several state groups based on the component tracking feature vector, the group traceability threshold and the dynamic traceability threshold of each construction state are calculated, so that the traceability threshold can be dynamically adjusted according to the change of the construction state, and the accuracy and adaptability of the traceability judgment are improved. The traceability state index output by the traceability prediction model is compared with the dynamic traceability threshold of the construction state, and when it exceeds, the corresponding tracking adjustment instruction is generated and the Internet of Things execution terminal action is triggered, which can timely discover and solve the problems of the components in the construction process, and ensure the construction quality and progress of the components. Regularly updating the dynamic traceability threshold of each construction state can make the traceability management always adapt to various changes in the construction process, and maintain the accuracy and effectiveness of the traceability judgment.

[0055] Through the synergistic effect of the above links, the method realizes full-process, multi-dimensional tracking and dynamic management of steel structure construction components, improves the efficiency and accuracy of construction management, reduces construction cost and risk, and provides strong support for smooth construction and quality assurance of steel structure engineering. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 The working principle diagram of the steel structure construction component tracking and tracing method based on the Internet of Things is described.

[0057] Figure 2 The flowchart for multi-dimensional tracking network construction is described.

[0058] Figure 3 The flowchart for training the traceability prediction model is described.

[0059] Figure 4 The flowchart for updating the dynamic traceability threshold is described. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the 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.

[0061] Please refer to Figures 1-4 The present application provides a steel structure construction component tracking and tracing method based on the Internet of Things, and the specific implementation steps are as follows:

[0062] Collect the Internet of Things sensing information and construction process information of the steel structure construction components, associate the sensing information with the process information, and construct the component tracking feature set;

[0063] Based on the component tracking feature set, define the tracking node, define the associated edge and edge weight according to the construction process information, construct the multi-dimensional tracking network, learn the evolution law of the tracking node in the multi-dimensional tracking network based on the time sequence tracking network model, and generate the component tracking feature vector;

[0064] Train the trace prediction model for predicting the real-time trace state of the component, take the component tracking feature vector and real-time construction data as input data, and take the component trace state index as output data to predict the trace state index of the current construction stage of the steel structure construction component;

[0065] Based on the component tracking feature vector, all construction states are divided into several state groups, the group trace threshold of each state group is calculated, and based on the group trace threshold and the matching degree between the component tracking feature vector of each construction state and the typical vector of the state group it belongs to, the dynamic trace threshold of the construction state is calculated.

[0066] Compare the trace state index output by the trace prediction model with the dynamic trace threshold of the construction state. If the trace state index exceeds the dynamic trace threshold, generate the corresponding tracking adjustment instruction, and trigger the Internet of Things execution terminal action. Update the dynamic trace threshold of each construction state regularly.

[0067] Embodiment 1

[0068] When constructing the component tracking feature set, the perception information and construction process information need to be obtained from the Internet of Things perception terminal of the steel structure construction component. Among them, the perception information includes component unique identification code, installation coordinate positioning, and material detection data. The component unique identification code is generated by the RFID tag. The tag gives each component a globally unique identity through a specific coding rule, such as using the EPC coding system, which includes manufacturer identification code, object classification code, and serial number fields, which can accurately distinguish different components. The installation coordinate positioning is obtained with the help of the Beidou satellite navigation system. Through the signal transmission and positioning solution of multiple satellites, the position of the component in the construction site can be accurately positioned to the centimeter level. The specific acquisition method is to install a Beidou positioning module on the component, receive satellite signals in real time, and solve the three-dimensional coordinate values (X, Y, Z). The material detection data is collected by professional equipment such as a spectrum analyzer. The spectrum analyzer irradiates the surface of the component with light of a specific wavelength, analyzes the spectral characteristics of the reflected light, and obtains the material composition of the component, such as the content of elements such as iron, carbon, and manganese, as well as mechanical performance indicators such as strength and hardness.

[0069] The construction process information includes transportation time stamp, hoisting sequence record, and welding quality log. The transportation time stamp is recorded by the logistics management system, and when the component passes through each logistics node during transportation, the system automatically collects and stores the time information, which is accurate to seconds, such as the time when the component leaves the warehouse and the time when it arrives at the construction site. The hoisting sequence record is input by the construction personnel through the handheld terminal. During the hoisting operation of the component, the construction personnel inputs the hoisting number and corresponding hoisting time of each component in sequence according to the construction plan and the actual situation on site, and records the hoisting sequence of the component in detail. The welding quality log is automatically generated by the welding quality monitoring equipment. The equipment monitors parameters such as welding current, voltage, and temperature in real time during welding, such as the size of the welding current, the voltage fluctuation, and the temperature change curve of the welding area, and records these parameters in chronological order to form a complete welding quality log.

[0070] The feature attributes of the extracted perception information and construction process information are associated. For the unique identification code of the component in the perception information, the feature attributes such as the manufacturer identification code, object classification code, and serial number in the code are extracted; the specific X, Y, and Z coordinate values of the installation coordinate positioning are extracted; and the feature attributes such as the content ratio, strength value, and hardness value of each element of the material detection data are extracted. The specific time information of the transportation time stamp in the construction process information is extracted; the hoisting number and corresponding hoisting time of the hoisting sequence record are extracted; and the feature attributes such as the specific value of the welding current, the specific value of the voltage, and the change range of the temperature of the welding quality log are extracted.

[0071] These extracted feature attributes are associated, taking the unique identification code of the component as the core, associating it with the X, Y, and Z coordinate values of the corresponding installation coordinate positioning to determine the specific location of the component in the construction site; associating it with the feature attributes such as the element content, strength, and hardness of the material detection data to understand the material properties of the component; associating it with the specific time information of the transportation time stamp to master the transportation time node of the component; associating it with the hoisting number and hoisting time of the hoisting sequence record to clearly understand the hoisting sequence and time of the component; and associating it with the parameters such as welding current, voltage, and temperature of the welding quality log to know the quality situation of the component during the welding process.

[0072] In this way, a component tracking feature set of each construction state is formed. For example, in the transportation stage, the component tracking feature set of a certain component includes its unique identification code, the specific time of the transportation timestamp, the coordinate value of the warehouse in the installation coordinate positioning, material detection data, etc.; in the hoisting stage, the component tracking feature set of the component includes its unique identification code, the hoisting number and time recorded in the hoisting sequence, the coordinate value of the hoisting position in the installation coordinate positioning, the transportation timestamp, etc. The component tracking feature set of each construction state covers all key information of the component in that construction stage, and these information are correlated to form a complete feature data set, which provides rich feature data support for subsequent tracking and tracing of the component, so that in the subsequent tracking process, the state and information of the component in each construction stage can be accurately understood through these feature data.

[0073] Embodiment 2

[0074] In building a multi-dimensional tracking network, the perception information needs to be taken as the basic attribute of the tracking node, and each tracking node is composed of its component tracking feature set and the corresponding basic attribute. Taking a steel structure construction component as an example, the unique identification code of the component in the perception information is “GS-001-20250702-001”, which is generated by a specific rule, where “GS” represents steel structure, “001” is the manufacturer code, “20250702” indicates the production time, and “001” is the serial number, which is unique. The installation coordinate positioning is obtained by the Beidou navigation system, and the three-dimensional coordinate value (100.234, 200.567, 10.345) is obtained, which accurately reflects the position of the component in the construction site. The material detection data is collected by a spectrum analyzer, which shows that the component is Q355B steel material with carbon content of 0.2%, manganese content of 1.4%, yield strength of 355 MPa, tensile strength of 510 MPa, etc. These perception information constitutes the basic attribute of the tracking node, and combined with the construction process information of the component in the transportation, hoisting, welding and other stages to form a component tracking feature set, which together constitutes a complete tracking node.

[0075] According to the construction process information, the association edges between the tracking nodes are constructed. First, the time sequence correlation degree is calculated. Taking tracking node p and tracking node q as an example, the time sequence record data of tracking node p is the information sequence in a certain continuous construction period, such as in the transportation stage, the transportation time stamp from 8:00 to 9:30 on July 2, 2025 is 8:00 (out of the warehouse), 8:30 (passing through node A), 9:30 (arriving at the site), and the corresponding transportation vehicle information, etc.; the time sequence record data of tracking node q in the corresponding period is 8:15 (another component out of the warehouse), 9:00 (passing through node A), and 9:45 (arriving at the site). The time sequence alignment algorithm (such as dynamic time warping algorithm) is used to align the two time sequence record data, and the events in different periods are corresponded in time sequence. Then, the association degree of time sequence data in different periods is given an adjustment weight based on the time decay parameter. The time decay parameter can be set to decrease exponentially with the increase of time interval, for example, the weight is multiplied by 0.8 every 1 hour from the current time. In this way, the time sequence correlation degree of tracking nodes p and q in the transportation period is calculated. Specifically, the time interval of each event is determined first, then the weight is adjusted according to the decay parameter, and finally the overall time sequence correlation degree is obtained by comprehensively considering the correlation degrees of all events.

[0076] The time sequence correlation degree is converted into tracking correlation degree. The conversion method can use a linear function, such as tracking correlation degree = time sequence correlation degree x 0.7 + 0.3 (where 0.7 and 0.3 are empirical coefficients, which can be adjusted according to the actual construction situation), so that the tracking correlation degree is in the interval [0, 1], which is convenient for subsequent calculation.

[0077] The collaborative correlation degree is calculated. According to the historical collaborative construction record of the component, the adjacent node set of tracking nodes p and q in the collaborative construction network is obtained. The adjacent node set includes other nodes that have cooperated with p and q, for example, in the hoisting stage, nodes p and q may have a collaborative relationship with hoisting equipment nodes, command nodes, etc. According to the adjacent node set, the collaborative matching index between the tracking nodes is calculated, which considers factors such as the number of common adjacent nodes and the frequency of collaborative construction. For example, if p and q have 3 common adjacent nodes, and have cooperated 6 times in the past 10 constructions, the collaborative matching index can be set to (3 / maximum number of common adjacent nodes) x 0.5 + (6 / 10) x 0.5. Different weights are given to nodes with different collaborative distances. The weight of nodes with short collaborative distance (such as directly adjacent collaborative nodes) is set to 0.8, and the weight of nodes with long collaborative distance (such as indirectly collaborative nodes) is set to 0.2, to obtain the weighted collaborative matching index. Then the weighted collaborative matching index is mapped to the interval [0, 1] through a normalization function, for example, using the Min-Max normalization method, to obtain the collaborative correlation degree between tracking nodes p and q.

[0078] The edge weight of the associated edge is calculated based on the tracking correlation degree and the collaborative correlation degree, and the calculation of the edge weight adopts a weighted summation manner, such as edge weight = tracking correlation degree x 0.6 + collaborative correlation degree x 0.4 (wherein 0.6 and 0.4 are weight coefficients, which can be adjusted according to the importance of the time sequence correlation and the collaborative correlation in the construction process). All the tracking nodes, the associated edges and the corresponding edge weights are combined together to form a multi-dimensional tracking network. In the network, each node contains the sensing information and the construction process characteristics of the component, and the edges between the nodes reflect the connection strength through the time sequence correlation degree and the collaborative correlation degree, thereby comprehensively reflecting the time sequence relationship and the collaborative operation relationship of the components in the construction process.

[0079] For example, in the hoisting link of steel structure construction, the tracking nodes of multiple components form time sequence associated edges through the time sequence of hoisting order, and at the same time form collaborative associated edges due to the common use of the same crane, and the edge weights are calculated according to the respective correlation degrees. The multi-dimensional tracking network constructed in this way can depict the mutual relationship of the components in the construction process from multiple dimensions, and provide structured data support for subsequent model learning of the evolution law of the tracking nodes based on the time sequence tracking network, so that the model can better understand the state change and mutual influence of the components in the construction process. In the entire construction process, the parameter settings of each step are determined based on the actual construction situation and historical data, so as to ensure that the network structure can accurately reflect the real relationship of the construction process.

[0080] Embodiment 3

[0081] When training the time sequence tracking network model, a time sequence tracking network model needs to be constructed, which includes a node encoding layer, a time sequence memory layer, an attention module, an output layer and a parameter optimization module. The node encoding layer adopts a multi-layer perception structure, which is used to map the features of the tracking nodes and the associated edges to a unified feature space. For example, for the component unique identification code, installation coordinate positioning and other features of the tracking nodes, through linear transformation and activation function in the node encoding layer, they are converted into feature vectors with the same dimension. The time sequence memory layer is introduced on the basis of the node encoding layer, which can adopt a long short-term memory network (LSTM) structure, takes the multi-dimensional tracking network sequence of each time point as input, saves historical information through the memory unit, thereby capturing the dynamic evolution mode of the tracking features, and outputs the evolution features of each tracking node at each time point, so as to depict the trajectory of the evolution features of the tracking nodes changing with time. The attention module is used to adaptively assign the influence weight of different adjacent nodes, and highlights the role of key adjacent nodes by calculating the feature similarity and other indicators between adjacent nodes and the current node. The output layer is used to output the evolution features of each tracking node at each time point, and the parameter optimization module calculates the gradient of the model parameters with respect to the loss function using the back propagation algorithm.

[0082] For each tracking node p, it is necessary to construct a positive sample set and a negative sample set based on the tracking correlation and the collaborative correlation. First, set the correlation threshold θ1 and the matching threshold θ2, which can be determined according to the construction history data and actual demand. The tracking nodes with tracking correlation greater than θ1 or collaborative correlation greater than θ2 with the tracking node p are regarded as positive samples, and the tracking nodes not belonging to the positive sample set are included in the negative sample set. Then, taking the positive samples and the negative samples as training data, using unsupervised training method, taking the nodes in the multi-dimensional tracking network as input and the evolution features of the nodes as output, applying the time series tracking network model on the multi-dimensional tracking network, and generating the evolution features of all tracking nodes at each time point t through forward propagation.

[0083] Using the evolution features of the tracking node p at the time point t, the feature similarity of the tracking node p and the tracking nodes in the positive sample set and the negative sample set is calculated. The feature similarity is calculated by using cosine similarity, and the formula is:

[0084] wherein, represents the evolution feature vector of the tracking node p and the feature vector of the tracking node q cosine similarity of is the dot product of two vectors; and are the L2 norms of the vectors and .

[0085] Based on the feature similarity, the contrast loss function at the current time point t is calculated. The expression of the contrast loss function is:

[0086]

[0087] wherein, is the contrast loss function of the tracking node p at the time point t; is the total number of samples; is the label, when q is a positive sample , otherwise ; is the feature distance between the tracking node p and q, ; is a preset distance margin.

[0088] The average value of the contrast loss functions of all tracking nodes and all time points is obtained, and the final loss function is obtained. Taking minimizing the loss function as the training target. The gradient of the loss function with respect to the model parameters is calculated by the back propagation algorithm, and the model parameters are updated using the optimizer (such as stochastic gradient descent method).

[0089] Taking a certain steel structure construction project as an example, it is assumed that there are 100 tracking nodes, and the time point t is from 1 to 10. For each tracking node p, at time point t = 1, its positive sample set may contain 20 tracking nodes, and its negative sample set contains 80 tracking nodes. After obtaining the evolution characteristics of each node through forward propagation, the similarity of each positive sample and negative sample with the characteristics of p is calculated, and then the loss value is calculated by substituting it into the contrast loss function. Assuming that the loss value calculated at this time is 0.8, after back propagation and parameter update, the loss value may decrease to 0.75 again at time point t = 1. Repeat the above steps of constructing sample set, forward propagation, calculating loss function, and updating parameters through back propagation until the loss function converges, for example, when the change amplitude of the loss function in the last 5 iterations is less than 0.001, it is considered that the model training is completed.

[0090] During the entire training process, the parameters of the node encoding layer, the weights of the time sequence memory layer, and the coefficients of the attention module will be continuously adjusted with iterations, so that the model can better learn the evolution rules of the tracking nodes in the multi-dimensional tracking network. For example, the weight of the forgetting gate in the time sequence memory layer will be adjusted according to the feedback of the loss function, so as to more reasonably forget unimportant historical information and retain key evolution characteristics. The weight coefficients in the attention module will also be gradually optimized, so that the model can more accurately identify the adjacent nodes that have a greater impact on the current node evolution, thereby improving the training effect of the model.

[0091] Embodiment 4:

[0092] When generating the tracking feature vector of the component on the multi-dimensional tracking network, the trained time sequence tracking network model is needed to learn the evolution characteristics of the tracking node. Taking a steel beam component in a certain steel structure construction project as an example, the component is divided into multiple tracking nodes during the construction process, and each tracking node corresponds to a different construction stage, such as raw material arrival, processing and manufacturing, transportation, hoisting, welding, etc.

[0093] The multi-dimensional tracking network is input into the trained time sequence tracking network model, and the evolution characteristics of tracking node p at the nth layer are obtained through forward propagation processing. Assuming that n = 3, the evolution characteristics of the 3rd layer have fused the multi-layer feature information of the node in the network, including the perception information from the bottom layer to the time sequence correlation and collaborative correlation features of the high layer. For example, the tracking node p at the transportation stage may contain the time sequence characteristics of the transportation timestamp, the collaborative correlation characteristics with other transportation component nodes, and the abstract representation of the basic features of the component itself, such as material and identification.

[0094] An attention module is introduced to calculate the attention weight of the adjacent tracking node of different collaborative relationships to the tracking node p. The adjacent tracking node includes other component nodes, transportation vehicle nodes, logistics nodes, etc. that have a collaborative relationship with the component node during transportation. The attention module determines the weight of each adjacent node according to the collaborative correlation degree of these adjacent nodes and the tracking node p, historical collaborative construction records, etc. For example, the component node q has a high collaborative correlation degree with the tracking node p in the same transportation batch and multiple collaborative transportation, and the attention module gives it a higher attention weight, such as 0.7. The component node r has only occasional collaborative transportation with the tracking node p, and the collaborative correlation degree is low, and the attention weight may be 0.3. In this way, the model can focus on the adjacent nodes that have a greater impact on the evolution of the tracking node p.

[0095] In the time sequence memory layer, a gating memory unit aggregation function is defined to fuse the evolution features of the adjacent tracking node q at the current time point and the evolution features of the tracking node p at the previous time point. Taking time point t as an example, the evolution features of the adjacent tracking node q at the current time point include its state information at time t, such as the transportation state change from the previous time point t-1 to t; the evolution features of the tracking node p at the previous time point record the state of p at time t-1. The gating memory unit aggregation function controls the inflow and outflow of information through a gating mechanism. For example, the input gate determines how much information in the evolution features of the current adjacent node q can flow into the evolution features of the current time point p, the forgetting gate determines how much information in the evolution features of the previous time point p needs to be forgotten, and the output gate determines which information in the final evolution features of the current time point p needs to be output. Through this fusion, the evolution features of the tracking node p at the current time point t are obtained, which not only contains the latest information of the current adjacent node, but also retains the key information in the historical evolution process. For example, at time t in the transportation stage, the evolution features of the tracking node p fuse the transportation position change information of the adjacent node q at time t and the transportation state information of p at time t-1, thereby forming the complete transportation state evolution features of p at time t.

[0096] The evolution feature of the tracking node p at time point t is spliced with the perception information of the tracking node. The perception information of the tracking node includes basic attributes such as component unique identification code, installation coordinate positioning, material detection data, etc. For example, the evolution feature of the tracking node p is a 128-dimensional vector, which contains the time sequence and coordination features of the transportation stage, while the component unique identification code in the perception information can be converted into a 64-dimensional one-hot encoding vector, the installation coordinate positioning is a 3-dimensional coordinate value vector, and the material detection data is a 16-dimensional material composition and mechanical property vector. These vectors are spliced together in order to form a 128+64+3+16=211-dimensional component tracking feature vector. The vector integrates the evolution feature of the tracking node p at time point t and its inherent perception information, and can comprehensively reflect all key features of the component in the current construction state.

[0097] Taking the process of transporting the steel beam component from the processing plant to the construction site as an example, at time point t1 when the transportation starts, the component tracking feature vector generated through the above steps contains the unique identification code of the component “GL-005-20250705”, the initial installation coordinate positioning (coordinate value of the processing plant warehouse), the material detection data (composition and strength of Q355B steel), and the evolution feature at the start of transportation (such as the coordination association feature with the transportation vehicle node, the initial time sequence feature of the transportation timestamp). As the transportation process progresses, at time point t2, the evolution feature of the tracking node p integrates the position change information and transportation state of the adjacent other transportation component nodes at t2, and the transportation state of p at t1, and is spliced with the current perception information (such as the real-time updated installation coordinate positioning, i.e. the position coordinate in the transportation process) to generate a new component tracking feature vector. In this way, the component tracking feature vector generated at each time point can dynamically reflect the state change of the component in the construction process, providing accurate and comprehensive input features for the subsequent traceability prediction model, so that the model can predict the real-time traceability state of the component based on these feature vectors. In the whole process, the processing of each step is based on actual construction data and model training results, ensuring that the generated component tracking feature vector can truly and effectively represent the construction state of the component.

[0098] Example 5:

[0099] In calculating the dynamic traceability threshold, first, based on the component tracking feature vector, all construction states are divided into several state groups through the density clustering algorithm. Taking a steel structure construction project as an example, the project contains various types of components, such as steel beams, steel columns, supports, etc. Each component will generate a large amount of construction state data during the construction process, and the component tracking feature vectors corresponding to these data form the basis for clustering. The DBSCAN density clustering algorithm is used, which divides groups according to the density distribution of sample points, and divides sample points with reachable density in the feature space into the same state group. For example, clustering the feature vectors corresponding to the processing and manufacturing state, transportation state, hoisting state, etc. of the steel beam may obtain the processing and manufacturing state group, transportation state group, hoisting state group, etc.

[0100] For each state group, the component tracking feature vector and historical construction record of the construction state need to be extracted. Taking the transportation state group as an example, this group contains the construction states of all components during the transportation phase, and each construction state corresponds to a component tracking feature vector, and also contains the historical construction record of the component during the transportation phase, such as transportation time, transportation route, environmental data during transportation, etc. The component tracking feature vectors of these construction states and each historical construction data are input into the trained traceability prediction model as input data, and the traceability prediction model outputs the traceability state indicators of each historical construction. These traceability state indicators reflect the traceability of the construction state in historical construction, such as transportation delay, component damage, etc.

[0101] The traceability state indicators of each historical construction of each construction state are counted to obtain the traceability state indicator distribution of the construction state. For example, a component has 100 historical construction records in the transportation phase, and its traceability state indicators may include transportation time deviation, component positioning deviation, etc. Statistical analysis of these indicators may find that the transportation time deviation presents a normal distribution, with a mean of 15 minutes and a standard deviation of 5 minutes. The statistical characteristics of the traceability state indicator distribution of the construction state are calculated, including the mean and the dispersion. The mean can reflect the overall level of the construction state traceability state indicators, such as the mean transportation time deviation of 15 minutes; the dispersion can be represented by the standard deviation, which reflects the fluctuation degree of the traceability state indicators, such as the standard deviation of 5 minutes, indicating that the transportation time deviation fluctuates between 10 to 20 minutes more often.

[0102] The statistical characteristics of the traceability state index distribution of each construction state in the state group are calculated, and then the group traceability threshold of the state group is obtained by synthesizing these statistical characteristics. The determination of the group traceability threshold needs to consider the statistical characteristics of all construction states in the state group. For example, the group traceability threshold can be set as the average value plus a certain multiple of the standard deviation. Assuming that the average value of the transportation time deviation of all construction states in the transportation state group is 20 minutes, the standard deviation is 8 minutes, and the multiple is 2, the group traceability threshold can be set to 20+2×8=36 minutes, that is, when the transportation time deviation exceeds 36 minutes, it is considered that the traceability state of the construction state may be abnormal.

[0103] For each construction state, the matching degree between its component tracking feature vector and the typical vector of the state group to which it belongs is calculated. The typical vector can be determined by calculating the average value of the component tracking feature vectors of all construction states in the state group, which represents the typical feature set of the state group. The component tracking feature vector represents the feature set of the construction state. By calculating the matching degree between the two vectors, the similarity of the construction state to the typical features of the state group can be reflected. The matching degree can be calculated by using the cosine similarity method. For example, the cosine similarity between the component tracking feature vector of a certain construction state and the typical vector of the state group to which it belongs is 0.85, which means that the matching degree of the construction state to the typical features of the group is high.

[0104] According to the matching degree between the component tracking feature vector of the construction state and the typical vector of the state group to which it belongs and the group traceability threshold, the dynamic traceability threshold of the construction state is calculated. For example, when the matching degree is high, it means that the construction state is similar to the typical features of the group, and a more stringent dynamic traceability threshold can be used; when the matching degree is low, a relatively loose dynamic traceability threshold is used. The specific calculation method can be dynamic traceability threshold = group traceability threshold × (1-k×(1-matching degree)), where k is an adjustment coefficient, which can be set according to actual conditions. Assuming that the group traceability threshold is 36 minutes and k is 0.5, the matching degree of a certain construction state is 0.85, then dynamic traceability threshold = 36×(1-0.5×(1-0.85)) = 36×(1-0.075) = 36×0.925 = 33.3 minutes. In this way, the dynamic traceability threshold of the construction state is adjusted according to its matching degree with the typical features of the group, realizing the setting of the dynamic threshold.

[0105] In the actual construction process, as the construction proceeds, factors such as component loss and construction process adjustment can cause the typical vector of the state group to deviate significantly. Therefore, it is necessary to update the dynamic traceability threshold of each construction state regularly. A fixed update period is set, such as updating once a week. Within each update period, the deviation data of the actual traceability state indicators of all construction states in the period from the dynamic traceability threshold is collected. For example, within a week, the actual transportation time deviation of a certain construction state is 35 minutes, its dynamic traceability threshold is 33.3 minutes, and the deviation is 1.7 minutes. The number of times and the magnitude of the deviation data exceeding the threshold are counted, such as within a week, the actual traceability state indicators of the construction state exceed the dynamic traceability threshold 2 times, and the magnitudes are 1.7 minutes and 2.5 minutes respectively. In combination with the change trend of the component tracking feature vector in the current state group, such as finding that the component tracking feature vectors of multiple construction states have a trend of overall delay in the time characteristics of the transportation stage, it is indicated that there may be transportation route adjustment, traffic congestion and the like, and the dynamic traceability threshold needs to be adjusted. If the typical vector of the state group deviates significantly due to component loss or construction process adjustment, such as after the construction process adjustment, the hoisting time of the component is generally shortened, causing the typical vector of the hoisting state group to change significantly, then the dynamic traceability threshold of all construction states in the group needs to be recalculated to ensure that the dynamic traceability threshold can accurately reflect the current construction state. Through this regular updating method, the dynamic traceability threshold can be dynamically adjusted as the construction process changes, improving the accuracy and reliability of traceability prediction.

[0106] It should be noted that, in this document, the terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0107] Although the embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A steel structure construction component tracking and tracing method based on Internet of Things, characterized in that, The application relates to a steel structure construction component real-time tracing method based on an Internet of Things (IOT) and a construction process. The method comprises the following steps: collecting IOT sensing information and construction process information of a steel structure construction component, associating the sensing information with the process information, and constructing a component tracking feature set; Based on the component tracking feature set, a tracking node is defined, an associated edge and an edge weight are defined according to the construction process information, a multi-dimensional tracking network is constructed, an evolution law of the tracking node in the multi-dimensional tracking network is learned based on a time sequence tracking network model, and a component tracking feature vector is generated, the time sequence tracking network model comprises a node coding layer, a time sequence memory layer, an attention module, an output layer and a parameter optimization module, each time point multi-dimensional tracking network is taken as input, and the evolution feature of each time point tracking node is taken as output; A tracing prediction model for predicting a component real-time tracing state is trained, component tracking feature vectors and real-time construction data are taken as input data, and a component tracing state index is taken as output data, so that the tracing state index of the current construction stage of the steel structure construction component is predicted; Based on the component tracking feature vector, all construction states are divided into a plurality of state groups, a group tracing threshold of each state group is calculated, a dynamic tracing threshold of the construction state is calculated based on the group tracing threshold and the matching degree between the component tracking feature vector of each construction state and a typical vector of the state group to which the construction state belongs, and the typical vector is determined by calculating the average value of the component tracking feature vectors of all construction states in the state group; The tracing state index output by the tracing prediction model is compared with the dynamic tracing threshold of the construction state, if the tracing state index exceeds the dynamic tracing threshold, corresponding tracking adjustment instructions are generated, and an IOT execution terminal is triggered to act, and the dynamic tracing threshold of each construction state is regularly updated.

2. The IoT-based tracking and tracing method of steel construction members as claimed in claim 1, wherein, The specific method for collecting IOT sensing information and construction process information of a steel structure construction component, associating the sensing information with the process information, and constructing a component tracking feature set is as follows: Sensing information and construction process information are obtained from an IOT sensing terminal of a steel structure construction component, the sensing information comprises a component unique identification code, installation coordinate positioning and material detection data, and the construction process information comprises a transportation time stamp, hoisting sequence records and welding quality logs; The feature attributes of the sensing information and the construction process information are extracted and associated, and a component tracking feature set of each construction state is obtained. 3.The steel structure construction component tracking and tracing method based on the Internet of Things according to claim 2, wherein, The specific method for defining a tracking node based on the component tracking feature set, defining an associated edge and an edge weight based on the construction process information, and constructing a multi-dimensional tracking network is as follows: The sensing information is taken as basic attributes of the tracking node, and the tracking node is formed based on the component tracking feature set of each tracking node and the corresponding basic attributes; An associated edge between the tracking nodes is constructed according to the construction process information, and a time sequence correlation degree between the tracking nodes is calculated based on time sequence record data in the construction process information; the time sequence record data of the tracking node p is an information sequence in a continuous construction period, the time sequence record data of the tracking node q is an information sequence in a corresponding period, a time sequence alignment algorithm is adopted, an adjustment weight is given to the time sequence data correlation degrees of different periods based on a time attenuation parameter, and the time sequence correlation degree between the tracking node p and the tracking node q is calculated. convert the time sequence correlation degree into a tracking correlation degree; obtaining a set of adjacent nodes of the tracking nodes p and q in the collaborative construction network according to the historical collaborative construction records of the component, calculating a collaborative matching index between the tracking nodes according to the set of adjacent nodes, assigning different weights to the nodes with different collaborative distances to obtain a weighted collaborative matching index, and normalizing the weighted collaborative matching index to the interval [0, 1] to obtain a collaborative correlation degree between the tracking nodes p and q; calculating an edge weight of an associated edge between the tracking nodes p and q based on the tracking correlation degree and the collaborative correlation degree, and constructing a multi-dimensional tracking network with the tracking nodes, the associated edges and the edge weights.

4. The IoT-based tracking and tracing method of steel construction members as claimed in claim 3, wherein, The training method of the time sequence tracking network model is: Step B1: constructing a time sequence tracking network model, the node encoding layer is used to map the features of each tracking node and associated edge to a unified feature space; on the basis of the node encoding layer, a time sequence memory layer is introduced to capture the dynamic evolution pattern of the tracking features, and the multi-dimensional tracking network sequence at each time point is used as the input, and the evolution features of each tracking node at each time point are output, and the trajectory of the evolution features of the tracking nodes changing with time is described; using an attention module to adaptively assign the influence weights of different adjacent nodes to highlight the role of key adjacent nodes; the output layer outputs the evolution features of each tracking node at each time point; the parameter optimization module is used to calculate the gradient of the loss function with respect to the model parameters using the back propagation algorithm; Step B2: for each tracking node p, constructing a positive sample set and a negative sample set based on the tracking correlation degree and the collaborative correlation degree, setting a correlation degree threshold and a matching degree threshold, regarding the tracking nodes with a tracking correlation degree greater than the correlation degree threshold or a collaborative correlation degree greater than the matching degree threshold as positive samples, and regarding the tracking nodes not belonging to the positive sample set as negative samples, regarding the positive samples and the negative samples as training data, using an unsupervised training method, using the nodes in the multi-dimensional tracking network as the input, using the evolution features of the nodes as the output, applying the time sequence tracking network model on the multi-dimensional tracking network, and generating the evolution features of all tracking nodes at each time point t through forward propagation; Step B3: using the evolution features of the tracking node p at the time point t, calculating the feature similarity between the tracking node p and the tracking nodes in the positive sample set and the negative sample set, calculating a comparison loss function at the current time point t based on the feature similarity, averaging the comparison loss functions of all tracking nodes and all time points to obtain a final loss function, and minimizing the loss function as the training target; Step B4: calculating the gradient of the loss function with respect to the model parameters through the back propagation algorithm, and updating the model parameters using an optimizer; Step B5: repeating steps B2 to B4 until the loss function converges. 5.The steel structure construction component tracking and tracing method based on Internet of Things according to claim 4, wherein, The specific method for learning the evolution law of the tracking nodes in the multi-dimensional tracking network based on the time sequence tracking network model to generate the component tracking feature vector is: on the multi-dimensional tracking network, using the time sequence tracking network model to learn the evolution features of the tracking nodes, obtaining the evolution features of the tracking node p at the nth layer through forward propagation, introducing an attention module to calculate the attention weights of the adjacent tracking nodes with different collaborative relationships on the tracking node p; The time sequence memory layer defines a gating memory unit aggregation function for fusing the evolution characteristics of the adjacent tracking node q at the current time point and the evolution characteristics of the tracking node p at the last time point to obtain the evolution characteristics of the tracking node p at the current time point t; The evolution characteristics of the tracking node p at the time point t are spliced with the perception information of the tracking node to obtain the component tracking feature vector of the tracking node p. 6.The steel structure construction component tracking and tracing method based on the Internet of Things according to claim 5, wherein, The traceability prediction model for predicting the real-time traceability state of the component is trained, the component tracking feature vector and the real-time construction data are taken as input data, and the component traceability state index is taken as output data. A batch of historical construction records of steel structure construction components are obtained, the component tracking feature vector is extracted, the state index of abnormal traceability is marked as an abnormal traceability label, training samples are constructed based on the component tracking feature vector and the abnormal traceability label, the component tracking feature vector and the real-time construction data are taken as input data, and the abnormal traceability label is taken as output data to train the traceability prediction model; For each training sample, the accurate prediction of the abnormal traceability label is taken as a prediction target, a cross-entropy loss function is used as a loss function of the training model, the value of the loss function is minimized as a training target, and the training is completed when the loss function converges; The trained traceability prediction model is used, the component tracking feature vector and the real-time construction data of the corresponding tracking node are extracted as input data for the current real-time construction state, and the traceability state index of the real-time construction state is output. 7.The steel construction member tracking and tracing method based on the Internet of Things according to claim 6, wherein, The component tracking feature vector is used to divide all construction states into a plurality of state groups, the group traceability threshold of each state group is calculated, and the dynamic traceability threshold of each construction state is calculated based on the group traceability threshold and the matching degree between the component tracking feature vector of each construction state and the typical vector of the state group to which it belongs. All construction states are divided into a plurality of state groups based on the component tracking feature vector through a density clustering algorithm. For each state group, the component tracking feature vector and the historical construction record of the construction state are extracted, the component tracking feature vector of the construction state and each historical construction data are taken as input data to input the traceability prediction model, and the traceability state index of each historical construction is obtained. The traceability state index of each historical construction of the construction state is counted to obtain the traceability state index distribution of the construction state, and the statistical characteristics of the traceability state index distribution of the construction state are calculated, including the mean value and the dispersion. The statistical characteristics of the traceability state index distribution of each construction state in the state group are calculated to obtain the group traceability threshold of the state group. For each construction state, the matching degree between the component tracking feature vector and the typical vector of the state group to which it belongs is calculated, wherein the component tracking feature vector represents the feature set of the construction state, and the typical vector represents the typical feature set of the state group. The dynamic traceability threshold of the construction state is calculated according to the matching degree between the component tracking feature vector of the construction state and the typical vector of the state group to which it belongs and the group traceability threshold.

8. The IoT based tracking and tracing method of steel construction elements as claimed in claim 1 wherein, The specific method for periodically updating the dynamic tracing threshold of each construction state is as follows: A fixed updating period is set, and in each updating period, the deviation data of the actual tracing state index and the dynamic tracing threshold of all construction states in the period are collected; The number and amplitude of the deviation data exceeding the threshold are counted, and the dynamic tracing threshold of the construction state is adjusted in combination with the change trend of the component tracking feature vector in the current state group; If the typical vector of the state group significantly deviates due to component loss or construction process adjustment, the dynamic tracing threshold of all construction states in the group is recalculated. 9.The steel structure construction component tracking and tracing method based on Internet of Things according to claim 1, wherein, The specific method for generating corresponding tracking adjustment instructions is as follows: According to the type and degree of the tracing state index exceeding the dynamic tracing threshold, a preset adjustment strategy table is called; If the exceeding type is identification loss, an adjustment instruction for component identification supplement is generated; if it is positioning deviation, an adjustment instruction for installation coordinate correction is generated; if it is material abnormality, an adjustment instruction for component batch replacement is generated. 10.The steel structure construction component tracking and tracing method based on the Internet of Things according to claim 9, wherein, The tracking adjustment instruction includes specific parameter adjustment value and execution time node.

Images