A highway informationization wisdom management and control system

Abstract

The application discloses a highway informationization intelligent management and control system and relates to the technical field of intelligent construction. The system comprises the following steps: collecting multi-source data in real time, uploading the multi-source data to an edge computing node for time stamp alignment and format standardization, inputting the multi-source data into a space-time Transform fusion model, extracting space-time features of sensor data and semantic features of a video picture through a multi-head attention mechanism, generating a dynamic risk score, updating a virtual construction progress model of a digital twin, superimposing adjusted construction standards on a real scene through an AR terminal to guide workers to make deviation correction, and encrypting and storing closed-loop processing records to a blockchain. The application adopts a pulse coding driven distributed learning framework to greatly improve the abnormal detection efficiency, and through an event triggered feature compression and adaptive weight updating mechanism, realizes rapid response and global optimization of device cluster behavior.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent construction, and in particular to a highway information-based intelligent management and control system. BACKGROUND

[0002] In the field of intelligent construction, construction safety monitoring is achieved through multi-modal sensing technology, and the collaborative analysis of vibration, temperature and video data is the core evaluation method. The existing technology uses independent sensors and video systems to build a risk assessment model, and realizes multi-source data synchronization through linear interpolation or fixed time window, for example, fusing vibration spectrum and video target detection results to generate risk scores. However, such methods have significant defects in complex construction scenarios (such as multi-device collaborative operation, dynamic environmental interference).

[0003] The core problem of traditional methods is the insufficient spatio-temporal alignment accuracy of multi-source heterogeneous data: there is a millisecond clock deviation between sensor data and video stream, and existing interpolation algorithms cause distortion of key features in non-uniform sampling scenarios; at the same time, multi-modal fusion relies on static weight distribution, which cannot dynamically adapt to sudden changes in the construction environment, resulting in a significant increase in the false detection rate of abnormal detection. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a highway information-based intelligent management and control system to solve the problem of construction safety risk assessment lag and high false detection rate caused by insufficient spatio-temporal alignment accuracy of multi-source heterogeneous data.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] The present application provides a highway information-based intelligent management and control system, which comprises: real-time acquisition of multi-source data, uploading to an edge computing node for timestamp alignment and format standardization, inputting into a spatio-temporal Transformer fusion model, extracting spatio-temporal features of sensor data and semantic features of video pictures through a multi-head attention mechanism, and generating a dynamic risk score; triggering a hierarchical early warning signal based on the dynamic risk score, executing AR projection instructions, red alert pushing, rectification work order generation and metering payment locking; uploading the device cluster anomaly pattern features of the current project to the cloud through a federated learning platform, and comparing and learning with the global feature library shared by historical projects for cross-project optimized group anomaly detection; performing online inference on real-time construction data stream, and automatically generating a set of instructions containing adjustment parameters when detecting that the device cluster behavior deviates from the safety threshold optimized by federated learning; updating the virtual construction progress model of the digital twin, and superimposing the adjusted construction standard on the real scene through the AR terminal to guide workers to correct the deviation, and encrypting the closed-loop processing records and storing them in the blockchain.

[0008] As a preferred scheme of the highway informatization intelligent management and control system, the specific steps of generating a dynamic risk score are as follows,

[0009] A hierarchical feature fusion architecture is adopted as the spatio-temporal Transformer fusion model.

[0010] The spatial attention flow is used to capture the spatial relationship between devices, and topological features are generated.

[0011] The temporal dependence of the device operating parameters is extracted by applying a causal mask, and a dynamic feature representation across time steps is established.

[0012] The spatio-temporal features of the sensor data and the semantic features of the video picture are dynamically fused through a gating mechanism.

[0013] As a preferred scheme of the highway informatization intelligent management and control system, the hierarchical early warning signal refers to constructing a three-level dynamic threshold, and dynamically adjusting the threshold range according to the real-time construction stage to trigger corresponding measures.

[0014] As a preferred scheme of the highway informatization intelligent management and control system, the device cluster anomaly pattern feature refers to the sensor topological features extracted by the spatial attention flow and the time sequence dynamic features extracted by the time attention flow, combined with the multi-modal feature vector after gating fusion of the video semantic features, and the spatio-temporal pulse graph pattern features generated by the pulse coding layer.

[0015] As a preferred scheme of the highway informatization intelligent management and control system, the specific steps of performing cross-project optimization group anomaly detection are as follows,

[0016] Collect the operating index data of multiple projects, and construct a multi-project spatio-temporal feature matrix after preprocessing the data.

[0017] Analyze the spatio-temporal correlation of indicators between projects, and extract cross-project spatio-temporal correlation features.

[0018] Based on transfer learning, an anomaly detection model for cross-project optimization is constructed, and cross-project feature transfer is performed by dynamically adjusting the parameters of the anomaly detection model.

[0019] The anomaly detection model is trained in combination with group behavior patterns and historical anomaly data to generate an anomaly probability distribution.

[0020] The anomaly threshold is dynamically adjusted according to the real-time detection data, and the group anomaly detection result is output.

[0021] As a preferred scheme of the highway informatization intelligent management and control system, the specific steps of the virtual construction progress model of the digital twin are as follows,

[0022] Collecting multi-source heterogeneous data of construction site, integrating BIM model, sensor monitoring data and resource scheduling plan, constructing dynamic updated virtual construction scene;

[0023] Based on the principle of discrete event simulation, a construction progress logical relationship model is established, an initial progress plan is generated through time series analysis and resource constraint optimization, task node parameters in the virtual construction scene are dynamically corrected by combining real-time construction data and historical deviation law, and progress prediction and risk early warning results are generated;

[0024] The resource allocation and task priority are adjusted through multi-objective optimization algorithm, and the optimal construction progress scheme is output.

[0025] As a preferred scheme of the highway information intelligent management and control system, the timestamp alignment and format standardization include defining a reference time sequence, synchronizing sensor data and video streams using a reference signal interpolation alignment method, and integrating video semantic features and sensor data into a unified data tensor.

[0026] As a preferred scheme of the highway information intelligent management and control system, the multi-source data includes road roller vibration frequency spectrum, paver temperature curve, pre-stressed tension force value, and construction pictures captured by video monitoring equipment.

[0027] As a preferred scheme of the highway information intelligent management and control system, the device cluster abnormal mode features include a space-time distribution map of road roller vibration frequency spectrum deviating from the mean value , and paver temperature gradient abnormal area coordinates.

[0028] As a preferred scheme of the highway information intelligent management and control system, the safety threshold is a dynamic discriminant boundary optimized by federated learning;

[0029] The instruction set includes road roller vibration frequency correction value and paver travel speed upper limit, and is issued to the corresponding device controller through the edge computing node to perform dynamic regulation and control.

[0030] The present application has the beneficial effects that: through the high-precision space-time alignment algorithm and the adaptive multi-modal fusion architecture, the real-time performance and the accuracy of the construction safety risk assessment are significantly optimized. Specifically, the space-time feature alignment mechanism based on physical constraints effectively suppresses the noise interference in the multi-source data synchronization process, ensuring the integrity of the equipment cluster vibration conduction characteristics; the dynamic gating fusion module automatically adjusts the contribution weight of the sensor and the video semantic features through environmental perception, realizing the collaborative optimization of multi-modal information under complex working conditions; the distributed learning framework driven by pulse coding greatly improves the efficiency of anomaly detection, and through the event-triggered feature compression and adaptive weight update mechanism, the rapid response and global optimization of the equipment cluster behavior are realized. Thus, the safety risk misjudgment probability is significantly reduced, and the execution timeliness of the warning instruction is improved. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0032] Fig. 1 The flowchart of the highway information intelligent management and control system.

[0033] Fig. 2 The multi-modal feature fusion flowchart of the highway information intelligent management and control system.

[0034] Fig. 3 The federal optimization interaction schematic diagram of the highway information intelligent management and control system.

[0035] Fig. 4 The virtual-real synchronous working principle diagram of the highway information intelligent management and control system. DETAILED DESCRIPTION

[0036] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0037] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0038] Second, the "one embodiment" or "an embodiment" referred to herein means a particular feature, structure, or characteristic including an implementation that can be included in at least one implementation of the application. The appearances of "in one embodiment" or "in an embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily all referring to a single, alternative embodiment, or a single, alternative implementation.

[0039] Embodiment 1, refer to Figs. 1-4 , as the first embodiment of the application, the embodiment provides a highway information intelligent management and control system, comprising the following steps:

[0040] S1, real-time acquisition of multi-source data, uploaded to the edge computing node for time stamp alignment and format standardization, input into the space-time Transformer fusion model, through the multi-head attention mechanism to extract the space-time characteristics of sensor data and the semantic characteristics of video pictures, and generate dynamic risk score.

[0041] Specifically, it includes the following steps:

[0042] Deploy sensor network and video stream equipment at the hardware layer for multi-source data acquisition.

[0043] Among them, the multi-source data includes the vibration spectrum of the road roller, the temperature curve of the paver, the prestressed tension force value, and the construction picture captured by the video monitoring equipment.

[0044] Define the data transmission protocol of sensor data and video stream:

[0045] Specifically, the sensor data: the vibration spectrum, temperature curve and tension force value are packaged into Cap'n Proto binary format, uploaded to the edge node through 5G NR Uu interface, and the end-to-end delay is ≤30ms. Video stream: using H.265 encoding, slice is 1 second HLS slice, transmitted to the edge node through RTSP protocol.

[0046] Data preprocessing is performed on the sensor data and video stream.

[0047] Specifically, the vibration spectrum of the road roller is subjected to FFT transformation, and the energy value of the 0-100Hz frequency band is extracted. The temperature curve of the paver is subjected to moving average filtering (window length 10 seconds). The YOLOv7-tiny model is used to detect the key objects (tower crane, worker, material pile) in the construction picture in real time, and output the Bounding Box coordinates and class label.

[0048] Based on the time characteristics of the sensor data and video stream after data preprocessing, the starting point of the reference time sequence is determined And the step , and define the discrete time sequence :

[0049] ;

[0050] wherein, the alignment accuracy, is the time index, is the length of the reference time series, indicating that there are time points in the reference time series (since starting from 0);

[0051] The interpolation of the sensor data on is calculated using the reference signal interpolation alignment method, and is expressed as:

[0052] ;

[0053] wherein, represents the interpolation of the aligned sensor data on the reference time point , represents the measured value of the sensor data at the time point , represents the time point in the reference time series, represents the time point of the sensor data, i.e. the time stamp of the sensor data, represents the number of sensor data points calculated by interpolation, represents the value of the constant pi, represents the time index of the sensor data; It should be noted that,

[0054] represents the Sinc function used for interpolation calculation.

[0055] Specifically, according to the time difference between the sensor data and the reference time point , the interpolation weight is calculated;

[0056] When , the value of the Sinc function is 1.

[0057] When , the value of the Sinc function gradually decays, indicating that the data points farther away from the target time point contribute less to the interpolation result.

[0058] Using the model to extract the Bounding Box coordinates of the key objects (such as tower cranes and workers) in each frame, and synchronizing their time stamps with the sensor data to construct a unified data tensor ;

[0059] wherein, represents the set of real numbers, ​(sensor channel number + video semantic feature number), data block length (corresponding to 1 second window).

[0060] After standardization and block processing of the unified data tensor , input the spatio-temporal Transformer fusion model, extract features through spatial attention flow, temporal attention flow and multi-modal fusion.

[0061] It should be noted that standardization and block processing refer to calculating the mean and standard deviation of each column of the unified data tensor, i.e. each feature dimension, then normalizing the data by subtracting the mean and dividing by the standard deviation of each feature value, obtaining the standardized data tensor; then block processing is performed, the purpose is to divide the data into fixed length blocks, which is convenient for subsequent model processing. The specific operation is to set the block length according to the requirements, for example, two hundred corresponding to one second window, divide the standardized data tensor into multiple fixed length data blocks according to the time dimension, each data block contains a certain number of time points and feature dimensions; finally, the divided data blocks are stored or passed to the subsequent model for feature extraction and fusion.

[0062] Further, after data preprocessing is completed, sensor data and video data are converted into feature matrices for subsequent fusion and analysis. Specifically,

[0063] Sensor data is represented as , and video data is represented as after extracting features through the ResNet-18 model.

[0064] wherein, represents the dimension of sensor features (such as 10 physical quantities measured by sensors), represents the length of the feature vector output by ResNet-18 (such as 1024 dimensions).

[0065] Specifically, the spatial attention flow is used to capture the spatial relationship between devices (such as device position coordinates) to extract the topological features of sensor data, represented as:

[0066] ;

[0067] wherein, represents the output of the attention mechanism, is the query matrix, representing the target features to be focused on, is the key matrix, representing the feature representation of the input data, is the value matrix, representing the actual feature value of the input data, represents the query matrix and the key matrix The product of the transposes of the matrices, The symbol for matrix transpose. Key matrix The square root of the feature dimension, express The subscript is used to represent the key matrix. Dimensions Represents a mask matrix based on the physical distance between devices (set when distance > 10m). ), Represents the value matrix based on attention weights Perform a weighted summation;

[0068] Temporal attention flow is employed to capture the temporal dependencies of multi-sensor data and extract the temporal dynamic features of the sensor data, represented as:

[0069] ;

[0070] In the formula, Indicates a causal mask (only historical information is allowed to participate);

[0071] The video semantic features (1024-dimensional vectors extracted by ResNet-18) are fused with sensor features through a gating mechanism, and are represented as follows:

[0072] ;

[0073] ;

[0074] In the formula, This represents the gating weight vector, used to dynamically adjust the fusion ratio of sensor features and video semantic features. This represents the Sigmoid activation function. This represents the weight matrix, used to map the concatenated feature vectors to the gated weight space. This represents the concatenated vector of sensor features and video semantic features. This represents the bias vector, used to adjust the baseline value of the gate weights. This represents the fused multimodal feature vector, which contains a dynamically weighted combination of sensor features and video semantic features. This represents the weighted result of the sensor features, where Control the contribution ratio of sensor features. This represents the weighted result of the video's semantic features, where Controlling the contribution ratio of video features, This represents element-wise multiplication (Hadamard product), which is the multiplication of corresponding elements of two vectors.

[0075] It should be noted that the sensor topology and time dynamic characteristics extracted by the spatial attention flow and the time attention flow are the main sources of sensor features in multi-modal fusion. Through the gating mechanism, the fusion ratio of sensor features and video semantic features is dynamically adjusted to ensure that the fused features can fully utilize the information of the two modalities.

[0076] Based on the fused multi-modal feature vector , a dynamic risk score is generated , which is expressed as:

[0077] ;

[0078] In the formula, represents a weight coefficient for adjusting the contribution of the difference between the current feature and the safety feature, is a summation symbol, indicating summation over all feature dimensions, represents a weight coefficient for adjusting the difference contribution of the th feature dimension, represents the value of the th feature dimension at time , is a safety feature, representing the feature value in the ideal safety state, is an L2 norm, representing the Euclidean distance between the current feature and the safety feature , represents a weight coefficient for adjusting the contribution of the difference between the current feature distribution and the historical feature distribution, is a KL divergence, representing the difference between the current feature distribution and the historical feature distribution .

[0079] It should be noted that the weight coefficients and can be determined by empirical values or optimization algorithms. For example, in a construction scene, the values of , can be set according to historical data or expert experience, indicating that the contribution of the current feature difference is greater, and can also be optimized by machine learning: for example, using cross-validation or gradient descent method, based on training data to optimize the values of and .

[0080] The weight coefficient can be determined by a time decay function or expert experience. Specifically, an exponential decay function is used, where is the decay coefficient, indicating that the weight of the time point is smaller the farther away from the current time; for expert experience determination: different time points can be assigned fixed weights according to the importance of the construction stage.

[0081] For safety features It can be determined by historical data statistics. For example, based on past normal construction data, the mean or median of the feature value is calculated as , and it can also be set by expert experience. For example, according to construction specifications or safety standards, the specific value of is defined.

[0082] For KL divergence It can be determined by probability distribution estimation. For example, using kernel density estimation (KDE) or Gaussian mixture model (GMM) to fit the current and historical feature distribution, calculate the KL divergence, and it can also be determined by historical data statistics. For example, based on past normal construction data, construct the historical feature distribution .

[0083] S2, trigger a graded early warning signal based on the dynamic risk score, execute AR projection instructions, red alert push, rectification work order generation and metering payment lock.

[0084] Specifically, the following steps are included:

[0085] Through the scoring formula, the grading threshold is constructed:

[0086] ;

[0087] It should be noted that indicates that there is potential risk and needs attention but has not reached the emergency level, indicates that the risk is high and immediate measures need to be taken, indicates that the risk is extremely high and the construction must be stopped immediately and emergency measures must be taken;

[0088] Among them, the grading threshold can be dynamically adjusted according to the real-time construction stage (such as pouring, hoisting) to adapt to the risk characteristics and tolerance of different stages. The specific implementation manner can be to preset the threshold range of different construction stages, and automatically switch the corresponding threshold according to the real-time construction stage.

[0089] Further, when the dynamic risk score reaches a certain grading threshold, the corresponding graded early warning signal will be triggered, and the following operations will be performed:

[0090] AR projection instructions: through augmented reality (AR), project risk information or rectification instructions to the construction site to guide workers' operation.

[0091] Red alert push: Send a red alert to relevant personnel, reminding them to take immediate action.

[0092] Rectification work order generation: Automatically generate rectification work orders to clearly identify the content and responsible person that needs to be rectified.

[0093] Metering payment lock: Lock the metering payment of the relevant project before the risk is resolved, ensuring that the rectification measures are implemented in place.

[0094] It should be noted that the grading threshold is the basis for triggering the grading warning signal, when a certain threshold is reached, the corresponding warning signal will be automatically triggered and the corresponding operation will be executed.

[0095] For example:

[0096] If , trigger , execute AR projection instructions and rectification work order generation.

[0097] If , trigger , execute red alert push and metering payment lock.

[0098] If , trigger , execute all warning and rectification measures.

[0099] Preferably, through this mechanism, the corresponding warning and rectification measures can be automatically executed according to the risk level, ensuring construction safety.

[0100] S3, upload the device cluster anomaly pattern features of the current project to the cloud through the federated learning platform, and compare them with the global feature library shared by the historical projects to generate a cross-project optimized group anomaly detection model.

[0101] Specifically, the following steps are included:

[0102] Extract the device cluster anomaly pattern features, including the time-space distribution map of the deviation of the vibratory roller vibration frequency spectrum from the mean value , and the coordinates of the abnormal temperature gradient area of the paver.

[0103] Combine the dynamic risk score sequence and the time-space Transformer features and input them into the pulse coding layer. The time difference threshold algorithm is used to convert continuous features into pulse sequences. Only when the feature change rate exceeds the threshold will a pulse be triggered.

[0104] The pulse sequence is represented as:

[0105] ;

[0106] wherein, represents the pulse state of the i-th feature at time , represents the value of the i-th fused feature at time , is a time variable representing a certain time within the integration interval, represents the rate of change (derivative) of the i-th fused feature at time , is a pulse trigger threshold, only when the feature rate of change exceeds the threshold, a pulse is emitted, significantly reducing the data transmission volume (compression rate ≥ 90%), is the length of the time integration interval, indicating the integration of the feature rate of change within the recent time, represents the integration interval;

[0107] The pulse sequence and sensor topology relationship (device space coordinates) are dynamically constructed into a spatiotemporal pulse graph :

[0108] wherein, the pulse sequence refers to a plurality of pulse states within a period of time.

[0109] The spatiotemporal pulse graph is represented as:

[0110] ;

[0111] wherein, is a node, representing a single device in the device cluster, carrying a pulse sequence, is an edge, representing the physical connection or signal coupling relationship (such as the vibration conduction path) between devices, represents the weight;

[0112] Further, the weight will be dynamically updated based on the physical distance between devices and pulse synchronicity, represented as:

[0113] ;

[0114] wherein, represents the weight value between device and device at time , represents the physical distance between device and device , is the time length of the summation interval, ​​​Indicates the time variable At that time, equipment The pulse state, Indicates the time variable At that time, equipment The pulse state, Represents the time variable at any point within the summation interval. Indicates from the current time Push forward Location of time;

[0115] It should be noted that weights The larger the value, the more likely the device is to be larger. and equipment The stronger the behavioral correlation, the more likely the device... and equipment This refers to two separate pieces of construction equipment (such as a road roller and a paver).

[0116] Encrypted upload of spatiotemporal pulse diagram The cloud-based federated learning platform adopts an event-driven mechanism (uploads are triggered only when a significant impulse pattern is detected), replacing traditional periodic updates.

[0117] The cloud-based federated platform employs an asynchronous update mechanism, where each edge node only triggers model uploads (event-driven) when a significant pulse pattern is detected, rather than at a fixed period.

[0118] The cloud updates the global model using the Impulse Temporal Dependency Plasticity (STDP) rule, with the following weight adjustment rules:

[0119] ;

[0120] In the formula, Indicates time At that time, equipment and equipment The change in weights between them This represents the learning rate (update step size). Indicates time At that time, equipment The pulse state, Indicates the current time Push forward Location of time;

[0121] The pulse pattern of the current project is compared with the global feature library of historical projects to perform cross-project optimized group anomaly detection.

[0122] S4. Perform online inference on the real-time construction data stream. When the behavior of the equipment cluster deviates from the safety threshold of federated learning optimization, automatically generate an instruction set containing adjustment parameters.

[0123] Specifically, the following steps are included:

[0124] Modeling the group anomaly detection as a quantum Markov decision process includes a state space: the pulse synchronization matrix of the device cluster ;

[0125] Action space: adjustment parameter instruction set (such as vibration amplitude, temperature threshold)

[0126] Reward function: based on the decline rate of dynamic risk score .

[0127] Further, using quantum approximate optimization algorithm (QAOA) to solve the optimal instruction strategy, breaking through the local optimal limit of classical reinforcement learning.

[0128] When the synchronization matrix deviates from the safety threshold optimized by federated learning, generate the instruction set through the quantum annealing processor;

[0129] Specifically, the construction and optimization of the safety threshold first constructs the joint probability distribution of the device feature parameters as the basic threshold based on the historical normal construction data through kernel density estimation, and relies on the federated learning platform to aggregate the topological feature weight matrix of the spatiotemporal pulse graph of each project., using pulse timing-dependent plasticity rules to dynamically update global model parameters, combined with device correlation weight increment to calculate the benchmark value of the safety threshold; On this basis, introduce the construction stage adaptation coefficient to dynamically adjust the threshold range, load the preset threshold coefficient according to the real-time construction stage and fuse the environmental compensation factor of the temperature and humidity sensor for threshold compensation optimization, and finally realize the dynamic optimization and deviation detection of the safety threshold through a multi-dimensional joint determination mechanism, including the Euclidean distance of the spatiotemporal Transformer fusion feature vector and the safety benchmark exceeding the dynamic compensation threshold, the KL divergence of the current feature distribution and the global feature library exceeding the preset probability deviation, and the device cluster pulse synchronization index breaking through the weighted sum of the correlation weight. Three conditions of cooperative judgment form the dynamic safety threshold system optimized by federated learning, providing a quantitative basis for device cluster behavior deviation detection.

[0130] Among them, the instruction set includes the vibration frequency correction value of the road roller and the upper limit of the paver travel speed, and is issued to the corresponding device controller through the edge computing node to perform dynamic regulation and control.

[0131] S5, update the virtual construction progress model of the digital twin, and superimpose the adjusted construction standard on the real scene through the AR terminal to guide the workers to correct the deviation, and encrypt the closed-loop processing record and store it to the blockchain.

[0132] Specifically, the following steps are included:

[0133] updating the virtual construction progress model of the digital twin according to the instruction set;

[0134] Specifically, the virtual construction progress model of the digital twin is a dynamic system for simulating and predicting construction progress, and its state is described by the following variables:

[0135] : the construction state vector at time , including equipment state, construction progress, environmental parameters, etc.

[0136] : the control input vector at time , including equipment parameter adjustment instructions (such as vibration frequency, temperature threshold).

[0137] : the output vector at time , including construction progress, risk score, AR projection path, etc.

[0138] The state update equation of the virtual construction progress model is represented as:

[0139]

[0140] In the formula: is the state transition function, which describes how the construction state changes over time, control input and external disturbance, is the external disturbance vector, including environmental noise, equipment abnormality, etc.

[0141] It should be noted that the acquisition of the state transition function needs to be realized through multi-source data modeling and hybrid driving framework. Specifically, based on the physical characteristics of construction equipment (such as vibration energy transmission, material mechanics parameters), differential equations or discrete event models are constructed, combined with the historical abnormal data of the federal learning platform (such as equipment failure mode, environmental disturbance record), the spatio-temporal Transformer is used to extract the correlation characteristics of multi-device clusters, and the data-driven state prediction logic is trained through LSTM or Neural ODE time series model;

[0142] Furthermore, real disturbances (such as temperature and humidity mutation, equipment abnormal shutdown) are injected into the digital twin environment, the actual construction progress data fed back by the AR terminal is used to close-loop calibrate the model parameters, and through the federal learning framework, the multi-project global feature optimization weight is aggregated, finally deployed to the edge computing node, real-time receiving control instructions and external disturbances detected by sensors, realizing dynamic iterative update, ensuring that the state transition process meets both physical laws and data generalization.

[0143] The output equation of the virtual construction progress model is represented as:

[0144]

[0145] wherein: represents an output function, describing how the construction state and control input are mapped to the output (e.g. construction progress, risk score).

[0146] It should be noted that the acquisition of the output function relies on the multi-modal feature fusion and rule-adaptive hybrid architecture. First, the output target (e.g. risk score formula, AR path planning rule) is defined according to the construction safety specification, combined with sensor data (vibration, temperature) and video semantic features, and a gating fusion mechanism is used to dynamically weight the generated fusion features, and a lightweight regression model (e.g. random forest or lightweight neural network) is trained to map to specific outputs (e.g. risk score, projected path). At the same time, based on the KL divergence to measure the distribution deviation of the construction state and the safety benchmark, the risk threshold and the path avoidance strategy are dynamically optimized through reinforcement learning;

[0147] Extreme scenarios (e.g. equipment failure, sudden rainfall) are simulated in the digital twin model to verify the robustness of the output, and finally deployed to a cloud-edge collaborative architecture to support real-time calculation of risk scores and trigger AR path dynamic re-planning. The warning records are automatically encrypted and stored in the IOTA Tangle blockchain through smart contracts, ensuring the compliance, auditability and real-time response capability of the output logic.

[0148] The adjusted construction standards (e.g. roller vibration frequency) are superimposed on the real scene through the AR terminal;

[0149] The AR projected path is dynamically re-planned to avoid high-risk areas;

[0150] The closed-loop processing records (e.g. adjusted construction parameters, worker operation records) are encrypted and stored in the blockchain, realizing fee-free and high-concurrency storage based on the IOTA Tangle architecture.

[0151] In summary, the present application significantly optimizes the real-time and accuracy of construction safety risk assessment through high-precision spatio-temporal alignment algorithm and adaptive multi-modal fusion architecture. Specifically, the spatio-temporal feature alignment mechanism based on physical constraints effectively suppresses noise interference in the synchronization process of multi-source data, ensuring the integrity of the vibration conduction characteristics of the device cluster; the dynamic gating fusion module automatically adjusts the contribution weight of sensor and video semantic features through environmental perception, achieving collaborative optimization of multi-modal information in complex working conditions; the pulse coding driven distributed learning framework significantly improves the efficiency of anomaly detection, and through the event triggered feature compression and adaptive weight update mechanism, the device cluster behavior is quickly responded and globally optimized. Thus, the probability of false safety risk judgment is significantly reduced, and the execution timeliness of the warning instruction is improved.

[0152] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A highway information-based intelligent management and control system, characterized in that: Comprise, Real-time acquisition of multi-source data, uploaded to the edge computing node for timestamp alignment and format standardization, input into the spatio-temporal Transformer fusion model, extraction of spatio-temporal features of sensor data and semantic features of video pictures through multi-head attention mechanism, generation of dynamic risk score; Trigger hierarchical early warning signals based on dynamic risk score, execute AR projection instructions, red alert push, rectification work order generation and metering payment lock; Through the federal learning platform, the device cluster abnormal mode characteristics of the current project are encrypted and uploaded to the cloud, and compared with the global feature library shared by the historical projects for comparative learning, and cross-project optimized group anomaly detection is performed; The specific steps are as follows: Extracting equipment cluster anomaly pattern features, including road roller vibration frequency spectrum deviating from the mean spatiotemporal distribution map, paver temperature gradient anomaly region coordinates Combining dynamic risk score sequences And spatiotemporal transformer features Input to the pulse coding layer, convert continuous features into pulse sequences through a time difference threshold algorithm, only when the feature change rate exceeds the threshold Time triggered pulses The pulse sequence is represented as: ； wherein denotes the pulse state of the i-th feature at time , denotes the value of the i-th fused feature at time , is a time variable denoting a certain time within the integration interval, denotes the rate of change of the i-th fused feature at time , is a pulse trigger threshold, a pulse is emitted only when the rate of change of the feature exceeds the threshold, is the length of the time integration interval, indicating that the rate of change of the feature in the recent time is integrated, denotes the integration interval;​​​ The pulse sequence and sensor topological relations are dynamically constructed into a space-time pulse graph : Wherein, the pulse sequence refers to a plurality of pulse states within a period of time; spatiotemporal impulse diagram is represented as: ； In the formula, is a node, representing a single device in the device cluster, carrying a pulse sequence, is an edge, representing a physical connection or signal coupling relationship between devices, represents the weight; Weights The weights are dynamically updated based on the physical distance between devices and the pulse synchronicity, denoted as: ； wherein denotes the weight value between the device and the device at the time , denotes the physical distance between the device and the device , is the time length of the summation interval, denotes the pulse state of the device at the time variable , denotes the pulse state of the device at the time variable , denotes the time variable at an arbitrary time within the summation interval, denotes the position from the current time forward by the time ; The cloud updates the global model through the pulse timing dependent plasticity STDP rule, and the weight adjustment rule is: ； In the formula, represents the change amount of the weight between the device and the device at time , represents the update step learning rate, represents the pulse state of the device at time , represents the position of the current time forward by time ; Compare the pulse mode of the current project with the global feature library of historical projects to perform cross-project optimized group anomaly detection; Online inference is performed on real-time construction data stream, and when the device cluster behavior deviates from the safety threshold optimized by federal learning, an instruction set containing adjustment parameters is automatically generated; Update the virtual construction progress model of the digital twin, and superimpose the adjusted construction standard on the real scene through the AR terminal to guide the workers to correct the deviation, and at the same time, the closed-loop processing records are encrypted and stored in the blockchain.

2. The highway informationized intelligent management and control system of claim 1, wherein: The specific steps of generating a dynamic risk score are as follows, A hierarchical feature fusion architecture is used as the spatio-temporal Transformer fusion model; Capture the spatial relationship between devices through spatial attention flow to generate topological features; Apply causal mask to extract the time sequence dependence of device operating parameters, and establish dynamic feature expression across time steps; Through the gating mechanism, dynamically fuse the spatio-temporal features of sensor data and the semantic features of video pictures. 3.The highway informationized intelligent management and control system according to claim 1, characterized in that: The hierarchical early warning signal refers to constructing a three-level dynamic threshold, and dynamically adjusting the threshold range according to the real-time construction stage to trigger corresponding measures.

4. The highway informationized intelligent management and control system of claim 2, wherein: The device cluster abnormal mode characteristics refer to the sensor topological features extracted by spatial attention flow and the time sequence dynamic features extracted by time attention flow, combined with the multi-modal feature vector after gating fusion of video semantic features, and the spatio-temporal pulse graph mode features generated by the pulse coding layer. 5.The highway informationized and intelligent management and control system according to claim 4, characterized in that: The specific steps of cross-project optimized group anomaly detection are as follows, Collect the operating index data of multiple projects, and construct a multi-project spatio-temporal feature matrix after preprocessing the data; Analyze the spatio-temporal correlation of indicators between projects, and extract cross-project spatio-temporal correlation features; Based on transfer learning, construct an anomaly detection model optimized across projects, and perform cross-project feature transfer by dynamically adjusting the parameters of the anomaly detection model; Train the anomaly detection model in combination with group behavior patterns and historical anomaly data to generate an anomaly probability distribution; Dynamically adjust the anomaly threshold according to the real-time detection data, and output the group anomaly detection result. 6.The highway informationized and intelligent management and control system according to claim 5, characterized in that: The specific steps of the virtual construction progress model of the digital twin are as follows, Acquire multi-source heterogeneous data of the construction site, integrate BIM model, sensor monitoring data and resource scheduling plan, and construct a dynamically updated virtual construction scene; Based on the principle of discrete event simulation, a construction schedule logical relationship model is established, an initial schedule is generated through time series analysis and resource constraint optimization, task node parameters in the virtual construction scene are dynamically corrected by combining real-time construction data and historical deviation rules, and schedule prediction and risk warning results are generated; Through multi-objective optimization algorithm, resource allocation and task priority are adjusted, and the optimal construction schedule scheme is output. 7.The highway informationized and intelligent management and control system according to claim 1, characterized in that: The multi-source data includes road roller vibration frequency spectrum, paver temperature curve, prestressed tension force value, and construction pictures captured by video monitoring equipment. 8.The highway informationized and intelligent management and control system according to claim 1, characterized in that: The timestamp alignment and format standardization include defining a reference time sequence and using a reference signal interpolation alignment method to synchronize sensor data and video streams. 9.The highway informationized and intelligent management and control system of claim 1, wherein: The device cluster anomaly mode features include road roller vibration frequency spectrum deviating from the mean spatiotemporal distribution map, paver temperature gradient abnormal region coordinates. 10.The highway informationized intelligent management and control system according to claim 1, characterized in that: The safety threshold is a dynamic discriminant boundary optimized through federated learning; The instruction set includes road roller vibration frequency correction value and paver travel speed upper limit, and is issued to the corresponding equipment controller through the edge computing node to perform dynamic regulation and control.

Images