Highway intelligent monitoring and management system and method and electronic equipment

Abstract

The invention relates to the field of intelligent transportation, and discloses an intelligent monitoring and management system and method for an expressway and electronic equipment, and the system collects global spatial-temporal data of the expressway to construct digital twins synchronized with the physical world; in the twinborn body, performing prediction and deduction based on a space-time causal map to identify potential risks; responding to the risk, generating an optimal intervention strategy through anti-fact deduction and executing the optimal intervention strategy, and recording a predicted intervention effect of the optimal intervention strategy; and after intervention, comparing a real traffic state with a prediction effect, calculating an anti-fact error, and carrying out dynamic self-correction on the space-time causal map according to the anti-fact error. According to the invention, links of perception, prediction, decision making, execution and feedback are fused into a self-adaptive control loop, and a self-correction mechanism based on an anti-fact error is introduced, so that the system can continuously learn and self-evolve from interaction with the physical world, and the problems of model solidification and poor adaptability of a traditional traffic management system are solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to intelligent monitoring and management systems, methods and electronic devices for highways. Background Technology

[0002] Highways are the arteries of modern transportation, and their operational efficiency and safety are directly related to the national economy and people's livelihoods. With the rapid development of information technology and artificial intelligence, intelligent transportation systems have become a key means to improve highway management. Current highway management systems generally collect traffic flow data by deploying sensors such as cameras, loop detectors, and millimeter-wave radar along the route, and combine this data with analysis models to monitor, predict, and guide traffic conditions.

[0003] However, existing intelligent highway management technologies still reveal several deep-seated limitations in practice. Many current systems are essentially reactive, with control measures often triggered only after negative events such as congestion or accidents have occurred, falling into a passive, post-event remedial mode and lacking the ability to proactively anticipate and mitigate risks. While some systems have introduced machine learning models for traffic condition prediction, these models are mostly static models trained offline on historical data, and their parameters tend to be fixed once deployed. This static characteristic makes them difficult to adapt to the highly dynamic and non-stationary nature of real traffic environments. When encountering atypical scenarios such as severe weather, sudden accidents, or holidays, the model's predictive accuracy drops significantly, thus affecting the effectiveness of subsequent decisions.

[0004] More importantly, existing technologies often lack the ability to deeply assess the complex consequences of intervention measures when making intervention decisions. The models they employ typically only capture statistical correlations between data points, failing to reveal the deep causal transmission mechanisms between various elements within the transportation system. Therefore, the system struggles to accurately answer key counterfactual questions about the future impact of specific interventions, leading to suboptimal control strategies and, in some cases, unexpected negative effects. In summary, existing technologies generally lack a closed-loop intelligent control architecture capable of self-evolution, proactively anticipating, and accurately extrapolating the consequences of interventions. Summary of the Invention

[0005] The technical problem that this invention aims to solve is that the prediction models used in existing highway monitoring and management systems are usually static. After the models are trained offline, they cannot adapt to the changing traffic environment online, nor can they obtain feedback from the actual effects of the intervention measures they implement to optimize themselves. As a result, their prediction accuracy and control effectiveness decrease over time.

[0006] To address these issues, this invention provides an intelligent monitoring and management system, method, and electronic device for highways, aiming to construct a closed-loop control system with dynamic self-correction capabilities to solve the aforementioned technical problems.

[0007] To achieve the above objectives, the intelligent monitoring and management system for highways provided in the first aspect of the present invention includes: a data acquisition module, a digital twin module, a prediction and inference module, an intervention decision module, and a self-correction module.

[0008] The data acquisition module is designed to collect data from various sensors along the highway in real time, forming structured spatiotemporal data covering the entire region. These sensors include, but are not limited to, high-definition cameras, millimeter-wave radar, lidar, and weather sensors.

[0009] The digital twin module has its input end connected to the output end of the data acquisition module. This module is used to receive the global spatiotemporal data and, based on the data, generate and maintain a digital twin that is synchronized in real time with the physical highway network status, on the basis of a pre-constructed three-dimensional static model of the highway.

[0010] The prediction and inference module has its input connected to the output of the digital twin module. This module internally stores a spatiotemporal causal graph. The spatiotemporal causal graph is a directed graph structure comprising: a set of nodes representing various traffic variables, a set of directed edges representing the causal relationships between these traffic variables, and a set of weights quantifying the strength of these causal relationships.

[0011] The function of this module is to perform forward simulation in the digital twin environment based on the internal logic of the spatiotemporal causal graph, using the current state of the digital twin as the initial condition, in order to output a prediction result of the future traffic state, which includes the identified potential traffic risks.

[0012] The intervention decision-making module has its input connected to the output of the prediction and inference module. The function of this module is to generate and execute an optimal counterfactual intervention strategy aimed at avoiding or mitigating the potential traffic risk upon receiving the information.

[0013] Specifically, the module performs the following operations: First, it generates a set containing multiple candidate intervention measures; second, for each candidate intervention measure in the set, it performs counterfactual deduction in the digital twin to obtain the corresponding predicted traffic state as the predicted intervention effect; finally, for each predicted traffic state, it calculates a utility score, selects the candidate intervention measure with the highest utility score as the optimal counterfactual intervention strategy, and converts the strategy into a control command to be issued to the roadside equipment for execution.

[0014] In one specific embodiment, the utility score U is calculated using a utility calculation formula, which is: U = w E ·M E +w S ·M S ; Among them, M E M is the traffic efficiency index extracted from the predicted traffic conditions. S For the traffic safety indicators extracted from the predicted traffic conditions, w E and w S These are the preset weights for the traffic efficiency index and the traffic safety index, respectively.

[0015] Furthermore, the traffic efficiency indicator may include at least one of average traffic speed or road segment throughput; the traffic safety indicator may include at least one of collision time or traffic conflict event count.

[0016] The self-correction module has its inputs connected to the data acquisition module, the digital twin module, and the intervention decision module, respectively, and its output connected to the prediction and inference module. This module constitutes the closed-loop feedback core of the system described in this invention. Its function is to acquire, through the data acquisition module and the digital twin module, the actual traffic state reflecting the actual intervention effect after the optimal counterfactual intervention strategy is executed, and simultaneously receive the predicted intervention effect generated by the intervention decision module during the decision-making process, corresponding to the optimal counterfactual intervention strategy. Subsequently, based on the counterfactual error between the actual traffic state and the predicted intervention effect, this module dynamically self-corrects the spatiotemporal causal graph used by the prediction and inference module.

[0017] Specifically, the self-correction module compares the actual traffic state with the predicted traffic state, which is the effect of the predictive intervention, to calculate the counterfactual error; then, the counterfactual error is used as a loss signal to adjust the causal relationship weights or causal relationship structures defined in the spatiotemporal causal graph.

[0018] In one specific embodiment, the self-correction module calculates the counterfactual error ε using the counterfactual error formula. cf The counterfactual error formula is as follows: εcf=D(S real (t+Δt),S pred (t+Δt∣do(A opt ))); Where t is the current time of performing the predictive inference, Δt is the time span of the predictive inference, and A opt For the optimal counterfactual intervention strategy, S real(t+Δt) represents the actual traffic state observed at time t+Δt, S pred (t+Δt∣do(A opt To match the optimal counterfactual intervention strategy A opt The corresponding predicted traffic state at time t+Δt represents the effect of the predicted intervention. do(·) is the counterfactual intervention quantifier, and D(·,·) is a preset distance or divergence metric function used to quantify the difference between the two states.

[0019] In one specific embodiment, the spatiotemporal causal graph includes causal relationship weights W. c The self-correction module uses a weight update formula to adjust the causal relationship weight W. c The weight update formula is as follows: Where t is the self-correcting discrete time step, and W c (t+1) represents the causal relationship weight updated in step (t+1), W c (t) represents the causal relationship weights before the t-th step update, η is the preset learning rate, and ε cf For the counterfactual error, The counterfactual error is the causal weight W at step t. c The gradient of (t). This embodiment of the invention, by setting a self-correction module, utilizes the deviation between the actual results generated by the system's intervention decisions in the physical world and the previously predicted results to form a counterfactual error, and uses this error to correct the system's core cognitive model: the spatiotemporal causal graph. This mechanism enables the system to acquire information from every interaction with the physical world, continuously optimize its predictive capabilities, and thus achieve adaptation to dynamic traffic environments.

[0020] The intelligent monitoring and management method for highways provided in the second aspect of the present invention, applied to the system described in any of the foregoing embodiments, includes the following steps: S1. Real-time collection of spatiotemporal data of the entire highway area; S2. Based on the aforementioned global spatiotemporal data, construct a digital twin that is synchronized with the physical highway in real time; S3. Based on a preset spatiotemporal causal graph, predictive simulations of future traffic conditions are performed in the digital twin to output the identified potential traffic risks. S4. In response to the potential traffic risk, generate the corresponding predicted intervention effect and the optimal counterfactual intervention strategy aimed at avoiding or mitigating the potential traffic risk, and execute the optimal counterfactual intervention strategy. S5. After the optimal counterfactual intervention strategy is executed, obtain the real traffic state that reflects the actual intervention effect, and perform dynamic self-correction on the spatiotemporal causal map based on the counterfactual error between the real traffic state and the predicted intervention effect.

[0021] Since the method embodiments described above correspond one-to-one with the aforementioned system embodiments, their specific technical details can be found in the description of the aforementioned system embodiments, and will not be repeated here.

[0022] A third aspect of the present invention provides an electronic device including a memory and a processor. The memory stores a computer program, and the processor is configured to execute the computer program to implement the highway intelligent monitoring and management method described in the foregoing method embodiments.

[0023] In summary, the present invention has at least one of the following beneficial technical effects: 1. This invention sets up a self-correction module to compare the actual traffic conditions generated by the intervention strategy in the physical world with the system's own predicted intervention effect to form a counterfactual error. This error is then used to continuously and dynamically self-correct the spatiotemporal causal graph, thereby achieving continuous adaptation to the dynamic traffic environment and ensuring the long-term effectiveness of the system. This closed-loop learning mechanism enables the system's core prediction model to continuously learn from the actual control effect and automatically adapt to changes in traffic patterns, overcoming the technical defects of traditional static models that lead to performance degradation due to environmental changes.

[0024] 2. By constructing a digital twin and using spatiotemporal causal graphs for predictive extrapolation, this invention can identify potential traffic risks in advance and perform counterfactual extrapolation on multiple candidate strategies before implementing any intervention to estimate their possible effects, thereby improving the initiative and accuracy of traffic risk management. This mechanism realizes the transformation from passive response to proactive prevention and enables the system to select the optimal timing and intensity of intervention, avoiding blind management.

[0025] 3. This invention evaluates the merits of different intervention strategies by introducing a quantitative utility score that integrates multiple dimensions such as traffic efficiency and traffic safety, making the decision-making process more objective and scientific, thereby enhancing the reliability and optimality of intervention decisions. By systematically comparing the expected utility of all candidate strategies, it can ensure that the optimal counterfactual intervention strategy selected in the end is theoretically the best choice under the current traffic conditions. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the structure of a highway intelligent monitoring and management system according to an embodiment of the present invention; Figure 2This is a flowchart illustrating an embodiment of the intelligent monitoring and management method for highways according to the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0027] See attached document Figure 1 , Figure 1 This is a schematic diagram of the structure of a highway intelligent monitoring and management system according to an embodiment of the present invention.

[0028] The system includes: a data acquisition module 10, a digital twin module 20, a prediction and inference module 30, an intervention decision module 40, and a self-correction module 50.

[0029] Data acquisition module 10 is used to collect real-time spatiotemporal data of the entire highway network. The spatiotemporal data is transmitted to digital twin module 20. Based on the spatiotemporal data, digital twin module 20 constructs and maintains a digital twin that is synchronized in real time with the physical highway network status.

[0030] The prediction and simulation module 30 is connected to the digital twin module 20. It receives the current state of the digital twin and performs forward simulation based on an internally preset spatiotemporal causal graph to output a prediction result of the future traffic state, including potential traffic risks. This prediction result is then transmitted to the intervention decision module 40.

[0031] The intervention decision module 40 is connected to the prediction and inference module 30. Upon receiving a potential traffic risk, this module generates and selects the optimal counterfactual intervention strategy, and then converts this strategy into a control command to be issued to the roadside traffic control equipment for execution. Simultaneously, the intervention decision module 40 transmits the predicted intervention effect corresponding to the optimal counterfactual intervention strategy to the self-correction module 50.

[0032] The self-calibration module 50 forms the core of the system's closed-loop feedback. Its inputs are connected to the digital twin module 20 and the intervention decision module 40, respectively, and its output is connected to the prediction and inference module 30. After the optimal counterfactual intervention strategy is executed, the self-calibration module 50 obtains the real traffic status reflecting the actual intervention effect from the digital twin module 20 and obtains the predicted intervention effect from the intervention decision module 40.

[0033] The self-correction module 50 calculates the counterfactual error between the actual traffic conditions and the predicted intervention effects, and uses this error as an adjustment signal to dynamically self-correct the spatiotemporal causal map used by the prediction and extrapolation module 30. This module uses the counterfactual error formula to calculate the counterfactual error ε. cf The counterfactual error formula is: Where t is the current time of performing the predictive simulation, Δt is the time span of the predictive simulation, and A opt For the optimal counterfactual intervention strategy, S real (t+Δt) represents the actual traffic state observed at time t+Δt, and S pred (t+Δt∣do(A opt For optimal counterfactual intervention strategy A opt The corresponding predicted traffic state at time t+Δt is used to predict the effect of intervention. do(·) is the counterfactual intervention quantifier, and D(·,·) is a preset distance or divergence metric function used to quantify the difference between the two states.

[0034] When selecting the optimal counterfactual intervention strategy, the intervention decision module 40 uses a utility score. The utility score U is calculated using the utility calculation formula: U = w E ·M E +w S ·M S ; Among them, M E M is a traffic efficiency index extracted from predicted traffic conditions. S For traffic safety indicators extracted from predicted traffic conditions, w E and w S These are the preset weights for traffic efficiency indicators and traffic safety indicators, respectively.

[0035] Through the above structure, this system forms a complete closed-loop workflow of data-driven decision-making, model prediction, intervention and decision-making, and feedback correction, enabling the system model to continuously self-correct based on feedback from the real physical world.

[0036] See attached document Figure 1 The data acquisition module 10 is the sensing foundation of the system of the present invention. Its function is to provide accurate and comprehensive real-time data input for subsequent digital twin construction, prediction and decision-making.

[0037] In one specific embodiment, the data acquisition module 10 integrates various sensing devices deployed along the roadside of the highway. These sensing devices include, but are not limited to: high-definition cameras, millimeter-wave radar, lidar, and weather sensors. The high-definition cameras are used to acquire continuous video image information of the road segment. The millimeter-wave radar is used to detect the position, speed, and other information of moving targets. The lidar is used to acquire high-precision three-dimensional point cloud data. The weather sensors are used to collect environmental parameters such as visibility, temperature, humidity, and wind speed.

[0038] The data acquisition module 10 performs a series of preprocessing operations on the heterogeneous raw data acquired from various sensing devices to form a unified data format usable by the system. The preprocessing operations include data cleaning, spatiotemporal alignment, and data fusion.

[0039] In the data cleaning step, the module filters the raw data to remove abnormal data points caused by equipment noise or environmental interference, and interpolates and fills in missing non-critical data to ensure the integrity of the data stream.

[0040] In the spatiotemporal alignment step, the module unifies data streams from different sensing devices to a common time reference and spatial coordinate system. Temporally, all acquired data is assigned a precise and consistent timestamp using Network Time Protocol (NTP) or an equivalent synchronization technology. Spatially, based on a pre-calibrated high-precision map, the observation data from all sensors is transformed into a unified geographic coordinate system.

[0041] In the data fusion step, the module correlates and merges multi-source data after spatiotemporal alignment to generate a comprehensive description of the traffic environment. For example, it fuses vehicle type identification results from cameras with speed and location information of the same vehicle detected by millimeter-wave radar to form a comprehensive traffic participant object containing multi-dimensional attributes such as category, location, speed, and acceleration.

[0042] By performing the above operations, the data acquisition module 10 ultimately outputs a series of structured global spatiotemporal data streams. These data streams accurately represent the state of each traffic participant and the relevant environmental conditions on a specific section of the highway at a specific moment, and are transmitted to the digital twin module 20.

[0043] See attached document Figure 1 The input of the digital twin module 20 is connected to the output of the data acquisition module 10, and its output is connected to the prediction and inference module 30. The function of this module is to receive structured global spatiotemporal data and construct a digital twin that is mapped to the physical world in real time and with high fidelity.

[0044] In one specific embodiment, the digital twin module 20 pre-loads a static, high-precision 3D model. This model accurately describes all the fixed elements in the highway environment, including but not limited to the precise spatial location and physical properties of road geometry, lane topology, traffic signs and markings, gantries, toll booths, bridges and tunnels, and slope protection barriers.

[0045] The digital twin module 20 continuously receives spatiotemporal data streams from the data acquisition module 10. The module injects information describing dynamic traffic participants (such as vehicles and pedestrians) and the environment (such as weather and lighting) from the data stream into the static high-precision 3D model in real time.

[0046] The injection process involves precisely mapping and presenting each identified traffic participant object onto the corresponding virtual lane or location in the static model based on its spatiotemporally aligned coordinates and posture information. Simultaneously, the object's dynamic attributes, such as real-time speed, acceleration, and vehicle type, are also bound to the virtual object. Environmental parameters, such as visibility or road surface slippage coefficient for specific road sections, are updated to reflect the scene attributes of the corresponding area in the digital twin.

[0047] Through the above operations, the digital twin module 20 generates and continuously maintains a dynamic, multi-dimensional virtual highway environment, i.e., a digital twin. This digital twin not only geometrically replicates the physical highway but also achieves real-time synchronization in terms of operational status. The complete state of the digital twin, as a structured dataset or state vector, is continuously transmitted to the prediction and inference module 30 as the initial conditions and computational environment for its subsequent predictive inference.

[0048] See attached document Figure 1 The input of the prediction and extrapolation module 30 is connected to the digital twin module 20, and its output is connected to the intervention decision module 40. The core function of this module is to perform forward extrapolation on the current traffic state represented by the digital twin based on a preset spatiotemporal causal graph, so as to generate a prediction of the traffic state in the future and identify potential traffic risks.

[0049] In one specific embodiment, the spatiotemporal causal graph is a directed acyclic graph that structurally encodes the causal relationships between variables in a transportation system. Specifically, the graph includes: a set of nodes representing various types of transportation variables, a set of directed edges representing the causal relationships between these variables, and a set of weights quantifying the strength of these causal relationships.

[0050] The nodes in the node set represent various discrete or continuous variables that affect traffic conditions. These variables include, but are not limited to: micro-state variables of a single vehicle (such as position, speed, acceleration), macro-traffic flow variables of a specific road segment (such as average speed, density, flow rate), environmental state variables (such as visibility, road surface friction coefficient), and state variables of traffic control measures (such as variable speed limits, ramp merging rates).

[0051] Each edge in the directed edge set points from a parent node to a child node, indicating that the variable represented by the parent node is one of the reasons why the variable represented by the child node changes. For example, a directed edge from the road friction coefficient node to the vehicle braking distance node represents the direct causal influence of road conditions on vehicle braking performance. The weight set assigns a numerical value to each directed edge, which quantifies the strength and direction of the corresponding causal relationship.

[0052] The prediction and extrapolation module 30 uses the traffic state at the current time t provided by the digital twin as the initial condition to initialize the state of all nodes in the spatiotemporal causal graph. Subsequently, the module performs iterative forward simulation with discrete time steps Δt. At each time step, the module calculates the state value of each node at the next time t+Δt based on the graph's topology and weights. Specifically, the state value of any child node at time t+Δt is a function of the state values ​​of all its parent nodes at time t and the weights of the corresponding connecting edges.

[0053] By continuously executing the aforementioned forward simulation, the prediction and extrapolation module 30 can generate a continuous evolution sequence of traffic conditions over a future period (e.g., the next 5 minutes). The module then analyzes this predicted sequence to identify potential traffic risks. Risk identification is based on a series of preset rules or thresholds. For example, when the predicted longitudinal time distance between any two vehicles is less than a safety threshold, it is identified as a rear-end collision risk; when the predicted traffic density of a road segment exceeds its capacity threshold, it is identified as a congestion risk.

[0054] The prediction and simulation module 30 ultimately outputs a structured risk information object. This object details the type, time of occurrence, spatial location, severity level, and associated traffic participants or road segments for each identified potential risk, and transmits it to the intervention decision module 40 as the basis for the latter to initiate the intervention decision process.

[0055] See attached document Figure 1 The input of the intervention decision module 40 is connected to the prediction and inference module 30, and its output is connected to the roadside traffic control equipment and the self-correction module 50, respectively. The function of this module is to generate and execute the optimal counterfactual intervention strategy in response to the potential traffic risks output by the prediction and inference module 30 through a systematic decision-making process.

[0056] In one specific embodiment, when the intervention decision module 40 receives the risk information object, it first generates a set containing multiple candidate intervention measures. These candidate intervention measures are feasible solutions for the identified risks, such as adjusting the variable speed limit for a specific road segment, changing the signal timing at ramp entrances to control merging rates, or issuing warning information on variable message signs.

[0057] Subsequently, for each candidate intervention measure in the set, the intervention decision module 40 performs a counterfactual simulation. This simulation process involves calling the prediction simulation module 30 to perform a simulation with the same current traffic state (provided by the digital twin module 20) as the initial condition. Unlike the basic prediction, in the counterfactual simulation, the spatiotemporal causal graph node value corresponding to the candidate intervention measure is forcibly set to the intervention value. For example, when evaluating the strategy of a speed limit of 80 km / h, the value of the variable speed limit node in the graph is fixed at 80 throughout the simulation. The output of this process is the predicted traffic state corresponding to each candidate intervention measure, i.e., the predicted intervention effect.

[0058] After obtaining the predicted intervention effects of all candidate intervention measures, the intervention decision module 40 will quantitatively evaluate each predicted traffic state to calculate its utility score.

[0059] The utility score U is calculated using the utility calculation formula, which is: U = w E ·M E +w S ·M S ; In this formula, M E This refers to traffic efficiency metrics extracted from predicted traffic conditions, such as the average speed of a road segment or the vehicle throughput per unit time. M S This refers to traffic safety metrics extracted from predicted traffic conditions, such as predicted minimum collision time or the number of traffic conflict events. E and w S These are the preset weights for traffic efficiency indicators and traffic safety indicators, and their values ​​determine the degree of emphasis the system places on efficiency and safety when making decisions.

[0060] The intervention decision module 40 compares the utility scores U generated by all candidate interventions and determines the candidate intervention with the highest score as the optimal counterfactual intervention strategy.

[0061] Finally, the intervention decision module 40 performs two output operations. First, it transforms the selected optimal counterfactual intervention strategy into specific equipment control commands and sends them to the corresponding highway roadside equipment for execution via the communication interface. Second, it transmits the predicted intervention effect corresponding to the optimal strategy (i.e., the predicted traffic state that produces the most efficient score) to the self-correction module 50 as a benchmark for subsequent counterfactual error calculation.

[0062] See attached document Figure 1The input of the self-calibration module 50 is connected to both the digital twin module 20 and the intervention decision module 40, while its output is connected to the prediction and inference module 30. The function of this module is to adjust the core prediction model of the system after each intervention by comparing the difference between the prediction and reality, thereby achieving closed-loop feedback and adaptive optimization.

[0063] In one specific embodiment, the operation of the self-correction module 50 is triggered after an intervention cycle ends. First, it obtains the actual traffic state S in the physical world during the corresponding time period after the optimal counterfactual intervention strategy is executed, through the digital twin module 20. real At the same time, it receives from the intervention decision module 40 the predicted intervention effect S corresponding to the implemented strategy, generated during the decision-making phase. pred .

[0064] After obtaining the above two data points, the self-correction module 50 calculates the counterfactual error ε between them. cf This error quantifies the accuracy with which the system predicts the effectiveness of a specific intervention.

[0065] Counterfactual error ε cf The counterfactual error is calculated using the counterfactual error formula, which is as follows: εcf=D(S real (t+Δt),S pred (t+Δt∣do(A opt ))); In this formula, t is the initial time of performing the predictive extrapolation, Δt is the prediction time span, and A opt For the optimal counterfactual intervention strategy, S real (t+Δt) represents the actual traffic state observed at time t+Δt, and S pred (t+Δt∣do(A opt For optimal counterfactual intervention strategy A opt The corresponding predicted traffic state at time t+Δt is used to predict the effect of intervention. do(·) is the counterfactual intervention quantifier, and D(·,·) is a preset metric function, such as mean squared error or cross-entropy, used to quantify the difference between the two state vectors.

[0066] The calculated counterfactual error ε cf It is used as a loss signal to drive adjustments to the spatiotemporal causal graph within the prediction and extrapolation module 30.

[0067] In one embodiment, the adjustment primarily targets the causal weights W in the graph. c The self-correction module 50 uses a weight update formula to adjust the causal relationship weights W. c The weight update formula is adjusted as follows: Where t is the self-correcting discrete time step, and W c (t+1) represents the causal relationship weight updated in step (t+1), W c (t) represents the causal relationship weights before the t-th step update, η is the preset learning rate, and ε cf For counterfactual error, The counterfactual error is related to the causal weight W at step t. c The gradient of (t) indicates the direction in which the weights need to be adjusted to reduce counterfactual error.

[0068] By performing the aforementioned weight update operation, the self-correction module 50 solidifies the experience gained from this intervention into the parameters of the spatiotemporal causal graph. The adjusted graph is then used by the prediction and inference module 30 for future prediction tasks, enabling the system to exhibit higher accuracy in subsequent predictions and decisions. This process, repeated over time, endows the system with the ability to continuously learn and self-optimize from interactions with the physical world.

[0069] The core of this invention lies in the spatiotemporal causal graph used by the prediction and inference module 30.

[0070] In one specific embodiment, the construction and representation of the graph follow the paradigm of a structured causal model. The set of nodes in the graph contains various variables necessary to describe the state of the traffic system, such as microstate variables representing an individual vehicle i, like its position vector p. i and velocity vector v i ; represents the macroscopic traffic flow variables of road segment j, such as the traffic density ρ of that road segment. j and traffic q j Variables representing environmental conditions, such as visibility V and road surface friction coefficient μ; and control variables representing intervention measures, such as the variable speed limit SL for road segment j. j and the merging rate R of ramp k k .

[0071] The set of directed edges in the graph represents the causal relationship assumptions between these variables, based on traffic engineering theory, vehicle dynamics physics, or statistical associations mined from historical data. For example, a path from node SL... j Pointing to node v i The directed edges (where vehicle i is located on road segment j) represent the causal effect of the speed limit on the driver's speed selection behavior. The weight set W in the graph... cEach of the above causal relationships has been quantified. During initial system deployment, these weights can be set based on theoretical formulas or obtained by performing system identification or machine learning training on offline historical datasets. These weights are the core objects for subsequent online optimization by the self-calibration module 50.

[0072] The core of the intervention decision module 40 in this embodiment of the invention is a decision-making process based on counterfactual inference. This process aims to answer the question of what will happen if a certain intervention is implemented. This process is implemented through the counterfactual intervention quantifier do(·). Technically, executing the operation do(A=a) is not equivalent to observing the event A=a in the graph. It represents an external forced intervention on the system that cuts off all original causal paths pointing to the variable node A and forcibly sets its value to a. For example, when evaluating the strategy of setting the speed limit to 80km / h, the system executes do(SL·). j =80) operation, that is, temporarily removing all pointers to node SL in the spacetime causal graph. j The input edge is fixed at 80, and then forward extrapolation is performed based on this to obtain the prediction of the future traffic state under this specific intervention.

[0073] For each counterfactual scenario, the intervention decision module 40 evaluates the predicted traffic conditions obtained through a utility function.

[0074] The utility formula is: U = w E ·M E +w S ·M S ; In the utility formula: Traffic efficiency index M E This can be specified as the average speed of all vehicles in the target area within a predicted time period, or the total traffic flow at key cross-sections. Traffic safety indicator M S This can be specifically defined as the reciprocal of the minimum collision time obtained after evaluating the interaction states of all vehicles within the prediction area, or the total count of predicted traffic conflict events (e.g., weaving or following events below a preset safety threshold). The intervention decision module 40 iterates through all candidate intervention measures, calculates their respective utility scores, and finally selects the measure with the highest utility score as the optimal counterfactual intervention strategy A. opt .

[0075] The core of the self-correction module 50 in this embodiment of the invention is an online learning mechanism based on counterfactual error. Its purpose is to reduce the prediction bias of the system model by utilizing the results of each real intervention. When the real traffic state S is obtained... real And the corresponding predictive intervention effect S predThen, the module calculates the counterfactual error ε. cf The choice of the distance metric function D(·,·) depends on the form of the state vector. If the state consists of a set of continuous variables (such as velocity and density), then the mean square error function can be used for D; if the state is described by a probability distribution, then cross-entropy or Kullback-Leibler divergence can be used.

[0076] The calculated counterfactual error ε cf It is used as a loss function to guide the spatiotemporal causal graph weights W c Update.

[0077] Weight update formula: The execution of this formula depends on the gradient. The calculation of the gradient. This gradient represents the sensitivity of the counterfactual error to each causal weight parameter. In a concrete implementation, due to the weight W... c To the final predicted state S pred The entire forward inference process can be constructed as a differentiable computational graph, thus the gradient can be efficiently computed by applying backpropagation or other automatic differentiation techniques to this time-series model. This gradient guides the direction and magnitude of weight updates, enabling the adjusted weights to reduce the probability of recurring similar prediction errors in future situations. See attached document Figure 2 , Figure 2 This is a flowchart illustrating a highway intelligent monitoring and management method according to an embodiment of the present invention. The method is a specific implementation of the aforementioned system embodiment and may include the following steps: S1. Real-time acquisition of spatiotemporal data of the entire highway. This step is performed by the data acquisition module 10. By integrating various sensors such as cameras, millimeter-wave radar, and lidar deployed along the highway, real-time, multimodal traffic and environmental data is acquired. The data acquisition module 10 cleans, aligns, and fuses the raw data to generate a structured spatiotemporal data stream.

[0078] S2. Based on the spatiotemporal data of the entire domain, construct a digital twin that is synchronized with the physical highway in real time. This step is performed by the digital twin module 20. The module receives the structured data stream generated in step S1 and maps and updates the dynamic traffic participant information and environmental parameter information in real time to a pre-built high-precision static three-dimensional model, thereby generating and maintaining a digital twin that can accurately represent the current traffic operation status of the physical world.

[0079] S3. Based on a preset spatiotemporal causal graph, predictive simulations of future traffic conditions are performed in the digital twin to output identified potential traffic risks. This step is executed by the prediction and simulation module 30. The module uses the digital twin state generated in step S2 as initial conditions, utilizes the spatiotemporal causal graph internally encoded with causal relationships between traffic system variables, performs iterative forward simulation to generate a traffic state evolution sequence over a future period, and identifies potential traffic risks from this sequence according to preset risk discrimination rules.

[0080] S4. In response to potential traffic risks, generate corresponding predicted intervention effects and optimal counterfactual intervention strategies aimed at avoiding or mitigating potential traffic risks, and execute the optimal counterfactual intervention strategy. This step is performed by the intervention decision module 40. After receiving the risk identified in step S3, the module first generates a set of candidate intervention measures. For each candidate measure, the module performs a counterfactual deduction to obtain its corresponding predicted intervention effect. Subsequently, the module quantifies and scores each predicted intervention effect using a utility function, and determines the measure with the highest score as the optimal counterfactual intervention strategy. During this process, the predicted intervention effect associated with the optimal strategy is also determined. Finally, the module converts the optimal strategy into a control command and issues it to the roadside equipment for execution.

[0081] S5. After the optimal counterfactual intervention strategy is executed, the actual traffic state reflecting the actual intervention effect is obtained, and the spatiotemporal causal graph is dynamically self-corrected based on the counterfactual error between the actual traffic state and the predicted intervention effect. This step is performed by the self-correction module 50. After the intervention is executed, the module obtains the actual traffic state through the digital twin module 20 and compares it with the predicted intervention effect determined in step S4 to calculate the counterfactual error. This error is then used as a loss signal, and the causal weights of the spatiotemporal causal graph used by the prediction and inference module 30 are updated through optimization algorithms such as gradient descent.

[0082] By iteratively executing steps S1 to S5, the method of this invention constitutes an adaptive control loop that integrates perception, prediction, decision-making, execution, and feedback. Each cycle not only completes an active control of potential traffic risks but also optimizes the system's own model using the actual effect of that control.

[0083] To make the technical solution of the present invention easier to understand, a specific application scenario will be used as an example for illustration below. This embodiment does not limit the scope of protection of the present invention in any way.

[0084] The scenario is set on a highway with initially clear weather and smooth traffic flow. At a certain moment, the weather sensor in the data acquisition module 10 detects that rainfall is gradually increasing in intensity in a specific section of the road. Simultaneously, video stream analysis from a high-definition camera shows that visibility in that section is rapidly decreasing. The data acquisition module 10 transmits this structured environmental data, along with real-time location and speed information of vehicles on that section of the road, to the digital twin module 20.

[0085] Upon receiving the aforementioned data, the digital twin module 20 immediately updates the state of the digital twin. Specifically, the environmental attributes of the road segment in the digital twin are updated to rainfall and low visibility, and the friction coefficient of the road surface model is correspondingly reduced. Simultaneously, the dynamic information of all vehicles in the physical world, such as their positions and speeds, is also synchronized in real time within the digital twin.

[0086] The prediction and simulation module 30 initiates a forward simulation based on the updated digital twin state as the initial condition. Its internal spatiotemporal causal graph includes the causal relationships between environmental nodes such as visibility and road friction coefficient and micro-level driving behavior nodes such as safe following distance and emergency braking distance. Based on this, the forward simulation results show that although the current driving behavior of each vehicle has not yet caused immediate danger, if the vehicles continue to maintain their current speed within the next three minutes, the probability of potential rear-end collisions, especially consecutive rear-end collisions, will exceed the preset risk threshold due to slippery road surfaces and obstructed visibility. The prediction and simulation module 30 then outputs this high-probability rear-end collision risk as a potential traffic risk to the intervention decision module 40.

[0087] After receiving the risk information, the intervention decision module 40 generates a set of candidate intervention measures. This set may include: Measure A: Reduce the variable speed limit on this section of road to 100 km / h; Measure B: Reduce the variable speed limit on this section of road to 80 km / h; Measure C: Only use variable message signs to issue warnings about slippery roads in rainy weather and to maintain a safe following distance.

[0088] Subsequently, the intervention decision module 40 performs counterfactual reasoning for each of the above measures. For example, to evaluate measure B, the module calls the predictive reasoning module 30 and executes do(SL) j=80) Operation, the simulation results (i.e., the predicted intervention effect of measure B) show that the average speed of the road segment decreases, the average following distance of vehicles increases significantly, and the predicted rear-end collision risk probability drops to a safe level. The module calculates the utility score of the predicted intervention effect of all measures. After comparison, although measure B scores lower than measures A and C in the traffic efficiency index, it scores extremely high in the traffic safety index, making its comprehensive utility score U the highest. Therefore, measure B is determined to be the optimal counterfactual intervention strategy. The intervention decision module 40 then issues a control command to the variable speed limit sign on the roadside, updating its speed limit value to 80km / h, and transmits the predicted intervention effect of measure B to the self-correction module 50.

[0089] Five minutes after the speed limit adjustment is implemented, the self-calibration module 50 begins operation. It obtains the actual traffic conditions for that road segment during that time period from the digital twin module 20; for example, the actual average speed is 82 km / h and the actual minimum following distance is 2.5 seconds. The self-calibration module 50 compares this actual traffic condition with the corresponding predicted intervention effect previously recorded by the intervention decision module 40 (e.g., predicted average speed of 80 km / h and predicted minimum following distance of 2.8 seconds), and calculates the counterfactual error ε between the two. cf .

[0090] Finally, this counterfactual error is used to adjust the weights of the spatiotemporal causal graph in the prediction inference module 30. For example, the calculated gradient indicates that the causal weight of the variable speed limit value on the driver's speed choice in the graph is slightly underestimated in rainy scenarios. By updating the weights based on this gradient, the system's predictions will be closer to reality when encountering similar rainy speed limit scenarios in the future.

[0091] This application scenario demonstrates how the present invention achieves a closed-loop working mode, from proactive risk identification to decision-making, execution, and finally self-learning and optimization based on real-world feedback.

[0092] See attached document Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention.

[0093] The present invention also provides an electronic device 300, which may be a server, a node in a computer cluster, an edge computing unit, or a device with similar computing capabilities. For example... Figure 3 As shown, the electronic device 300 may include one or more processors 310, a memory 320, and a bus 330 for connecting the above components.

[0094] Memory 320 may include non-volatile storage media, such as read-only memory (ROM), and / or volatile storage media, such as random access memory (RAM). Memory 320 is used to store computer program instructions.

[0095] Processor 310 may be a central processing unit (CPU), graphics processing unit (GPU), or other general-purpose or special-purpose processors. Processor 310 is used to call and execute computer program instructions stored in memory 320.

[0096] In one embodiment of the present invention, when the processor 310 executes computer program instructions, it implements the functions of each module in the intelligent highway monitoring and management system described in the above embodiments, or implements the steps of the intelligent highway monitoring and management method described in the above embodiments.

[0097] The electronic device 300 may also include a communication interface 340 for communicating with other devices. For example, the communication interface 340 may receive data from external sensors or send control commands to downstream traffic control equipment.

[0098] Those skilled in the art will understand that Figure 3 The structure shown is for illustrative purposes only. Electronic devices may also include other components not shown in the figure, or may not include certain components shown in the figure.

[0099] The present invention also provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the steps of the intelligent monitoring and management method for highways described in the foregoing method embodiments.

Claims

1. A smart monitoring and management system for highways, characterized in that, include: The data acquisition module is used to collect real-time spatiotemporal data of the entire highway area; The digital twin module is used to receive the global spatiotemporal data and construct a digital twin that is synchronized with the physical highway in real time based on the global spatiotemporal data. The prediction and extrapolation module is used to predict and extrapolate future traffic conditions in the digital twin based on a preset spatiotemporal causal graph, so as to output the identified potential traffic risks. An intervention decision module is used to receive the potential traffic risks, generate corresponding predicted intervention effects and optimal counterfactual intervention strategies aimed at avoiding or mitigating the potential traffic risks, and execute the optimal counterfactual intervention strategies. The self-correction module is used to obtain the actual traffic state reflecting the actual intervention effect after the optimal counterfactual intervention strategy is executed, and to receive the predicted intervention effect generated by the intervention decision module. Based on the counterfactual error between the actual traffic state and the predicted intervention effect, the module performs dynamic self-correction on the spatiotemporal causal map.

2. The intelligent monitoring and management system for highways according to claim 1, characterized in that, The self-correction module performs dynamic self-correction on the spatiotemporal causal graph, specifically for: The counterfactual error is calculated by comparing the actual traffic conditions with the predicted traffic conditions, which represent the effect of the predicted intervention. The counterfactual error is used as a loss signal to adjust the causal relationship weights or causal relationship structures defined in the spatiotemporal causal graph.

3. The intelligent monitoring and management system for highways according to claim 2, characterized in that, The spatiotemporal causal graph includes causal relationship weights W. c The self-correction module uses a weight update formula to adjust the causal relationship weight W. c The weight update formula is as follows: Where t is the discrete time step of self-correction, and W c (t+1) represents the causal relationship weight updated in step (t+1), W c (t) represents the causal relationship weights before the t-th step update, η is the preset learning rate, and ε cf For the counterfactual error, The counterfactual error is the causal weight W at step t. c The gradient of (t).

4. The intelligent monitoring and management system for highways according to claim 2, characterized in that, The self-correction module calculates the counterfactual error ε using the counterfactual error formula. cf The counterfactual error formula is as follows: εcf=D(S real (t+Δt),S pred (t+Δt∣do(A opt ))); Where t is the current time of performing the predictive inference, Δt is the time span of the predictive inference, and A opt For the optimal counterfactual intervention strategy, S real (t+Δt) represents the actual traffic state observed at time t+Δt, S pred (t+Δt∣do(A opt To match the optimal counterfactual intervention strategy A opt The corresponding predicted traffic state at time t+Δt represents the effect of the predicted intervention. do(·) is the counterfactual intervention quantifier, and D(·,·) is a preset distance or divergence metric function used to quantify the difference between the two states.

5. The intelligent monitoring and management system for highways according to claim 1, characterized in that, The intervention decision module generates the predicted intervention effect and the optimal counterfactual intervention strategy, specifically for: Generate a set containing multiple candidate interventions; In the digital twin, counterfactual reasoning is performed for each candidate intervention measure to obtain the corresponding predicted traffic state as the predicted intervention effect; For each predicted traffic state, a utility score is calculated, and the candidate intervention with the highest utility score is selected as the optimal counterfactual intervention strategy.

6. The intelligent monitoring and management system for highways according to claim 5, characterized in that, The utility score U is calculated using a utility calculation formula, which is: U=w E ·M E +w S ·M S ; Among them, M E M is the traffic efficiency index extracted from the predicted traffic conditions. S For the traffic safety indicators extracted from the predicted traffic conditions, w E and w S These are the preset weights for the traffic efficiency index and the traffic safety index, respectively.

7. The intelligent monitoring and management system for highways according to claim 6, characterized in that, The traffic efficiency indicators include at least one of average traffic speed or road segment throughput; the traffic safety indicators include at least one of collision time or traffic conflict event count.

8. The intelligent monitoring and management system for highways according to claim 1, characterized in that, The spatiotemporal causal graph includes: A set of nodes representing various traffic variables; The set of directed edges representing the causal relationships between the traffic variables; A set of weights that quantifies the strength of the causal relationship.

9. A method for intelligent monitoring and management of highways, applied to the intelligent monitoring and management system for highways as described in any one of claims 1-8, characterized in that, Includes the following steps: S1. Real-time collection of spatiotemporal data of the entire highway area; S2. Based on the aforementioned global spatiotemporal data, construct a digital twin that is synchronized with the physical highway in real time; S3. Based on a preset spatiotemporal causal graph, predictive simulations of future traffic conditions are performed in the digital twin to output the identified potential traffic risks. S4. In response to the potential traffic risk, generate the corresponding predicted intervention effect and the optimal counterfactual intervention strategy aimed at avoiding or mitigating the potential traffic risk, and execute the optimal counterfactual intervention strategy. S5. After the optimal counterfactual intervention strategy is executed, obtain the real traffic state that reflects the actual intervention effect, and perform dynamic self-correction on the spatiotemporal causal map based on the counterfactual error between the real traffic state and the predicted intervention effect.

10. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the computer program to implement the intelligent monitoring and management method for highways as described in claim 9.

Images