Intelligent navigation traffic control system for construction site

Abstract

The invention discloses a construction site intelligent navigation traffic control system, and relates to the field of construction site traffic control, and the system comprises a data collection module which is used for collecting construction site traffic environment data in real time through employing an Internet of Things sensor, an unmanned plane, a high-definition camera and laser radar equipment; comprising vehicle access flow, motion trails of construction equipment, construction area division and road conditions; according to the invention, the map precision can be improved, navigation errors caused by map errors can be effectively avoided, dependence on a high-performance server is reduced, the overall deployment cost is reduced, popularization and application in a construction site environment with limited resources are facilitated, efficient storage and transmission are realized, and the influence of emergency situations on traffic control is effectively reduced; the intelligent level of traffic signal regulation and control can be improved, traffic resource waste or congestion caused by traditional fixed-time-length signals is avoided, the uncertainty of the decision making process is reduced, the traffic accident rate is effectively reduced, and the overall traffic operation efficiency of a construction site is improved.

Description

Technical Field

[0001] The present invention relates to the field of construction site traffic management, and particularly to an intelligent navigation traffic management system for construction sites. Background Art

[0002] In the field of engineering construction, how to efficiently manage the complex traffic flow inside the construction site and seamlessly connect it with the surrounding road environment has become an important issue urgently to be solved in the industry. With the acceleration of the urbanization process, the number and scale of infrastructure construction sites are increasing day by day, and the traffic management problems inside and around the construction sites are becoming more prominent. The traffic environment of the construction site is complex. The traffic conditions of the construction site are affected by multiple factors such as the construction stage, the activities of mechanical equipment, and the flow of personnel. The traffic flow has significant dynamic characteristics. In the coexistence environment of personnel and mechanical equipment, illegal operations, dangerous behaviors, and blind spot accidents occur frequently, posing a serious threat to construction safety and traffic order. The traffic of the construction site needs to be closely coordinated with the urban traffic to avoid traffic congestion in the surrounding areas caused by construction, and at the same time optimize the flow of vehicles and personnel inside the construction site and other adjustments. Traditional traffic management methods mainly rely on manual command and simple signal control, lacking the perception and intelligent analysis of real-time data, and it is difficult to meet the requirements of modern construction sites for high efficiency, safety, and dynamic adjustment. With the rapid development of technologies such as the Internet of Things (IoT), big data, and artificial intelligence (AI), intelligent traffic management provides a new way to solve this problem. Therefore, there is an urgent need for an intelligent navigation traffic management system for construction sites to improve the scientific and intelligent level of construction site traffic management.

[0003] The existing intelligent navigation traffic management systems for construction sites have poor map accuracy, cannot avoid navigation errors caused by map errors, rely on high-performance servers, and increase the overall deployment cost. In addition, the intelligent level of traffic signal regulation in the existing intelligent navigation traffic management systems for construction sites is generally low, and the accident rate is relatively high, reducing the overall traffic operation efficiency of the construction site. For this reason, we propose an intelligent navigation traffic management system for construction sites. Summary of the Invention

[0004] The object of the present invention is to solve the above-mentioned problems, and provide an intelligent navigation traffic management system for construction sites.

[0005] The present invention proposes an intelligent navigation traffic management system for construction sites, and the system includes:

[0006] The data acquisition module is used to collect real-time traffic environment data of the construction site by using Internet of Things sensors, drones, high-definition cameras, and lidar devices, including vehicle inflow and outflow, the movement trajectories of construction equipment, construction area division, and road conditions;

[0007] The simulation module is used to build a high-precision lightweight construction site map model, perform real-time rendering, display the dynamic traffic conditions of the construction site, simulate the traffic flow, equipment movement and personnel flow in different construction stages, and evaluate traffic congestion points and potential safety hazards;

[0008] The analysis and prediction module is used to analyze traffic data, predict the traffic flow change trends in various areas within the construction site, and provide a basis for traffic planning;

[0009] The path planning module is used to dynamically adjust the path planning of construction vehicles and equipment inside the construction site in combination with real-time traffic data;

[0010] The signal optimization module is used to dynamically adjust the traffic signals and command lights inside the construction site according to the real-time traffic conditions of the construction site, and optimize the traffic flow distribution;

[0011] The risk control module is used to combine the simulation results of the construction site, detect dangerous behaviors in the construction site traffic in real time, and issue warnings or take automatic control measures;

[0012] The monitoring and optimization module is used to monitor the impact of the construction site traffic activities on the surrounding environment;

[0013] The information interaction module is used to perform visual processing of real-time data and push personalized information;

[0014] The support optimization module is used to provide global optimization suggestions for construction managers to improve the overall operation efficiency of the construction site;

[0015] The emergency response module is used to quickly identify emergencies and anomalies and generate optimized adjustment strategies.

[0016] Optionally, the specific steps for the simulation module to build a high-precision lightweight construction site map model are as follows:

[0017] S101: Extract the geometric information of the building structure, road network and construction area of the corresponding construction site, store the geometric data in the form of vertices, edges and faces as a three-dimensional grid model, add the corresponding nodes, connection relationships and attributes, and define traffic areas and dynamic objects to build the corresponding traffic network diagram, set the behavior rules of vehicles, pedestrians and construction equipment, establish a rule-based behavior model, set the departure conditions of each traffic event, and then set the initial traffic flow distribution according to the construction stage and task plan, determine the flow entry and exit points, and the initial carrying capacity of each road;

[0018] S102: According to the collected data of each group, generate corresponding map feature representations through a high-precision BIM model. Then, construct a set of hybrid learning models based on the CNN network and the RNN network, and use the generated three-dimensional geometric data, semantic labels, and dynamic attribute map feature representations as input data to train the performance of the hybrid learning model until the model performance meets the preset requirements. After that, use the map features output by the hybrid learning model as soft labels;

[0019] S103: Use low-rank decomposition technology to reduce the number of parameters and computational complexity of the hybrid learning model, and construct a set of lightweight models. This model generates corresponding map features through partial map feature representations and the same softmax function as the hybrid learning model, and calculates the KL divergence between the map features of the lightweight model and the soft labels, as well as the cross-entropy loss between the map features output by this model and the true values;

[0020] S104: Combine the KL divergence and the cross-entropy loss to construct a distillation loss function. Use the defined distillation loss function to optimize the parameters of the lightweight model through backpropagation. Then, use the uninvoked map feature representations as the validation set to evaluate the model performance. Repeatedly train and update the lightweight model until the distillation loss function converges within the preset range. Then, send the trained lightweight model to different edge devices, and use this model to generate the construction site map model of the corresponding area.

[0021] Optionally, the specific steps for the path planning module to dynamically adjust the path planning of construction vehicles and equipment inside the construction site are as follows:

[0022] S201: Generate a corresponding weighted graph G=(N, L, W) according to the construction site map model, where N represents the node set, indicating positions, L represents the edge set, indicating paths, and W represents the weight set, indicating the costs of the paths. Initialize the pheromone concentration on each edge, and set the population size, maximum number of iterations, and evaporation coefficient;

[0023] S202: Initialize the initial node positions of each search body in the population by random selection. Calculate the heuristic value of each edge according to the reciprocal of the path distance. Then, calculate the selection probability of transferring from the current state to the next state through the pheromone concentration and the heuristic value. Each search body selects paths in turn according to the selection probability to form a complete path planning scheme;

[0024] S203: After each round of path selection, each search body updates the pheromone concentration of the path it passes through using the evaporation mechanism based on the preset evaporation coefficient, and updates the heuristic value and weight of the path based on the real-time traffic data. The path selection and pheromone concentration update are repeated until the preset maximum number of iterations is reached. The path with the highest pheromone concentration is selected as the optimal path, and the path and total cost are output and the corresponding path planning solution is generated at the same time.

[0025] Optionally, the specific calculation formula of the selection probability in S202 is as follows:

[0026]

[0027] Where P ij represents the probability that the searcher selects path i→j; τ ij represents the pheromone concentration of path i→j; η ij represents the heuristic value of path i→j; τ ik represents the pheromone concentration of path i→k; η ik represents the heuristic value of path i→k; N allowed represents the set of nodes that the searcher has not visited yet; α and β represent weight parameters, which control the importance of pheromone and heuristic value respectively;

[0028] The specific calculation formula of the evaporation mechanism described in S203 is as follows:

[0029]

[0030] In the formula, ρ represents the pheromone evaporation coefficient; m represents the number of search bodies; Represents the information of the release of the y-th group of search bodies on the path i→j.

[0031] Optionally, the specific steps of the signal optimization module to optimize traffic flow distribution are as follows:

[0032] S301: The traffic signal control in the construction site is modeled as a decision-making process, and the state of the construction site traffic system and the strategy for controlling the switching of traffic lights are collected, and the corresponding state space S = {s1, s2, ..., s n1} and action space A = {a1, a2, ..., a n2}, where s n1 represents the n1th system state in the state space S, a n2 Represents the strategy for switching the n2th control signal light in the state space A;

[0033] S302: Calculate the transition probability of reaching state s' by executing action a from state s, construct a state transition matrix based on the calculation result, then initialize the state s0 ∈ S of the traffic signal, including the signal configurations and traffic conditions at each intersection, and calculate the initial reward value according to the optimization objective of traffic flow distribution;

[0034] S303: According to the current state s t , generate all sets of traffic signal switching actions, where each action corresponds to a specific signal configuration, and then execute action a t After that, the system enters the next state s t+1 , and then, through the state transition probability P(s t+1 |s t , a t ), simulate the change of traffic flow, and according to the simulation result, use the Bellman equation to iteratively calculate the cumulative reward expectation when each state adopts the optimal strategy;

[0035] S304: Continuously update the cumulative reward expectation of the strategy adopted by each state through policy iteration until the cumulative expected reward converges to within a preset range, stop the iteration, and at the same time output the optimal strategy, update the current state and reward value according to the real-time traffic data, and dynamically adjust the signal configuration according to the optimal strategy to optimize traffic flow distribution.

[0036] Optionally, the specific steps for the risk control module to detect dangerous behaviors in construction site traffic are as follows:

[0037] S401: Collect traffic behavior data from the simulation module and actual monitoring devices, including video frames, personnel trajectories, and vehicle movement states, annotate the video frames, define the categories of dangerous behaviors, and then preprocess each group of data;

[0038] S402: Extract the spatial features and temporal features of video frames or trajectory data through a pre-trained convolutional neural network, then construct a behavior detection model, and use the annotated data to train the behavior detection model, and at the same time introduce the data generated by the construction site simulation to enrich the training set scenarios;

[0039] S403: The behavior detection model inputs the predicted values of each annotated data through the forward propagation algorithm, and calculates the loss value between the predicted value and the actual value through the cross-entropy loss function. Starting from the output layer, based on the chain rule, calculate the gradient of the loss value for each layer of the behavior detection model through the backpropagation algorithm, and then use the gradient descent algorithm to optimize the model parameters, and repeatedly train and update the behavior detection model until the loss value converges to a preset range;

[0040] S404: Input the extracted spatial features and temporal features into the behavior detection model. The behavior detection model transfers the received spatial features and temporal features layer by layer, predicts each frame or behavior segment, and outputs the probability of dangerous behavior. If the probability of dangerous behavior is higher than the preset threshold, an alarm is immediately triggered, and the on-site traffic flow is dynamically adjusted in combination with the construction site simulation.

[0041] Advantages of the present invention:

[0042] 1. The construction site map constructed based on the BIM technology in the present invention has high precision and high visualization level, and can describe in detail the terrain, facility layout, access path and other complex information inside the construction site. Then, through the knowledge distillation technology, the complex high-precision map model is compressed into a lightweight version, greatly reducing the storage and transmission costs of the model, while maintaining the integrity of key information. Combined with the lightweight processed map model, real-time rendering is performed to reflect the dynamic changes in the construction site in the map in real time, which can improve the map accuracy, effectively avoid navigation errors caused by map errors, reduce the dependence on high-performance servers, lower the overall deployment cost, and facilitate popularization and application in the construction site environment with limited resources, realizing efficient storage and transmission, and effectively reducing the impact of emergencies on traffic management.

[0043] 2. The present invention models the traffic signal control in the construction site as a decision-making process, collects the state of the construction site traffic system and the strategy for controlling the signal light switching, constructs the corresponding state space and action space, initializes the state of the traffic signal, calculates the initial reward value according to the optimization goal of traffic flow distribution, and according to the current state, then executes the optional action, and the system enters the next state. The traffic flow change is simulated through the state transition probability. According to the simulation result, the cumulative reward expectation when each state adopts the optimal strategy is calculated iteratively using the Bellman equation. The cumulative reward expectation of the strategy adopted by each state is continuously updated through policy iteration until the cumulative expected reward converges to the preset range, and the iteration is stopped. At the same time, the optimal strategy is output, and the current state and reward value are updated according to the real-time traffic data, and the signal light configuration is dynamically adjusted according to the optimal strategy to optimize the traffic flow distribution, which can improve the intelligent level of traffic signal regulation, avoid the waste of traffic resources or congestion caused by traditional fixed-duration signals, reduce the uncertainty of the decision-making process, effectively reduce the incidence of traffic accidents, and improve the overall traffic operation efficiency of the construction site. Description of the Drawings

[0044] The present invention will be further described below with reference to the drawings.

[0045] Figure 1 It is a framework diagram of an intelligent navigation traffic management system for a construction site. Detailed Embodiments

[0046] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0047] All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0048] Embodiment 1

[0049] The embodiment of the present invention provides a smart navigation traffic governance system for construction sites. Refer to Figure 1 , Figure 1 which is a framework diagram of a smart navigation traffic governance system for construction sites provided by the embodiment of the present invention. The system includes the following steps

[0050] The data acquisition module is used to collect real-time traffic environment data of the construction site by using Internet of Things sensors, unmanned aerial vehicles, high-definition cameras and lidar devices, including vehicle in-and-out flow, movement trajectories of construction equipment, construction area division and road conditions.

[0051] The simulation module is used to build a high-precision lightweight construction site map model, perform real-time rendering, display the dynamic traffic conditions of the construction site, simulate traffic flow, equipment movement and personnel flow in different construction stages, and evaluate traffic congestion points and potential safety hazards.

[0052] Specifically, extract the geometric information of the building structure, road network, and construction area of the corresponding construction site, store the geometric data in the form of vertices, edges, and faces as a three-dimensional grid model, add the corresponding nodes, connection relationships, and attributes, define the traffic areas and dynamic objects to construct the corresponding traffic network diagram, set the behavior rules for vehicles, pedestrians, and construction equipment, establish a rule-based behavior model, set the departure conditions for each traffic event, and then according to the construction stage and task plan, set the initial traffic flow distribution, determine the flow entry and exit points, and the initial carrying capacity of each road. Generate the corresponding map feature representation through the high-precision BIM model based on the collected data groups. Then, construct a set of hybrid learning models based on the CNN network and the RNN network, and use the generated three-dimensional geometric data, semantic labels, and dynamic attribute map feature representations as input data to train the performance of the hybrid learning model until the model performance meets the preset requirements. Then, use the map features output by the hybrid learning model as soft labels, adopt low-rank decomposition technology to reduce the number of parameters and computational complexity of the hybrid learning model, construct a set of lightweight models. This model generates the corresponding map features through partial map feature representations and the same softmax function as the hybrid learning model, calculates the KL divergence between the map features of the lightweight model and the soft labels, and the cross-entropy loss between the map features output by this model and the true values. Combine the KL divergence and the cross-entropy loss to construct a distillation loss function. Use the defined distillation loss function to optimize the parameters of the lightweight model through backpropagation. Then, use the uninvoked map feature representations as the validation set to evaluate the model performance, and repeatedly train and update the lightweight model until the distillation loss function converges within the preset range. Send the trained lightweight model to different edge devices, and then generate the construction site map model of the corresponding area through this model.

[0053] The analysis and prediction module is used to analyze traffic data and predict the traffic flow change trend in each area of the construction site, providing a basis for traffic planning.

[0054] Example 2

[0055] An embodiment of the present invention provides a construction site intelligent navigation traffic governance system. Refer to Figure 1 , Figure 1 which is a framework diagram of a construction site intelligent navigation traffic governance system provided by an embodiment of the present invention. The system includes:

[0056] The path planning module is used to dynamically adjust the path planning of construction vehicles and equipment inside the construction site in combination with real-time traffic data.

[0057] Specifically, a weighted graph G=(N, L, W) is generated based on the construction site map model, where N represents the set of nodes, indicating positions; L represents the set of edges, indicating paths; W represents the set of weights, indicating the costs of the paths. The pheromone concentration on each edge is initialized, and the population size, maximum number of iterations, and evaporation coefficient are set. The initial node positions of each search body in the population are initialized by random selection. The heuristic value of each edge is calculated based on the reciprocal of the path distance. Then, the selection probability of transferring from the current state to the next state is calculated based on the pheromone concentration and the heuristic value. Each search body selects a path in turn according to the selection probability to form a complete path planning scheme. After each round of path selection, each search body updates the pheromone concentration of the path it has passed through using the evaporation mechanism according to the preset evaporation coefficient, and simultaneously updates the heuristic value and weight of the path according to the real-time traffic data. The path selection and pheromone concentration update are repeated until the preset maximum number of iterations is reached. The path with the highest pheromone concentration is selected as the optimal path, and the path and total cost are output, and the corresponding path planning scheme is generated at the same time.

[0058] In the embodiment, the specific calculation formula for the selection probability is as follows:

[0059]

[0060] In the formula, P ij represents the probability that the search body selects the path i→j; τ ij represents the pheromone concentration of the path i→j; η ij represents the heuristic value of the path i→j; τ ik represents the pheromone concentration of the path i→k; η ik represents the heuristic value of the path i→k; N allowed represents the set of nodes that the search body has not visited yet; α and β represent weight parameters, controlling the importance of pheromone and heuristic value respectively;

[0061] The specific calculation formula for the evaporation mechanism is as follows:

[0062]

[0063] In the formula, ρ represents the pheromone evaporation coefficient; m represents the number of search bodies; represents the information released by the y-th group of search bodies on the path i→j.

[0064] The signal optimization module is used to dynamically adjust the traffic signals and command lights in the construction site according to the real-time traffic conditions in the construction site, and optimize the traffic flow distribution.

[0065] Specifically, the traffic signal control in the construction site is modeled as a decision-making process, and the state of the construction site traffic system and the strategy for controlling the signal light switching are collected, and the corresponding state space S={s1, s2,..., s n1}, and the action space \(A=\{a_1,a_2,\cdots,a\}\) n2} where \(s\) n1 represents the \(n_1\)-th system state in the state space \(S\), and \(a\) n2 represents the \(n_2\)-th strategy for controlling the traffic signal switching in the state space \(A\). Calculate the transition probability of reaching the state \(s'\) from the state \(s\) by performing the action \(a\), and construct a state transition matrix based on the calculation results. Then initialize the state \(s_0\in S\) of the traffic signal, including the signal lamp configuration and traffic conditions at each intersection, and calculate the initial reward value according to the optimization objective of traffic flow distribution. According to the current state \(s\) t , generate all sets of traffic signal switching actions, where each action corresponds to a specific signal lamp configuration. Then perform the action \(a\) t , and the system enters the next state \(s\) t+1 . Then, through the state transition probability \(P(s\) t+1 |s\) t ,a\) t ), simulate the change of traffic flow. According to the simulation results, use the Bellman equation to iteratively calculate the cumulative reward expectation when taking the optimal strategy for each state. Continuously update the cumulative reward expectation of the strategy adopted by each state through policy iteration until the cumulative expected reward converges to within a preset range, then stop the iteration. At the same time, output the optimal strategy, and update the current state and reward value according to the real-time traffic data, and dynamically adjust the signal lamp configuration according to the optimal strategy to optimize the traffic flow distribution.

[0066] The risk control module is used to combine the construction site simulation results, detect dangerous behaviors in the construction site traffic in real time, and issue warnings or take automatic control measures.

[0067] Specifically, traffic behavior data, including video frames, personnel trajectories, and vehicle motion states, are collected from the simulation module and actual monitoring devices. The video frames are annotated to define the categories of dangerous behaviors. Then, each group of data is preprocessed, and the spatial and temporal features of the video frames or trajectory data are extracted through a pre-trained convolutional neural network. After that, a behavior detection model is constructed and trained using the annotated data. Meanwhile, the data generated by the construction site simulation is introduced to enrich the scenarios in the training set. The behavior detection model inputs the predicted values of each annotated data through the forward propagation algorithm and calculates the loss value between the predicted value and the actual value through the cross-entropy loss function. Starting from the output layer, based on the chain rule, the gradient of the loss value with respect to each layer of the behavior detection model is calculated through the backpropagation algorithm. Then, the gradient descent algorithm is used to optimize the model parameters, and the behavior detection model is repeatedly trained and updated until the loss value converges to a preset range. The extracted spatial and temporal features are input into the behavior detection model, which sequentially transmits the received spatial and temporal features and predicts each frame or behavior segment to output the probability of dangerous behaviors. If the probability of dangerous behaviors is higher than the preset threshold, an alarm is immediately triggered, and the on-site traffic flow is dynamically adjusted in combination with the construction site simulation.

[0068] The monitoring and optimization module is used to monitor the impact of construction site traffic activities on the surrounding environment; the information interaction module is used for visualizing real-time data and pushing personalized information; the support and optimization module is used to provide global optimization suggestions for construction managers to improve the overall operation efficiency of the construction site; the emergency response module is used to quickly identify emergencies and anomalies and generate optimization and adjustment strategies.

[0069] The above has described in detail an embodiment of the present invention, but the content described is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the application of the present invention shall still fall within the scope covered by the patent of the present invention.

Claims

1. An intelligent navigation traffic management system for construction sites, characterized in that, Including: The data acquisition module is used to collect real-time traffic environment data of the construction site by using Internet of Things sensors, drones, high-definition cameras and lidar devices, including vehicle entry and exit flow, movement trajectories of construction equipment, construction area division and road conditions; The simulation module is used to build a high-precision lightweight construction site map model, perform real-time rendering, display the dynamic traffic conditions of the construction site, simulate traffic flow, equipment movement and personnel flow in different construction stages, and evaluate traffic congestion points and potential safety hazards; The analysis and prediction module is used to analyze traffic data, predict the changing trends of traffic flow in each area of the construction site, and provide a basis for traffic planning; The path planning module is used to dynamically adjust the path planning of construction vehicles and equipment inside the construction site in combination with real-time traffic data; The signal optimization module is used to dynamically adjust traffic signals and command lights inside the construction site according to the real-time traffic conditions of the construction site, and optimize traffic flow distribution; The risk control module is used to combine the simulation results of the construction site, detect dangerous behaviors in the construction site traffic in real time, and issue warnings or take automatic control measures; The monitoring and optimization module is used to monitor the impact of construction site traffic activities on the surrounding environment; The information interaction module is used to visually process real-time data and push personalized information; The support and optimization module is used to provide global optimization suggestions for construction managers to improve the overall operation efficiency of the construction site; The emergency response module is used to quickly identify emergencies and anomalies and generate optimization and adjustment strategies.

2. The intelligent navigation traffic governance system for construction sites according to claim 1, wherein The specific steps for the simulation module to build a high-precision lightweight construction site map model are as follows: S101: Extract the geometric information of the building structure, road network and construction area of the corresponding construction site, store the geometric data in the form of vertices, edges and faces as a three-dimensional grid model, add corresponding nodes, connection relationships and attributes, define traffic areas and dynamic objects to build a corresponding traffic network diagram, set the behavior rules of vehicles, pedestrians and construction equipment, establish a rule-based behavior model, set the departure conditions of each traffic event, and then set the initial traffic flow distribution according to the construction stage and task plan, determine the flow entry and exit points, and the initial carrying capacity of each road; S102: Generate corresponding map feature representations through a high-precision BIM model according to the collected data groups. Then, build a set of hybrid learning models based on the CNN network and the RNN network, and use the generated three-dimensional geometric data, semantic labels and dynamic attribute map feature representations as input data to train the performance of the hybrid learning model through the input data until the model performance meets the preset requirements. Then, use the map features output by the hybrid learning model as soft labels; S103: Use low-rank decomposition technology to reduce the number of parameters and computational complexity of the hybrid learning model, build a set of lightweight models. This model generates corresponding map features through partial map feature representations and the same softmax function as the hybrid learning model, calculates the KL divergence between the map features of the lightweight model and the soft labels, and the cross-entropy loss between the map features output by this model and the true values; S104: Construct a distillation loss function by combining the KL divergence and the cross-entropy loss. Use the defined distillation loss function to optimize the parameters of the lightweight model through backpropagation. Then, use the uninvoked map feature representations as the validation set to evaluate the model performance. Repeatedly train and update the lightweight model until the distillation loss function converges within a preset range. Send the trained lightweight model to different edge devices, and then generate the construction site map model for the corresponding area through this model.

3. The intelligent navigation traffic governance system for construction sites according to claim 2, characterized in that, The specific steps for the path planning module to dynamically adjust the path planning of construction vehicles and equipment inside the construction site are as follows: S201: Generate a corresponding weighted graph G = (N, L, W) based on the construction site map model, where N represents the set of nodes, indicating positions; L represents the set of edges, indicating paths; W represents the set of weights, indicating the costs of paths. Initialize the pheromone concentration on each edge, and set the population size, the maximum number of iterations, and the evaporation coefficient. S202: Initialize the initial node positions of each search body in the population by random selection. Calculate the heuristic value of each edge according to the reciprocal of the path distance. Then, calculate the selection probability of transferring from the current state to the next state based on the pheromone concentration and the heuristic value. Each search body selects a path in turn according to the selection probability to form a complete path planning scheme. S203: After each round of path selection, each search body updates the pheromone concentration of the path it has passed through using the evaporation mechanism according to the preset evaporation coefficient. At the same time, update the heuristic value and weight of the path according to the real-time traffic data. Repeatedly perform path selection and pheromone concentration update until the preset maximum number of iterations is reached. Select the path with the highest pheromone concentration as the optimal path, and output the path and the total cost, and generate the corresponding path planning scheme at the same time.

4. The intelligent navigation traffic governance system for construction sites according to claim 3, characterized in that, The specific calculation formula for the selection probability in S202 is as follows: Where, P ij represents the probability of the search body selecting the path i→j; τ ij represents the pheromone concentration of the path i→j; η ij represents the heuristic value of the path i→j; τ ik represents the pheromone concentration of the path i→k; η ik represents the heuristic value of path i→k; N allowed represents the set of nodes not yet visited by the search body; α and β represent weight parameters that control the importance of pheromone and heuristic value respectively; The specific calculation formula for the evaporation mechanism in S203 is as follows: where ρ represents the pheromone evaporation coefficient; m represents the number of search bodies; represents the information released by the y-th group of search bodies on the path i→j.

5. An intelligent navigation traffic governance system for construction sites according to claim 3, characterized in that, The specific steps for the signal optimization module to optimize the traffic flow distribution are as follows: S301: Model the traffic signal control within the construction site as a decision-making process, collect the status of the construction site traffic system and the strategies for controlling the traffic signal switching, and construct the corresponding state space S = {s1, s2,..., s n1} and action space A = {a1, a2,..., a n2}, where s n1 represents the n1-th system state in the state space S, and a n2 represents the n2-th strategy for controlling the traffic signal switching in the action space A; S302: Calculate the transition probability of executing action a from state s to reach state s′, and construct a state transition matrix according to the calculation result. Then, initialize the state s0 ∈ S of the traffic signal, including the signal light configuration and traffic conditions at each intersection, and calculate the initial reward value according to the optimization goal of the traffic flow distribution. S303: Generate all sets of signal light switching actions according to the current state s t , where each action corresponds to a specific signal light configuration, and then execute action a t . After that, the system enters the next state s t+1 , and then simulate the traffic flow change through the state transition probability P(s t+1 |s t ,a t ). According to the simulation results, use the Bellman equation to iteratively calculate the expected cumulative reward when taking the optimal strategy for each state; S304: Continuously update the cumulative reward expectation of the policy adopted in each state through policy iteration until the cumulative expected reward converges and improves within a preset range, stop the iteration, and output the optimal policy at the same time. Update the current state and reward value according to the real-time traffic data, and dynamically adjust the signal light configuration according to the optimal policy to optimize the traffic flow distribution.

6. The intelligent navigation traffic governance system for construction sites according to claim 5, characterized in that, The specific steps for the risk control module to detect dangerous behaviors in the construction site traffic in real time are as follows: S401: Collect traffic behavior data from the simulation module and actual monitoring devices, including video frames, personnel trajectories, and vehicle movement states. Annotate the video frames, define the categories of dangerous behaviors, and then preprocess each group of data. S402: Extract the spatial features and temporal features of video frames or trajectory data through a pre-trained convolutional neural network. Then, construct a behavior detection model and use the labeled data to train the behavior detection model. At the same time, introduce the data generated by the construction site simulation to enrich the training set scenarios; S403: The behavior detection model inputs the predicted values of each labeled data through the forward propagation algorithm, and calculates the loss value between the predicted value and the actual value through the cross-entropy loss function. Starting from the output layer, based on the chain rule, calculate the gradient of the loss value for each layer of the behavior detection model through the backpropagation algorithm. Then, use the gradient descent algorithm to optimize the model parameters, and repeatedly train and update the behavior detection model until the loss value converges to a preset range; S404: Input the extracted spatial features and temporal features into the behavior detection model. The behavior detection model transmits the received spatial features and temporal features layer by layer, and makes predictions for each frame or behavior segment, outputting the probability of dangerous behavior. If the probability of dangerous behavior is higher than the preset threshold, immediately trigger an alarm and dynamically adjust the on-site traffic flow in combination with the construction site simulation.

Images