Intelligent early warning method for bridge hoisting traffic guide and change for urban complex environment

Abstract

The invention discloses an urban complex environment-oriented bridge hoisting traffic guide and change intelligent early warning method, and relates to the technical field of traffic control, and the method mainly comprises the steps: carrying out the space-time alignment and noise filtering of original data, and carrying out the fusion construction of a construction influence domain dynamic model for quantifying the interaction relation between the motion track of a hoisting machine and a traffic flow; predicting a traffic jam probability and an accident risk level in a future preset time period through LSTM based on a spatio-temporal feature vector obtained by the construction influence domain dynamic model; based on the prediction result of the LSTM, generating a signal lamp timing optimization strategy and a bypassing path scheme through a reinforcement learning algorithm, and calculating a strategy execution priority; and according to the generated signal lamp timing optimization strategy, the bypassing path scheme and the execution priority, controlling the roadside equipment to execute a self-adaptive guide and change action. According to the method, the defect that a traditional static guiding and changing scheme cannot respond to sudden mechanical displacement and traffic flow changes is overcome, and passive lag optimization is changed into active pre-judgment.

Description

Technical Field

[0001] This invention relates to the field of traffic control system technology, specifically to an intelligent early warning method for traffic diversion during bridge hoisting in complex urban environments. Background Technology

[0002] During the hoisting and installation of steel box girders for urban bridges, traditional traffic diversion schemes primarily rely on manual on-site command and fixed traffic signs for lane control. This static management model has significant drawbacks: firstly, fixed signs cannot respond in real time to sudden changes in traffic flow (such as abnormal congestion caused by accidents or sudden displacement of construction machinery), resulting in diversion strategies lagging behind actual road conditions; secondly, manual command is limited by blind spots and experience-based judgment, making it difficult to accurately quantify the cumulative risks of the hoisting machinery's movement trajectory (such as dynamic changes in turning radius) on surrounding traffic flow. More importantly, existing systems lack the ability to integrate and analyze multi-source dynamic data (machine positioning, traffic density, vehicle acceleration, etc.), making it impossible to predict the probability of congestion and accident risk levels in the construction impact area in future periods, nor to generate coordinated optimization schemes for traffic light timing and detour routes. These deficiencies lead to decreased traffic efficiency in the construction area and an increased incidence of secondary accidents such as scrapes and rear-end collisions caused by response delays, severely restricting bridge construction safety and traffic operation efficiency in complex urban environments. Summary of the Invention

[0003] To ensure bridge construction safety and improve traffic efficiency, this invention proposes an intelligent early warning method for traffic diversion during bridge hoisting in complex urban environments, comprising the following steps: S1: Collects raw data on the positioning of hoisting machinery, vehicle dynamics, and road monitoring in the construction area via IoT protocols; S2: After spatiotemporal alignment and noise filtering of the original data, the data is fused and constructed into a dynamic model of the construction influence domain that quantifies the interaction between the hoisting machinery's trajectory and traffic flow. S3: Based on the spatiotemporal feature vectors obtained from the construction impact domain dynamic model, the probability of traffic congestion and the level of accident risk within a preset time period are predicted by an LSTM neural network. S4: Based on the prediction results of the LSTM neural network, a traffic light timing optimization strategy and detour route scheme are generated through a reinforcement learning algorithm, and the execution priority of the strategy is calculated. S5: Based on the generated traffic light timing optimization strategy, detour route plan, and execution priority, the roadside equipment is linked through the V2X communication protocol to execute adaptive guidance and modification actions.

[0004] Furthermore, in step S2, the fusion and construction of the dynamic model of the construction influence domain specifically involves: Centered on the hoisting machinery, establish grid unit dimensions based on the hoisting machinery's turning radius. An adaptively adjusted polar coordinate system dynamic mesh is used, and mechanical parameters and traffic flow parameters are recorded in the mesh cells; After assigning dynamic weights to mechanical parameters and traffic flow parameters, a comprehensive risk value is calculated, and the impact of mechanical motion on traffic flow is quantified through time series analysis. The formula is expressed as follows: In the formula, According to the grid Distance from the vehicle to the hoisting machinery center The weights of dynamically adjusted mechanical parameters and traffic flow parameters. For grid Traffic density inside For grid Average speed within the area For grid Standard deviation of vehicle acceleration within the range, For grid The overall risk value, As a factor influencing the trajectory of mechanical motion, The rotational angular velocity, For grid Rate of change of internal vehicle speed gradient It is a spatiotemporal feature vector.

[0005] Furthermore, when the turning radius of the hoisting machinery encroaches on the lane, if the grid... Internal composite risk value If the value exceeds the preset warning threshold, the alarm system and variable message signs will be activated to issue a warning.

[0006] Furthermore, in step S3, the LSTM neural network predicts the probability of traffic congestion and the level of accident risk within a preset future time period using the following formula: In the formula, For the current moment, For the previous moment, To store the unit state vector of long-term traffic temporal features, the traffic temporal features include traffic congestion probability. and accident risk level , Forget gates are used to control the proportion of historical information retained. An input gate used to control the proportion of new information input. This is the neuron connection weight matrix of the LSTM unit. For bias vectors, This is the hidden layer output at the previous time step t-1. The spatiotemporal feature vector is the input at the current time t.

[0007] Furthermore, in step S4, the reinforcement learning algorithm is the PPO algorithm, and its objective loss function for policy optimization is: In the formula, Loss to the target This is the expectation operation for calculating the statistical average of the policy gradient estimate at time t. The pruning factor is the pruning factor for the PPO algorithm. This is the advantage estimate for evaluating the value of the action at time t. Let be the ratio of the probability of the old and new strategies performing the same action at time t. This means limiting the probability ratio of the new and old strategies to a certain value. Interval.

[0008] Furthermore, in step S4, the strategy execution priority is based on the congestion probability. Accident level The weighted score is obtained.

[0009] Furthermore, when generating the detour route plan in step S4, the detour route is dynamically weighted based on the ratio between the maximum traffic capacity of the road per unit time and the minimum straight-line distance from the center point of the detour route to the hoisting machinery.

[0010] Furthermore, in step S5, the adaptive guidance action includes: pushing real-time guidance information to the driver's AR-HUD device, the information including virtual lane lines, restricted areas, and dynamic indications of detour routes.

[0011] Furthermore, in step S1, vehicle dynamic data is collected by smart road studs deployed within a preset range around the construction area. The smart road studs sense vehicle trajectory and acceleration data in real time.

[0012] Compared with the prior art, the present invention has at least the following beneficial effects: (1) The present invention proposes an intelligent early warning method for traffic diversion during bridge hoisting in complex urban environments. Based on the construction influence domain analysis of dynamic grid modeling, it deeply couples the hoisting machinery movement trajectory with real-time traffic flow parameters, which solves the fundamental defect that traditional static diversion schemes cannot respond to sudden mechanical displacement and traffic flow changes, and realizes the leap from passive lag to active prediction in risk perception. (2) Relying on the temporal analysis capability of LSTM neural network for spatiotemporal feature vectors, the system can predict the congestion trend and accident risk level in the construction impact domain in advance, so as to make prevention and control decisions in advance; (3) Combining the traffic light timing strategy and detour route scheme generated by the reinforcement learning algorithm, the optimal allocation of road network resources is achieved through a dynamic weighting mechanism; (4) By leveraging augmented reality guidance and multi-source alarm linkage, a three-dimensional guidance and modification network for human-machine collaboration is built between physical roadside facilities and vehicle terminals, reducing the risk of driver misjudgment. Attached Figure Description

[0013] Figure 1 This is a step diagram of an intelligent early warning method for traffic diversion during bridge hoisting in complex urban environments. Detailed Implementation

[0014] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.

[0015] To make the above-mentioned objectives, features, and beneficial effects of the present invention more readily understood, the technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific construction scenarios. It should be noted that this embodiment uses the hoisting of steel box girders for urban elevated bridges as an application scenario. An IoT sensing network consisting of intelligent road studs, geomagnetic vehicle flow detectors, and roadside units (RSUs) is deployed within a 500-meter radius of the construction area, simultaneously accessing the GPS positioning system of the hoisting machinery and traffic management video surveillance data. The core implementation logic of this solution is as follows: First, a dynamic construction impact domain model is constructed through spatiotemporal alignment of multi-source heterogeneous data to quantify the interaction risks between the mechanical rotation motion and traffic flow; second, based on the spatiotemporal feature vector output by this model, an LSTM neural network is driven to predict the probability of congestion and accidents within the next 15-minute window; finally, a signal light timing optimization strategy and an AR-visualized detour guidance scheme are generated through reinforcement learning algorithms to achieve adaptive and coordinated execution of traffic diversion measures, such as... Figure 1 As shown, the main steps include the following: S1: Collects raw data on the positioning of hoisting machinery, vehicle dynamics, and road monitoring in the construction area via IoT protocols; S2: After spatiotemporal alignment and noise filtering of the original data, the data is fused and constructed into a dynamic model of the construction influence domain that quantifies the interaction between the hoisting machinery's trajectory and traffic flow. S3: Based on the spatiotemporal feature vectors obtained from the construction impact domain dynamic model, the probability of traffic congestion and the level of accident risk within a preset time period are predicted by an LSTM neural network. S4: Based on the prediction results of the LSTM neural network, a traffic light timing optimization strategy and detour route scheme are generated through a reinforcement learning algorithm, and the execution priority of the strategy is calculated. S5: Based on the generated traffic light timing optimization strategy, detour route plan, and execution priority, the roadside equipment is linked through the V2X communication protocol to execute adaptive guidance and modification actions.

[0016] During the multi-source perception phase of the construction impact domain, the system collaboratively collects raw data through heterogeneous IoT devices: real-time positioning data of the hoisting machinery is continuously captured by a GPS module integrated into the crane's slewing mechanism, uploading the machinery's latitude and longitude, slewing angle, and dynamic slewing radius at a fixed sampling frequency; simultaneously, intelligent road stud arrays deployed within a preset distance range around the construction area, using fusion detection technology of geomagnetic sensing and millimeter-wave radar, perceive the trajectory coordinates, instantaneous speed, and acceleration fluctuation characteristics of passing vehicles in real time, and transmit this data back to edge nodes via a low-power wide-area network; raw road monitoring data is extracted by parsing traffic management video streams and combining computer vision algorithms to extract lane-level traffic flow density parameters. These three types of data streams undergo preliminary timestamp synchronization and protocol conversion through the edge gateway, thus forming a spatiotemporally correlated raw data input set.

[0017] Based on the collected spatiotemporal correlated raw data input set, the system calls the dynamic mesh engine on the edge server to perform multi-level processing, specifically executing the following process: establishing a polar coordinate system with the real-time positioning coordinates of the hoisting machinery as the pole, and based on the hoisting machinery's rotation radius... Dynamically divide the grid into ring-shaped cells, ensuring that the smaller the distance r from the machine center, the smaller the size of the grid cell. Record two types of parameters simultaneously: Mechanical parameters, including: rotational angular velocity radial displacement ; Traffic flow parameters, including: vehicle density A draw, your car speed Acceleration standard deviation .

[0018] Calculate the grid composite risk value using a dynamic weighting function. : , Among them, mechanical parameter weights Traffic flow parameter weights satisfy: , In the formula, For grid cells Real-time European distance from the vehicle to the hoisting machinery center. To control the adjustment factor for the rate of decay of mechanical parameter risk weights with distance, It is a natural constant. Distance from vehicle to hoisting machinery center Increases and then decays exponentially. Then increase accordingly. Further along the rotation angle Integrating the partial derivative of risk, the influencing factor of the mechanical motion trajectory is generated: , Ultimately, and the rate of change of vehicle speed gradient Concatenate them into a spatiotemporal feature vector: .

[0019] When the comprehensive risk value of any grid cell in the dynamic mesh model Exceeding the preset warning threshold Upon that, the system immediately activates the three-dimensional linkage response mechanism: firstly, based on the risk grid coordinates... Spatial positioning activates the nearest audible and visual alarm, emitting a directional strobe beam. The focused area of ​​the beam covers a radius radius centered on the machine's center. High-risk sector areas are designated as physical warning barriers. Simultaneously, upstream variable message signs in the construction area dynamically load detour route codes, which strictly match detour schemes generated by reinforcement learning algorithms. Scrolling text prompts (such as "Lifting operations in progress, please detour via route B") guide macroscopic traffic diversion. Furthermore, roadside units (RSUs) broadcast the GPS boundary coordinates of high-risk grids to vehicles within the affected area. After receiving the data, onboard AR-HUD devices overlay a semi-transparent red no-entry zone virtual layer 20 meters in front of the driver's field of vision. The layer's transparency increases with distance. The value increases, thus visually mapping the risk intensity. This response mechanism will continue until the overall risk value falls back to a safe range. The protection will automatically be lifted after the protection measures are activated, ensuring that the protection measures are strictly synchronized with the dynamic risk status.

[0020] Spatiotemporal feature vectors generated from the construction influence domain dynamic model The system performs multi-step time series analysis in the LSTM prediction engine deployed in the edge computing cluster: first, it analyzes the vector sequences of 20 consecutive time steps. The input tensors are stacked with a sampling period of 200ms, and long-term dependency features of traffic states are extracted through a three-layer gated recurrent unit. The forget gate is one such feature. Dynamically filter historical information, input gate Combine current input features to jointly update the storage congestion probability. Accident risk level Unit state vector Its mathematical evolution follows: , in, This is the neuron connection weight matrix of the LSTM unit. For bias vectors, This represents the hidden state at the previous time step t-1. For the current input vector (That is, the spatiotemporal feature vector at the current time t). After passing through the fully connected layer's dual-channel output, the system obtains a sequence of predicted values ​​every 3 minutes within a 15-minute window: Traffic congestion probability Identify the risk of traffic congestion in each grid area; Accident Risk Level Quantify the probability of a collision. When the LSTM prediction engine outputs the accident risk level... When the risk level is medium or higher, the system will activate a tiered intervention strategy: based on weighted scores. ( The strategies for improvement are ranked using preset weighting coefficients. High-risk areas ( It automatically obtains the highest priority and forces queue-jumping to execute the traffic light timing optimization strategy.

[0021] When generating detour route plans, the maximum traffic capacity of the road per unit time is taken into account. Minimum straight-line distance from the center point of the path to the hoisting machinery ratio Adjust road weighting. The higher the risk level, the higher the weighting. The greater the increase in the weight of the value path, the better. Similarly, for high-risk areas ( The system pushes a pulse warning box via AR-HUD (with the center point bound to the coordinates of a high-risk grid) and links roadside equipment to execute the highest priority policy: turning off the associated traffic light phase and writing the high-weight detour route plan into the variable message sign.

[0022] Regarding the selection of traffic light timing optimization strategies and detour route schemes, this invention generates a policy pool through the Proximal Policy Optimization (PPO) algorithm: First, a policy space containing 500 sets of policy vectors is initialized, with each set of vectors... Numbered traffic light timing scheme and bypass topology Then, the target loss function is used. Iterative optimization of the strategy: , Among them, the cutting factor Constraining the differences between old and new strategies, advantage estimation Based on dynamic weighting of accident risk levels ( (Time weight increased to 2.5 times). After policy optimization, the policy execution priority is calculated: by... For strategy ranking, priority increases by 0.1 for every 0.1 increase in score. The system is upgraded to a higher level; simultaneously, for detour route options, the road selection preference is adjusted based on a dynamic weighting mechanism, which means calculating the capacity-distance ratio of each route. Path weight Set as Sigmoid function output The weight of high-capacity paths is adjusted accordingly. It increases and approaches 1. The final output is a policy package with priority labels.

[0023] Upon receiving the policy packet with priority tags output from step S4, the system completes the linkage of roadside devices within 50ms via the V2X communication protocol: first, according to priority... The traffic light timing optimization strategy is implemented in descending order, and phase transition technology is used to optimize the timing scheme. The signal controller at the target intersection is injected with a 20% extension of the yellow light duration during the transition period to ensure a smooth traffic flow transition; simultaneously, based on the weight of detour paths... Filtering the optimal path The topology data is converted into a set of virtual lane coordinates that can be resolved by AR-HUD, and broadcast to vehicles within the construction impact area at a rate of 10 frames per second via roadside units (RSUs). A green arrow directional trajectory is dynamically rendered 30 meters in front of the driver's field of vision, with the trajectory width varying accordingly. The value is increased and expanded to 1.5 times the base width to enhance visual guidance. This is for grids prone to exceeding safety limits (…). ), synchronously activate the directional flashing sequence of the audible and visual alarm, the flashing frequency and The value is positively correlated (up to 10Hz), and text instructions with path codes and AR arrows are displayed on the upstream variable message signs (e.g., "Follow the green arrow to enter route B"). During all guidance and modification actions, the smart road beacon monitors vehicle response data in real time. If the actual diversion rate deviates from the expected value by more than 15%, the strategy package dynamic refresh loop is triggered.

[0024] In summary, the present invention proposes an intelligent early warning method for traffic diversion during bridge hoisting in complex urban environments. Based on construction influence domain analysis using dynamic grid modeling, it deeply couples the hoisting machinery's trajectory with real-time traffic flow parameters, solving the fundamental defect of traditional static diversion schemes that cannot respond to sudden mechanical displacement and traffic flow changes. This enables a leap from passive and delayed risk perception to proactive prediction.

[0025] Meanwhile, leveraging the temporal analysis capabilities of LSTM neural networks for spatiotemporal feature vectors, the system can predict congestion trends and accident risk levels within the construction impact domain in advance, thereby enabling early prevention and control decisions. Furthermore, by combining traffic light timing strategies and detour route schemes generated through reinforcement learning algorithms, a dynamic weighting mechanism is used to achieve optimal allocation of road network resources.

[0026] Ultimately, by leveraging augmented reality guidance and multi-source alarm linkage, a three-dimensional guidance and modification network with human-machine collaboration was built between physical roadside facilities and vehicle terminals, reducing the risk of driver misjudgment.

[0027] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0028] Furthermore, in this invention, descriptions involving terms such as "first," "second," and "a" are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0029] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0030] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

Claims

1. A bridge hoisting traffic guide intelligent early warning method for urban complex environment, characterized in that, The method comprises the steps of: S1: collecting hoisting machinery positioning data, vehicle dynamic data and original data of road monitoring in the construction area through an Internet of Things protocol; S2: after time-space alignment and noise filtering of the original data, a construction influence domain dynamic model is constructed by fusing the interaction relationship between the quantitative hoisting machinery motion trajectory and the traffic flow; S3: based on the time-space feature vector obtained from the construction influence domain dynamic model, the traffic congestion probability and accident risk level in the future preset time period are predicted through an LSTM neural network; S4: based on the prediction result of the LSTM neural network, a signal light timing optimization strategy and a detour path scheme are generated through a reinforcement learning algorithm, and the strategy execution priority is calculated; S5: according to the generated signal light timing optimization strategy and detour path scheme and the execution priority, adaptive guiding and changing actions are executed through V2X communication protocol and roadside equipment.

2. The bridge hoisting traffic guide intelligent early warning method for urban complex environment according to claim 1, characterized in that, In the S2 step, the fusion construction of the construction influence domain dynamic model is specifically as follows: With the lifting machinery as the center, the grid cell scale is established according to the swing radius of the lifting machinery The adaptive adjustment polar coordinate system dynamic grid, and record the mechanical parameters and traffic flow parameters in the grid cell; After the mechanical parameters and traffic flow parameters are given dynamic weights, the comprehensive risk value is calculated, and the influence of mechanical motion on traffic flow is quantified through time series analysis, which is expressed by the following formula: In the formula, respectively according to the grid The distance from the vehicle to the center of the hoisting machine in the grid , dynamically adjust the weights of the mechanical parameters and the traffic flow parameters, The vehicle flow density in the grid , The average vehicle speed in the grid , The vehicle acceleration standard deviation in the grid , The comprehensive risk value of the grid , The influence factor of the mechanical motion trajectory, The rotation angular velocity, The vehicle speed gradient change rate in the grid , The space-time feature vector.

3. The bridge hoisting traffic guide intelligent early warning method for urban complex environment according to claim 2, characterized in that, When the hoisting machinery radius of gyration invades the lane, if the grid The comprehensive risk value in the grid is greater than the preset warning threshold, the sound and light alarm and the variable information board are linked to issue a warning message.

4. The bridge hoisting traffic guide intelligent early warning method for urban complex environment according to claim 1, characterized in that, In the S3 step, the LSTM neural network predicts the traffic congestion probability and accident risk level in the future preset time period through the following formula: In the formula, is the current time, is the last time, is the unit state vector for storing long-term traffic time sequence characteristics, the traffic time sequence characteristics including traffic congestion probability and accident risk level , is a forgetting gate for controlling the retention ratio of historical information, is an input gate for controlling the input ratio of new information, is a neural connection weight matrix of the LSTM unit, is a bias vector, is the hidden layer output of the previous time t-1, is the space-time feature vector input at the current time t.

5. The bridge hoisting traffic guide intelligent early warning method for urban complex environment according to claim 1, characterized in that, In the S4 step, the reinforcement learning algorithm is a PPO algorithm, and the target loss function of the strategy optimization is: In the formula, is the target loss, is the expected operation of statistical average calculation of policy gradient estimation at the current t moment, is the clipping coefficient of the PPO algorithm, is the advantage estimation of the action value evaluated at the current t moment, is the probability ratio of the same action executed by the new and old policies at the current t moment, indicates that the probability ratio of the new and old policies is limited to the interval.

6. The bridge hoisting traffic guide intelligent early warning method for urban complex environment according to claim 1, characterized in that, In the S4 step, the policy execution priority is acquired according to the congestion probability and the weighted score of the accident level .

7. The bridge hoisting traffic guide intelligent early warning method for urban complex environment according to claim 1, characterized in that, When the S4 step generates the detour path scheme, according to the maximum traffic capacity per unit time of the road and the minimum straight line distance from the detour path center point to the hoisting machinery, the detour path is dynamically weighted based on the ratio between the two.

8. The bridge hoisting traffic guide intelligent early warning method for urban complex environment according to claim 1, characterized in that, In the S5 step, the adaptive guiding and changing actions include: real-time guiding and changing information is pushed to the driver's AR-HUD device, and the information includes virtual lane lines, no-entry areas and dynamic indications of detour paths.

9. The bridge hoisting traffic guide intelligent early warning method for urban complex environment according to claim 1, characterized in that, In the S1 step, the vehicle dynamic data is collected by the intelligent road stud deployed in the preset range around the construction area, which can sense vehicle trajectory and acceleration data in real time.

Images