A special transportation path online planning method based on real-time risk dynamic impedance

Abstract

The application discloses a kind of based on real-time risk dynamic impedance special transportation path online planning method, it is related to intelligent path planning technical field, including the following steps: extracting road environment prior vector and visual semantic feature are cross-processed, generate abnormal dangerous degree and abnormal coordinate area;Based on vehicle geometric parameter, road slope and abnormal coordinate area, the longitudinal variable slope passing rate of vehicle and the lateral turning passing rate are calculated and weighted to obtain comprehensive passing performance index;According to the topological graph of road network and the performance index, construct global generalized impedance function, and obtain the optimal benchmark path by optimization;When advancing along benchmark path, calculate road segment congestion index based on traffic state data and vehicle lane occupation characteristics, and obtain local dynamic impedance in combination with abnormal dangerous degree;Based on the transformation characteristics of local dynamic impedance, execute path online re-planning.The application is used to solve the problem that existing path planning method is prone to physical interference and frequent recalculation causes decision oscillation.

Description

Technical Field

[0001] This invention relates to the field of intelligent route planning technology, and more specifically, to a special transportation route online planning method based on real-time risk dynamic impedance. Background Technology

[0002] Special vehicle transportation tasks often involve the long-distance, cross-regional allocation of oversized cargo. Existing route planning methods mostly rely on static digital maps and conventional optimization algorithms to achieve end-to-end basic navigation guidance.

[0003] However, the existing technology still has the following technical problems in practical applications: 1. In the process of basic route output, due to the complex and ever-changing road environment, the static routes generated by existing technologies often have a low degree of adaptability to actual road conditions. At the same time, existing technologies rarely take into account the geometric dimensions of special vehicles and the spatial constraints of actual terrain when outputting routes, which makes vehicles prone to physical interference such as bottoming out and scraping on steep slopes, sharp bends or narrow road sections, thereby reducing the reliability of transportation and the success rate of tasks in complex environments. 2. In the dynamic execution phase of long-distance transportation, existing technologies are prone to over-responding to minor high-frequency traffic fluctuations or occasional situations in the road network, leading to frequent route jumps and recalculations, which in turn cause decision oscillations and reduce the smoothness of vehicle operation.

[0004] Therefore, there is an urgent need for a special transportation route planning method that can integrate multi-dimensional physical traffic constraints to generate a baseline path and has the ability to resist oscillations and perform online dynamic replanning. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an online planning method for special transportation routes based on real-time risk dynamic impedance. This method generates a benchmark route by combining visual perception positioning with vehicle geometric parameters to construct a comprehensive performance index, and performs online replanning based on the transformation characteristics of local dynamic impedance. This addresses the problems of physical interference and decision oscillation caused by frequent recalculation in existing route planning methods.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A special transportation route online planning method based on real-time risk dynamic impedance includes the following steps: Extract the environmental prior vectors and visual semantic features of the road and perform cross-processing to generate the road anomaly danger level and anomaly coordinate area; Based on vehicle geometric parameters, road slope, and the abnormal coordinate area, the longitudinal slope change rate and lateral turning rate of the vehicle are calculated, and the comprehensive passability index is obtained by weighted summation. A global generalized impedance function is constructed based on the road network topology and the performance index, and an optimization calculation is performed to obtain the optimal reference path; When traveling along the baseline path, the road segment congestion index is calculated based on traffic status data and vehicle lane occupation characteristics, and the local dynamic impedance is obtained in combination with the abnormal danger level. Based on the transformation characteristics of the local dynamic impedance, online replanning of the transportation path is performed.

[0007] In a preferred embodiment, generating the abnormal road hazard level and abnormal coordinate region includes: introducing a detection guidance marker as the query subject, performing cross-attention interaction with the visual semantic features and the environmental prior vector to obtain a risk assessment vector and a guidance vector; performing spatial attention mapping and residual fusion on the guidance vector and the deep semantic features contained in the visual semantic features to obtain an enhanced feature map; and decoding the risk assessment vector and the enhanced feature map respectively to obtain the abnormal road hazard level and abnormal coordinate region.

[0008] In a preferred embodiment, the longitudinal slope pass rate includes: calculating the front failure coefficient and the rear failure coefficient of the vehicle based on the vertical dimension parameters of the vehicle chassis and the slope angle data of the road; calculating the ratio of the front failure coefficient and the rear failure coefficient based on a preset characteristic length, and combining a preset longitudinal sensitivity coefficient to perform nonlinear mapping on the ratio that meets the preset conditions to obtain the longitudinal slope pass rate.

[0009] In a preferred embodiment, the formula for calculating the front failure coefficient and the rear failure coefficient of the vehicle is as follows: , , in, , These are the front and rear failure coefficients of the vehicle, respectively. , These are the radii of the front and rear wheels of the vehicle, respectively. , These refer to the front suspension height and the rear chassis height of the vehicle, respectively. The slope angle of the slope where the vehicle is located; The slope angle at which the vehicle will travel.

[0010] In a preferred embodiment, the lateral turning throughput includes: calculating the spatial distance between the vehicle's planar geometric parameters and the turning hinge angle based on trigonometric functions to obtain the trajectory radius of the vehicle when turning, and calculating the actual width of the turning lane occupied by the vehicle when turning by combining a preset safety margin; obtaining the physical width of the current road segment, and removing the lateral occupancy dimensions of the abnormal coordinate areas to obtain the effective road width; and nonlinearly mapping the ratio of the effective road width to the turning lane width to the lateral turning throughput by combining a preset lateral sensitivity coefficient.

[0011] In a preferred embodiment, obtaining the optimal baseline path includes: dividing the comprehensive throughput performance index into restricted and normal zones using a preset restricted threshold, and then constructing a global generalized impedance function; the impedance function maps the road segment impedances belonging to the restricted and normal zones to preset restricted and normal impedances, respectively; the normal impedance is obtained by weighted summation of the road segment travel time cost and a safety penalty term constructed based on the comprehensive throughput performance index; the impedance function is used to calculate the generalized impedance of the road segments in the road network topology map, and network node expansion and optimization are performed to obtain the optimal baseline path.

[0012] In a preferred embodiment, the road segment congestion index is calculated based on traffic state data and vehicle lane occupancy characteristics, using the following formula: , in, The congestion index of the road segment For road section exist Real-time equivalent standard traffic flow at any given moment. and For calibration parameters; The actual effective traffic capacity of a road segment is obtained by reducing the lane occupancy characteristics based on vehicle lane occupancy.

[0013] In a preferred embodiment, obtaining the local dynamic impedance includes: extracting the time-varying features of the abnormal danger level, and generating a sudden event penalty term based on the comparison results of the time-varying features and the preset blocking judgment threshold, combined with the preset event penalty amplification coefficient and the global generalized impedance of the corresponding road segment; and performing superposition calculation on the global generalized impedance of the corresponding road segment, the road segment congestion index and the sudden event penalty term to obtain the local dynamic impedance.

[0014] In a preferred embodiment, the online replanning of the path based on the transformation characteristics of the local dynamic impedance includes: calculating the cumulative impedance increment of the local dynamic impedance of the road segment within a preset window relative to the corresponding global generalized impedance, and constructing a multi-dimensional replanning criterion that includes at least a first triggering condition based on the sudden event penalty term and a second triggering condition based on the cumulative impedance increment; and, based on the triggering determination result of the replanning criterion, suspending the execution of the current optimal baseline path and using the road network node mapped to the current vehicle position as the new starting point to perform online replanning of the transportation path.

[0015] A special transportation route online planning system based on real-time risk dynamic impedance includes the following modules: a perception and positioning module, used to extract the environmental prior vector and visual semantic features of the road and perform cross-processing to generate the road anomaly hazard level and anomaly coordinate area; a performance evaluation module, used to calculate the vehicle's longitudinal slope change throughput rate and lateral turning throughput rate based on vehicle geometric parameters, road slope, and the anomaly coordinate area, and obtain a comprehensive throughput performance index through weighted summation; a global planning module, used to construct a global generalized impedance function based on the road network topology map and the performance index, and perform optimization calculation to obtain the optimal baseline path; a dynamic impedance module, used to calculate the road segment congestion index based on traffic state data and vehicle lane occupation characteristics while traveling along the baseline path, and obtain the local dynamic impedance in combination with the anomaly hazard level; and a replanning module, used to perform online replanning of the transportation route based on the transformation characteristics of the local dynamic impedance.

[0016] The technical effects and advantages of the online planning method for special transportation routes based on real-time risk dynamic impedance of the present invention are as follows: 1. This invention obtains abnormal coordinate regions by cross-processing prior vectors of the road environment with visual semantic features, and calculates a comprehensive throughput performance index by combining vehicle geometric parameters and road slope data, and embeds it into the road network topology map. This enables a refined characterization of the passage capability of special transport vehicles in complex road environments, effectively improving the accuracy and environmental adaptability of special transport baseline path generation.

[0017] 2. This invention constructs an online dynamic replanning mechanism that responds to the evolution of road network risk conditions by using local dynamic impedance transformation characteristics determined based on traffic state data, vehicle lane occupancy characteristics, and abnormal risk levels. This achieves closed-loop decision-making with global path guidance and local adaptive adjustment, reduces the risk of path failure and decision oscillation caused by dynamic traffic fluctuations, and ensures the robustness of online path planning and the reliability of transportation execution throughout the special transportation process. Attached Figure Description

[0018] Figure 1This is a schematic diagram of a special transportation route online planning method based on real-time risk dynamic impedance, provided as an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of a special transportation route online planning system module based on real-time risk dynamic impedance, provided for an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1, Figure 1 This invention presents an online planning method for special transportation routes based on real-time risk dynamic impedance, comprising the following steps: S1. Extract the environmental prior vectors and visual semantic features of the road and perform cross-interaction to generate perceptual localization results of road anomalies, including: S101. Obtain a multi-dimensional road event dataset and construct a joint loss function to train the perception model using multi-modal fusion, as detailed below: Acquire a number of road event image samples (e.g., more than 9,700 cases) covering fixed monitoring perspectives (such as tunnels and bridges), aerial perspectives from drone inspections (such as landslides and collapses), and perspectives from vehicle recorders (such as collisions and debris). The image samples include extreme weather conditions such as rain, snow, and fog, as well as scenes with changes in light and shadow. Spatial bounding box annotations and semantic risk level labels are applied to each image sample to construct a road event dataset. During the model training phase, image samples from the road event dataset are input into the perception model. Through feature extraction and cross-attention interaction, the model outputs the predicted risk level and predicted coordinate region. A joint loss function is constructed to calculate the predicted risk level. Corresponding to the actual risk level label The risk prediction error between them, and the prediction coordinate region. With the corresponding true bounding box The positioning term error between the risk prediction error and the positioning term error is calculated, and the risk prediction error is weighted and summed with the positioning term error to obtain the joint loss. The formula for the joint loss function is as follows: , , , in, For risk prediction error, For positioning error, For joint losses; , , , All are preset weighting coefficients, for example, weighting coefficients are set to 1.0, 0.1, 1.0, and 2.0 respectively; the risk prediction error is enhanced by the mean square error (MSE) to improve the perception model's ability to quantify the degree of danger, and the location item error is... The geometric alignment accuracy between the predicted coordinate region and the true bounding box is optimized by using binary cross-entropy loss (BCE) and KL divergence. Backpropagation is performed with the joint loss as the target, and the parameters of the perceptual model are iteratively updated using the AdamW optimizer until the joint loss function converges or reaches the preset maximum number of training rounds. The learning rate of the AdamW optimizer can be set to 0.0003, the gradient accumulation step can be set to 5, the maximum number of training rounds can be set to 10, and the number of steps per round can be set to 500. The convergence condition can be set to the change in the joint loss between two adjacent iterations being less than a preset change threshold of 10⁻⁴.

[0022] S102. Extract visual semantic features from real-time road images, and encode the basic features of the road scene and real-time environmental data into an environmental prior vector, as follows: Acquire real-time road images and perform resolution scaling (e.g., unify to a higher resolution). After pixel normalization preprocessing, the CLIP visual encoder in the trained perception model is used to extract the visual semantic feature vector of the preprocessed real-time road image. Simultaneously, to suppress perceptual interference caused by drastic changes in light and shadow under different road environments and extreme weather, the environmental information encoding module in the perception model is invoked to construct structured text prompts from the basic features of the road scene and real-time environmental data obtained from environmental monitoring stations. The word segmenter and multilayer perceptron (MLP) in the perception model are then used for word segmentation encoding and feature mapping, transforming the result into an environmental prior vector. The basic scene features include road type features, road morphology features, lane environment features, traffic facility features, and terrain area features. The real-time environmental data includes weather conditions, time period information, precipitation intensity, visibility, and lighting conditions. The calculation formulas for the visual semantic features and the environmental prior vector are as follows: , , in, It is a visual semantic feature, and ; Real-time images of the road; Let be the prior vector of the environment, and MLP stands for Multilayer Perceptron Network; These are the basic features of the scene, including road type and terrain area data; This provides real-time environmental data, including precipitation intensity and visibility data.

[0023] S103. Perform cross-attention interaction between the visual semantic features and the environmental prior vector, and output the perceptual localization result of the road anomaly through spatial attention mapping, as follows: To achieve the mapping from high-dimensional semantic logic to low-dimensional spatial coordinate regression, the detection guidance marker (DTTN) embedded in the inference layer during perceptual model training is used as the query subject. This DTTN undergoes deep cross-interaction with the visual semantic features and the environment prior vector, i.e., cross-attention interaction. During this deep cross-interaction, the DTTN is used as the query matrix, and the environment prior vector is used as a conditional constraint to construct a key matrix and a value matrix. Attention weights are adaptively adjusted by calculating dot product similarity, thereby aggregating pixel features related to road anomaly morphology within the visual semantic features. Finally, the DTTN itself is transformed into a risk assessment vector and a guidance vector with scene semantics. The calculation formula is as follows: , in, For risk assessment vector and , For guiding vector and CA stands for Cross-Attention Mechanism; Subsequently, the guiding vector is used as a query signal to perform a second cross-attention interaction globally with the visual semantic features, so as to use the guiding vector to perform spatial attention mapping on the original pixel features; further, the semantic features after attention mapping are residually fused with the visual semantic features to obtain an enhanced feature map; using the risk decoder and detection decoder configured at the output end of the perception model, regression calculations are performed on the risk assessment vector and the enhanced feature map respectively, directly outputting the degree of abnormal risk and the abnormal coordinate region on the road, as shown in the following formula: , , in, The degree of abnormal risk, and ; This is an abnormal coordinate region, and ,in The four vector components are defined as representing the [top left x-coordinate, top left y-coordinate, bottom right x-coordinate, bottom right y-coordinate] of the bounding box.

[0024] This step effectively suppresses background noise caused by extreme weather and changes in light and shadow by fusing environmental prior vectors and visual features and adaptively adjusting attention weights using detection-guided markers. At the same time, it improves the localization accuracy of small-scale abnormal targets in complex backgrounds by combining cross-attention interaction and residual fusion mechanisms. In addition, it uses a decoder to directly regress the output of the abnormal category and spatial coordinates, reducing the computational delay of the inference process and achieving efficient end-to-end perception and localization.

[0025] S2. Based on vehicle geometric parameters, road slope data, and the aforementioned abnormal coordinate region, calculate the vehicle's longitudinal slope change rate and lateral turning rate respectively, and sum them by weight to obtain a comprehensive passability index, including: S201. Calculate the longitudinal gradient pass rate of the vehicle, as follows: When heavy transport vehicles pass through longitudinal slope change areas (such as slope change points or gentle transition sections at slope change points), due to limitations in vehicle wheelbase and chassis ground clearance, the bottom of the vehicle body may experience spatial interference with the road structure, such as "bottoming out" or "collision." This embodiment uses the vehicle's front wheels, rear wheels, and floor boundaries as feature analysis points to perform vertical geometric failure determination on heavy transport vehicles, thereby quantitatively assessing the vehicle's passing safety. Specific steps include: First, based on the obtained vertical dimensions of the vehicle chassis and the slope angle data of the road, the failure coefficients of the front and rear of the vehicle are calculated using the following formulas: , , in, , These are the failure coefficients for the front and rear of the vehicle, respectively. , These are the radii of the front and rear wheels of the vehicle, respectively. , These refer to the front suspension height and the rear chassis height of the vehicle, respectively. The slope angle of the slope where the vehicle is located. The slope angle at which the vehicle will travel; Next, preset front and rear characteristic lengths are introduced. The front characteristic length can be taken as the front overhang length of the tractor, that is, the horizontal distance from the center of the front axle to the frontmost part of the vehicle body (for example, 1.5m); the rear characteristic length can be taken as the rear overhang length of the semi-trailer, that is, the horizontal distance from the center of the rear axle to the rearmost part of the vehicle body (for example, 2.5m). The front failure coefficient and the rear failure coefficient are divided by the front characteristic length and the rear characteristic length, respectively, and the resulting ratios are used as the front ramp pass index and the rear ramp pass index, respectively. The reciprocal of the maximum value of the front ramp pass index and the rear ramp pass index is used as the longitudinal matching coefficient to characterize the ratio of road supply to vehicle demand. Finally, based on the preset longitudinal sensitivity coefficient, the longitudinal matching coefficient is nonlinearly mapped using the Logistic function to obtain the longitudinal gradient throughput. The formula for the nonlinear mapping is as follows: , in, This refers to the longitudinal slope passability; The preset longitudinal sensitivity coefficient is used to adjust the mapping slope when the longitudinal matching coefficient is converted into a performance index near the critical point (i.e., the longitudinal matching coefficient equals 1). The longitudinal matching coefficient is used. Since heavy cargo transportation operations have high safety redundancy requirements for road throughput parameters, when the road supply capacity is equal to the vehicle throughput demand (i.e., the longitudinal matching coefficient is equal to 1), the actual operation is already in a critical risk state. Therefore, this embodiment configures a large longitudinal sensitivity coefficient (recommended value range is 12 to 20) to ensure that when the longitudinal matching coefficient is slightly less than 1, the longitudinal slope change throughput rate drops rapidly to below 0.35 and is judged as impassable. When the longitudinal matching coefficient is slightly greater than 1.2 (i.e., the road supply exceeds the vehicle demand by 20%), the longitudinal slope change throughput rate quickly approaches 1, thereby realizing nonlinear real-time early warning of transportation risks.

[0026] S202. Calculate the lateral turning pass rate for vehicles traveling on curves, as follows: First, spatial distance calculations based on trigonometric functions are performed on the vehicle's planar layout geometry parameters and turning articulation angles to obtain the innermost turning radius (usually the trajectory of the trailer's inner wheel) and the outermost turning radius (corresponding to the trajectory of the tractor's front overhang) when the vehicle turns. The calculation formulas are as follows: , , in, , These are the innermost turning radius and the outermost turning radius, respectively. This is the distance from the towing pin to the center axis of the trailer's rotation. For trailer width, For the width of the tractor unit, This is the wheelbase from the front axle to the rear axle of the tractor unit. This is the offset distance between the traction pin and the rear axle of the tractor vehicle. This refers to the front overhang length of the tractor unit. The dynamic articulation angle of the vehicle when turning is obtained by interpolating the turning radius of the current road centerline obtained from the map database and inputting it into a preset "curvature-articulation angle" mapping table. This is used to characterize the vehicle's true folding posture. The mapping table is constructed for a specific vehicle type (such as a tractor-semi-trailer) through offline kinematic simulation or real vehicle trajectory calibration. Next, based on the preset safety margin, the actual width of the turning lane occupied by the vehicle in the curve is calculated using the following formula: , in, The actual width of the turning lane occupied; the constant 0.5 is the preset minimum safety margin on one side, used to compensate for driver operation deviations and dynamic sweeping errors under heavy loads; Then, the original physical width of the current road segment is retrieved from the map database. Based on the pre-calibrated camera intrinsic and extrinsic parameter matrices of the visual perception device, the abnormal coordinate region identified in step S1 is projected onto the ground three-dimensional coordinate system of the road cross-section through homography transformation. The actual lateral width occupied by the abnormal obstacle is calculated. The occupied lateral width is subtracted from the original physical width of the road to obtain the effective width of the road segment after removing the obstacle. The ratio of the effective width of the road to the width of the dynamic turning channel is used as the lateral matching coefficient. Based on the preset lateral sensitivity coefficient, the lateral matching coefficient is mapped to the lateral turning pass rate using the Logistic function, as shown in the following formula: , in, This refers to the lateral turning passability. The preset horizontal sensitivity coefficient (recommended value range is 12 to 20). The lateral matching coefficient is used; when the lateral turning pass rate approaches 0, it indicates that the vehicle's sweeping trajectory has encroached on the space outside the road surface boundary. At this time, the system triggers a path optimization warning or a widening modification suggestion. When the lateral turning pass rate approaches 1, it is determined that the lateral traffic requirements are met.

[0027] S203. The longitudinal gradient throughput rate and the lateral turning throughput rate are weighted and summed to obtain the comprehensive throughput performance index, as follows: Because mountain roads are often characterized by sharp bends and steep slopes, any mismatch in any dimension could lead to the failure of the entire heavy-haul transportation mission. To objectively assess the overall road adaptability, a comprehensive throughput performance index, including penalty mechanisms and weighted compensation, is constructed. , Wherein, PI is the overall pass performance index; The weight for the penalty for the weakest link is [0,1], for example, 0.7. The higher the value, the more the overall evaluation result is dominated by the worst-performing item in each dimension, which is used to prevent the good performance of a single dimension from masking the security bottleneck of another dimension. , These are the dimensional compensation weights, representing the importance of the vertical and horizontal evaluation dimensions, respectively, and satisfying the following conditions: ,For example , Each is 0.5.

[0028] This step establishes geometric constraint mappings in the vertical and horizontal dimensions and uses a nonlinear logistic function to capture and quantify traffic risks. By calculating a comprehensive index that includes a bottleneck penalty mechanism, it effectively avoids the masking of global bottleneck risks by local safety margins, providing route optimization decision support with high safety redundancy for heavy cargo transportation.

[0029] S3. Based on the road network topology and the comprehensive throughput index, a global generalized impedance function is constructed and the optimal baseline path is obtained through optimization, as follows: Obtain the topology map of a complex mountain road network and construct a global generalized impedance function for multi-objective optimization. Specifically, for any adjacent nodes in the road network... and Construct its directed road segments Global generalized impedance function The global generalized impedance function uses a preset forbidden threshold to divide the synthesis pass performance index into forbidden intervals and normal intervals, and is calculated using a piecewise function form. The formula for the global generalized impedance function is as follows: , in, For road section The original physical length; The expected average speed of heavy transport vehicles is preset based on expert experience. For regular road sections, it can be 30km / h to 60km / h, while for bridge sections and mountain curves, it should not exceed 20km / h. For road section The overall performance index; A preset no-entry threshold (e.g., set to 0.35) is used to define the no-entry zone (i.e., ...). ) and regular interval (i.e. ); and These are the preset time cost weight and security penalty weight, respectively. If safety requirements outweigh traffic efficiency, then... , ; This is a safety penalty term that utilizes an inverse proportional offset characteristic to characterize road conditions (i.e., The worse the conditions, the higher the hidden costs of slow maneuvering or physical elevation required for vehicle movement; when When a road segment falls into the restricted area, the generalized impedance of the corresponding road segment is a preset restricted impedance (i.e., infinity) to perform physical blocking of the road segment at the topology level; when When the path falls within the conventional range, the generalized impedance of the corresponding road segment is mapped to the conventional impedance obtained by weighted calculation. In a complex mountain road network topology with a high-dimensional impedance matrix, based on the generalized impedance calculated using the impedance function, an improved A* algorithm is employed to perform network-level node expansion for rapid global optimization, yielding an optimal baseline path with the lowest overall impedance. The comprehensive evaluation function of the improved A* algorithm is as follows: , in, For the currently traversed node The overall estimated cost; the actual cost This represents the distance from the starting point to the current node. The cumulative global generalized impedance (i.e., all impedances on the planned path) (cumulative value); heuristic function Indicates starting from the current node The estimated cost to reach the target node is the Manhattan distance (km) between the current node and the target node. ,in and These are the spatial coordinates of the current node and the target node, respectively. This is the scale conversion factor, used to convert spatial distance into the estimated cost of the cumulative global generalized impedance at the same scale. Its value can be set according to the time cost per unit distance, the historical statistical impedance mean, or the simulation calibration results, and can be 0.05. Specifically, the improved A* algorithm's pathfinding iterative process is based on maintaining the node's table of unupdated nodes and the table of already updated nodes, and the steps are as follows: (1) Clear the table to be updated and the table that has been updated, add the initial node (i.e. the planning starting point) to the table to be updated, and set the actual cost value of the initial node to 0; (2) Extract the comprehensive estimated cost from the table to be updated. The smallest node is taken as the current node; if the current node is the target node, the path finding is determined to be successful and the termination condition is triggered to proceed to step (5); otherwise, find all reachable neighboring nodes of the current node in the road network, skip the nodes that already exist in the updated table and the nodes whose global generalized impedance of the corresponding directed road segment is the forbidden impedance (i.e., infinity), and extract the set of remaining reachable neighboring nodes. (3) Calculate the actual cost and the comprehensive estimated cost of each neighboring node in the set of neighboring nodes; if the neighboring node does not exist in the table to be updated, add the neighboring node to the table to be updated and set the current node as the parent node of the neighboring node; if the neighboring node already exists in the table to be updated, and the verification finds that the actual cost of reaching the neighboring node through the current node is smaller, force the current node to be updated as the parent node of the neighboring node, and recalculate the comprehensive estimated cost of the neighboring node. (4) Delete the current node from the table to be updated and add it to the updated table, then return to step (2) to continue the loop iteration; (5) Repeat steps (2) to (4) until the current node extracted from the table to be updated is the target node (pathfinding successful), or the table to be updated is empty (pathfinding failed); after the pathfinding is successful, start from the target node and backtrack along the parent node pointer of each node to the initial node, and the optimal reference path with the lowest comprehensive impedance can be output.

[0030] This step constructs a piecewise global generalized impedance function and uses a preset no-entry threshold to directly and physically block high-risk road sections at the road network topology level, ensuring basic traffic safety. At the same time, by combining preset time cost weights and safety penalty weights, an objective quantitative balance between traffic efficiency and implicit traffic costs is achieved. Furthermore, an improved pruning and skipping mechanism for no-entry impedance nodes in the A* algorithm is adopted, effectively avoiding the expansion of invalid nodes and the waste of computing power. While reducing the dimensionality of solving complex mountain road networks, this step improves the computational efficiency and rationality of searching for the global optimal baseline path.

[0031] S4. While traveling along the baseline path, the road segment congestion index is calculated based on traffic state data and vehicle lane occupancy characteristics, and then superimposed on the corresponding global generalized impedance to obtain the local dynamic impedance, as follows: The heavy vehicle travels to the current node according to the optimal baseline path. At that time, the system obtains the directional road segment ahead in real time. Traffic status data is used, and a traffic flow density assessment model based on road resistance function is adopted to calculate the real-time congestion index. The calculation formula is as follows: , in, Real-time congestion index for the road segment; For road section exist Real-time equivalent standard traffic flow (unit: pcu / h) can be obtained in real time through road IoT sensors or checkpoint monitoring data. and To calibrate the parameters for the model, where , It can be set to 0.1; The actual effective traffic capacity of this road section is calculated by reducing the capacity based on vehicle lane occupancy characteristics. The specific reduction formula is as follows: ,in, This refers to the original design capacity of a road segment obtained by reading high-precision map data. This refers to the total number of lanes in one direction on the road segment. This refers to the actual number of lanes occupied by large vehicles; based on the calculation formula of the real-time congestion index, when... When the current traffic flow is in a free-flow or steady-flow state, the congestion index is calculated. If the value approaches 0, it indicates that the road segment is in a state of forced flow (oversaturation), and the congestion index increases exponentially. Based on the real-time congestion index and this road segment The global generalized impedance, superimposed with a sudden event penalty term, yields the value for this road segment. The local dynamic impedance, and the calculation formulas for the sudden event penalty term and the local dynamic impedance are as follows: , , in, This is a penalty item for emergencies; This is a preset event penalty amplification factor, used to adjust the sensitivity of sudden events to the overall impedance. Its value range can be set to... ; For the corresponding road section The global generalized impedance; The degree of abnormal risk output for step S1 exist The quantization characteristics of time, with a value range of This is used to characterize the potential impact of a sudden event on vehicle traffic. The preset blocking threshold can be set from 0.85 to 0.95. For road section exist Local dynamic impedance at time t, The real-time congestion index; The preset congestion impedance weighting coefficient can be set from 0.8 to 1.8.

[0032] This step calculates local dynamic impedance based on global impedance, real-time congestion index, and sudden event penalties. It objectively quantifies the capacity reduction and real-time congestion risk caused by large vehicles occupying lanes, giving the system dynamic risk perception capabilities and effectively avoiding the risk of sudden failure of static paths.

[0033] S5. Based on the transformation characteristics of the local dynamic impedance, online replanning of the path is performed, as follows: In actual large-item transportation, in order to avoid frequent invalid replanning (i.e. path oscillation) caused by small high-frequency fluctuations in traffic flow, the system constructs a local dynamic replanning mechanism that responds to multi-dimensional replanning criteria; the system calculates the difference between the local dynamic impedance and the global generalized impedance of the road segment ahead in real time as the dynamic impedance increment to assess the degree of deterioration of traffic conditions. The multidimensional reprogramming criterion is composed of the sudden event penalty term and the ratio of the cumulative value of dynamic impedance increment to the corresponding global generalized impedance sum, specifically including: A strong triggering mechanism based on the sudden event penalty term: When the system detects a complete blockage-type sudden event in the direction of the vehicle's travel along the optimal baseline path, the abnormal risk level of the corresponding road segment is greater than the blockage judgment threshold. Furthermore, the sudden event penalty term tends to infinity, directly triggering path replanning. (2) Weak triggering mechanism based on ratio: Calculate the ratio of the cumulative value of dynamic impedance increment within the preset window to the sum of global generalized impedance, and compare the ratio with the preset tolerance threshold (e.g., 0.2). When the ratio is greater than the tolerance threshold, path replanning is triggered. The preset window is set based on actual working conditions. When the road network nodes are dense and the length of a single road segment is short, 4 to 6 road segments can be selected as the cumulative window. When the node coefficient and the length of a single road segment are long enough to characterize the change in road status, a single road segment is preferred for replanning judgment. When any of the triggering conditions in the multidimensional replanning criteria is activated, the system suspends the execution of the current optimal baseline path. At this time, the system takes the node of the vehicle's current perceived location in the road network topology as the starting point for the new path planning, while retaining the target node of the original transportation task, and calls the improved A* algorithm in step S3 again to perform network-level node expansion and decision update, thereby outputting a new optimal path that avoids the current high-risk area, and realizing online dynamic replanning of the transportation path.

[0034] Example 2, Figure 2 This invention presents an online special transportation route planning system based on real-time risk dynamic impedance, comprising the following modules: The perception and localization module is used to extract the environmental prior vectors and visual semantic features of the road and perform cross-processing to generate the road anomaly danger level and anomaly coordinate area. The performance evaluation module is used to calculate the vehicle's longitudinal slope change rate and lateral turning rate based on the vehicle's geometric parameters, road slope, and the abnormal coordinate area, and obtain the comprehensive passability index by weighted summation. The global planning module is used to construct a global generalized impedance function based on the road network topology map and the performance index, and to perform optimization calculations to obtain the optimal baseline path; The dynamic impedance module is used to calculate the road congestion index based on traffic state data and vehicle lane occupation characteristics when traveling along the reference path, and to obtain the local dynamic impedance in combination with the abnormal danger level. The replanning module is used to perform online replanning of the transportation path based on the transformation characteristics of the local dynamic impedance.

[0035] The above formulas are all dimensionless calculations. The formulas are derived from software simulations using a large amount of collected data, and are the closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0036] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0037] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0038] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0039] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0040] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A special transportation route online planning method based on real-time risk dynamic impedance, characterized in that, Includes the following steps: Extract the environmental prior vectors and visual semantic features of the road and perform cross-processing to generate the road anomaly danger level and anomaly coordinate area; Based on vehicle geometric parameters, road slope, and the abnormal coordinate area, the longitudinal slope change rate and lateral turning rate of the vehicle are calculated, and the comprehensive passability index is obtained by weighted summation. A global generalized impedance function is constructed based on the road network topology and the performance index, and an optimization calculation is performed to obtain the optimal reference path; When traveling along the baseline path, the road segment congestion index is calculated based on traffic status data and vehicle lane occupation characteristics, and the local dynamic impedance is obtained in combination with the abnormal danger level. Based on the transformation characteristics of the local dynamic impedance, online replanning of the transportation path is performed.

2. The method according to claim 1, characterized in that, The generated road anomaly hazard level and anomaly coordinate area include: A detection guidance marker is introduced and used as the query subject. It interacts with the visual semantic features and the environmental prior vector across attention to obtain the risk assessment vector and the guidance vector. The guided vector and the deep semantic features contained in the visual semantic features are subjected to spatial attention mapping and residual fusion to obtain an enhanced feature map; The risk assessment vector and the enhanced feature map are decoded respectively to obtain the abnormal danger level and abnormal coordinate area of ​​the road.

3. The method according to claim 1, characterized in that, The longitudinal slope passability includes: Based on the vehicle chassis vertical dimension parameters and road slope angle data, calculate the front and rear failure coefficients of the vehicle. Based on the preset characteristic length, the ratio of the front failure coefficient to the rear failure coefficient is calculated respectively, and combined with the preset longitudinal sensitivity coefficient, the ratio that meets the preset conditions is nonlinearly mapped to obtain the longitudinal slope change rate.

4. The method according to claim 3, characterized in that, The formulas for calculating the front and rear failure coefficients of a vehicle are as follows: ， ， in, , These are the front and rear failure coefficients of the vehicle, respectively. , These are the radii of the front and rear wheels of the vehicle, respectively. , These refer to the front suspension height and the rear chassis height of the vehicle, respectively. The slope angle of the slope where the vehicle is located; The slope angle at which the vehicle will travel.

5. The method according to claim 1, characterized in that, The lateral turning passability includes: The spatial distance is calculated based on trigonometric functions using the vehicle's planar layout geometric parameters and turning hinge angle to obtain the trajectory radius of the vehicle when turning. Combined with a preset safety margin, the actual width of the turning lane occupied by the vehicle when turning is calculated. Obtain the physical width of the current road segment, and remove the lateral occupancy of the abnormal coordinate area to obtain the effective road width; By combining a preset lateral sensitivity coefficient, the ratio of the effective road width to the turning lane width is non-linearly mapped to the lateral turning pass rate.

6. The method according to claim 1, characterized in that, The process of obtaining the optimal baseline path includes: By using a preset no-entry threshold, the comprehensive pass performance index is divided into a no-entry interval and a normal interval, and then a global generalized impedance function is constructed. The impedance function maps the road segment impedances belonging to the restricted and regular zones to preset restricted and regular impedances, respectively; the regular impedance is obtained by weighted summation of the road segment travel time cost and a safety penalty term constructed based on the comprehensive throughput performance index; The generalized impedance of road segments in the road network topology is calculated using the impedance function, and network node expansion and optimization are performed to obtain the optimal baseline path.

7. The method according to claim 1, characterized in that, The road segment congestion index is calculated based on traffic condition data and vehicle lane occupation characteristics, using the following formula: ， in, The congestion index of the road segment For road section exist Real-time equivalent standard traffic flow at any given moment. and For calibration parameters, The actual effective traffic capacity of a road segment is obtained by reducing the lane occupancy characteristics based on vehicle lane occupancy.

8. The method according to claim 1, characterized in that, The acquisition of local dynamic impedance includes: Extract the time-varying features of the abnormal danger level, and based on the comparison results of the time-varying features and the preset blocking judgment threshold, combine the preset event penalty amplification coefficient and the global generalized impedance of the corresponding road segment to generate a sudden event penalty item; The local dynamic impedance is obtained by superimposing the global generalized impedance of the corresponding road segment, the road segment congestion index, and the sudden event penalty term.

9. The method according to claim 8, characterized in that, The online replanning of the path based on the transformation characteristics of the local dynamic impedance includes: Calculate the cumulative value of the local dynamic impedance of the road segment within the preset window relative to the corresponding global generalized impedance, and construct a multi-dimensional reprogramming criterion that includes at least a first triggering condition based on the sudden event penalty term and a second triggering condition based on the cumulative value of impedance increment. Based on the triggering determination result of the replanning criterion, the execution of the current optimal baseline path is stopped, and the online replanning of the transportation path is performed with the road network node mapped to the current position of the vehicle as the new starting point.

10. A special transportation route online planning system based on real-time risk dynamic impedance, characterized in that, Includes the following modules: The perception and localization module is used to extract the environmental prior vectors and visual semantic features of the road and perform cross-processing to generate the road anomaly danger level and anomaly coordinate area. The performance evaluation module is used to calculate the vehicle's longitudinal slope change rate and lateral turning rate based on the vehicle's geometric parameters, road slope, and the abnormal coordinate area, and obtain the comprehensive passability index by weighted summation. The global planning module is used to construct a global generalized impedance function based on the road network topology map and the performance index, and to perform optimization calculations to obtain the optimal baseline path; The dynamic impedance module is used to calculate the road congestion index based on traffic state data and vehicle lane occupation characteristics when traveling along the reference path, and to obtain the local dynamic impedance in combination with the abnormal danger level. The replanning module is used to perform online replanning of the transportation path based on the transformation characteristics of the local dynamic impedance.

Images