Logistics and supply chain management method based on compass positioning and path planning algorithm

Abstract

The application discloses a logistics and supply chain management method based on a Beidou positioning and path planning algorithm, and comprises the following steps: receiving a transportation task issued by a dispatching center, and generating task initialization parameters; calling a map service engine to generate a candidate path set and an initial execution path; in the process that a vehicle travels according to the initial execution path, a running state vector is generated in combination with a Beidou real-time position; threshold determination and time sequence identification are performed on the running state vector to generate event features; the running state vector and the event features are input into a hierarchical reinforcement learning model to obtain an emergency action scheme; scheduling execution is performed based on the emergency action scheme to form emergency execution result information; and the running state vector, the event features, the emergency action scheme and the emergency execution result information are visually displayed to generate a transportation report. The application adopts hierarchical reinforcement learning to realize logistics and supply chain management.

Description

Technical Field

[0001] The present invention relates to the technical field of logistics and supply chain management, and particularly to a logistics and supply chain management method based on Beidou positioning and path planning algorithms. Background Art

[0002] With the rapid development of logistics and supply chain operations, path planning and operation monitoring of transport vehicles in complex environments have become the key factors restricting efficiency and cost. Existing logistics path planning mostly relies on GPS positioning and traditional algorithms, usually with distance or time as the main optimization objectives. Some systems introduce traffic information for correction, but lack comprehensive consideration of cargo attributes, vehicle energy consumption characteristics, and positioning errors, making it difficult to meet the refined requirements of cold chain transportation and heavy haul transportation. In terms of operation monitoring, existing systems mainly rely on OBD or single-point sensors, and perform anomaly detection through threshold judgment methods, which can identify situations such as excessive rotation speed and insufficient tire pressure. However, they lack analysis of continuous time series data, resulting in prominent false alarm and missed alarm problems, and are unable to achieve hierarchical identification and intelligent disposal in case of emergencies.

[0003] In emergency handling, some platforms adopt the method of manually setting rules to execute detours, repairs, or vehicle replacements. However, the response depends on manual intervention and lacks the ability to dynamically balance timeliness, energy consumption, and risks, resulting in less refined emergency decisions. Existing research has proposed methods based on reinforcement learning for scheduling optimization, but most are single-layer models and are difficult to balance global decision-making and local execution in multi-level linkage scenarios. Regarding the storage of transportation data, there are currently blockchain applications for logging on the chain, but most stay at the level of result recording, lack integration with operation status, abnormal events, and emergency results, and are difficult to fully reflect the entire transportation process, nor can they meet the requirements of supply chain management for transparency and traceability.

[0004] Therefore, how to provide a logistics and supply chain management method based on Beidou positioning and path planning algorithms is an urgent problem for those skilled in the art. Summary of the Invention

[0005] An object of the present invention is to propose a logistics and supply chain management method based on Beidou positioning and path planning algorithms. The present invention makes full use of Beidou high-precision positioning, path cost function modeling, time-series anomaly identification, and hierarchical reinforcement learning multi-level linkage mechanism, and details the whole process of realizing path optimization, abnormal event identification, emergency action generation, and result storage in transportation tasks, with the advantages of more refined path planning, more intelligent abnormal handling, and transparent and traceable supply chain management.

[0006] The logistics and supply chain management method based on Beidou positioning and path planning algorithms according to an embodiment of the present invention includes the following steps:

[0007] Receive the transportation task issued by the dispatch center, start the vehicle-mounted Beidou terminal to obtain the vehicle's initial latitude, longitude, speed and heading information, and generate task initialization parameters;

[0008] The map service engine is invoked to generate a set of candidate paths, and a path cost function is constructed based on the task initialization parameters to generate the initial execution path;

[0009] During the vehicle's journey along the initial execution path, data from OBD, inertial measurement unit, and onboard camera are continuously collected. Combined with BeiDou real-time location and road conditions, an operational status vector is generated.

[0010] Threshold determination and time sequence identification are performed on the running status vector to obtain abnormal event types and severity level candidates, and event features are generated;

[0011] The running state vector and event features are input into a hierarchical reinforcement learning model. The high-level policy network outputs the linkage level, and the low-level execution network generates emergency action parameters to obtain an emergency action plan.

[0012] Based on the emergency response plan, dispatch and execute the plan, and generate emergency execution result information;

[0013] The system visualizes and stores information on operational status vectors, event characteristics, emergency action plans, and emergency execution results in a chain, generates transportation reports, and sends them back to the supply chain management platform.

[0014] Optionally, the process of receiving a transportation task from the dispatch center, activating the vehicle-mounted BeiDou terminal to obtain the vehicle's initial latitude, longitude, speed, and heading information, and generating task initialization parameters specifically includes:

[0015] Receive transportation task data packets issued by the dispatch center. The transportation task data packets include the starting point coordinates, the ending point coordinates, the set of necessary nodes, cargo attribute parameters, and time constraint parameters.

[0016] Decode and transform the starting point coordinates, ending point coordinates, and the set of necessary nodes, convert the address code into latitude and longitude coordinates and unify them to the same coordinate reference system to generate a geographic location vector;

[0017] The vehicle-mounted Beidou terminal is activated to complete self-test and time synchronization, receive signals from at least four satellites and complete positioning calculation, obtain the vehicle's initial latitude and longitude, velocity parameters calculated based on Doppler frequency shift, and heading angle parameters calculated based on heading angle or fused by inertial measurement unit, and combine them to form the vehicle's initial state vector;

[0018] The geographical location vector, cargo attribute parameters, time constraint parameters, and vehicle initial state vector are combined to generate task initialization parameters.

[0019] Optionally, the step of calling the map service engine to generate a candidate path set, constructing a path cost function based on task initialization parameters, and generating the initial execution path specifically includes:

[0020] A set of candidate routes is generated by calling a map service engine, which includes a digital map database, a real-time traffic data interface, and a route search module.

[0021] The digital map database stores road topology, road node sets, road edge sets, road lengths, road speed limits, and road grade information. The real-time traffic data interface receives road congestion status, accident closure information, and traffic control rules.

[0022] The path search module constructs a directed graph on the road topology. The directed graph includes a set of road nodes and a set of road edges. The set of road nodes includes the starting point coordinates, the ending point coordinates, and the set of necessary nodes. The set of road edges includes the connecting edges between nodes and attribute parameters.

[0023] The path search module performs multi-path search on the directed graph based on the starting node and the ending node, maps the congestion level in real-time traffic data to the speed correction coefficient of the edge, performs connectivity verification and road level screening on the generated multiple paths, and obtains a set of candidate paths containing all necessary nodes.

[0024] Based on the cargo attribute parameters, time constraint parameters, and vehicle initial state vector in the task initialization parameters, a path cost function is constructed. The path cost function includes a distance term, an energy consumption term, a time penalty term, and a positioning deviation penalty term. Each term is weighted and summed according to a preset weight to obtain the cost of the candidate path.

[0025] In the path cost function, the distance term is defined as the sum of the lengths of all road segments on the path, and the energy consumption term is defined as the sum of the products of the length of each road segment on the path and the road segment energy consumption coefficient calculated by combining cargo attribute parameters and vehicle initial state vector.

[0026] The timeliness penalty term is defined as the difference between the estimated arrival time and the arrival time limit given by the timeliness constraint parameter when the estimated arrival time of the path exceeds the arrival time limit. When the estimated arrival time does not exceed the arrival time limit, the penalty value is set to zero. The positioning deviation penalty term is defined as the sum of the deviations obtained from the Beidou positioning error estimation of all road segments on the path.

[0027] The cost of each candidate path in the candidate path set is calculated using the path cost function to obtain the corresponding cost value, and all cost values ​​are summarized to form a cost value set.

[0028] The candidate path with the smallest cost value in the cost value set is selected as the initial execution path and sent to the vehicle navigation terminal. At the same time, the initial execution path is synchronized to the dispatch center to complete the path allocation.

[0029] Optionally, the step of continuously collecting data from the OBD, inertial measurement unit, and onboard camera during the vehicle's journey along the initial execution path, and combining this data with the real-time BeiDou location and road conditions to generate an operational state vector specifically includes:

[0030] During the vehicle's journey along the initial execution path, engine speed, instantaneous fuel consumption, tire pressure, and coolant temperature are collected via the onboard OBD interface to form a set of vehicle mechanical state parameters.

[0031] The longitudinal acceleration, lateral acceleration, vertical acceleration, and angular velocity are collected by the vehicle-mounted inertial measurement unit to form a set of kinematic state parameters;

[0032] By continuously collecting video frame streams during vehicle operation using an in-vehicle camera, image clarity parameters and inter-frame jitter parameters of the video frame stream are extracted to form a set of video monitoring parameters.

[0033] The vehicle's longitude, latitude, speed, and heading angle information are obtained in real time through the vehicle-mounted Beidou terminal. Longitude and latitude are used as the vehicle's spatial position parameters, speed is used as the vehicle's motion speed parameters, and heading angle is used as the vehicle's direction of travel parameters. The above parameters are combined to form a set of positioning status parameters.

[0034] The vehicle's mechanical state parameters, kinematic state parameters, video monitoring parameters, and positioning parameters are combined to construct an operational state vector.

[0035] Optionally, the step of threshold determination and time sequence identification of the running state vector to obtain abnormal event types and severity level candidates, and generating event features specifically includes:

[0036] Threshold judgment is performed on the vehicle mechanical state parameters in the operating state vector. When the engine speed exceeds the preset speed threshold, the instantaneous fuel consumption exceeds the preset fuel consumption threshold, the tire pressure is lower than the preset pressure threshold, or the coolant temperature exceeds the preset temperature threshold, the corresponding mechanical abnormality label is output.

[0037] Threshold determination is performed on the vehicle kinematic state parameters in the running state vector. When the absolute value of longitudinal acceleration, lateral acceleration or angular velocity exceeds the corresponding threshold, the corresponding kinematic anomaly label is output.

[0038] Threshold determination is performed on the video monitoring parameters in the running status vector. When the image clarity parameter is lower than the preset clarity threshold or the inter-frame jitter parameter exceeds the preset jitter threshold, the corresponding video anomaly label is output.

[0039] The system continuously collects vehicle operation status vectors within a fixed time window, forming an operation status sequence that includes the current time and historical time. After normalizing the operation status sequence, it extracts time features and compares the extracted time features with a preset abnormal pattern library one by one. When the comparison result reaches a preset similarity threshold, it outputs the corresponding abnormal event type and severity level candidate.

[0040] The abnormal labels obtained from threshold determination are fused with the abnormal event types and severity level candidates output by comparing with the running status sequence to generate event features.

[0041] Optionally, the step of inputting the running state vector and event features into the hierarchical reinforcement learning model, outputting the linkage level by the high-level policy network, and generating emergency action parameters by the low-level execution network under the constraint of the linkage level to obtain the emergency action plan specifically includes:

[0042] At each decision moment, the running state vector and event features are fed into the hierarchical reinforcement learning model as decision input pairs. The hierarchical reinforcement learning model includes a high-level policy network, multiple low-level execution networks, a high-level value function, a low-level action value function, and a termination function.

[0043] The high-level policy network first extracts features from the decision input pairs to generate comprehensive representation information reflecting the vehicle's operating status and abnormal situations;

[0044] Then, the comprehensive representation information is compared and evaluated with the pre-trained strategy parameters. The applicability scores of the four linkage levels—continue execution, dynamic detour, maintenance dispatch, and replacement vehicle shuttle—are calculated in the current scenario, and the linkage level with the highest applicability score is selected.

[0045] The linkage level serves as the emergency response level for the current task, and the pre-trained strategy parameters are a set of parameters formed during the training phase through continuous iteration and updates using historical operational data and simulation scenarios.

[0046] Invoke the lower-level execution network corresponding to the linkage level, output emergency action parameters, and form an emergency action plan;

[0047] Execute the emergency action plan, obtain execution feedback information and calculate the reward value, which is obtained by weighting the task delay time, operational risk measurement value, cold chain timeout loss, energy consumption and event cancellation indication.

[0048] The high-level value function and low-level action value function are updated based on the reward value, and a termination judgment function is used for judgment. Based on the judgment result, an emergency action plan is output.

[0049] Optionally, the invocation of the lower-level execution network corresponding to the linkage level, outputting emergency action parameters, and forming an emergency action plan specifically includes:

[0050] When the linkage level is set to continue execution, no adjustments are made to the vehicle's current transport route and operating speed. The emergency action parameters are set to empty, and an emergency action plan with the linkage level set to continue execution is generated directly.

[0051] When the linkage level is dynamic detour, the corresponding low-level execution network is invoked to output the target candidate path number and the speed band range of the path based on the current running state vector and event characteristics, thus forming an emergency action plan for dynamic detour.

[0052] When the linkage level is maintenance dispatch, the corresponding low-level execution network outputs the target maintenance point and the expected arrival time window based on the running state vector and event characteristics. The selected maintenance point is taken as the service location that the vehicle needs to go to, and the start time and end time of the time window are taken as the time constraints that the vehicle must arrive at the maintenance point. The above maintenance point and time window are combined to form emergency action parameters, thereby generating an emergency action plan for maintenance dispatch.

[0053] When the linkage level is to replace vehicle shuttle, the corresponding low-level execution network is invoked to generate a shuttle plan. The shuttle plan includes selecting a shuttle node and determining a backup vehicle. The shuttle node is a node in the road node set, and the backup vehicle is a vehicle in the backup vehicle set.

[0054] The connection node and the backup vehicle number are combined to form emergency action parameters, and an emergency action plan for replacing the vehicle connection is generated accordingly.

[0055] Optionally, the step of updating the high-level value function and the low-level action value function based on the reward value, and using a termination judgment function to make a judgment, and outputting an emergency action plan based on the judgment result, specifically includes:

[0056] After obtaining the reward value, the improvement amount of the high-level value function is calculated using the time-series difference method, and the high-level value function corresponding to the current state is corrected based on the improvement amount of the high-level value function so that it can more accurately reflect the long-term cumulative reward.

[0057] The high-level value function represents the expected cumulative reward that can be obtained over a long period of time under the vehicle operation state vector and event characteristics;

[0058] The current state refers to the vehicle's operating state after executing the emergency action plan, including the operating state vector, event characteristics, and execution results fed back by the action execution process;

[0059] After obtaining the reward value, the improvement amount of the low-level action value function corresponding to the linkage level is calculated using the temporal difference method. Based on the improvement amount of the low-level action value function, the low-level action value function corresponding to the current state and emergency action is corrected so that it can reflect the merits and demerits of action selection in the current emergency scenario.

[0060] The low-level action value function represents the expected cumulative reward that can be obtained under the linkage level for specific emergency action parameters under the vehicle operation state vector and event characteristics.

[0061] After the value function is updated, the high-level policy network and the low-level execution network corresponding to the linkage level are updated based on the reward value. The update process uses the policy gradient method to adjust the network parameters so that the output linkage level and emergency action parameters tend to be optimal under long-term cumulative rewards.

[0062] After updating the value function and policy network, the state after execution is determined using a termination decision function. When the determination result is completion, the final result of the emergency action plan is output. When the determination result is incomplete, new running state vectors and event characteristics are received.

[0063] Optionally, the scheduling and execution based on the emergency action plan to generate emergency execution result information specifically includes:

[0064] When the linkage level is dynamic detour, the path planning engine is invoked to select a new optimal path from the candidate path set, the optimal path is used as the updated initial execution path, and the updated path is sent to the vehicle navigation terminal. At the same time, the update result is recorded synchronously in the dispatch center.

[0065] When the linkage level is maintenance dispatch, a maintenance task order is generated based on the maintenance point and arrival time window in the emergency action parameters. The maintenance task order is pushed to the maintenance end, the transportation plan is adjusted and the estimated arrival time is recalculated.

[0066] When the linkage level is to replace the vehicle for shuttle, a shuttle instruction is generated based on the shuttle node and the standby vehicle number in the emergency action parameters. The shuttle instruction is then sent to the standby vehicle, a shuttle task is established, and the task allocation information is updated synchronously in the dispatch center.

[0067] Updated paths, maintenance task orders, or connection instructions are treated as emergency execution results and uniformly written into the execution record set, and an execution status identifier is generated.

[0068] Optionally, the step of visualizing and chain-basedly storing the operational status vector, event characteristics, emergency action plans, and emergency execution results, generating a transportation report, and transmitting it back to the supply chain management platform specifically includes:

[0069] The system uses operational status vectors, event characteristics, emergency action plans, and emergency execution results as input data for visualization, generating a visualization interface that includes vehicle driving trajectory curves, timelines of abnormal event occurrence and handling, execution records of each linked action, and corresponding execution status indicators.

[0070] A chain-like evidence storage structure is established for the visual input data. A hash function is used to calculate the hash value of each data record and generate a block header. The hash value of the previous block header and the hash value of the current data record are used to form a blockchain evidence storage chain, ensuring the immutability and traceability of the visual input data.

[0071] After the vehicle completes delivery and the receiving party completes electronic signature, a transportation report is generated and transmitted back to the supply chain management platform via a communication interface.

[0072] The transportation report includes route execution status, anomaly handling process, energy consumption indicators, and timeliness indicators. The energy consumption indicators are calculated based on the energy consumption term in the route cost function, and the timeliness indicators are calculated based on the timeliness penalty term in the route cost function.

[0073] The beneficial effects of this invention are:

[0074] This invention proposes a path cost function in path planning, which integrates transportation distance, vehicle energy consumption, task timeliness and BeiDou positioning deviation into a unified model. This enables path selection to consider both transportation conditions and environmental impacts, solving the problem that traditional methods rely solely on distance and time, leading to discrepancies between the results and actual needs. This improves the reliability of paths in complex scenarios.

[0075] In anomaly identification, this invention constructs an operational state vector from sensor and BeiDou terminal data, and generates event features by combining threshold judgment and time series analysis. This enables the identification of mechanical and kinematic anomalies, and reduces false alarms and missed alarms through trend analysis, thereby improving the accuracy of operational monitoring.

[0076] In terms of emergency response, this invention proposes a multi-level linkage mechanism for a hierarchical reinforcement learning model. The high-level policy network is responsible for selecting the emergency level based on event characteristics, while the low-level execution network outputs specific parameters, such as route number, maintenance point, and connection node, based on the high-level decision. By incorporating cold chain loss, energy consumption, time delay, and risk level into the reward function, this invention enables the model to achieve dynamic trade-offs and adaptive optimization in complex transportation environments. This mechanism not only improves the automation level of emergency response but also ensures the continuity and reliability of transportation tasks in emergency situations, exhibiting a higher level of intelligence and scalability compared to emergency mechanisms relying on manual rules.

[0077] In data management, this invention visualizes the running status vector, event characteristics, emergency plans and execution results, and uses chain-based evidence storage to ensure the integrity of data throughout the process. After the task is completed, a transportation report is generated that includes path execution status, anomaly handling, energy consumption and timeliness indicators, providing support for the transparency and traceability of the supply chain. Attached Figure Description

[0078] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0079] Figure 1 The flowchart shows the logistics and supply chain management method based on BeiDou positioning and path planning algorithm proposed in this invention.

[0080] Figure 2 This is a schematic diagram illustrating the construction of the path cost function in the logistics and supply chain management method based on BeiDou positioning and path planning algorithm proposed in this invention.

[0081] Figure 3 This is a schematic diagram of the multi-level linkage decision structure based on hierarchical reinforcement learning in the logistics and supply chain management method based on Beidou positioning and path planning algorithm proposed in this invention. Detailed Implementation

[0082] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0083] refer to Figure 1-3 The logistics and supply chain management method based on BeiDou positioning and path planning algorithms includes the following steps:

[0084] Receive the transportation task issued by the dispatch center, start the vehicle-mounted Beidou terminal to obtain the vehicle's initial latitude, longitude, speed and heading information, and generate task initialization parameters;

[0085] The map service engine is invoked to generate a set of candidate paths, and a path cost function is constructed based on the task initialization parameters to generate the initial execution path;

[0086] During the vehicle's journey along the initial execution path, data from OBD, inertial measurement unit, and onboard camera are continuously collected. Combined with BeiDou real-time location and road conditions, an operational status vector is generated.

[0087] Threshold determination and time sequence identification are performed on the running status vector to obtain abnormal event types and severity level candidates, and event features are generated;

[0088] The running state vector and event features are input into a hierarchical reinforcement learning model. The high-level policy network outputs the linkage level, and the low-level execution network generates emergency action parameters to obtain an emergency action plan.

[0089] Based on the emergency response plan, dispatch and execute the plan, and generate emergency execution result information;

[0090] The system visualizes and stores information on operational status vectors, event characteristics, emergency action plans, and emergency execution results in a chain, generates transportation reports, and sends them back to the supply chain management platform.

[0091] In this embodiment, receiving the transportation task issued by the dispatch center, activating the vehicle-mounted Beidou terminal to obtain the vehicle's initial latitude, longitude, speed, and heading information, and generating task initialization parameters specifically include:

[0092] Receive transportation task data packets issued by the dispatch center. The transportation task data packets include the starting point coordinates, the ending point coordinates, the set of necessary nodes, cargo attribute parameters, and time constraint parameters.

[0093] Decode and transform the starting point coordinates, ending point coordinates, and the set of necessary nodes, convert the address code into latitude and longitude coordinates and unify them to the same coordinate reference system to generate a geographic location vector;

[0094] The vehicle-mounted Beidou terminal is activated to complete self-test and time synchronization, receive signals from at least four satellites and complete positioning calculation, obtain the vehicle's initial latitude and longitude, velocity parameters calculated based on Doppler frequency shift, and heading angle parameters calculated based on heading angle or fused by inertial measurement unit, and combine them to form the vehicle's initial state vector;

[0095] The geographical location vector, cargo attribute parameters, time constraint parameters, and vehicle initial state vector are combined to generate task initialization parameters.

[0096] In this embodiment, the step of calling the map service engine to generate a candidate path set, constructing a path cost function based on task initialization parameters, and generating the initial execution path specifically includes:

[0097] A set of candidate routes is generated by calling a map service engine, which includes a digital map database, a real-time traffic data interface, and a route search module.

[0098] The digital map database stores road topology, road node sets, road edge sets, road lengths, road speed limits, and road grade information. The real-time traffic data interface receives road congestion status, accident closure information, and traffic control rules.

[0099] The path search module constructs a directed graph on the road topology. The directed graph includes a set of road nodes and a set of road edges. The set of road nodes includes the starting point coordinates, the ending point coordinates, and the set of necessary nodes. The set of road edges includes the connecting edges between nodes and attribute parameters.

[0100] The path search module performs multi-path search on the directed graph based on the starting node and the ending node, maps the congestion level in real-time traffic data to the speed correction coefficient of the edge, performs connectivity verification and road level screening on the generated multiple paths, and obtains a set of candidate paths containing all necessary nodes.

[0101] Based on the cargo attribute parameters, time constraint parameters, and vehicle initial state vector in the task initialization parameters, a path cost function is constructed. The path cost function includes a distance term, an energy consumption term, a time penalty term, and a positioning deviation penalty term. Each term is weighted and summed according to a preset weight to obtain the cost of the candidate path.

[0102] In the path cost function, the distance term is defined as the sum of the lengths of all road segments on the path, and the energy consumption term is defined as the sum of the products of the length of each road segment on the path and the road segment energy consumption coefficient calculated by combining cargo attribute parameters and vehicle initial state vector.

[0103] The timeliness penalty term is defined as the difference between the estimated arrival time and the arrival time limit given by the timeliness constraint parameter when the estimated arrival time of the path exceeds the arrival time limit. When the estimated arrival time does not exceed the arrival time limit, the penalty value is set to zero. The positioning deviation penalty term is defined as the sum of the deviations obtained from the Beidou positioning error estimation of all road segments on the path.

[0104] The cost of each candidate path in the candidate path set is calculated using the path cost function to obtain the corresponding cost value, and all cost values ​​are summarized to form a cost value set.

[0105] The candidate path with the smallest cost value in the cost value set is selected as the initial execution path and sent to the vehicle navigation terminal. At the same time, the initial execution path is synchronized to the dispatch center to complete the path allocation.

[0106] In this embodiment, the step of continuously collecting data from the OBD, inertial measurement unit, and vehicle camera during the vehicle's journey along the initial execution path, and combining this data with the real-time BeiDou location and road conditions to generate an operational state vector specifically includes:

[0107] During the vehicle's journey along the initial execution path, engine speed, instantaneous fuel consumption, tire pressure, and coolant temperature are collected via the onboard OBD interface to form a set of vehicle mechanical state parameters.

[0108] The longitudinal acceleration, lateral acceleration, vertical acceleration, and angular velocity are collected by the vehicle-mounted inertial measurement unit to form a set of kinematic state parameters;

[0109] By continuously acquiring video frame streams during vehicle operation using an in-vehicle camera, image clarity parameters and inter-frame jitter parameters are extracted from the video frame streams. The image clarity parameters reflect the identifiability of the acquired images, and the inter-frame jitter parameters reflect the stability during the video acquisition process, thereby forming a set of video monitoring parameters.

[0110] The vehicle's longitude, latitude, speed, and heading angle information are obtained in real time through the vehicle-mounted Beidou terminal. Longitude and latitude are used as the vehicle's spatial position parameters, speed is used as the vehicle's motion speed parameters, and heading angle is used as the vehicle's direction of travel parameters. The above parameters are combined to form a set of positioning status parameters.

[0111] The vehicle's mechanical state parameters, kinematic state parameters, video monitoring parameters, and positioning parameters are combined to construct an operational state vector.

[0112] In this embodiment, the step of threshold determination and time sequence identification of the running state vector to obtain abnormal event types and severity level candidates, and generating event features specifically includes:

[0113] Threshold judgment is performed on the vehicle mechanical state parameters in the operating state vector. When the engine speed exceeds the preset speed threshold, the instantaneous fuel consumption exceeds the preset fuel consumption threshold, the tire pressure is lower than the preset pressure threshold, or the coolant temperature exceeds the preset temperature threshold, the corresponding mechanical abnormality label is output.

[0114] Threshold determination is performed on the vehicle kinematic state parameters in the running state vector. When the absolute value of longitudinal acceleration, lateral acceleration or angular velocity exceeds the corresponding threshold, the corresponding kinematic anomaly label is output.

[0115] Threshold determination is performed on the video monitoring parameters in the running status vector. When the image clarity parameter is lower than the preset clarity threshold or the inter-frame jitter parameter exceeds the preset jitter threshold, the corresponding video anomaly label is output.

[0116] The system continuously collects vehicle operation status vectors within a fixed time window, forming an operation status sequence that includes the current time and historical time. After normalizing the operation status sequence, it extracts time features and compares the extracted time features with a preset abnormal pattern library one by one. When the comparison result reaches a preset similarity threshold, it outputs the corresponding abnormal event type and severity level candidate.

[0117] The abnormal labels obtained from threshold determination are fused with the abnormal event types and severity level candidates output by comparing with the running status sequence to generate event features.

[0118] In this embodiment, the step of inputting the running state vector and event features into the hierarchical reinforcement learning model, outputting the linkage level by the high-level policy network, and generating emergency action parameters by the low-level execution network under the constraint of the linkage level to obtain the emergency action plan specifically includes:

[0119] At each decision moment, the running state vector and event features are fed into the hierarchical reinforcement learning model as decision input pairs. The hierarchical reinforcement learning model includes a high-level policy network, multiple low-level execution networks, a high-level value function, a low-level action value function, and a termination function.

[0120] The high-level policy network first extracts features from the decision input pairs to generate comprehensive representation information reflecting the vehicle's operating status and abnormal situations;

[0121] Then, the comprehensive representation information is compared and evaluated with the pre-trained strategy parameters. The applicability scores of the four linkage levels—continue execution, dynamic detour, maintenance dispatch, and replacement vehicle shuttle—are calculated in the current scenario, and the linkage level with the highest applicability score is selected.

[0122] The linkage level serves as the emergency response level for the current task, and the pre-trained strategy parameters are a set of parameters formed during the training phase through continuous iteration and updates using historical operational data and simulation scenarios.

[0123] Invoke the lower-level execution network corresponding to the linkage level, output emergency action parameters, and form an emergency action plan;

[0124] Execute the emergency response plan, obtain execution feedback information, and calculate the reward value. The reward value is obtained by weighting the task delay time, operational risk metric, cold chain timeout loss, energy consumption, and event resolution indication.

[0125] ;

[0126] in, For a moment The reward value, This is a weighting factor for delay time, used to adjust the degree of influence of task delays on reward value calculation. For normalized time The time delay incurred after implementing the emergency response plan, dimensionless. This is a weighting coefficient for operational risk, used to adjust the degree of influence of operational risk in the calculation of reward value. For normalized time The operational risk metric generated after implementing the emergency response plan is dimensionless. This is a weighting coefficient for cold chain timeout losses, used to adjust the impact of cold chain cargo losses on the calculation of reward values. For normalized time The quantity of losses incurred due to cold chain goods exceeding the time limit after the implementation of the emergency response plan, dimensionless. This is a weighting coefficient for energy consumption, used to adjust the degree of influence of vehicle energy consumption in the calculation of reward values. For normalized time Energy consumption of vehicles after implementing emergency response plans (dimensionless). This is a weighting coefficient for successful event resolution, used to adjust the positive contribution of successful event resolution to the reward calculation. For normalized time A dimensionless indicator of whether the abnormal event was successfully resolved after the emergency response plan was executed.

[0127] The high-level value function and low-level action value function are updated based on the reward value, and a termination judgment function is used for judgment. Based on the judgment result, an emergency action plan is output.

[0128] In this embodiment, the step of calling the lower-level execution network corresponding to the linkage level, outputting emergency action parameters, and forming an emergency action plan specifically includes:

[0129] When the linkage level is set to continue execution, no adjustments are made to the vehicle's current transport route and operating speed. The emergency action parameters are set to empty, and an emergency action plan with the linkage level set to continue execution is generated directly.

[0130] When the linkage level is dynamic detour, the corresponding low-level execution network is invoked to output the target candidate path number and the speed band range of the path based on the current running state vector and event characteristics. The target candidate path number is used to select a specific path from the candidate path set, and the speed band range is used to limit the lower and upper limits of the vehicle speed on the path. The target candidate path number and speed band range are used together as emergency action parameters to form a dynamic detour emergency action plan.

[0131] When the linkage level is maintenance dispatch, the corresponding low-level execution network outputs the target maintenance point and the expected arrival time window based on the running state vector and event characteristics. The selected maintenance point is taken as the service location that the vehicle needs to go to, and the start time and end time of the time window are taken as the time constraints that the vehicle must arrive at the maintenance point. The above maintenance point and time window are combined to form emergency action parameters, thereby generating an emergency action plan for maintenance dispatch.

[0132] When the linkage level is to replace vehicle shuttle, the corresponding low-level execution network is invoked to generate a shuttle plan. The shuttle plan includes selecting a shuttle node and determining a backup vehicle. The shuttle node is a node in the road node set, and the backup vehicle is a vehicle in the backup vehicle set.

[0133] The connection node and the backup vehicle number are combined to form emergency action parameters, and an emergency action plan for replacing the vehicle connection is generated accordingly.

[0134] In this embodiment, the step of updating the high-level value function and the low-level action value function based on the reward value, and using a termination judgment function to make a judgment, and outputting an emergency action plan based on the judgment result, specifically includes:

[0135] After obtaining the reward value, the improvement amount of the high-level value function is calculated using the time-series difference method, and the high-level value function corresponding to the current state is corrected based on the improvement amount of the high-level value function so that it can more accurately reflect the long-term cumulative reward.

[0136] The high-level value function represents the expected cumulative reward that can be obtained over a long period of time under the vehicle operation state vector and event characteristics;

[0137] The current state refers to the vehicle's operating state after executing the emergency action plan, including the operating state vector, event characteristics, and execution results fed back by the action execution process;

[0138] After obtaining the reward value, the improvement amount of the low-level action value function corresponding to the linkage level is calculated using the temporal difference method. Based on the improvement amount of the low-level action value function, the low-level action value function corresponding to the current state and emergency action is corrected so that it can reflect the merits and demerits of action selection in the current emergency scenario.

[0139] The low-level action value function represents the expected cumulative reward that can be obtained under the linkage level for specific emergency action parameters under the vehicle operation state vector and event characteristics.

[0140] After the value function is updated, the high-level policy network and the low-level execution network corresponding to the linkage level are updated based on the reward value. The update process uses the policy gradient method to adjust the network parameters so that the output linkage level and emergency action parameters tend to be optimal under long-term cumulative rewards.

[0141] After updating the value function and policy network, the state after execution is determined using a termination decision function. When the determination result is completion, the final result of the emergency action plan is output. When the determination result is incomplete, new running state vectors and event features are received to carry out subsequent decision iterations.

[0142] In this embodiment, the scheduling and execution based on the emergency action plan to generate emergency execution result information specifically includes:

[0143] When the linkage level is dynamic detour, the path planning engine is invoked to select a new optimal path from the candidate path set, the optimal path is used as the updated initial execution path, and the updated path is sent to the vehicle navigation terminal. At the same time, the update result is recorded synchronously in the dispatch center.

[0144] When the linkage level is maintenance dispatch, a maintenance task order is generated based on the maintenance point and arrival time window in the emergency action parameters. The maintenance task order is pushed to the maintenance end, the transportation plan is adjusted and the estimated arrival time is recalculated.

[0145] When the linkage level is to replace the vehicle for shuttle, a shuttle instruction is generated based on the shuttle node and the standby vehicle number in the emergency action parameters. The shuttle instruction is then sent to the standby vehicle, a shuttle task is established, and the task allocation information is updated synchronously in the dispatch center.

[0146] Updated paths, maintenance task orders, or connection instructions are treated as emergency execution results and uniformly written into the execution record set, and an execution status identifier is generated.

[0147] In this embodiment, the visualization and chain-based storage of operational status vectors, event characteristics, emergency action plans, and emergency execution results, generating a transportation report and transmitting it back to the supply chain management platform specifically includes:

[0148] The system uses operational status vectors, event characteristics, emergency action plans, and emergency execution results as input data for visualization, generating a visualization interface that includes vehicle driving trajectory curves, timelines of abnormal event occurrence and handling, execution records of each linked action, and corresponding execution status indicators.

[0149] A chain-like evidence storage structure is established for the visual input data. A hash function is used to calculate the hash value of each data record and generate a block header. The hash value of the previous block header and the hash value of the current data record are used to form a blockchain evidence storage chain, ensuring the immutability and traceability of the visual input data.

[0150] After the vehicle completes delivery and the receiving party completes electronic signature, a transportation report is generated and transmitted back to the supply chain management platform via a communication interface.

[0151] The transportation report includes route execution status, anomaly handling process, energy consumption indicators, and timeliness indicators. The energy consumption indicators are calculated based on the energy consumption term in the route cost function, and the timeliness indicators are calculated based on the timeliness penalty term in the route cost function.

[0152] Example 1:

[0153] To verify the feasibility of the present invention in implementation, the present invention is applied to the urban distribution link of a cold chain logistics enterprise. This enterprise undertakes the warehousing and transportation tasks of daily fresh food. During the transportation process, it is not only necessary to ensure the quality of the goods under low-temperature conditions, but also to complete on-time delivery in urban areas with complex traffic and easily interfered positioning signals. The traditional scheduling method only relies on the navigation system to generate the shortest path and processes vehicle anomalies through manual rules. Once traffic congestion, equipment failure or positioning drift occurs, it often leads to delivery delays, a high cold chain breakage rate, and a decline in customer satisfaction.

[0154] During the application of the present invention, first, the dispatching center issues a transportation task, which includes the starting warehouse, multiple distribution terminals, time limit requirements, and goods attributes. After the vehicle starts, the on-vehicle Beidou terminal real-time collects the initial position, speed, and heading information, and generates multiple candidate paths in combination with the map service engine. The system constructs a path cost function based on the task initialization parameters, takes the transportation distance, vehicle energy consumption, time limit requirements, and positioning deviation as constraint conditions, calculates the comprehensive cost of each candidate path, and selects the optimal path. This path is sent to the on-vehicle navigation terminal and synchronized with the dispatching center.

[0155] During the vehicle driving process, the OBD records the engine speed, fuel consumption, and tire pressure, the inertial measurement unit records the acceleration and angular velocity, the on-vehicle camera detects lane deviation, and the Beidou terminal continuously provides the position coordinates. The above data is uniformly converted into an operating state vector. The system identifies engine overheating and emergency braking anomalies through threshold judgment, and at the same time combines time series identification methods to conduct trend analysis on the slow decline of tire pressure and continuous position drift, outputs the type and severity level of abnormal events, and generates event characteristics. During one distribution, the system detected a downward trend in the tire pressure of a certain vehicle and determined it as an abnormal event of medium level.

[0156] After the abnormal event is triggered, the hierarchical reinforcement learning model receives the input of the operating state vector and event characteristics. The high-level policy network determines that the maintenance dispatch policy needs to be executed, and the low-level execution network generates emergency action parameters, outputs the recommended repair point and the arrival time window. The system schedules the vehicle to drive into the nearest repair service station and calculates the alternative vehicle connection path to ensure the uninterrupted cold chain transportation. In the reward function, the system simultaneously considers cold chain losses, delay time, energy consumption changes, and risk levels, and finally automatically selects a plan to complete tire replacement within 15 minutes and continue distribution by the alternative vehicle. This plan shortens the time by an average of 27 minutes compared with the traditional manual processing method, and the temperature deviation of the cold chain goods is controlled within 0.3°C.

[0157] After transportation is completed, the system visualizes the operational status vector, abnormal events, emergency response plans, and execution results, and writes them to the block through chain-like evidence storage to generate an immutable execution log. After electronic signature is completed at the delivery terminal, the system automatically generates a transportation report. The report shows that the actual transportation time for this task was 3 hours and 28 minutes, 8 minutes later than the planned time, but the cold chain of the goods remained intact, average energy consumption decreased by 6.7%, and the anomaly handling completion rate reached 100%. These results demonstrate that the present invention can effectively ensure the continuity and transparency of transportation tasks in complex environments.

[0158] Table 1 Comparison of Transportation Task Execution

[0159]

[0160] As can be seen from Table 1 above, the present invention significantly improves upon traditional methods in several key performance indicators of logistics transportation tasks. Firstly, regarding transportation time, the average transportation time under the traditional route planning and anomaly handling model is 3 hours and 55 minutes. The present invention, by optimizing route selection through a comprehensive cost function and combining it with a hierarchical reinforcement learning model for emergency decision-making, reduces the average transportation time to 3 hours and 28 minutes, improving overall efficiency by nearly 12%. Secondly, regarding cold chain temperature maintenance, the temperature deviation of traditional methods reaches ±1.2℃, while the present invention effectively controls temperature fluctuations throughout the transportation process, reducing the average deviation to ±0.3℃, improving the stability of the cold chain environment by approximately 75%, and effectively ensuring the quality of perishable goods.

[0161] Secondly, regarding energy consumption, the traditional method has an average energy consumption of 12.8 liters per 100 kilometers. This invention, by combining cargo attributes and vehicle status to optimize energy consumption, reduces unnecessary detours and idling in route planning and anomaly handling, lowering energy consumption to 11.9 liters per 100 kilometers, representing an energy saving of 6.7%. In terms of anomaly handling efficiency, traditional manual or rule-driven anomaly handling takes an average of 45 minutes. This invention, through real-time state vector analysis and intelligent linkage mechanisms, shortens the handling time to 18 minutes, a reduction of 60%, significantly improving handling efficiency.

[0162] Finally, regarding task stability, the success rate of anomaly handling under traditional methods is 82%, while this invention achieves a 100% completion rate, improving anomaly handling capabilities by approximately 22%. Regarding the proportion of delayed orders, 14% of orders were delayed under the traditional model, while after implementing this invention, the delay rate decreased to 3%, an improvement of 11%. These results demonstrate that this invention can effectively improve the efficiency, energy consumption, and cold chain security of logistics transportation in complex environments, while enhancing anomaly handling and task stability, providing higher reliability and transparency for supply chain management.

[0163] Table 2. Results of Abnormal Event Identification and Handling

[0164]

[0165] As can be seen from Table 2 above, the present invention significantly improves the accuracy and efficiency of abnormal event identification compared to traditional methods. Regarding identification accuracy, the traditional method achieves average identification rates of 85%, 70%, 88%, and 65% for tire blowouts, tire pressure drops, emergency braking, and position drift, respectively. In contrast, the present invention achieves identification rates of 97%, 94%, 96%, and 91% for the same scenarios, representing an overall accuracy improvement of 15% to 26%. This indicates that the combination of operational state vector analysis with threshold determination and time series analysis can effectively capture continuous characteristics, reducing false alarms and missed alarms.

[0166] Regarding handling time, traditional methods often rely on manual rules and offline coordination, with an average processing time between 30 and 55 minutes. This invention, however, utilizes a hierarchical reinforcement learning model to achieve multi-level linkage, directly generating detour, repair, or connection solutions after system judgment, reducing handling time to 10 to 25 minutes. Overall, this invention reduces time by approximately 62% in tire blowout handling, approximately 60% in tire pressure drop handling, and approximately 55% in position drift handling. These results demonstrate that this invention outperforms traditional methods in both abnormal event identification and emergency response, effectively improving the safety and continuity of transportation tasks.

[0167] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A logistics and supply chain management method based on Beidou positioning and path planning algorithm, characterized in that, The method comprises the following steps: receiving a transportation task issued by a dispatch center, starting a vehicle-mounted Beidou terminal to obtain initial latitude and longitude, speed and heading information of the vehicle, and generating task initialization parameters; calling a map service engine to generate a candidate path set, constructing a path cost function based on the task initialization parameters, and generating an initial execution path; during the driving of the vehicle according to the initial execution path, continuously collecting data of OBD, an inertial measurement unit and a vehicle-mounted camera, combining the real-time position of Beidou and the road state to generate a running state vector; performing threshold judgment and time sequence recognition on the running state vector to obtain abnormal event types and severity level candidates, and generating event features; inputting the running state vector and the event features into a hierarchical reinforcement learning model, outputting a linkage level from a high-level policy network, generating emergency action parameters from a low-level execution network, and obtaining an emergency action scheme; based on the emergency action scheme, performing scheduling and execution to form emergency execution result information; visualizing and chain-attesting the running state vector, the event features, the emergency action scheme and the emergency execution result information, generating a transportation report and returning it to a supply chain management platform; the specific process of inputting the running state vector and the event features into the hierarchical reinforcement learning model, outputting the linkage level from the high-level policy network, and generating the emergency action parameters from the low-level execution network to obtain the emergency action scheme comprises: at each decision-making time, inputting the running state vector and the event features as a decision input pair into the hierarchical reinforcement learning model, the hierarchical reinforcement learning model comprising a high-level policy network, a plurality of low-level execution networks, a high-level value function, a low-level action value function and a termination function; the high-level policy network performs feature extraction on the decision input pair to generate comprehensive feature information, compares and evaluates the comprehensive feature information with pre-trained policy parameters, sequentially calculates the applicability scores of four linkage levels, i.e. continuous execution, dynamic detour, maintenance dispatch and replacement vehicle connection, in the current scene, and outputs the linkage level with the highest applicability score; calling the low-level execution network corresponding to the linkage level to output the emergency action parameters and form the emergency action scheme; executing the emergency action scheme, obtaining execution feedback information and calculating a reward value, the reward value being obtained by weighting the task delay time, the running risk metric value, the cold chain overtime loss, the energy consumption and the event release indication; updating the high-level value function and the low-level action value function based on the reward value, and using the termination judgment function to judge, and outputting the emergency action scheme according to the judgment result.

2. The method according to claim 1, wherein, the task initialization parameters comprise a geographic position vector, a cargo attribute parameter, a time constraint parameter and a vehicle initial state vector. 3.The method of claim 1, wherein, the specific process of calling the map service engine to generate the candidate path set, constructing the path cost function based on the task initialization parameters, and generating the initial execution path comprises: calling the map service engine to generate the candidate path set, the map service engine comprising a digital map database, a real-time traffic data interface and a path search module; the path search module constructs a directed graph on a road topology structure, the directed graph comprising a road node set and a road edge set; The path searching module performs multi-path searching based on the start node and the end node on the directed graph to obtain a candidate path set; A path cost function is constructed based on the cargo attribute parameters, time limit constraint parameters and vehicle initial state vector in the task initialization parameters, the path cost function including a distance item, an energy consumption item, a time limit penalty item and a positioning deviation penalty item, each item being weighted and summed according to a preset weight to obtain a value of the candidate path; The path cost function is used to calculate the cost of each candidate path in the candidate path set to obtain a corresponding value, and all the values are summarized to form a value set; The candidate path corresponding to the minimum value in the value set is selected as an initial execution path. 4.The method of claim 1, wherein, The running state vector includes vehicle mechanical state parameters, kinematic state parameters, video monitoring parameters and positioning state parameters. 5.The method of claim 1, wherein, The threshold determination and time sequence identification of the running state vector are performed to obtain abnormal event types and severity level candidates, and event features are generated, which specifically include: The vehicle mechanical state parameters in the running state vector are subjected to threshold determination, and when the engine speed exceeds a preset speed threshold, the instantaneous fuel consumption exceeds a preset fuel consumption threshold, the tire air pressure is lower than a preset air pressure threshold or the coolant temperature exceeds a preset temperature threshold, a mechanical abnormality label is output; The vehicle kinematic state parameters in the running state vector are subjected to threshold determination, and when the absolute values of longitudinal acceleration, transverse acceleration or angular velocity exceed corresponding thresholds, a kinematic abnormality label is output; The video monitoring parameters in the running state vector are subjected to threshold determination, and when the image definition parameter is lower than a preset definition threshold or the interframe jitter parameter exceeds a preset jitter threshold, a video abnormality label is output; The running state vector of the vehicle is continuously collected within a fixed time window to form a running state sequence, and the running state sequence is compared to output abnormal event types and severity level candidates; The abnormality labels obtained by threshold determination and the abnormal event types and severity level candidates output by comparison of the running state sequence are fused to generate event features. 6.The method of claim 1, wherein, The low-level execution network corresponding to the linkage level is called to output emergency action parameters to form an emergency action scheme, which specifically includes: When the linkage level is continue execution, no adjustment is made to the current transportation path and running speed of the vehicle, the emergency action parameters are set to null, and a continue execution emergency action scheme is generated; When the linkage level is dynamic detour, the corresponding low-level execution network is called, the target candidate path number and speed band range are taken as the emergency action parameters, and a dynamic detour emergency action scheme is generated; When the linkage level is repair dispatching, the corresponding low-level execution network is called, the repair point and time window are combined to form the emergency action parameters, and a repair dispatching emergency action scheme is generated; When the linkage level is replacement vehicle connection, the corresponding low-level execution network is called, the connection node and backup vehicle number are combined to form the emergency action parameters, and a replacement vehicle connection emergency action scheme is generated. 7.The method of claim 1, wherein, The high-level value function and the low-level action value function are updated based on the reward value, and a termination determination function is used for determination, and an emergency action scheme is output according to the determination result, which specifically includes: Based on the reward value, the improvement amount of the high-level value function is calculated by using the time difference method, and the high-level value function corresponding to the current state is corrected based on the improvement amount of the high-level value function; Based on the reward value, the improvement amount of the low-level action value function corresponding to the linkage level is calculated by using the time difference method, and the low-level action value function corresponding to the current state and the emergency action is corrected based on the improvement amount of the low-level action value function; After completing the value function update, the high-level policy network and the low-level execution network corresponding to the linkage level are updated based on the reward value; After completing the value function and policy network update, the termination determination function is used for determination, and when the determination result is complete, the final result of the emergency action scheme is output, and when the determination result is not complete, a new running state vector and event feature are continuously received. 8.The method of claim 1, wherein, The scheduling execution based on the emergency action scheme forms emergency execution result information, which specifically includes: When the linkage level is dynamic bypass, the path planning engine is called to select a new optimal path from the candidate path set, and the optimal path is used as the updated initial execution path; When the linkage level is maintenance dispatching, a maintenance task list is generated according to the maintenance point and arrival time window in the emergency action parameters; When the linkage level is replacement vehicle connection, a connection instruction is generated according to the connection node and backup vehicle number in the emergency action parameters; The updated path, maintenance task list or connection instruction is written into the execution record set as the emergency execution result information, and an execution state identifier is generated. 9.The method of claim 1, wherein, The visual display and chain storage of the running state vector, event feature, emergency action scheme and emergency execution result information, the generation of the transportation report and the return to the supply chain management platform specifically include: The running state vector, event feature, emergency action scheme and emergency execution result information are used as input data for visual display to generate a visual interface; A chain storage structure is established for the visual input data, a hash value of each data record is calculated by using a hash function to generate a block header, and a block chain storage link list is formed by using the hash value of the previous block header and the hash value of the current data record; After the vehicle completes the distribution and the electronic receipt is completed by the receiving end, a transportation report is generated, and the transportation report is returned to the supply chain management platform through the communication interface.

Images