Intelligent logistics park unmanned vehicle cluster cooperative distribution method and system

Abstract

The application discloses a smart logistics park unmanned vehicle cluster cooperative distribution method and system, relates to the related technical field of logistics distribution cooperative control, and comprises the following steps: based on real-time logistics monitoring data of the smart logistics park, a first predicted emergency is obtained by using an emergency prediction model; an initial cooperative distribution strategy group is generated in combination with target distribution orders and vehicle cluster state information; a flexible confrontation index is introduced to evaluate the initial cooperative distribution strategy group, and the first cooperative distribution strategy group is output by optimizing the difference between the flexible confrontation index and a preset flexible confrontation index. The technical problems that the path planning in the prior art has poor adaptability to emergencies and unpredictable events, response is slow, the overall distribution efficiency of the unmanned vehicle cluster is reduced, and operation reliability is insufficient are solved, the anti-interference ability and adaptability of the cooperative distribution strategy of the unmanned vehicle cluster are dynamically enhanced through predictive evaluation and flexible optimization, and the overall distribution efficiency and operation reliability are improved.

Description

Technical Field

[0001] This invention relates to the technical field of collaborative control in logistics distribution, specifically to a method and system for collaborative distribution of unmanned vehicle clusters in smart logistics parks. Background Technology

[0002] As a core node in the modern logistics system, the operational efficiency and service quality of smart logistics parks directly impact the effectiveness of the entire logistics chain. Traditional logistics parks rely on manual scheduling and fixed route planning, which often exhibit slow response, low resource utilization, and poor anti-interference capabilities when facing ever-increasing order volumes and complex and ever-changing operating environments. Especially in scenarios involving high-concurrency orders, seasonal peaks, equipment failures, traffic congestion, weather changes, and other unforeseen events, dynamic optimization and real-time adjustments are difficult to achieve, leading to delivery delays, increased costs, and even service interruptions. In recent years, unmanned vehicle technology has been widely applied in logistics delivery. However, the dynamic and uncertain nature of smart logistics parks places higher demands on the collaborative scheduling of unmanned vehicle clusters. For example, the unpredictability of unforeseen events may disrupt pre-set delivery plans. How to achieve dynamic route replanning without interrupting overall operations has become a key challenge in the field of smart logistics.

[0003] Therefore, current technologies suffer from poor adaptability and slow response to sudden and unpredictable events, leading to a decline in the overall delivery efficiency and insufficient operational reliability of unmanned vehicle clusters. Summary of the Invention

[0004] This application provides a method and system for collaborative delivery of unmanned vehicle clusters in smart logistics parks. It solves the technical problems in existing technologies, such as poor adaptability and slow response of path planning to sudden and unpredictable events, which leads to a decline in the overall delivery efficiency and insufficient operational reliability of unmanned vehicle clusters. It achieves the technical effect of improving the anti-interference ability and adaptability of unmanned vehicle cluster collaborative delivery strategies through predictive evaluation and flexible optimization, thereby improving the overall delivery efficiency and operational reliability.

[0005] This application provides a method for collaborative delivery of unmanned vehicle clusters in a smart logistics park. The method includes: acquiring a real-time logistics monitoring dataset of a target smart logistics park; inputting the real-time logistics monitoring dataset into a contingency prediction model for prediction to obtain a first predicted contingency, wherein the contingency prediction model is obtained by training on a set of historical contingencies of the target smart logistics park; performing collaborative delivery analysis of the unmanned vehicle cluster according to target delivery order information and unmanned vehicle cluster status information to obtain an initial collaborative delivery strategy group, wherein the initial collaborative delivery strategy group includes the delivery planning paths obtained by optimizing each unmanned vehicle; evaluating the flexibility resistance index of the initial collaborative delivery strategy group according to the first predicted contingency, optimizing the initial collaborative delivery strategy group according to the index difference between the flexibility resistance index and a preset flexibility resistance index, and outputting a first collaborative delivery strategy group.

[0006] In a possible implementation, the unmanned vehicle cluster collaborative delivery method in the smart logistics park further performs the following processing: extracting historical logistics monitoring dataset samples corresponding to the historical emergency event set; extracting features from the historical logistics monitoring dataset samples to obtain logistics monitoring time-series feature samples, logistics monitoring flow feature samples, and logistics monitoring environment feature samples; labeling training samples based on the logistics monitoring time-series feature samples, logistics monitoring flow feature samples, and logistics monitoring environment feature samples to obtain training sample data based on historical time windows; and training a time series prediction network according to the training sample data to obtain an emergency event prediction model.

[0007] In a possible implementation, the unmanned vehicle cluster collaborative delivery method in the smart logistics park also performs the following processing: acquiring training sample data based on the historical time window; the training sample data includes extracting event type labels for each historical emergency, binary classification labels representing the occurrence of the event, and logistics monitoring time sequence feature alignment samples, logistics monitoring flow feature alignment samples, and logistics monitoring environment feature alignment samples based on each historical emergency in the corresponding historical time window.

[0008] In a possible implementation, the unmanned vehicle cluster collaborative delivery method in the smart logistics park further performs the following processing: parsing the order delivery demand vector of the target delivery order information; parsing the vehicle delivery capacity vector of the unmanned vehicle cluster status information; constructing an order-vehicle matching benefit matrix based on the order delivery demand vector and the vehicle delivery capacity vector, wherein the benefit scoring items of the order-vehicle matching benefit matrix include path reachability score, energy consumption cost score, and time window fit score; and performing group collaborative delivery path planning for the unmanned vehicle cluster according to the order-vehicle matching benefit matrix to output an initial collaborative delivery strategy group.

[0009] In a possible implementation, the unmanned vehicle cluster collaborative delivery method in the smart logistics park further performs the following processing: identifying the event type of the first predicted emergency; determining an emergency disturbance vector according to the event type, the emergency disturbance vector including at least one of path disturbance range, delay disturbance duration, and vehicle state disturbance; evaluating multiple flexible adversarial factors of each delivery strategy in the initial collaborative delivery strategy group based on the emergency disturbance vector, including path reconfigurability, delivery task portability, vehicle state redundancy, and affected path proportion; and weighting and summing the multiple flexible adversarial factors to obtain a flexible adversarial index.

[0010] In a possible implementation, the unmanned vehicle cluster collaborative delivery method in the smart logistics park further performs the following processing: calculating the difference between the flexible countermeasure index and the preset flexible countermeasure index; when the difference is greater than a preset threshold, an optimization instruction is triggered, wherein the optimization instruction includes at least one of path reconstruction optimization instruction, task migration optimization instruction, state compensation optimization instruction, and collaborative reconstruction optimization instruction.

[0011] In a possible implementation, the unmanned vehicle cluster collaborative delivery method in the smart logistics park also performs the following processing: generating task optimization constraints based on the triggered optimization instructions; replanning the group collaborative delivery path of the unmanned vehicle cluster based on the task optimization constraints and the order-vehicle matching benefit matrix, and outputting a first collaborative delivery strategy group.

[0012] In a possible implementation, the unmanned vehicle cluster collaborative delivery method in the smart logistics park further performs the following processing: inputting the real-time logistics monitoring dataset into a contingency prediction model for prediction, and obtaining the top N predicted contingencies with a probability greater than the expected occurrence; evaluating N flexible countermeasure indicators of the initial collaborative delivery strategy group according to the N predicted contingencies; calculating the N indicator differences between the N flexible countermeasure indicators and the preset flexible countermeasure indicators; and performing collaborative optimization of the initial collaborative delivery strategy group based on the N indicator differences to output a first collaborative delivery strategy group.

[0013] In a possible implementation, the unmanned vehicle cluster collaborative delivery method in the smart logistics park further performs the following processing: identifying N optimization instructions corresponding to the N index differences; extracting overlapping optimization instructions from the N optimization instructions; performing collaborative optimization of the group collaborative delivery path of the unmanned vehicle cluster according to the overlapping optimization instructions; and outputting a first collaborative delivery strategy group.

[0014] This application also provides a collaborative delivery system for unmanned vehicle clusters in a smart logistics park. The system includes: a logistics monitoring data acquisition module for acquiring real-time logistics monitoring datasets of a target smart logistics park; a contingency prediction acquisition module for inputting the real-time logistics monitoring datasets into a contingency prediction model to predict and acquire a first predicted contingency, wherein the contingency prediction model is acquired by training on a set of historical contingencies of the target smart logistics park; a collaborative delivery analysis module for performing collaborative delivery analysis of unmanned vehicle clusters according to target delivery order information and unmanned vehicle cluster status information to acquire an initial collaborative delivery strategy group, wherein the initial collaborative delivery strategy group includes the delivery planning paths obtained by optimizing each unmanned vehicle; and a collaborative delivery strategy optimization module for evaluating the flexibility resistance index of the initial collaborative delivery strategy group according to the first predicted contingency, optimizing the initial collaborative delivery strategy group according to the index difference between the flexibility resistance index and a preset flexibility resistance index, and outputting a first collaborative delivery strategy group.

[0015] This application proposes a method and system for collaborative delivery of unmanned vehicle clusters in smart logistics parks. Based on real-time logistics monitoring data from the smart logistics park, it uses a contingency prediction model to predict and obtain the first predicted contingency event. Combining target delivery orders and vehicle cluster status information, it generates an initial collaborative delivery strategy group based on the optimal path. A flexible adversarial index is introduced to evaluate the initial collaborative delivery strategy group, and optimization is performed by comparing it with a preset flexible adversarial index, resulting in the output of the first collaborative delivery strategy group. This solves the technical problems in existing technologies, such as poor adaptability and slow response of path planning to sudden and unpredictable events, leading to a decrease in the overall delivery efficiency and insufficient operational reliability of the unmanned vehicle cluster. It achieves the technical effect of dynamically enhancing the anti-interference capability and adaptability of the collaborative delivery strategy of the unmanned vehicle cluster through predictive evaluation and flexible optimization, thereby improving overall delivery efficiency and operational reliability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a schematic diagram of the unmanned vehicle cluster collaborative delivery method in a smart logistics park provided in an embodiment of this application.

[0018] Figure 2This is a schematic diagram of the unmanned vehicle cluster collaborative delivery system for a smart logistics park provided in an embodiment of this application.

[0019] Figure labeling: 10 Logistics monitoring data acquisition module, 20 Predictive emergency event acquisition module, 30 Collaborative delivery analysis module, 40 Collaborative delivery strategy optimization module. Detailed Implementation

[0020] To further illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose, the following detailed description is provided in conjunction with the accompanying drawings and preferred embodiments, based on the specific implementation methods, structures, features and effects of the present invention.

[0021] This application provides a method for collaborative delivery of unmanned vehicles in a smart logistics park, such as... Figure 1 As shown, the method includes:

[0022] Step S100: Obtain the real-time logistics monitoring dataset of the target smart logistics park.

[0023] Preferably, various sensors, devices, and information units deployed in the target smart logistics park collect logistics monitoring data in real time, forming a real-time logistics monitoring dataset. This dataset reflects the current status of various elements of logistics operations within the park, including structured or unstructured data. This data primarily includes environmental and infrastructure status data, unmanned vehicle cluster status data, order and cargo dynamic data, and external environmental data. Environmental and infrastructure status data includes traffic flow data, i.e., real-time vehicle and personnel density and flow rate in key areas such as main roads, intersections, and loading / unloading areas, obtained through analysis of geomagnetic sensors, lidar, and camera video streams; road status data, i.e., whether critical paths are closed or congested due to temporary construction, cargo stacking, water damage, etc., obtained through cameras combined with computer vision or road obstacle sensors; and facility availability data, i.e., the current occupancy status, number of idle facilities, and estimated waiting time of facilities such as loading / unloading platforms, charging piles, and parcel sorting cabinets.

[0024] Preferably, the unmanned vehicle cluster status data includes position and attitude data, namely the real-time GPS / BeiDou coordinates, speed, heading angle, and acceleration of each unmanned vehicle, acquired through the onboard GNSS module and inertial measurement unit (IMU); operational status data, namely the onboard battery / fuel level, temperature and health status of core components such as drive motors, sensors, and computing units, and current load weight / volume; and task execution data, namely whether a task is currently being executed, the mileage traveled, the estimated remaining completion time of the current task, and the deviation from the planned path. Order and cargo dynamic data includes order demand data, namely detailed information on orders requiring delivery, such as cargo size / weight / special requirements, origin coordinates, destination coordinates, and required delivery time window; and cargo status data, such as the real-time location in the warehouse, whether outbound picking has been completed, and whether it has been loaded onto the unmanned vehicle. External environmental data includes real-time meteorological data within the smart logistics park area, such as precipitation, wind speed, visibility, and temperature, acquired through the park's micro-weather stations.

[0025] Step S200: Input the real-time logistics monitoring dataset into the emergency prediction model for prediction to obtain the first predicted emergency event. The emergency prediction model is obtained by training on the historical emergency event set of the target smart logistics park.

[0026] Step S200 further includes step S210, extracting historical logistics monitoring dataset samples corresponding to the historical emergency event set; step S220, performing feature extraction on the historical logistics monitoring dataset samples to obtain logistics monitoring time-series feature samples, logistics monitoring flow feature samples, and logistics monitoring environment feature samples; step S230, labeling training samples based on the logistics monitoring time-series feature samples, logistics monitoring flow feature samples, and logistics monitoring environment feature samples to obtain training sample data based on historical time windows; and step S240, training a time series prediction network according to the training sample data to obtain an emergency event prediction model.

[0027] Preferably, the emergency prediction model is obtained by training on a set of historical emergency events in the target smart logistics park. Specifically, samples of historical logistics monitoring datasets corresponding to the set of historical emergency events are extracted. The set of historical emergency events is an event log that records all known past emergency events. Each record includes at least the type of emergency event, the location / event of the event, such as "Main Road A is severely congested", "Charging Station No. 3 is malfunctioning", "Unmanned Vehicle No. 5 battery overheating alarm", "Order backlog in Area B exceeds the threshold", and the severity level of the event, such as minor, severe, or fatal. For each emergency event in the event log, all relevant logistics monitoring data within a certain period before the occurrence of the emergency event are retrieved from the historical database. For example, when analyzing the "Main Road A is severely congested" event, all data on vehicle flow, vehicle speed, camera footage, etc., related to Main Road A within 30 minutes before the congestion occurred are extracted.

[0028] Preferably, features are extracted from historical logistics monitoring datasets to determine monitoring time-series features, monitoring flow features, and monitoring environment features, which respectively form logistics monitoring time-series feature samples, logistics monitoring flow feature samples, and logistics monitoring environment feature samples. The logistics monitoring time-series feature samples are extracted from timestamped data, reflecting the patterns and trends of data changes over time. Examples include the average vehicle speed change trend in a specific area over the past 5 minutes (stable / increasing / decreasing), the periodicity of charging pile occupancy throughout the day, and the deviation of order arrival rate from the average level. The logistics monitoring flow feature samples mainly come from traffic flow and order flow data, reflecting the load and pressure at a certain moment. Examples include the instantaneous number of vehicles at key path nodes, the number of orders entering a certain area per unit time, and the average waiting queue length of loading and unloading platforms. The logistics monitoring environment feature samples come from meteorological sensors, equipment status sensors, etc., reflecting the state of the external physical environment, such as real-time rainfall, ambient temperature, the average operating temperature of key unmanned vehicle components, and the average wind speed within the park.

[0029] Preferably, the historical time window refers to the data period before each emergency, such as 30 minutes before the emergency occurs. Then, training samples are labeled based on logistics monitoring time-series feature samples, logistics monitoring flow feature samples, and logistics monitoring environmental feature samples. That is, all time-series, flow, and environmental features belonging to the historical time window are used to form input samples and are labeled to indicate the emergency that occurs after the end of the time window. For example, if the input of the sample is that the speed of vehicles on main road A continues to decrease, the vehicle density exceeds the threshold, and there is light rain, the label of the sample is that main road A is severely congested. In this way, a large number of paired samples of pre-event data features and emergency results are determined to form training sample data for training the emergency prediction model. Then, a time series prediction network is used to train the training sample data. The time series prediction network is a deep learning model used to process data with time sequence, such as LSTM (Long Short-Term Memory) network and Transformer. It can learn potential patterns that predict future outcomes from data arranged in time. The training sample data is input into the time series prediction network, and through repeated learning and adjustment of internal parameters, it eventually learns to predict the likely occurrence of a corresponding sudden event in the future based on real-time data features similar to a certain pre-event pattern in history, thereby realizing proactive early warning of sudden events.

[0030] Furthermore, step S230 also includes acquiring training sample data based on the historical time window; the training sample data includes extracting event type labels for each historical emergency, binary classification labels representing the occurrence of the event, and logistics monitoring time sequence feature alignment samples, logistics monitoring flow feature alignment samples, and logistics monitoring environment feature alignment samples based on each historical emergency in the corresponding historical time window.

[0031] Preferably, the training sample data includes an event type label extracted for each historical emergency and a binary classification label representing the occurrence of the event. The event type label explicitly indicates the specific type of emergency that actually occurred after the corresponding historical time window ended, such as traffic congestion, charging pile failure, autonomous vehicle battery alarm, order backlog, etc., and is used to train the model for multi-class prediction of emergency events. The binary classification label representing the occurrence of the event is a Boolean value of yes / no or 1 / 0 code, indicating whether any emergency occurred after the given historical time window. For example, if any emergency such as "traffic congestion" or "vehicle failure" occurred, it is 1; if everything is normal and no emergency occurred, it is 0. This is used to train the model for anomaly detection, that is, to first determine whether the state is normal. Based on the logistics monitoring time-series feature alignment samples, logistics monitoring flow feature alignment samples, and logistics monitoring environment feature alignment samples within the corresponding historical time window for each historical emergency, all different types of feature data are segmented and matched according to the same time benchmark to ensure that they describe the same time period. Specifically, the logistics monitoring time-series feature alignment samples refer to feature data collected at fixed time intervals and arranged in chronological order within a specific time window before the event occurs; the logistics monitoring flow feature alignment samples are feature data reflecting system load collected within the same time window; and the logistics monitoring environment feature alignment samples are environmental data collected within the same time window.

[0032] Preferably, the real-time logistics monitoring dataset is input into the emergency prediction model for prediction. Specifically, real-time time-series data, real-time traffic data, and real-time environmental data for the current time window are obtained. For example, the average vehicle speed per minute at key intersections, the number of vehicles queuing per minute in front of the loading and unloading platform, and the rainfall reported per minute by the park's weather station in the past 10 minutes are collected. Features are extracted from these real-time data to obtain the same real-time time-series features, real-time traffic features, and real-time environmental features as the training data. Then, these are input into the trained and deployed emergency prediction model. Its internal time-series prediction network analyzes the real-time features based on the prediction patterns learned from historical data and outputs prediction results, including the probability distribution of event types formed by the possible occurrence probabilities of various emergencies and a binary classification judgment to determine whether an emergency has occurred. Finally, the first predicted emergency event is obtained, which is the most likely emergency event, that is, the event type with the highest probability in the event type probability distribution.

[0033] Step S300: Perform unmanned vehicle cluster collaborative delivery analysis according to the target delivery order information and the unmanned vehicle cluster status information to obtain an initial collaborative delivery strategy group, which includes the delivery planning path obtained by each unmanned vehicle through optimization.

[0034] Preferably, the system acquires target delivery order information and unmanned vehicle cluster status information. The target delivery order information is a list of tasks to be completed, including cargo attributes such as volume, weight, and category for each order, precise location information of pick-up / unloading points, delivery time constraints and corresponding priorities for each order. The unmanned vehicle cluster status information is a list of currently available transportation resources, including vehicle attributes such as unique identifier, maximum load, maximum volume, and vehicle type for each unmanned vehicle, dynamic status such as real-time location, current battery / fuel level, current load, and operating status, as well as performance constraints such as maximum speed and driving range. Then, based on the target delivery order information and the status information of the unmanned vehicle cluster, an unmanned vehicle cluster collaborative delivery analysis is performed, including delivery task allocation and path planning. This involves considering the current location of the vehicles, their load capacity, and the matching degree with the orders, rationally allocating order tasks to unmanned vehicle resources, and planning the optimal delivery route for each unmanned vehicle to ensure that all pickup and unloading points are visited in sequence, thereby minimizing the overall cost and maximizing efficiency. For example, this avoids multiple vehicles going to the same congested loading and unloading point at the same time, assigns orders in the same direction to the same vehicle to reduce the total mileage, and prioritizes assigning tasks to vehicles with sufficient battery power and close to the delivery point. This results in the output of an initial collaborative delivery strategy group, which includes the optimized delivery planning path obtained by each unmanned vehicle. That is, each collaborative delivery strategy includes the delivery task sequence of each unmanned vehicle and the detailed driving path of each unmanned vehicle, ensuring the highest logistics delivery efficiency under ideal conditions.

[0035] Furthermore, step S300 also includes step S310, parsing the order delivery demand vector of the target delivery order information; step S320, parsing the vehicle delivery capability vector of the unmanned vehicle cluster status information; step S330, constructing an order-vehicle matching benefit matrix based on the order delivery demand vector and the vehicle delivery capability vector, wherein the benefit scoring items of the order-vehicle matching benefit matrix include path reachability score, energy consumption cost score, and time window fit score; and step S340, performing group collaborative delivery path planning for the unmanned vehicle cluster according to the order-vehicle matching benefit matrix, and outputting an initial collaborative delivery strategy group.

[0036] Preferably, the algorithm analyzes the order delivery demand vector of the target delivery order information, that is, transforms each order demand into a vector representing all key attributes of the order, including delivery node location, delivery time window, and delivery priority, such as pick-up / unloading point coordinates, earliest available service time / latest delivery time; and analyzes the vehicle delivery capability vector of the autonomous vehicle cluster status information, that is, transforms the status capability information of each autonomous vehicle into a digital feature vector, which may include current location coordinates, remaining battery power, maximum load / volume, current load weight, and average driving speed. Then, based on the order delivery demand vector and the vehicle delivery capability vector, an order-vehicle matching benefit matrix is ​​constructed, which quantitatively evaluates the merits of assigning each order to each autonomous vehicle. In this matrix, the rows represent all delivery task orders to be assigned, the columns represent all available autonomous vehicles, and the elements in the matrix represent the comprehensive benefit score obtained by assigning delivery task orders to autonomous vehicles. The higher the score, the better the matching.

[0037] Preferably, the revenue scoring items of the order-vehicle matching revenue matrix include path accessibility score, energy cost score, and time window fit score. Specifically, the path accessibility score refers to calculating the additional driving distance or time required to allocate the vehicle for the order based on the current location of the autonomous vehicle and the pickup / unloading location of the order. The shorter the distance, the higher the path accessibility score. The energy cost score refers to estimating the energy consumption required to complete the order by combining the vehicle's remaining battery power. The lower the energy consumption and the more sufficient the vehicle's battery power, the higher the energy cost score. The time window fit score refers to evaluating whether the time when the autonomous vehicle is expected to arrive at the order pickup or unloading point after being allocated an existing task order falls within the time window required by the order. If it arrives early with a short waiting time or is exactly within the time window, the fit is higher and the time window fit score is higher. If it is expected to be late, the score is negative or zero.

[0038] Preferably, the autonomous vehicle cluster is used for collaborative delivery path planning based on the order-vehicle matching benefit matrix. This involves using integer programming or heuristic algorithms to solve for the globally optimal or near-optimal task allocation and path planning scheme using the order-vehicle matching benefit matrix. Specifically, considering the current driving paths and loads of all autonomous vehicles, the algorithm analyzes the possible positions to insert new delivery orders into the existing paths of the autonomous vehicles and evaluates the impact of each insertion method on the scores of each item in the order-vehicle matching benefit matrix. The algorithm then decides which autonomous vehicles should be assigned to the delivery orders and the order in which they should be inserted into the existing driving paths of the autonomous vehicles to maximize the global total benefit. Finally, the algorithm outputs an initial collaborative delivery strategy group, which includes the sequence of delivery orders assigned to each autonomous vehicle and the detailed planned path calculated based on the order sequence.

[0039] Step S400: Evaluate the flexibility resistance index of the initial collaborative delivery strategy group according to the first predicted emergency event, optimize the initial collaborative delivery strategy group according to the index difference between the flexibility resistance index and the preset flexibility resistance index, and output the first collaborative delivery strategy group.

[0040] Step S400 further includes step S410, identifying the event type of the first predicted emergency; step S420, determining an emergency disturbance vector according to the event type, the emergency disturbance vector including at least one of path disturbance range, delay disturbance duration, and vehicle state disturbance; step S430, evaluating multiple flexible adversarial factors of each delivery strategy in the initial collaborative delivery strategy group based on the emergency disturbance vector, including path reconfigurability, delivery task portability, vehicle state redundancy, and proportion of affected paths; and step S440, weighting and summing the multiple flexible adversarial factors to obtain a flexible adversarial index.

[0041] Preferably, the event type of the first predicted emergency is accurately identified, such as route congestion, vehicle breakdown, severe weather, etc. The abstract emergency event type is transformed into an emergency disturbance vector whose impact can be quantified and calculated. This vector is used to define the specific impact parameters of the emergency event on the logistics system. The emergency disturbance vector includes at least one of the following: route disturbance range, delay disturbance duration, and vehicle status disturbance. The route disturbance range is a geospatial parameter that defines the specific area affected by the event; the delay disturbance duration is a time parameter that defines the expected delay caused by the event; and the vehicle status disturbance is a resource parameter that defines the impact of the event on the vehicle's own state.

[0042] Preferably, multiple resilience factors are evaluated for each delivery strategy in the initial collaborative delivery strategy group based on the sudden disturbance vector. These resilience factors are quantitative indicators used to measure the buffering, adjustment, and recovery capabilities of the initial collaborative delivery strategy in the face of sudden disturbances. Specifically, they include path reconfigurability, delivery task portability, vehicle state redundancy, and the proportion of affected paths. Path reconfigurability measures whether paths in the strategy can be easily adjusted to bypass disturbances, examines the proportion of all path segments in the initial collaborative delivery strategy within the affected area, and assesses whether there are other feasible alternative paths for the affected path segments that do not drastically increase costs. The more alternative paths and the smaller the cost increase, the higher the path reconfigurability score. Delivery task portability measures whether tasks in the initial collaborative delivery strategy can be easily transferred from one vehicle to another, identifies the tasks carried by vehicles whose paths are affected, and assesses whether there are other idle or low-task vehicles in the autonomous vehicle cluster that can take over these tasks. The more available backup vehicles and the lower the difficulty of task handover, the higher the delivery task portability score.

[0043] Preferably, vehicle state redundancy is used to measure whether vehicles have sufficient resource reserves to cope with additional consumption. It examines whether affected vehicles have sufficient surplus in terms of battery power, load capacity, and time buffer. For example, if a predicted emergency will cause a 20% increase in energy consumption, a vehicle with only 15% battery remaining has low redundancy, while a vehicle with 80% battery remaining has high redundancy. The affected path ratio measures the direct impact of the emergency on the initial collaborative delivery strategy. It is calculated directly as the percentage of path lengths intersecting with the path disturbance range out of the total path length in all planned paths of the initial collaborative delivery strategy. The lower the percentage, the better the affected path ratio. Based on the importance of each flexible countermeasure factor under the current emergency type, corresponding weights are assigned. The scores of each flexible countermeasure factor are then normalized. Finally, multiple flexible countermeasure factors are weighted and summed to output a total score representing the ability of each initial delivery strategy to cope with the specific predicted emergency, which is the flexible countermeasure index. The higher the score, the more resilient the initial collaborative delivery strategy.

[0044] Furthermore, step S400 also includes step S450, calculating the difference between the flexible countermeasure index and the preset flexible countermeasure index; when the difference is greater than a preset threshold, an optimization instruction is triggered, wherein the optimization instruction includes at least one of path reconstruction optimization instruction, task migration optimization instruction, state compensation optimization instruction, and collaborative reconstruction optimization instruction.

[0045] Preferably, the preset flexible resistance index is a preset resilience qualification line, representing the lowest acceptable risk level. The difference between the flexible resistance index and the preset flexible resistance index is calculated to clearly indicate the degree of unacceptability of the current initial collaborative delivery strategy's resilience. Then, a preset threshold is set based on historical data to define the range of acceptable small differences, usually a very small positive number or 0. The index difference is compared with the preset threshold. If the index difference is less than or equal to the preset threshold, it means that the resilience of the current initial collaborative delivery strategy has reached or is very close to the safety requirements, and no optimization is needed; the strategy can be directly adopted. If the index difference is greater than the preset threshold, it means that the resilience of the current initial collaborative delivery strategy is significantly insufficient, and there is an unacceptable risk. In this case, an optimization instruction is automatically triggered. The optimization instruction includes at least one of the following: path reconstruction optimization instruction, task migration optimization instruction, state compensation optimization instruction, and collaborative reconstruction optimization instruction.

[0046] Preferably, when the evaluation finds that the path reconfigurability factor score is low and the proportion of affected paths is high, a path reconfiguration optimization command is triggered, instructing the path planning algorithm to find alternative routes for the affected vehicles to bypass the impact area of ​​the sudden event; when the evaluation finds that the delivery task portability factor score is low, a task migration optimization command is triggered, such as when a critical vehicle has an excessive workload and no other vehicle can easily take over, instructing the task allocation algorithm to reallocate some delivery tasks from one vehicle to another to disperse risk and balance the load; when the evaluation finds that the vehicle state redundancy factor score is low, a state compensation optimization command is triggered, such as when the vehicle's battery cannot support the additional energy consumption caused by detours or weather, instructing compensatory measures to be taken, such as instructing the vehicle to charge before performing the task or planning a speed curve with lower energy consumption for it; when multiple flexible adversarial factors score low at the same time, a collaborative reconfiguration optimization command is triggered, that is, instructing a global joint optimization, which may combine path reconfiguration, task migration and state compensation, etc., to regenerate a new, highly resilient collaborative delivery strategy.

[0047] Furthermore, step S400 also includes step S460, generating task optimization constraints according to the triggered optimization instructions; step S470, replanning the group collaborative delivery path of the unmanned vehicle cluster according to the task optimization constraints and the order-vehicle matching benefit matrix, and outputting the first collaborative delivery strategy group.

[0048] Preferably, each triggered optimization instruction is converted into a corresponding task optimization constraint. Specifically, if a path reconstruction optimization instruction is triggered, the generated task optimization constraint may explicitly prohibit certain path options, such as the path of vehicle V1 must not pass through the event-affected area Z; if a task migration optimization instruction is triggered, the generated task optimization constraint may forcibly change the binding relationship between the task and the vehicle, such as removing task T1 from the task list of vehicle V1 and adding task T1 to the task list of vehicle V2; if a state compensation optimization instruction is triggered, the generated task optimization constraint may insert fixed, mandatory task points into the vehicle path sequence, such as requiring vehicle V1 to visit charging station Y1 before starting subsequent tasks, and staying there for no less than 20 minutes; if a collaborative reconstruction optimization instruction is triggered, the generated constraint may be a combination of all task optimization constraints and may include global constraints, such as the total additional mileage of all vehicles must not exceed X kilometers.

[0049] Preferably, the task optimization constraints and the order-vehicle matching benefit matrix are used as inputs to re-execute the group collaborative delivery route optimization. That is, using improved genetic algorithms, large-scale neighborhood search algorithms and other route planning algorithms, under the premise of satisfying all task optimization constraints, we once again look for task allocation and route planning schemes that can maximize the total benefit score. That is, we find a balance between meeting resilience requirements and pursuing operational efficiency. For example, we identify a route that is longer than the initial route but perfectly avoids congested areas, or we identify a route that assigns the task to another vehicle that is farther away and has higher overall efficiency. Finally, the replanned delivery scheme is output, which is the first collaborative delivery strategy group, which contains multiple comprehensive schemes that maintain high operational efficiency and have resilience to cope with specific predictive risks.

[0050] Furthermore, step S470 also includes step S471, inputting the real-time logistics monitoring dataset into the emergency prediction model for prediction, and obtaining the top N predicted emergencies with a probability greater than the expected occurrence; step S472, evaluating N flexible countermeasure indicators of the initial collaborative delivery strategy group according to the N predicted emergencies; step S473, calculating the N indicator differences between the N flexible countermeasure indicators and the preset flexible countermeasure indicators; step S474, performing collaborative optimization on the initial collaborative delivery strategy group based on the N indicator differences, and outputting the first collaborative delivery strategy group.

[0051] Preferably, the real-time logistics monitoring dataset is processed and input into the emergency prediction model for prediction. The model outputs a list of potential events sorted in descending order of probability of occurrence. The top N predicted emergencies with a probability greater than the expected probability are selected from this list, where N is a positive integer representing the number of predicted emergencies. Then, N independent evaluations are performed on each of the N predicted emergencies. For each selected predicted emergency, a corresponding emergency disturbance vector is determined based on the event type. This vector is then used to evaluate each strategy in the initial collaborative strategy group. N flexible countermeasures indicators for the initial collaborative delivery strategy group against predicted emergencies are calculated, each indicator representing its ability to cope with congestion risk. The system assesses the ability to handle risks from harsh environments, charging pile malfunctions, or autonomous vehicle failures. For each risk scenario, it calculates the differences between N flexible countermeasure indicators and N preset flexible countermeasure indicators to clearly demonstrate the multiple weaknesses of the initial collaborative delivery strategy when facing different risks. Then, based on the differences of the N indicators, it performs collaborative optimization on the initial collaborative delivery strategy group, identifying and determining the collaborative delivery strategy group with the strongest global resilience. This ensures that all N flexible countermeasure indicators are as high as possible under various risk scenarios, and that the differences between them and the preset indicators are all less than preset thresholds. Finally, it obtains the first collaborative delivery strategy group, which can guarantee the highest stability in task completion when congestion, heavy rain, and charging uncertainties coexist.

[0052] Furthermore, step S474 also includes identifying the N optimization instructions corresponding to the N index differences; extracting the overlapping optimization instructions of the N optimization instructions; performing group collaborative delivery path optimization on the unmanned vehicle cluster according to the overlapping optimization instructions; and outputting the first collaborative delivery strategy group.

[0053] Preferably, the system identifies N corresponding optimization instructions generated for the differences in N indicators, determines an optimization instruction list, including at least one of path reconstruction optimization instructions, task migration optimization instructions, state compensation optimization instructions, and collaborative reconstruction optimization instructions. Then, it analyzes the optimization instruction list to identify overlapping optimization instructions that are triggered multiple times among the N optimization instructions. This means that the optimization instruction is simultaneously required by multiple unexpected events with different risks; for example, the path reconstruction optimization instruction is triggered simultaneously by unexpected event A and unexpected event B. Finally, the system performs collaborative optimization of the unmanned vehicle cluster's delivery path according to the overlapping optimization instructions. This involves generating comprehensive task optimization constraints based on the overlapping optimization instructions, performing global collaborative delivery path planning under the premise of satisfying comprehensive task priority constraints, and processing non-overlapping optimization instructions with lower priority or as alternative optimization schemes. Through collaborative optimization of the group's collaborative delivery path centered on overlapping optimization instructions, the system ultimately outputs a first collaborative delivery strategy group. This achieves maximum protection against multiple potential risks through minimal and critical adjustments, thereby ensuring improved overall delivery efficiency and operational reliability.

[0054] In the above text, refer to Figure 1 This paper describes in detail a method for collaborative delivery of unmanned vehicles in a smart logistics park according to an embodiment of the present invention. Next, we will refer to... Figure 2 This invention describes an unmanned vehicle cluster collaborative delivery system for a smart logistics park according to an embodiment of the present invention.

[0055] The unmanned vehicle cluster collaborative delivery system for smart logistics parks according to embodiments of the present invention addresses the technical problems in existing technologies, such as poor adaptability and slow response of path planning to sudden and unpredictable events, leading to a decline in the overall delivery efficiency and insufficient operational reliability of the unmanned vehicle cluster. It achieves the technical effect of dynamically enhancing the anti-interference capability and adaptability of the unmanned vehicle cluster collaborative delivery strategy through predictive evaluation and flexible optimization, thereby improving overall delivery efficiency and operational reliability. Figure 2 As shown, the unmanned vehicle cluster collaborative delivery system of the smart logistics park includes: a logistics monitoring data acquisition module 10, a predictive emergency acquisition module 20, a collaborative delivery analysis module 30, and a collaborative delivery strategy optimization module 40.

[0056] The logistics monitoring data acquisition module 10 is used to acquire real-time logistics monitoring datasets of the target smart logistics park; the emergency prediction acquisition module 20 is used to input the real-time logistics monitoring datasets into an emergency prediction model for prediction, and acquire a first predicted emergency, wherein the emergency prediction model is acquired by training on a set of historical emergencies of the target smart logistics park; the collaborative delivery analysis module 30 is used to perform collaborative delivery analysis of the unmanned vehicle cluster according to the target delivery order information and the unmanned vehicle cluster status information, and acquire an initial collaborative delivery strategy group, wherein the initial collaborative delivery strategy group includes the delivery planning paths obtained by each unmanned vehicle through optimization; the collaborative delivery strategy optimization module 40 is used to evaluate the flexibility resistance index of the initial collaborative delivery strategy group according to the first predicted emergency, optimize the initial collaborative delivery strategy group according to the index difference between the flexibility resistance index and the preset flexibility resistance index, and output a first collaborative delivery strategy group.

[0057] The specific configuration of the emergency prediction acquisition module 20 will be described in detail below. The emergency prediction acquisition module 20 further includes: extracting historical logistics monitoring dataset samples corresponding to the historical emergency set; performing feature extraction on the historical logistics monitoring dataset samples to obtain logistics monitoring time-series feature samples, logistics monitoring flow feature samples, and logistics monitoring environment feature samples; labeling training samples based on the logistics monitoring time-series feature samples, logistics monitoring flow feature samples, and logistics monitoring environment feature samples to obtain training sample data based on historical time windows; and training a time series prediction network according to the training sample data to obtain an emergency prediction model.

[0058] The specific configuration of the predictive emergency acquisition module 20 will be described in detail below. The predictive emergency acquisition module 20 further includes: acquiring training sample data based on the historical time window; the training sample data includes extracted event type labels for each historical emergency, binary classification labels representing the occurrence of the event, and aligned samples of logistics monitoring time-series features, logistics monitoring flow features, and logistics monitoring environment features based on each historical emergency within the corresponding historical time window.

[0059] The specific configuration of the collaborative delivery analysis module 30 will be described in detail below. The collaborative delivery analysis module 30 further includes: parsing the order delivery demand vector of the target delivery order information; parsing the vehicle delivery capacity vector of the unmanned vehicle cluster status information; constructing an order-vehicle matching benefit matrix based on the order delivery demand vector and the vehicle delivery capacity vector, wherein the benefit scoring items of the order-vehicle matching benefit matrix include path reachability score, energy consumption cost score, and time window fit score; and performing group collaborative delivery path planning for the unmanned vehicle cluster according to the order-vehicle matching benefit matrix, outputting an initial collaborative delivery strategy group.

[0060] The specific configuration of the collaborative delivery strategy optimization module 40 will be described in detail below. The collaborative delivery strategy optimization module 40 further includes: identifying the event type of the first predicted emergency; determining a sudden disturbance vector according to the event type, the sudden disturbance vector including at least one of path disturbance range, delay disturbance duration, and vehicle state disturbance; evaluating multiple flexible adversarial factors for each delivery strategy in the initial collaborative delivery strategy group based on the sudden disturbance vector, including path reconfigurability, delivery task portability, vehicle state redundancy, and the proportion of affected paths; and weighted summing the multiple flexible adversarial factors to obtain a flexible adversarial index.

[0061] The specific configuration of the collaborative delivery strategy optimization module 40 will be described in detail below. The collaborative delivery strategy optimization module 40 further includes: calculating the difference between the flexible countermeasure index and a preset flexible countermeasure index; triggering an optimization instruction when the index difference is greater than a preset threshold, wherein the optimization instruction includes at least one of a path reconstruction optimization instruction, a task migration optimization instruction, a state compensation optimization instruction, and a collaborative reconstruction optimization instruction.

[0062] The specific configuration of the collaborative delivery strategy optimization module 40 will be described in detail below. The collaborative delivery strategy optimization module 40 further includes: generating task optimization constraints based on triggered optimization instructions; replanning the group collaborative delivery path of the unmanned vehicle cluster based on the task optimization constraints and the order-vehicle matching benefit matrix, and outputting a first collaborative delivery strategy group.

[0063] The specific configuration of the collaborative delivery strategy optimization module 40 will be described in detail below. The collaborative delivery strategy optimization module 40 further includes: inputting the real-time logistics monitoring dataset into a contingency prediction model for prediction, obtaining the top N predicted contingencies with a probability greater than the expected occurrence; evaluating N flexible countermeasure indicators of the initial collaborative delivery strategy group according to the N predicted contingencies; calculating the N indicator differences between the N flexible countermeasure indicators and preset flexible countermeasure indicators; and performing collaborative optimization of the initial collaborative delivery strategy group based on the N indicator differences to output a first collaborative delivery strategy group.

[0064] The specific configuration of the collaborative delivery strategy optimization module 40 will be described in detail below. The collaborative delivery strategy optimization module 40 further includes: identifying N optimization instructions corresponding to the N index differences; extracting overlapping optimization instructions from the N optimization instructions; performing collaborative optimization of the group collaborative delivery path for the unmanned vehicle cluster according to the overlapping optimization instructions; and outputting a first collaborative delivery strategy group.

[0065] The unmanned vehicle cluster collaborative delivery system for smart logistics parks provided in this embodiment of the invention can execute the unmanned vehicle cluster collaborative delivery method for smart logistics parks provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for collaborative delivery of unmanned vehicles in a smart logistics park, characterized in that, The method includes: Obtain the real-time logistics monitoring dataset of the target smart logistics park; The real-time logistics monitoring dataset is input into the emergency prediction model for prediction to obtain the first predicted emergency event. The emergency prediction model is obtained by training on the historical emergency event set of the target smart logistics park. Based on the target delivery order information and the status information of the unmanned vehicle cluster, an analysis of the unmanned vehicle cluster collaborative delivery is performed to obtain an initial collaborative delivery strategy group, which includes the delivery planning path obtained by each unmanned vehicle through optimization. The initial collaborative delivery strategy group is evaluated based on the first predicted emergency event. The initial collaborative delivery strategy group is then optimized based on the difference between the initial collaborative delivery strategy group and the preset collaborative delivery strategy group, and a first collaborative delivery strategy group is output. The method for evaluating the flexibility resistance index of the initial collaborative delivery strategy group according to the first predicted emergency includes: Identify the event type of the first predicted emergency; The sudden disturbance vector is determined according to the event type, and the sudden disturbance vector includes at least one of the following: path disturbance range, delay disturbance duration, and vehicle state disturbance. The initial collaborative delivery strategy group is evaluated based on the sudden disturbance vector, including multiple flexible countermeasure factors for each delivery strategy, such as path reconfigurability, delivery task portability, vehicle state redundancy, and the proportion of affected paths. The multiple flexible countermeasure factors are weighted and summed to obtain the flexible countermeasure index; Among them, the difference between the flexible countermeasure index and the preset flexible countermeasure index is calculated; When the difference in the indicators is greater than a preset threshold, an optimization instruction is triggered, wherein the optimization instruction includes at least one of path reconstruction optimization instruction, task migration optimization instruction, state compensation optimization instruction, and collaborative reconstruction optimization instruction; The optimization of the initial collaborative delivery strategy group based on the triggered optimization instructions includes the following methods: Generate task optimization constraints based on the triggered optimization instructions; Based on the task optimization constraints and the order-vehicle matching benefit matrix, the unmanned vehicle cluster is replanned for group collaborative delivery path replanning, and the first collaborative delivery strategy group is output.

2. The unmanned vehicle cluster collaborative delivery method for smart logistics parks as described in claim 1, characterized in that, The emergency prediction model is obtained by training on a set of historical emergencies in the target smart logistics park, and the method includes: Extract samples of the historical logistics monitoring dataset corresponding to the set of historical emergencies; Feature extraction is performed on the historical logistics monitoring dataset samples to obtain logistics monitoring time-series feature samples, logistics monitoring flow feature samples, and logistics monitoring environment feature samples. Training sample annotation is performed based on the logistics monitoring time-series feature samples, logistics monitoring flow feature samples, and logistics monitoring environment feature samples to obtain training sample data based on historical time windows. The time series prediction network is trained using the training sample data to obtain the event prediction model.

3. The unmanned vehicle cluster collaborative delivery method for smart logistics parks as described in claim 2, characterized in that, Obtain training sample data based on the historical time window; The training sample data includes event type labels extracted for each historical emergency, binary classification labels representing the occurrence of the event, and logistics monitoring time-series feature alignment samples, logistics monitoring flow feature alignment samples, and logistics monitoring environment feature alignment samples based on each historical emergency in the corresponding historical time window.

4. The unmanned vehicle cluster collaborative delivery method for smart logistics parks as described in claim 1, characterized in that, Based on the target delivery order information and the autonomous vehicle cluster status information, analyze the collaborative delivery of the autonomous vehicle cluster to obtain an initial collaborative delivery strategy group. The methods include: Parse the order delivery demand vector of the target delivery order information; Parse the vehicle delivery capability vector of the unmanned vehicle cluster status information; Based on the order delivery demand vector and the vehicle delivery capacity vector, an order-vehicle matching benefit matrix is ​​constructed. The benefit scoring items of the order-vehicle matching benefit matrix include path accessibility score, energy consumption cost score, and time window fit score. Based on the order-vehicle matching benefit matrix, the unmanned vehicle cluster is used to plan the group collaborative delivery path and output the initial collaborative delivery strategy group.

5. The unmanned vehicle cluster collaborative delivery method for smart logistics parks as described in claim 1, characterized in that, After inputting the real-time logistics monitoring dataset into the emergency prediction model for prediction and obtaining the first predicted emergency, the method further includes: The real-time logistics monitoring dataset is input into the emergency prediction model for prediction, and the top N predicted emergencies with a probability greater than the expected occurrence are obtained. Evaluate the N flexible countermeasure indicators of the initial collaborative delivery strategy group based on the N predicted contingency events; Calculate the differences between the N flexible countermeasure indicators and the N preset flexible countermeasure indicators; Based on the differences of the N indicators, the initial collaborative delivery strategy group is optimized to output the first collaborative delivery strategy group.

6. The unmanned vehicle cluster collaborative delivery method for smart logistics parks as described in claim 5, characterized in that, The initial collaborative delivery strategy group is collaboratively optimized based on the differences of the N indicators to output a first collaborative delivery strategy group. The method includes: Identify the N optimization instructions corresponding to the N index differences; Extract the overlapping optimization instructions from the N optimization instructions, perform collaborative optimization of the group delivery path of the unmanned vehicle cluster according to the overlapping optimization instructions, and output the first collaborative delivery strategy group.

7. A collaborative delivery system for unmanned vehicles in a smart logistics park, characterized in that: The system is used to implement the unmanned vehicle cluster collaborative delivery method for smart logistics parks according to any one of claims 1 to 6, and the system includes: The logistics monitoring data acquisition module is used to acquire real-time logistics monitoring datasets of the target smart logistics park. The emergency event prediction module is used to input the real-time logistics monitoring dataset into the emergency event prediction model for prediction and to obtain the first predicted emergency event. The emergency event prediction model is obtained by training on the historical emergency event set of the target smart logistics park. The collaborative delivery analysis module is used to perform collaborative delivery analysis of the unmanned vehicle cluster according to the target delivery order information and the status information of the unmanned vehicle cluster, and to obtain an initial collaborative delivery strategy group, which includes the delivery planning path obtained by each unmanned vehicle through optimization. The collaborative delivery strategy optimization module is used to evaluate the flexibility resistance index of the initial collaborative delivery strategy group according to the first predicted emergency, optimize the initial collaborative delivery strategy group according to the index difference between the flexibility resistance index and the preset flexibility resistance index, and output the first collaborative delivery strategy group.

Images