Logistics transportation process multi-task collaborative management method based on deep learning

Abstract

The application discloses a logistics transportation process multi-task cooperative management and control method based on deep learning, which comprises the following steps: collecting logistics transportation basic operation data, order pickup and delivery node time window, route coincidence relationship and loading and unloading site connection relationship, constructing vehicle main body capability record, order urgency record and vehicle task bundle execution period; extracting historical execution deviation and processing it into a single-value deviation intensity sequence; obtaining window truncation position, effective window length and drift trigger mark by using ADWIN algorithm; determining stable vehicle task bundle and to-be-reconstructed order task section when drift triggering; calculating order drift contribution value and order coupling strength, and extending to obtain a reconstructed order task set; generating candidate order results by improving CBBA algorithm and locally consensus resolution, and merging to obtain a cooperative management and control result. The application forms logistics transportation multi-task dynamic cooperative management and control based on deviation drift detection and local bidding reconstruction.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling technology, and in particular to a multi-task collaborative management and control method for logistics transportation processes based on deep learning. Background Technology

[0002] Logistics transportation process control involves multiple aspects, including order dispatch, vehicle scheduling, driver status management, pickup and delivery time window constraints, load constraints, route connections, and station waiting control. With the expansion of trunk line transportation, urban distribution transportation, and on-demand delivery services, the time dependencies, route overlaps, loading and unloading connections, and vehicle capacity constraints between transportation tasks are constantly increasing. Single order sorting or single vehicle route optimization methods are insufficient to cover the dynamic coupling relationships under the parallel execution of multiple tasks. Existing intelligent logistics systems have begun to introduce data-driven scheduling, predicted arrival time analysis, vehicle status monitoring, and multi-agent task allocation algorithms to assist in forming order dispatch results and vehicle task bundles during the transportation process.

[0003] In existing technologies, one approach constructs a static scheduling model based on order time windows, vehicle load, and distance costs, and generates vehicle task plans through rule engines or heuristic search. Another approach establishes an arrival time prediction model based on historical transportation data, and then performs rolling order dispatch based on vehicle availability. A third approach employs a multi-agent bidding mechanism, enabling vehicle entities to engage in local bidding based on task benefits and route insertion costs, and resolves the problem of multiple vehicles competing for the same order through a consensus process. These approaches can handle vehicle allocation, route insertion, and time window verification in general transportation scheduling scenarios, but they still rely on preset rules for the continuous evolution of operational deviations during transportation, the reconstruction of local task bundles, and the dynamic prioritization of conflicting orders.

[0004] However, when faced with multi-order time window compression, local disturbances in vehicle task bundles, changes in route connections, and the continuous accumulation of execution deviations, these methods tend to treat deviation events as isolated anomalies, lacking a unified approach to transform arrival time deviations, station waiting deviations, load occupancy deviations, and route deviations into detectable time sequences. Existing bidding allocation methods primarily rely on task revenue and execution costs, failing to fully reflect the constraints of order drift contribution, order coupling strength, and order urgency in candidate order ranking, local consensus resolution, and stable task bundle merging. This results in difficulties in forming unified intelligent collaborative decisions in multi-task dynamic coupling scenarios, leading to lags in transportation process control and response.

[0005] Therefore, how to provide a multi-task collaborative management and control method for logistics transportation processes based on deep learning is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a multi-task collaborative management and control method for logistics transportation processes based on deep learning. This invention forms multi-task collaborative decision-making by detecting deviation drift and improving CBBA local reconstruction, thereby reducing the response lag in transportation process management and control.

[0007] The deep learning-based multi-task collaborative management and control method for logistics transportation processes according to embodiments of the present invention includes:

[0008] Collect basic logistics and transportation operation data, order pickup and delivery node time windows, route overlap relationships, and loading and unloading station connection relationships to construct vehicle main capacity records, order urgency records, and vehicle task execution time periods;

[0009] Historical execution deviations are extracted from basic logistics and transportation operation data and processed to obtain a single-value deviation intensity sequence.

[0010] The ADWIN algorithm is used to process the single-valued bias intensity sequence to obtain the window truncation position, effective window length, and drift trigger flag;

[0011] When the drift trigger flag is set, the reserved transportation period is determined based on the window truncation position and the effective window length. The start and end times of the order pickup and delivery node time windows, the vehicle task bundle execution period and the reserved transportation period are compared respectively to obtain the stable vehicle task bundle and the order task segment to be reconstructed.

[0012] The order drift contribution value is determined based on the cumulative difference in the single-value deviation strength of the same order before and after the window truncation position. The order coupling strength is calculated based on the route overlap relationship, loading and unloading station connection relationship and order urgency record. The task segment of the order to be reconstructed is expanded to obtain the set of reconstructed order tasks.

[0013] The reconstructed order task set is written into the improved CBBA algorithm, a basic competitive value is constructed based on the vehicle's main capability record, and a comprehensive penalty term is written into the basic competitive value to obtain candidate order results.

[0014] Based on the candidate order results, a candidate vehicle task bundle is constructed. The consensus order of conflicting orders is determined according to the order drift contribution value, order coupling strength and order urgency. Local consensus resolution is then performed on the candidate vehicle task bundle to obtain the resolved candidate vehicle task bundle.

[0015] The candidate vehicle task bundles are merged with the stable vehicle task bundles to obtain the logistics transportation collaborative management results.

[0016] Optionally, the construction of vehicle main capability records, order urgency records, and vehicle task bundle execution time periods specifically includes:

[0017] By collecting logistics transportation orders, vehicle and driver status, vehicle remaining load, estimated arrival time and actual arrival time according to order identifiers, basic logistics transportation operation data is obtained.

[0018] Write the order pickup and delivery node time window to the corresponding order identifier to obtain the order time window index record;

[0019] Based on the vehicle and driver status, vehicle remaining load, and order window index records, candidate matching is performed on the orders that each vehicle can accept, and candidate matching records are obtained.

[0020] Write the route overlap relationship and loading / unloading station connection relationship into the candidate matching record, and sort them according to the length of the overlapping road segment and the order of the loading / unloading station connection between the candidate orders corresponding to the same vehicle to obtain the candidate task connection order;

[0021] Based on the candidate task connection order, candidate orders with vehicle remaining load capacity less than the candidate order load capacity requirement and vehicle and driver status not in a dispatchable state are removed to obtain the vehicle main capacity record.

[0022] Based on the order pickup and delivery node time windows, estimated arrival time and actual arrival time, calculate the remaining time of pickup node, remaining time of delivery node and arrival deviation time respectively, and calculate the connection waiting time according to the connection order of loading and unloading stations of adjacent candidate orders in the candidate task connection priority.

[0023] The smaller of the remaining time at the pickup node and the remaining time at the delivery node is taken as the time window remaining benchmark. Candidate orders are sorted according to the time window remaining benchmark, arrival deviation time, and connection waiting time to obtain the order urgency record.

[0024] Based on the vehicle's main capability record and the order urgency record, the execution time of the candidate tasks is reorganized to obtain the vehicle task bundle execution time.

[0025] Optionally, the step of extracting and processing historical execution deviations from basic logistics and transportation operation data specifically involves:

[0026] Historical execution deviations are extracted from basic logistics and transportation operation data according to order identifiers and execution times. The historical execution deviations are then linked with the order pickup and delivery node time windows, route overlap relationships, and vehicle task execution periods to obtain deviation extraction index records.

[0027] Based on the deviation extraction index records, the arrival time deviation, station waiting deviation, load occupancy deviation, and route deviation deviation are determined respectively, and written into the same deviation item set according to the corresponding order identifier and deviation type;

[0028] The set of deviation terms is unified into positive values ​​according to the direction of increasing deviation, thus obtaining the set of positive deviation terms;

[0029] Based on the maximum and minimum deviation values ​​of the same deviation item in historical execution deviations, the set of positive deviation items is normalized using the same scale to obtain the normalized set of deviation items.

[0030] Based on the order urgency record, determine the order in which the deviations corresponding to each order in the normalized deviation item set are written, and perform sequential arrangement on the normalized deviation item set to obtain the candidate sequence of deviation intensity;

[0031] Establish deviation weight rankings according to deviation type, and perform forward shift correction on the deviation weight rankings corresponding to the same order based on the order urgency record to obtain deviation weight marking results;

[0032] The candidate sequence of deviation intensity is rearranged according to the execution time, and the normalized deviation terms corresponding to the same execution time are weighted and summarized according to the deviation weight labeling results to obtain a single-value deviation intensity sequence.

[0033] Optionally, the ADWIN algorithm is used to process the single-valued bias intensity sequence, specifically as follows:

[0034] The single-value deviation intensity sequence is written into the adaptive detection window of the ADWIN algorithm according to the order of execution time to obtain the candidate detection window;

[0035] Based on the execution time sequence and the deviation corresponding to the order urgency record in the candidate detection window, write the order change position to generate a candidate segmentation position set;

[0036] The candidate detection window is divided into a front deviation window and a back deviation window based on the candidate segmentation positions in the candidate segmentation position set, and the mean difference and fluctuation difference between the windows are calculated to obtain the candidate drift judgment record.

[0037] From the candidate drift determination records, select candidate segmentation positions where the mean of the latter segment deviation window is higher than the mean of the former segment deviation window and the fluctuation difference reaches the preset fluctuation threshold, and obtain the effective segmentation positions.

[0038] The execution times corresponding to the effective segmentation positions are compared, and the effective segmentation position with the largest execution time is determined as the window truncation position. The single-value deviation intensity sequence after the window truncation position is retained to obtain the effective window length.

[0039] A drift trigger flag is generated when the effective window length reaches the preset detection length and the average value of the subsequent deviation window is continuously higher than the average value of the preceding deviation window.

[0040] Optionally, obtaining the stable vehicle task bundle and the order task segment to be reconstructed specifically involves:

[0041] When the drift trigger flag is set, the last execution time of the retained part in the single-value deviation intensity sequence is determined according to the effective window length, and the time period between the execution time corresponding to the window truncation position and the last execution time is determined as the retained transportation period.

[0042] Extract the start and end times of each vehicle task bundle from the vehicle task bundle execution time period according to the vehicle identification, and compare them with the start and end times of the retained transportation time period to obtain the task bundle time period comparison record.

[0043] Extract the start and end times of the pickup node and the start and end times of the delivery node from the order pickup and delivery node time window according to the order identifier, and compare them with the start and end times of the retained transportation period to obtain the order time window comparison record;

[0044] Based on the task bundle time period comparison record, vehicle task bundles whose task start time and task end time are both within the retained transportation time period are retained to obtain candidate stable vehicle task bundles.

[0045] Based on the order time window comparison record, extract orders whose pickup node start and end times or delivery node start and end times cross the boundary of the reserved transportation period to obtain candidate reconstructed orders;

[0046] Remove the task segments containing candidate reconstruction orders from the candidate stable vehicle task bundle to obtain the stable vehicle task bundle, and merge the removed task segments with the candidate reconstruction orders to obtain the task segments to be reconstructed.

[0047] Optionally, obtaining the reconstructed order task set specifically involves:

[0048] The single-value deviation intensity sequence is divided into a pre-truncation deviation segment and a post-truncation deviation segment according to the window truncation position, and the cumulative value of single-value deviation intensity in the pre-truncation deviation segment and the post-truncation deviation segment is calculated according to the order identifier.

[0049] The cumulative difference of single-value deviation intensity is calculated based on the cumulative value of single-value deviation intensity in the pre-truncation deviation segment and post-truncation deviation segment corresponding to the same order, and the order drift contribution value is determined.

[0050] Extract the length of overlapping road segments between the task segment to be reconstructed and adjacent orders based on the route overlap relationship, and extract the connection waiting time between corresponding adjacent orders based on the connection relationship of loading and unloading stations;

[0051] Based on the length of overlapping road segments, connection waiting time, and order urgency, adjacent orders are sorted in a progressive manner, and the sorted position is used as the order coupling strength.

[0052] Orders whose drift contribution value is greater than the preset contribution threshold are identified as drift expansion orders, and adjacent orders whose order coupling strength corresponds to a sorting position not greater than the preset coupling position threshold are identified as coupling expansion orders.

[0053] Write the drift-extended orders and coupled-extended orders into the task segment of the orders to be reconstructed, and delete duplicate orders according to the order identifier to obtain the set of reconstructed order tasks.

[0054] Optionally, the improved CBBA algorithm is specifically as follows:

[0055] The task set for reconstructing orders is split into task items according to the order identifier, and the order drift contribution value, order coupling strength and order urgency are recorded and written into the corresponding task items to obtain the local bidding task sequence.

[0056] Based on the vehicle's main capabilities record, establish vehicle task pairing relationships between each task item in the local bidding task sequence and the vehicles that can undertake the task, thereby obtaining the local reconstruction bidding domain.

[0057] Extract the occupied task time slots and assigned order order sequence from the stable vehicle task bundle according to the vehicle identifier, and determine the candidate insertion positions corresponding to the vehicle task pairing relationship in the local reconstruction bidding domain according to the order pickup and delivery node time window to obtain the candidate insertion position record;

[0058] Based on the vehicle's main capability record, candidate insertion position record, and local bidding task sequence, perform basic bidding sorting on the vehicle task pairing relationship, and write the sorted result into the basic bidding value.

[0059] The load constraint penalty is calculated based on the comparison between the load requirement of the candidate order and the remaining load of the vehicle, and the pickup and delivery time window default penalty is calculated based on the time difference between the expected arrival time corresponding to the candidate insertion position and the order pickup and delivery node time window.

[0060] The driving route deviation penalty is calculated based on the length of the road segment between the candidate order and the adjacent orders before and after the candidate insertion position that do not fall into the route overlap relationship, and the vehicle task switching penalty is calculated based on the number of times the inserted candidate order causes changes in the order of the assigned orders.

[0061] Calculate the drift coupling penalty based on the order drift contribution value corresponding to the candidate order and the order coupling strength between the candidate order and the adjacent orders before and after the candidate insertion position;

[0062] The load constraint penalty, delivery window violation penalty, route deviation penalty, vehicle task switching penalty, and drift coupling penalty are combined into a comprehensive penalty item and written into the basic competitive value;

[0063] The candidate order competitive value is obtained by deducting the base competitive value according to the direction of increasing comprehensive penalty items;

[0064] Based on the vehicle identifier, order identifier, and candidate insertion position record, the candidate order competitive value, basic competitive value, and comprehensive penalty item are linked to obtain the candidate order result.

[0065] Optionally, obtaining the candidate vehicle task bundle specifically involves:

[0066] Extract the competitive value of candidate orders from the candidate order results according to the vehicle identifier, sort the candidate orders corresponding to the same vehicle, and write the sorted candidate orders into the task bundle of the corresponding vehicle according to the candidate insertion position record to obtain the candidate vehicle task bundle.

[0067] Extract orders that appear simultaneously under at least two vehicle identifiers from the candidate vehicle task bundle to obtain a set of conflicting orders, and write the vehicle task pairing relationship containing conflicting orders into the conflict pairing record;

[0068] Based on the order drift contribution value, order coupling strength and order urgency records corresponding to the conflicting orders, the conflicting order set is sorted in a progressive manner to obtain the consensus order of the conflicting orders.

[0069] Conflicting orders are extracted sequentially according to the consensus order. Based on the conflict pairing records, the competitive value of candidate orders corresponding to the vehicle task pairing relationship of the same conflicting order is compared, and the pairing relationship to be retained is determined.

[0070] Remove vehicle task pairings that are not determined to be retained from the candidate vehicle task bundle, and simultaneously delete the corresponding candidate insertion position records to obtain the candidate vehicle task bundle to be verified.

[0071] The remaining candidate orders in the candidate vehicle task bundle to be verified are re-executed to extract the conflict order set. If there are no conflict orders, the candidate vehicle task bundle to be verified is determined as the candidate vehicle task bundle to be resolved.

[0072] Optionally, merging the candidate vehicle task bundle with the stabilizing vehicle task bundle specifically involves:

[0073] According to the vehicle identifier, the order identifier, task start time, task end time and candidate insertion position record are extracted and decrypted from the candidate vehicle task bundle and the stable vehicle task bundle, and a stable source mark and a candidate source mark are established for each task segment to obtain the vehicle task merging base sequence.

[0074] Write the task segments in the stable vehicle task bundle and the task segments in the candidate vehicle task bundle into the vehicle task merging base sequence according to the vehicle identifier to obtain the merged task bundle to be verified.

[0075] The start and end times of adjacent task segments corresponding to the same vehicle in the task bundle to be checked and merged are compared to obtain the time period overlap record.

[0076] Based on the overlapping time period records, extract the overlapping task segment pairs. When the task segment pair contains both stable source markers and candidate source markers, retain the task segment corresponding to the stable source marker and mark the task segment corresponding to the candidate source marker as the task segment to be adjusted.

[0077] When all task segments contain candidate source tags, compare the competitive value of candidate orders for the corresponding task segments, retain the task segments with higher competitive value, and mark the unretained task segments as task segments to be adjusted.

[0078] Based on the candidate insertion position record, re-match unoccupied candidate insertion positions for the task segment to be adjusted in the task bundle to be verified and merged to obtain the corrected merge task bundle;

[0079] Based on the order pickup and delivery node time window verification and correction, merge the task segments in the task bundle to obtain the target vehicle task bundle;

[0080] The target vehicle task bundle is recorded and output according to vehicle identifier, order identifier, task start time, task end time and candidate insertion position, to obtain the logistics transportation collaborative management and control results.

[0081] The beneficial effects of this invention are:

[0082] (1) This invention unifies the arrival time deviation, station waiting deviation, load occupancy deviation and driving route deviation into a single-value deviation intensity sequence, and uses the ADWIN algorithm to generate window truncation position, effective window length and drift trigger mark, thereby realizing the time-series detection and drift trigger identification of logistics transportation execution deviation, thus providing a reliable trigger condition for subsequent local reconstruction of vehicle task bundle.

[0083] (2) The present invention determines the order drift contribution value by the cumulative difference of the single value deviation intensity of the same order before and after the window truncation position, and calculates the order coupling strength by combining the route overlap relationship, loading and unloading station connection relationship and order urgency record, thereby realizing the extended filtering of the order task segment to be reconstructed, so that the reconstructed order task set can cover the deviation diffusion order and the coupled related order.

[0084] (3) The improved CBBA algorithm in this invention introduces order drift contribution value, order coupling strength, order urgency record and comprehensive penalty term in the process of candidate order bidding and local consensus resolution, realizes the linkage constraint on candidate order results, conflict order consensus order and resolution candidate vehicle task bundle, thereby forming a logistics transportation collaborative management result for multi-task dynamic coupling scenario. Attached Figure Description

[0085] 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:

[0086] Figure 1 This is a flowchart of the multi-task collaborative management and control method for logistics transportation based on deep learning proposed in this invention.

[0087] Figure 2 This is a flowchart of the improved CBBA algorithm for the multi-task collaborative management and control method of logistics transportation based on deep learning proposed in this invention. Detailed Implementation

[0088] Example 1: Reference Figures 1-2 A deep learning-based multi-task collaborative management and control method for logistics transportation processes includes:

[0089] Collect basic logistics and transportation operation data, order pickup and delivery node time windows, route overlap relationships, and loading and unloading station connection relationships to construct vehicle main capacity records, order urgency records, and vehicle task execution time periods;

[0090] Historical execution deviations are extracted from basic logistics and transportation operation data and processed to obtain a single-value deviation intensity sequence. The processing involves determining arrival time deviation, station waiting deviation, load occupancy deviation, and route deviation from the historical execution deviations, and then unifying each deviation into a positive value according to the direction of deviation increase before performing same-scale normalization.

[0091] The ADWIN algorithm is used to process the single-valued bias intensity sequence to obtain the window truncation position, effective window length, and drift trigger flag;

[0092] When the drift trigger flag is set, the reserved transportation period is determined based on the window truncation position and the effective window length. The start and end times of the order pickup and delivery node time windows, the vehicle task bundle execution period and the reserved transportation period are compared respectively to obtain the stable vehicle task bundle and the order task segment to be reconstructed.

[0093] The order drift contribution value is determined based on the cumulative difference in the single-value deviation strength of the same order before and after the window truncation position. The order coupling strength is calculated based on the route overlap relationship, loading and unloading station connection relationship and order urgency record. The order task segment to be reconstructed is expanded based on the order drift contribution value and the order coupling strength to obtain the reconstructed order task set.

[0094] The reconstructed order task set is written into the improved CBBA algorithm. The basic competitive value is constructed based on the vehicle's main capability record. The comprehensive penalty term is written into the basic competitive value to obtain the candidate order result. The comprehensive penalty term includes load constraint penalty, pickup and delivery time window violation penalty, driving route deviation penalty, vehicle task switching penalty and drift coupling penalty.

[0095] Based on the candidate order results, a candidate vehicle task bundle is constructed. The consensus order of conflicting orders is determined according to the order drift contribution value, order coupling strength and order urgency. Local consensus resolution is then performed on the candidate vehicle task bundle to obtain the resolved candidate vehicle task bundle.

[0096] The candidate vehicle task bundles are merged with the stable vehicle task bundles to obtain the collaborative management and control results of logistics transportation.

[0097] In this embodiment, the construction of vehicle main capability records, order urgency records, and vehicle task bundle execution time periods specifically includes:

[0098] By collecting logistics transportation orders, vehicle and driver status, vehicle remaining load, estimated arrival time and actual arrival time according to order identifiers, basic logistics transportation operation data is obtained.

[0099] Write the order pickup and delivery node time window to the corresponding order identifier to obtain the order time window index record;

[0100] Based on the vehicle and driver status, vehicle remaining load, and order window index records, candidate matching is performed on the orders that each vehicle can accept, and candidate matching records are obtained.

[0101] The overlapping relationships of routes and the connection relationships of loading and unloading stations are written into the candidate matching record. The candidate tasks are sorted according to the length of overlapping road segments between candidate orders corresponding to the same vehicle and the order of connection between loading and unloading stations. The sorting order is as follows: the first arrangement is performed according to the length of overlapping road segments between candidate orders corresponding to the same vehicle from large to small, and the second arrangement is performed according to the order of connection between loading and unloading stations when the length of overlapping road segments is the same.

[0102] Based on the candidate task connection order, candidate orders with vehicle remaining load capacity less than the candidate order load capacity requirement and vehicle and driver status not in a dispatchable state are removed to obtain the vehicle main capacity record.

[0103] Based on the order pickup and delivery node time windows, estimated arrival time and actual arrival time, calculate the remaining time of pickup node, remaining time of delivery node and arrival deviation time respectively, and calculate the connection waiting time according to the connection order of loading and unloading stations of adjacent candidate orders in the candidate task connection priority.

[0104] In Example 1, for each candidate order, the deadline times for pickup and delivery are extracted from the order pickup and delivery node time windows, and the estimated arrival time and actual arrival time are extracted from the basic logistics and transportation operation data. The remaining time for pickup is obtained by subtracting the estimated arrival time from the deadline time for pickup, and the remaining time for delivery is obtained by subtracting the estimated arrival time from the deadline time for delivery. The arrival deviation time is obtained by subtracting the estimated arrival time from the actual arrival time and taking the positive time. Adjacent candidate orders are extracted according to the candidate task connection order. The time difference between the completion time of the loading and unloading station of the previous candidate order and the start time of the loading and unloading station of the next candidate order is calculated. When the start time of the next candidate order is later than the completion time of the previous candidate order, the difference is used as the connection waiting time. When the start time of the next candidate order is not later than the completion time of the previous candidate order, the connection waiting time is recorded as 0.

[0105] The smaller of the remaining time at the pickup node and the remaining time at the delivery node is taken as the time window remaining benchmark. Candidate orders are sorted according to the time window remaining benchmark, arrival deviation time, and connection waiting time to obtain an order urgency record. The sorting order is as follows: the candidate orders are arranged in ascending order of the time window remaining benchmark; when the time window remaining benchmarks are the same, they are arranged in descending order of arrival deviation time; when the arrival deviation time is the same, they are arranged in ascending order of connection waiting time. The ranking after the fifth arrangement is used as the order urgency record.

[0106] Based on the vehicle's main capability record and the order urgency record, the execution time of the candidate tasks is reorganized to obtain the vehicle task bundle execution time.

[0107] In Example 1, candidate orders are extracted from the vehicle's main capability record based on vehicle identification, and these candidate orders are arranged according to the priority of the corresponding candidate tasks for the same vehicle. The order urgency record is written into the candidate orders for the same vehicle. When candidate orders with adjacent task priorities but discontinuous urgency rankings are encountered, their adjacent positions are adjusted from smallest to largest according to their urgency ranking. For each adjusted candidate order, the estimated arrival time is used as the task start reference time. The remaining time of the pickup node, the remaining time of the delivery node, and the connection waiting time are sequentially written into the task start reference time to obtain the candidate order start time and candidate order end time. The start times and end times of each candidate order under the same vehicle are arranged in chronological order to obtain the vehicle task bundle execution time period.

[0108] In this embodiment, the step of extracting and processing historical execution deviations from basic logistics and transportation operation data specifically involves:

[0109] Historical execution deviations are extracted from basic logistics and transportation operation data according to order identifiers and execution times. The historical execution deviations are then linked with the order pickup and delivery node time windows, route overlap relationships, and vehicle task execution periods to obtain deviation extraction index records.

[0110] In Example 1, when no historical execution deviations are extracted from the basic logistics and transportation operation data, the order pickup and delivery node time windows, estimated arrival time, actual arrival time, vehicle remaining load, candidate order load requirements, route overlap, and loading / unloading station connection relationships are retrieved according to the order identifier. The time difference between the estimated arrival time and the deadline of the order pickup and delivery node time window is recorded as the arrival timeliness deviation; the connection interval between adjacent candidate order loading / unloading stations is recorded as the station waiting deviation; the difference between the candidate order load requirements and the vehicle's remaining load is recorded as the load occupancy deviation; and the length of the road segment where there is no route overlap between candidate orders is recorded as the driving route deviation. The above four types of deviations are written into the same order identifier and execution time to form the initial historical execution deviation.

[0111] Based on the deviation extraction index records, the arrival time deviation, station waiting deviation, load occupancy deviation, and route deviation deviation are determined respectively, and written into the same deviation item set according to the corresponding order identifier and deviation type;

[0112] In Example 1, the order identifier and execution time in the deviation extraction index record are used to retrieve the corresponding order pickup and delivery node time window, route overlap relationship, and vehicle task bundle execution period. The time difference between the actual arrival time and the end time of the order pickup and delivery node time window is recorded as the arrival timeliness deviation. The time difference between the end time of the previous order and the start time of the next order in adjacent candidate orders is recorded as the station waiting deviation. The absolute value corresponding to the remaining difference after the candidate order's load demand occupies the vehicle's remaining load is less than 0 is recorded as the load occupancy deviation. The cumulative value of the length of the road segment that does not fall into the route overlap relationship is recorded as the driving route deviation deviation.

[0113] The arrival time deviation, station waiting deviation, load occupancy deviation, and route deviation deviation in the deviation item set are unified into positive values ​​according to the direction of increase of deviation, thus obtaining the positive deviation item set;

[0114] Based on the maximum and minimum deviation values ​​of the same deviation item in historical execution deviations, the set of positive deviation items is normalized using the same scale to obtain the normalized set of deviation items.

[0115] Based on the order urgency record, determine the order in which the deviations corresponding to each order in the normalized deviation item set are written, and perform sequential arrangement on the normalized deviation item set to obtain the candidate sequence of deviation intensity;

[0116] Establish a deviation weight order according to the deviation type, place the weight order corresponding to the arrival time deviation before the route deviation deviation, station waiting deviation and load occupancy deviation, and perform forward shift correction on the deviation weight order corresponding to the same order according to the order urgency record to obtain the deviation weight marking result;

[0117] In Example 1, a four-level deviation weight ranking table is established according to deviation type. The arrival time deviation is recorded as the first weight ranking, the route deviation deviation as the second weight ranking, the station waiting deviation as the third weight ranking, and the load occupancy deviation as the fourth weight ranking. The order urgency record is extracted by order identifier, and orders with a ranking lower than a preset urgency threshold are marked as urgent orders. For the same set of deviation items corresponding to urgent orders, the route deviation deviation, station waiting deviation, and load occupancy deviation are each shifted forward by one weight ranking. If the shifted value exceeds the first weight ranking, the first weight ranking is maintained. For the same set of deviation items not marked as urgent orders, the original weight ranking in the four-level deviation weight ranking table is maintained. The corrected weight ranking is written into the deviation weight marking result.

[0118] The candidate sequence of deviation intensity is rearranged according to the execution time, and the normalized deviation terms corresponding to the same execution time are weighted and summarized according to the deviation weight labeling results to obtain a single-value deviation intensity sequence.

[0119] In this embodiment, the ADWIN algorithm is used to process the single-valued bias intensity sequence, specifically as follows:

[0120] The single-value deviation intensity sequence is written into the adaptive detection window of the ADWIN algorithm according to the order of execution time to obtain the candidate detection window;

[0121] Based on the execution time sequence and the deviation corresponding to the order urgency record in the candidate detection window, write the order change position to generate a candidate segmentation position set;

[0122] In Example 1, order identifiers corresponding to two adjacent execution times are extracted sequentially according to the arrangement order of the single-value deviation intensity sequence within the candidate detection window; the deviation writing order of the same order identifier in two adjacent execution times is extracted from the order urgency record; the deviation writing order of the same order identifier in two adjacent execution times is compared, and when the deviation writing order moves forward, backward, enters the window, or exits the window, the next execution time is marked as a candidate segmentation position; when the deviation writing order does not change, no candidate segmentation position is generated; duplicate marks are removed according to the order of execution times, and all candidate segmentation positions are summarized to obtain a set of candidate segmentation positions.

[0123] The candidate detection window is divided into a front deviation window and a back deviation window based on the candidate segmentation positions in the candidate segmentation position set, and the mean difference and fluctuation difference between the windows are calculated to obtain the candidate drift judgment record.

[0124] In Example 1, for each candidate segmentation position in the candidate segmentation position set, the single-value deviation intensity earlier than the candidate segmentation position within the candidate detection window is written into the front segment deviation window, and the single-value deviation intensity of the candidate segmentation position and later than the candidate segmentation position is written into the back segment deviation window; the arithmetic mean and the average absolute value of the deviation intensity of the single-value deviation in the front segment deviation window are calculated, and the arithmetic mean and the average absolute value of the deviation intensity of the single-value deviation in the back segment deviation window are calculated; the arithmetic mean of the back segment deviation is subtracted from the arithmetic mean of the front segment to obtain the mean difference, and the absolute value of the deviation intensity of the back segment deviation is subtracted from the absolute value of the deviation intensity of the front segment deviation to obtain the fluctuation difference; the candidate segmentation position, the front segment deviation window, the back segment deviation window, the mean difference, and the fluctuation difference are written into the same candidate drift determination record;

[0125] From the candidate drift determination records, select candidate segmentation positions where the mean of the latter segment deviation window is higher than the mean of the former segment deviation window and the fluctuation difference reaches a preset fluctuation threshold to obtain effective segmentation positions. The preset fluctuation threshold is obtained by taking the first 30% of the single-value deviation intensity in the candidate detection window arranged according to the execution time as the reference segment, calculating the difference between each single-value deviation intensity in the reference segment and the mean of the reference segment, and taking the absolute value of each difference, and then multiplying the average of the absolute values ​​by 2.

[0126] The execution times corresponding to the effective segmentation positions are compared, and the effective segmentation position with the largest execution time is determined as the window truncation position. The single-value deviation intensity sequence after the window truncation position is retained to obtain the effective window length.

[0127] When the effective window length reaches the preset detection length and the average value of the subsequent deviation window is continuously higher than the average value of the preceding deviation window, a drift trigger mark is generated. The preset detection length is obtained by rounding up 30% of the number of single-value deviation intensities in the candidate detection window to obtain the length candidate value. When the length candidate value is less than 5, the preset detection length is set to 5. When the length candidate value is not less than 5, the length candidate value is set to the preset detection length.

[0128] In this embodiment, obtaining the stable vehicle task bundle and the order task segment to be reconstructed specifically involves:

[0129] When the drift trigger flag is set, the last execution time of the retained part in the single-value deviation intensity sequence is determined according to the effective window length, and the time period between the execution time corresponding to the window truncation position and the last execution time is determined as the retained transportation period.

[0130] Extract the start and end times of each vehicle task bundle from the vehicle task bundle execution time period according to the vehicle identification, and compare them with the start and end times of the retained transportation time period to obtain the task bundle time period comparison record.

[0131] In Example 1, the vehicle task bundle number, task start time, and task end time corresponding to the same vehicle are extracted according to the vehicle identifier within the vehicle task bundle execution period. The task start time is compared with the start time and end time of the reserved transportation period to obtain the start comparison results: the start time is earlier than the period, the start time is within the period, and the start time is later than the period. The task end time is compared with the start time and end time of the reserved transportation period to obtain the termination comparison results: the termination time is earlier than the period, the termination time is within the period, and the termination time is later than the period. The vehicle identifier, vehicle task bundle number, task start time, task end time, start comparison result, and termination comparison result are written into the same record to obtain the task bundle period comparison record.

[0132] Extract the start and end times of the pickup node and the start and end times of the delivery node from the order pickup and delivery node time window according to the order identifier, and compare them with the start and end times of the retained transportation period to obtain the order time window comparison record;

[0133] In Example 1, the start time, end time, start time, and end time of the pickup node are extracted from the pickup and delivery node time windows of the order according to the order identifier. The start time and end time of the pickup node are compared with the start time and end time of the reserved transportation period, respectively, to obtain the pickup comparison results for pickup time windows that are earlier than the time period, within the time period, later than the time period, and cross the time period boundary. The start time and end time of the delivery node are compared according to the same rules to obtain the delivery comparison results. The order identifier, the four node times, the pickup comparison results, and the delivery comparison results are written into the same record to obtain the order time window comparison record.

[0134] Based on the task bundle time period comparison record, vehicle task bundles whose task start time and task end time are both within the retained transportation time period are retained to obtain candidate stable vehicle task bundles.

[0135] Based on the order time window comparison record, extract orders whose pickup node start and end times or delivery node start and end times cross the boundary of the reserved transportation period to obtain candidate reconstructed orders;

[0136] Remove the task segments containing candidate reconstruction orders from the candidate stable vehicle task bundle to obtain the stable vehicle task bundle, and merge the removed task segments with the candidate reconstruction orders to obtain the task segments to be reconstructed.

[0137] In this embodiment, obtaining the reconstructed order task set specifically refers to:

[0138] The single-value deviation intensity sequence is divided into a pre-truncation deviation segment and a post-truncation deviation segment according to the window truncation position, and the cumulative value of single-value deviation intensity in the pre-truncation deviation segment and the post-truncation deviation segment is calculated according to the order identifier.

[0139] In Example 1, the single-value deviation intensity sequence is traversed in chronological order of execution time, and the truncation execution time corresponding to the window truncation position is extracted. Single-value deviation intensities earlier than the truncation execution time and their associated order identifiers are written into the deviation segment before truncation. Single-value deviation intensities equal to and later than the truncation execution time and their associated order identifiers are written into the deviation segment after truncation. During the writing process, the execution time order of the single-value deviation intensities in the original sequence is maintained, and the segment ownership mark of each single-value deviation intensity under the same order identifier is retained. When there are multiple order identifiers at the same execution time, the corresponding segments are written according to the order identifiers.

[0140] The cumulative difference of single-value deviation intensity is calculated based on the cumulative value of single-value deviation intensity in the deviation segments before and after truncation corresponding to the same order, and the cumulative difference of single-value deviation intensity is determined as the order drift contribution value.

[0141] Extract the length of overlapping road segments between the task segment to be reconstructed and adjacent orders based on the route overlap relationship, and extract the connection waiting time between corresponding adjacent orders based on the connection relationship of loading and unloading stations;

[0142] Adjacent orders are sorted in ascending order based on the length of overlapping road segments (from largest to smallest), connection waiting time (from smallest to largest), and order urgency (from smallest to largest). The sorted order position is then used as the order coupling strength.

[0143] Orders with a drift contribution value greater than a preset contribution threshold are identified as drift-extended orders. Adjacent orders whose order coupling strength corresponds to a sorting position not greater than a preset coupling position threshold are identified as coupling-extended orders. The preset contribution threshold is determined by arranging order drift contribution values ​​greater than 0 in ascending order, selecting the value at the 70th percentile after arrangement as a candidate threshold, and using the average value of the single-value deviation strength within the truncated deviation segment as the segment baseline value. When the candidate threshold is greater than the segment baseline value, the candidate threshold is set as the preset contribution threshold; when the candidate threshold is not greater than the segment baseline value, the segment is removed from the preset contribution threshold. The segment baseline value is set as a preset contribution threshold. When there is no order drift contribution value greater than 0, the preset contribution threshold is set to 0. The preset coupling position threshold is obtained by counting the number of adjacent orders corresponding to the order task segment to be reconstructed and rounding up 30% of the number of adjacent orders to obtain the coupling position candidate value. When the number of adjacent orders is less than 3, the number of adjacent orders is set as the preset coupling position threshold. When the number of adjacent orders is not less than 3 and the coupling position candidate value is less than 3, the preset coupling position threshold is set to 3. When the coupling position candidate value is not less than 3, the coupling position candidate value is set as the preset coupling position threshold.

[0144] Write the drift-extended orders and coupled-extended orders into the task segment of the orders to be reconstructed, and delete duplicate orders according to the order identifier to obtain the set of reconstructed order tasks.

[0145] In this embodiment, the improved CBBA algorithm is specifically as follows:

[0146] The task set for reconstructing orders is split into task items according to the order identifier, and the order drift contribution value, order coupling strength and order urgency are recorded and written into the corresponding task items to obtain the local bidding task sequence.

[0147] In Example 1, the order task set is reconstructed by traversing according to the order identifier. The order task segments to be reconstructed, drifted extended orders, and coupled extended order records associated with the same order identifier are merged into the same order record. For each order record, one task item is created, and the order identifier, order pickup and delivery node time window, candidate order load requirement, source record, and corresponding vehicle task segment number are written into the task item. When there are duplicate records with the same order identifier, the record whose execution time is closest to the window truncation position is retained. All task items are arranged according to the order identifier to complete the task item splitting.

[0148] Based on the vehicle's main capabilities record, establish vehicle task pairing relationships between each task item in the local bidding task sequence and the vehicles that can undertake the task, thereby obtaining the local reconstruction bidding domain.

[0149] Extract the occupied task time slots and assigned order order sequence from the stable vehicle task bundle according to the vehicle identifier, and determine the candidate insertion positions corresponding to the vehicle task pairing relationship in the local reconstruction bidding domain according to the order pickup and delivery node time window to obtain the candidate insertion position record;

[0150] In Example 1, the occupied task time period is the time from the start time to the end time of the reserved task segment of the same vehicle in the stable vehicle task bundle; the allocated order order order order identifiers within the same stable vehicle task bundle are arranged according to the order of the task start time; in the local reconstruction bidding domain, the order pickup and delivery node time windows of the task items are extracted according to the vehicle task pairing relationship, and candidate idle time periods are formed between adjacent occupied task time periods, at the beginning of the first occupied task time period, and at the end of the last occupied task time period; the start and end times of the candidate idle time periods are compared with the start and end times of the pickup node and the delivery node, respectively, and the time period positions that can accommodate the pickup node time window and the delivery node time window are reserved to obtain the candidate insertion positions;

[0151] Based on the vehicle's main capability record, candidate insertion position record, and local bidding task sequence, perform basic bidding sorting on the vehicle task pairing relationship, and write the sorted result into the basic bidding value.

[0152] In Example 1, for each vehicle task pairing, the remaining load and schedulable status of the vehicle are extracted from the vehicle's main capacity record, the start and end times of the candidate insertion position are extracted from the candidate insertion position record, and the order drift contribution value, order coupling strength, and order urgency record are extracted from the local bidding task sequence. Pairings with remaining load less than the candidate order load requirement, schedulable status not met, or candidate insertion positions unable to accommodate the order pickup and delivery node time windows are deleted. For the remaining pairings, the first sorting is performed according to the order urgency record from smallest to largest. If the sorting ranks are the same, the second sorting is performed according to the order drift contribution value from largest to smallest. If the order drift contribution values ​​are the same, the third sorting is performed according to the order coupling strength corresponding to the sorting ranks from smallest to largest, thus obtaining the basic bidding ranking.

[0153] The load constraint penalty is calculated based on the comparison between the load requirement of the candidate order and the remaining load of the vehicle, and the pickup and delivery time window default penalty is calculated based on the time difference between the expected arrival time corresponding to the candidate insertion position and the order pickup and delivery node time window.

[0154] In Example 1, the load constraint penalty is calculated by comparing the load requirement of the candidate order with the remaining load of the vehicle. If the load requirement of the candidate order is no greater than the remaining load of the vehicle, it is recorded as 0. If the load requirement of the candidate order is greater than the remaining load of the vehicle, the difference is recorded as the load constraint penalty. The estimated arrival time corresponding to the candidate insertion position is marked as the estimated arrival time for pickup and delivery according to the node type. If the estimated arrival time for pickup is later than the pickup node deadline, the time difference is recorded as the pickup time window default value; otherwise, it is recorded as 0. If the estimated arrival time for delivery is later than the delivery node deadline, the time difference is recorded as the delivery time window default value; otherwise, it is recorded as 0. The pickup time window default value and the delivery time window default value are added together to obtain the pickup and delivery time window default penalty.

[0155] The driving route deviation penalty is calculated based on the length of the road segment between the candidate order and the adjacent orders before and after the candidate insertion position that do not fall into the route overlap relationship, and the vehicle task switching penalty is calculated based on the number of times the inserted candidate order causes changes in the order of the assigned orders.

[0156] In Example 1, the preceding and following adjacent orders are extracted according to the candidate insertion position, and the planned driving segments between the candidate order and the preceding and following adjacent orders are extracted; the planned driving segments are matched for route overlap, and the lengths of segments that do not match the route overlap are accumulated to obtain the driving route deviation penalty; the original order arrangement is formed according to the order of the assigned orders before the candidate order is inserted, and the new order arrangement is formed according to the vehicle task bundle after the candidate order is inserted; the position of the same order identifier in the original order arrangement and the new order arrangement are compared, and the number of orders whose position changes is counted to obtain the vehicle task switching penalty;

[0157] Calculate the drift coupling penalty based on the order drift contribution value corresponding to the candidate order and the order coupling strength between the candidate order and the adjacent orders before and after the candidate insertion position;

[0158] In Example 1, the order drift contribution value corresponding to the candidate order is extracted from the local bidding task sequence, and the order drift contribution value is used as the drift base value; the preceding adjacent order and the following adjacent order are extracted from the candidate insertion position record, and the order coupling strength between the candidate order and the preceding adjacent order and the following adjacent order are extracted respectively; each sorting position is compared with a preset coupling position threshold. When the sorting position is not greater than the preset coupling position threshold, the drift base value is deducted in ascending order of sorting position to obtain the adjacency coupling deduction value; when the sorting position is greater than the preset coupling position threshold, the adjacency coupling deduction value is recorded as 0; the adjacency coupling deduction value corresponding to the preceding adjacent order and the following adjacent order is deducted from the drift base value, and the deducted value is used as the drift coupling penalty.

[0159] The load constraint penalty, delivery window violation penalty, route deviation penalty, vehicle task switching penalty, and drift coupling penalty are combined into a comprehensive penalty item and written into the basic competitive value;

[0160] In Example 1, the load constraint penalty, delivery window violation penalty, route deviation penalty, vehicle task switching penalty, and drift coupling penalty are sorted in ascending order of value within the same local reconstruction bidding domain. When the values ​​are the same, the sorting order is determined by the arrangement of vehicle identifier, order identifier, and candidate insertion position record. The sorting order of the five types of penalties is accumulated according to the same vehicle task pairing relationship to obtain the comprehensive penalty item.

[0161] The candidate order competitive value is obtained by deducting the base competitive value according to the direction of increasing comprehensive penalty items;

[0162] Based on the vehicle identifier, order identifier, and candidate insertion position record, the candidate order competitive value, basic competitive value, and comprehensive penalty item are linked to obtain the candidate order result.

[0163] In this embodiment, obtaining the candidate vehicle task bundle specifically involves:

[0164] Extract the competitive value of candidate orders from the candidate order results according to the vehicle identifier, sort the candidate orders corresponding to the same vehicle in descending order of competitive value, and write the sorted candidate orders into the task bundle of the corresponding vehicle according to the candidate insertion position record to obtain the candidate vehicle task bundle.

[0165] Extract orders that appear simultaneously under at least two vehicle identifiers from the candidate vehicle task bundle to obtain a set of conflicting orders, and write the vehicle task pairing relationship containing conflicting orders into the conflict pairing record;

[0166] Based on the order drift contribution value, order coupling strength, and order urgency records corresponding to the conflicting orders, a progressive sorting is performed on the conflicting order set to obtain the consensus order of the conflicting orders. The progressive sorting is performed as follows: first sorting is performed according to the order drift contribution value from large to small; second sorting is performed according to the order coupling strength from small to large when the order drift contribution values ​​are the same; and third sorting is performed according to the order urgency record from small to large when the order coupling strength is the same.

[0167] Conflicting orders are extracted sequentially according to the consensus order. Based on the conflict pairing records, the competitive value of candidate orders for each vehicle task pairing relationship corresponding to the same conflicting order is compared, and the vehicle task pairing relationship with the highest competitive value of candidate orders is determined as the retained pairing relationship.

[0168] In Example 1, conflicting orders are extracted in ascending order of their ranking in the consensus sequence of conflicting orders, and vehicle-task pairing relationships containing the current conflicting order are filtered in the conflict pairing record. For each selected vehicle-task pairing relationship, the competitive value of candidate orders associated with the same order identifier, the same vehicle identifier, and the same candidate insertion position is extracted from the candidate order results. The competitive values ​​of candidate orders are arranged in descending order of their values, and the vehicle-task pairing relationship corresponding to the candidate order competitive value with the largest value is determined as the retained pairing relationship. When the values ​​are the same, the vehicle-task pairing relationship is selected in ascending order of its ranking in terms of order urgency.

[0169] Remove vehicle task pairings that are not determined to be retained from the candidate vehicle task bundle, and simultaneously delete the corresponding candidate insertion position records to obtain the candidate vehicle task bundle to be verified.

[0170] The remaining candidate orders in the candidate vehicle task bundle to be verified are re-executed to extract the conflict order set. If there are no conflict orders, the candidate vehicle task bundle to be verified is determined as the candidate vehicle task bundle to be resolved.

[0171] In this embodiment, merging the candidate vehicle task bundle with the stabilizing vehicle task bundle specifically means:

[0172] According to the vehicle identifier, the order identifier, task start time, task end time and candidate insertion position record are extracted and decrypted from the candidate vehicle task bundle and the stable vehicle task bundle, and a stable source mark and a candidate source mark are established for each task segment to obtain the vehicle task merging base sequence.

[0173] The task segments in the stable vehicle task bundle and the task segments in the candidate vehicle task bundle are written into the vehicle task merging base sequence according to the vehicle identifier, and arranged in chronological order of task start time to obtain the merged task bundle to be verified.

[0174] The start and end times of adjacent task segments corresponding to the same vehicle in the task bundle to be checked and merged are compared. Task segment pairs in which the end time of the previous task segment is later than the start time of the next task segment are filtered to obtain time period overlap records.

[0175] Based on the overlapping time period records, extract the overlapping task segment pairs. When the task segment pair contains both stable source markers and candidate source markers, retain the task segment corresponding to the stable source marker and mark the task segment corresponding to the candidate source marker as the task segment to be adjusted.

[0176] When all task segments contain candidate source tags, compare the competitive value of candidate orders for the corresponding task segments, retain the task segments with higher competitive value, and mark the unretained task segments as task segments to be adjusted.

[0177] Based on the candidate insertion position record, re-match unoccupied candidate insertion positions for the task segments to be adjusted in the task bundle to be verified and merged. Write back the successfully matched task segments to be adjusted to the task bundle to be verified and merged, and delete the unmatched task segments to be adjusted from the task bundle to be verified and merged to obtain the corrected merge task bundle.

[0178] According to the order pickup and delivery node time window, the task segments in the merged task bundle are verified, corrected and merged. The task segments whose task start time and task end time exceed the corresponding order pickup and delivery node time window are deleted to obtain the target vehicle task bundle.

[0179] The target vehicle task bundle is recorded and output according to vehicle identifier, order identifier, task start time, task end time and candidate insertion position, to obtain the logistics transportation collaborative management and control results.

[0180] Example 2:

[0181] To verify the feasibility of this invention in practice, it was applied to a smart logistics transportation scheduling scenario involving multiple vehicles and multiple orders executing in parallel. The application targets were order tasks generated within a continuous transportation cycle, vehicle operating status, driver dispatchability status, order pickup and delivery node windows, vehicle remaining load, estimated arrival time, and actual arrival time. In this scenario, orders have overlapping routes and loading / unloading station connections. Some vehicle task bundles experience arrival time deviations, station waiting deviations, load occupancy deviations, and route deviations during execution. If execution continues according to the existing fixed vehicle task bundles, local delays will propagate along adjacent orders and loading / unloading station connections, leading to untimely vehicle task bundle reconstruction, repeated competition of conflicting orders, and delayed response in transportation process control.

[0182] In implementation, the system acquires basic logistics and transportation operation data in 5-minute intervals, and simultaneously accesses order pickup and delivery node time windows, route overlap relationships, and loading / unloading station connection relationships. This embodiment selects 48 orders and 12 vehicles as verification objects. The system collects logistics and transportation order information, vehicle and driver status, vehicle remaining load, estimated arrival time, and actual arrival time according to the order identifier, obtaining basic logistics and transportation operation data. The order pickup and delivery node time windows are written into the corresponding order identifier, obtaining order time window index records. Based on the vehicle and driver status, vehicle remaining load, and order time window index records, candidate matching is performed on the orders that each vehicle can accept, obtaining candidate matching records. Subsequently, route overlap relationships and loading / unloading station connection relationships are written into the candidate matching records. The first arrangement is performed according to the length of the overlapping road segment between candidate orders corresponding to the same vehicle, from largest to smallest. When the length of the overlapping road segment is the same, the second arrangement is performed according to the order of loading / unloading station connections, obtaining the candidate task connection priority.

[0183] Regarding the candidate task connection priority, the system eliminates candidate orders where the vehicle's remaining load capacity is less than the candidate order's load capacity requirement, or where the vehicle and driver are not in a dispatchable state, thus obtaining a vehicle main capacity record. Based on the order's pickup and delivery node time windows, estimated arrival time, and actual arrival time, the system calculates the remaining time for pickup, delivery, and arrival deviation, respectively, and calculates the connection waiting time based on the connection order of adjacent candidate orders' loading and unloading stations. Specifically, the system takes the smaller value between the remaining time of pickup and delivery as the time window remainder benchmark, sorts candidate orders in ascending order of time window remainder benchmark, descending order of arrival deviation time, and ascending order of connection waiting time, and records the sorting position as the order urgency. The system then reorganizes the candidate task connection priority execution time periods according to the vehicle main capacity record and the order urgency record to obtain the vehicle task bundle execution time periods.

[0184] During the deviation processing phase, the system extracts historical execution deviations from basic logistics and transportation operation data according to order identifiers and execution times. It then links these historical deviations with order pickup and delivery node time windows, route overlap relationships, and vehicle task execution periods, resulting in deviation extraction index records. Based on these index records, the system determines arrival timeliness deviations, station waiting deviations, load occupancy deviations, and route deviation deviations, and writes them into the same deviation item set according to the corresponding order identifier and deviation type. The four types of deviations in the deviation item set are then standardized to positive values ​​according to their increasing direction, resulting in a positive deviation item set. This set is then normalized using the maximum and minimum deviation values ​​for the same deviation item in historical execution deviations, resulting in a normalized deviation item set. The system determines the order in which deviations are written based on order urgency records and establishes deviation weight ordering according to deviation type. The weight corresponding to arrival timeliness deviations is placed before route deviation deviations, station waiting deviations, and load occupancy deviations. Finally, a forward shift correction is performed based on the order urgency records to obtain the deviation weight marking results. The system performs temporal rearrangement of the candidate sequence of deviation intensity according to the execution time, and performs weighted summation of the normalized deviation terms corresponding to the same execution time according to the deviation weight labeling results to obtain a single-value deviation intensity sequence.

[0185] During the drift detection phase, the system writes the sequence of single-value deviation intensity into the adaptive detection window of the ADWIN algorithm according to the execution time sequence, and the detection window retains the most recent 96 execution times. The system generates a candidate segmentation position set by writing the deviation corresponding to the execution time sequence and order urgency record in the candidate detection window according to the order of execution time in the candidate detection window. The candidate detection window is divided into a front-segment deviation window and a back-segment deviation window based on the candidate segmentation positions, and the mean difference and fluctuation difference between the windows are calculated to obtain the candidate drift judgment record. In this embodiment, the preset fluctuation threshold uses the top 30% of the single-value deviation intensity in the candidate detection window arranged by execution time as the benchmark segment. The average of the absolute values ​​of the differences between each single-value deviation intensity in the benchmark segment and the mean of the benchmark segment is calculated, and this average is multiplied by 2. The preset detection length is 30% of the number of single-value deviation intensities in the candidate detection window, rounded up. When the candidate length value is less than 5, it is set to 5. The system filters candidate segmentation positions where the mean of the latter segment deviation window is higher than the mean of the former segment deviation window and the fluctuation difference reaches a preset fluctuation threshold, thus obtaining effective segmentation positions. The effective segmentation position with the largest value at the execution time is determined as the window truncation position, and a drift trigger flag is generated when the effective window length reaches the preset detection length and the mean of the latter segment deviation window is continuously higher than the mean of the former segment deviation window.

[0186] After the drift trigger flag is established, the system determines the last execution time of the retained portion in the single-value deviation intensity sequence based on the effective window length, and defines the time period between the execution time corresponding to the window truncation position and the last execution time as the retained transportation period. The system extracts the task start and end times of each vehicle task bundle from the vehicle task bundle execution period according to the vehicle identifier, compares them with the start and end times of the retained transportation period, and obtains a task bundle time period comparison record. It also extracts the start and end times of the pickup and delivery nodes from the order pickup and delivery node time windows according to the order identifier, and compares them with the start and end times of the retained transportation period, and obtains an order time window comparison record. The system retains vehicle task bundles whose task start and end times are both within the retained transportation period, obtaining candidate stable vehicle task bundles; it extracts orders whose pickup or delivery node start and end times cross the boundary of the retained transportation period, obtaining candidate reconstructed orders; it removes task segments containing candidate reconstructed orders from the candidate stable vehicle task bundles, obtaining stable vehicle task bundles, and merges the removed task segments with the candidate reconstructed orders to obtain the task segments of the orders to be reconstructed.

[0187] The system divides the single-value deviation intensity sequence into pre-truncation and post-truncation deviation segments based on the window truncation position. It then calculates the cumulative single-value deviation intensity value in each segment according to the order identifier, determines the cumulative difference in single-value deviation intensity as the order drift contribution value, and extracts the length of overlapping road segments between the task segment to be reconstructed and adjacent orders based on route overlap relationships. It also extracts the connection waiting time between corresponding adjacent orders based on loading / unloading station connection relationships. Adjacent orders are progressively sorted according to their overlapping road segment length (from largest to smallest), connection waiting time (from smallest to largest), and order urgency (from smallest to largest) ranking, and this ranking is used as the order coupling strength. Orders with a drift contribution value greater than a preset contribution threshold are identified as drift-extended orders, and adjacent orders with a ranking corresponding to their order coupling strength not greater than a preset coupling ranking threshold are identified as coupling-extended orders. These are written into the task segment to be reconstructed. After deleting duplicate orders, the reconstructed order task set is obtained.

[0188] In the improved CBBA algorithm execution phase, the system splits the reconstructed order task set into task items according to the order identifier, and records the order drift contribution value, order coupling strength, and order urgency into the corresponding task items to obtain a local bidding task sequence. The system establishes vehicle task pairing relationships based on vehicle capability records to obtain a local reconstruction bidding domain; it extracts occupied task time slots and allocated order sequences from stable vehicle task bundles, and determines candidate insertion position records based on order pickup and delivery node time windows. The system performs basic bidding sorting based on vehicle capability records, candidate insertion position records, and the local bidding task sequence, and writes the sorting position into the basic bidding value; it calculates load constraint penalties, pickup and delivery time window violation penalties, route deviation penalties, vehicle task switching penalties, and drift coupling penalties respectively, and accumulates the five types of penalties according to the same vehicle task pairing relationship, combining them into a comprehensive penalty item and writing it into the basic bidding value; it subtracts the basic bidding value according to the increasing direction of the comprehensive penalty item to obtain the candidate order bidding value, and establishes a link between the candidate order bidding value, the basic bidding value, and the comprehensive penalty item to obtain the candidate order result.

[0189] The system constructs candidate vehicle task bundles based on candidate order results. Orders appearing simultaneously under at least two vehicle identifiers are extracted from these bundles to obtain a set of conflicting orders. Vehicle task pairings containing conflicting orders are written into the conflict pairing record. The system determines the consensus order of conflicting orders according to the following order order drift contribution value (from largest to smallest), order coupling strength (from smallest to largest), and order urgency (from smallest to largest). The system then compares the competitive value of candidate orders for each vehicle task pairing corresponding to the same conflicting order, identifying the vehicle task pairing with the highest competitive value as the retained pairing. The system deletes vehicle task pairings not identified as retained pairings and their corresponding candidate insertion position records, resulting in a candidate vehicle task bundle to be verified. If no conflicting orders exist, the candidate vehicle task bundle to be verified is identified as the resolution candidate vehicle task bundle. The system merges the resolution candidate vehicle task bundle with the stable vehicle task bundle, verifies it using overlapping time periods and order pickup / delivery node time windows to form the target vehicle task bundle, and outputs the logistics transportation collaborative management results.

[0190] Table 1 Comparison of Implementation Results of Multi-Task Collaborative Management and Control in Logistics Transportation Processes

[0191] Sample batch Before average arrival time deviation processing After average arrival time deviation processing Average site wait time before processing Average site wait time after processing Before handling conflicting order quantities After handling conflicting order quantities Before route deviation processing After route deviation mileage processing Task bundle reconstruction order quantity Batch 1 28.4 minutes 17.6 minutes 22.1 minutes 14.3 minutes 9 2 46.5 km 29.2 km 11 Batch 2 31.7 minutes 19.8 minutes 25.4 minutes 15.9 minutes 11 3 52.8 km 34.1 km 13 Batch 3 26.9 minutes 16.2 minutes 20.7 minutes 13.5 minutes 8 2 43.6 km 28.7 km 10 Batch 4 34.2 minutes 21.5 minutes 27.9 minutes 17.8 minutes 12 4 58.3 km 39.6 km 15 Batch 5 29.6 minutes 18.1 minutes 23.5 minutes 14.8 minutes 10 3 49.7 km 31.4 km 12

[0192] As shown in Table 1, the average arrival time deviation, average station waiting time, number of conflicting orders, and route deviation mileage all decreased after processing across the five sample batches. The average arrival time deviation in batch 4 decreased from 34.2 minutes to 21.5 minutes, indicating that drift trigger identification based on the single-value deviation intensity sequence can pinpoint the transportation periods requiring reconstruction and reduce delay transmission by separating stable vehicle task bundles from the order task segments to be reconstructed. The number of conflicting orders in batch 2 decreased from 11 to 3, indicating that after forming a consensus order for conflicting orders based on order drift contribution value, order coupling strength, and order urgency, local consensus resolution can reduce redundant competition. The route deviation mileage in batches 1, 3, and 5 all decreased, indicating that after incorporating route overlap and loading / unloading station connection relationships into the order coupling strength calculation, the reconstructed order task set covers both deviation orders and adjacent coupled orders.

[0193] The results of this embodiment show that the present invention can convert multiple types of execution deviations in the logistics transportation process into a single-value deviation intensity sequence, and form a window truncation position, effective window length, and drift trigger mark through the ADWIN algorithm. After drift triggering, the present invention can distinguish between stable vehicle task bundles and order task segments to be reconstructed, and form a set of reconstructed order tasks by combining the order drift contribution value and the order coupling strength. The improved CBBA algorithm generates a resolution candidate vehicle task bundle through local bidding task sequence, basic bidding value, comprehensive penalty term, and local consensus resolution, so that the logistics transportation collaborative management results can adapt to the changes in the transportation process under the dynamic coupling scenario of multiple tasks.

[0194] 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 multi-task collaborative management and control method for logistics transportation processes based on deep learning, characterized in that, include: Collect basic logistics and transportation operation data, order pickup and delivery node time windows, route overlap relationships, and loading and unloading station connection relationships to construct vehicle main capacity records, order urgency records, and vehicle task execution time periods; Historical execution deviations are extracted from basic logistics and transportation operation data and processed to obtain a single-value deviation intensity sequence. The ADWIN algorithm is used to process the single-valued bias intensity sequence to obtain the window truncation position, effective window length, and drift trigger flag; When the drift trigger flag is set, the reserved transportation period is determined based on the window truncation position and the effective window length. The start and end times of the order pickup and delivery node time windows, the vehicle task bundle execution period and the reserved transportation period are compared respectively to obtain the stable vehicle task bundle and the order task segment to be reconstructed. The order drift contribution value is determined based on the cumulative difference in the single-value deviation strength of the same order before and after the window truncation position. The order coupling strength is calculated based on the route overlap relationship, loading and unloading station connection relationship and order urgency record. The task segment of the order to be reconstructed is expanded to obtain the set of reconstructed order tasks. The reconstructed order task set is written into the improved CBBA algorithm, a basic competitive value is constructed based on the vehicle's main capability record, and a comprehensive penalty term is written into the basic competitive value to obtain candidate order results. Based on the candidate order results, a candidate vehicle task bundle is constructed. The consensus order of conflicting orders is determined according to the order drift contribution value, order coupling strength and order urgency. Local consensus resolution is then performed on the candidate vehicle task bundle to obtain the resolved candidate vehicle task bundle. The candidate vehicle task bundles are merged with the stable vehicle task bundles to obtain the logistics transportation collaborative management results.

2. The deep learning-based multi-task collaborative management and control method for logistics transportation processes according to claim 1, characterized in that, The construction of vehicle main capability records, order urgency records, and vehicle task bundle execution time periods are specifically as follows: By collecting logistics transportation orders, vehicle and driver status, vehicle remaining load, estimated arrival time and actual arrival time according to order identifiers, basic logistics transportation operation data is obtained. Write the order pickup and delivery node time window to the corresponding order identifier to obtain the order time window index record; Based on the vehicle and driver status, vehicle remaining load, and order window index records, candidate matching is performed on the orders that each vehicle can accept, and candidate matching records are obtained. Write the route overlap relationship and loading / unloading station connection relationship into the candidate matching record, and sort them according to the length of the overlapping road segment and the order of the loading / unloading station connection between the candidate orders corresponding to the same vehicle to obtain the candidate task connection order; Based on the candidate task connection order, candidate orders with vehicle remaining load capacity less than the candidate order load capacity requirement and vehicle and driver status not in a dispatchable state are removed to obtain the vehicle main capacity record. Based on the order pickup and delivery node time windows, estimated arrival time and actual arrival time, calculate the remaining time of pickup node, remaining time of delivery node and arrival deviation time respectively, and calculate the connection waiting time according to the connection order of loading and unloading stations of adjacent candidate orders in the candidate task connection priority. The smaller of the remaining time at the pickup node and the remaining time at the delivery node is taken as the time window remaining benchmark. Candidate orders are sorted according to the time window remaining benchmark, arrival deviation time, and connection waiting time to obtain the order urgency record. Based on the vehicle's main capability record and the order urgency record, the execution time of the candidate tasks is reorganized to obtain the vehicle task bundle execution time.

3. The deep learning-based multi-task collaborative management and control method for logistics transportation processes according to claim 1, characterized in that, The process of extracting and processing historical execution deviations from basic logistics and transportation operation data specifically involves: Historical execution deviations are extracted from basic logistics and transportation operation data according to order identifiers and execution times. The historical execution deviations are then linked with the order pickup and delivery node time windows, route overlap relationships, and vehicle task execution periods to obtain deviation extraction index records. Based on the deviation extraction index records, the arrival time deviation, station waiting deviation, load occupancy deviation, and route deviation deviation are determined respectively, and written into the same deviation item set according to the corresponding order identifier and deviation type; The set of deviation terms is unified into positive values ​​according to the direction of increasing deviation, thus obtaining the set of positive deviation terms; Based on the maximum and minimum deviation values ​​of the same deviation item in historical execution deviations, the set of positive deviation items is normalized using the same scale to obtain the normalized set of deviation items. Based on the order urgency record, determine the order in which the deviations corresponding to each order in the normalized deviation item set are written, and perform sequential arrangement on the normalized deviation item set to obtain the candidate sequence of deviation intensity; Establish deviation weight rankings according to deviation type, and perform forward shift correction on the deviation weight rankings corresponding to the same order based on the order urgency record to obtain deviation weight marking results; The candidate sequence of deviation intensity is rearranged according to the execution time, and the normalized deviation terms corresponding to the same execution time are weighted and summarized according to the deviation weight labeling results to obtain the single-value deviation intensity sequence.

4. The deep learning-based multi-task collaborative management and control method for logistics transportation processes according to claim 1, characterized in that, The ADWIN algorithm is used to process the single-valued bias intensity sequence, specifically as follows: The single-value deviation intensity sequence is written into the adaptive detection window of the ADWIN algorithm according to the order of execution time to obtain the candidate detection window; Based on the execution time sequence and the deviation corresponding to the order urgency record in the candidate detection window, write the order change position to generate a candidate segmentation position set; The candidate detection window is divided into a front deviation window and a back deviation window based on the candidate segmentation positions in the candidate segmentation position set, and the mean difference and fluctuation difference between the windows are calculated to obtain the candidate drift judgment record. From the candidate drift determination records, select candidate segmentation positions where the mean of the latter segment deviation window is higher than the mean of the former segment deviation window and the fluctuation difference reaches the preset fluctuation threshold, and obtain the effective segmentation positions. The execution times corresponding to the effective segmentation positions are compared, and the effective segmentation position with the largest execution time is determined as the window truncation position. The single-value deviation intensity sequence after the window truncation position is retained to obtain the effective window length. A drift trigger flag is generated when the effective window length reaches the preset detection length and the average value of the subsequent deviation window is continuously higher than the average value of the preceding deviation window.

5. The deep learning-based multi-task collaborative management and control method for logistics transportation processes according to claim 1, characterized in that, The process of obtaining the stable vehicle task bundle and the order task segment to be reconstructed is as follows: When the drift trigger flag is set, the last execution time of the retained part in the single-value deviation intensity sequence is determined according to the effective window length, and the time period between the execution time corresponding to the window truncation position and the last execution time is determined as the retained transportation period. Extract the start and end times of each vehicle task bundle from the vehicle task bundle execution time period according to the vehicle identification, and compare them with the start and end times of the retained transportation time period to obtain the task bundle time period comparison record. Extract the start and end times of the pickup node and the start and end times of the delivery node from the order pickup and delivery node time window according to the order identifier, and compare them with the start and end times of the retained transportation period to obtain the order time window comparison record; Based on the task bundle time period comparison record, vehicle task bundles whose task start time and task end time are both within the retained transportation time period are retained to obtain candidate stable vehicle task bundles. Based on the order time window comparison record, extract orders whose pickup node start and end times or delivery node start and end times cross the boundary of the reserved transportation period to obtain candidate reconstructed orders; Remove the task segments containing candidate reconstruction orders from the candidate stable vehicle task bundle to obtain the stable vehicle task bundle, and merge the removed task segments with the candidate reconstruction orders to obtain the task segments to be reconstructed.

6. The deep learning-based multi-task collaborative management and control method for logistics transportation processes according to claim 1, characterized in that, The obtained reconstructed order task set is specifically as follows: The single-value deviation intensity sequence is divided into a pre-truncation deviation segment and a post-truncation deviation segment according to the window truncation position, and the cumulative value of single-value deviation intensity in the pre-truncation deviation segment and the post-truncation deviation segment is calculated according to the order identifier. The cumulative difference of single-value deviation intensity is calculated based on the cumulative value of single-value deviation intensity in the pre-truncation deviation segment and post-truncation deviation segment corresponding to the same order, and the order drift contribution value is determined. Extract the length of overlapping road segments between the task segment to be reconstructed and adjacent orders based on the route overlap relationship, and extract the connection waiting time between corresponding adjacent orders based on the connection relationship of loading and unloading stations; Based on the length of overlapping road segments, connection waiting time, and order urgency, adjacent orders are sorted in a progressive manner, and the sorted position is used as the order coupling strength. Orders whose drift contribution value is greater than the preset contribution threshold are identified as drift expansion orders, and adjacent orders whose order coupling strength corresponds to a sorting position not greater than the preset coupling position threshold are identified as coupling expansion orders. Write the drift-extended orders and coupled-extended orders into the task segment of the orders to be reconstructed, and delete duplicate orders according to the order identifier to obtain the set of reconstructed order tasks.

7. The deep learning-based multi-task collaborative management and control method for logistics transportation processes according to claim 1, characterized in that, The improved CBBA algorithm is as follows: The task set for reconstructing orders is split into task items according to the order identifier, and the order drift contribution value, order coupling strength and order urgency are recorded and written into the corresponding task items to obtain the local bidding task sequence. Based on the vehicle's main capabilities record, establish vehicle task pairing relationships between each task item in the local bidding task sequence and the vehicles that can undertake the task, thereby obtaining the local reconstruction bidding domain. Extract the occupied task time slots and assigned order order sequence from the stable vehicle task bundle according to the vehicle identifier, and determine the candidate insertion positions corresponding to the vehicle task pairing relationship in the local reconstruction bidding domain according to the order pickup and delivery node time window to obtain the candidate insertion position record; Based on the vehicle's main capability record, candidate insertion position record, and local bidding task sequence, perform basic bidding sorting on the vehicle task pairing relationship, and write the sorted result into the basic bidding value. The load constraint penalty is calculated based on the comparison between the load requirement of the candidate order and the remaining load of the vehicle, and the pickup and delivery time window default penalty is calculated based on the time difference between the expected arrival time corresponding to the candidate insertion position and the order pickup and delivery node time window. The driving route deviation penalty is calculated based on the length of the road segment between the candidate order and the adjacent orders before and after the candidate insertion position that do not fall into the route overlap relationship, and the vehicle task switching penalty is calculated based on the number of times the inserted candidate order causes changes in the order of the assigned orders. Calculate the drift coupling penalty based on the order drift contribution value corresponding to the candidate order and the order coupling strength between the candidate order and the adjacent orders before and after the candidate insertion position; The load constraint penalty, delivery window violation penalty, route deviation penalty, vehicle task switching penalty, and drift coupling penalty are combined into a comprehensive penalty item and written into the basic competitive value; The candidate order competitive value is obtained by deducting the base competitive value according to the direction of increasing comprehensive penalty items; Based on the vehicle identifier, order identifier, and candidate insertion position record, the candidate order competitive value, basic competitive value, and comprehensive penalty item are linked to obtain the candidate order result.

8. The deep learning-based multi-task collaborative management and control method for logistics transportation processes according to claim 1, characterized in that, The process of obtaining the candidate vehicle task bundle is as follows: Extract the competitive value of candidate orders from the candidate order results according to the vehicle identifier, sort the candidate orders corresponding to the same vehicle, and write the sorted candidate orders into the task bundle of the corresponding vehicle according to the candidate insertion position record to obtain the candidate vehicle task bundle. Extract orders that appear simultaneously under at least two vehicle identifiers from the candidate vehicle task bundle to obtain a set of conflicting orders, and write the vehicle task pairing relationship containing conflicting orders into the conflict pairing record; Based on the order drift contribution value, order coupling strength and order urgency records corresponding to conflicting orders, the conflicting order set is sorted in a progressive manner to obtain the consensus order of conflicting orders; Conflicting orders are extracted sequentially according to the consensus order. Based on the conflict pairing records, the competitive value of candidate orders corresponding to the vehicle task pairing relationship of the same conflicting order is compared, and the pairing relationship to be retained is determined. Remove vehicle task pairings that are not determined to be retained from the candidate vehicle task bundle, and simultaneously delete the corresponding candidate insertion position records to obtain the candidate vehicle task bundle to be verified. The remaining candidate orders in the candidate vehicle task bundle to be verified are re-executed to extract the conflict order set. If there are no conflict orders, the candidate vehicle task bundle to be verified is determined as the candidate vehicle task bundle to be resolved.

9. The deep learning-based multi-task collaborative management and control method for logistics transportation processes according to claim 1, characterized in that, The merging of the candidate vehicle task bundle with the stable vehicle task bundle specifically involves: According to the vehicle identifier, the order identifier, task start time, task end time and candidate insertion position record are extracted and decrypted from the candidate vehicle task bundle and the stable vehicle task bundle, and a stable source mark and a candidate source mark are established for each task segment to obtain the vehicle task merging base sequence. Write the task segments in the stable vehicle task bundle and the task segments in the candidate vehicle task bundle into the vehicle task merging base sequence according to the vehicle identifier to obtain the merged task bundle to be verified. The start and end times of adjacent task segments corresponding to the same vehicle in the task bundle to be checked and merged are compared to obtain the time period overlap record. Based on the overlapping time period records, extract the overlapping task segment pairs. When the task segment pair contains both stable source markers and candidate source markers, retain the task segment corresponding to the stable source marker and mark the task segment corresponding to the candidate source marker as the task segment to be adjusted. When all task segments contain candidate source tags, compare the competitive value of candidate orders for the corresponding task segments, retain the task segments with higher competitive value, and mark the unretained task segments as task segments to be adjusted. Based on the candidate insertion position record, re-match unoccupied candidate insertion positions for the task segment to be adjusted in the task bundle to be verified and merged to obtain the corrected merge task bundle; Based on the order pickup and delivery node time window verification and correction, merge the task segments in the task bundle to obtain the target vehicle task bundle; The target vehicle task bundle is recorded and output according to vehicle identifier, order identifier, task start time, task end time and candidate insertion position, to obtain the logistics transportation collaborative management and control results.

Images