An adaptive logistics task allocation method for multi-warehouse collaboration

Abstract

The present invention discloses an adaptive logistics task allocation method for multi-warehouse collaboration, which specifically relates to the field of task allocation in logistics transportation. The method comprises the following steps: constructing a behavioral deviation map of logistics warehouses in a logistics alliance, solving the path inertia function and the response rigidity vector, and generating a response model baseline for logistics warehouses under a non-profit orientation; generating a path inertia superposition risk value based on the fusion of the path repeatability matching value and the response model baseline, and using it as a sorting intervention factor to perform dynamic scheduling weight adjustment; constructing a rejection chain structure and a rejection behavior frequency table based on the rejection behavior of each task, and forming a responsibility-sensitive sorting function for sorting penalties. By constructing a multi-dimensional scheduling weight dynamic adjustment mechanism that integrates the warehouse's natural response behavior, path preference trend, rejection behavior record, strategy imitation degree, and fulfillment performance, the problem of scheduling imbalance and long-term delay in the transportation of some orders caused by the alliance warehouse's tendency to accept orders can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of task allocation for logistics transportation, and more specifically, to an adaptive logistics task allocation method for multi-warehouse collaboration. Background Art

[0002] In a multi-brand logistics alliance, warehouses from different companies participate in the unified scheduling of orders. This may seem like resource sharing, but problems can easily arise in actual operations.

[0003] In practice, we've found that each warehouse is more willing to take orders with high profits and simple delivery, while being indifferent to low-profit, complex delivery tasks. The end result is a decline in overall delivery efficiency, increasingly unbalanced platform scheduling, and even orders that remain unattended for extended periods of time.

[0004] Therefore, in an alliance consisting of multiple corporate warehouses, how to ensure that each warehouse can share tasks fairly is an urgent problem that needs to be solved. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an adaptive logistics task allocation method for multi-warehouse collaboration. By constructing a multi-dimensional scheduling weight dynamic adjustment mechanism that integrates the natural response behavior of warehouses, path preference trends, rejection behavior records, strategy imitation degree and fulfillment performance, it solves the problem of scheduling imbalance and long-term delay of some orders caused by the tendency of alliance warehouses to accept orders proposed in the above-mentioned background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive logistics task allocation method for multi-warehouse collaboration, comprising the following steps:

[0007] S1: By constructing a behavioral deviation map of logistics warehouses in the logistics alliance, solving the path inertia function and response rigidity vector, and generating a response model baseline of logistics warehouses under non-profit orientation;

[0008] S2: Generate the path inertia superposition risk value based on the fusion of the path repeatability matching value and the response model baseline, and use it as a ranking intervention factor to perform dynamic scheduling weight adjustment;

[0009] S3: Based on the rejection behavior of each task, a rejection chain structure and a rejection behavior frequency table are constructed to form a responsibility-sensitive ranking function for sorting penalties;

[0010] S4: Using the similarity of strategy feature vectors to construct a strategy imitation trend map between warehouses, calibrate the spreading risk boundary through the diffusion coefficient, and perform sequenced interruption intervention on the starting warehouse;

[0011] S5: By integrating the performance status, a responsibility performance vector is constructed, and the initial ranking weight and alliance credit record are updated based on the incentive adaptation weight to achieve two-way binding of performance incentives.

[0012] In a preferred embodiment, S1 further includes: extracting historical order record data of each warehouse from the logistics task log of the logistics alliance, the historical order record data including task path, task response delay, and inventory flow record, and the historical order record data serving as the warehouse behavior basic data set;

[0013] A structured classification process is performed on the basic warehouse behavior dataset. The volatility of the path direction distribution, the spatial gradient characteristics of the inventory flux density field, and the response frequency variation trend are sequentially solved and defined as a warehouse behavior indicator triple. The warehouse behavior indicator triple is input into the graph construction process. Combined with the structural path of the task execution graph, a warehouse behavior deviation map is established. The warehouse behavior deviation map is used to reflect the natural response differences of the warehouse to different types of tasks.

[0014] Based on the node connection strength of the warehouse behavior deviation graph, the warehouse path inertia function is derived, and combined with the response period distribution to form a response rigidity vector, which serves as the baseline of the warehouse response model under non-profit constraints.

[0015] In a preferred embodiment, S2 further includes: extracting a path trajectory set for each warehouse from historical order record data, and extracting a path attribute set for the current scheduling task from the current set of tasks to be assigned, then performing path vectorization processing on the warehouse path trajectory set and the path set of the current scheduling task, respectively, constructing a standard vector structure representing the direction of the historical path and the current path, and calculating the path repetition matching value between each warehouse and the current scheduling task set based on the path direction overlap; statistically calculating the path selection concentration trend based on the matching value distribution of each warehouse, and finally generating a path aggregation density index representing the degree of path repetition locking risk;

[0016] The path aggregation density index is vector-fused with the path inertia function in the warehouse response model baseline to output the path inertia superposition risk value, which is used to measure the repeated path locking tendency.

[0017] The path inertia superposition risk value is injected into the scheduling and sorting model as a dynamic intervention factor, and weight mapping is performed in conjunction with the current task value vector to generate a matching sorting table from the initial task to the warehouse.

[0018] In a preferred embodiment, S3 further includes: setting a unique task identifier for each task to be assigned, and establishing a corresponding rejection chain structure accordingly, wherein each node in the rejection chain structure records the warehouse number, response status, and response time corresponding to the task being rejected, forming a warehouse response behavior track for tracing;

[0019] During the task allocation process, if a scheduling failure or response timeout occurs, the rejection behavior corresponding to the task is appended to the rejection chain structure of the task in chronological order, and expanded in round order to form a rejection behavior chain list;

[0020] Aggregate the rejection chain structures of all tasks by task ID, use warehouse number as index to count the cumulative frequency of rejection behavior, and construct a rejection behavior frequency table based on warehouse to characterize the historical density characteristics of rejection responses;

[0021] The rejection impact factor of each warehouse is determined based on the rejection behavior frequency table, and the rejection impact factor is included in the scheduling ranking weight calculation. By constructing a responsibility-sensitive ranking function, its ranking priority is dynamically lowered in the next round of task allocation to achieve punitive ranking adjustment.

[0022] In a preferred embodiment, S4 further includes: extracting task information received by each warehouse in a similar scheduling cycle from the task allocation history, extracting three types of task attributes based on the profit range, path direction attribute, and trigger response time period corresponding to each task, and encoding the three types of task attributes to form a warehouse scheduling strategy feature vector set as the strategic behavior expression of each warehouse in the current cycle;

[0023] The warehouse scheduling strategy vector set is used as input to perform similarity comparison. Based on the similarity distribution between the vectors, a strategy imitation trend map between warehouses is constructed. In the strategy imitation trend map, strategy trajectory paths with similarity scores exceeding a preset similarity score threshold are marked as high-risk imitation paths.

[0024] Calculate the strategy diffusion coefficient for each high-risk imitation path. The strategy diffusion coefficient is measured based on the number of warehouses in the imitation path, the degree of strategy update synchronization, and the duration. Paths where the strategy diffusion coefficient exceeds the intervention threshold are marked as the spreading risk boundary.

[0025] Execute strategic intervention actions on the starting warehouse node in the spreading risk boundary, set the corresponding scheduling penalty coefficient according to its imitation propagation level, and use the scheduling penalty coefficient in the scheduling sorting weight adjustment process to dynamically suppress the behavior pattern replication trend.

[0026] In a preferred embodiment, S5 further includes: after each scheduling cycle ends, extracting three task completion attributes, namely, response success rate, low-profit task completion ratio, and path selection diversity, based on the scheduling response results of tasks completed by each warehouse in the corresponding scheduling cycle, and integrating the three task completion attributes by warehouse dimension to generate a responsibility fulfillment vector;

[0027] The responsibility fulfillment vector is used as input data and mapped to a preset set of incentive factors. Based on the weights of the incentive factors in the incentive factor set on different fulfillment dimensions, the incentive adaptation weight allowed for each warehouse in the next scheduling cycle is calculated.

[0028] The incentive adaptation weight is injected into the task scheduling and sorting process of the next cycle as the initial sorting weight, and the incentive adaptation weight corresponding to each warehouse is written into the cycle credit record table to update the warehouse's credit status information within the alliance.

[0029] Technical effects and advantages of the present invention:

[0030] By constructing a response model baseline that includes a path inertia function and a response rigidity vector, the present invention reflects the warehouse's natural task response behavior under non-profit orientation, thereby suppressing the unbalanced tendency of prioritizing high-profit tasks in the joint transportation scheduling of multi-enterprise warehouses and ensuring fair distribution of low-profit tasks.

[0031] The present invention extracts the overlap trend between the warehouse path trajectory and the current task path, and combines it with the response model to generate the path inertia superposition risk value, dynamically identifying the degree of path duplication preference, thereby achieving early intervention in high-frequency path locking behavior in scheduling and sorting, and avoiding excessive resource concentration;

[0032] This paper constructs a task rejection chain structure, counts the behavioral trajectory and frequency of task rejection, and derives the rejection impact factor by combining task level and response delay, thereby achieving a quantitative expression of the warehouse's degree of responsibility and implementing responsible punishment and adjustment for negative response behaviors during the sorting process.

[0033] By constructing strategy feature vectors and analyzing their similarity distribution across warehouses, we generate a strategy imitation trend map. We then combine the strength of the strategy diffusion path to construct a risk boundary, thereby identifying and intervening in convergent trends in inter-warehouse scheduling behaviors and improving response diversity.

[0034] By integrating the fulfillment results of warehouses within a cycle, a fulfillment vector based on response success rate, proportion of low-profit tasks and path diversity is constructed, and incentive adaptation weights are introduced to update the ranking weights, thereby realizing the integrated closed loop of positive feedback on fulfillment and dynamic incentive mechanism in task scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flow chart of the steps of the present invention. DETAILED DESCRIPTION

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0037] Refer to the instruction manual Figure 1 An adaptive logistics task allocation method for multi-warehouse collaboration according to an embodiment of the present invention includes the following steps:

[0038] S1: By constructing a behavioral deviation map of logistics warehouses in the logistics alliance, solving the path inertia function and response rigidity vector, and generating a response model baseline of logistics warehouses under non-profit orientation;

[0039] S2: Generate the path inertia superposition risk value based on the fusion of the path repeatability matching value and the response model baseline, and use it as a ranking intervention factor to perform dynamic scheduling weight adjustment;

[0040] S3: Based on the rejection behavior of each task, a rejection chain structure and a rejection behavior frequency table are constructed to form a responsibility-sensitive ranking function for sorting penalties;

[0041] S4: Using the similarity of strategy feature vectors to construct a strategy imitation trend map between warehouses, calibrate the spreading risk boundary through the diffusion coefficient, and perform sequenced interruption intervention on the starting warehouse;

[0042] S5: By integrating the performance status, a responsibility performance vector is constructed, and the initial ranking weight and alliance credit record are updated based on the incentive adaptation weight to achieve two-way binding of performance incentives.

[0043] S1 also includes: extracting historical order record data for each warehouse from the logistics alliance's logistics task log. The historical order record data includes task paths, task response delays, and inventory flow records. The historical order record data serves as the basic dataset of warehouse behavior;

[0044] A structured classification process is performed on the basic warehouse behavior dataset. The volatility of the path direction distribution, the spatial gradient characteristics of the inventory flux density field, and the response frequency variation trend are sequentially solved and defined as a warehouse behavior indicator triple. The warehouse behavior indicator triple is input into the graph construction process. Combined with the structural path of the task execution graph, a warehouse behavior deviation map is established. The warehouse behavior deviation map is used to reflect the natural response differences of the warehouse to different types of tasks.

[0045] Based on the node connection strength of the warehouse behavior deviation graph, the warehouse path inertia function is derived, and combined with the response period distribution to form a response rigidity vector, which serves as the baseline of the warehouse response model under non-profit constraints.

[0046] It should be further explained that in the formula structure involved in this scheme, dimensionless terms can serve as proportionality or structural adjustment factors. When combined with quantities with units, they only play a numerical scaling role and do not introduce new physical dimensions. Therefore, they will not change or confuse the overall unit system of expression. This combination of "dimensionless terms and units" can be understood as a composite structural expression commonly used in mathematical and physical modeling, conforming to the principle of dimensional consistency and having a clear physical interpretation basis.

[0047] Secondly, in the formula structure of this scheme, if multiple variables with different physical units are involved, including but not limited to time, mass, or energy variables, their joint appearance is to express the collaborative modeling relationship of multiple physical mechanisms. Each variable is formed into a unified structure through function mapping, ratio combination, or normalization adjustment. The units and meanings are clear, and the overall expression conforms to the principle of dimensional consistency and the common formula of engineering modeling.

[0048] The "S1 also includes" solution involves extracting historical order record data from logistics task logs, sequentially solving the volatility of path direction distribution, the spatial gradient characteristics of the inventory flux density field, and the response frequency variation trend, and constructing a triplet of warehouse behavior indicators. Based on the behavior deviation map, the path inertia function and response rigidity vector are derived to construct a response model baseline under non-profit orientation. ;

[0049] in For warehouse The response model baseline, [unit: ]; For warehouse In time The path direction fluctuation function, [unit: number of path segments, segments]; Observation period for modeling behavior, [unit: hours]; is the stock flux density function, [unit: ]; is the stock flux direction vector field, [unit: direction vector, dimensionless]; is the area of inventory operation area, [unit: ]; is the function of the number of responses per unit time, unit: [times / hour]; is the path scale normalization constant, which is defined as the normalization factor of the maximum number of path segments, unit: [segment];

[0050] For the first term of the response model baseline is the path inertia term, [unit: ], which is achieved by integrating the square of the second-order rate of change of the path;

[0051] Item 2 is the inventory flux intensity term, with the unit of [pieces / hour]; in order to maintain the same dimension as the path term, a normalized scale is used It cancels out the inventory density dimension;

[0052] The third exponential part refers to the response frequency fluctuation term, which is dimensionless;

[0053] The overall unit closure results in the response model baseline are: ;

[0054] The formula of the response model baseline is constructed to extract the response tendency under non-profit conditions from the warehouse behavior. First, the path function The second-order derivative reflects the stability of path selection, and its integral reciprocal represents the inertia strength. The inventory gradient multiplied by the direction vector represents the local inventory flow intensity, and the integral reflects the overall activity. In order to avoid unit mismatch, the inventory flux term is introduced Normalize the path scale; the response frequency fluctuation term decays exponentially, making the response more unstable and the overall baseline lower; the final output It represents the multiplicative effect of responsiveness, path inertia, and inventory activity, which is also the baseline value used for risk suppression and capacity guidance in scheduling.

[0055] S2 also includes: extracting a path trajectory set for each warehouse from historical order record data, and extracting a path attribute set for the current scheduling task from the current set of tasks to be assigned, then performing path vectorization processing on the warehouse path trajectory set and the path set of the current scheduling task, respectively, to construct a standard vector structure representing the direction of the historical path and the current path, and calculating the path repetition matching value between each warehouse and the current scheduling task set based on the path direction overlap; statistically analyzing the path selection concentration trend based on the matching value distribution of each warehouse, and finally generating a path aggregation density index representing the degree of path repetition locking risk;

[0056] The path aggregation density index is vector-fused with the path inertia function in the warehouse response model baseline to output the path inertia superposition risk value, which is used to measure the repeated path locking tendency.

[0057] The path inertia superposition risk value is injected into the scheduling and sorting model as a dynamic intervention factor, and weight mapping is performed in conjunction with the current task value vector to generate a matching sorting table from the initial task to the warehouse.

[0058] The solution based on "S2 also includes" includes: extracting the historical path direction vector of each warehouse from the historical path direction field, calculating the direction coincidence integral and aggregate gradient with the current scheduling task path direction, and combining it with the response model baseline in S1 to generate the path inertia superposition risk value as the sorting intervention factor:

[0059] ;

[0060] in is the path inertia superposition risk value, unit: [ ]; is the historical path direction vector field, unit: [dimensionless direction vector]; The direction vector of the current task path, unit: [dimensionless direction vector]; Defines the spatial domain for the path, units: [ ]; is the direction vector gradient tensor field, unit: [1 / m]; Generated by the formula in S1, unit: [ ];

[0061] for It should be noted that the first term is the normalized path direction coincidence integral term, and the dimension is [dimensionless]; the second term is the square root of the path gradient modulus integral, and the dimension is , is dimensionless; the third term is , the unit is [ ]; the overall unit is [ ];

[0062] in The core of the formula is to comprehensively evaluate the coincidence trend of warehouse path selection and path distribution density, expressed as a risk intervention factor. The first term captures the path reuse trend by integrating the dot product of two direction fields and normalizing it; the second term calculates the square of the rate of change of the path direction field and calculates the root mean square to reflect the gradient strength of path aggregation; the two terms are combined to express the path inertia strength; and finally multiplied by the response model baseline , to achieve the fusion judgment of response capability and path tendency, as Applied to scheduling sorting weight adjustment to achieve early intervention in path competition behavior.

[0063] S3 also includes: setting a unique task identifier for each task to be assigned, and establishing a corresponding rejection chain structure based on the identifier. Each node in the rejection chain structure records the warehouse number, response status, and response time corresponding to the task being rejected, forming a warehouse response behavior track for traceability;

[0064] During the task allocation process, if a scheduling failure or response timeout occurs, the rejection behavior corresponding to the task is appended to the rejection chain structure of the task in chronological order, and expanded in round order to form a rejection behavior chain list;

[0065] Aggregate the rejection chain structures of all tasks by task ID, use warehouse number as index to count the cumulative frequency of rejection behavior, and construct a rejection behavior frequency table based on warehouse to characterize the historical density characteristics of rejection responses;

[0066] The rejection impact factor of each warehouse is determined based on the rejection behavior frequency table, and the rejection impact factor is included in the scheduling ranking weight calculation. By constructing a responsibility-sensitive ranking function, its ranking priority is dynamically lowered in the next round of task allocation to achieve punitive ranking adjustment.

[0067] The "S3 also includes" solution includes: recording the warehouse's rejection response to tasks during the scheduling process based on a rejection chain structure, building a dynamic and normalized rejection penalty score based on task scheduling urgency, task level, and scheduling density, and generating the warehouse's rejection impact factor: ;

[0068] in For warehouse The rejection impact factor, [unit: dimensionless]; The duration of the scheduling evaluation cycle, [unit: hours]; For in time Time Warehouse The number of rejected tasks, [unit: times]; For the task The scheduling urgency factor [unit: dimensionless, defined as the normalized value of the urgency level score]; For the task The cumulative waiting time before being responded to, [unit: hours]; For the task The task level number, [unit: level number, dimensionless (integer value)]; is the mean reference item of the system task level, [unit: level number, dimensionless]

[0069] For the components, The unit is hour / hour = dimensionless; is dimensionless; It has been normalized to dimensionless; The total amount of the item is dimensionless, The unit of is hour, the unit of integral result is [hour]; the overall multiplication ( ) completely cancels out the time dimension; the final result is dimensionless;

[0070] The formula is used to evaluate the impact of the warehouse's rejection behavior on the scheduling task in the current cycle. The penalty term is constructed by combining each rejection with the task urgency, response delay time and task level complexity, and then accumulated over time. Finally, it is normalized by dividing it by the cycle length and output as the rejection impact factor that can be directly called in the scheduling sorting. .

[0071] S4 also includes: extracting task information received by each warehouse in a similar scheduling cycle from the task allocation history, extracting three types of task attributes based on the profit range, path direction attribute, and trigger response time period corresponding to each task, and encoding the three types of task attributes to form a warehouse scheduling strategy feature vector set as the strategic behavior expression of each warehouse in the current cycle;

[0072] The warehouse scheduling strategy vector set is used as input to perform similarity comparison. Based on the similarity distribution between the vectors, a strategy imitation trend map between warehouses is constructed. In the strategy imitation trend map, strategy trajectory paths with similarity scores exceeding a preset similarity score threshold are marked as high-risk imitation paths.

[0073] Calculate the strategy diffusion coefficient for each high-risk imitation path. The strategy diffusion coefficient is measured based on the number of warehouses in the imitation path, the degree of strategy update synchronization, and the duration. Paths where the strategy diffusion coefficient exceeds the intervention threshold are marked as the spreading risk boundary.

[0074] Execute strategic intervention actions on the starting warehouse node in the spreading risk boundary, set the corresponding scheduling penalty coefficient according to its imitation propagation level, and use the scheduling penalty coefficient in the scheduling sorting weight adjustment process to dynamically suppress the behavior pattern replication trend.

[0075] The solution based on "S4 also includes": extracting the scheduling strategy feature vector of each warehouse within the scheduling cycle, comprehensively calculating the strategy diffusion intensity through behavioral similarity, strategy update speed and structural behavior path distance, and outputting the strategy diffusion coefficient: ;

[0076] in For warehouse The strategic diffusion coefficient of [unit: dimensionless]; , is the warehouse scheduling strategy feature vector (composed of three types of task attribute encodings: profit interval, path direction attribute, and trigger response time period), [unit: dimensionless vector]; Representation and Warehouse There is a set of neighboring warehouses with highly similar scheduling behaviors, For collection The number index of The strategy evaluation time window, [unit: hour]; is the rate of change of the strategy vector, [unit: 1 / hour]; is the perturbation constant, [unit: dimensionless], which is used to prevent the denominator from being zero; is the length of the inter-warehouse behavior structure path, [unit: segment]; is the strategy synchronization factor, [unit: dimensionless], which represents the ratio of the update time overlap of the two warehouse scheduling strategies; is the upper limit of points, is the lower limit of integral and the upper limit of integral The length of the sliding window for evaluating the diffusion of scheduling policy behaviors, expressed in hours, and is set to a constant for all warehouses.

[0077] In the units of the above formula:

[0078] The square of the vector difference is dimensionless;

[0079] in the denominator = [1 / hour] × [hour] = dimensionless;

[0080] is the path segment, and the reciprocal is ;

[0081] is dimensionless;

[0082] therefore The unit result in the formula is: dimensionless × dimensionless × [1 / segment] × dimensionless = [1 / segment]. If normalized (i.e., divided by the maximum path length), the dimensionless result is obtained.

[0083] The formula aims to characterize whether there is a tendency for the scheduling strategy to imitate and spread within the alliance; high similarity, synchronous update and structural proximity are the three major channels for strategy diffusion. By combining behavioral similarity (vector difference), strategy stability (integral of change rate) and path connection density, a strategy diffusion score is formed, and finally Used to determine whether to implement scheduling and sorting interruption intervention for the warehouse.

[0084] S5 also includes: after each scheduling cycle ends, extracting three task completion attributes, namely, response success rate, low-profit task completion ratio, and path selection diversity, based on the scheduling response results of each warehouse's completed tasks within the corresponding scheduling cycle, and integrating the three task completion attributes by warehouse dimension to generate a responsibility fulfillment vector;

[0085] The responsibility fulfillment vector is used as input data and mapped to a preset set of incentive factors. Based on the weights of the incentive factors in the incentive factor set on different fulfillment dimensions, the incentive adaptation weight allowed for each warehouse in the next scheduling cycle is calculated.

[0086] The incentive adaptation weight is injected into the task scheduling and sorting process of the next cycle as the initial sorting weight, and the incentive adaptation weight corresponding to each warehouse is written into the cycle credit record table to update the warehouse's credit status information within the alliance.

[0087] The "S5 also includes" solution includes: combining the warehouse's response success rate to scheduling tasks, the completion ratio of low-profit tasks, and the diversity of path selection, constructing a fulfillment vector through time integration, and normalizing it to form a dimensionless incentive adaptation weight: ;

[0088] in For warehouse The incentive adaptation weight of [unit: dimensionless]; is the length of the scheduling evaluation cycle, [unit: hours]; is the frequency of successful task response per unit time, [unit: times / hour]; is the reference maximum response frequency, [unit: times / hour]; is the completion ratio of low-profit tasks, [unit: dimensionless]; is the path diversity factor, [unit: dimensionless], the path diversity factor includes the path coverage; Taking the square root is still dimensionless;

[0089] Composition structure description:

[0090] The integrand unit is: times / hour × dimensionless × dimensionless = [times / hour]; after integration, it is: times; the denominator The unit is hour × times / hour = times; the final unit is: times / times = dimensionless;

[0091] The formula is used to measure the actual fulfillment quality of the warehouse in the current scheduling cycle; the three indicators constitute the core performance characteristics, which are convolved and integrated over time to form a complete cycle score; then normalized with the maximum success frequency to output a dimensionless incentive weight , used to influence the initial sorting weight of the next cycle and synchronously used to update the warehouse reputation factor.

[0092] It should be noted that, in a multi-brand logistics alliance, warehouse resources may superficially achieve synergy. However, in the actual scheduling of specific tasks, since each warehouse belongs to a different enterprise, its scheduling decisions are constrained by its own profit orientation, resulting in structural imbalances in resource allocation. When faced with large-scale parallel task allocation, the alliance platform often finds that some warehouses actively prefer to accept high-profit, low-complexity orders, while passively selecting or even cold-shouldering low-profit or non-optimal routing orders. The long-term existence of this phenomenon can lead to a deterioration in task allocation: high-value orders face competition and low-value orders are delayed for a long time, resulting in a loss of fairness and task acceptance stability in the overall scheduling system. Furthermore, the scheduling system lacks a feedback mechanism, making it impossible to proactively identify and intervene in behavioral imbalances. Therefore, to achieve fair contract fulfillment and responsible order acceptance behavior among warehouses within the alliance, this solution constructs a five-level progressive adaptive task allocation method consisting of response model construction, routing risk suppression, behavioral punishment, imitation trend intervention, and responsibility incentive feedback to achieve the dual goals of improving alliance scheduling fairness and overall responsiveness.

[0093] The goal of S1 is to build a response model baseline as a starting point for scheduling behavior;

[0094] The S1 solution first collects historical order-taking behavior logs from each warehouse in the logistics alliance, extracts its task path characteristics, response delay records, and inventory flow status, and uses this to construct a basic warehouse behavior dataset. After structured classification, this dataset is extracted into three core indicators: the volatility of the path direction distribution, the spatial gradient characteristics of the inventory flux density field, and the response frequency change trend, and then a warehouse behavior indicator triple is constructed. By mapping the triple to the task execution graph, a warehouse behavior deviation map is generated to describe the warehouse's natural response distribution to different types of tasks in a non-profit-oriented situation. On this basis, the path inertia function is extracted based on the second-order derivative of the path direction change, and the response rigidity vector is constructed based on the response frequency change. The two are combined to form the warehouse's response model baseline, which serves as a reference benchmark for the platform to evaluate the scheduling stability and behavior naturalness of the warehouse.

[0095] The goal of S2 is to output risk intervention factors based on the fusion response model of pathway overlap;

[0096] The warehouse's historical path direction field and the current scheduled task's path direction are vectorized, and their direction overlap integral and path aggregation density index are calculated to characterize whether the current task has the risk of "path inertia repetitive lock." This path repeatability index is then integrated with the path inertia function in the response model baseline to output a path inertia superposition risk value. This risk value serves as a dynamic intervention factor in the sorting and scheduling process to identify possible path monopoly behavior in the warehouse in advance. It is then combined with the task value vector to perform weight mapping and output the initial sorting weight to ensure that the warehouse does not continuously seize scheduling priority due to path matching advantages.

[0097] The goal of S3 is to build a rejection impact factor based on responsibility backtracking to achieve sorting and punishment;

[0098] During the scheduling process, a rejection chain structure is constructed for each task, recording the time, status, and warehouse number of each task rejected by a warehouse. These rejection events are appended in rounds, ultimately forming a rejection behavior chain list. After the scheduling cycle ends, the system counts the cumulative number of rejections for each warehouse and calculates its rejection impact factor based on the task's scheduling urgency, response delay time, and task level. This rejection impact factor is injected into the responsibility-sensitive ranking function as a penalty item in the scheduling sorting process, which is used to lower the scheduling priority of warehouses that frequently reject orders or evade responsibility, thereby implementing scheduling penalties for warehouses with deviating behavior.

[0099] The goal of S4 is to globally monitor warehouse strategy imitation behavior and implement trend interruption intervention;

[0100] To prevent the spread of high-yield strategies among warehouses, the platform extracts the strategy vectors (three-dimensional attribute encoding vectors) expressed by warehouse task fulfillment during each scheduling cycle and constructs a strategy similarity map. If it detects that certain warehouses form highly similar strategy paths within a short period of time, it identifies a "strategy imitation trend" and calculates the strategy diffusion coefficient to identify the potential risk boundary for this spread. The system then performs punitive sorting intervention on the warehouse node that initiated the spread, adjusting its scheduling weight to suppress the excessive spread of isomorphic scheduling patterns between warehouses and restore the differentiated responsiveness of warehouses.

[0101] The goal of S5 is to integrate the performance results after the cycle ends and form incentive adaptation weight feedback;

[0102] After each scheduling cycle, the platform conducts a quantitative assessment of all warehouses' fulfillment behavior, extracting three indicators: response success rate, completion rate of low-profit tasks, and path selection diversity, to construct a fulfillment vector. This vector is then matched with a set of preset incentive factors to form a corresponding incentive adaptation weight. This weight is then injected into the scheduling engine as the initial factor for the next cycle's scheduling ranking. Simultaneously, the warehouse's reputation record within the alliance is updated, achieving a positive connection between scheduling incentives and responsibility fulfillment, motivating warehouses to actively participate in scheduling all task types.

[0103] This solution is applicable to distributed warehousing and distribution platforms jointly operated by multiple brands and multiple enterprises, and is particularly suitable for shared platform logistics organizations, such as regional joint distribution alliances, e-commerce platform warehousing and distribution integrated systems, or cross-regional inventory allocation collaborative networks. In such organizational structures, although the participating warehouses share a scheduling platform, inconsistent interests often lead to systematic task bias or performance tilt. Traditional polling, weighted, or static scoring scheduling schemes cannot dynamically identify changing trends in warehouse behavior, nor can they effectively punish order rejection or suppress the spread of strategy imitation. The five-level progressive mechanism proposed in this invention, from response baseline modeling to rejection behavior punishment, to strategy diffusion control and performance incentive feedback, achieves the control goals of scheduling fairness, response diversity, and differentiated behavior between warehouses.

[0104] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An adaptive logistics task allocation method for multi-warehouse collaboration, characterized by: The following steps are involved: S1: By constructing a behavioral deviation map of logistics warehouses in the logistics alliance, solving the path inertia function and response rigidity vector, and generating a response model baseline of logistics warehouses under non-profit orientation; S2: Generate the path inertia superposition risk value based on the fusion of the path repeatability matching value and the response model baseline, and use it as a ranking intervention factor to perform dynamic scheduling weight adjustment; S3: Based on the rejection behavior of each task, a rejection chain structure and a rejection behavior frequency table are constructed to form a responsibility-sensitive ranking function for sorting penalties; S4: Using the similarity of strategy feature vectors to construct a strategy imitation trend map between warehouses, calibrate the spreading risk boundary through the diffusion coefficient, and perform sequenced interruption intervention on the starting warehouse; S5: By integrating the performance status, a responsibility performance vector is constructed, and the initial ranking weight and alliance reputation record are updated based on the incentive adaptation weight to achieve a two-way binding of performance incentives; S1 also includes: extracting historical order record data for each warehouse from the logistics alliance's logistics task log. The historical order record data includes task paths, task response delays, and inventory flow records. The historical order record data serves as the basic dataset of warehouse behavior; A structured classification process is performed on the basic warehouse behavior dataset. The volatility of the path direction distribution, the spatial gradient characteristics of the inventory flux density field, and the response frequency variation trend are sequentially solved and defined as a warehouse behavior indicator triple. The warehouse behavior indicator triple is input into the graph construction process. Combined with the structural path of the task execution graph, a warehouse behavior deviation map is established. The warehouse behavior deviation map is used to reflect the natural response differences of the warehouse to different types of tasks. Based on the node connection strength of the warehouse behavior deviation graph, the warehouse path inertia function is derived, and combined with the response period distribution to form a response rigidity vector, which serves as the baseline of the warehouse response model under non-profit constraints.

2. The method for adaptive logistics task allocation for multi-warehouse collaboration according to claim 1 is characterized by: S2 also includes: extracting a path trajectory set for each warehouse from historical order record data, and extracting a path attribute set for the current scheduling task from the current set of tasks to be assigned, then performing path vectorization processing on the warehouse path trajectory set and the path set of the current scheduling task, respectively, to construct a standard vector structure representing the direction of the historical path and the current path, and calculating the path repetition matching value between each warehouse and the current scheduling task set based on the path direction overlap; statistically analyzing the path selection concentration trend based on the matching value distribution of each warehouse, and finally generating a path aggregation density index representing the degree of path repetition locking risk; The path aggregation density index is vector-fused with the path inertia function in the warehouse response model baseline to output the path inertia superposition risk value, which is used to measure the repeated path locking tendency. The path inertia superposition risk value is injected into the scheduling and sorting model as a dynamic intervention factor, and weight mapping is performed in conjunction with the current task value vector to generate a matching sorting table from the initial task to the warehouse.

3. The method for adaptive logistics task allocation for multi-warehouse collaboration according to claim 2 is characterized by: S3 also includes: setting a unique task identifier for each task to be assigned, and establishing a corresponding rejection chain structure based on the identifier. Each node in the rejection chain structure records the warehouse number, response status, and response time corresponding to the task being rejected, forming a warehouse response behavior track for traceability; During the task allocation process, if a scheduling failure or response timeout occurs, the rejection behavior corresponding to the task is appended to the rejection chain structure of the task in chronological order, and expanded in round order to form a rejection behavior chain list; Aggregate the rejection chain structures of all tasks by task ID, use warehouse number as index to count the cumulative frequency of rejection behavior, and construct a rejection behavior frequency table based on warehouse to characterize the historical density characteristics of rejection responses; The rejection impact factor of each warehouse is determined based on the rejection behavior frequency table, and the rejection impact factor is included in the scheduling ranking weight calculation. By constructing a responsibility-sensitive ranking function, its ranking priority is dynamically lowered in the next round of task allocation to achieve punitive ranking adjustment.

4. The method for adaptive logistics task allocation for multi-warehouse collaboration according to claim 3 is characterized by: S4 also includes: extracting task information received by each warehouse in a similar scheduling cycle from the task allocation history, extracting three types of task attributes based on the profit range, path direction attribute, and trigger response time period corresponding to each task, and encoding the three types of task attributes to form a warehouse scheduling strategy feature vector set as the strategic behavior expression of each warehouse in the current cycle; The warehouse scheduling strategy vector set is used as input to perform similarity comparison. Based on the similarity distribution between the vectors, a strategy imitation trend map between warehouses is constructed. In the strategy imitation trend map, strategy trajectory paths with similarity scores exceeding a preset similarity score threshold are marked as high-risk imitation paths. Calculate the strategy diffusion coefficient for each high-risk imitation path. The strategy diffusion coefficient is measured based on the number of warehouses in the imitation path, the degree of strategy update synchronization, and the duration. Paths where the strategy diffusion coefficient exceeds the intervention threshold are marked as the spreading risk boundary. Execute strategic intervention actions on the starting warehouse node in the spreading risk boundary, set the corresponding scheduling penalty coefficient according to its imitation propagation level, and use the scheduling penalty coefficient in the scheduling sorting weight adjustment process to dynamically suppress the behavior pattern replication trend.

5. The method for adaptive logistics task allocation for multi-warehouse collaboration according to claim 4 is characterized by: S5 also includes: after each scheduling cycle ends, extracting three task completion attributes, namely, response success rate, low-profit task completion ratio, and path selection diversity, based on the scheduling response results of each warehouse's completed tasks within the corresponding scheduling cycle, and integrating the three task completion attributes by warehouse dimension to generate a responsibility fulfillment vector; The responsibility fulfillment vector is used as input data and mapped to a preset set of incentive factors. Based on the weights of the incentive factors in the incentive factor set on different fulfillment dimensions, the incentive adaptation weight allowed for each warehouse in the next scheduling cycle is calculated. The incentive adaptation weight is injected into the task scheduling and sorting process of the next cycle as the initial sorting weight, and the incentive adaptation weight corresponding to each warehouse is written into the cycle credit record table to update the warehouse's credit status information within the alliance.

Images