Logistics waybill allocation method, device, medium and product based on dynamic feature compensation
The logistics waybill allocation method with dynamic feature compensation utilizes a dual-tower neural network to decouple waybill and capacity features, generates dynamic compensation coefficients, optimizes logistics allocation, solves the problems of uneven resource allocation and task starvation, and improves system response performance and allocation efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI GANGSOFT INFORMATION TECH ENG CO LTD
- Filing Date
- 2026-05-06
- Publication Date
- 2026-07-31
AI Technical Summary
Existing logistics distribution systems suffer from local optimum allocation traps, leading to uneven resource allocation and task starvation, making it difficult to adapt to complex and dynamic market environments and affecting the stability of the transportation capacity ecosystem and the robustness of the system.
A logistics waybill allocation method based on dynamic feature compensation is adopted. By decoupling the features of the waybill side and the capacity side through a dual-tower neural network representation structure, an independent embedding space for the waybill side and the capacity side is constructed. The matching relationship is calculated and a basic score is generated. Dynamic compensation coefficients are generated by combining the allocation hindrance feature vector and the gating compensation network, and weighted fusion is performed to optimize the allocation result.
It improves the accuracy of vehicle-cargo matching and the efficiency of feature processing, reduces the complexity of model inference, alleviates the problem of uneven resource allocation, and improves the system's response performance and allocation efficiency in high-concurrency scenarios.
Smart Images
Figure CN122491772A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of transportation technology, and in particular to a logistics waybill allocation method, equipment, medium and product based on dynamic feature compensation. Background Technology
[0002] In the modern logistics and distribution sector, the quality of waybill allocation schemes directly impacts a platform's fulfillment efficiency and the stability of its transportation capacity ecosystem. Early logistics platforms primarily relied on manual dispatching or a simple order-grabbing model. Under manual dispatching, allocation decisions heavily depend on the dispatcher's subjective experience, resulting in slow response times and a lack of unified fairness standards in large-scale concurrent scenarios. The simple order-grabbing model, on the other hand, depends entirely on the response speed of the transportation terminals, easily leading to the loss of core, high-quality capacity and the inability of newly added capacity to survive due to a lack of competitive advantage, ultimately causing an imbalance in the transportation capacity structure.
[0003] With the development of artificial intelligence technology, although some platforms have introduced tiered order dispatching or guaranteed return strategies based on simple logic, these solutions are often rigid, manually set rules that are difficult to adapt to complex and dynamic market environments. Furthermore, existing deep learning-based matching models, in their pursuit of maximizing real-time matching efficiency, result in high-scoring capacity terminals occupying the vast majority of resources, while capacity terminals in the initial or marginal stages face the problem of not being able to obtain tasks for extended periods. This technical flaw of uneven resource allocation not only inhibits the metabolism of the capacity pool but also reduces the system's robustness and global scheduling capabilities under extreme loads. Summary of the Invention
[0004] One objective of this application is to provide a logistics waybill allocation method, device, medium, and product based on dynamic feature compensation, which at least solves the problems of allocation traps in logistics allocation systems that fall into local optima, uneven resource allocation at edge nodes, and task starvation caused by logistics scheduling in the prior art.
[0005] To achieve the above objectives, some embodiments of this application provide the following aspects:
[0006] In a first aspect, some embodiments of this application provide a logistics waybill allocation method based on dynamic feature compensation, the method comprising:
[0007] Obtain the attribute information of the target waybill, and obtain a set of candidate transport terminals for the target waybill based on preset filtering conditions;
[0008] Extract historical transport data of each transport terminal in the candidate transport terminal set, and construct a transport capacity profile feature vector for each transport terminal;
[0009] The attribute information of the target waybill and the feature vector of the capacity profile are respectively embedded and represented to construct the feature vector on the waybill side and the feature vector on the capacity side;
[0010] The feature vectors on the waybill side are input into the pre-trained waybill representation network and mapped to the task embedding space, and the feature vectors on the capacity side are input into the pre-trained capacity representation network and mapped to the capacity embedding space.
[0011] Based on the mapped task embedding space and the capability embedding space, the matching relationship between the transport-side feature vector and the transport capacity-side feature vector is calculated to generate a basic matching score;
[0012] Extract the allocation stagnation feature vector corresponding to each of the transport terminals, and input the allocation stagnation feature vector into a preset gating compensation network for nonlinear mapping to generate dynamic compensation coefficients;
[0013] The basic matching score is weighted and fused based on the dynamic compensation coefficient to generate a comprehensive score.
[0014] The comprehensive score is input into the ranking model to generate the allocation probability of each of the transport capacity terminals, and the candidate transport capacity terminal set is ranked according to the allocation probability.
[0015] Based on the ranking results and combined with the preset exploration strategy, the target transportation terminal is determined from the candidate transportation terminal set, a final dispatch set is generated, and the target waybill is issued based on the final dispatch set.
[0016] Secondly, some embodiments of this application also provide an electronic device, the electronic device comprising: one or more processors; and a memory storing computer program instructions, which, when executed, cause the processor to perform the steps of the method described above.
[0017] Thirdly, some embodiments of this application also provide a computer-readable medium having computer program instructions stored thereon, which can be executed by a processor to implement the method described above.
[0018] Fourthly, some embodiments of this application also provide a computer program product, including a computer program / instructions that, when executed by a processor, implement the method described above.
[0019] Compared with related technologies, the solution provided in this application introduces a dual-tower neural network representation structure, which achieves decoupled modeling of waybill-side features and capacity-side features, significantly improving the accuracy of vehicle-cargo matching and feature processing efficiency. Compared to traditional feature-based allocation methods, which struggle to fully exploit the nonlinear relationships between high-dimensional heterogeneous data, this embodiment maps the attribute information of the target waybill and the capacity profile feature vector to independent task embedding spaces and capability embedding spaces, respectively, and calculates their matching relationship in a continuous vector space to generate a basic matching score, thereby improving the discriminative power of feature representation and the accuracy of matching results.
[0020] Meanwhile, since the features of the transport unit side and the transport capacity side adopt a decoupled encoding method, the distribution interference problem caused by the direct interaction of different types of features in the same space is avoided. When facing a large number of candidate transport capacity terminals, the matching process can be completed efficiently through vectorized calculation, thereby effectively reducing the complexity of model inference and improving the system's response performance in high-concurrency scheduling scenarios.
[0021] Furthermore, by constructing a feature vector representing allocation bottlenecks and introducing a gating compensation network to model the resource acquisition status of capacity terminals, dynamic compensation coefficients are generated. These dynamic compensation coefficients are then applied to the basic matching score through a weighted fusion mechanism. This allows the system to adaptively adjust capacity terminals with insufficient resource acquisition while maintaining the stability of the original matching ranking. Thus, without significantly reducing overall matching efficiency, the long-tail hunger problem in the capacity allocation process is effectively alleviated, achieving synergistic optimization of allocation efficiency and resource allocation balance.
[0022] Finally, after the target waybill is issued, the allocation obstruction feature vector of the capacity terminal in the candidate capacity terminal set is incremented or cleared and reset, thus constructing a state machine update mechanism that dynamically evolves over time. Attached Figure Description
[0023] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.
[0024] Figure 1 An exemplary flowchart of a logistics waybill allocation method based on dynamic feature compensation provided for some embodiments of this application;
[0025] Figure 2 An exemplary structural diagram of the electronic device provided for some embodiments of this application. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0027] Figure 1 A flowchart illustrating an exemplary logistics waybill allocation method based on dynamic feature compensation, provided for some embodiments of this application, is included in the method:
[0028] S101. Obtain the attribute information of the target waybill and obtain a set of candidate transport terminals for the target waybill based on preset filtering conditions.
[0029] Specifically, the attribute information includes the coordinates of the loading start point, the coordinates of the loading end point, the weight of the cargo, the volume of the cargo, and the transportation time window. The system publishes the target waybill and obtains the registration records submitted by the transportation terminals for the target waybill. Based on preset filtering conditions and the registration records, the system extracts a set of candidate transportation terminals for the target waybill; the preset filtering conditions include geofence range restrictions and the online idle status of the transportation terminals.
[0030] S102. Extract the historical transport data of each transport terminal in the candidate transport terminal set, and construct the transport capacity profile feature vector of each transport terminal.
[0031] Specifically, the system quantifies and cleans the historical transportation data to construct a capacity profile feature vector for each of the transportation terminals. The capacity profile feature vector may include a timeliness performance indicator sequence, a service quality evaluation indicator sequence, a dynamic behavior feature sequence, vehicle type features, vehicle energy type features, driver rating features, and permanent route features, etc.
[0032] S103. The attribute information of the target waybill and the capacity profile feature vector are respectively embedded and represented to construct the waybill-side feature vector and the capacity-side feature vector.
[0033] Specifically, the system extracts discrete categorical features and continuous numerical features from the attribute information of the target waybill. The discrete categorical features include geographical location features processed by discretization encoding, which at least include the loading start point and loading end point. The continuous numerical features include cargo weight and cargo volume. The system inputs the discrete categorical features into a preset word embedding layer for lookup table mapping calculation to generate a first dense vector. A fully connected layer nonlinear transformation operation is performed on the continuous numerical features to generate a second dense vector. The system concatenates the first dense vector and the second dense vector to generate the waybill-side feature vector. Simultaneously, the system inputs the capacity profile feature vector into a preset multilayer perceptron for high-dimensional feature extraction and linear transformation mapping, mapping the multidimensional combined profile features into a continuous numerical representation of a unified dimension, generating the capacity-side feature vector. Through the above embedding representation processing, the system transforms the discretely combined attribute information and the capacity profile feature vector in the aforementioned steps into a multidimensional continuous tensor format that can be computed by a deep representation network.
[0034] S104. Input the feature vector of the waybill side into the pre-trained waybill representation network and map it to the task embedding space. Input the feature vector of the capacity side into the pre-trained capacity representation network and map it to the capacity embedding space.
[0035] Specifically, the system inputs the waybill-side feature vector constructed in the aforementioned steps into a pre-trained waybill representation network, and the capacity-side feature vector into a pre-trained capacity representation network. The waybill representation network and the capacity representation network employ independent network weight matrices and hidden layer topologies. The system inputs the waybill-side feature vector into the first multilayer perceptron of the waybill representation network; through linear mapping transformations and nonlinear activation function operations within the first multilayer perceptron, high-order semantic feature extraction is performed on the waybill-side feature vector; the extracted high-order semantic features are then dimensionally aligned through the output layer of the waybill representation network to generate a task representation vector with a preset target dimension, thereby mapping the waybill-side feature vector to the task embedding space.
[0036] Synchronously, the system inputs the capacity-side feature vector into the second multilayer perceptron of the capacity representation network; performs forward propagation sequence calculation through the second multilayer perceptron to extract nonlinear combination features from the capacity-side feature vector; and performs the same dimension alignment processing on the extracted nonlinear combination features through the output layer of the capacity representation network to generate a capability representation vector with the same preset target dimension as the task representation vector, thereby mapping the capacity-side feature vector to the capability embedding space.
[0037] S105. Based on the mapped task embedding space and the capability embedding space, calculate the matching relationship between the transport-side feature vector and the transport-capacity-side feature vector, and generate a basic matching score.
[0038] Specifically, the system extracts the first multidimensional continuous coordinate values of the task representation vector mapped to the task embedding space; and extracts the second multidimensional continuous coordinate values of the capability representation vector mapped to the capability embedding space. The first and second multidimensional continuous coordinate values are input into a preset spatial distance calculation function; this function is configured to perform a sequence of vector inner product operations or cosine similarity calculation logic. The system performs a matrix multiplication operation on the first and second multidimensional continuous coordinate values using the spatial distance calculation function, outputting a spatial distance parameter. Subsequently, the system inputs the spatial distance parameter into a preset normalization mapping function, performs numerical domain compression calculation on the spatial distance parameter, and converts the spatial distance parameter into a basic matching score within a preset continuous numerical range of zero to one. The system outputs the basic matching score corresponding to each of the capacity terminals and stores it in the calculation cache queue of the current scheduling cycle as input parameters for subsequent compensation network calculations and ranking models.
[0039] S106. Extract the allocation stagnation feature vectors corresponding to each of the transport capacity terminals, and input the allocation stagnation feature vectors into a preset gating compensation network for nonlinear mapping to generate dynamic compensation coefficients.
[0040] Specifically, the system reads the number of consecutive missed orders and the silence duration of each of the transport terminals; it then concatenates the consecutive missed order count and the silence duration into a numerical array to construct the allocation delay feature vector. The constructed allocation delay feature vector is input into a preset gating compensation network. The gating compensation network contains a multi-layer fully connected layer structure and a nonlinear activation function. The allocation delay feature vector is then input into the input layer of the gating compensation network, where a linear weighted summation operation is performed through the fully connected layer of the feedforward neural network. The result is then input into the nonlinear activation function to perform a nonlinear mapping calculation; the nonlinear mapping calculation outputs an initial compensation weight for a specified numerical range. Subsequently, the system extracts the pre-generated service quality score for each of the transport terminals and calls the system's preset hierarchical constraint function. The service quality score is a basic feature value quantified based on multi-dimensional evaluation data (including historical cargo loss rate, on-time arrival rate, and customer complaint records) of each transport terminal within its historical fulfillment period, used to objectively characterize the comprehensive fulfillment reliability of the transport terminal. The initial compensation weight and the service quality score are synchronously input into the hierarchical constraint function. Based on the numerical stratification range of the service quality score, the system can perform a multiplicative decay operation or a numerical upper limit truncation operation on the initial compensation weight. For example, when the service quality score is detected to be lower than a preset quality benchmark threshold, the system forcibly performs a multiplicative decay operation or a numerical upper limit truncation operation on the initial compensation weight output by the gating compensation network to generate a corresponding dynamic compensation coefficient. Through the above constraint processing, the finally generated dynamic compensation coefficient retains the anti-starvation compensation effect for high-quality long-tail capacity terminals, and also achieves precise circuit breaking of the compensation weight for inferior capacity terminals from the algorithm structure, ensuring that the waybill allocation system strictly follows the global service quality priority scheduling principle when performing compensation adjustment.
[0041] S107. Based on the dynamic compensation coefficient, the basic matching score is weighted and fused to generate a comprehensive score.
[0042] Specifically, the system extracts the dynamic compensation coefficients for each of the aforementioned capacity terminals and retrieves the corresponding basic matching scores from the calculation cache queue. The system calls a preset weighted fusion model or a preset residual network weighting formula to perform a fusion calculation on the basic matching scores and the dynamic compensation coefficients. The weighted fusion model is used to adjust the matching results by introducing the dynamic compensation coefficients while maintaining the original ranking characteristics of the basic matching scores, thereby generating a comprehensive score for each of the aforementioned capacity terminals. The residual network weighting formula can be configured to perform numerical summation calculation logic using a jump connection mechanism. The system sets the basic matching scores as the baseline parameter of the identity mapping and the dynamic compensation coefficients as the residual parameter of the residual compensation module. Subsequently, the system substitutes the baseline parameter and the residual parameter into the residual network weighting formula for linear combination calculation. The combination operation includes, but is not limited to, linear weighting. Afterward, the system outputs the generated comprehensive score and stores it in a preset ranking calculation buffer pool for subsequent steps of the ranking model to call and arrange.
[0043] S108. Input the comprehensive score into the sorting model to generate the allocation probability of each of the transport capacity terminals, and sort the candidate transport capacity terminal set according to the allocation probability.
[0044] Specifically, the system extracts the comprehensive scores for each of the aforementioned capacity terminals from the sorting calculation buffer pool, and calls the exponential normalization function built into the sorting model to perform exponential mapping and proportional summation operations on all the comprehensive scores. Through the exponential normalization function, the system transforms the unbounded distribution of comprehensive scores into discrete probability distribution values with a sum of one, generating the allocation probability for each of the aforementioned capacity terminals. The system binds and stores the generated allocation probabilities with the corresponding device identifiers of the capacity terminals as key-value pairs. Subsequently, according to a preset array sorting algorithm, the magnitude of the allocation probability is set as the sorting key, and a descending sorting traversal operation is performed on all capacity terminals in the candidate capacity terminal set to generate a sorted result list containing a determined sorting order. The normalization function includes, but is not limited to, the Softmax function.
[0045] S109. Based on the sorting results and combined with the preset exploration strategy, determine the target transportation terminal from the candidate transportation terminal set, generate the final dispatch set, and issue the target waybill based on the final dispatch set.
[0046] Specifically, the system retrieves the exploration probability threshold parameter stored in the scheduling configuration database. At the decision trigger moment of the current scheduling cycle, the system calls the underlying pseudo-random number generator to generate a random value within a preset range of zero to one. The system inputs the generated random value and the exploration probability threshold parameter into a preset value comparator logic unit for comparison. When the random value is greater than the exploration probability threshold parameter, the system triggers robust branch logic, extracting a preset number of transport terminals as target transport terminals sequentially from the top node according to the descending order of the sorting result list. When the random value is less than or equal to the exploration probability threshold parameter, the system triggers probability sampling branch logic. Based on the allocated probability value corresponding to each transport terminal, the system constructs a one-dimensional discrete cumulative probability distribution interval in memory and projects a random sampling probe into the one-dimensional discrete cumulative probability distribution interval to perform a sampling operation, identifying the transport terminal hit by the probe as the target transport terminal. The system encodes the device identifier of the determined target transport terminal into the final dispatch set. The system parses the final dispatch set, calls the data encapsulation protocol to generate a dispatch instruction data packet for the target waybill, and sends the dispatch instruction data packet to the target transportation terminal through the wireless communication interface.
[0047] In this embodiment, by introducing a dual-tower neural network representation structure, decoupled modeling of waybill-side features and capacity-side features is achieved, significantly improving the accuracy of vehicle-cargo matching and feature processing efficiency. Compared to traditional feature-based allocation methods, which struggle to fully exploit the nonlinear relationships between high-dimensional heterogeneous data, this embodiment maps the attribute information of the target waybill and the capacity profile feature vector to independent task embedding spaces and capability embedding spaces, respectively, and calculates their matching relationship in a continuous vector space to generate a basic matching score, thereby improving the discriminative power of feature representation and the accuracy of matching results.
[0048] Meanwhile, since the features of the transport unit side and the transport capacity side adopt a decoupled encoding method, the distribution interference problem caused by the direct interaction of different types of features in the same space is avoided. When facing a large number of candidate transport capacity terminals, the matching process can be completed efficiently through vectorized calculation, thereby effectively reducing the complexity of model inference and improving the system's response performance in high-concurrency scheduling scenarios.
[0049] Furthermore, by constructing a feature vector representing allocation bottlenecks and introducing a gating compensation network to model the resource acquisition status of capacity terminals, dynamic compensation coefficients are generated. These dynamic compensation coefficients are then applied to the basic matching score through a weighted fusion mechanism. This allows the system to adaptively adjust capacity terminals with insufficient resource acquisition while maintaining the stability of the original matching ranking. Thus, without significantly reducing overall matching efficiency, the long-tail hunger problem in the capacity allocation process is effectively alleviated, achieving synergistic optimization of allocation efficiency and resource allocation balance.
[0050] In one embodiment, after issuing the target waybill based on the final dispatch set, the method further includes:
[0051] Extract the allocation obstruction feature vector corresponding to each of the aforementioned transportation terminals;
[0052] Traverse the set of candidate transport terminals;
[0053] Perform a step-by-step accumulation operation on the allocation obstruction feature vector of the capacity terminal that is not included in the final dispatch set;
[0054] A reset operation is performed on the allocation obstruction feature vector of the capacity terminal included in the final dispatch set.
[0055] Specifically, the system first extracts the allocation stall feature vectors corresponding to each capacity terminal within the current scheduling period from the storage module or cache queue. These allocation stall feature vectors characterize the resource allocation status of the capacity terminal during continuous scheduling, and include at least information such as the number of consecutive missed orders and the time interval since the last successful order acceptance. Through the unified extraction of these feature vectors, the system can obtain the initial stall status data of each capacity terminal in the current candidate capacity terminal set, providing basic input for subsequent update operations.
[0056] After extracting the allocation delay feature vector, the system iterates through the candidate capacity terminal set. The system can sequentially access each capacity terminal in the candidate capacity terminal set according to a preset data structure, and perform a status determination operation on each capacity terminal based on the member information in the final dispatch set. The status determination operation is used to determine whether the current capacity terminal has been included in the final dispatch set, thereby distinguishing between capacity terminals that have received order allocations and those that have not.
[0057] The system incrementally updates the number of consecutive missed orders, increasing it by a preset step size based on the current scheduling cycle. Simultaneously, it continuously accumulates or advances the time interval features to reflect the continuous increase in waiting time for the terminal since its last successful order acceptance. Through this incremental accumulation, the system gradually strengthens the characterization of the stalled state of unallocated terminals, giving them higher compensation priority in subsequent scheduling processes.
[0058] For the transport terminals included in the final dispatch set, the system performs a reset operation on their corresponding allocation congestion feature vector. The system can restore the number of consecutive missed orders to a preset initial value and reset the time interval feature to an initial timestamp or zero value, indicating that the transport terminal has successfully obtained a waybill allocation within the current scheduling cycle. Through this reset process, the system avoids transport terminals that have already obtained resources from being judged as being in a high-congestion state in subsequent scheduling, thereby ensuring that the congestion feature vector can accurately reflect the latest resource acquisition status of the transport terminal.
[0059] After completing the above traversal and update operations, the system writes the updated allocation stall feature vector back to the storage module or state cache as input data for the gating compensation network in the next scheduling cycle. Through the above update mechanism, the allocation stall feature vector can continuously evolve during multiple scheduling rounds, forming a dynamically changing state sequence over time, thereby supporting the subsequent allocation model's refined modeling of the resource acquisition status of transport terminals.
[0060] In this embodiment, a dynamic update mechanism for allocation bottleneck feature vectors is introduced after waybill issuance, enabling the system to continuously track changes in the resource acquisition status of each capacity terminal. By performing cumulative updates on capacity terminals that have not received allocations and resetting those that have received allocations, the bottleneck characteristics of capacity terminals that have not received orders for a long time can be gradually amplified, making them easier for the compensation mechanism to identify in subsequent scheduling processes, thereby effectively alleviating the long-tail hunger problem in resource allocation. Simultaneously, this update mechanism has a simple structure, low computational overhead, and is easy to implement in real-time updates in distributed scheduling systems, helping to improve overall allocation fairness and scheduling stability without increasing system complexity.
[0061] In one embodiment, constructing the capacity profile feature vector for each of the capacity terminals includes:
[0062] The historical transportation data is cleaned and processed to extract multidimensional transportation feature data; the multidimensional transportation feature data includes at least one of the following: timeliness performance index sequence, service quality evaluation index sequence, dynamic behavior feature sequence, vehicle type feature, vehicle energy type feature, driver rating feature, and permanent route feature.
[0063] Normalization processing is performed on the multidimensional carrier feature data;
[0064] The normalized multidimensional carrier feature data are spliced together to generate the capacity profile feature vector corresponding to each of the capacity terminals.
[0065] Specifically, the system first performs data cleaning on the historical transportation data. This historical transportation data originates from actual transportation records of transportation terminals within a preset statistical period, including order fulfillment information, service feedback information, and operational behavior data. The system uses preset data cleaning rules to remove outliers, fill in missing values, and standardize the format of the historical transportation data to eliminate data noise and inconsistencies, thereby ensuring the accuracy and stability of subsequent feature extraction.
[0066] After data cleaning, the system extracts multi-dimensional transportation feature data from the processed historical transportation data. This multi-dimensional transportation feature data is used to characterize the comprehensive transportation capacity of a transport terminal from multiple dimensions, including at least one of the following: timeliness fulfillment indicator sequence, service quality evaluation indicator sequence, dynamic behavior feature sequence, vehicle type feature, vehicle energy type feature, driver rating feature, and resident route feature. Specifically, the timeliness fulfillment indicator sequence reflects the on-time completion status of the transport terminal in historical transportation tasks; the service quality evaluation indicator sequence characterizes customer feedback ratings and cargo damage; the dynamic behavior feature sequence characterizes the order-accepting behavior and activity level of the transport terminal in recent scheduling; the vehicle type and vehicle energy type features describe the transportation vehicle attributes of the transport terminal; the driver rating feature reflects the overall service level of drivers; and the resident route feature characterizes the long-term operational preferences of the transport terminal in specific regions or routes.
[0067] After extracting the multidimensional carrier feature data, the system performs normalization processing on various features. For continuous numerical features, the system scales their numerical range using min-max normalization or standardization methods to eliminate the impact of differences in feature dimensions. For discrete categorical features, the system transforms them into a unified numerical representation through encoding mapping or numerical processing. Through the above normalization processing, the system can convert multidimensional features from different sources and with large distribution differences into a unified numerical space representation with strong comparability.
[0068] After normalization, the system concatenates the processed multidimensional carrier feature data in a preset order to generate a fixed-dimensional capacity profile feature vector. This concatenation process integrates various heterogeneous features into a unified vector representation, facilitating input processing for subsequent deep learning models. Through this step, each capacity terminal generates a set of structurally consistent and dimensionally unified capacity profile feature vectors, thereby achieving a standardized characterization of the comprehensive capabilities of the capacity terminal.
[0069] In this embodiment, a multi-dimensional capacity profile feature vector that comprehensively reflects the carrying capacity of transport terminals is constructed by systematically cleaning, extracting features, and normalizing and concatenating historical transport data. Compared with traditional methods that rely on only a single indicator or a small number of features, this embodiment effectively improves the completeness and distinguishability of capacity feature expression by integrating multi-dimensional information, enabling a more refined depiction of the capacity differences between different transport terminals. Simultaneously, through normalization and unified vectorization processing, the interference between different dimensions of features is reduced, improving the stability and convergence efficiency of subsequent model calculations, thereby providing a reliable data foundation for achieving high-precision waybill allocation.
[0070] In one embodiment, the step of calculating the matching relationship between the transport-side feature vector and the transport-capacity-side feature vector based on the mapped task embedding space and the capability embedding space, and generating a basic matching score, includes:
[0071] Extract the first multidimensional spatial coordinates of the one-sided feature vector mapped to the task embedding space;
[0072] Extract the second multidimensional spatial coordinates of the capacity-side feature vector mapped to the capability embedding space;
[0073] Perform an inner product operation on the first multidimensional spatial coordinates and the second multidimensional spatial coordinates to generate spatial distance parameters;
[0074] The spatial distance parameter is converted into the basic matching score.
[0075] Specifically, the system first extracts the first multidimensional spatial coordinates of the waybill-side feature vector mapped to the task embedding space. The first multidimensional spatial coordinates are high-dimensional continuous numerical representations output by the waybill representation network, used to characterize the positional distribution features of the target waybill in the task semantic space.
[0076] Simultaneously, the system extracts the second multidimensional spatial coordinates of the capacity-side feature vector mapped to the capacity embedding space. These second multidimensional spatial coordinates are high-dimensional continuous numerical representations output by the capacity representation network, used to characterize the comprehensive carrying capacity of the capacity terminal in the capacity semantic space. Through these steps, the system obtains two sets of vector representations under a unified dimensional constraint, thus providing an input basis for subsequent spatial matching calculations.
[0077] After obtaining the first and second multidimensional spatial coordinates, the system performs an inner product operation on them to generate a corresponding spatial distance parameter. Following the rules of vector dot product, the system performs element-wise multiplication and summation on each dimension component of the first and second multidimensional spatial coordinates to obtain a scalar value characterizing the degree of matching between them. The spatial distance parameter reflects the similarity between the waybill demand characteristics and the capacity characteristics in the embedding space; a larger value indicates a higher degree of matching.
[0078] After calculating the spatial distance parameters, the system performs numerical mapping processing to generate the basic matching score. The system inputs the spatial distance parameters into a preset mapping function and converts them into score values suitable for subsequent sorting through monotonic transformation. The mapping function is used to scale or normalize the original inner product result to improve the comparability and numerical stability between different samples. Through the above mapping processing, the system finally obtains the basic matching score for each transportation terminal corresponding to the target waybill and outputs the basic matching score to the subsequent processing module.
[0079] In this embodiment, by performing inner product operations on the features of the transport terminal and the transport capacity side in a unified embedding space, an efficient characterization of their matching relationship is achieved. Compared with traditional rule-based or low-dimensional feature matching methods, this embodiment utilizes high-dimensional continuous vector representations to fully express complex nonlinear feature relationships, thereby improving the accuracy of matching calculations. Simultaneously, the vector inner product-based calculation method has good computational efficiency, is suitable for batch matching calculations of a large number of candidate transport capacity terminals, helps reduce the computational overhead of the system in high-concurrency scheduling scenarios, and improves overall response performance.
[0080] In one embodiment, the allocation blocking feature vector includes a value for the number of consecutive missed orders and a value for the duration of silence; the step of inputting the allocation blocking feature vector into a preset gating compensation network for nonlinear mapping to generate dynamic compensation coefficients includes:
[0081] The consecutive number of missed orders and the silence duration are input into the input layer of the gating compensation network;
[0082] The data of the input layer is mapped and calculated using the activation function in the gated compensation network to output the initial compensation weights;
[0083] Extract the service quality score of the transportation terminal, and perform hierarchical constraint processing on the initial compensation weight based on the service quality score to generate the dynamic compensation coefficient.
[0084] Specifically, the system first concatenates the number of consecutive missed orders and the duration of silence to construct a unified allocation bottleneck feature vector, and then inputs this feature vector into the input layer of the gating compensation network. The gating compensation network is a pre-trained or pre-structured neural network model used to represent and map the resource acquisition status of the transportation terminal.
[0085] After input is received, the system performs linear weighting calculations on the allocated stagnation feature vectors through a multi-layer fully connected structure in the gated compensation network. Then, it performs nonlinear mapping processing on the linear transformation result using a preset activation function, thereby outputting the corresponding initial compensation weights. The activation function constrains the network output within a preset numerical range, ensuring that the initial compensation weights stably reflect the differences in compensation intensity under different stagnation states. Through this nonlinear mapping process, the system can adaptively generate compensation weights of varying magnitudes based on changes in the number of consecutive missed orders and the duration of silence.
[0086] After obtaining the initial compensation weight, the system further extracts the service quality score corresponding to the capacity terminal and performs hierarchical constraint processing on the initial compensation weight based on the service quality score. The system divides the service quality score into corresponding level intervals according to preset service quality stratification intervals and configures differentiated constraint rules for different level intervals. The constraint rules include numerical scaling or upper limit restrictions on the initial compensation weight to prevent low-service-quality capacity terminals from receiving excessive compensation due to high congestion levels. Through the above hierarchical constraint processing, the system introduces a fair compensation mechanism while taking into account the impact of service quality factors on scheduling results, thereby generating the final dynamic compensation coefficient.
[0087] After generating the dynamic compensation coefficient, the system outputs it to the subsequent weighted fusion module for adjusting the basic matching score. Through the above processing flow, the dynamic compensation coefficient can form a differentiated compensation intensity distribution among different transportation terminals, thereby providing a more refined adjustment basis for subsequent allocation decisions.
[0088] In this embodiment, dynamic modeling of the resource acquisition status of transport terminals is achieved by introducing allocation obstruction feature vectors and constructing a gated compensation network. By combining nonlinear mapping with service quality constraints, the compensation mechanism can increase the allocation opportunities for transport terminals with long-term no-order periods while avoiding excessive bias towards low-service-quality transport terminals, thus achieving a balance between allocation fairness and service quality. This method has a clear structure and controllable computational process, making it suitable for large-scale transport scheduling scenarios and contributing to improved stability and rationality of the overall scheduling system.
[0089] In one embodiment, the step of weighting and fusing the basic matching scores based on the dynamic compensation coefficient to generate a comprehensive score includes:
[0090] Extract the preset residual network weighting formula;
[0091] The basic matching score is used as the benchmark, and the dynamic compensation coefficient is used as the residual term. These are then substituted into the residual network weighting formula for summation to generate the comprehensive score.
[0092] Specifically, the system first extracts a preset residual network weighting formula. This formula is used to adjust the results by introducing a compensation term while retaining the original matching score information. The residual network weighting formula corresponds to a weighted model with a skip connection structure, and its core lies in achieving a smooth correction of the original score through the combination of the baseline term and the compensation term.
[0093] After obtaining the residual network weighting formula, the system inputs the basic matching score as a benchmark term into the residual network weighting formula, and simultaneously inputs the dynamic compensation coefficient as a residual term into the residual network weighting formula. The benchmark term reflects the original matching degree between the waybill and the transport capacity, and the residual term characterizes the compensation information generated based on the allocation delay state. Through the above parameter substitution operation, the system integrates information from two different sources into the same fusion calculation framework.
[0094] In the specific fusion calculation process, the system multiplies the dynamic compensation coefficient with a preset scaling factor to obtain a compensation adjustment value, and then adds the compensation adjustment value to the basic matching score to obtain the final comprehensive score. The scaling factor is used to control the influence of the compensation item on the overall score, so as to avoid the compensation item from excessively interfering with the original matching results.
[0095] After calculating the comprehensive score, the system outputs the comprehensive score to the subsequent sorting processing module and stores it in a preset calculation cache structure for use by the sorting model. Through the above weighted fusion processing, the system maintains the dominant role of the original matching result while introducing a compensation mechanism to appropriately adjust specific capacity terminals, thereby forming a comprehensive evaluation result that balances matching accuracy and allocation fairness.
[0096] In this embodiment, a weighted fusion method based on residual structure is adopted to organically combine the basic matching score with the dynamic compensation coefficient. This allows for dynamic adjustment of the allocation priority of transport terminals without compromising the stability of the original matching order. Compared to direct replacement or forced rearrangement, this method can gradually compensate transport terminals that have not been allocated for a long time while maintaining the overall matching effect. This effectively alleviates the problem of uneven resource allocation during the scheduling process and improves the overall stability and fairness of the system.
[0097] In one embodiment, determining the target transportation terminal from the candidate transportation terminal set based on the ranking results and in conjunction with a preset exploration strategy, and generating the final dispatch set, includes:
[0098] Obtain the preset exploration probability threshold and generate random values within the scheduling period;
[0099] Compare the random value with the exploration probability threshold;
[0100] When the random value is greater than the exploration probability threshold, a preset number of transportation terminals are sequentially selected from the candidate transportation terminal set according to the sorting result as the target transportation terminal.
[0101] When the random value is less than or equal to the exploration probability threshold, a probability sampling mechanism is triggered to sample the candidate capacity terminal set according to the allocation probability, and the capacity terminal hit by the sampling is taken as the target capacity terminal.
[0102] The identified target transportation terminals are incorporated into the final dispatch set.
[0103] Specifically, the system first obtains a preset exploration probability threshold from the scheduling configuration module and generates a random value when the current scheduling cycle is triggered. The random value is located within a preset value range and is used to switch between deterministic selection and random exploration, thereby realizing dynamic selection of different decision paths.
[0104] After obtaining the random value and the exploration probability threshold, the system performs a comparison operation between the two. This comparison operation determines whether the current scheduling process has entered the exploration phase. When the random value is greater than the exploration probability threshold, the system executes a deterministic selection strategy based on the ranking results. The system can sequentially select a preset number of target terminals from the candidate terminal set according to the ranking result list generated in the previous steps, arranged in descending order of allocation probability. This process ensures that terminals with higher matching scores are prioritized in most scheduling cycles to maintain overall allocation efficiency and service quality.
[0105] When the random value is less than or equal to the exploration probability threshold, the system triggers a probability sampling mechanism. The system can construct a discrete probability distribution based on the allocation probability corresponding to each of the aforementioned capacity terminals, and perform a sampling operation on the candidate capacity terminal set according to the discrete probability distribution. During the sampling process, the system randomly selects candidate capacity terminals according to the probability distribution weights, and uses the sampled capacity terminal as the target capacity terminal. Through this probability sampling mechanism, the system can introduce a certain degree of randomness while ensuring the overall matching trend, providing opportunities for capacity terminals with low exposure or insufficient historical data to participate in the allocation.
[0106] After identifying the target transportation terminals, the system aggregates their identification information and incorporates it into the final dispatch set. This final dispatch set serves as input to the scheduling execution module, used to generate corresponding scheduling instructions and send them to the target transportation terminals. Through this process, the system completes the entire process from sorting decision to target selection within a single scheduling cycle.
[0107] In this embodiment, an exploration strategy is introduced based on the ranking results, achieving an organic combination of deterministic allocation and random exploration. While ensuring that high-matching capacity terminals receive priority in waybill allocation, a probabilistic sampling mechanism provides participation opportunities for potentially high-quality capacity terminals with insufficient historical exposure. This effectively alleviates the model's dependence on historical data and reduces path dependency risks during the allocation process. This method helps improve the system's adaptability to new and long-tail capacity terminals, while enhancing the model's generalization performance and overall scheduling effectiveness over long-term operation.
[0108] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, but without changing the core design of the algorithm and process, are also within the scope of protection of this application.
[0109] Furthermore, some embodiments of this application also provide an electronic device. The electronic device can be various forms of digital computer, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, etc. The electronic device can also be various forms of mobile devices, such as cellular phones, smartphones, wearable devices, and other similar computing devices.
[0110] The electronic device includes: one or more processors; and a memory storing computer program instructions that, when executed, cause the processor to perform the steps of the methods provided in any one or more of the above embodiments. Figure 2 An exemplary structural diagram of the electronic device is disclosed. The electronic device includes one or more processors 1101, a memory 1102, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components are interconnected via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some other embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple electronic devices can be connected, each providing some of the necessary operations. The components, their connections and relationships, and their functions shown herein are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.
[0111] The electronic device may further include an input device 1103 and an output device 1104. The processor 1101, memory 1102, input device 1103 and output device 1104 may be connected by a bus or other means, as shown in the figure, which is connected by a bus.
[0112] Input device 1103 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the electronic device, such as a touch screen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 1104 may include a display device, auxiliary lighting device (e.g., LED), and haptic feedback device (e.g., vibration motor). The display device may include, but is not limited to, a liquid crystal display, a light-emitting diode display, and a plasma display. In some embodiments, the display device may be a touch screen.
[0113] To provide interaction with the user, the electronic device can be a computer. The computer has: a display device (e.g., a cathode ray tube or LCD monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback); and input from the user can be received in any form (e.g., voice input or tactile input).
[0114] In this embodiment, a computer-readable medium stores a computer program / instructions that, when executed by a processor, implement the steps of the methods provided in any one or more of the above embodiments. This computer-readable medium may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into that device. The aforementioned computer-readable medium carries one or more computer-readable instructions.
[0115] The memory 1102 can serve as a non-transitory computer-readable storage medium, used to store non-transitory software programs, non-transitory computer-executable programs, and modules. The processor 1101 executes various functional applications and data processing of the server by running the non-transitory software programs, instructions, and modules stored in the memory 1102, thereby implementing the program instructions / modules corresponding to the methods provided in any one or more of the embodiments described above in this application.
[0116] The memory 1102 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the electronic device. Furthermore, the memory 1102 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 1102 may optionally include memory remotely located relative to the processor 1101, and these remote memories can be connected to the electronic device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
[0117] It should be noted that the computer-readable medium described in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. Computer-readable media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections having one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
[0118] Computer-readable media include permanent and non-permanent, removable and non-removable media, which can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory, static random access memory, dynamic random access memory, other types of random access memory, read-only memory, electrically erasable programmable read-only memory, flash memory or other memory technologies, read-only optical discs, digital versatile optical discs or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.
[0119] Computer program code for performing the operations of this application can be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, and C++, and conventional procedural programming languages such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including local area networks (LANs) or wide area networks (WANs), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0120] In the above embodiments, all or part of the implementation can be achieved through software, hardware, firmware, or any combination thereof. For example, it can be implemented using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In some embodiments, the software program of this application can be executed by a processor to implement the above steps or functions. Similarly, the software program of this application (including related data structures) can be stored in a computer-readable recording medium, such as RAM memory, magnetic or optical drives, floppy disks, and similar devices. In addition, some steps or functions of this application can be implemented in hardware, for example, as circuitry that cooperates with a processor to perform the various steps or functions.
[0121] The computer program product provided in this application includes one or more computer programs / instructions. When executed by a processor, these computer programs / instructions generate, in whole or in part, the processes or functions described in this application. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.
[0122] The flowcharts or block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-specific system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0123] The scope of this application is defined by the appended claims rather than the foregoing description, and is therefore intended to encompass all variations falling within the meaning and scope of equivalents of the claims. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a device claim may also be implemented by a single unit or device in software or hardware. Terms such as "first," "second," etc., are used only for distinguishing descriptions and do not indicate any particular order, nor should they be construed as indicating or implying relative importance.
[0124] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily made by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims, and the above embodiments should be regarded as exemplary and non-limiting.
Claims
1. A logistics waybill allocation method based on dynamic feature compensation, characterized in that, The method includes: Obtain the attribute information of the target waybill, and obtain a set of candidate transport terminals for the target waybill based on preset filtering conditions; Extract historical transport data of each transport terminal in the candidate transport terminal set, and construct a transport capacity profile feature vector for each transport terminal; The attribute information of the target waybill and the feature vector of the capacity profile are respectively embedded and represented to construct the feature vector on the waybill side and the feature vector on the capacity side; The feature vectors on the waybill side are input into the pre-trained waybill representation network and mapped to the task embedding space, and the feature vectors on the capacity side are input into the pre-trained capacity representation network and mapped to the capacity embedding space. Based on the mapped task embedding space and the capability embedding space, the matching relationship between the transport-side feature vector and the transport capacity-side feature vector is calculated to generate a basic matching score; Extract the allocation stagnation feature vector corresponding to each of the transport terminals, and input the allocation stagnation feature vector into a preset gating compensation network for nonlinear mapping to generate dynamic compensation coefficients; The basic matching score is weighted and fused based on the dynamic compensation coefficient to generate a comprehensive score. The comprehensive score is input into the ranking model to generate the allocation probability of each of the transport capacity terminals, and the candidate transport capacity terminal set is ranked according to the allocation probability. Based on the ranking results and combined with the preset exploration strategy, the target transportation terminal is determined from the candidate transportation terminal set, a final dispatch set is generated, and the target waybill is issued based on the final dispatch set.
2. The method according to claim 1, characterized in that, After issuing the target waybill based on the final dispatch set, the process also includes: Extract the allocation obstruction feature vector corresponding to each of the aforementioned transportation terminals; Traverse the set of candidate transport terminals; Perform a step-by-step accumulation operation on the allocation obstruction feature vector of the capacity terminal that is not included in the final dispatch set; A reset operation is performed on the allocation obstruction feature vector of the capacity terminal included in the final dispatch set.
3. The method according to claim 1, characterized in that, The construction of the capacity profile feature vector for each of the aforementioned capacity terminals includes: The historical transportation data is cleaned and processed to extract multidimensional transportation feature data; the multidimensional transportation feature data includes at least one of the following: timeliness performance index sequence, service quality evaluation index sequence, dynamic behavior feature sequence, vehicle type feature, vehicle energy type feature, driver rating feature, and permanent route feature. Normalization processing is performed on the multidimensional carrier feature data; The normalized multidimensional carrier feature data are spliced together to generate the capacity profile feature vector corresponding to each of the capacity terminals.
4. The method according to claim 1, characterized in that, The process involves calculating the matching relationship between the single-side feature vector and the capacity-side feature vector based on the mapped task embedding space and the capacity embedding space, and generating a basic matching score, including: Extract the first multidimensional spatial coordinates of the one-sided feature vector mapped to the task embedding space; Extract the second multidimensional spatial coordinates of the capacity-side feature vector mapped to the capability embedding space; Perform an inner product operation on the first multidimensional spatial coordinates and the second multidimensional spatial coordinates to generate spatial distance parameters; The spatial distance parameter is converted into the basic matching score.
5. The method according to claim 1, characterized in that, The allocation blocking feature vector includes the number of consecutive missed orders and the duration of silence; the step of inputting the allocation blocking feature vector into a preset gating compensation network for nonlinear mapping to generate dynamic compensation coefficients includes: The consecutive number of missed orders and the silence duration are input into the input layer of the gating compensation network; The data of the input layer is mapped and calculated using the activation function in the gated compensation network to output the initial compensation weights; Extract the service quality score of the transportation terminal, and perform hierarchical constraint processing on the initial compensation weight based on the service quality score to generate the dynamic compensation coefficient.
6. The method according to claim 1, characterized in that, The step of weighting and fusing the basic matching scores based on the dynamic compensation coefficient to generate a comprehensive score includes: Extract the preset residual network weighting formula; The basic matching score is used as the benchmark, and the dynamic compensation coefficient is used as the residual term. These are then substituted into the residual network weighting formula for summation to generate the comprehensive score.
7. The method according to claim 1, characterized in that, The process of determining target transportation terminals from the candidate transportation terminal set based on the ranking results and in conjunction with a preset exploration strategy, and generating a final dispatch set, includes: Obtain the preset exploration probability threshold and generate random values within the scheduling period; Compare the random value with the exploration probability threshold; When the random value is greater than the exploration probability threshold, a preset number of transportation terminals are sequentially selected from the candidate transportation terminal set according to the sorting result as the target transportation terminal. When the random value is less than or equal to the exploration probability threshold, a probability sampling mechanism is triggered to sample the candidate capacity terminal set according to the allocation probability, and the capacity terminal hit by the sampling is taken as the target capacity terminal. The identified target transportation terminals are incorporated into the final dispatch set.
8. An electronic device, characterized in that, The electronic device includes: One or more processors; and A memory storing computer program instructions, which, when executed, cause the processor to perform the steps of the method as described in any one of claims 1 to 7.
9. A computer-readable medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.
10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.