Multi-task intelligent scheduling management platform

Abstract

The present application relates to the technical field of task scheduling management, and specifically discloses a multi-task intelligent scheduling management platform. By adopting data processing and natural language processing algorithms based on artificial intelligence and deep learning, it analyzes each data sample in the historical task dataset and the task to be processed, so as to capture the embedded semantic features of the task to be processed and the aggregated coding features between the semantics of multiple data samples. Then, through the task semantic matching technology, it assigns task priorities to the task to be processed, thereby contributing to the realization of efficient task scheduling and resource allocation.

Description

Technical Field

[0001] This application relates to the technical field of task scheduling management, and more specifically, to a multi-task intelligent scheduling management platform. Background Art

[0002] With the rapid development of the e-commerce and retail industries, the efficiency of logistics distribution has become one of the key factors affecting customer satisfaction. Especially during peak periods, such as promotional activities, logistics companies face the pressure of processing a huge number of orders. How to efficiently schedule resources, optimize distribution routes, and improve the priority of task processing has become an urgent problem for logistics enterprises to solve.

[0003] Traditional scheduling management methods mainly rely on manual experience or simple rule engines and are difficult to handle complex multi-task scenarios. For example, in the face of multiple orders, different delivery destinations, time requirements, and resource limitations (such as vehicles, personnel, etc.), traditional methods often cannot achieve a globally optimal resource allocation, resulting in low distribution efficiency, increased costs, and even affecting the customer experience.

[0004] Therefore, a multi-task intelligent scheduling management platform is desired. Summary of the Invention

[0005] This application provides a multi-task intelligent scheduling management platform that can achieve efficient task scheduling and resource allocation.

[0006] According to one aspect of this application, there is provided a multi-task intelligent scheduling management platform, including: a to-be-processed task extraction module for extracting to-be-processed tasks from a task queue; a historical task data set extraction module for extracting a historical task data set from a background database, where each data sample in the historical task data set includes task content and task priority; a data set splitting module for splitting the historical task data set based on task priority to obtain multiple historical task data subsets; a historical task data semantic aggregation and encoding module for performing semantic embedding encoding and semantic aggregation analysis on each data sample in each historical task data subset to obtain multiple historical task data subset semantic aggregation and encoding vectors; a to-be-processed task semantic embedding encoding module for performing semantic embedding encoding on the to-be-processed task to obtain a to-be-processed task semantic embedding encoding vector; a task semantic matching calculation module for calculating the task semantic matching degrees between the to-be-processed task semantic embedding encoding vector and each historical task data subset semantic aggregation and encoding vector in the multiple historical task data subset semantic aggregation and encoding vectors to obtain multiple task semantic matching degrees; and a task priority designation module for designating the task priority corresponding to the largest one among the multiple task semantic matching degrees as the task priority of the to-be-processed task.

[0007] In the above multi-task intelligent scheduling management platform, the historical task data semantic aggregation and encoding module includes: a historical task semantic embedding and encoding unit, which is used to perform semantic embedding and encoding on each data sample in each historical task data subset to obtain a subset of historical task semantic embedding and encoding vectors; a feature aggregation and analysis unit, which is used to perform feature aggregation and analysis based on information kernel aggregation on the subset of historical task semantic embedding and encoding vectors to obtain the semantic aggregation and encoding vector of the historical task data subset.

[0008] In the above multi-task intelligent scheduling management platform, the feature aggregation and analysis unit includes: a historical task semantic coarse-grained aggregation and encoding subunit, which is used to input the subset of historical task semantic embedding and encoding vectors into an information kernel coarse-grained aggregation network to obtain a historical task semantic coarse-grained aggregation and encoding vector; a kernel aggregation compensation weight factor determination subunit, which is used to determine the kernel aggregation compensation weights of each historical task semantic embedding and encoding vector in the subset of historical task semantic embedding and encoding vectors based on the historical task semantic coarse-grained aggregation and encoding vector to obtain a set of kernel aggregation compensation weight factors; a dynamic compensation aggregation and analysis subunit, which is used to perform dynamic compensation aggregation analysis based on node fine-grained features on the historical task semantic coarse-grained aggregation and encoding vector and the subset of historical task semantic embedding and encoding vectors based on the set of kernel aggregation compensation weight factors to obtain a historical task semantic fine-grained compensation and aggregation encoding vector; a historical task data aggregation subunit, which is used to input the historical task semantic fine-grained compensation and aggregation encoding vector and the historical task semantic coarse-grained aggregation and encoding vector into a residual unit to obtain the semantic aggregation and encoding vector of the historical task data subset.

[0009] In the above multi-task intelligent scheduling management platform, the kernel aggregation compensation weight factor determination subunit includes: a kernel aggregation compensation factor calculation secondary subunit, which is used to calculate the kernel aggregation compensation factors of each historical task semantic embedding and encoding vector in the subset of historical task semantic embedding and encoding vectors relative to the historical task semantic coarse-grained aggregation and encoding vector to obtain a set of kernel aggregation compensation factors; a compensation explicit modeling processing secondary subunit, which is used to perform compensation explicit modeling based on a gating function on the set of kernel aggregation compensation factors to obtain the set of kernel aggregation compensation weight factors.

[0010] In the above multi-task intelligent scheduling management platform, the core convergence compensation factor calculation secondary subunit includes: a vector modulation tertiary subunit for modulating the historical task semantic embedding encoding vector and the historical task semantic coarse-grained convergence encoding vector to obtain a historical task semantic embedding encoding modulation vector and a historical task semantic coarse-grained convergence encoding modulation vector; a compensation information calculation tertiary subunit for calculating the compensation information between the historical task semantic embedding encoding modulation vector and the historical task semantic coarse-grained convergence encoding modulation vector to obtain a historical task semantic core convergence compensation weight vector; a historical task semantic deviation compensation factor calculation tertiary subunit for calculating a historical task semantic deviation compensation factor based on the historical task semantic embedding encoding vector and the historical task semantic coarse-grained convergence encoding vector; and a core convergence compensation factor calculation tertiary subunit for calculating a core convergence compensation factor based on the historical task semantic core convergence compensation weight vector and the deviation compensation factor.

[0011] In the above multi-task intelligent scheduling management platform, the historical task semantic deviation compensation factor calculation tertiary subunit is used to: in response to the two-norm of the historical task semantic embedding encoding vector being less than the two-norm of the historical task semantic coarse-grained convergence encoding vector, calculate the logarithm function value with base 2 of the sum of the quotient of the two-norm of the historical task semantic embedding encoding vector divided by the two-norm of the historical task semantic coarse-grained convergence encoding vector and the constant one; in response to the two-norm of the historical task semantic embedding encoding vector being greater than or equal to the two-norm of the historical task semantic coarse-grained convergence encoding vector, calculate the quotient of the two-norm of the historical task semantic embedding encoding vector divided by the two-norm of the historical task semantic coarse-grained convergence encoding vector.

[0012] In the above multi-task intelligent scheduling management platform, the compensation explicit modeling processing secondary subunit is used to: in response to the core convergence compensation factor in the set of core convergence compensation factors being greater than or equal to a predetermined threshold, input the core convergence compensation factor into the sigmoid function; in response to the core convergence compensation factor in the set of core convergence compensation factors being less than the predetermined threshold, set the core convergence compensation factor to 0.

[0013] In the above multi-task intelligent scheduling management platform, the dynamic compensation convergence analysis subunit is used to: input the set of core convergence compensation weight factors, the historical task semantic coarse-grained convergence encoding vector, and a subset of the historical task semantic embedding encoding vectors into a node fine-grained dynamic compensation convergence network to obtain a historical task semantic fine-grained compensation convergence encoding vector.

[0014] In the above multi-task intelligent scheduling management platform, inputting the set of the nuclear convergence compensation weight factors, the historical task semantic coarse-grained convergence coding vector, and a subset of the historical task semantic embedding coding vectors into the node fine-grained dynamic compensation convergence network to obtain the historical task semantic fine-grained compensation convergence coding vector includes: calculating the position-wise differences between each of the historical task semantic embedding coding vectors in the subset of the historical task semantic embedding coding vectors and the historical task semantic coarse-grained convergence coding vector to obtain a set of historical task semantic convergence deviation vectors; using the set of the nuclear convergence compensation weight factors as weights, calculating the weighted sum between each of the historical task semantic convergence deviation vectors in the set of the historical task semantic convergence deviation vectors to obtain the historical task semantic fine-grained compensation convergence coding vector.

[0015] In the above multi-task intelligent scheduling management platform, the task semantic matching calculation module is configured to: input the to-be-processed task semantic embedding coding vector and each of the historical task data subset semantic aggregation coding vectors in the multiple historical task data subset semantic aggregation coding vectors into the task semantic matching network to obtain the multiple task semantic matching degrees.

[0016] A multi-task intelligent scheduling management platform provided by the present application analyzes each data sample in the historical task dataset and the to-be-processed task by using data processing and natural language processing algorithms based on artificial intelligence and deep learning, so as to capture the embedded semantic features of the to-be-processed task and the aggregation coding features between the semantics of multiple data samples, and then assigns a task priority to the to-be-processed task through the task semantic matching technology, thereby helping to achieve efficient task scheduling and resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] To more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application and do not limit the present application.

[0018] Figure 1 It is a schematic block diagram of the multi-task intelligent scheduling management platform according to the embodiment of the present application.

[0019] Figure 2 It is a schematic diagram of the data flow of the multi-task intelligent scheduling management platform according to the embodiment of the present application.

[0020] Figure 3 It is a schematic block diagram of the historical task data semantic aggregation coding module in the multi-task intelligent scheduling management platform according to the embodiment of the present application.

[0021] Figure 4Schematic block diagram of the feature aggregation analysis unit in the multi-task intelligent scheduling management platform according to an embodiment of the present application.

[0022] Figure 5 Schematic block diagram of the core convergence compensation weight factor determination subunit in the multi-task intelligent scheduling management platform according to an embodiment of the present application.

[0023] Figure 6 Schematic block diagram of the second-level subunit for calculating the core convergence compensation factor in the multi-task intelligent scheduling management platform according to an embodiment of the present application. Detailed implementation manners

[0024] Next, the technical solutions in the embodiments of the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts also belong to the scope of protection of the present application.

[0025] In recent years, with the rapid development of artificial intelligence and machine learning technologies, intelligent scheduling systems have gradually become effective tools to solve this problem. By introducing technologies such as natural language processing, semantic analysis, and deep learning, intelligent scheduling systems can learn task priority patterns from historical task data and make dynamic scheduling decisions in combination with real-time task features. This data-driven scheduling method can not only improve the efficiency of task processing but also significantly reduce the cost of manual intervention.

[0026] Specifically, the technical concept of the present application is to analyze each data sample in the historical task dataset and the task to be processed by using data processing and natural language processing algorithms based on artificial intelligence and deep learning, so as to capture the embedded semantic features of the task to be processed and the aggregation coding features between the semantics of multiple data samples, and then assign a task priority to the task to be processed through task semantic matching technology, thereby helping to achieve efficient task scheduling and resource allocation.

[0027] More specifically, as Figure 1 and Figure 2As shown, the multi-task intelligent scheduling management platform 1 includes: a to-be-processed task extraction module 10 for extracting to-be-processed tasks from a task queue; a historical task dataset extraction module 20 for extracting a historical task dataset from a background database, where each data sample in the historical task dataset includes a task content and a task priority; a data set splitting module 30 for splitting the historical task dataset based on the task priority to obtain multiple historical task data subsets; a historical task data semantic aggregation and encoding module 40 for performing semantic embedding encoding and semantic aggregation analysis on each data sample in each historical task data subset to obtain multiple historical task data subset semantic aggregation encoding vectors; a to-be-processed task semantic embedding encoding module 50 for performing semantic embedding encoding on the to-be-processed task to obtain a to-be-processed task semantic embedding encoding vector; a task semantic matching calculation module 60 for calculating the task semantic matching degrees between the to-be-processed task semantic embedding encoding vector and each historical task data subset semantic aggregation encoding vector in the multiple historical task data subset semantic aggregation encoding vectors to obtain multiple task semantic matching degrees; and a task priority specifying module 70 for specifying the task priority corresponding to the maximum value among the multiple task semantic matching degrees as the task priority of the to-be-processed task.

[0028] Exemplarily, in the to-be-processed task extraction module 10, to-be-processed tasks are extracted from the task queue. It should be understood that the purpose of extracting to-be-processed tasks from the task queue is to ensure that all tasks can be recognized, analyzed by the system, and reasonably allocated resources for processing according to the priority. This process not only helps improve the efficiency of task processing but also ensures optimal task scheduling and resource allocation in the case of limited resources. Specifically, in the e-commerce and retail industries, especially during promotional activities, enterprises may face the pressure of processing a large number of orders. These tasks are not only numerous but also have different levels of urgency, time requirements, and resource limitations (such as vehicles, personnel, etc.). Without systematic management and scheduling of these tasks, it is easy to cause problems such as resource waste and delivery delays, thus affecting the customer experience and service quality. Therefore, to-be-processed tasks are first extracted from the task queue.

[0029] Exemplarily, in the historical task dataset extraction module 20, a historical task dataset is extracted from the background database. Each data sample in the historical task dataset contains task content and task priority. It should be understood that in a complex business environment, such as the e-commerce or logistics distribution industry, a large number of task requests are generated every day. These tasks are not only numerous but also have different levels of urgency, time requirements, and resource needs. If only relying on manual experience or simple rules to determine task priorities and resource allocation, it is often difficult to cope with such complex and changeable situations, easily leading to resource waste or task delays. Therefore, by extracting the historical task dataset from the background database, the system can analyze based on real historical records and discover potential patterns and rules. For example, certain specific types of tasks may always be assigned a higher priority because they are usually associated with high-value customers or urgent needs; while other tasks can be flexibly arranged according to their completion time windows. In this way, through learning from historical data, the system can make more informed decisions when facing new tasks, ensure optimal utilization of resources, and at the same time improve the overall service efficiency and quality.

[0030] Exemplarily, in the data set splitting module 30, the historical task dataset is split into multiple historical task data subsets based on task priority. It should be understood that in the intelligent scheduling management platform, the historical task dataset usually contains a large number of task samples, and each sample has different task content and priority labels. To more efficiently process new tasks and assign priorities to them, the platform needs to learn the patterns of task priorities from the historical task data. Based on this, in the technical solution of this application, first, the historical task dataset is split into multiple historical task data subsets based on task priority, and semantic embedding encoding and semantic aggregation analysis are performed on each data sample in each historical task data subset to obtain multiple historical task data subset semantic aggregation encoding vectors. That is to say, the historical task dataset is divided into multiple historical task data subsets according to priority labels (such as high, medium, low), where each subset contains task samples with the same priority. This splitting method can reduce data complexity, enabling subsequent semantic analysis to be carried out separately for each priority category, thus more accurately capturing the semantic features of historical task data in different priority categories.

[0031] Exemplarily, in the historical task data semantic aggregation encoding module 40, semantic embedding encoding and semantic aggregation analysis are performed on each data sample in each subset of historical task data to obtain multiple semantic aggregation encoding vectors of the subsets of historical task data. It should be understood that semantic embedding encoding and semantic aggregation analysis are performed on each data sample in each subset of historical task data, so as to map each data sample in the subsets of historical task data with different priorities into a common semantic space and convert it into a high-dimensional vector representation, and then capture the aggregation features and common representations between the embedded semantics of each historical task data sample in each subset of historical task data with each priority, providing a basis for the subsequent priority assignment of the tasks to be processed.

[0032] In one embodiment, as Figure 3 shown, the historical task data semantic aggregation encoding module 40 includes: a historical task semantic embedding encoding unit 41, configured to perform semantic embedding encoding on each data sample in each subset of historical task data to obtain a subset of historical task semantic embedding encoding vectors; and a feature aggregation analysis unit 42, configured to perform feature aggregation analysis based on information kernel aggregation on the subset of historical task semantic embedding encoding vectors to obtain the semantic aggregation encoding vectors of the subsets of historical task data.

[0033] Specifically, in the historical task semantic embedding encoding unit 41, for each subset of historical task data with each priority, first, semantic embedding encoding is performed on each data sample in each subset of historical task data to obtain a subset of historical task semantic embedding encoding vectors. That is to say, semantic embedding encoding is performed on each data sample in each subset of historical task data to convert the task content into a high-dimensional vector representation. For example, the delivery address is mapped to different positions in the high-dimensional vector space using the BERT or Word2Vec model, and the commodity information is mapped to another high-dimensional vector. The semantic embedding vector of each task data sample captures the embedded semantic information of its task content, providing a basis for subsequent aggregation analysis and priority matching and assignment of the tasks to be processed.

[0034] Specifically, in the feature aggregation analysis unit 42, in order to capture the common representation among the historical task data semantics in each task priority set, it is necessary to perform aggregation analysis on the historical task data in each task priority set. However, directly performing global aggregation (such as mean pooling) on a subset of the historical task semantic embedding vectors will lose the personalized information of key nodes. For example, a high-priority task may need special processing due to special delivery addresses (such as remote areas), but simple global averaging will dilute this feature. Based on this, in the technical solution of this application, further perform feature aggregation analysis based on information kernel aggregation on the subset of the historical task semantic embedding coding vectors to obtain the semantic aggregation coding vectors of the historical task data subset. Specifically, the processing core of the feature aggregation analysis based on information kernel aggregation is phased processing, from coarse-grained to fine-grained, combining the global and local historical task data semantic association features. That is to say, construct global features through coarse-grained aggregation, and then use the kernel aggregation compensation factor and the fine-grained dynamic compensation mechanism to retain local details, and finally generate a global and local semantic aggregation feature representation of the historical task data subset. In this way, the balance problem of the global and local features of the semantic association of the historical task data in each priority set is solved, and the advantages are multi-level feature integration and dynamic adjustment. It not only effectively captures the statistical representation of global features, but also carefully enhances the significance of the personalized features of the semantics of each historical task data, thus realizing the integration of multi-level and multi-scale complex historical task data semantic features, providing a basis for the subsequent priority matching task of new tasks.

[0035] In one embodiment, as Figure 4 shown, the feature aggregation analysis unit 42 includes: a historical task semantic coarse-grained aggregation coding subunit 421, configured to input a subset of the historical task semantic embedding coding vectors into an information kernel coarse-grained aggregation network to obtain a historical task semantic coarse-grained aggregation coding vector; a kernel aggregation compensation weight factor determination subunit 422, configured to determine the kernel aggregation compensation weights of the respective historical task semantic embedding coding vectors in the subset of the historical task semantic embedding coding vectors based on the historical task semantic coarse-grained aggregation coding vector to obtain a set of kernel aggregation compensation weight factors; a dynamic compensation aggregation analysis subunit 423, configured to perform dynamic compensation aggregation analysis based on the node fine-grained features on the historical task semantic coarse-grained aggregation coding vector and the subset of the historical task semantic embedding coding vectors based on the set of kernel aggregation compensation weight factors to obtain a historical task semantic fine-grained compensation aggregation coding vector; a historical task data aggregation subunit 424, configured to input the historical task semantic fine-grained compensation aggregation coding vector and the historical task semantic coarse-grained aggregation coding vector into a residual unit to obtain the semantic aggregation coding vector of the historical task data subset.

[0036] Specifically, the calculation process of the historical task semantic coarse-grained aggregation encoding subunit 421 can be expressed by the formula as follows:

[0037]

[0038] Wherein, is a subset of the historical task semantic embedding encoding vectors, are respectively the 1st, 2nd, th, and th historical task semantic embedding encoding vectors in the subset of the historical task semantic embedding encoding vectors, and represent taking the maximum and minimum values in the vector, represents the global feature information kernel factor, is function, is the global feature information kernel weight, is the number of vectors in the subset of the historical task semantic embedding encoding vectors, is the historical task semantic coarse-grained aggregation encoding vector.

[0039] It should be understood that the historical task semantic embedding encoding vectors in these subsets are input into the information kernel coarse-grained aggregation network. The role of this network is to generate a coarse-grained aggregation encoding vector representing the overall characteristics of the subset through compressive modeling of the global features of all historical tasks in each subset. Specifically, the information kernel coarse-grained aggregation network captures the complex similarity structure between nodes (i.e., the semantic embedding encoding vectors of each historical task) through a non-linear mapping mechanism, thereby forming a highly condensed global feature representation. This step is equivalent to providing an "overview map" for each priority category, enabling the overall characteristics of a certain type of historical task to be quickly identified at the macroscopic level. In this process, the concept of the information kernel plays a key role. It not only helps to establish a node relationship model in the high-dimensional space but also effectively captures the similarities and differences between different historical tasks. For example, in the subset of high-priority historical tasks, certain specific types of historical tasks (such as emergency deliveries for VIP customers) may always occupy an important position, and this pattern can be accurately identified by the information kernel coarse-grained aggregation network. In this way, it is possible to better understand which factors determine the high priority of a certain historical task and accordingly formulate corresponding scheduling strategies.

[0040] In one embodiment, as Figure 5As shown, the kernel convergence compensation weight factor determination subunit 422 includes: a kernel convergence compensation factor calculation secondary subunit 422-1, configured to calculate kernel convergence compensation factors of respective historical task semantic embedding encoding vectors in a subset of the historical task semantic embedding encoding vectors relative to the historical task semantic coarse-grained convergence encoding vector to obtain a set of kernel convergence compensation factors; and a compensation explicit modeling processing secondary subunit 422-2, configured to perform compensation explicit modeling based on a gating function on the set of kernel convergence compensation factors to obtain a set of the kernel convergence compensation weight factors.

[0041] In one embodiment, as Figure 6 shown, the kernel convergence compensation factor calculation secondary subunit 422-1 includes: a vector modulation tertiary subunit 422-11, configured to modulate the historical task semantic embedding encoding vector and the historical task semantic coarse-grained convergence encoding vector to obtain a historical task semantic embedding encoding modulation vector and a historical task semantic coarse-grained convergence encoding modulation vector; a compensation information calculation tertiary subunit 422-12, configured to calculate compensation information between the historical task semantic embedding encoding modulation vector and the historical task semantic coarse-grained convergence encoding modulation vector to obtain a historical task semantic kernel convergence compensation weight vector; a historical task semantic deviation compensation factor calculation tertiary subunit 422-13, configured to calculate a historical task semantic deviation compensation factor based on the historical task semantic embedding encoding vector and the historical task semantic coarse-grained convergence encoding vector; and a kernel convergence compensation factor calculation tertiary subunit 422-14, configured to calculate a kernel convergence compensation factor based on the historical task semantic kernel convergence compensation weight vector and the deviation compensation factor.

[0042] It should be understood that relying solely on the coarse-grained aggregation coding vector is not sufficient to fully describe the unique characteristics of each task. To preserve the personalized information of each task, a kernel aggregation compensation mechanism is introduced. Specifically, based on the historical task semantic coarse-grained aggregation coding vector, the deviation between each historical task semantic embedding coding vector and this coarse-grained aggregation coding vector is calculated to determine the kernel aggregation compensation weight for each task. This process can be regarded as a correction rule for global features, aiming to ensure that the unique information of each task is not lost due to the compression of global features. In this process, by measuring the difference between each node feature vector and the coarse-grained aggregation coding vector, a compensation factor for each task is dynamically generated. These compensation factors are essentially a quantitative description of the deviation degree of each task relative to the overall pattern. For example, in a subset of high-priority tasks, certain specific types of tasks (such as the urgent delivery of VIP customers) may be significantly different from most tasks, and these differences can be accurately captured by the compensation factors. In this way, even when faced with a large number of similar tasks, those tasks that require special attention can be identified and corresponding resource tilts can be given. Next, these compensation factors are used to construct a set of kernel aggregation compensation weight factors. This set provides a personalized adjustment coefficient for each task, enabling more accurate decisions to be made in subsequent task scheduling and resource allocation. In this way, not only can the understanding of the overall pattern be maintained, but also the unique needs of each task can be flexibly addressed. This dynamic adjustment mechanism based on kernel aggregation compensation weights is similar to the attention mechanism, which allows for the coordinated expression of features at both the global and local levels. For example, when processing high-priority tasks, resources can be flexibly allocated according to the compensation weight factors to ensure that critical tasks are processed in a timely manner while reasonably arranging the scheduling order of other tasks. In this way, not only can the overall service efficiency be improved, but also resource waste and delays can be effectively avoided. In addition, the application of kernel aggregation compensation weight factors also enhances the robustness and adaptability of the model. In actual operation, the requirements and priorities of tasks may fluctuate over time. By introducing a compensation mechanism, these changes can be responded to more quickly and corresponding adjustments can be made.

[0043] Specifically, the calculation process of the vector modulation three-level subunit 422-11 can be expressed by the formula as follows:

[0044]

[0045] Among them, is the processing of the point convolution layer, is the first weight matrix, is the second weight matrix, is function, is the A semantic embedding encoding modulation vector for historical tasks, is a semantic coarse-grained aggregation encoding modulation vector for historical tasks.

[0046] Specifically, the calculation process of the compensation information calculation three-level subunit 422-12 can be expressed by the formula as:

[0047]

[0048] Wherein, is subtraction by position, is taking the absolute value, is a semantic kernel aggregation compensation weight vector for historical tasks.

[0049] In one embodiment, the historical task semantic deviation compensation factor calculation three-level subunit is used to: in response to the two-norm of the historical task semantic embedding encoding vector being less than the two-norm of the historical task semantic coarse-grained aggregation encoding vector, calculate the result of dividing the two-norm of the historical task semantic embedding encoding vector by the two-norm of the historical task semantic coarse-grained aggregation encoding vector and then adding the constant one, and calculate the base-2 logarithmic function value of the obtained value; in response to the two-norm of the historical task semantic embedding encoding vector being greater than or equal to the two-norm of the historical task semantic coarse-grained aggregation encoding vector, calculate the result of dividing the two-norm of the historical task semantic embedding encoding vector by the two-norm of the historical task semantic coarse-grained aggregation encoding vector. Specifically, this process can be expressed by the formula as:

[0050]

[0051] Wherein, represents the two-norm of the vector, is the base-2 logarithmic function value, represents the historical task semantic deviation compensation factor.

[0052] Specifically, here, for the deviation compensation factor between the historical task semantic embedding coding vector and the historical task semantic coarse-grained aggregation coding vector, the performance deviation of the kernel aggregation strategy as a scenario strategy can be measured by quantifying the regret metric based on the information kernel compression hypothesis in the kernel aggregation decision-making process, that is, the game-theoretic counterfactual regret value. Specifically, through the vector norm representation, a normalized decision point loss description based on the policy action is provided for the counterfactual regret value, that is, the vector norm representation of the historical task semantic embedding coding vector and the historical task semantic coarse-grained aggregation coding vector. Then, for the possible differences in the vector distribution action game scenarios, the compensation rule correction of the node personalized information is carried out respectively with the information distribution degree of the regret value and the relative distribution amplitude of the regret value, so as to consider the node personalized information as the unselected action in the decision-making, and perform the bias compensation in the way of assuming its potential benefit based on the information kernel aggregation hypothesis.

[0053] Specifically, the calculation process of the kernel aggregation compensation factor of the three-level subunit 422-14 can be expressed by the formula as follows:

[0054]

[0055] Wherein, and respectively represent the compensation weight matrix and the compensation bias vector, is the compensation modulation vector, is the corresponding kernel aggregation compensation factor.

[0056] In one embodiment, the compensation explicit modeling processing two-level subunit 422-2 is configured to: in response to the kernel aggregation compensation factor in the set of kernel aggregation compensation factors being greater than or equal to a predetermined threshold, input the kernel aggregation compensation factor into the sigmoid function; in response to the kernel aggregation compensation factor in the set of kernel aggregation compensation factors being less than the predetermined threshold, set the kernel aggregation compensation factor to 0. Specifically, this process can be expressed by the formula as follows:

[0057]

[0058] Wherein, is the preset threshold, is the compensation explicit modeling operation, is the corresponding kernel aggregation compensation weight factor.

[0059] Specifically, further optimization is performed on these kernel convergence compensation factors, namely, explicit modeling of compensation based on a gating function. In this process, the role of the gating function is to dynamically select and regulate information, ensuring that only those local features that make significant contributions to the overall model are highlighted, while irrelevant or redundant information is appropriately suppressed. In this way, coordinated expression of features can be achieved at both the global and local levels, ensuring that the unique requirements of each task are fully considered. Specifically, the gating function explicitly models the compensation factors, screens out those compensation factors that have a significant impact on the final encoding result, and assigns them appropriate weights. This step not only improves the compactness and significance of feature aggregation but also enhances the model's ability to capture complex relationships. For example, in a subset of high-priority tasks, certain specific types of tasks (such as emergency deliveries for VIP customers) may have unique attributes that can be accurately identified and emphasized by the gating function. In this way, even when faced with a large number of similar tasks, those tasks that require special attention can be identified and given corresponding resource allocation.

[0060] In one embodiment, the dynamic compensation convergence analysis sub-unit 423 is configured to: input the set of the kernel convergence compensation weight factors, the historical task semantic coarse-grained convergence encoding vector, and the subset of the historical task semantic embedded encoding vectors into a node fine-grained dynamic compensation convergence network to obtain a historical task semantic fine-grained compensation convergence encoding vector. Specifically, this process can be represented by the formula:

[0061]

[0062] where is the historical task semantic fine-grained compensation convergence encoding vector.

[0063] It should be understood that in the process of performing dynamic compensation aggregation analysis based on node fine-grained features on a subset of the historical task semantic coarse-grained aggregation encoding vector and the historical task semantic embedding encoding vector based on this set of kernel aggregation compensation weight factors, by dynamically adjusting each node feature vector, a historical task semantic fine-grained compensation aggregation encoding vector is generated. The core of this process lies in introducing high fidelity, enabling the compensated features to more accurately describe the local details of the nodes while adapting to the dynamic changes of the global feature constraints. Specifically, the position-wise difference between each historical task semantic embedding encoding vector and its corresponding coarse-grained aggregation encoding vector is calculated to form a set of historical task semantic aggregation deviation vectors. Then, using the kernel aggregation compensation weight factors as weights, the weighted sum of these deviation vectors is calculated to generate the historical task semantic fine-grained compensation aggregation encoding vector. This process is similar to the attention mechanism, allowing for coordinated expression of features at both the global and local levels. For example, in a subset of high-priority tasks, certain specific types of tasks (such as emergency deliveries for VIP customers) may have unique attributes that can be accurately captured through fine-grained compensation aggregation analysis. Even in the face of a large number of similar tasks, those tasks that require special attention can be identified and corresponding resource allocation can be given. In this way, not only can the overall pattern be understood, but the unique needs of each task can also be flexibly addressed.

[0064] In one embodiment, inputting the set of the kernel aggregation compensation weight factors, the historical task semantic coarse-grained aggregation encoding vector, and the subset of the historical task semantic embedding encoding vector into a node fine-grained dynamic compensation aggregation network to obtain a historical task semantic fine-grained compensation aggregation encoding vector includes: calculating the position-wise difference between each historical task semantic embedding encoding vector in the subset of the historical task semantic embedding encoding vector and the historical task semantic coarse-grained aggregation encoding vector to obtain a set of historical task semantic aggregation deviation vectors; using the set of the kernel aggregation compensation weight factors as weights, calculating the weighted sum of the historical task semantic aggregation deviation vectors in the set of the historical task semantic aggregation deviation vectors to obtain the historical task semantic fine-grained compensation aggregation encoding vector.

[0065] Specifically, the calculation process of the historical task data aggregation sub-unit 424 can be expressed by the formula as follows:

[0066]

[0067] Wherein, and are residual weight coefficients, is the historical task data subset semantic aggregation encoding vector.

[0068] It should be understood that the historical task semantic fine-grained compensation aggregated encoding vector and the historical task semantic coarse-grained aggregated encoding vector are input into the residual unit to generate the historical task data subset semantic aggregation encoding vector. The core of this process lies in fusing global and local features to ensure that both the overall pattern can be captured and the unique attributes of each task can be reflected in detail. Specifically, the design of the residual unit is not only widely used in deep learning, but its significance here lies more in an effective mechanism for information fusion. Through residual connections, the problem of gradient disappearance can be avoided, enabling the features after fine-grained dynamic adjustment to not only inherit the global perspective of the coarse-grained features but also retain the locally significant details. This design forms a refined feature expression form that combines the global and the local. In actual operation, assume that we have a high-priority task subset that includes a large number of urgent delivery tasks for VIP customers. Although these tasks have some common features (such as high priority), the specific requirements of each task may vary (such as specific delivery time windows or special delivery requirements). By inputting the historical task semantic fine-grained compensation aggregated encoding vector and the historical task semantic coarse-grained aggregated encoding vector into the residual unit, it is possible to maintain an understanding of the overall features of high-priority tasks at the macroscopic level while accurately capturing the unique needs of each task at the microscopic level.

[0069] Exemplarily, in the to-be-processed task semantic embedding encoding module 50, the to-be-processed task is subjected to semantic embedding encoding to obtain a to-be-processed task semantic embedding encoding vector. That is, similarly, the to-be-processed task is subjected to semantic embedding encoding to obtain a to-be-processed task semantic embedding encoding vector. Through semantic embedding encoding, the to-be-processed task can be embedded and mapped into a high-dimensional space, thereby being transformed into a vector representation and capturing the embedding semantics of the to-be-processed task.

[0070] Exemplarily, in the task semantic matching calculation module 60, the task semantic matching degrees between the to-be-processed task semantic embedding encoding vector and each of the historical task data subset semantic aggregation encoding vectors in the multiple historical task data subset semantic aggregation encoding vectors are calculated respectively to obtain multiple task semantic matching degrees. It should be understood that the to-be-processed task semantic embedding encoding vector represents the semantic information of a new task (such as delivery address, commodity information, etc.), and is transformed into a high-dimensional vector through semantic embedding encoding technologies (such as BERT, Word2Vec). Each of the historical task data subset semantic aggregation encoding vectors in the multiple historical task data subset semantic aggregation encoding vectors represents the representative semantic features of each priority subset (for example, the high-priority subset may include features such as "urgent delivery" and "VIP customer"), and is generated through semantic aggregation analysis. In order to determine which priority subset the new task best matches, in the technical solution of this application, the task semantic matching calculation module is configured to: input the to-be-processed task semantic embedding encoding vector and each of the historical task data subset semantic aggregation encoding vectors in the multiple historical task data subset semantic aggregation encoding vectors into a task semantic matching network respectively to obtain the multiple task semantic matching degrees. In a specific embodiment, the task semantic matching network calculates the task semantic matching degrees between the to-be-processed task semantic embedding encoding vector and each of the historical task data subset semantic aggregation encoding vectors in the multiple historical task data subset semantic aggregation encoding vectors through an attention mechanism or cosine similarity.

[0071] Exemplarily, in the task priority specifying module 70, the task priority corresponding to the maximum value among the multiple task semantic matching degrees is specified as the task priority of the to-be-processed task. It should be understood that the historical task data set contains a large number of tasks that have been successfully processed in the past and their corresponding priority labels. These historical data provide a valuable reference basis for the priority assignment of the current task. By matching the historical task patterns, it is possible to draw on past experience and make more scientific and reasonable decisions. At the same time, matching the to-be-processed task with the most similar historical task subset and assigning the corresponding priority can ensure the optimal allocation of resources. For example, during a promotional event, the system can reasonably arrange delivery vehicles and personnel according to the urgency and special requirements of the tasks, avoiding resource waste and delays.

[0072] In a specific embodiment, first, find the maximum value from multiple task semantic matching degrees. Then, according to the priority of the historical task data subset corresponding to the maximum matching degree, assign the corresponding priority to the task to be processed. For example, if the maximum matching degree appears in the high-priority subset, set the priority of the task to be processed as "high". In this way, the task priority can be assigned to the task to be processed through the task semantic matching technology, which helps to achieve efficient task scheduling and resource allocation.

[0073] In summary, the multi-task intelligent scheduling management platform according to the embodiments of the present application is elucidated. It analyzes each data sample in the historical task dataset and the task to be processed by adopting data processing and natural language processing algorithms based on artificial intelligence and deep learning, so as to capture the embedded semantic features of the task to be processed and the aggregated coding features between the semantics of multiple data samples. Then, the task priority is assigned to the task to be processed through the task semantic matching technology, which helps to achieve efficient task scheduling and resource allocation.

[0074] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraint conditions of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but this implementation should not be considered to exceed the scope of the present application.

[0075] It should be understood that the specific examples herein are only to help those skilled in the art better understand the embodiments of the present application, rather than limiting the scope of the embodiments of the present application.

[0076] It should also be understood that in various embodiments of the present application, the magnitudes of the serial numbers of the various processes do not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0077] It should also be understood that the various implementation manners described in this specification can be implemented alone or in combination, and the embodiments of the present application do not limit this.

[0078] Unless otherwise specified, all technical and scientific terms used in the embodiments of this application have the same meanings as those commonly understood by those skilled in the technical field of this application. The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit the scope of this application. The term "and / or" used in this application includes any and all combinations of one or more of the related listed items. The singular forms "a", "above-mentioned", and "the" used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0079] In several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0080] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0081] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0082] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0083] As described above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A multi-task intelligent scheduling management platform, characterized in that: include: A pending task extraction module is used to extract pending tasks from the task queue; A historical task data set extraction module is used to extract a historical task data set from a background database, wherein each data sample in the historical task data set includes task content and task priority; A data set splitting module, used for splitting the historical task data set based on task priority to obtain multiple historical task data subsets; A historical task data semantic aggregation coding module is used to perform semantic embedding coding and semantic aggregation analysis on each data sample in each historical task data subset to obtain multiple historical task data subset semantic aggregation coding vectors; A semantic embedding coding module for tasks to be processed, used for performing semantic embedding coding on the tasks to be processed to obtain semantic embedding coding vectors for the tasks to be processed; A task semantic matching calculation module is used to respectively calculate the task semantic matching degree between the semantic embedding coding vector of the task to be processed and each historical task data subset semantic aggregation coding vector in the multiple historical task data subset semantic aggregation coding vectors to obtain multiple task semantic matching degrees; A task priority designation module, configured to designate a task priority corresponding to a task with the largest semantic matching degree among the plurality of tasks as the task priority of the task to be processed; The historical task data semantic aggregation coding module includes: A historical task semantic embedding coding unit, used for performing semantic embedding coding on each data sample in each historical task data subset to obtain a subset of historical task semantic embedding coding vectors; A feature aggregation analysis unit, used for performing feature aggregation analysis based on information core aggregation on a subset of the historical task semantic embedding coding vector to obtain a semantic aggregation coding vector of the historical task data subset; Wherein, the feature aggregation analysis unit includes: A historical task semantics coarse-grained convergence encoding subunit, used for inputting a subset of the historical task semantics embedding encoding vectors into an information core coarse-grained convergence network to obtain a historical task semantics coarse-grained convergence encoding vector; A core convergence compensation weight factor determination subunit is used to determine the core convergence compensation weight of each historical task semantic embedding coding vector in the subset of the historical task semantic embedding coding vector based on the historical task semantic coarse-grained convergence coding vector to obtain a set of core convergence compensation weight factors; A dynamic compensation convergence analysis subunit is used to perform a dynamic compensation convergence analysis based on node fine-grained features on a subset of the historical task semantics coarse-grained converged coding vector and the historical task semantics embedded coding vector based on the set of the core convergence compensation weight factors to obtain a historical task semantics fine-grained compensation convergence coding vector; The historical task data aggregation subunit is used to input the historical task semantic fine-grained compensation aggregation coding vector and the historical task semantic coarse-grained aggregation coding vector into the residual unit to obtain the historical task data subset semantic aggregation coding vector.

2. The multi-task intelligent scheduling management platform according to claim 1 is characterized in that: The core convergence compensation weight factor determination subunit includes: A secondary subunit for calculating a kernel convergence compensation factor is used to calculate a kernel convergence compensation factor of each historical task semantic embedding coding vector in a subset of the historical task semantic embedding coding vector relative to the historical task semantic coarse-grained convergence coding vector to obtain a set of kernel convergence compensation factors; The compensation explicit modeling processing secondary subunit is used to perform compensation explicit modeling based on a gating function on the set of core convergence compensation factors to obtain the set of core convergence compensation weight factors.

3. The multi-task intelligent scheduling management platform according to claim 2 is characterized in that: The core convergence compensation factor calculation secondary subunit includes: A vector modulation tertiary subunit, used for modulating the historical task semantic embedding coding vector and the historical task semantic coarse-grained aggregation coding vector to obtain a historical task semantic embedding coding modulation vector and a historical task semantic coarse-grained aggregation coding modulation vector; The compensation information calculation tertiary subunit is used to calculate the compensation information between the historical task semantic embedding coding modulation vector and the historical task semantic coarse-grained convergence coding modulation vector to obtain the historical task semantic core convergence compensation weight vector; A historical task semantic deviation compensation factor calculation tertiary subunit is used to calculate the historical task semantic deviation compensation factor based on the historical task semantic embedding coding vector and the historical task semantic coarse-grained convergence coding vector; The third-level sub-unit for calculating the kernel convergence compensation factor is used to calculate the kernel convergence compensation factor based on the historical task semantic kernel convergence compensation weight vector and the deviation compensation factor.

4. The multi-task intelligent scheduling management platform according to claim 3 is characterized in that: The historical task semantic deviation compensation factor calculation three-level sub-unit is used to: In response to the binary norm of the historical task semantic embedding coding vector being less than the binary norm of the historical task semantic coarse-grained converged coding vector, the binary norm of the historical task semantic embedding coding vector is calculated and divided by the binary norm of the historical task semantic coarse-grained converged coding vector, and then added to a constant 1, and a logarithmic function value with base 2 of the numerical value is calculated; In response to the binary norm of the historical task semantic embedding coding vector being greater than or equal to the binary norm of the historical task semantic coarse-grained aggregation coding vector, the binary norm of the historical task semantic embedding coding vector is calculated by dividing the binary norm of the historical task semantic coarse-grained aggregation coding vector.

5. The multi-task intelligent scheduling management platform according to claim 4 is characterized in that: The compensation explicit modeling processing secondary subunit is used to: in response to the core convergence compensation factor in the set of core convergence compensation factors being greater than or equal to a predetermined threshold, input the core convergence compensation factor into a sigmoid function; in response to the core convergence compensation factor in the set of core convergence compensation factors being less than a predetermined threshold, set the core convergence compensation factor to 0.

6. The multi-task intelligent scheduling management platform according to claim 5, characterized in that: The dynamic compensation convergence analysis subunit is used to: input the set of core convergence compensation weight factors, the historical task semantic coarse-grained convergence coding vector and a subset of the historical task semantic embedded coding vector into the node fine-grained dynamic compensation convergence network to obtain the historical task semantic fine-grained compensation convergence coding vector.

7. The multi-task intelligent scheduling management platform according to claim 6, characterized in that: Inputting the set of the core convergence compensation weight factors, the historical task semantics coarse-grained converged coding vector and a subset of the historical task semantics embedded coding vector into a node fine-grained dynamic compensation convergence network to obtain a historical task semantics fine-grained compensation converged coding vector, including: Calculate the positional difference between each historical task semantic embedding coding vector in the subset of the historical task semantic embedding coding vector and the historical task semantic coarse-grained converged coding vector to obtain a set of historical task semantic converged deviation vectors; Taking the set of core convergence compensation weight factors as weights, the weighted sum of each historical task semantic convergence deviation vector in the set of historical task semantic convergence deviation vectors is calculated to obtain the historical task semantic fine-grained compensation convergence encoding vector.

8. The multi-task intelligent scheduling management platform according to claim 7 is characterized in that: The task semantic matching calculation module is used to: input the semantic embedding coding vector of the task to be processed and each historical task data subset semantic aggregation coding vector in the multiple historical task data subset semantic aggregation coding vectors into the task semantic matching network respectively to obtain the multiple task semantic matching degrees.

Images