Multi-task processing method and system based on large model

Abstract

The invention relates to the technical field of task scheduling, in particular to a multi-task processing method and system based on a large model, and the method comprises the following steps: extracting resource parameters by using the large model to generate a conflict intensity matrix, screening resource coincident node pairs, judging structural conflicts, constructing a dependency graph, calculating complexity, scoring and cutting a dense region, and extracting a high-frequency path division scheduling unit, and generating a task priority sequence. According to the method, the conflict intensity matrix is constructed, the task resource conflict relation is recognized in combination with node aggregation features, the sensitivity and precision of task division to resource distribution are improved, upper and lower bound nodes are divided based on median deviation, structure optimization and recombination are achieved in combination with the resource type mutual difference rate, and the path depth and branch factors form a complexity scoring basis. The method comprises the following steps: dynamically adjusting weight identification and cutting a dense area in combination with an occupation time length, calling a path with the highest frequency to form a main path set, dividing non-main path tasks according to a node coverage relationship, and enhancing scheduling boundary definition and resource isolation capability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of task scheduling, in particular to a multi-task processing method and system based on a large model. BACKGROUND

[0002] The technical field of task scheduling includes systematic research and implementation of the execution order, resource allocation, and optimization of the execution process of computing tasks. The core content of this technical field is to design a reasonable task allocation mechanism to address the problems of resource conflicts, increased delays, and decreased system efficiency caused by multi-task parallel execution. It mainly involves task dispatching strategies of operating systems, dynamic allocation methods of computing resources, and concurrency control in multi-processor systems. Task scheduling is widely used in multi-core processor environments, distributed computing architectures, and cloud computing platforms. Through the setting of different task priorities, execution timing control, and scheduling algorithm selection, it effectively supports the stable and efficient operation of computing systems.

[0003] Among them, the multi-task processing method refers to the task organization and scheduling strategy designed to improve the efficiency of computing resources and task execution in a multi-core processing or distributed system environment. This patent subject addresses the resource competition between multiple tasks, processing priority determination, and dynamic load balancing issues. It uses a task division mechanism based on pre-set strategies, constructs a task dependency graph to clarify the execution order, and uses a classification discrimination model to prioritize tasks based on historical task execution data. During task execution, it dynamically monitors resource occupation and adjusts the scheduling strategy to achieve reasonable distribution of tasks among different computing nodes and real-time management during the execution process.

[0004] Traditional multi-task processing technology handles resource conflicts through fixed task allocation strategies, relies on static data for task priority division, lacks real-time perception of task running state and behavior, and cannot accurately depict resource overlap relationships. Task structure splitting is based on logical processes rather than resource characteristics, and node distribution has a high stacking problem, increasing the probability of conflicts. The dependency graph only reflects the static order between tasks and does not have complexity dimension analysis, making it difficult to implement fine-grained optimization in path structure-intensive areas. The call frequency in task path identification is not considered, leading to a deviation of the scheduling main path from the real execution hotspot, resulting in resource scheduling deviation and path scheduling conflicts, and significant increases in task processing chain delay fluctuations. SUMMARY

[0005] To solve the technical problems existing in the prior art, the embodiments of the present application provide a multi-task processing method and system based on a large model. The technical solution is as follows: To achieve the above purpose, the present application adopts the following technical scheme: a multi-task processing method based on a large model, comprising the following steps: S1: obtaining resource type code, access frequency and occupation time of the task node by using a large model, calculating the Euclidean distance of the resource demand vector to obtain a conflict intensity matrix, screening resource overlapping node pairs by using a preset threshold, judging structural conflict according to node aggregation characteristics, calling a density clustering algorithm to evaluate density, and generating a reconfigurable task node structure body; S2: according to the reconfigurable task node structure body, dividing upper and lower boundary nodes by using a conflict intensity median deviation parameter, extracting upper boundary nodes to construct independent units, matching lower boundary nodes and fusing according to resource type mutual difference rate, and generating a primary task scheduling unit set; S3: extracting resource access channel number and scheduling trigger order based on the primary task scheduling unit set, constructing a dependency graph and calculating node path depth and branch factor to obtain a complexity score, adjusting weight according to occupation time, pruning dense area execution subgraph and node aggregation according to the score, and generating a dependency subgraph set; S4: according to the node identification mapping relationship between the dependency subgraph set and the primary task scheduling unit set, extracting the highest call frequency path to form a main path set, dividing the remaining nodes to form a scheduling group, and generating a task structure division result.

[0006] As a further scheme of the application, the reconfigurable task node structure body includes node clustering labels, resource conflict labels and structure adjustment indication information, the primary task scheduling unit set includes fusion node indexes, independent unit identifiers and unit structure configuration parameters, the dependency subgraph set includes node dependency relationship graphs, task path hierarchical structures and aggregated subgraph identifiers, and the task structure division result includes main path node indexes, grouping boundary lists and structure level numbers.

[0007] As a further scheme of the application, obtaining resource type code, access frequency and occupation time of the task node by using a large model, calculating the Euclidean distance of the resource demand vector to obtain a conflict intensity matrix, screening resource overlapping node pairs by using a preset threshold, judging structural conflict according to node aggregation characteristics, calling a density clustering algorithm to evaluate density, and generating a reconfigurable task node structure body, the steps are specifically as follows: S101: obtaining resource type code, access frequency and occupation time of the task node by using a large model, obtaining vector expression according to resource dimension group, calculating distance values of a plurality of node pairs on each resource dimension, constructing a symmetric conflict intensity matrix by combining the distance of each dimension, and generating a resource distance intensity matrix value; S102: According to the resource distance strength matrix value, a resource coincidence judgment threshold is called, task node combinations with a resource distance less than the resource coincidence judgment threshold in the node pair are screened, and aggregation classification is performed according to the resource type coding information of the node mapping, the number of aggregated nodes under each resource code is identified, the frequency distribution of each resource combination in the task set is calculated, and a resource coincidence node distribution rate is obtained; The resource coincidence judgment threshold is the mean value of the resource distance of all node pairs minus the standard deviation. S103: The resource coincidence node distribution rate is called, the access frequency and occupation time length data of each node in the resource combination are called, the DBSCAN density clustering algorithm is called, the local resource access density value is calculated, the high-density resource structure is marked, the connected relationship graph is constructed and the high-density structure is classified and aggregated, and the reconfigurable task node structure body is obtained. The DBSCAN density clustering algorithm determines the search radius based on the k-neighbor distance curve inflection point method, sets the minimum neighborhood node number by combining the dimension multiple method and the task size adjustment method, calculates the local density between nodes, identifies the high-density resource structure, and aggregates the reconfigurable task node structure body. The local resource access density value is obtained by constructing a two-dimensional vector of the access frequency and occupation time length of the task node, setting a search radius and a minimum neighborhood node number in the vector space, counting the number of nodes in the neighborhood and analyzing the local density.

[0008] As a further scheme of the application, according to the reconfigurable task node structure body, the upper and lower boundary nodes are divided by using the conflict strength median deviation parameter, the upper boundary nodes are extracted to construct independent units, the lower boundary nodes are matched and fused according to the resource type mutual difference rate, and the steps of generating the primary task scheduling unit set are as follows: S201: According to the resource conflict strength value corresponding to each group of nodes in the reconfigurable task node structure body, the median value of the conflict strength is set according to the median deviation parameter, the offset difference between the resource conflict value of each node and the median value is calculated, the offset direction is identified, and the nodes with a positive offset direction are extracted as upper boundary nodes, and the nodes with a negative offset direction are extracted as lower boundary nodes. The number and corresponding structure position information of the upper and lower boundary nodes are recorded, and the upper and lower boundary node distribution quantity is generated. The median deviation parameter is obtained by extracting the resource conflict strength value of each node, calculating the median, extracting the difference between the value of each node and the median, obtaining the offset difference and dividing the offset direction, and identifying the upper boundary node and the lower boundary node. S202: Based on the distribution of the upper and lower bound nodes, extract the set of upper bound nodes, identify the resource type code and original position information in the structure of each upper bound node, establish independent index relationships between nodes and mark the corresponding resource attribute items in the index path, call the resource usage frequency difference between nodes, calculate the resource isolation ratio, construct the resource independent partition structure of the upper bound nodes according to the isolation ratio, and generate the resource utilization separation index. The resource isolation ratio is calculated by taking the difference in access frequency of each node in the upper bound node set for the same resource type, and normalizing the difference to the ratio of the average access frequency. This ratio is used to quantify the heterogeneity of resource usage and the construction of an independent resource partitioning structure.

[0009] S203: Call the index information and node resource code content corresponding to each upper bound node in the resource utilization separation index, identify the lower bound nodes that have direct dependency connections with the upper bound nodes and continuous structure numbers according to the structure number and connected edge relationship in the index path, extract the resource type code of the corresponding lower bound node, calculate the resource type dissimilarity rate according to the number of resource type differences and the total number of resource items, and perform fusion matching and node reorganization on the upper and lower bound nodes according to the dissimilarity rate value to generate a primary task scheduling unit set; The resource utilization separation index is obtained by extracting the resource type code and resource usage frequency of the upper bound node, calculating the difference in access frequency between nodes on the same resource type and normalizing it to the isolation ratio, constructing an independent resource partitioning structure, and obtaining the resource utilization separation index. The resource type dissimilarity rate is obtained by identifying the proportion of non-overlapping resource types between upper and lower bound nodes to the total number of resource types, and by pairing and merging them according to the node resource code and structure number.

[0010] As a further aspect of the present invention, the specific formula for calculating the resource type dissimilarity rate based on the number of resource type differences and the total number of resource items is as follows: ; Calculate the resource type dissimilarity rate; in, The resource type distinctness rate is expressed in dimensionless form. The total number of resource types defined in the system. For the index number of the resource type, For the index number of the resource type, This is the index number of the lower bound node. This represents the total number of lower bound nodes extracted so far. For the first Does the nth lower bound node have the nth lower bound node? The status value of the resource class is in dimensionless form. For the first whether the first lower bound node has the first a state value of the resource of the first the first lower bound node a proportion value of the resource of the first in the node resource structure or an importance weight of the resource type, which is dimensionless, the first lower bound node a proportion value of the resource of the first in the node resource structure or an importance weight of the resource type, which is dimensionless.

[0011] As a further scheme of the application, the steps of extracting a resource access channel number and a scheduling trigger sequence based on the primary task scheduling unit set, constructing a dependency graph and calculating a node path depth and a branch factor to obtain a complexity score, adjusting a weight in combination with an occupation time length, pruning a dense area execution subgraph and node aggregation according to the score, and generating a dependency subgraph set are specifically as follows: S301: Extracting a resource access channel number and a scheduling trigger sequence based on the primary task scheduling unit set, extracting a directed connection path between nodes according to a trigger sequence, constructing a position index matrix of each pair of nodes in a path graph, setting a weight and a connection direction of an edge between nodes, and generating a task node dependency graph; S302: Calling a path index and an edge connection relationship in the task node dependency graph, identifying a number of nodes in each path as a path depth and identifying a number of subsequent nodes connected to each node as a branch factor, calling the path depth and the branch factor to calculate an original complexity score, adjusting the original complexity score according to an occupation time length information recorded in a scheduling table and according to an occupation time length, and obtaining a node scheduling complexity score value; S303: Filtering node score data according to the node scheduling complexity score value, extracting a node index with a score greater than a complexity score quantile threshold value, and identifying a corresponding aggregation distribution section in the dependency graph, constructing and pruning a continuous aggregation path interval in the graph structure according to a score difference between adjacent nodes, and each node in the aggregation section being a subgraph merging object to generate a dependency subgraph set; The complexity score quantile threshold value is obtained by constructing a sorting array according to scheduling complexity scores of all nodes, comparing scheduling performances of each quantile value on a plurality of task sets by using a grid search method, and selecting an 80th percentile score value as a threshold for screening dense path nodes.

[0012] As a further scheme of the application, the steps of extracting a highest call frequency path to form a main path set, dividing remaining nodes to form a scheduling group, and generating a task structure division result according to a node identification mapping relationship of the dependency subgraph set and the primary task scheduling unit set are specifically as follows: S401: Extract the node number included in each path according to the node identification mapping relationship of the dependent subgraph set and the primary task scheduling unit set, collect the trigger frequency of the corresponding node in the scheduling log, identify the total call number of each node in each path, extract the path number with the highest trigger frequency in each path, and construct a dataset to generate a high-frequency path trigger sequence; The high-frequency path trigger sequence is a frequent path set obtained by identifying the total trigger frequency of each node in each path in the scheduling log and selecting the path number set with the highest trigger frequency; S402: Call the path number in the high-frequency path trigger sequence, and obtain all the node numbers, eliminate the main path node numbers in the node number set of the primary task scheduling unit set, and extract the remaining node number set, perform edge traversal on the resource dependency relationship between the internal nodes of the remaining node number set, perform hierarchical clustering analysis on the remaining nodes according to the connection directionality and the resource channel number, and generate a node scheduling partition structure; S403: Construct a path attribution index table according to the remaining nodes in the node scheduling partition structure, construct a grouping number mapping table using the main path node set in the high-frequency path trigger sequence, merge the path attribution index table and the grouping number mapping table into a unified number index structure, and map it to the scheduling structure diagram to generate a task structure division result; The node scheduling partition structure is obtained by performing edge traversal on the resource dependency relationship between the remaining nodes after eliminating the main path nodes, and performing dependency clustering in combination with the resource channel number and the connection directionality to form a grouping structure of task nodes.

[0013] As a further scheme of the application, the method further comprises a step S5: S5: Based on the task structure division result, collect the resource usage rate and processing time length of each task unit, construct a task processing behavior vector, evaluate the resource change rate and adjust and sort the priority of each task unit to obtain a task execution priority sequence; The task execution priority sequence specifically refers to a task unit number, a sorting weight coefficient, and a priority arrangement index.

[0014] As a further scheme of the application, based on the task structure division result, collect the resource usage rate and processing time length of each task unit, construct a task processing behavior vector, evaluate the resource change rate and adjust and sort the priority of each task unit to obtain a task execution priority sequence. The step specifically comprises: S501: Based on the task structure division result, collect the resource usage rate and processing time length of each task unit, perform interval normalization processing and construct a joint data structure, combine the resource dimension and the time dimension, form a two-dimensional behavior vector, and generate a task scheduling feature vector set; The task scheduling feature vector set is constructed by interval normalization processing of the resource utilization rate and processing duration of each task unit, and combining into a two-dimensional behavior vector; S502: The resource dimension data in the task scheduling feature vector set is called, the sliding window difference method is used to calculate the resource utilization rate difference value of the continuous task units on the same resource dimension, the absolute value sequence of the difference value is extracted and used as the fluctuation amplitude, the fluctuation mean value of each resource dimension is calculated, the resource items with a fluctuation value less than the average offset value are removed, the resource dimension variation coefficient is calculated, and the stability index is summarized to generate a resource utilization variation coefficient group; The sliding window difference method uses a window size of 3 and a sliding step of 1 to perform step-by-step difference calculation on the resource utilization rate sequence, extracts the difference value between adjacent task units and forms a fluctuation amplitude sequence; The specific formula for calculating the resource dimension variation coefficient is: ; The resource dimension variation coefficient is calculated. Wherein, is the resource dimension variation coefficient of the jth resource dimension, is the resource utilization rate of the jth resource dimension in the kth task unit, is the arithmetic mean of the resource utilization rate of the jth resource dimension in all n task units, is the total number of continuous task units participating in resource utilization rate statistics, is a constant form of a small positive real disturbance factor for the denominator to prevent zero division error, is the index identification of the resource dimension, is the index identification of the jth resource dimension, is the time sequence index identification of the task unit; The resource dimension variation coefficient is calculated by statistics of the utilization rate sequence of each resource in each task unit, and the ratio of the standard deviation to the arithmetic mean of the utilization rate sequence, which is used to reflect the fluctuation degree of resource utilization; The resource utilization variation coefficient group is constructed by extracting the variation coefficient of the calculated utilization rate of each resource dimension, which is used to reflect the resource utilization stability and fluctuation amplitude; S503: According to the resource utilization variation coefficient group, the time proportion weight is calculated combined with the processing duration normalization value in the task scheduling feature vector set, and the priority of each task unit is adjusted and sorted by using the resource fluctuation value and the processing duration value, to generate a task execution priority sequence.

[0015] In another aspect, a large model-based multi-task processing system is provided, which is applied to a large model-based multi-task processing method, and the system comprises: A resource conflict analysis module obtains resource type codes, access frequencies and occupation time lengths of task nodes by using a large model, calculates the Euclidean distance of resource demand vectors and constructs a conflict intensity matrix, screens node pairs in combination with a set threshold, calls a density clustering algorithm to evaluate density to determine structural conflicts, and generates a reconfigurable task node structure body to be transmitted to a scheduling structure generation module; A scheduling structure generation module divides upper and lower boundary nodes according to the reconfigurable task node structure body, extracts upper boundary nodes by using a conflict intensity median deviation parameter to construct independent units, matches lower boundary nodes and fuses them according to resource type mutual difference rates, generates a primary task scheduling unit set to be transmitted to a dependency graph construction module; A dependency graph construction module extracts resource access channel numbers and scheduling trigger sequences based on the primary task scheduling unit set, constructs a dependency graph, calculates complexity scores by using path depth and branch factor, adjusts score weights in combination with occupation time lengths, trims dense areas and performs subgraph and node aggregation, and generates a dependency subgraph set to be transmitted to a structure division module; A structure division module identifies path call frequencies and extracts the highest frequency paths according to the node identification mapping relationship between the dependency subgraph set and the primary task scheduling unit set, constructs a main path set, divides remaining nodes and forms scheduling groups, and generates a task structure division result to be transmitted to a priority sorting module; A priority sorting module constructs task processing behavior vectors based on the task structure division result, collects resource usage rates and processing time lengths of each task unit, evaluates resource change rates and sorts by using a priority adjustment function, and generates a task execution priority sequence.

[0016] The technical scheme provided by the embodiment of the application has at least the following beneficial effects: The multi-task processing system based on a large model considers resource usage differences and structural distribution characteristics in the task division stage, enhances the resource independence and structural boundary recognition capability of the scheduling unit, and avoids resource competition caused by high coincidence between nodes. In the scheduling path construction process, the structural complexity and execution overhead are considered, the redundant nodes in the dense scheduling area are accurately identified and trimmed, and the path execution efficiency and controllability are improved. The task division result is driven by call frequency to classify main and secondary paths, and the non-main path task grouping is realized in combination with node coverage relationships, the division structure is clear, and the interference between tasks is reduced. The task sorting mechanism simultaneously considers resource fluctuations and processing time lengths, dynamically schedules the order to reflect the current resource state, and overall improves the scheduling accuracy and concurrent execution efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0018] Figure 1 Workflow diagram of the present application; Figure 2 System flowchart of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the present application will be described below in combination with the drawings.

[0020] In the embodiments of the present application, the words such as "example", "for example" are used to represent as an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0021] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent. "Of", "corresponding" and "corresponding" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent.

[0022] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1. When the distinction is not emphasized, the meanings expressed are consistent.

[0023] In order to make the technical problems, technical solutions and advantages to be solved by the present application more clear, the following will be described in detail in combination with the drawings and specific embodiments.

[0024] Please refer to Figure 1 The present application provides a technical solution, a multi-task processing method based on a large model, comprising the following steps: S1: using a large model to obtain resource type coding, access frequency and occupation time length of a task node, calculating the Euclidean distance of a resource demand vector to obtain a conflict intensity matrix, using a preset threshold to screen resource overlapping node pairs, judging structure conflict according to node aggregation characteristics, calling a density clustering algorithm to evaluate density, and generating a reconfigurable task node structure body; S2: According to the reconfigurable task node structure, the upper and lower bound nodes are divided by using the conflict intensity median deviation parameter, the upper bound node is extracted to construct an independent unit, the lower bound node is matched and fused according to the resource type mutual difference rate, and a primary task scheduling unit set is generated; S3: Based on the primary task scheduling unit set, the resource access channel number and scheduling trigger order are extracted, a dependency graph is constructed, the node path depth and branch factor are calculated to obtain a complexity score, the weight is adjusted combined with the occupation time length, the dense area execution subgraph and node aggregation are cut according to the score, and a dependency subgraph set is generated; S4: According to the node identification mapping relationship between the dependency subgraph set and the primary task scheduling unit set, the highest call frequency path is extracted to form a main path set, the remaining nodes are divided to form a scheduling group, and a task structure division result is generated; S5: Based on the task structure division result, the resource usage rate and processing time length of each task unit are collected, a task processing behavior vector is constructed, the resource change rate is evaluated, and the priority of each task unit is adjusted and sorted to obtain a task execution priority sequence; The reconfigurable task node structure includes node clustering labels, resource conflict labels, and structure adjustment indication information. The primary task scheduling unit set includes fusion node indexes, independent unit identifiers, and unit structure configuration parameters. The dependency subgraph set includes node dependency relationship graphs, task path hierarchy structures, and aggregated subgraph identifiers. The task structure division result includes main path node indexes, grouping boundary lists, and structure level numbers. The task execution priority sequence specifically includes task unit numbers, sorting weight coefficients, and priority arrangement indexes.

[0025] Please refer to Figure 1 , the resource type code, access frequency and occupation time length of the task node are obtained by using a large model, the Euclidean distance of the resource demand vector is calculated to obtain the conflict intensity matrix, the resource overlap node pairs are screened by using a preset threshold, the structure conflict is judged according to the node aggregation characteristics, the density clustering algorithm is called to evaluate the density, and the specific steps of generating the reconfigurable task node structure include: S101: The resource type code, access frequency and occupation time length of the task node are obtained by using a large model, the vector expression is obtained according to the resource dimension group, the distance value of a plurality of node pairs in each resource dimension is calculated, the symmetric conflict intensity matrix is constructed by combining the distance of each dimension, and the resource distance intensity matrix value is generated; To obtain the resource type code, access frequency and occupation duration of the task node using a large model, the task scheduling system first needs to call each task node in the task list, and for each task node, extract the corresponding resource call information from the system resource log. Record the resource code (such as the type identification code of CPU, GPU, memory, disk IO, etc.), the number of resource calls per unit time (i.e. access frequency, unit: times / sec), and the average duration of resource occupation by the task (in seconds). Establish a mapping vector for these three types of data according to the task node number and resource type, for example, the access frequency of node A on the CPU is 5 times / sec, the average occupation duration is 0.2 seconds, and the resource type is code 01. Its vector form is represented as [01, 5, 0.2]. After arranging the data structure of each task node, the task-resource triple set is obtained. On this basis, for any two nodes (such as nodes A and B), the Euclidean distance is calculated for their access frequency and occupation duration in the same resource code dimension, i.e. the access frequency and occupation duration of node A are , and the access frequency and occupation duration of node B are , then the distance in this resource dimension is: ; Repeat the process to calculate the distance values in all resource dimensions, then combine the dimension distances in the order of resource codes to form a vector with a length of the total number of resources. Finally, perform corresponding operations on all task node pairs to construct a symmetric matrix D, where each element represents the combined distance of task node i and node j in all resource dimensions, which represents the difference between their resource access characteristics. The following is a sample data explanation: Table 1 Task node resource call feature table

[0026] As shown in Table 1, the distance between nodes A and B in the CPU resource is: ; The distance in the GPU resource is: ; The resource distance vector of node pair A-B after combination is [2.002, 1.044]. According to multiple node pairs, construct a vector matrix of this form, and store the resource distance vectors of each node pair. Finally, output the resource distance strength value matrix between nodes in the form of a matrix, where the diagonal elements are always 0, and the other symmetric distributed elements represent the distance combination values between any two nodes in all resource dimensions.

[0027] S102: According to the resource distance strength matrix value, call the resource coincidence judgment threshold, filter the task node combination in which the resource distance of the node pair is less than the resource coincidence judgment threshold, and aggregate and classify according to the resource type coding information of the node mapping, identify the number of aggregated nodes under each resource code, calculate the frequency distribution of each resource combination in the task set, and obtain the resource coincidence node distribution rate; According to the resource distance strength matrix value, first read the resource distance combination value of each task node pair in the matrix row by row, extract the specific resource distance of each node pair in each dimension, and set the combination distance of a task node pair A-B in the CPU and GPU dimensions as [2.002, 1.044]. The resource coincidence judgment threshold set by the system needs to be set based on the average distance and standard deviation between resources. The resource coincidence threshold is set as the average distance of all node pairs minus a standard deviation, which is calculated by taking the square difference average of all distance values and then taking the square root. If the average distance of all node pairs in the CPU dimension is 2.3 and the standard deviation is 0.5, the threshold is set to 1.8. The judgment condition is: if the distance of a node pair in any resource dimension is lower than the threshold value of that dimension, it is considered that the resource may coincide, and the judgment action is performed for each resource dimension, that is, the comparison is performed. The node pairs that meet the conditions are extracted to form a candidate node group. The resource type coding information of all candidate node groups is read respectively, and the number of node pairs appearing under the same code is counted. When performing this counting, all candidate node groups need to be traversed, and a counter is set for each resource code. If the CPU code is 01, there are 8 node pairs involving this resource in all candidate node groups, then the aggregated node number of 01 is set to 8, and the node distribution frequency under this code is calculated. The frequency is represented by the ratio of the number of times the resource code appears in the candidate nodes to the total number of node pairs. For example, if all candidate node pairs are 50 pairs and 8 pairs involve code 01, then the frequency is The frequency statistics need to be performed for all resource codes in turn to form a frequency array, such as the frequency of resource code 01 being 0.16 and the frequency of code 02 being 0.22. On this basis, the resource coincidence node distribution rate is obtained. The following is an example of data: Table 2 Resource code frequency statistics table

[0028] As shown in Table 2, resource code 02 (GPU) has the highest distribution frequency, indicating that the GPU resource has a high degree of coincidence in the current task node, and the distribution frequency will be used as the initial input item in the density clustering, which will directly affect the density evaluation of the node structure. The "high frequency" in the above statistical step should be reasonably interval, if the mean of all resource distribution frequencies is 0.15, and the standard deviation is 0.05, then the high frequency resource threshold is set to 0.2 or more, and the low frequency resource threshold is set to 0.1 or less, which is used as the basis for subsequent density estimation. The process finally outputs the resource coincidence node distribution rate under each resource type code, which is used to guide the density evaluation and structure aggregation operation of high coincidence resources; The resource coincidence determination threshold is the mean of all node resource distances minus the standard deviation.

[0029] S103: Call the resource coincidence node distribution rate, based on the access frequency and occupation time data of each node in the resource combination, call the DBSCAN density clustering algorithm, calculate the local resource access density value, mark the high density resource structure, construct the connected relationship graph and classify and aggregate the high density structure, and get the reconfigurable task node structure body; The resource coincidence node distribution rate is called, first read the distribution frequency corresponding to each resource code in Table 2, and execute aggregation analysis for the resource type with frequency greater than the set threshold value, for example, if the frequency of GPU resource code 02 is 0.22, which has exceeded the preset high density threshold value 0.2, then all task nodes involving this resource are extracted from the task set, and a resource access record table is established according to the node number, which records the access frequency and average occupation time of each node to the resource, for example, the access frequency of node A is 2 times per second, and the occupation time is 1.5 seconds, the access frequency of node B is 1 time per second, and the occupation time is 1.2 seconds, the system constructs a two-dimensional feature vector space according to the node resource combination, and represents each task node as a two-dimensional vector point , form a coordinate set for density calculation, the density evaluation adopts fixed radius method, set the search radius as 0.5, and the minimum neighborhood node number MinPts is set as 3, for any node, execute neighborhood search in this two-dimensional space, and count the number of nodes falling within the radius centered on it, if the number ≥ MinPts, the node is defined as a density core point, if the falling number < MinPts, but it is located in the neighborhood of a core point, it is a boundary point, if neither of them is satisfied, it is a noise point, the specific search action is: calculate the Euclidean distance between the current node and all other nodes, whenever the distance is less than , add it to the neighborhood set, if the number of elements in the set reaches the threshold value, mark it as a core point, for example: the coordinates of node A are (2, 1.5), the other nodes are (1.5, 1.4), (2.2, 1.6) and (3.0, 1.1), the distances between these nodes are calculated as follows: ; ; ; wherein d1 and d2 are less than 0.5, fall into the neighborhood, the number of nodes is 2, and MinPts = 3 is not satisfied, so A is not a core point, and if A falls into the neighborhood of other core points, A is a boundary point. By traversing all nodes and establishing a neighborhood graph, a plurality of connected subgraphs are finally formed, each subgraph is composed of density-related nodes, all nodes in the connected graph are classified into the same class of resource structure bodies by the system, a task node reconfigurable structure body is formed, and the structure body is stored with a label. Each node in the structure body still retains the original resource call parameters for subsequent reconstruction and optimization of task scheduling operations. This process takes resource overlap frequency as an input condition and local density value as a screening basis to realize structure mapping and calibration of resource-intensive areas, complete the classification operation of connected subgraphs, and output the task node structure body identification set; The DBSCAN density clustering algorithm calculates the Euclidean distance between the two-dimensional vector composed of the task node resource access frequency and the occupation time, determines the search radius based on the k-nearest neighbor distance curve inflection point method, sets the minimum number of neighborhood nodes by combining the dimension multiple method and the task size adjustment method, calculates the local density between nodes, identifies high-density resource structures, and aggregates them into a reconfigurable task node structure body; The local resource access density value is obtained by constructing a two-dimensional vector from the task node access frequency and occupation time, setting a search radius and a minimum number of neighborhood nodes in the vector space, counting the number of nodes in the neighborhood, and analyzing the local density.

[0030] Please refer to Figure 1 , according to the reconfigurable task node structure body, the upper and lower boundary nodes are divided by using the conflict intensity median deviation parameter, the upper boundary nodes are extracted to construct independent units, the lower boundary nodes are matched and fused according to the resource type mutual difference rate, and the specific steps of generating the primary task scheduling unit set include: S201: According to the resource conflict intensity value of each group of nodes in the reconfigurable task node structure body, the median value of the conflict intensity is set according to the median deviation parameter, the offset difference between the resource conflict value of each node and the median value is calculated, the offset direction is identified, and the nodes with positive offset direction are extracted as upper boundary nodes, and the nodes with negative offset direction are extracted as lower boundary nodes. Record the number of upper and lower boundary nodes and the corresponding structure position information to generate the upper and lower boundary node distribution; According to the resource conflict intensity value corresponding to each group of nodes in the reconfigurable task node structure, the conflict intensity values of all nodes in the node set are first extracted, a one-dimensional array set is constructed to represent the node conflict index sequence, then the median is calculated, and the five node conflict intensity values are 2.3, 1.7, 2.9, 1.5, and 2.1, respectively. The sorted sequence is [1.5, 1.7, 2.1, 2.3, 2.9], and the median value is 2.1. Set the median value as the reference benchmark, and then traverse the node set. Subtract the median value from the conflict intensity value of each node to obtain the offset value. Set node A as 2.3, and the offset value is , which is a positive offset. Node B is 1.7, and the offset value is , which is a negative offset. The judgment standard is that the offset value greater than 0 is classified as an upper bound node, and the offset value less than 0 is classified as a lower bound node. In this embodiment, the offset values of nodes A and C (2.9) are 0.2 and 0.8, respectively, which are classified as upper bound nodes. The offset values of nodes B and D (1.5) are -0.4 and -0.6, respectively, which are classified as lower bound nodes. The value of node E is 2.1, and the offset value is 0, which is not included in the two categories. Record the node number and its position in the structure after classification. Set the original position of node C as the 3rd layer, the 2nd group, and the 1st node. The position identifier is (3-2-1). The structure position information needs to be labeled with two levels of group marks to form a position array. Meanwhile, the number of upper and lower bound nodes is counted. Here, the number of upper bound nodes is 2, and the number of lower bound nodes is 2. The output is the node identifier set and the distribution of upper and lower bound nodes. Table 3 Task node resource conflict intensity table

[0031] As shown in Table 3, the offset direction is obtained by judging the difference between the conflict intensity value and the median value, and the upper and lower bound nodes are extracted and the position information is recorded, so as to obtain the distribution of upper and lower bound nodes. The median deviation parameter is obtained by extracting the resource conflict intensity value of each node, calculating the median, extracting the difference between each node value and the median, obtaining the offset difference, and dividing the offset direction to identify the upper bound node and the lower bound node.

[0032] S202: Based on the distribution of upper and lower bound nodes, the upper bound node set is extracted, the resource type code and the original position information in the structure of each upper bound node are identified, the independent index relationship between nodes is established, and the corresponding resource attribute item in the index path is marked. The resource usage frequency difference between nodes is called to calculate the resource isolation ratio, the resource independent partition structure of the upper bound node is constructed according to the isolation ratio, and the resource utilization separation index is generated. Based on the upper and lower bound node distribution, first extract the upper bound node set, in table 3 has identified the node A and C for positive offset node, set its resource type code is 01 (CPU) and 03 (memory), and according to the original structure position in the structure body mark its belonging to hierarchical and grouping information, for example, the original structure position of node A is 1-1-1, and the original structure position of node C is 3-2-1, on this basis, the independent index path is constructed, that is, the index structure is formed by generating node number, resource code and original position triplets for each upper bound node, such as (A, 01, 1-1-1), (C, 03, 3-2-1), and the resource attribute item involved in each index path needs to be identified, that is, the specific resource item corresponding to the resource type code is extracted, for example, 01 is CPU and 03 is memory, and the resource usage frequency between nodes is further read, and the pairing of nodes in the same upper bound set is performed according to the resource code, and the difference operation between the access frequencies of each paired node pair under each resource type is performed, assuming that the access frequencies of nodes A and C are 4 times / second and 2 times / second respectively, the difference is 2 times / second, the average difference ratio is used to evaluate the resource isolation degree, that is, the ratio of the sum of the access frequency difference values of all paired nodes of the type resource to the average access frequency, assuming that the sum of the access frequency difference values is 4 and the average access frequency is 3, the isolation ratio is The same calculation operation is performed in the node set, the isolation ratio under all resource codes is calculated and an independent partition index is formed, and the resource with an isolation ratio higher than the set threshold value is regarded as an independent resource, assuming that the isolation threshold value is 1.2, the above ratio 1.33 meets the independent condition, and the node pair is classified as a resource independent unit in the resource utilization structure, and the above process is repeated for all node pairs under all resource codes, and the values of the resource isolation ratio are composed into an array, assuming that the array is [1.33, 0.95, 1.5], which represents the isolation strength of three types of resources, and the threshold value array [1.2, 1.2, 1.2] is compared, wherein the first and third items meet the independent condition, so the resource independent partition structure of the upper bound node is generated, and a separation index matrix is generated with the resource code as the key value, to reflect the independence degree under different resources; The process finally forms the resource partition structure of the independent upper bound node, identifies the resource utilization difference degree between each node, and further constitutes a resource utilization separation index set. The index set will be used in the lower bound node matching and fusion structure construction logic subsequently as a key reference factor; The resource isolation ratio is obtained by calculating the access frequency difference value of each node in the upper bound node set under the same resource type, and normalizing the difference value to the ratio of the average access frequency, which is used to quantify the heterogeneity degree of resource use and the construction of resource independent partition structure.

[0033] S203: Call the index information and node resource code content corresponding to each upper bound node in the resource utilization separation index. Based on the structure number and connected edge relationship in the index path, identify the lower bound nodes that have direct dependency connection with the upper bound node and continuous structure number. Extract the resource type code of the corresponding lower bound node. Calculate the resource type dissimilarity rate based on the number of resource type differences and the total number of resource items. Based on the dissimilarity rate value, perform fusion matching and node reorganization on the upper and lower bound nodes to generate a primary task scheduling unit set. The specific formula for calculating the resource type difference rate based on the number of resource type differences and the total number of resource items is as follows: ; Calculate the resource type dissimilarity rate; in, The resource type distinctness rate is expressed in dimensionless form. The total number of resource types defined in the system. For the index number of the resource type, For the index number of the resource type, This is the index number of the lower bound node. This represents the total number of lower bound nodes extracted so far. For the first Does the nth lower bound node have the nth lower bound node? The status value of the resource class is in dimensionless form. For the first Does the nth lower bound node have the nth lower bound node? The status value of the resource class is in dimensionless form. For the first The lower bound node of the nth node The proportion or importance weight of a resource type within the node's resource structure, expressed in dimensionless units. For the first The lower bound node of the nth node The proportion of a resource type in the node's resource structure or the importance weight of a resource type, in dimensionless units; formula: ; Detailed explanation of the formula and the derivation process of the formula calculation; The formula is used to calculate the resource type dissimilarity rate, and the result is used to measure the degree of difference in resource type distribution among lower bound nodes; Parameter meanings and settings: The resource type dissimilarity rate reflects the degree of difference in resource type distribution among lower bound nodes; The total number of resource types is set to 4; For the lower bound node total number, set to 3; For the state value of whether the first lower bound node has the first type of resource; For the proportion value of the first type of resource in the node resource structure or the importance weight of the resource type in the first lower bound node, set as follows: For resource type 1, set to 0.4; For resource type 2, set to 0.3; For resource type 3, set to 0.2; For resource type 4, set to 0.1; Substitute the parameters into the formula to calculate: Set the relationship between the lower bound node and the resource type as follows: Node 1: resource type 1 ( , ), resource type 2 ( , ); Node 2: resource type 2 ( , ), resource type 3 ( , ); Node 3: resource type 3 ( , ), resource type 4 ( , ); Calculate the weighted sum of each resource type: Resource type 1: ; Resource type 2: ; Resource type 3: ; Resource type 4: ; Calculate the total sum of the weighted sums of all resource types: ; Calculate the average weighted value: ; Calculate the numerator part: ; Calculate the denominator part: ; Finally, calculate the resource type mutual difference rate: ; The result 0.0292 indicates that the distribution difference of the resource type in the lower bound node is low, and the resource type distribution is relatively balanced. The result can be used to evaluate the rationality of resource allocation and guide the subsequent resource optimization and scheduling strategy; The resource utilization separation index is obtained by extracting the resource type code and resource usage frequency of the upper bound node, calculating the difference of access frequency between nodes on the same resource type and normalizing it to the isolation ratio, constructing the resource independent partition structure, and calculating the resource utilization separation index. The resource type mutual difference rate is obtained by identifying the proportion of the number of non-overlapping resource types between the upper and lower bound nodes to the total number of resource types, and calculating according to the node resource code and structure number.

[0034] Please refer to Figure 1 , based on the primary task scheduling unit set, extract the resource access channel number and scheduling trigger sequence, construct the dependency graph and calculate the node path depth and branch factor to get the complexity score, adjust the weight according to the occupation time, according to the score, cut the dense area execution subgraph and node aggregation, and generate the specific steps of the dependency subgraph set as follows: S301: Based on the primary task scheduling unit set, extract the resource access channel number and scheduling trigger sequence, extract the directed connection path between nodes according to the trigger sequence, construct the position index matrix of each pair of nodes in the path graph, set the weight and connection direction of the edge between nodes, and generate the task node dependency graph; Based on the primary task scheduling unit set to extract resource access channel number and scheduling trigger sequence, first traverse all scheduling unit node set, read the channel number and corresponding trigger sequence of each task node, in ascending order according to trigger sequence to construct scheduling execution sequence, in the example data, the trigger sequence of node A to E is 1 to 5 in turn, form node trigger sequence [A, B, C, D, E], for adjacent nodes in the sequence, in order to define the directed connection path, such as A→B, B→C, C→D, D→E, for each pair of connected nodes, generate position index matrix, the element marked as 1 in the ith row and jth column of the matrix indicates that there is A(i)→B(j) path, construct 5×5 path matrix for all node pairs and fill in the directed connection relationship, and record the departure node and target node channel number in each path, and set the connection weight according to the node trigger sequence difference value, set the connection weight as the inverse of the trigger sequence difference value, for example, the trigger difference value of A to B is 1, then the weight is 1, the trigger difference value of C to D is 1, then the weight is also 1, if the trigger interval of two nodes exceeds 1, the corresponding weight will be reduced, such as the trigger difference value of A to C is 2, then the weight is 0.5, in the path, the connection direction is from the low order node to the high order node, annotate the weight and direction of all connection edges, finally construct the task node dependency graph, wherein the graph structure adopts adjacency list representation, each node stores the subsequent node number and corresponding edge weight pointed by it, forming a complete dependent directed graph structure; Table 4 scheduling unit node trigger and channel information table

[0035] As shown in table 4, according to the trigger sequence and channel number, the node dependent directed path can be constructed and the edge weight and connection direction can be assigned, and finally the task node dependency graph is generated.

[0036] S302: call the path index and edge connection relationship in the task node dependency graph, identify the number of nodes in each path as the path depth, identify the number of subsequent nodes connected to each node as the branch factor, call the path depth and branch factor to calculate the original complexity score, combine the occupation time length information recorded in the scheduling table of each node, and set the score adjustment coefficient according to the occupation time length, adjust the original complexity score, and obtain the node scheduling complexity score value; The path index in the task node dependency graph is connected with the edge. First, the constructed task dependency graph structure is read, the number of nodes contained in each path in the graph is counted, and the path depth is taken as the number of nodes. The node counting operation is performed on each complete path from the source node to the terminal node, for example, in the path A→B→C→D→E, the number of nodes contained is 5, and the path depth is 5. The outgoing edge analysis is performed on each node, that is, the number of directly subsequent nodes connected out of the node is counted as the branch factor of the node. For example, node A points to node B only, and the branch factor of A is 1. If a node points to multiple subsequent nodes, the branch factor is increased accordingly, for example, node F points to nodes G and H, and the branch factor is 2. The original complexity score is represented by the product of the path depth and the branch factor. Let the path depth be , the node branch factor be , and the complexity score be . The average of the scores of all nodes in each path is taken as the path complexity score. For example, the branch factors of the 5 nodes in the path A→E are 1, 1, 1, 1, and 0, and the average is 0.8. The original score is . The resource occupation time recorded in the scheduling table of each node is combined to adjust the score. Let the occupation time of node A be 0.5 seconds, that of B be 1.0 second, that of C be 0.8 second, that of D be 1.2 seconds, and that of E be 0.5 second. The average occupation time is 0.8 seconds. The score adjustment coefficient is calculated according to the ratio of the occupation time to the average value. Let the occupation time of node C be 0.8 second, and the coefficient be 1.0. The occupation time of node D is 1.2 seconds, and the coefficient is 1.5. The final adjusted score is obtained by multiplying the original score by the weight coefficient to form the final scheduling complexity score. If the original score is 4.0 and the average coefficient is 1.1, the final score is . The operation is performed on all node paths to form a complete node scheduling complexity score table, which provides data support for subsequent screening and structure pruning.

[0037] S303: According to the node scheduling complexity score value, the node score data is screened, the node index with a score greater than the complexity score quantile threshold value is extracted, and the corresponding aggregated distribution section in the dependency graph is identified. According to the score difference between adjacent nodes, the continuous aggregated path interval in the graph structure is constructed and pruned. Each node in the aggregated section is a subgraph merging object, and a dependency subgraph set is generated. According to the node scheduling complexity score value, the scoring data of all nodes is filtered. First, all calculated scheduling complexity score values are extracted to form a one-dimensional score array, for example, the score array is [2.3, 4.4, 3.1, 5.0, 3.6]. According to the array, the statistical quantile value of the score distribution is calculated, and the complexity score quantile threshold is set to 80 quantiles. That is, the score value at the 80% position after sorting is extracted as the screening threshold. If the sorted array is [2.3, 3.1, 3.6, 4.4, 5.0], the 80% position is the fourth value 4.4. Set this value as the score quantile threshold. Compare all scores one by one, and filter out the nodes with scores greater than 4.4. The node number is E (5.0). Extract the topological structure position of the node in the dependency graph. According to the connected path index, obtain the path section where node E is located. Record the connection state and score difference between the node and the previous and next nodes. For example, the score of node D is 4.4, the score of node E is 5.0, and the difference is 0.6. If the adjacent node score difference is less than 1.0, it is considered as a continuous aggregation section. Then D and E form a continuous path section. The path section is aggregated. The aggregation process includes merging all node labels in the section, merging edge relationships, and forming a subgraph object. If nodes E and F also meet the score continuity and the difference is less than 1.0, the path section is extended to D→E→F. A complete continuous path interval is generated, and it is extracted as an independent subgraph in the dependency graph. Record all node numbers and connection edge sets in the subgraph. After executing the filtering and aggregation process, all aggregated node substructures with outstanding scores are extracted. Finally, a plurality of aggregated subgraph sets are constructed. Each subgraph represents the structure expression of the score dense section, which is used for subsequent structure fusion and scheduling compression processing. This process realizes the pruning and subgraph generation logic of the dependency graph structure driven by the score. The output result is a dependency subgraph set and its structure list. The complexity score quantile threshold is calculated by constructing a sorted array of all node scheduling complexity scores, and then using grid search to compare the scheduling performance of each quantile value on multiple task sets. The score value at the 80th percentile is selected as the threshold for screening dense path nodes.

[0038] Please refer to Figure 1 According to the node identification mapping relationship between the dependency subgraph set and the primary task scheduling unit set, the highest call frequency path is extracted to form a main path set. The remaining nodes are divided to form a scheduling group. The specific steps to generate the task structure division result include: S401: According to the node identification mapping relationship between the dependency subgraph set and the primary task scheduling unit set, the node number included in each path is extracted. The trigger frequency of the corresponding node in the scheduling log is collected. The total call number of each node in each path is identified. The path number with the highest trigger frequency in each path is extracted and a data set is constructed to generate a high-frequency path trigger sequence. According to the node identification mapping relationship of the dependency subgraph set and the primary task scheduling unit set, first, each dependency subgraph path is numbered and all node numbers contained therein are extracted. In the example, path P1 contains nodes A and B, P2 contains nodes C and D, and P3 contains nodes E and F. Then, the scheduling log is accessed, and the trigger frequency of each node is extracted. Assuming that the trigger frequency of node A is 15 times, that of B is 20 times, that of C is 12 times, that of D is 10 times, that of E is 25 times, and that of F is 18 times. Next, the frequencies of all nodes in each path are summed. The total frequency of path P1 is 15+20=35, that of P2 is 22, and that of P3 is 43. A path frequency mapping table is formed by combining all path numbers and corresponding frequency values. A maximum value extraction operation is performed, that is, the highest frequency value is selected from all paths, and the main path number is determined to be P3. The path with the highest trigger frequency is set as the high-frequency path. A trigger sequence is constructed according to the path number. If multiple paths have the same frequency, the main path sequence is processed in ascending order of the number. This sequence is used to represent the dominant execution line in the subsequent task scheduling logic, and a high-frequency path trigger sequence set is finally generated. Table 5: Scheduling path trigger frequency table

[0039] As shown in Table 5, path P3 is identified as the main path because the total call frequency of its node set is the largest, and the main path trigger sequence is formed from the frequent path set. The high-frequency path trigger sequence is obtained by identifying the total trigger frequency of each node in each path in the scheduling log, and selecting the path number set with the highest trigger frequency to obtain the frequent path set.

[0040] S402: Call the path number in the high-frequency path trigger sequence to obtain the corresponding full node number, remove the main path node number in the node number set of the primary task scheduling unit set, extract the remaining node number set, and perform edge traversal on the resource dependency relationship between the nodes in the remaining node number set. According to the connection directionality and resource channel number, perform hierarchical clustering analysis on the remaining nodes to generate a node scheduling partition structure. Call the path number in the high-frequency path trigger sequence to get its corresponding all node numbers, in the example, the main path is P3, the node numbers are E and F, extract all task node number sets from the primary task scheduling unit set, set the universal set as [A, B, C, D, E, F], eliminate the main path nodes E and F from it, the remaining node set is [A, B, C, D], then perform resource dependency relationship extraction operation in the remaining node set, perform resource channel number and connection direction comparison analysis on any two nodes, set the node A channel as CH1 and the node B channel as CH2, if A is earlier than B in the scheduling trigger sequence, and A and B share resources or are connected indirectly through resource transfer, then A→B constitutes a dependent edge, and the dependent edge is marked by judging and marking all node pairs, and finally the dependent subgraph structure between the remaining nodes is constituted, then the dependent clustering is performed on the subgraph, and the clustering process is performed according to the following logic: first, take any node as the starting point, traverse all directly connected subsequent nodes, and record the resource channel number, if multiple nodes form a chain path and use the same channel number, they are classified into the same group, for example, nodes A→B→C all use CH1 channel, then A, B and C are clustered into a group, if a node uses different channels, it is split at the boundary and allocated to another group, for example, node D uses CH3 channel, then an independent partition is constituted, repeat the process to complete the dependent grouping of all nodes, finally generate multiple node scheduling partition structures, each partition represents a task node set whose resource channel and scheduling sequence logic are coupled, the structure is an important intermediate product of task scheduling structure division, which is used for subsequent task number mapping.

[0041] S403: Construct a path attribution index table according to the remaining nodes in the node scheduling partition structure, construct a group number mapping table using the main path node set in the high-frequency path trigger sequence, merge the path attribution index table and the group number mapping table into a unified number index structure, and map it to the scheduling structure diagram to generate a task structure division result; According to the remaining nodes in the node scheduling partition structure, a path attribution index table is constructed. First, a unique group number is set for each scheduling partition. Assuming that nodes A, B and C are divided into partition 1, and node D is divided into partition 2, a path attribution index table is established in the format of {A: G1, B: G1, C: G1, D: G2}. The index table indicates the scheduling group number to which each node belongs in the structure partition. Then, a master path node set E and F extracted from the high-frequency path trigger sequence is used to construct a master path mapping number. Assuming that the master path number is M1, the mapping relationship is {E: M1, F: M1}. The path attribution index table and the master path mapping table are merged to generate a unified number index structure. The structure assigns a unique number label to all nodes, in which the master path nodes are the master path numbers, and the non-master path nodes are the scheduling partition numbers. The unified structure is represented as {A: G1, B: G1, C: G1, D: G2, E: M1, F: M1}. Then, the number structure is mapped to the scheduling structure diagram, and a number replacement operation is performed on the nodes in the original diagram to display the number attribution of each node in the structure diagram. Finally, the entire task diagram is processed according to the number attribution, and nodes with the same number are classified as a group of node substructures. A structure partition diagram is generated in the diagram, and each subdiagram is represented as a structure block. The nodes in the structure block are connected according to the original dependency edge, and the structure blocks are connected as logical modules. Finally, a complete task structure partition result is output. The result has complete number attribution, scheduling relationship and resource logic, which facilitates the execution of structure-level scheduling planning by the scheduling system. The node scheduling partition structure is obtained by performing edge traversal on the resource dependency relationship between the remaining nodes after removing the master path nodes, clustering the dependencies according to the resource channel number and connection directionality, and forming a group structure of task nodes.

[0042] Please refer to Figure 1 Based on the task structure partition result, the resource usage rate and processing time of each task unit are collected, a task processing behavior vector is constructed, the resource change rate is evaluated, and the priority of each task unit is adjusted and sorted to obtain a specific step of a task execution priority sequence, which includes: S501: Based on the task structure partition result, the resource usage rate and processing time of each task unit are collected, interval normalization is performed, a joint data structure is constructed, resource dimensions and time dimensions are combined, a two-dimensional behavior vector is formed, a task scheduling feature vector set is generated, and a task scheduling feature vector set is generated. Based on the task structure division result, first, the resource utilization rate and the corresponding processing time of each task unit are extracted, the resource utilization rate column vector and the processing time column vector are constructed and normalized, and the normalization method is linear interval normalization to the range of [0, 1]. The resource utilization rate normalization is performed according to the formula. Assuming that the original value of the resource utilization rate is [80, 65, 90, 50, 70], the maximum value is 90, and the minimum value is 50, after normalization, it is [(80-50) / (90-50)=0.75, (65-50) / 40=0.375, (90-50) / 40=1.0, (50-50) / 40=0.0, (70-50) / 40=0.5]. The original value of the processing time is [6.0, 4.5, 8.0, 3.0, 5.5], the maximum value is 8.0, and the minimum value is 3.0, after normalization, it is [(6.0-3.0) / 5=0.6, (4.5-3.0) / 5=0.3, (8.0-3.0) / 5=1.0, (3.0-3.0) / 5=0.0, (5.5-3.0) / 5=0.5]. The normalized resource value and time value are combined to form a two-dimensional vector pair, and each task unit can be represented as (resource utilization normalization value, time duration normalization value), such as task T1 (0.75, 0.6) and T3 (1.0, 1.0). The two-dimensional vectors of all task units form a scheduling feature vector set, such as {T1: [0.75, 0.6], T2: [0.375, 0.3], T3: [1.0, 1.0], T4: [0.0, 0.0], T5: [0.5, 0.5]}, which is used as basic data for subsequent resource change rate analysis and priority sorting. Table 6 Task unit resource and time duration normalization vector table

[0043] As shown in Table 6, each task unit forms a two-dimensional scheduling behavior vector by normalizing the resource and time duration data, which is used as a basic data set in subsequent priority adjustment calculations. The task scheduling feature vector set is obtained by interval normalizing the resource utilization rate and processing time of each task unit, and combining it into a two-dimensional behavior vector.

[0044] S502: Call the resource dimension data in the task scheduling feature vector set, use the sliding window difference method to calculate the resource utilization rate difference value of the continuous task unit in the same resource dimension, extract the absolute value sequence of the difference value as the fluctuation amplitude, calculate the fluctuation mean value of each resource dimension, eliminate the resource items with a fluctuation value less than the average offset value, calculate the resource dimension coefficient of variation, and summarize it as a stability index to generate a resource utilization variation group. The sliding window difference method uses a window size of 3 and a sliding step of 1 to perform step-by-step difference calculation on the resource usage rate sequence, extracts the difference between adjacent task units and forms a fluctuation amplitude sequence; The specific formula for calculating the resource dimension coefficient of variation is: ; Calculate the resource dimension coefficient of variation; Wherein, is the resource dimension coefficient of variation of the jth resource dimension, is the resource usage rate of the jth resource dimension in the kth task unit, is the arithmetic mean of the resource usage rate of the jth resource dimension in all n task units, is the total number of consecutive task units participating in resource usage rate statistics, is a constant positive real number perturbation factor existing in the form to prevent zero division error, is the index of the resource dimension, is the index of the jth resource dimension, is the time sequence index of the task unit; Formula: ; Formula details and formula calculation derivation process: The formula is used to calculate the dispersion degree of the usage rate of each type of resource dimension in the task scheduling process, and the result is used to identify the fluctuation stability characteristics of resource usage, supporting the stability induction of subsequent resource structure; Parameter meaning and setting value: is the resource dimension coefficient of variation of the jth resource dimension, and the sampling results recorded in the task scheduling log are set to: 62%, 65%, 58%, 60%, 55%, i.e. , , , , ; is the arithmetic mean of the usage rate of the resource dimension in the 5 task units, which is calculated as: ; is the number of consecutive sampling task units, which is set to 5; is a small positive number perturbation term, which is set to 0.001 to prevent the denominator from being zero; Substitute the parameters into the formula to calculate: Calculate the square difference term: ; Average: ; Square root: ; Substitute the formula to calculate the coefficient of variation: ; The result 0.05667 indicates that the usage rate of the jth resource dimension in the current task scheduling process has a low variation degree, with a fluctuation range within 6%. This value is used to filter the dimension items with significant fluctuation amplitude in the resource utilization coefficient of variation matrix, further summarize the stability index, and improve the coordination of resource scheduling distribution. The coefficient of variation of the resource dimension is calculated by statistics of the usage rate sequence of each resource in each task unit, and the ratio of the standard deviation to the arithmetic mean of the usage rate sequence. It is used to reflect the fluctuation degree of resource usage. The resource utilization coefficient of variation array is a numerical set constructed by extracting the coefficient of variation of the usage rate of each resource dimension. It is used to reflect the stability and fluctuation amplitude of resource usage.

[0045] S503: According to the resource utilization coefficient of variation array, combined with the processing time normalization value in the task scheduling feature vector set, calculate the time proportion weight, and adjust and sort the priority of each task unit using the resource fluctuation value and processing time value, generate the task execution priority sequence; According to the resource utilization coefficient of variation array, combined with the processing time normalization value in the task scheduling feature vector set, first, the resource usage normalization value of all task units needs to be evaluated. The coefficient of variation formula is , where is the standard deviation, is the mean, the standard deviation of the resource normalization value array [0.75, 0.375, 1.0, 0.0, 0.5] is calculated, the mean is (0.75+0.375+1.0+0.0+0.5) / 5=0.525, the square difference is [(0.75-0.525)²+(0.375-0.525)²+(1.0-0.525)²+(0.0-0.525)²+(0.5-0.525)²] / 5=(0.0506+0.0225+0.2256+0.2756+0.0006) / 5=0.5749 / 5=0.1150, the standard deviation , the coefficient of variation , then the time proportion weight of each task unit is calculated, and the weight is set as the proportion of the task unit processing time length normalized value to the sum of all task processing time length normalized values, the total normalized time value is 0.6+0.3+1.0+0.0+0.5=2.4, the time length weight of T1 is 0.6 / 2.4=0.25, and the time length weight of T3 is 1.0 / 2.4≈0.417, and so on, the resource fluctuation value is set as the absolute value of the difference between the task resource normalized value and the average value 0.525, for example, the fluctuation value of T3 is |1.0-0.525|=0.475, the priority score is calculated by combining the value and the processing time length value, and the score is set as the weighted sum of the resource fluctuation and the time length weight, and the weighted coefficients are set as α=0.6(resource fluctuation) and β=0.4(time length weight), and the priority score of T3 is The same calculation is performed on all task units, and the score is obtained, and finally the task execution priority sequence is formed in descending order of the score, such as [T3, T1, T5, T2, T4], which is used by the scheduler to determine the task start order in the execution phase. The sorting result is based on the resource fluctuation intensity and the time proportion, and the quantized arrangement logic of the task priority is realized.

[0046] Please refer to Figure 2 The multi-task processing system based on a large model is used to execute the multi-task processing method based on a large model, and the system comprises: A resource conflict analysis module acquires the resource type code, access frequency and occupation time length of the task node by using a large model, calculates the Euclidean distance of the resource demand vector and constructs a conflict intensity matrix, screens node pairs by combining a set threshold, evaluates the density to judge the structure conflict by calling a density clustering algorithm, and generates a reconfigurable task node structure body to be transmitted to a scheduling structure generation module; A scheduling structure generation module divides upper and lower boundary nodes according to the reconfigurable task node structure body, extracts an upper boundary node to construct an independent unit by calling a conflict intensity median deviation parameter, matches a lower boundary node and fuses according to the resource type mutual difference rate, generates a primary task scheduling unit set to be transmitted to a dependency graph construction module; A dependency graph construction module extracts a resource access channel number and a scheduling trigger order based on the primary task scheduling unit set, constructs a dependency graph, calculates a complexity score by calling a path depth and a branch factor, adjusts the score weight by combining the occupation time length, trims a dense area and performs subgraph and node aggregation, and generates a dependency subgraph set to be transmitted to a structure division module; A structure division module identifies a path call frequency and extracts a highest frequency path according to the node identification mapping relationship of the dependency subgraph set and the primary task scheduling unit set, constructs a main path set, divides the remaining nodes and forms a scheduling group, and generates a task structure division result to be transmitted to a priority sorting module; The priority sorting module constructs a task processing behavior vector based on the resource usage and processing time length of each task unit according to the task structure division result, evaluates the resource change rate, calls a priority adjustment function for sorting, and generates a task execution priority sequence.

[0047] The above-described embodiments can be implemented, in whole or in part, by software, hardware (such as a circuit), firmware or any combination thereof. When implemented using software, the above-described embodiments can be implemented, in whole or in part, in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated, in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transferred from one website, computer, server or data center to another website, computer, server or data center through wired (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid state disk.

[0048] It should be understood that the term "and / or" herein merely describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the following three cases: A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. In addition, the character " / " herein generally represents that the associated objects before and after it are in an "or" relationship, but it can also represent an "and / or" relationship, which can be understood according to the context before and after it.

[0049] In the present application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following" or the like means any combination of the items, including any combination of single item or multiple items. For example, at least one of a, b, or c can represent a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

[0050] It should be understood that the size of the sequence number of each process described above does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the application.

[0051] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0052] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the devices, apparatuses and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0053] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0054] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0055] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically independently, or two or more units can be integrated into one unit.

[0056] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the present application that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of software products. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0057] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A multi-task processing method based on a large model, characterized in that, The method includes: S1: Use a large model to obtain the resource type encoding, access frequency and occupation duration of task nodes, calculate the Euclidean distance of resource demand vectors to obtain the conflict intensity matrix, use a preset threshold to filter resource overlapping node pairs, judge structural conflicts based on node clustering characteristics, call density clustering algorithm to evaluate density, and generate recombinable task node structures. S2: Based on the reconfigurable task node structure, divide the upper and lower bound nodes using the conflict intensity median deviation parameter, extract the upper bound nodes to construct independent units, match the lower bound nodes and merge them according to the resource type dissimilarity rate to generate a primary task scheduling unit set. S3: Based on the primary task scheduling unit set, extract the resource access channel number and scheduling trigger order, construct a dependency graph and calculate the node path depth and branch factor to obtain a complexity score, adjust the weights in combination with the occupied time, and perform subgraph and node aggregation in dense areas according to the score to generate a dependency subgraph set; S4: Based on the node identifier mapping relationship between the dependency subgraph set and the primary task scheduling unit set, extract the path with the highest call frequency to form the main path set, divide the remaining nodes to form scheduling groups, and generate the task structure partitioning result.

2. The multi-task processing method based on a large model according to claim 1, characterized in that, The reconfigurable task node structure includes node clustering labels, resource conflict labels, and structure adjustment indication information. The primary task scheduling unit set includes fusion node index, independent unit identifier, and unit structure configuration parameters. The dependency subgraph set includes node dependency relationship graph, task path hierarchy structure, and aggregation subgraph identifier. The task structure partitioning result includes main path node index, group boundary list, and structure level number.

3. The multi-task processing method based on a large model according to claim 1, characterized in that, The specific steps for generating a recombinable task node structure are as follows: Utilizing a large model to obtain the resource type encoding, access frequency, and duration of task nodes; calculating the Euclidean distance of resource demand vectors to obtain the conflict intensity matrix; using a preset threshold to filter resource-overlapping node pairs; judging structural conflicts based on node clustering characteristics; and calling a density clustering algorithm to evaluate density. S101: Use a large model to obtain the resource type encoding, access frequency and duration of task nodes, construct and obtain vector representations according to resource dimensions, calculate the distance values ​​of multiple node pairs on each resource dimension, construct a symmetric conflict intensity matrix by combining the distances of each dimension, and generate resource distance intensity matrix values. S102: Based on the resource distance strength matrix value, call the resource overlap judgment threshold, filter the task node combinations in the node pair whose resource distance is less than the resource overlap judgment threshold, and perform aggregation and classification according to the resource type encoding information mapped by the node, identify the number of aggregated nodes under each type of resource encoding, calculate the frequency distribution of each group of resource combinations in the task set, and obtain the resource overlap node distribution rate. The resource overlap determination threshold is the mean of the distances between all nodes and the resource, minus the standard deviation. S103: Call the resource overlap node distribution rate, and based on the access frequency and occupancy time data of each node in the resource combination, call the DBSCAN density clustering algorithm to calculate the local resource access density value, mark the high-density resource structure, construct the connectivity graph and classify and aggregate the high-density resource structure to obtain the recombinable task node structure. The DBSCAN density clustering algorithm calculates the Euclidean distance between two-dimensional vectors composed of the resource access frequency and occupation duration of task nodes, determines the search radius based on the inflection point method of the k-th nearest neighbor distance curve, and sets the minimum number of neighboring nodes by combining the dimension multiple method and the task size adjustment method. It calculates the local density between nodes, identifies high-density resource structures, and aggregates them into recombinable task node structures. The local resource access density value is obtained by constructing a two-dimensional vector based on the access frequency and duration of task nodes, setting a search radius and a minimum number of neighboring nodes in the vector space, counting the number of nodes in the neighborhood and analyzing the local density.

4. The multi-task processing method based on a large model according to claim 3, characterized in that, Based on the reconfigurable task node structure, the steps of dividing upper and lower bound nodes using the conflict intensity median deviation parameter, extracting upper bound nodes to construct independent units, matching lower bound nodes, and merging them according to the resource type dissimilarity rate to generate a primary task scheduling unit set are as follows: S201: Based on the resource conflict intensity value corresponding to each group of nodes in the reconfigurable task node structure, set the median value of the conflict intensity according to the median deviation parameter, calculate the offset difference between the resource conflict value and the median value of each node, identify the offset direction, extract nodes with positive offset direction as upper bound nodes, extract nodes with negative offset direction as lower bound nodes, record the number of upper and lower bound nodes and the corresponding structural position information, and generate the upper and lower bound node distribution. The median deviation parameter is obtained by extracting the resource conflict intensity value of each node, calculating the median, extracting the difference between each node's value and the median, obtaining the offset difference, dividing the offset direction, and identifying the upper and lower bound nodes. S202: Based on the distribution of the upper and lower bound nodes, extract the set of upper bound nodes, identify the resource type code and original position information in the structure of each upper bound node, establish independent index relationships between nodes and mark the corresponding resource attribute items in the index path, call the resource usage frequency difference between nodes, calculate the resource isolation ratio, construct the resource independent partition structure of the upper bound nodes according to the isolation ratio, and generate the resource utilization separation index. The resource isolation ratio is calculated by taking the difference in access frequency of each node in the upper bound node set on the same resource type, and normalizing the difference to the average access frequency ratio. This ratio is used to quantify the heterogeneity of resource usage and the construction of an independent resource partitioning structure. S203: Call the index information and node resource code content corresponding to each upper bound node in the resource utilization separation index, identify the lower bound nodes that have direct dependency connections with the upper bound nodes and continuous structure numbers according to the structure number and connected edge relationship in the index path, extract the resource type code of the corresponding lower bound node, calculate the resource type dissimilarity rate according to the number of resource type differences and the total number of resource items, and perform fusion matching and node reorganization on the upper and lower bound nodes according to the dissimilarity rate value to generate a primary task scheduling unit set; The resource utilization separation index is obtained by extracting the resource type code and resource usage frequency of the upper bound node, calculating the difference in access frequency between nodes on the same resource type and normalizing it to the isolation ratio, constructing an independent resource partitioning structure, and obtaining the resource utilization separation index. The resource type dissimilarity rate is obtained by identifying the proportion of non-overlapping resource types between upper and lower bound nodes to the total number of resource types, and by pairing and merging them according to the node resource code and structure number.

5. The multi-task processing method for large models according to claim 4, characterized in that, The specific formula for calculating the resource type difference rate based on the number of resource type differences and the total number of resource items is as follows: ； Calculate the resource type dissimilarity rate; in, The resource type distinctness rate is expressed in dimensionless form. The total number of resource types defined in the system. For the index number of the resource type, For the index number of the resource type, This is the index number of the lower bound node. This represents the total number of lower bound nodes extracted so far. For the first Does the nth lower bound node have the nth lower bound node? The status value of the resource class is in dimensionless form. For the first Does the nth lower bound node have the nth lower bound node? The status value of the resource class is in dimensionless form. For the first The lower bound node of the nth node The proportion or importance weight of a resource type within the node's resource structure, expressed in dimensionless units. For the first The lower bound node of the nth node The proportion of a resource type in the node's resource structure or the importance weight of a resource type, expressed in dimensionless units.

6. The multi-task processing method based on a large model according to claim 4, characterized in that, The steps for extracting resource access channel numbers and scheduling trigger orders based on the primary task scheduling unit set, constructing a dependency graph and calculating node path depth and branch factor to obtain a complexity score, adjusting weights based on occupied time, and pruning dense regions according to the score to execute subgraphs and node aggregation to generate a dependency subgraph set are as follows: S301: Extract the resource access channel number and scheduling trigger order based on the primary task scheduling unit set, extract the directed connection path between nodes according to the trigger sequence, construct the position index matrix of each pair of nodes in the path graph, and set the weight and connection direction of the edges between nodes to generate a task node dependency graph. S302: Call the path index and edge connection relationship in the task node dependency graph, identify the number of nodes in each path and use it as the path depth, identify the number of subsequent nodes connected to each node and use it as the branch factor, call the path depth and branch factor to calculate the original complexity score, combine the occupied time information of each node recorded in the scheduling table, and set the score adjustment coefficient according to the occupied time to adjust the original complexity score and obtain the node scheduling complexity score value. S303: Based on the node scheduling complexity score, filter the node score data, extract the node indexes with scores greater than the complexity score percentile threshold, and identify the corresponding aggregation distribution segments in the dependency graph. Based on the score difference between adjacent nodes, construct and trim continuous aggregation path intervals in the graph structure. Each node in the aggregation segment is a subgraph merging object, generating a dependency subgraph set. The complexity score percentile threshold is constructed by creating a sorted array based on the scheduling complexity scores of all nodes, and then using a grid search method to compare the scheduling performance of each percentile value on multiple task sets, selecting the score value at the 80th percentile as the threshold for filtering dense path nodes.

7. The multi-task processing method based on a large model according to claim 6, characterized in that, Based on the node identifier mapping relationship between the dependency subgraph set and the primary task scheduling unit set, the steps for extracting the most frequently invoked paths to form the main path set, dividing the remaining nodes into scheduling groups, and generating the task structure partitioning result are as follows: S401: Based on the node identifier mapping relationship between the dependency subgraph set and the primary task scheduling unit set, extract the node number included in each path, collect the trigger frequency of the corresponding node in the scheduling log, identify the total number of calls for each node in each path, extract the path number with the highest trigger frequency in each path and construct a dataset to generate a high-frequency path trigger sequence. The high-frequency path trigger sequence is a set of frequent paths obtained by identifying the total trigger frequency of each node in the scheduling log of each path and selecting the set of path numbers with the highest trigger frequency. S402: Call the path number in the high-frequency path trigger sequence and obtain the corresponding all node numbers. Remove the main path node number from the node number set of the primary task scheduling unit set and extract the remaining node number set. Perform edge traversal on the resource dependency relationship between nodes within the remaining node number set. Perform hierarchical clustering analysis on the remaining nodes according to the connection directionality and resource channel number to generate a node scheduling partition structure. S403: Construct a path attribution index table based on the remaining nodes in the node scheduling partition structure, construct a group number mapping table using the set of main path nodes in the high-frequency path trigger sequence, merge the path attribution index table and the group number mapping table into a unified number index structure, and map it to the scheduling structure diagram to generate the task structure partitioning result; The node scheduling partitioning structure is formed by traversing the resource dependencies between the remaining nodes after removing the main path nodes, and then performing dependency clustering based on the resource channel number and connection directionality to form a grouping structure for task nodes.

8. The multi-task processing method based on a large model according to claim 1, characterized in that, The method further includes step S5: S5: Based on the task structure division results, collect the resource utilization rate and processing time of each task unit, construct a task processing behavior vector, evaluate the resource change rate, and adjust and sort the priority of each task unit to obtain a task execution priority sequence. The task execution priority sequence specifically consists of task unit number, sorting weight coefficient, and priority sorting index.

9. The multi-task processing method based on a large model according to claim 8, characterized in that, Based on the task structure partitioning results, the following steps are taken to collect the resource utilization rate and processing time of each task unit, construct a task processing behavior vector, evaluate the resource change rate, and adjust and sort the priority of each task unit to obtain the task execution priority sequence: S501: Based on the task structure division results, collect the resource utilization rate and processing time of each task unit, perform interval normalization processing and construct a joint data structure, combine the resource dimension and time dimension, form a two-dimensional behavior vector, and generate a task scheduling feature vector set. The task scheduling feature vector set is constructed by performing interval normalization on the resource utilization rate and processing time of each task unit and combining them into a two-dimensional behavior vector. S502: Call the resource dimension data in the task scheduling feature vector set, use the sliding window difference method to calculate the resource utilization difference value of consecutive task units in the same resource dimension, extract the absolute value sequence of the difference value as the fluctuation amplitude, calculate the fluctuation mean of each resource dimension, remove resource items with fluctuation values ​​less than the average offset value, calculate the resource dimension variation coefficient, summarize it into a stability index, and generate a resource utilization variation coefficient group. The sliding window difference method uses a window size of 3 and a sliding step size of 1 to perform stepwise difference calculations on the resource utilization rate sequence, extract the difference between adjacent task units and form a fluctuation amplitude sequence. The specific formula for calculating the coefficient of variation of the resource dimension is as follows: ； Calculate the coefficient of variation for the resource dimension; in, Let be the coefficient of variation of the j-th resource dimension. Let the resource utilization rate of the j-th resource dimension be the resource utilization rate in the k-th task unit. Let be the arithmetic mean of the resource utilization rate of the j-th resource dimension across all n task units. This represents the total number of consecutive task units participating in resource utilization statistics. A minimal positive real perturbation factor, existing in constant form, is used in the denominator to prevent division by zero errors. This serves as the index identifier for the resource dimension. This is the index identifier for the j-th resource dimension. Time-series index identifier for task units; The coefficient of variation for the resource dimension is obtained by statistically analyzing the usage rate sequence of each type of resource in each task unit and calculating the ratio of the standard deviation to the arithmetic mean of the usage rate sequence, which is used to reflect the degree of fluctuation in resource usage. The resource utilization coefficient of variation set is a numerical set constructed by extracting the coefficient of variation of the utilization rate for each resource dimension and is used to reflect the stability and fluctuation range of resource utilization. S503: Based on the resource utilization variation coefficient group and the normalized value of processing time in the task scheduling feature vector set, calculate the time proportion weight, and use the resource fluctuation value and processing time value to adjust and sort the priority of each task unit to generate a task execution priority sequence.

10. A multi-task processing system based on a large model, characterized in that, The system is used to implement the multi-task processing method based on a large model as described in any one of claims 1-9, and the system comprises: The resource conflict analysis module uses a large model to obtain the resource type code, access frequency and occupation duration of task nodes, calculates the Euclidean distance of resource demand vectors and constructs a conflict intensity matrix, combines a set threshold to filter node pairs, calls a density clustering algorithm to evaluate density and judge structural conflicts, and generates a recombinable task node structure to be passed to the scheduling structure generation module. The scheduling structure generation module divides the upper and lower bound nodes according to the recombinable task node structure, calls the conflict intensity median deviation parameter to extract the upper bound node to construct an independent unit, matches the lower bound node and merges them according to the resource type dissimilarity rate, and generates a primary task scheduling unit set which is then passed to the dependency graph construction module. The dependency graph construction module extracts the resource access channel number and scheduling trigger order based on the primary task scheduling unit set, constructs the dependency graph, calls path depth and branch factor to calculate the complexity score, adjusts the score weight in combination with the occupied time, prunes dense areas and performs subgraph and node aggregation, generates a dependency subgraph set and passes it to the structure partitioning module. The structure partitioning module identifies the path call frequency and extracts the most frequent path based on the node identifier mapping relationship between the dependency subgraph set and the primary task scheduling unit set, constructs the main path set, partitions the remaining nodes and forms scheduling groups, and generates the task structure partitioning result, which is then passed to the priority sorting module. The priority sorting module collects the resource utilization rate and processing time of each task unit based on the task structure division result to construct a task processing behavior vector, evaluates the resource change rate, calls the priority adjustment function to sort, and generates a task execution priority sequence.

Images