A resource scheduling optimization method and system for intelligent fusion terminals

Abstract

This invention discloses a resource scheduling optimization method and system for intelligent converged terminals, relating to the field of data processing. The method includes: constructing a terminal task feature chain by acquiring a multi-resource state sequence and a current task set; performing value analysis and multi-resource coupled scheduling on the multi-resource state sequence to obtain the terminal resource scheduling space; obtaining a first terminal scheduling strategy by performing service experience-oriented optimization based on a service experience evaluation architecture; obtaining a second terminal scheduling strategy by performing task offloading decisions and resource reallocation based on a cloud-edge-device coordinator; obtaining a third terminal scheduling strategy by performing uncertainty modeling and uncertainty adversarial compensation; and executing dynamic multi-resource adjustment of the terminal device. This application solves the problem of low scheduling efficiency caused by the inability of existing terminal scheduling to dynamically adapt to changes in task requirements and fluctuations in resource states, achieving the effect of dynamically adjusting resource allocation according to changes in task requirements and fluctuations in resource states, and improving the efficiency of multi-resource collaborative scheduling.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a resource scheduling optimization method and system for intelligent converged terminals. Background Technology

[0002] In the process of intelligent converged terminal devices running multiple types of concurrent tasks, the coordinated scheduling of various resources such as computing, storage, and communication directly determines the overall performance and task execution efficiency of the system. Currently, the industry commonly adopts resource allocation methods based on fixed priorities. For example, different tasks are given static priorities, and resources are queued and allocated according to these priorities. Alternatively, when resource contention occurs, common strategies such as first-come, first-served or round-robin scheduling are used to uniformly manage computing and storage resources. However, these methods cannot dynamically perceive changes in resource demands during task execution and lack a comprehensive consideration of the coupling relationships between resources. This can easily lead to high-priority tasks excessively consuming resources, causing low-priority tasks to wait for extended periods. Furthermore, they cannot adjust resource allocation strategies based on the real-time status of the terminal devices, resulting in low overall system resource utilization and increased task response latency.

[0003] At present, the resource scheduling of intelligent converged terminals suffers from the technical problem of low scheduling efficiency due to the inability to dynamically adapt to changes in task requirements and fluctuations in resource status. Summary of the Invention

[0004] This application provides a resource scheduling optimization method and system for intelligent converged terminals. By acquiring the multi-resource state sequence and current task set of the terminal device, a terminal task feature chain is constructed. Based on this feature chain, the multi-resource state sequence is analyzed for value and multi-resource coupling scheduling is performed to generate a terminal resource scheduling space. Using a service experience evaluation architecture as a criterion, experience-oriented optimization is performed on the scheduling space to select a first terminal scheduling strategy. A cloud-edge-device coordinator then performs task offloading decisions and resource reallocation based on the first scheduling strategy, offloading transferable tasks to the edge or cloud side and reallocating locally released resources to obtain a second terminal scheduling strategy. Uncertainty modeling and adversarial compensation are applied to the second scheduling strategy. Through resource redundancy compensation and dependency propagation suppression, a robust third terminal scheduling strategy is generated. Based on this, dynamic adjustment of multiple resources is implemented. This solves the technical problem of low scheduling efficiency in existing intelligent converged terminals due to the inability to dynamically adapt to changes in task requirements and resource state fluctuations. It achieves the technical effect of dynamically adjusting resource allocation according to changes in task requirements and resource state fluctuations, and improving the efficiency of multi-resource collaborative scheduling.

[0005] This application provides a resource scheduling optimization method for intelligent converged terminals, comprising: acquiring a multi-resource state sequence and a current task set of a terminal device, and constructing a terminal task feature chain based on the current task set; performing value analysis and multi-resource coupling scheduling on the multi-resource state sequence based on the terminal task feature chain to obtain a terminal resource scheduling space; performing service experience-oriented optimization on the terminal resource scheduling space based on a service experience evaluation architecture to obtain a terminal first scheduling strategy; performing task offloading decisions and resource reallocation on the terminal first scheduling strategy based on a cloud-edge-device coordinator to obtain a terminal second scheduling strategy; performing uncertainty modeling and uncertainty adversarial compensation on the terminal second scheduling strategy to obtain a terminal third scheduling strategy, and executing dynamic multi-resource adjustment of the terminal device according to the terminal third scheduling strategy.

[0006] In a possible implementation, the multi-resource state sequence is analyzed for value and coupled for multi-resource scheduling based on the terminal task feature chain to obtain a terminal resource scheduling space. The following processes are then performed: the multi-resource state sequence is analyzed for value based on the terminal task feature chain to obtain a resource contribution distribution; the multi-resource state sequence is adapted, filtered, and its coupling relationships are sorted out based on the resource contribution distribution to obtain a candidate resource map; and a joint scheduling decision is made based on the candidate resource map according to the terminal task feature chain to generate the terminal resource scheduling space.

[0007] In a possible implementation, the terminal resource scheduling space is optimized based on the service experience evaluation architecture to obtain a first terminal scheduling strategy. The following processes are then performed: demand characteristics are analyzed based on the terminal task feature chain to determine multi-dimensional service experience evaluation indicators; reward and penalty configurations are performed based on the multi-dimensional service experience evaluation indicators to generate the service experience evaluation architecture; a scheme-by-scheme service experience evaluation is performed on the terminal resource scheduling space based on the service experience evaluation architecture to obtain a service experience evaluation graph; the terminal resource scheduling space is filtered using the expected service experience evaluation as a constraint, combined with the service experience evaluation graph, to obtain a resource scheduling filtering domain; energy consumption minimization optimization is performed on the resource scheduling filtering domain to generate the first terminal scheduling strategy.

[0008] In a possible implementation, based on the cloud-edge-device coordinator, the terminal's first scheduling strategy is used to make task offloading decisions and resource reallocations, resulting in a second scheduling strategy. The following processes are then performed: Based on the cloud-edge-device coordinator, the terminal's task feature chain is analyzed for migration costs and experience gains are identified to obtain a task migration projection graph; based on the task migration projection graph, the terminal's task feature chain is assessed for portability to obtain a task migration assessment result; based on the task migration assessment result, the first scheduling strategy is assessed for impact to obtain a task migration resource impact domain; based on the task migration resource impact domain, the first scheduling strategy is used to make task offloading decisions and resource reallocations to generate the second scheduling strategy.

[0009] In a possible implementation, the following processes are performed: Based on the cloud-edge-device coordinator, migration cost analysis and experience gain identification are performed on the terminal task feature chain to obtain a task migration projection graph. Then, the following steps are executed: Cloud-edge migration decision is made on the terminal task feature chain based on the cloud-edge-device coordinator to obtain an initial task migration graph; path optimization is performed on the initial task migration graph to obtain an optimized task migration graph; simulated migration of the terminal task feature chain is performed based on the optimized task migration graph to obtain a task migration simulation dataset; migration cost analysis and experience gain identification are performed on the simulated task migration dataset to generate the task migration projection graph.

[0010] In a possible implementation, uncertainty modeling and uncertainty adversarial compensation are performed on the second terminal scheduling strategy to obtain a third terminal scheduling strategy. The following processes are then performed: uncertainty modeling is performed on the second terminal scheduling strategy to obtain a resource scheduling uncertainty model; a terminal scheduling regret value is generated based on the resource scheduling uncertainty model, and it is determined whether the terminal scheduling regret value is greater than or equal to a predetermined scheduling regret value; if the terminal scheduling regret value is greater than or equal to the predetermined scheduling regret value, resource redundancy compensation and task dependency propagation suppression processing are performed on the second terminal scheduling strategy based on the resource scheduling uncertainty model to generate the third terminal scheduling strategy.

[0011] In a possible implementation, the terminal's second scheduling strategy is processed for resource redundancy compensation and task dependency propagation suppression based on the resource scheduling uncertainty model to generate the terminal's third scheduling strategy. The following processes are then performed: The key uncertain tasks of the terminal's second scheduling strategy are determined based on the resource scheduling uncertainty model, along with the resource fluctuation range and task dependencies corresponding to these key uncertain tasks; multi-dimensional resource redundancy compensation is configured for the key uncertain tasks based on the resource fluctuation range, and the resource redundancy compensation configuration result is obtained; downstream related tasks affected by the key uncertain tasks are identified based on the task dependencies, and a dependency propagation path from the key uncertain tasks to the downstream related tasks is constructed; propagation suppression is configured for the dependency propagation path, and the propagation suppression configuration result is obtained, including delay triggering configuration parameters, execution isolation configuration parameters, and resource reservation configuration parameters; the terminal's second scheduling strategy is updated and optimized based on the resource redundancy compensation configuration result and the propagation suppression configuration result to obtain the terminal's third scheduling strategy.

[0012] In a possible implementation, the following processing is performed: the multi-resource state sequence includes a computing resource state sequence, a storage resource state sequence, a communication resource state sequence, and an energy resource state sequence.

[0013] In a possible implementation, a terminal task feature chain is constructed based on the current task set, and the following processing is performed: task attribute features are extracted from the current task set to obtain each task feature set; the task feature sets are chained and encoded according to the data dependency relationship, execution order relationship and resource competition relationship between the current task sets to generate the terminal task feature chain.

[0014] This application also provides a resource scheduling optimization system for intelligent converged terminals, comprising: a terminal task feature chain construction module, used to acquire a multi-resource state sequence and a current task set of the terminal device, and construct a terminal task feature chain based on the current task set; a terminal resource scheduling space acquisition module, used to perform value analysis and multi-resource coupling scheduling on the multi-resource state sequence based on the terminal task feature chain, and acquire a terminal resource scheduling space; a service experience-oriented optimization module, used to perform service experience-oriented optimization on the terminal resource scheduling space based on a service experience evaluation architecture, and acquire a first terminal scheduling strategy; a resource reallocation module, used to perform task offloading decision and resource reallocation on the first terminal scheduling strategy based on a cloud-edge-device coordinator, and acquire a second terminal scheduling strategy; and an uncertainty adversarial compensation module, used to perform uncertainty modeling and uncertainty adversarial compensation on the second terminal scheduling strategy, acquire a third terminal scheduling strategy, and execute dynamic multi-resource adjustment of the terminal device according to the third terminal scheduling strategy.

[0015] This application proposes a resource scheduling optimization method and system for intelligent converged terminals. First, it acquires the multi-resource state sequence and current task set of the terminal device, and constructs a terminal task feature chain based on the current task set. Next, it performs value analysis and multi-resource coupling scheduling on the multi-resource state sequence based on the terminal task feature chain to obtain the terminal resource scheduling space. Then, it performs service experience-oriented optimization on the terminal resource scheduling space according to a service experience evaluation architecture to obtain a first terminal scheduling strategy. Following this, it performs task offloading decisions and resource reallocation on the first terminal scheduling strategy using a cloud-edge-device coordinator to obtain a second terminal scheduling strategy. Finally, it performs uncertainty modeling and uncertainty adversarial compensation on the second terminal scheduling strategy to obtain a third terminal scheduling strategy, and executes the dynamic adjustment of the terminal device's multi-resources according to the third terminal scheduling strategy. The method and system proposed in this application achieve the technical effect of dynamically adjusting resource allocation based on changes in task requirements and fluctuations in resource state, and improving the efficiency of multi-resource collaborative scheduling. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating a resource scheduling optimization method for intelligent converged terminals provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the structure of a resource scheduling optimization system for intelligent converged terminals provided in an embodiment of this application.

[0019] Figure labeling: Terminal task feature chain construction module 10, terminal resource scheduling space acquisition module 20, service experience-oriented optimization module 30, resource reallocation module 40, uncertainty resistance compensation module 50. Detailed Implementation

[0020] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" can be the same or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0023] This application provides a resource scheduling optimization method for intelligent converged terminals, such as... Figure 1 As shown, the method includes:

[0024] Step S100: Obtain the multi-resource state sequence and current task set of the terminal device, and construct the terminal task feature chain based on the current task set. The multi-resource state sequence includes computing resource state sequence, storage resource state sequence, communication resource state sequence and energy resource state sequence.

[0025] Specifically, real-time operational status data of the terminal device is acquired. This data includes the idle rate and occupancy rate of computing resources, the read / write speed and remaining capacity of storage resources, the bandwidth utilization and signal strength of communication resources, and the current remaining battery power, instantaneous power consumption, and battery temperature of energy resources. The status data of these four types of resources are sampled at fixed time intervals to form computing resource status sequences, storage resource status sequences, communication resource status sequences, and energy resource status sequences, which together constitute a multi-resource status sequence. Simultaneously, all tasks currently queued for processing and those executing but not yet completed are read from the terminal device's task manager to form the current task set.

[0026] In one possible implementation, a terminal task feature chain is constructed based on the current task set. Step S100 further includes step S110, which extracts task attribute features from the current task set to obtain feature sets for each task. Specifically, metadata for each task in the current task set is extracted, including task identifier, task type, task arrival time, task deadline, task code size, input data volume, output data volume, and historical average execution time. For each task, the task type in its metadata is mapped to resource requirement features. For example, the computational resource requirement for image rendering tasks is mapped to high, the storage resource requirement to medium, and the communication resource requirement to low; the computational resource requirement for file upload and download tasks is mapped to low, the storage resource requirement to medium, and the communication resource requirement to high. The mapping result is represented by a three-dimensional vector, where the three dimensions correspond to the computational resource requirement level, storage resource requirement level, and communication resource requirement level, respectively. Each level takes an integer value from 1 to 3, where 1 represents low requirement, 2 represents medium requirement, and 3 represents high requirement. The difference between the task deadline and the task arrival time is used as a latency constraint feature, which is a scalar value. A service quality requirement feature is constructed by obtaining the historical average execution time of the task, where the historical average execution time is the arithmetic mean of the time taken from start to finish in the past ten executions. The total allowed execution time of the task is obtained by subtracting the task arrival time from the task deadline. An initial ratio is obtained by dividing the historical average execution time by the total allowed execution time. An ideal ratio threshold is set, and the difference between the initial ratio and the ideal ratio threshold is calculated to obtain the deviation. If the deviation is less than or equal to 0, the service quality requirement feature is set to the maximum value of 1.0. If the deviation is greater than 0, it means that the historical average execution time exceeds the ideal execution time. A value between 0 and 1 is calculated using the formula: Service Quality Requirement Feature Value = 1 / (1 + Deviation). The closer this value is to 1, the higher the service quality requirement of the task, meaning the system needs to provide sufficient resources to ensure that its execution time is as close to or lower than the historical average as possible; the closer this value is to 0, the lower the service quality requirement of the task. In the task type classification, system-level tasks are marked with the highest priority (5), user interaction tasks with high priority (4), real-time audio / video communication tasks with priority (3), background data synchronization tasks with medium priority (2), and maintenance tasks with low priority (1), forming a task priority feature. This feature is an integer scalar with a value from 1 to 5. The three-dimensional vector of resource requirement features, the scalar value of latency constraint features, the service quality requirement feature value, and the integer value of task priority features are sequentially concatenated to form a six-dimensional vector. This six-dimensional vector is the task feature set for each task. All tasks in the current task set are traversed to obtain the feature sets for each task.

[0027] Step S120: Chain-based associative encoding is performed on the feature sets of each task based on the data dependencies, execution order, and resource competition relationships among the current task sets to generate the terminal task feature chain. Specifically, data dependencies are determined based on the file read / write relationship between the input and output data of each task in the current task set. For example, if the output filename of task A appears in the input file list of task B, it is determined that there is a data dependency relationship between task A and task B, and task A is a predecessor task of task B. Execution order is determined based on the order of the tasks in the queue and the timestamps allocated by the executor. For example, the ready timestamp and scheduling timestamp of each task are read from the task manager, and the tasks are sorted according to the order of their scheduling timestamps to obtain the execution order. A resource competition matrix is ​​established based on the resource requirement characteristics in the task feature set of each task. The rows of this matrix represent tasks, the columns represent resource types, and the matrix elements are the numerical values ​​of the task's demand level for that type of resource. When the sum of the demand level values ​​of two or more tasks for the same type of resource exceeds a preset threshold, it is determined that there is a resource competition relationship between these tasks. An empty directed graph structure is created as the terminal task feature chain. The nodes of this directed graph represent the feature sets of each task, and the directed edges point from the preceding task to the following task. First, directed edges are added based on data dependencies. If task A and task B have a data dependency, a directed edge from A to B is added between nodes of task A and task B. Next, directed edges are added based on execution order. If the scheduling timestamp of task C is earlier than that of task D, and there is no data dependency between C and D, a directed edge from C to D is added between nodes of C and D. Finally, directed edges are added based on resource contention. Task pairs with resource contention are extracted from the resource contention matrix. For each task pair, directed edges are added according to the task priority feature values ​​in the task feature sets, from low to high. This means the node with the lower priority value points to the node with the higher priority value, indicating that the higher-priority task can preempt the resources of the lower-priority task. After adding all directed edges, the generated directed graph is topologically sorted, and the node order is rearranged according to the topological sorting result to obtain the terminal task feature chain.

[0028] Step S200: Perform value analysis and multi-resource coupling scheduling on the multi-resource state sequence based on the terminal task feature chain to obtain the terminal resource scheduling space.

[0029] Specifically, value analysis refers to assessing the importance of each type of resource to each task in the task chain at different points in time, in order to identify high-value resource allocation opportunities. Multi-resource coupling scheduling refers to comprehensively considering the mutual constraints and synergies among four types of resources: computing, storage, communication, and energy, to avoid optimizing one type of resource in isolation and causing other resources to become bottlenecks. In practice, firstly, based on the characteristics of each task in the terminal task feature chain, the value of the multi-resource state sequence is assessed, generating a resource contribution distribution reflecting the importance of resources. Then, based on this contribution distribution, the multi-resource state sequence is adapted and filtered, eliminating low-value resource states, and the coupling relationships between the four types of resources are analyzed to form a candidate resource map. Finally, using the tasks in the terminal task feature chain as scheduling units and the resource state combinations in the candidate resource map as optional allocation schemes, multiple feasible terminal resource scheduling schemes are generated through joint scheduling decisions. The set of these schemes constitutes the terminal resource scheduling space.

[0030] In one possible implementation, the terminal resource scheduling space is obtained by performing value analysis and multi-resource coupling scheduling on the multi-resource state sequence based on the terminal task feature chain. Step S200 further includes step S210, performing value analysis on the multi-resource state sequence based on the terminal task feature chain to obtain the resource contribution distribution. Specifically, for the i-th task in the terminal task feature chain, the computational resource demand level, storage resource demand level, and communication resource demand level, as well as the latency constraint feature value and service quality demand feature value, are obtained from its task feature set. The percentage of idle computational resources, storage resources, communication resources, and remaining available energy resources at the expected execution time of the task are read from the multi-resource state sequence, and these four percentage values ​​form a four-dimensional supply vector. Each dimension value in the supply vector is matched and mapped with the corresponding demand level, where the higher the demand level, the higher the corresponding supply percentage needs to be. According to the matching result, if the condition is met, the resource matching degree is set to 1; otherwise, it is set to 0, resulting in the four-dimensional resource satisfaction vector of the task. All tasks in the terminal task feature chain are traversed to obtain the resource satisfaction vector of each task. A value decay factor is defined, and the last task in the terminal task feature chain is taken as the starting point for calculation. The initial value of the resource contribution distribution of this task is equal to its resource satisfaction vector. The terminal task feature chain is traversed from back to front. For the j-th task, its resource contribution distribution is calculated by adding the resource satisfaction vector of the task itself to the product of the resource contribution distribution of its successor task and the value decay factor. The resource contribution distributions of each task are then summed according to resource type to obtain the total contribution value for each of the four resource types, forming the resource contribution distribution.

[0031] Step S220: Based on the resource contribution distribution, the multi-resource state sequence is adapted, filtered, and its coupling relationships are analyzed to obtain a candidate resource map. Specifically, a contribution threshold is set. For each type of resource, tasks with contribution values ​​greater than or equal to the contribution threshold are selected from the multi-resource state sequence and marked as adapted tasks for that type of resource. For each type of resource, resource state data for the corresponding time period is extracted from the multi-resource state sequence based on the timestamp and resource demand of the adapted tasks, forming a candidate resource subsequence for that type of resource. The candidate resource subsequences of computing resources, storage resources, communication resources, and energy resources are aligned along the same time axis to form a four-dimensional candidate resource matrix. The rows of this matrix represent time points, the columns represent resource types, and the matrix elements represent the available percentage of that type of resource at that time point. The coupling relationships of the candidate resource matrix are analyzed, and the Pearson correlation coefficient between the available percentages of any two types of resources at the same time point is calculated to obtain the pairwise correlation coefficients of the four types of resources. A strong coupling threshold is set, and resource pairs with correlation coefficients greater than or equal to the strong coupling threshold are marked as strongly coupled resource pairs. For each strongly coupled resource pair, a coupled resource dimension is added to the candidate resource matrix. The value of this dimension is the product of the available percentages of the resource pair, forming an extended candidate resource matrix. Each row in the extended candidate resource matrix is ​​combined as a resource state, and all rows constitute the candidate resource map.

[0032] Step S230: Perform joint scheduling decisions on the candidate resource map based on the terminal task feature chain to generate the terminal resource scheduling space. Specifically, each task feature set in the terminal task feature chain is treated as a scheduling unit, and each row of resource status combinations in the candidate resource map is used as an optional resource allocation scheme for that scheduling unit. A depth-first search algorithm is used to traverse all possible task-resource matching combinations to generate multiple terminal resource scheduling schemes. Specifically, this involves maintaining a current scheduling path list and a set of scheduled tasks. Starting from the first task in the terminal task feature chain, for the current task, select time points from the candidate resource map that simultaneously satisfy the task's computational resource requirement level, storage resource requirement level, communication resource requirement level, and remaining energy resource availability percentage. Combine the resource statuses corresponding to these time points as candidate allocation schemes for the task. Try each candidate allocation scheme sequentially, record the scheme in the current scheduling path list, add the task to the set of scheduled tasks, and then recursively process the next task. When processing the last task in the chain, store all schemes in the current scheduling path list as a complete terminal resource scheduling scheme in the scheduling scheme set. After completing all recursive explorations, schemes with conflicting resource allocation times are removed from the scheduling scheme set, i.e., schemes where the same type of resource at the same time point is allocated to two different tasks. This yields the final scheduling schemes for all terminal resources, which together constitute the terminal resource scheduling space.

[0033] Step S300: Based on the service experience evaluation architecture, perform service experience-oriented optimization on the terminal resource scheduling space to obtain the terminal's first scheduling strategy.

[0034] Specifically, a pre-built service experience evaluation architecture is used to optimize each scheduling scheme in the terminal resource scheduling space based on service experience, selecting the optimal scheme as the terminal's primary scheduling strategy. The service experience evaluation architecture is a scoring system with multi-dimensional evaluation indicators such as latency satisfaction, service quality satisfaction, scheduling fairness, and energy efficiency, mapping each resource scheduling scheme to a comprehensive service experience evaluation score. In practice, firstly, the demand characteristics of each task are analyzed from the terminal task feature chain to determine the multiple dimensions of indicators used for service experience evaluation. Then, reward and penalty rules are configured for each dimension of indicators; that is, reward points are given when the indicator performance is better than expected, and penalty points are given when it is worse than expected. The reward and penalty scores of all indicators are accumulated to form the service experience evaluation architecture. Next, this architecture is used to score each scheme in the terminal resource scheduling space, obtaining a service experience evaluation graph. Using a set service experience evaluation expectation value as a constraint, schemes that meet the expectation value are selected from the evaluation graph, forming the resource scheduling screening domain. Finally, within this selection domain, optimization is performed with the goal of minimizing total energy consumption, and the scheme with the lowest energy consumption is selected as the first scheduling strategy for the terminal.

[0035] In one possible implementation, service experience-oriented optimization is performed on the terminal resource scheduling space according to the service experience evaluation architecture to obtain the terminal's first scheduling strategy. Step S300 further includes step S310, which involves analyzing the demand characteristics based on the terminal task feature chain to determine multi-dimensional service experience evaluation indicators. Specifically, multi-dimensional service experience evaluation indicators are determined based on the latency constraint characteristics and service quality requirement characteristics of each task feature set in the terminal task feature chain. Specifically, the ratio of the actual completion time of a task to its deadline is defined as a latency satisfaction index. This index ranges from 0 to 1; when the ratio is less than or equal to 1, the value is 1 minus the ratio; when the ratio is greater than 1, the value is 0. The computing resource jitter rate, storage resource access latency, communication resource packet loss rate, and energy resource consumption rate during task execution are extracted and compared with the service quality requirement characteristic values ​​corresponding to each resource. Specifically, the comparison method is to calculate the ratio of the normalized value of the actual service quality of each resource to the service quality requirement characteristic value, and take the minimum value among the four ratios as the service quality satisfaction index. The scheduling fairness index is defined as whether a task is preempted or delayed. A value of 1 is assigned if the task is executed normally according to the scheduling order in the terminal task feature chain, and a value of 0 is assigned if a task is preempted by a higher-priority task, causing a delay. The energy efficiency index is defined as the ratio of the current remaining battery power to the total battery capacity in the terminal device's energy resource status sequence. These four indices are combined into a multi-dimensional service experience evaluation index, with each index corresponding to one dimension.

[0036] Step S320: Configure rewards and penalties based on the multi-dimensional service experience evaluation indicators to generate the service experience evaluation architecture. Specifically, configure reward and penalty parameters for each indicator in the multi-dimensional service experience evaluation indicators. For example, for the latency satisfaction indicator, set an expected satisfaction threshold. When the actual latency satisfaction is greater than or equal to the expected satisfaction threshold, a reward value is given; otherwise, a penalty value is given. The reward value is positive, and the penalty value is negative. For the service quality satisfaction indicator, set an expected matching degree threshold. When the service quality satisfaction is greater than or equal to the expected matching degree threshold, a reward value is given; otherwise, a penalty value is given. For the scheduling fairness indicator, a reward value is given when the value is 1, and a penalty value is given when the value is 0. For the energy efficiency indicator, set an expected efficiency threshold. When the energy efficiency indicator is greater than or equal to the expected efficiency threshold, a reward value is given; otherwise, a penalty value is given. The reward or penalty value for each indicator is summed to obtain the total service experience evaluation score for each task. The scoring rules for all tasks are combined to form the service experience evaluation architecture.

[0037] Step S330: Evaluate the service experience of the terminal resource scheduling space on a per-scheme basis according to the service experience evaluation architecture to obtain a service experience evaluation graph. Specifically, traverse each terminal resource scheduling scheme in the terminal resource scheduling space, and input the allocation time point and resource usage of each task in the scheme into the service experience evaluation architecture. The service experience evaluation architecture calculates the service experience evaluation score for each task under the scheme according to the reward and penalty rules configured in step S320. For any task, the service experience evaluation score is calculated as follows: extract the actual value of latency satisfaction, the actual value of service quality satisfaction, the scheduling fairness index value, and the energy efficiency index value from the execution result of the task. Substitute these four values ​​into the reward and penalty rules set in step S320, and sum the reward and penalty values ​​of the four indicators to obtain the service experience evaluation score for the task. Sum the service experience evaluation scores of all tasks in a scheme and divide by the total number of tasks in the scheme to obtain the comprehensive service experience evaluation score for the scheme. Arrange the comprehensive service experience evaluation scores of each scheme in order of scheme number to form a service experience evaluation graph.

[0038] Step S340: Using the expected service experience evaluation as a constraint, and combining the service experience evaluation graph, the terminal resource scheduling space is filtered to obtain a resource scheduling filtering domain. Specifically, a service experience evaluation expectation threshold is set. The comprehensive service experience evaluation score of each scheme is traversed from the service experience evaluation graph, and terminal resource scheduling schemes with a comprehensive score greater than or equal to the threshold are filtered out and placed into the resource scheduling filtering domain. If the resource scheduling filtering domain is empty after filtering, it means that no scheme can currently reach the expected level. The threshold is then lowered, and the above filtering process is repeated until the resource scheduling filtering domain is not empty. This adaptive adjustment mechanism ensures that at least one feasible scheduling scheme can be filtered out under any circumstances.

[0039] Step S350: Optimize the resource scheduling filtering domain by minimizing energy consumption to generate the terminal's first scheduling strategy. Specifically, for each terminal resource scheduling scheme in the resource scheduling filtering domain, calculate the total energy consumption of that scheme. The total energy consumption is calculated as follows: for each task in the scheme, read the average power consumption of the task during its allocated execution time period from the energy resource state sequence, multiply the average power consumption by the execution duration of the task to obtain the task's energy consumption value. Summate the energy consumption values ​​of all tasks to obtain the total energy consumption of the scheme. Traverse all schemes in the resource scheduling filtering domain, compare the total energy consumption of each scheme, and select the terminal resource scheduling scheme with the minimum total energy consumption as the terminal's first scheduling strategy. This strategy minimizes the power consumption of the terminal device while meeting service experience expectations.

[0040] Step S400: Based on the cloud-edge-device coordinator, the terminal performs task offloading decisions and resource reallocation on the terminal's first scheduling strategy to obtain the terminal's second scheduling strategy.

[0041] Specifically, the cloud-edge-device coordinator, deployed in the kernel layer of the terminal device's operating system, is invoked to perform task offloading decisions and resource reallocation for the terminal's first scheduling policy. Task offloading decision refers to determining which tasks in the terminal's task feature chain are suitable for migration from the terminal device to the edge server or cloud server for execution. Resource reallocation refers to redistributing the resources released locally on the terminal to downstream tasks affected by the task migration, based on the offloading decision results. The cloud-edge-device coordinator is a software module deployed within the terminal device, responsible for coordinating the allocation of computing tasks among the terminal, edge server, and cloud server. It maintains cloud-side resource lists and edge-side resource lists, recording the network addresses, computing resources, and storage resources of the remote cloud server and edge gateway, respectively. During execution, the cloud-edge-device coordinator first performs migration cost analysis and experience gain identification on the terminal's task feature chain, generating a task migration projection graph describing the feasibility and benefits of migration. Then, based on this graph, the portability of each task is determined, yielding a task migration determination result. Next, this determination result is compared with the terminal's first scheduling policy to identify the scope of the impact of task migration on the existing resource allocation, forming the task migration resource impact domain. Finally, based on the influence domain, an unloading decision is made for the tasks in the terminal's first scheduling strategy. Tasks that are determined to be migrateable are removed from the local machine and their resources are released. At the same time, the released resources are reallocated to the affected downstream tasks, thereby generating the terminal's second scheduling strategy.

[0042] In one possible implementation, based on the cloud-edge-device coordinator's first scheduling strategy for the terminal, a task offloading decision and resource reallocation are performed to obtain a second scheduling strategy for the terminal. Step S400 further includes step S410, where the cloud-edge-device coordinator performs migration cost analysis and experience gain identification on the terminal task feature chain to obtain a task migration projection graph. Specifically, migration cost analysis refers to quantifying the additional overhead required to migrate a task from the terminal to the edge or cloud, including transmission latency, computation latency, transmission energy consumption, computation energy consumption, and communication costs. Experience gain identification refers to evaluating the degree of improvement in user experience after task migration, including the percentage reduction in latency and the percentage improvement in service quality. The task migration projection graph is a graph structure with tasks as nodes and the ratio of migration cost to experience gain as edge weights, used for migrationability determination. In practice, the cloud-edge-device coordinator first makes cloud-edge migration decisions on the terminal task feature chain and generates an initial task migration graph. Then, the path of the initial migration graph is optimized to obtain an optimized task migration graph. Next, a simulated migration is performed according to the optimized graph and the data is recorded to form a task migration simulation dataset. Finally, the migration cost is analyzed and the experience gain is identified on the dataset to construct a task migration projection graph network.

[0043] Step S420: The portability of the terminal task feature chain is determined based on the task migration projection network to obtain the task migration determination result. Specifically, the portability score of each task is extracted from the task migration projection network. That is, for each task, all possible migration paths originating from that task are identified, the reciprocal of the ratio of migration cost to experience gain on each path is taken, and the sum of these reciprocals is obtained to obtain the portability score of the task. A portability threshold is set. If the portability score of a task is greater than or equal to the portability threshold, the task is determined to be portable; if it is less than the portability threshold, it is determined to be non-portable. The determination result of each task is marked according to its task identifier to form the task migration determination result corresponding to each task.

[0044] Step S430: Based on the task migration determination result, the impact identification of the terminal's first scheduling policy is performed to obtain the task migration resource impact domain. Specifically, each task in the terminal's first scheduling policy is compared task-by-task with the task migration determination result obtained in step 420. For tasks determined to be migrateable, the amount of computing resources, storage resources, communication resources, and energy resources originally allocated to the task are extracted from the terminal's first scheduling policy. These resources are marked as releaseable resources, indicating that if the task is migrated, these resources can be reclaimed and redistributed to other tasks. For tasks determined not to be migrateable, the resource dependencies of the task are extracted from the terminal's first scheduling policy, i.e., a list of resources occupied by all preceding tasks that the task depends on, including the specific amount and time period occupied for each resource. The list of releaseable resources for all migrateable tasks is merged with the list of resource dependencies for tasks not migrateable to form the task migration resource impact domain. This impact domain records the resource release amount for each migrateable task and a list of downstream tasks affected by the resource release.

[0045] Step S440: Based on the task migration resource influence domain, perform task unloading decisions and resource reallocation on the terminal's first scheduling policy to generate the terminal's second scheduling policy. Specifically, for each migrateable task in the task migration resource influence domain, according to the migration target recorded in its task migration judgment result (i.e., edge-side or cloud-side), the task is removed from the task execution list of the terminal's first scheduling policy, and the computing resources, storage resources, communication resources, and energy resources occupied by the task locally on the terminal are marked as released in the resource allocation table of the terminal's first scheduling policy. For downstream tasks affected by the migrated task, i.e., those tasks recorded in the task migration resource influence domain that depend on the output data of the migrated task, resources are reallocated from the released resource pool according to the resource requirement characteristics of each downstream task. All affected upstream and downstream tasks are sorted from high to low according to their task priority characteristics, with tasks with higher priority values ​​being allocated resources first. For each downstream task to be allocated, a resource block that can simultaneously meet the minimum value of its computing, storage, communication, and energy resource requirements is searched from the released resource pool, and this resource block is allocated to the downstream task. After completing the task unloading decisions for all migrated tasks and the resource reallocation for all affected downstream tasks, an updated complete scheduling policy is generated and used as the terminal's second scheduling policy.

[0046] In one possible implementation, the cloud-edge-device coordinator performs migration cost analysis and experience gain identification on the terminal task feature chain to obtain a task migration projection graph. Step S410 further includes step S411, which performs cloud-edge migration decision-making on the terminal task feature chain based on the cloud-edge-device coordinator to obtain an initial task migration graph. Specifically, for each task in the terminal task feature chain, the cloud-edge-device coordinator first determines whether it can be offloaded based on its latency constraint characteristics. For example, if the latency is very short, the task is determined not to be offloaded and must be kept for local execution. If the latency is at a medium level, the task is determined to be offloadable to edge resources. If the latency is relatively relaxed, the task is determined to be offloadable to cloud resources or edge resources. For tasks determined to be unloadable, the process of migrating them to the cloud or edge is simulated: the transmission latency is obtained by dividing the sum of the task code size and the input data volume by the communication bandwidth of the cloud or edge; the computation latency is obtained by dividing the computation resource requirement of the task on the cloud or edge by the available computation resources on the cloud or edge, and then multiplying by the computational complexity coefficient of the task; the sum of the transmission latency and the computation latency is taken as the total execution latency of the task on the cloud or edge. This total execution latency is compared with the total execution latency of the same task executed locally on the terminal. If the total execution latency after migration is less than the local total execution latency, the migration path is marked as a valid migration path. The path of each task executed locally and all the migration paths marked as valid are respectively regarded as nodes in a graph structure, and the data dependencies between tasks are regarded as directed edges between nodes, constructing a directed acyclic graph, which is the initial task migration graph.

[0047] Step S412: Perform path optimization processing on the initial task migration graph to obtain an optimized task migration graph. Specifically, for each possible migration path from the source task to the target task in the initial task migration graph, calculate the path weight of that path. That is, subtract the sum of the migration costs corresponding to each task node along the path from the sum of the service experience gains corresponding to each task node along the path, and use the difference as the weight of the path. The migration cost includes a weighted sum of three components: transmission energy consumption, computing energy consumption, and communication cost. The service experience gain includes a weighted sum of two components: latency reduction and service quality improvement. The Dijkstra algorithm is used to search for the shortest path in the initial task migration graph. The weight of each directed edge is used as the distance metric for that edge. Starting from all source nodes in the graph, the path with the minimum cumulative weight reaching each target node is searched. After the search is complete, paths with a cumulative weight greater than a predetermined threshold are removed from the graph, along with redundant migration decision nodes on these paths. The path with the minimum cumulative weight and its corresponding node are retained, resulting in an optimized directed acyclic graph, i.e., the optimized task migration graph.

[0048] Step S413: Perform simulated migration of the terminal task feature chain according to the task migration optimization graph to obtain a task migration simulation dataset. Specifically, according to the migration decisions recorded in the task migration optimization graph, simulated migration is performed sequentially for each task in the terminal task feature chain. For each migration decision node in the optimization graph, based on the migration target marked on the node, the corresponding resource allocator is invoked to allocate corresponding computing, storage, communication, and energy resources to the task. During the simulation execution, the actual latency, actual energy consumption, and actual service quality of each task are recorded. The simulation execution results of each task are arranged in chronological order in the terminal task feature chain to form a dataset. Each record in this dataset contains the task identifier, local execution latency before migration, actual execution latency after migration, local energy consumption before migration, actual energy consumption after migration, service quality before migration, and actual service quality after migration. Based on this, the changes in latency, energy consumption, and service quality before and after migration for each task are calculated. This dataset is the task migration simulation dataset.

[0049] Step S414: Perform migration cost analysis and experience gain identification on the task migration simulation dataset to generate the task migration projection graph. Specifically, traverse each simulated migration result record in the task migration simulation dataset. For each record, extract migration cost parameters, including the increase in transmission latency, the increase in computation latency, the increase in transmission energy consumption, the increase in computation energy consumption, and communication costs. Simultaneously, extract experience gain parameters, including the proportion of latency reduction and the proportion of service quality improvement. Combine the extracted migration cost parameters and experience gain parameters into a two-dimensional vector. The first dimension of this vector is the weighted sum of the five migration cost parameters, and the second dimension is the weighted sum of the two experience gain parameters. Each simulated migration result corresponds to a two-dimensional vector. Classify and summarize all the two-dimensional vectors of the simulated migration results according to the migration path. That is, in the directed graph structure, with tasks as nodes, for each valid migration path from task A to task B, the ratio of the first dimension to the second dimension of the two-dimensional vector corresponding to the path is used as the weight of the directed edge from node A to node B. After the aggregation is completed, a new graph structure is formed. The nodes of this graph structure are the tasks in the terminal task feature chain, and the weight of the directed edges between the nodes is the ratio of migration cost to experience gain. This graph structure is the task migration inference graph network.

[0050] Step S500: Perform uncertainty modeling and uncertainty adversarial compensation on the second scheduling strategy of the terminal to obtain the third scheduling strategy of the terminal, and execute the multi-resource dynamic adjustment of the terminal device according to the third scheduling strategy of the terminal.

[0051] Specifically, uncertainty modeling and uncertainty mitigation compensation are applied to the second scheduling strategy for the terminal to generate a robust third scheduling strategy. Uncertainty modeling refers to quantifying the resource fluctuations that the second scheduling strategy may encounter during execution. These fluctuations include variations in execution time of computing resources, access latency of storage resources, bandwidth of communication resources, and consumption rate of energy resources. Uncertainty mitigation compensation involves pre-configuring resource redundancy and propagation suppression mechanisms to reduce the impact of resource fluctuations on the scheduling strategy's execution effectiveness. In practice, a resource scheduling uncertainty model is first established for the second scheduling strategy to describe the probability distribution of the actual availability of various resources. Based on this model, the terminal scheduling regret value is calculated and compared with a preset scheduled regret value to determine the vulnerability of the current scheduling strategy to uncertainty. If the terminal scheduling regret value is greater than or equal to the preset scheduling regret value, the mitigation compensation process is initiated: key uncertain tasks in the strategy, their resource fluctuation ranges, and task dependencies are identified based on the uncertainty model; resource redundancy compensation is configured for these tasks, and dependency propagation suppression is configured for their downstream related tasks. The compensation and suppression configuration results are merged and updated into the terminal's second scheduling policy to obtain the terminal's third scheduling policy. Finally, according to the resource allocation parameters and scheduling sequence in the terminal's third scheduling policy, instructions are sent to each resource manager through the terminal device's resource scheduler to complete the dynamic adjustment of multiple resources and task execution.

[0052] In one possible implementation, uncertainty modeling and uncertainty adversarial compensation are performed on the second scheduling strategy of the terminal to obtain a third scheduling strategy for the terminal. Step S500 further includes step S510, performing uncertainty modeling on the second scheduling strategy of the terminal to obtain a resource scheduling uncertainty model. Specifically, for each resource allocation decision point in the second scheduling strategy of the terminal, the uncertainty factors that may arise during its execution are extracted. These uncertainty factors include: fluctuations in the execution time of computing resources, i.e., the variation in the actual computing resource usage time of the same task in different execution rounds; fluctuations in the access latency of storage resources; fluctuations in the bandwidth of communication resources; and fluctuations in the remaining power consumption rate of energy resources. For each type of uncertainty factor, a normal distribution model is used for description. Historical operating data of the terminal device in the past hour are collected, and the arithmetic mean and standard deviation of the execution time of computing resources, the arithmetic mean and standard deviation of the access latency of storage resources, the arithmetic mean and standard deviation of the bandwidth of communication resources, and the arithmetic mean and standard deviation of the energy consumption rate are calculated respectively. The four means and four standard deviations obtained above are used as the benchmark parameters of the uncertainty model to construct a four-dimensional joint normal distribution model, which is the resource scheduling uncertainty model. The model takes a specific point in time as input and outputs the probability distribution of the actual availability of four types of resources—computing, storage, communication, and energy—at that point in time.

[0053] Step S520: Generate a terminal scheduling regret value based on the resource scheduling uncertainty model, and determine whether the terminal scheduling regret value is greater than or equal to a predetermined scheduling regret value. Specifically, N possible actual resource state scenarios are sampled from the resource scheduling uncertainty model, for example, N is 1000. The sampling process is as follows: 1000 sets of random numbers are randomly selected from a four-dimensional joint normal distribution. Each set of random numbers contains four values, corresponding to the offset ratio of the actual available amount of four types of resources relative to the predicted available amount. For each sampled scenario, the terminal's second scheduling strategy is simulated and executed under that scenario. During the simulation, the actual resource amount obtained by each task is adjusted according to the resource offset ratio in the scenario. For each scenario, the absolute value of the deviation between the actual completion time and the planned completion time of each task under that scenario is calculated. Then, the absolute values ​​of the deviations of all tasks are summed to obtain the scheduling deviation value under that scenario. The scheduling deviation values ​​under 1000 scenarios are added together and divided by 1000 to obtain the expected scheduling deviation value. Simultaneously, in each scenario, the number of times a task fails due to insufficient resources (i.e., the task fails to obtain its minimum required resources) is counted. The sum of these failure counts across 1000 scenarios is divided by 1000 to obtain the expected number of failures. The expected scheduling deviation is multiplied by a first weighting coefficient, and this is added to the expected number of failures multiplied by a second weighting coefficient to obtain the terminal scheduling regret value. A predetermined scheduling regret value is set, and the calculated terminal scheduling regret value is determined to be greater than or equal to this predetermined regret value.

[0054] Step S530: If the terminal scheduling regret value is greater than or equal to the predetermined scheduling regret value, resource redundancy compensation and task dependency propagation suppression processing are performed on the terminal's second scheduling strategy according to the resource scheduling uncertainty model to generate the terminal's third scheduling strategy. Specifically, when the terminal scheduling regret value is greater than or equal to the predetermined scheduling regret value, it indicates that the current terminal's second scheduling strategy is insufficient to resist resource uncertainty, and an adversarial compensation process needs to be initiated. First, the resource types that cause the greatest uncertainty and the tasks that are most severely affected are identified from the resource scheduling uncertainty model, and these tasks are marked as key uncertainty tasks. Then, for each key uncertainty task, additional redundant resources are allocated to it according to its resource fluctuation range to form a resource redundancy compensation configuration result. At the same time, for downstream tasks affected by key uncertainty tasks, dependency propagation paths are constructed and propagation suppression parameters are configured, including delayed triggering, execution isolation, and resource reservation, to form a propagation suppression configuration result. Finally, the resource redundancy compensation configuration result and the propagation suppression configuration result are merged and updated into the terminal's second scheduling strategy to generate the terminal's third scheduling strategy.

[0055] In one possible implementation, the terminal's second scheduling strategy is processed for resource redundancy compensation and task dependency propagation suppression according to the resource scheduling uncertainty model to generate the terminal's third scheduling strategy. Step S530 further includes step S531, determining the key uncertain tasks of the terminal's second scheduling strategy, as well as the resource fluctuation range and task dependency relationships corresponding to the key uncertain tasks, based on the resource scheduling uncertainty model. Specifically, the variance value of the probability distribution corresponding to each resource type is extracted from the resource scheduling uncertainty model. The larger the variance value, the more drastic the fluctuation in the actual available quantity of that resource type. The variance values ​​of computing resources, storage resources, communication resources, and energy resources are sorted from largest to smallest, and the three resource types with the largest variance values ​​are selected as the resource types with the largest fluctuations. Then, each task in the terminal's second scheduling strategy is traversed. For each task, the percentage of its resource demand for the three resource types with the largest fluctuations relative to the total resource demand of that task is calculated. If this percentage exceeds a preset percentage, the task is marked as a key uncertain task. For each marked key uncertainty task, the fluctuation range of resources within the planned execution period is extracted from the resource scheduling uncertainty model. The lower limit of the fluctuation range is defined as negative twice the standard deviation of the resource prediction mean, and the upper limit is defined as positive twice the standard deviation of the resource prediction mean. Simultaneously, a list of task dependencies for this key uncertainty task is read from the terminal task feature chain. This list contains the task identifiers of all preceding tasks that this task depends on, as well as the task identifiers of all subsequent tasks that depend on this task.

[0056] Step S532: Configure multi-dimensional resource redundancy compensation for the key uncertain tasks based on the resource fluctuation range, and obtain the resource redundancy compensation configuration result. Specifically, for each key uncertain task, firstly, obtain the baseline demand of various resources originally allocated to the task in the terminal's second scheduling strategy. Then, calculate the resource redundancy compensation amount based on the resource fluctuation range of the task extracted in step S531. The redundancy compensation amount is divided into two parts: the first part is the basic redundancy amount, for example, 10% of the baseline demand amount; the second part is the fluctuation compensation amount, which is the difference between the upper limit of the resource fluctuation range and the baseline demand amount. Add the basic redundancy amount and the fluctuation compensation amount to obtain the resource redundancy compensation amount for the task. According to this redundancy compensation amount, additional computing resources, storage resources, communication resources, and energy resources are reserved for each key uncertain task. Record the identifier of each key uncertain task, the corresponding redundancy compensation amount, and the reserved resource information in a resource reservation table, which is the resource redundancy compensation configuration result.

[0057] Step S533: Identify downstream related tasks affected by the key uncertainty task based on the task dependency relationship, and construct a dependency propagation path from the key uncertainty task to the downstream related tasks. Specifically, for each key uncertainty task, extract all subsequent task identifiers from the task dependency relationship list obtained in step S531, and mark these subsequent tasks as downstream related tasks, indicating that the execution of these tasks depends on the output of the key uncertainty task. For each pair of key uncertainty tasks and their downstream related tasks, starting with the key uncertainty task as the starting node and the downstream related task as the ending node, traverse the intermediate nodes sequentially along the direction of the directed edges defined in the terminal task feature chain, connecting the starting node, each intermediate node, and the ending node in the traversal order to form a directed path, which is the dependency propagation path. Each key uncertainty task may correspond to multiple dependency propagation paths, and each path corresponds to a downstream related task.

[0058] Step S534: Configure propagation suppression for the dependency propagation path and obtain the propagation suppression configuration result. The propagation suppression configuration result includes delay triggering configuration parameters, execution isolation configuration parameters, and resource reservation configuration parameters. Specifically, for each dependency propagation path constructed in step S533, configure the following three types of propagation suppression parameters: The first type is delay triggering configuration parameters: Set a trigger delay threshold. When the actual execution time of the upstream task (i.e., the key uncertain task) on the dependency propagation path exceeds its planned execution time, the start of the upstream and downstream related tasks on that path is delayed. The second type is execution isolation configuration parameters: Set an execution isolation strategy to allocate the upstream task and the downstream related task to different processor cores for execution. At the same time, physically isolate the memory area used by the upstream task from the memory area used by the downstream related task to avoid memory pollution of the upstream task affecting the downstream task. The third type is resource reservation configuration parameters: Reserve spare resources for the downstream related task. When the upstream task consumes more resources than expected due to resource fluctuations, the downstream related task can directly use the reserved spare resources without resource contention. The three types of configuration parameters are packaged according to the organization structure of the propagation path to form a data structure, which is the propagation suppression configuration result.

[0059] Step S535: Update and optimize the terminal's second scheduling strategy based on the resource redundancy compensation configuration result and the propagation suppression configuration result to obtain the terminal's third scheduling strategy. Specifically, merge the resource reservation table in the resource redundancy compensation configuration result into the resource allocation table of the terminal's second scheduling strategy. For each key uncertain task recorded in the resource reservation table, update its original resource allocation in the terminal's second scheduling strategy to the original resource requirement plus the resource redundancy compensation amount. The updated resource allocation serves as the upper limit of the actual available resources for that task during execution. Write the delay trigger configuration parameters, execution isolation configuration parameters, and resource reservation configuration parameters from the propagation suppression configuration result into the task scheduling control block of the terminal's second scheduling strategy. The task scheduling control block is a data structure in the terminal operating system used to manage the scheduling information of each task. After writing the above parameters, the task scheduler will determine the start time of downstream tasks according to the delay trigger configuration parameters, allocate processor cores and memory regions according to the execution isolation configuration parameters, and reserve spare resources for downstream tasks according to the resource reservation configuration parameters during execution. After completing the update of the resource allocation table and the writing of the parameters in the task scheduling control block, a brand-new scheduling strategy is obtained, which is the terminal's third scheduling strategy. Finally, according to the updated resource allocation table and scheduling schedule in the terminal's third scheduling strategy, the terminal device's resource scheduler sends resource allocation instructions to the computing resource manager, storage resource manager, communication resource manager, and energy resource manager, respectively. Each resource manager allocates resources and dynamically adjusts them according to the parameters in the instructions until all tasks are completed.

[0060] This application embodiment acquires the multi-resource state sequence and current task set of the terminal device, constructs a terminal task feature chain, performs value analysis and multi-resource coupling scheduling based on the feature chain, generates a terminal resource scheduling space, and optimizes the scheduling space based on the service experience evaluation architecture to select the terminal's first scheduling strategy. The cloud-edge-device coordinator performs task offloading decisions and resource reallocation on the first scheduling strategy, offloading transferable tasks to the edge or cloud and reallocating locally released resources to obtain the terminal's second scheduling strategy. Uncertainty modeling and adversarial compensation are performed on the second scheduling strategy, and a robust terminal third scheduling strategy is generated through resource redundancy compensation and dependency propagation suppression. Based on this, the dynamic adjustment of multiple resources and other technical means are executed to solve the technical problem of low scheduling efficiency caused by the inability to dynamically adapt to changes in task requirements and fluctuations in resource status in existing intelligent converged terminals. It achieves the technical effect of dynamically adjusting resource allocation according to changes in task requirements and fluctuations in resource status and improving the efficiency of multi-resource collaborative scheduling.

[0061] In the above text, refer to Figure 1 A resource scheduling optimization method for intelligent converged terminals according to an embodiment of the present invention is described in detail. Next, reference will be made to... Figure 2 This invention describes a resource scheduling optimization system for intelligent converged terminals according to an embodiment of the present invention.

[0062] According to an embodiment of the present invention, a resource scheduling optimization system for intelligent converged terminals addresses the technical problem of low scheduling efficiency caused by the inability of existing intelligent converged terminals to dynamically adapt to changes in task requirements and fluctuations in resource status. The system achieves the technical effect of dynamically adjusting resource allocation based on changes in task requirements and fluctuations in resource status, thereby improving the efficiency of multi-resource collaborative scheduling. The resource scheduling optimization system for intelligent converged terminals includes: a terminal task feature chain construction module 10, a terminal resource scheduling space acquisition module 20, a service experience-oriented optimization module 30, a resource reallocation module 40, and an uncertainty mitigation compensation module 50.

[0063] The terminal task feature chain construction module 10 is used to acquire the multi-resource state sequence and current task set of the terminal device, and construct the terminal task feature chain based on the current task set; the terminal resource scheduling space acquisition module 20 is used to perform value analysis and multi-resource coupling scheduling on the multi-resource state sequence based on the terminal task feature chain to acquire the terminal resource scheduling space; the service experience-oriented optimization module 30 is used to perform service experience-oriented optimization on the terminal resource scheduling space based on the service experience evaluation architecture to acquire the terminal first scheduling strategy; the resource reallocation module 40 is used to perform task offloading decision and resource reallocation on the terminal first scheduling strategy based on the cloud-edge-device coordinator to acquire the terminal second scheduling strategy; the uncertainty adversarial compensation module 50 is used to perform uncertainty modeling and uncertainty adversarial compensation on the terminal second scheduling strategy to acquire the terminal third scheduling strategy, and execute the multi-resource dynamic adjustment of the terminal device according to the terminal third scheduling strategy.

[0064] The detailed description of the specific configuration of the terminal resource scheduling space acquisition module 20 is as follows: As described above, the terminal resource scheduling space is obtained by performing value analysis and multi-resource coupling scheduling on the multi-resource state sequence based on the terminal task feature chain. The terminal resource scheduling space acquisition module 20 may further include: a value analysis unit for performing value analysis on the multi-resource state sequence based on the terminal task feature chain to obtain the resource contribution distribution; an adaptation and filtering unit for performing adaptation and filtering and coupling relationship sorting on the multi-resource state sequence based on the resource contribution distribution to obtain a candidate resource map; and a joint scheduling decision unit for performing joint scheduling decision on the candidate resource map based on the terminal task feature chain to generate the terminal resource scheduling space.

[0065] The detailed description of the specific configuration of the service experience-oriented optimization module 30 is as follows: As mentioned above, the service experience-oriented optimization of the terminal resource scheduling space is performed according to the service experience evaluation architecture to obtain the terminal's first scheduling strategy. The service experience-oriented optimization module 30 may further include: a demand characteristic analysis unit for analyzing demand characteristics according to the terminal task feature chain to determine multi-dimensional service experience evaluation indicators; a reward and punishment configuration unit for configuring rewards and punishments according to the multi-dimensional service experience evaluation indicators to generate the service experience evaluation architecture; a scheme-by-scheme service experience evaluation unit for performing scheme-by-scheme service experience evaluation of the terminal resource scheduling space according to the service experience evaluation architecture to obtain a service experience evaluation graph; a terminal resource scheduling space filtering unit for filtering the terminal resource scheduling space based on the service experience evaluation expectation as a constraint and in combination with the service experience evaluation graph to obtain a resource scheduling filtering domain; and an energy consumption minimization optimization unit for performing energy consumption minimization optimization of the resource scheduling filtering domain to generate the terminal's first scheduling strategy.

[0066] The detailed configuration of the resource reallocation module 40 is explained below: As described above, the resource reallocation module 40 performs task offloading decisions and resource reallocation on the terminal's first scheduling strategy based on the cloud-edge-device coordinator to obtain the terminal's second scheduling strategy. The resource reallocation module 40 may further include: a migration cost analysis unit for performing migration cost analysis and experience gain identification on the terminal's task feature chain based on the cloud-edge-device coordinator to obtain a task migration projection graph; a portability determination unit for performing portability determination on the terminal's task feature chain based on the task migration projection graph to obtain a task migration determination result; an impact identification unit for performing impact identification on the terminal's first scheduling strategy based on the task migration determination result to obtain a task migration resource impact domain; and a task offloading decision unit for performing task offloading decisions and resource reallocation on the terminal's first scheduling strategy based on the task migration resource impact domain to generate the terminal's second scheduling strategy.

[0067] Specifically, the migration cost analysis and experience gain identification of the terminal task feature chain based on the cloud-edge-device coordinator are used to obtain a task migration projection graph. The migration cost analysis unit may further include: a cloud-edge migration decision subunit used to make cloud-edge migration decisions based on the terminal task feature chain of the cloud-edge-device coordinator to obtain an initial task migration graph; a path optimization processing subunit used to perform path optimization processing on the initial task migration graph to obtain an optimized task migration graph; a migration simulation subunit used to perform simulated migration of the terminal task feature chain based on the optimized task migration graph to obtain a task migration simulation dataset; and an experience gain identification subunit used to analyze migration costs and identify experience gains in the task migration simulation dataset to generate the task migration projection graph.

[0068] The detailed description of the specific configuration of the uncertainty countermeasure compensation module 50 is explained as follows: As described above, uncertainty modeling and uncertainty countermeasure compensation are performed on the second scheduling strategy of the terminal to obtain the third scheduling strategy of the terminal. The uncertainty countermeasure compensation module 50 may further include: an uncertainty modeling unit for performing uncertainty modeling on the second scheduling strategy of the terminal to obtain a resource scheduling uncertainty model; a terminal scheduling regret value generation unit for generating a terminal scheduling regret value according to the resource scheduling uncertainty model and determining whether the terminal scheduling regret value is greater than or equal to a predetermined scheduling regret value; and a resource redundancy compensation unit for performing resource redundancy compensation and task dependency propagation suppression processing on the second scheduling strategy of the terminal according to the resource scheduling uncertainty model if the terminal scheduling regret value is greater than or equal to the predetermined scheduling regret value, to generate the third scheduling strategy of the terminal.

[0069] The process involves performing resource redundancy compensation and task dependency propagation suppression on the second terminal scheduling strategy based on the resource scheduling uncertainty model to generate the third terminal scheduling strategy. The resource redundancy compensation unit may further include: a key uncertainty task determination subunit, used to determine the key uncertainty tasks of the second terminal scheduling strategy according to the resource scheduling uncertainty model, as well as the resource fluctuation range and task dependency relationship corresponding to the key uncertainty tasks; a multi-dimensional resource redundancy compensation configuration subunit, used to configure multi-dimensional resource redundancy compensation for the key uncertainty tasks according to the resource fluctuation range, and obtain resource redundancy compensation configuration results; a downstream related task identification subunit, used to identify downstream related tasks affected by the key uncertainty tasks according to the task dependency relationship, and construct a dependency propagation path from the key uncertainty tasks to the downstream related tasks; a propagation suppression configuration subunit, used to configure propagation suppression on the dependency propagation path, and obtain propagation suppression configuration results, the propagation suppression configuration results including delay trigger configuration parameters, execution isolation configuration parameters, and resource reservation configuration parameters; and a third terminal scheduling strategy generation subunit, used to update and optimize the second terminal scheduling strategy according to the resource redundancy compensation configuration results and the propagation suppression configuration results, and obtain the third terminal scheduling strategy.

[0070] The detailed description of the specific configuration of the terminal task feature chain construction module 10 is explained as follows: As mentioned above, the terminal task feature chain construction module 10 may further include: the multi-resource state sequence includes a computing resource state sequence, a storage resource state sequence, a communication resource state sequence, and an energy resource state sequence.

[0071] The terminal task feature chain construction module 10, which constructs a terminal task feature chain based on the current task set, may further include: a task attribute feature extraction unit for extracting task attribute features from the current task set to obtain each task feature set; and a chain-linked encoding unit for performing chain-linked encoding on each task feature set based on the data dependency relationship, execution order relationship, and resource competition relationship between the current task sets to generate the terminal task feature chain.

[0072] The resource scheduling optimization system for intelligent converged terminals provided in this embodiment of the invention can execute the resource scheduling optimization method for intelligent converged terminals provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0073] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0074] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A resource scheduling optimization method for intelligent converged terminals, characterized in that, The method includes: Obtain the multi-resource state sequence and current task set of the terminal device, and construct the terminal task feature chain based on the current task set; Based on the terminal task feature chain, the multi-resource state sequence is subjected to value analysis and multi-resource coupling scheduling to obtain the terminal resource scheduling space. The value analysis refers to evaluating the importance of each type of resource to each task in the task chain at different time points. The multi-resource coupling scheduling refers to comprehensively considering the mutual constraints and synergistic relationships among the four types of resources: computing, storage, communication and energy. Based on the service experience evaluation architecture, the terminal resource scheduling space is optimized in a service experience-oriented manner to obtain the terminal's first scheduling strategy. Based on the cloud-edge-device coordinator, the terminal's first scheduling strategy is used to make task offloading decisions and resource reallocation, and the terminal's second scheduling strategy is obtained. Uncertainty modeling and uncertainty adversarial compensation are performed on the second scheduling strategy of the terminal to obtain the third scheduling strategy of the terminal, and the multi-resource dynamic adjustment of the terminal device is executed according to the third scheduling strategy of the terminal. The step of performing uncertainty modeling and uncertainty adversarial compensation on the second scheduling strategy of the terminal to obtain the third scheduling strategy of the terminal includes: Uncertainty modeling is performed on the second scheduling strategy of the terminal to obtain a resource scheduling uncertainty model; The terminal scheduling regret value is generated based on the resource scheduling uncertainty model, and it is determined whether the terminal scheduling regret value is greater than or equal to the predetermined scheduling regret value. If the terminal scheduling regret value is greater than or equal to the predetermined scheduling regret value, the terminal second scheduling strategy is processed for resource redundancy compensation and task dependency propagation suppression according to the resource scheduling uncertainty model, and the terminal third scheduling strategy is generated. The step of performing resource redundancy compensation and task dependency propagation suppression processing on the second terminal scheduling strategy based on the resource scheduling uncertainty model to generate the third terminal scheduling strategy includes: The key uncertain tasks of the terminal's second scheduling strategy are determined based on the resource scheduling uncertainty model, as well as the resource fluctuation range and task dependency relationship corresponding to the key uncertain tasks. The key uncertain tasks are the resource types that cause the greatest uncertainty and the tasks that are most severely affected, identified from the resource scheduling uncertainty model. Based on the resource fluctuation range, perform multi-dimensional resource redundancy compensation configuration for the key uncertainty tasks, and obtain the resource redundancy compensation configuration results. Based on the task dependencies, identify downstream related tasks affected by the key uncertain task, and construct the dependency propagation path from the key uncertain task to the downstream related task; Configure propagation suppression for the dependent propagation path and obtain the propagation suppression configuration result, which includes delayed triggering configuration parameters, execution isolation configuration parameters, and resource reservation configuration parameters. The second scheduling strategy for the terminal is updated and optimized based on the resource redundancy compensation configuration result and the propagation suppression configuration result, and the third scheduling strategy for the terminal is obtained.

2. The resource scheduling optimization method for intelligent converged terminals as described in claim 1, characterized in that, Based on the terminal task feature chain, value analysis and multi-resource coupling scheduling are performed on the multi-resource state sequence to obtain the terminal resource scheduling space, including: The multi-resource state sequence is analyzed for value based on the terminal task feature chain to obtain the resource contribution distribution. Based on the resource contribution distribution, the multi-resource state sequence is adapted, filtered, and its coupling relationships are sorted out to obtain a candidate resource map. Based on the terminal task feature chain, a joint scheduling decision is made on the candidate resource map to generate the terminal resource scheduling space.

3. The resource scheduling optimization method for intelligent converged terminals as described in claim 1, characterized in that, Based on the service experience evaluation architecture, the terminal resource scheduling space is optimized in a service experience-oriented manner to obtain the terminal's first scheduling strategy, including: Based on the terminal task feature chain, demand characteristics are analyzed to determine multi-dimensional indicators for service experience evaluation. Based on the multi-dimensional service experience evaluation indicators, reward and punishment configurations are performed to generate the service experience evaluation architecture. Based on the service experience evaluation architecture, a service experience evaluation of the terminal resource scheduling space is performed on a case-by-case basis to obtain a service experience evaluation graph. Using the expected service experience evaluation as a constraint, and combining the service experience evaluation graph, the terminal resource scheduling space is filtered to obtain the resource scheduling filtering domain; The resource scheduling filtering domain is optimized by minimizing energy consumption to generate the first scheduling strategy for the terminal.

4. The resource scheduling optimization method for intelligent converged terminals as described in claim 1, characterized in that, Based on the cloud-edge-device coordinator's decision on task offloading and resource reallocation according to the terminal's first scheduling policy, a second scheduling policy for the terminal is obtained, including: Based on the cloud-edge-device coordinator, the migration cost analysis and experience gain identification are performed on the terminal task feature chain to obtain the task migration inference graph network. Based on the task migration inference graph, the portability of the terminal task feature chain is determined, and the task migration determination result is obtained. Based on the task migration determination result, the impact of the terminal's first scheduling strategy is identified to obtain the task migration resource impact domain; Based on the resource influence domain of the task migration, the terminal's first scheduling strategy is used to make task unloading decisions and resource reallocation, thereby generating the terminal's second scheduling strategy.

5. The resource scheduling optimization method for intelligent converged terminals as described in claim 4, characterized in that, Based on the cloud-edge-device coordinator, the migration cost is analyzed and the experience gain is identified for the terminal task feature chain, and a task migration projection network is obtained, including: Based on the cloud-edge-device coordinator, cloud-edge migration decisions are made on the terminal task feature chain to obtain an initial task migration map. The initial task migration graph is subjected to path optimization processing to obtain the optimized task migration graph; Perform simulated migration of the terminal task feature chain according to the task migration optimization graph to obtain a task migration simulation dataset; The task migration simulation dataset is analyzed for migration cost and identified for experience gain, and the task migration projection graph is generated.

6. The resource scheduling optimization method for intelligent converged terminals as described in claim 1, characterized in that, The multi-resource state sequence includes a computing resource state sequence, a storage resource state sequence, a communication resource state sequence, and an energy resource state sequence.

7. The resource scheduling optimization method for intelligent converged terminals as described in claim 1, characterized in that, Constructing a terminal task feature chain based on the current task set includes: Extract task attribute features from the current task set to obtain feature sets for each task; Based on the data dependencies, execution order, and resource competition relationships among the current task sets, the feature sets of each task are chained together and encoded to generate the terminal task feature chain.

8. A resource scheduling and optimization system for intelligent converged terminals, characterized in that, The system is used to implement the resource scheduling optimization method for intelligent converged terminals according to any one of claims 1-7, the system comprising: The terminal task feature chain construction module is used to obtain the multi-resource state sequence and current task set of the terminal device, and construct the terminal task feature chain based on the current task set; The terminal resource scheduling space acquisition module is used to perform value analysis and multi-resource coupling scheduling on the multi-resource state sequence based on the terminal task feature chain to acquire the terminal resource scheduling space. The service experience-oriented optimization module is used to perform service experience-oriented optimization on the terminal resource scheduling space according to the service experience evaluation architecture, and obtain the terminal's first scheduling strategy. The resource reallocation module is used to make task offloading decisions and resource reallocation based on the first scheduling strategy of the terminal by the cloud-edge-device coordinator, and to obtain the second scheduling strategy of the terminal. The uncertainty countermeasure compensation module is used to perform uncertainty modeling and uncertainty countermeasure compensation on the second scheduling strategy of the terminal, obtain the third scheduling strategy of the terminal, and perform multi-resource dynamic adjustment of the terminal device according to the third scheduling strategy of the terminal. The uncertainty mitigation compensation module includes: The uncertainty modeling unit is used to perform uncertainty modeling on the second scheduling strategy of the terminal to obtain a resource scheduling uncertainty model; The terminal scheduling regret value generation unit is used to generate a terminal scheduling regret value based on the resource scheduling uncertainty model, and to determine whether the terminal scheduling regret value is greater than or equal to a predetermined scheduling regret value. The resource redundancy compensation unit is used to perform resource redundancy compensation and task dependency propagation suppression processing on the second scheduling strategy of the terminal according to the resource scheduling uncertainty model if the terminal scheduling regret value is greater than or equal to the predetermined scheduling regret value, and generate the third scheduling strategy of the terminal. The resource redundancy compensation unit includes: The key uncertainty task determination subunit is used to determine the key uncertainty tasks of the terminal's second scheduling strategy according to the resource scheduling uncertainty model, as well as the resource fluctuation range and task dependency relationship corresponding to the key uncertainty tasks; The multi-dimensional resource redundancy compensation configuration subunit is used to perform multi-dimensional resource redundancy compensation configuration for the key uncertainty task according to the resource fluctuation range, and obtain the resource redundancy compensation configuration result. The downstream associated task identification subunit is used to identify downstream associated tasks affected by the key uncertainty task according to the task dependency relationship, and to construct the dependency propagation path from the key uncertainty task to the downstream associated task; The propagation suppression configuration subunit is used to configure propagation suppression for the dependent propagation path and obtain the propagation suppression configuration result, which includes delayed triggering configuration parameters, execution isolation configuration parameters, and resource reservation configuration parameters. The terminal third scheduling strategy generation subunit is used to update and optimize the terminal second scheduling strategy based on the resource redundancy compensation configuration result and the propagation suppression configuration result, and obtain the terminal third scheduling strategy.

Images