A task scheduling method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application disclose a task scheduling method and device, electronic equipment and a storage medium. Embodiments of the present application obtain a plurality of sub-tasks and a plurality of association relationships between the sub-tasks; generate a plurality of candidate task execution sequences corresponding to a task target based on the plurality of sub-tasks and the plurality of association relationships; determine a plurality of uncertainty matrices corresponding to the plurality of candidate task execution sequences based on the plurality of sub-tasks and the plurality of association relationships; generate a plurality of sequence risk information corresponding to the plurality of candidate task execution sequences based on the plurality of uncertainty matrices; determine a target entropy weight corresponding to a current time period based on an initial matrix corresponding to the current time period and a historical entropy weight corresponding to a previous time period; select a target task execution sequence from the plurality of candidate task execution sequences based on the plurality of uncertainty matrices, the plurality of sequence risk information and the target entropy weight; and execute the target task execution sequence to obtain a task execution result. The scheme can improve the success rate of task execution.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically to a task orchestration method, apparatus, electronic device, and storage medium. Background Technology

[0002] With the development of computer technology, task orchestration has been widely applied in fields such as data processing, distributed computing, and workflow management. Existing task orchestration methods typically rely on static configuration, employ heuristic scheduling, or use selection strategies based on a single confidence level.

[0003] However, the existing methods mentioned above have obvious technical defects in practical applications: over-reliance on static configuration makes task orchestration lack the ability to adapt to changes, heuristic scheduling strategies are prone to getting stuck in local optima and it is difficult to maximize resource utilization, and single-dimensional evaluation indicators cannot fully characterize the actual execution status of tasks, which will seriously affect the success rate of task execution after overall task orchestration. Summary of the Invention

[0004] This application provides a task orchestration method, apparatus, electronic device, and storage medium. It generates sequence risk information by determining the uncertainty matrix corresponding to multiple candidate task execution sequences, thereby improving the accuracy of risk assessment. Furthermore, through adaptive entropy weighting, it enhances the stability and adaptability of task orchestration, thereby increasing the overall task execution success rate.

[0005] In a first aspect, embodiments of this application provide a task orchestration method, including: Obtain multiple subtasks and the relationships between them, wherein the multiple subtasks are determined based on the task objective; Based on the multiple sub-tasks and multiple relationships, generate multiple candidate task execution sequences corresponding to the task objective; Based on multiple subtasks and multiple relationships, determine multiple first uncertainty matrices corresponding to multiple subtasks in multiple candidate task execution sequences; based on multiple subtasks and multiple relationships, determine multiple second uncertainty matrices corresponding to multiple relationships in multiple candidate task execution sequences; based on multiple first uncertainty matrices and multiple second uncertainty matrices, construct multiple uncertainty matrices corresponding to multiple candidate task execution sequences. Based on the multiple uncertainty matrices, multiple sequence risk information corresponding to the multiple candidate task execution sequences is generated; Based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period, the target entropy weight corresponding to the current time period is determined. The initial matrix is ​​determined based on the sequence risk information corresponding to the current time period. Based on the multiple uncertainty matrices, the multiple sequence risk information, and the target entropy weight, a target task execution sequence is selected from the multiple candidate task execution sequences; The target task execution sequence is executed to obtain the task execution result.

[0006] Accordingly, embodiments of this application provide a task orchestration apparatus, including: In some embodiments, the acquisition unit may be specifically used to acquire multiple subtasks and the association relationships between the multiple subtasks, wherein the multiple subtasks are determined based on the task objective; In some embodiments, the sequence generation unit can be specifically used to generate multiple candidate task execution sequences corresponding to the task target based on the multiple sub-tasks and multiple association relationships; In some embodiments, the matrix determination unit can be specifically used to determine multiple first uncertainty matrices corresponding to multiple subtasks in multiple candidate task execution sequences based on multiple subtasks and multiple relationships; determine multiple second uncertainty matrices corresponding to multiple relationships in multiple candidate task execution sequences based on multiple subtasks and multiple relationships; and construct multiple uncertainty matrices corresponding to multiple candidate task execution sequences based on multiple first uncertainty matrices and multiple second uncertainty matrices. In some embodiments, the information generation unit may be specifically used to generate multiple sequence risk information corresponding to the multiple candidate task execution sequences based on the multiple uncertainty matrices; In some embodiments, the entropy weight determination unit can be specifically used to determine the target entropy weight corresponding to the current time period based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period, wherein the initial matrix includes the uncertainty matrix corresponding to the current time period. In some embodiments, the filtering unit may be specifically used to filter a target task execution sequence from the plurality of candidate task execution sequences based on the plurality of uncertainty matrices, the plurality of sequence risk information and the target entropy weight; In some embodiments, the execution unit may be specifically used to execute the target task execution sequence to obtain task execution results.

[0007] Furthermore, embodiments of this application also provide an electronic device, including a processor and a memory, wherein the memory stores an application program, and the processor is used to run the application program in the memory to execute the task orchestration method provided in embodiments of this application.

[0008] Furthermore, embodiments of this application also provide a computer program product, including computer instructions, which, when executed by a processor, implement the steps in the task orchestration method provided in embodiments of this application.

[0009] Furthermore, embodiments of this application also provide a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to execute steps in any of the task orchestration methods provided in embodiments of this application.

[0010] This application embodiment obtains multiple subtasks and their interrelationships, where the subtasks are determined based on the task objective. Then, multiple candidate task execution sequences corresponding to the task objective are generated based on the multiple subtasks and their interrelationships. Next, multiple first uncertainty matrices are determined for each subtask in the candidate task execution sequences based on the multiple subtasks and their interrelationships; multiple second uncertainty matrices are also determined for each interrelationship in the candidate task execution sequences based on the multiple subtasks and their interrelationships; based on the multiple first and second uncertainty matrices, multiple uncertainty matrices corresponding to the candidate task execution sequences are constructed. Then, multiple sequence risk information corresponding to the candidate task execution sequences is generated based on the multiple uncertainty matrices. Next, the target entropy weight for the current time period is determined based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period, where the initial matrix is ​​determined based on the sequence risk information for the current time period. Finally, the target task execution sequence is selected from the multiple candidate task execution sequences based on the multiple uncertainty matrices, the multiple sequence risk information, and the target entropy weight. Finally, the target task execution sequence is executed to obtain the task execution result.

[0011] This scheme determines the uncertainty propagation mechanism in the candidate task execution sequence through multiple sub-tasks and relationships, thus deriving a more accurate uncertainty matrix. Furthermore, the determination of the uncertainty propagation mechanism improves the accuracy of the obtained sequence risk information, reducing the possibility of misjudgment. Additionally, by introducing adaptive entropy weights, the weights automatically adjust according to the task execution status, avoiding subjective bias caused by manually setting weights and enabling timely responses to sudden anomalies during task execution, thereby improving task execution stability. Therefore, task orchestration using this method can improve the overall success rate of task execution. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1This is a schematic diagram illustrating an application scenario of the task orchestration method provided in the embodiments of this application; Figure 2 This is a first flowchart illustrating the task orchestration method provided in an embodiment of this application; Figure 3 This is a second flowchart illustrating the task orchestration method provided in the embodiments of this application; Figure 4 This is a schematic diagram of the task diagram provided in the embodiments of this application; Figure 5 This is a schematic diagram of the task orchestration apparatus provided in the embodiments of this application; Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0014] The technical solutions described below, with reference to the accompanying drawings, will be clearly and completely described. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0015] The following explanations of some terms used in the embodiments of this application are provided to facilitate understanding by those skilled in the art.

[0016] This application provides a task orchestration method, apparatus, electronic device, and storage medium. The task orchestration apparatus can be integrated into an electronic device, which may be a server or a user terminal, etc.

[0017] The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud-preset databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery network (CDN), and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, smart TV, in-vehicle terminal, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein.

[0018] Figure 1 A schematic diagram illustrating an application scenario of the task orchestration method provided in this application embodiment is shown. For example... Figure 1 As shown, the task orchestration device is integrated into an electronic device, and the electronic device is a server, as an example.

[0019] The user's terminal (i.e., the client) can establish a communication connection with the server. Based on this, the user can send a task orchestration request to the server through their client. Upon receiving the task orchestration request and determining the task objective based on it, the server obtains multiple subtasks and their relationships. Then, based on these subtasks and relationships, it generates multiple candidate task execution sequences corresponding to the task objective. Next, it determines multiple uncertainty matrices corresponding to these candidate task execution sequences. Based on these uncertainty matrices, it generates multiple sequence risk information for each candidate task execution sequence. Then, based on the initial matrix corresponding to the current time period and the historical entropy weights of the previous time period, it determines the target entropy weight for the current time period. The initial matrix is ​​determined based on the sequence risk information for the current time period. Finally, based on the multiple uncertainty matrices, sequence risk information, and target entropy weights, it selects the target task execution sequence from the candidate sequences. The target task execution sequence is then executed, and the task execution result is returned to the client that made the task orchestration request.

[0020] It should be noted that, Figure 1 There can be multiple clients, and multiple clients can send task orchestration requests to the server at the same time or at different times.

[0021] The task orchestration method provided in this application can be applied to scenarios such as data pipeline processing, machine learning workflow scheduling, continuous integration / continuous deployment pipelines, and complex transaction processing. In these scenarios, by constructing a directed acyclic graph to describe the execution order and dependency constraints between tasks, the system can automatically identify parallel task groups without dependencies, thereby achieving parallel execution and failure recovery of tasks, significantly improving system throughput and resource utilization.

[0022] It should be noted that the embodiments of this application can be used in various scenarios, including but not limited to the above-mentioned data pipeline processing, machine learning workflow scheduling, continuous integration / continuous deployment pipelines, and complex transaction processing.

[0023] It is understood that in the specific implementation of this application, data related to task objectives, task execution specifications, etc. are involved. When the following embodiments of this application are applied to specific products or technologies, permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0024] The following sections provide detailed descriptions of each example. It should be noted that the order in which the embodiments are described is not intended to limit the preferred order of the embodiments.

[0025] This embodiment will be described from the perspective of a task orchestration device, which can be integrated into an electronic device, such as a server or a terminal. The terminal can include tablet computers, laptops, personal computers (PCs), wearable devices, virtual reality devices, or other smart devices that can generate image files.

[0026] A task orchestration method includes: acquiring multiple subtasks and their relationships, wherein the subtasks are determined based on a task objective; generating multiple candidate task execution sequences corresponding to the task objective based on the subtasks and relationships; determining multiple uncertainty matrices corresponding to the candidate task execution sequences based on the subtasks and relationships; generating sequence risk information for the candidate task execution sequences based on the uncertainty matrices; determining the target entropy weight for the current time period based on an initial matrix for the current time period and historical entropy weights for the previous time period, wherein the initial matrix is ​​determined based on the sequence risk information for the current time period; selecting a target task execution sequence from the candidate task execution sequences based on the uncertainty matrices, sequence risk information, and target entropy weights; and finally executing the target task execution sequence to obtain the task execution result.

[0027] Figure 2 A flowchart illustrating the task orchestration method provided in an embodiment of this application is shown. Figure 2 As shown, the specific process of this task orchestration method is as follows: 201. Electronic devices acquire multiple subtasks and the relationships between them.

[0028] Multiple subtasks are determined based on the task objective. For example, if the task objective is to troubleshoot smart device faults and complete fault location, repair, and acceptance testing to ensure the normal operation of the smart device, then the task objective can be broken down into multiple subtasks. For instance, the task objective can be divided into data acquisition subtasks, fault detection subtasks, fault location subtasks, repair subtasks, and acceptance subtasks. These subtasks have certain interrelationships. Specifically, the interrelationships include that the fault detection subtask can only be executed after the data acquisition subtask is completed, and the acceptance subtask can only be executed after the repair subtask is completed, and so on.

[0029] In some embodiments, a task graph can be used to visually represent multiple subtasks and the relationships between them. A task graph is a visualization tool used to show the multiple subtasks required to achieve a task objective and the relationships between them. In a task graph, nodes can be used to represent subtasks, and directed edges can be used to represent the relationships between them.

[0030] For example, refer to Figure 4 Medium task map, Figure 4 The directed acyclic task graph G=(V,E) includes 5 nodes v1-v5. Node v1 represents the data acquisition subtask, node v2 represents the fault detection subtask, node v3 represents the fault location subtask, node v4 represents the maintenance subtask, and node v5 represents the acceptance subtask. Figure 4 It can be seen that there is a directed edge between node v1 and node v2 pointing to node v2, indicating that the fault detection subtask can only be executed after the data acquisition subtask is completed.

[0031] In some embodiments, nodes can be used to determine information such as the task type, execution duration, required resources, and verification criteria of the subtasks they correspond to. Directed edges can be used to determine information such as execution order dependencies, constraints, and resource coupling relationships between connected nodes. Execution order dependencies include prerequisite dependencies and parallel dependencies. Prerequisite dependencies mean that a subsequent subtask can only execute after its predecessor has completed, while parallel dependencies mean that multiple subtasks can execute synchronously. Constraint relationships refer to the restrictions that nodes, as expressed by directed edges in the task graph, must adhere to. Resource coupling relationships refer to the constraint associations when multiple subtasks share the same type of resource.

[0032] In some embodiments, the task graph can be generated based on input information, which may include multiple or one of task objectives, constraint information, resource information, time limit information, compliance information, etc. The task objective is the final state or overdue result to be achieved, and may include information such as task completion deadlines and required accuracy. Constraint information may include hard constraints and soft constraints. Hard constraints refer to mandatory conditions that must be met during task execution; for example, hard constraints may include resource limits, compliance requirements, and security thresholds. Soft constraints refer to optimization objectives that are expected to be met but are allowed to deviate from during task execution; for example, soft constraints may include optimization objective priorities and resource conservation information. Resource information describes the resources required for task execution; for example, resource information may include information related to computing power resources, human resources, and equipment resources. Time limit information is information associated with the task execution period; for example, time limit information may include node deadlines and total project duration. Compliance information refers to the standards that must be followed to execute the task; for example, compliance information may include industry standards and security regulations.

[0033] Specifically, if the task objective is to troubleshoot intelligent devices and complete fault location, repair, and acceptance testing to ensure the normal operation of the intelligent devices, hard constraints include the conditions that the data acquisition equipment must meet, troubleshooting time limits, and repair / acceptance pass rates. Soft constraints include optimizing for shorter fault location time, reducing resource occupancy, and required resource information. Therefore, the task objective can be decomposed into data acquisition subtasks, fault detection subtasks, fault location subtasks, repair subtasks, and acceptance subtasks to determine five nodes v1-v5. Next, the relationships between the five nodes are determined based on the task flow and constraint information, and a reference diagram is drawn based on these relationships. Figure 4 The task map.

[0034] 202. Electronic devices generate multiple candidate task execution sequences corresponding to the task objective based on multiple sub-tasks and multiple relationships.

[0035] A task execution sequence is an executable plan generated based on task objectives and constraints. It includes at least information such as the execution order of subtasks, resource allocation, parallelism, timeout / retry strategy, verification strategy, and rollback strategy to ensure its executability and evaluability. One task objective can correspond to one or more task execution sequences. For example, refer to... Figure 4If the task objective is to troubleshoot intelligent devices and complete fault location, repair, and acceptance testing to ensure normal operation, with a hard constraint that the total troubleshooting time limit must not exceed 120 minutes, a candidate task execution sequence can be generated based on the serial execution logic, executing tasks in the order of node v1, node v2, node v3, node v4, and node v5. Furthermore, resources for the candidate task execution sequence can be centrally allocated, and the verification strategy is full verification for nodes v1, v3, and v5, sampling verification for nodes v2 and v4, with two retries for timeouts. Alternatively, based on the principle of prioritizing core nodes and reasonably compressing non-core nodes, the 120-minute total time limit can be allocated to each node to clarify node-level time limits, thereby ensuring that the total time limit for executing the candidate task sequence will not exceed 120 minutes. This rigid allocation of node time limits and timeout retry control ensures that the overall time limit is controllable.

[0036] In some embodiments, there can be multiple methods for generating candidate task execution sequences corresponding to a task objective based on multiple subtasks and multiple relationships. Therefore, different methods can generate various candidate task execution sequences. For example, candidate task execution sequences can be generated based on industry experience or rule templates, and the task execution sequences generated in this way can meet industry standards. Specifically, if the objective task is to troubleshoot intelligent device faults and complete fault location, fault repair, and fault acceptance to ensure the normal operation of the intelligent device, the execution flow of the fault troubleshooting task can be determined according to a preset set of structured configuration parameters: first, execute the data acquisition subtask; then, execute the fault detection subtask; then, execute the fault location subtask; then, execute the repair subtask; and finally, execute the acceptance subtask. After determining the execution order of the subtasks, the candidate task execution sequence can be obtained by adjusting the resource allocation details accordingly.

[0037] For example, to ensure that the generated candidate task execution sequences satisfy the constraint information, they can be generated based on the constraint information. The constraint information can include hard constraints, which must be met by the generated candidate task execution sequences. It can also include soft constraints; if a candidate task execution sequence does not meet the soft constraints, a penalty can be imposed. The higher the penalty value, the lower the priority of the candidate task sequence. This method ensures that all generated candidate task execution sequences meet the constraint information requirements.

[0038] For example, to ensure that the generated candidate task execution sequences meet diverse needs, heuristic rules can be applied to guide the sampling process, thereby generating multiple candidate task execution sequences with differentiation. Specifically, the execution order of parallel subtasks can be adjusted through different topological sorting to generate candidate task execution sequences. Alternatively, candidate task execution sequences can be generated by prioritizing the allocation of data acquisition devices to target subtasks or by performing balanced allocation through different resource allocations. Furthermore, candidate task execution sequences can be generated by performing full verification or sampling verification on multiple subtasks using different verification strategies.

[0039] In some embodiments, the method for generating candidate task execution sequences is not fixed or singular, but can be flexibly selected or multiple generation methods can be combined according to specific needs. These generation methods are not limited to the three methods described above, but can also include planner-based generation methods, large language model-based generation methods, optimization algorithm-based generation methods, and so on.

[0040] 203. The electronic device determines multiple first uncertainty matrices corresponding to multiple subtasks in multiple candidate task execution sequences based on multiple subtasks and multiple relationships; determines multiple second uncertainty matrices corresponding to multiple relationships in multiple candidate task execution sequences based on multiple subtasks and multiple relationships; and constructs multiple uncertainty matrices corresponding to multiple candidate task execution sequences based on multiple first uncertainty matrices and multiple second uncertainty matrices.

[0041] Uncertainty information, in the field of task orchestration, refers to information that cannot be accurately predicted before task execution but manifests after execution, affecting the task's state or outcome. To facilitate subsequent processing, this application uses an uncertainty matrix to represent uncertainty information. The uncertainty matrix transforms unknown fluctuations during task execution into quantifiable, optimizable, and decision-making information, thereby supporting more efficient and reliable task orchestration. This application utilizes a flexibly configurable estimator plugin to quantify uncertainty, allowing for flexible adaptation to different task scenarios and effectively addressing the problems of fuzzy and one-sided uncertainty quantification.

[0042] In some embodiments, the uncertainty matrix includes multiple dimensions, which correspond to multiple types of uncertainty factors. That is, each dimension can represent the quantitative information corresponding to an uncertainty factor.

[0043] In some embodiments, uncertainty factors may include model factors or knowledge factors. Uncertainty caused by model factors or knowledge factors refers to the degree to which the predictive model or industry knowledge cannot accurately determine the performance of the task or the stability of the correlation. This uncertainty matrix caused by model factors or knowledge factors can be represented as uepi .

[0044] Uncertainty factors can also include environmental factors. Uncertainty caused by environmental factors refers to the degree of deviation in task execution or instability in correlation caused by external environmental interferences, such as temperature fluctuations, noise, and network latency. This uncertainty matrix caused by environmental factors can be represented as u ale .

[0045] Uncertainty factors can also include constraints. Uncertainty caused by constraints refers to the degree to which hard and soft constraints in the candidate task execution sequence cannot be satisfied or conflict. This uncertainty matrix caused by constraints can be represented as u con .

[0046] Uncertainty factors can also include data factors. Uncertainty caused by data factors refers to the degree of defect in the data required for task execution, such as missing data, data drift, or insufficient quality. This uncertainty matrix caused by data factors can be represented as u data .

[0047] Uncertainty factors can also include resource factors. Uncertainty caused by resource factors refers to the degree of deviation in task execution caused by the inability to use resources allocated to subtasks on time or their performance failing to meet standards. This uncertainty matrix caused by data factors can be represented as u res .

[0048] Uncertainty factors can also include correlation factors. The uncertainty caused by correlation factors is the degree to which subsequent subtasks cannot start on time due to the delay or failure of the preceding subtask. This uncertainty matrix caused by correlation factors can be represented as u dep .

[0049] In some embodiments, one or more of the aforementioned uncertainty factors may be considered based on the actual application situation to obtain an uncertainty matrix that can more comprehensively and accurately reflect the uncertainty in the candidate task execution sequence, so as to guide the subsequent screening process of the target task execution sequence.

[0050] In some embodiments, since the task graph includes multiple nodes and multiple directed edges, and these nodes and edges exhibit a certain degree of randomness or variability during task execution, the uncertainty matrix corresponding to the candidate task execution sequence can be generated by calculating the node-level uncertainty matrix and the edge-level uncertainty matrix. Furthermore, since the calculation of the uncertainty matrix takes into account multiple types of uncertainty factors, the first uncertainty matrix corresponding to multiple subtasks can be represented as follows: u i(v)=[u1 epi u1 ale u1 con u1 data u1 res u1 dep ], where each element is represented by a quantized value, and the range of values ​​can be determined as [0,1]. The larger the quantized value, the stronger the uncertainty.

[0051] Specifically, the first uncertainty matrix u1 caused by model factors epi The degree of dispersion of the confidence level in the prediction model output can be determined by the following formula: , where p i The model predicts the probability distribution of node v's success, where n is the number of prediction result categories. For example, the prediction result can include success, partial success, and failure.

[0052] The first uncertainty matrix u1 caused by environmental factors ale The random fluctuation variance of data collected by environmental sensors can be calculated using the following formula: , where x j Let j be the value of the environmental parameter collected for the jth time. The average value collected during the same period in history is denoted as m, where m is the sample size. The results are then normalized to [0,1].

[0053] The first uncertainty matrix u1 caused by constraints con It can be calculated based on the probability of constraint satisfaction, and the calculation formula can be u1. con =1- , where a represents the number of constraints satisfied and b represents the total number of constraints.

[0054] The first uncertainty matrix u1 caused by data factors data The calculation can be performed by combining the missing data rate, noise variance, and data drift detection score. The calculation formula can be u1. data = ,in, These are the weighting coefficients. For data missing rate, For noise variance, The score is for drift detection.

[0055] The first uncertainty matrix u1 caused by related factors dep The success rate of the preceding node can be calculated using the formula u1. dep =1-p v’ , where p v’ This represents the success rate of the execution of the preceding node v'.

[0056] Similarly, since the calculation of the uncertainty matrix takes into account multiple factors, the second uncertainty matrix corresponding to a directed edge can be represented as follows: u i (e)=[u2 epi u2 ale u2 con u2 data u2 res u2 dep ], where each element is represented by a quantized value, and the range of values ​​can be determined as [0,1]. The larger the quantized value, the stronger the uncertainty.

[0057] Specifically, the second uncertainty matrix u2 caused by model factors epi This can be determined by the degree of dispersion in the confidence level of the prediction model output. The second uncertainty, u2, is due to environmental factors. ale The random fluctuation variance of data collected by environmental sensors can be calculated. The second uncertainty matrix u2 is due to constraints. con It can be calculated based on constraint conflict probabilities, such as resource coupling conflict probabilities. The second uncertainty matrix u2 is due to data factors. data It can be calculated by comprehensively considering data missing rate, noise variance, and data drift detection score. The second uncertainty matrix u2 is due to resource factors. res The probability of resource unavailability caused by resource coupling conflicts corresponding to directed edges can be calculated. The second uncertainty matrix u2 is due to the influence of related factors. dep The probability of a break in the association of directed edges can be calculated. For example, a break in the association caused by the failure of the preceding node can be calculated using the formula u2. dep =1-p v’ , where p v’ This represents the success rate of the execution of the preceding node v'.

[0058] In some embodiments, since the uncertainty matrix is ​​a comprehensive representation of node-level and edge-level uncertainties, after obtaining multiple first uncertainty matrices and multiple second uncertainty matrices, multiple uncertainty matrices corresponding to multiple candidate task execution sequences can be generated for application in subsequent global risk assessment steps. The calculation formula can be... ,in, These are the weighting coefficients. The total number of nodes. This represents the total number of directed edges.

[0059] 204. Electronic devices generate multiple sequence risk information corresponding to multiple candidate task execution sequences based on multiple uncertainty matrices.

[0060] In some embodiments, in order to select a more effective target task execution sequence from multiple candidate task execution sequences, it is necessary to obtain the sequence risk information corresponding to each candidate task execution sequence. This sequence risk information can reflect the risk status of the candidate task execution sequence. The sequence risk information can include multiple types of information, such as propagation risk information, cascade failure probability, information missing distribution, etc., which can show the risk level of different candidate task execution sequences from different dimensions.

[0061] In some embodiments, the electronic device can determine multiple preset propagation matrices and multiple first weights corresponding to multiple sub-tasks. Then, based on the multiple first uncertainty matrices, multiple preset propagation matrices, and multiple first weights corresponding to the multiple sub-tasks, multiple propagation matrices corresponding to the multiple sub-tasks are generated. Next, multiple second weights corresponding to multiple associations are determined, and then, based on the multiple second weights and multiple propagation matrices, multiple propagation risk information corresponding to multiple candidate task execution sequences is generated.

[0062] The predefined propagation matrix J is a transformation matrix that propagates the uncertainty matrix of a node or directed edge from the predecessor node to the successor node, or from the directed edge to the associated node. This predefined propagation matrix J can be represented as a k×k matrix, with matrix elements J... ij is the propagation weighting coefficient of the i-th type of uncertainty factor on the j-th type of uncertainty factor, used to control the propagation intensity of uncertainty caused by various factors.

[0063] After determining multiple preset propagation matrices, the first uncertainty matrix u corresponding to node v can be determined according to the topological order of the task graph. i (v) is multiplied by the preset propagation matrix J corresponding to node v to obtain the propagated matrix u corresponding to node v. i prop (v), and pass the propagated matrix to the successor node v' of node v. The propagated matrix u corresponding to the successor node v'. i prop (v') is based on the first uncertainty matrix u corresponding to node v'. i The weighted sum of the propagation matrices of (v') and all its predecessor nodes is called the first weight. The first weight can be determined based on the strength of the association relationship corresponding to the directed edge. The stronger the association relationship, the larger the first weight.

[0064] After determining multiple propagation matrices, multiple second weights corresponding to multiple associations can be determined. These second weights are determined by considering both the risk of the node itself and the risk arising from the associations between nodes; these second weights can be adjusted according to the specific task scenario. Then, based on the multiple second weights and multiple propagation matrices, a local risk matrix corresponding to each node and each directed edge can be calculated. Next, the multiple local risk matrices can be aggregated by maximizing or averaging the second weights to obtain multiple propagation risk information corresponding to multiple candidate task execution sequences. After determining multiple risk information, the candidate task execution sequence with the highest risk can be identified as the dominant risk sequence, and the propagation risk information corresponding to this dominant risk sequence is R. i path .

[0065] In some embodiments, the electronic device can determine multiple execution failure probabilities corresponding to multiple subtasks based on multiple first uncertainty matrices corresponding to multiple subtasks. Then, it determines multiple connection failure probabilities corresponding to multiple associations based on multiple second uncertainty matrices corresponding to multiple associations. Next, based on the multiple execution failure probabilities and the multiple connection failure probabilities, it generates multiple cascading failure probabilities corresponding to multiple candidate task execution sequences.

[0066] The cascading failure probability is the probability that the entire task execution sequence will fail due to node execution failure or the breakage of the association relationship corresponding to the directed edge in the candidate task execution sequence. Therefore, the cascading failure probability is related to the first uncertainty matrix corresponding to the node and the second uncertainty matrix corresponding to the directed edge.

[0067] After obtaining multiple first uncertainty matrices, their mean values ​​can be calculated using a monotonic mapping function to obtain the execution failure probability for each node. After obtaining multiple second uncertainty matrices, their mean values ​​can be calculated using a monotonic mapping function to obtain the connection failure probability for each directed edge. Then, the execution success probability can be obtained based on the execution failure probability, and the connection success probability can be obtained based on the connection failure probability. Next, the execution success probability for all nodes in the candidate task execution sequence and the connection success probability for all directed edges can be multiplied to obtain the task success probability for that candidate task execution sequence. Finally, the cascading failure probability is determined based on the obtained task success probabilities.

[0068] In some embodiments, the calculation of the cascade failure probability is not limited to the above scheme. The cascade failure probability can also be calculated by introducing weights that characterize the strength of the association or the importance weights of the nodes.

[0069] In some embodiments, the electronic device can determine multiple third weights corresponding to multiple dimensions in multiple first uncertainty matrices. Then, based on the multiple third weights and the multiple first uncertainty matrices, multiple information missing degrees corresponding to multiple sub-tasks are generated. Next, multiple information missing distributions corresponding to multiple candidate task execution sequences are constructed based on the multiple information missing degrees.

[0070] Information gaps are distributed across the nodes and edges of the task graph. Quantitative representations represent the set of numerical values ​​representing the portion of cognitive uncertainty that can be reduced through proactive probing or verification operations. This distribution is decoupled from data uncertainty, reflecting only the eliminateable uncertainty caused by incomplete knowledge. Its spatial distribution guides the optimal allocation of probing resources.

[0071] For example, since the first uncertainty matrix includes matrix elements resulting from multiple uncertainties, a corresponding third weight can be assigned to the elements corresponding to these uncertainties based on the actual situation. Specifically, if the uncertainties include correlation factors and data factors, a third weight can be determined to balance the impact of the two types of uncertainties on the information gap. Then, the information missing degree of each node is generated using the following formula:

[0072] The higher the value of the information missingness, the greater the information gap and the higher the information incompleteness corresponding to that node. Then, based on these information missingness values, multiple information missingness distributions can be constructed corresponding to multiple candidate task execution sequences.

[0073] In some embodiments, node uncertainty can also be precisely reduced through verification tasks. The electronic device can add a verification task to the candidate task execution sequence corresponding to the information missing distribution if the information missing distribution includes risky subtasks with an information missing degree exceeding a preset threshold. Then, the state information corresponding to the risky subtasks is obtained. Next, the information missing distribution is updated based on the state information until the information missing distribution no longer includes risky subtasks with an information missing degree exceeding the preset threshold.

[0074] For example, if the obtained information missing distribution contains risky subtasks with information missingness exceeding a preset threshold, then a verification task is needed to reduce the uncertainty of these risky subtasks. After adding a verification task, the current status information of the risky subtask can be obtained, and the information missing distribution can then be updated based on this status information. If the updated information missing distribution still contains risky subtasks with information missingness exceeding the preset threshold, then the above steps are repeated until the information missing distribution no longer contains risky subtasks with information missingness exceeding the preset threshold.

[0075] In some embodiments, the function for calculating the degree of missing information is not limited to the function corresponding to the above formula, but may also use monotonic functions such as product functions and weighted sum of squares functions.

[0076] 205. The electronic device determines the target entropy weight for the current time period based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period.

[0077] In some embodiments, the entropy weights can be adaptively assigned to enable the present application to handle sudden anomalies during task execution and adjust the plan in a timely manner according to the anomalies, thereby improving the adaptability and stability of the overall plan. Entropy weights are a method for calculating the weights of indicators using information entropy, used to measure the amount of information contained in different indicators in the evaluation system, thereby objectively determining the weight coefficients of each indicator.

[0078] For example, a sliding time window can be used to determine the current time period and multiple historical time periods. Sliding time window W t It is a time window used to collect historical uncertainty data, propagate result data, and calculate entropy weights. This sliding time window includes relevant data at time t and multiple consecutive historical times, and its size is determined in an adaptive manner.

[0079] Specifically, the task execution cycle can be determined based on the task graph, and the sampling interval can be determined in conjunction with the system's real-time requirements. For example, the sampling interval for high real-time tasks can be 1 minute, while the sampling interval for long-cycle industrial tasks can be 10 minutes, and so on. The number of task sampling points can be determined by the task execution cycle and sampling interval. Next, a scaling factor can be set according to the task's dynamic characteristics, and multiplied by the calculated number of task sampling points to obtain the size W of the sliding time window. o If the task changes rapidly, set a larger scaling factor so that the window can include more points, making it easier to capture changes. Additionally, upper and lower limits can be added to the sliding time window, setting W... o Limited to a preset range [W] min W max Next, the size of the sliding time window can be adjusted based on the previously determined information loss distribution to obtain a more suitable sliding time window.

[0080] In some embodiments, the initial matrix is ​​determined based on the sequence risk information corresponding to the current time period. Specifically, sequence risk information corresponding to the execution sequence of candidate tasks can be obtained. Then, an initial matrix U is constructed, where each row of the initial matrix U represents the indicator data at a moment within a sliding time window, and each column corresponds to the quantified value of the sequence risk information for a class of indicators. Next, the initial matrix U can be normalized to [0,1], mapping each element in the matrix to the same numerical range to eliminate differences in dimensions and scales. In addition, if the indicator has no fluctuations, all elements in that column are normalized to 0.5 to avoid errors.

[0081] In some embodiments, the electronic device can determine the initial matrix corresponding to the current time period based on the sequence risk information. Then, based on the information entropy of the initial matrix, the initial entropy weight corresponding to the initial matrix is ​​determined. Next, the historical entropy weight corresponding to the previous time period is obtained. Finally, based on the historical entropy weight and the initial entropy weight, the target entropy weight corresponding to the current time period is generated.

[0082] For example, after determining the initial matrix U, the entropy value E of each uncertainty index can be calculated using the information entropy formula. j (t), and through 1-E j (t) The difference coefficient is calculated, and then normalized again to obtain the initial entropy weight W corresponding to the current time period. j (t). Next, the volatility coefficient C of the composite risk value within the sliding time window can be used as a reference. v Determine the time decay coefficient The overall risk value is obtained based on the initial entropy weight and the initial matrix U, and the volatility coefficient C is... v The ratio of the standard deviation of the composite risk value to the mean of the composite risk value is determined by piecewise mapping. ,when hour =0.1; 0.1 hour ; hour =0.5. Then, the historical entropy weights corresponding to the previous time period are obtained through a sliding time window, and the initial entropy weights are updated with momentum using the exponential moving average formula, thereby achieving adaptive evolution of the weights over time and highlighting the influence of recent data on the weights. Then, all weights can be summed; if the sum is 1, the weight requirement is met, and the adaptive weight vector under this sliding window, which is the target entropy weight, is output for subsequent comprehensive credibility scoring.

[0083] 206. Electronic devices select target task execution sequences from multiple candidate task execution sequences based on multiple uncertainty matrices, multiple sequence risk information and target entropy weights.

[0084] For example, multiple candidate task execution sequences can be comprehensively evaluated based on the acquired uncertainty matrices, multiple sequence risk information, and target entropy weights. The target task execution sequence can then be selected from these candidate sequences based on the scoring results. This scoring method balances uncertainty, risk, and execution cost, resulting in a more scientific and optimal output. This approach can improve the success rate of execution while minimizing execution costs.

[0085] In some embodiments, the electronic device can generate global uncertainty information based on multiple uncertainty matrices and a target entropy weight. Then, it obtains multiple task orchestration cost information corresponding to multiple candidate task execution sequences. Next, based on the global uncertainty information, multiple sequence risk information, and multiple task orchestration cost information, it generates multiple sequence evaluation results corresponding to the multiple candidate task execution sequences. Finally, based on the multiple sequence evaluation results, it selects the target task execution sequence from the multiple candidate task execution sequences.

[0086] For example, the first uncertainty matrix corresponding to each node can be weighted and summed using the target entropy weight to obtain the comprehensive uncertainty for each node. Then, all nodes are traversed and calculated, and the uncertainty of all nodes is summed. This summation is then divided by the total number of nodes to obtain the global uncertainty information of the entire scheme, with a value range of [0,1]. This global uncertainty information is used to reflect the comprehensive uncertainty level of all nodes in the entire task orchestration scheme. The smaller the value, the lower the overall uncertainty and the stronger the stability of the scheme.

[0087] Then, the orchestration cost information for multiple candidate task execution sequences can be obtained. The task orchestration cost information needs to consider the comprehensive time cost, computing power cost, and resource consumption cost. The time cost is the ratio of the total execution time to the optimal execution time. The computing power cost is the ratio of the total computing power consumption to the computing power limit, which is obtained by summing the computing power consumption of each subtask. The resource consumption cost is obtained by averaging the sum of the resource consumption rates of each subtask. All types of costs are normalized to [0,1], and the weight coefficients sum to 1.

[0088] Next, the acquired sequence risk information can be normalized. Based on global uncertainty information, sequence risk information, and task orchestration cost information, the comprehensive risk and penalty coefficients for each risk are determined. A comprehensive weighted average is then used to calculate the comprehensive risk value R, which is then expressed by formula S. i=1-R is used to calculate the sequence evaluation result. The numerical range of the sequence evaluation result is [0,1]. The closer the value is to 1, the higher the credibility of the candidate task execution sequence. In addition, the penalty coefficients can be adjusted according to the scenario task configuration, and the sum is 1. Next, the candidate task execution sequence with the highest sequence evaluation result value can be determined as the target task execution sequence. If there are multiple candidate task execution sequences with the same score or the difference is less than the preset difference, the scheme with the lowest cascading failure probability is selected first; if the cascading failure probability is still the same, the scheme with the lowest task orchestration cost is selected first.

[0089] 207. The electronic device executes the target task execution sequence and obtains the task execution result.

[0090] In some embodiments, to improve task success rates, feedback data can be collected during task orchestration, and the underlying thinking or cognitive processes can be monitored and adjusted through a metacognitive controller. If the task execution sequence is detected as not guaranteeing smooth task execution, the task orchestration is re-executed to adaptively address various anomalies during execution, thereby ensuring the smooth progress of the task.

[0091] For example, feedback information generated during the execution of the target task sequence can be obtained. This feedback information can include one or more of the following: node execution information, edge association information, uncertainty dynamic information, resource status information, and cost consumption information. Node execution information includes node execution progress, execution success rate, and deviation between actual and planned execution time. Edge association information includes edge association latency and the number of associations. Uncertainty dynamic information includes real-time quantification values ​​of uncertainty indicators at the node or edge level and uncertainty propagation deviations. Resource status information includes actual resource availability, performance fluctuation values, and the number of resource conflicts. Cost consumption information includes actual time cost, computing power cost, and deviation between resource occupancy cost and planned cost. The frequency of feedback information collection can be determined based on the size of the sliding time window to ensure that the feedback information is synchronized with the time dimension of entropy weight updates and risk assessments. Then, outlier removal and normalization processing can be performed on the collected initial feedback information to obtain the feedback information set.

[0092] Then, the execution state deviation of the target task execution sequence can be monitored in real time through a metacognitive controller to identify potential anomalies. A deviation vector of monitoring indicators can be defined. ,in The real-time deviation of the k-th monitoring indicator is calculated from the difference between the actual feedback information and the predicted feedback information of the k-th indicator at time t. A monitoring deviation threshold can be preset. th When any thWhen this occurs, the indicator is marked as an "abnormal warning" and proceeds to the diagnostic stage; if all th Then the execution sequence of the target task will continue to be executed, and feedback information will be continuously collected.

[0093] If an indicator marked as "anomaly warning" is detected, it indicates that the target task execution sequence meets the warning conditions. Therefore, it is necessary to diagnose the root cause and level of the anomaly, and clarify the scope of its impact on task execution. Firstly, indicator correlation analysis can be used to locate the cause of the anomaly, and then, combined with the uncertainty propagation mechanism, the correlation between the anomaly indicator and uncertainty and risk indicators can be analyzed. Specifically, the Pearson correlation coefficient can be used to calculate the correlation, with a value range of […]. [1,1] The larger the absolute value, the higher the correlation. When the absolute value of the correlation exceeds 0.6, the uncertainty index can be determined as the cause of the anomaly.

[0094] Then, based on the uncertainty quantification value of the cause of the anomaly and the magnitude of the anomaly deviation, the causes of the anomaly can be divided into multiple anomaly levels to determine priority. Specifically, if the uncertainty quantification value of the cause of the anomaly is ≥0.6, and If the value is 0.3, the cause of this anomaly is defined as a Level 1 anomaly. This state may lead to cascading failure and requires immediate re-orchestration. If 0.4 ≤ the uncertainty quantification value of the cause of the anomaly < 0.6, and 0.15 ≤ If the uncertainty quantification value of the cause of the anomaly is <0.3, then the cause of this anomaly is defined as a level two anomaly. This state may affect task execution efficiency and requires parameter adjustment and continuous monitoring. If the uncertainty quantification value of the cause of the anomaly is <0.4, and If the value is less than 0.15, the cause of this exception is defined as a level 3 exception. This state does not affect the core process; only the feedback information needs to be recorded and updated.

[0095] After determining the cause of the anomaly and its corresponding level, anomaly analysis conclusions can be established, providing a basis for rearrangement or parameter adjustment. Specifically, if the cause of the anomaly corresponds to a Level 1 or Level 2 anomaly, the correlation between the anomaly and the parameters generated earlier can be analyzed to clarify the direction of parameter adjustment or the focus of rearrangement. Parameters can include propagation matrices, entropy weights, sliding time windows, penalty coefficients, etc. For example, if the anomaly is caused by resource factors and its level is Level 1, the anomaly analysis conclusion could include that the anomaly is due to statistical deviations in resource availability probability or an excessively low resource uncertainty weight in the entropy weight. If it is a Level 3 anomaly, the anomaly type and frequency can be summarized to analyze whether there are common patterns, such as a node repeatedly exhibiting slight deviations, thus providing data support for subsequent feedback updates. In this case, no parameter adjustment or rearrangement is required. The anomaly analysis conclusions are then output, including the root cause of the anomaly, the anomaly level, the core reason, and adjustment suggestions to ensure traceability and feasibility.

[0096] After obtaining the anomaly analysis conclusions, an adaptive strategy can be implemented based on the anomaly level to dynamically adjust the solution. For a Level 1 anomaly, a re-orchestration can be triggered immediately. Based on the collected feedback information, steps 102 to 106 are re-executed to replace the currently executing target task sequence, thereby mitigating the risk of cascading failure. For a Level 2 anomaly, a complete re-orchestration is not required; only parameters are adjusted based on the analysis conclusions, such as adjusting entropy weights, optimizing the propagation coefficients of propagation operators, and adjusting timeout / retry strategies. The anomaly is then continuously monitored to see if it is mitigated. For a Level 3 anomaly, no adjustments to the solution or parameters are needed; the anomaly data is simply included in the feedback update set for subsequent parameter iteration and optimization.

[0097] In addition to determining whether to re-orchestrate based on the anomaly level, trigger conditions can be set to initiate re-orchestrating when these conditions are met, ensuring the comprehensiveness of closed-loop control. For example, re-orchestrating can be triggered when the real-time calculated cascade failure probability exceeds a risk tolerance threshold. It can also be triggered when the deviation between the node-level or edge-level uncertainty matrix and the initial quantization value consistently exceeds a threshold. Furthermore, re-orchestrating can be triggered when the deviation between the actual task orchestration cost and the preset cost exceeds a threshold.

[0098] In some embodiments, triggering conditions, failure modes, effective orchestration strategies, and corresponding uncertainty distributions can be written into the database for retrieval and reuse or parameter updates in subsequent orchestration. Furthermore, after the target task execution sequence is completed, feedback information, anomalies, analysis conclusions, and re-orchestration results can also be written into the database to support continuous iterative optimization of the solution and ensure the adaptive capability of the entire task orchestration and uncertainty management system.

[0099] As described above, this embodiment of the application can obtain multiple subtasks and the relationships between them, where the multiple subtasks are determined based on the task objective. Then, multiple candidate task execution sequences corresponding to the task objective are generated based on the multiple subtasks and multiple relationships. Next, multiple first uncertainty matrices are determined based on the multiple subtasks and multiple relationships within the multiple candidate task execution sequences; multiple second uncertainty matrices are determined based on the multiple relationships within the multiple candidate task execution sequences; and multiple uncertainty matrices corresponding to the multiple candidate task execution sequences are constructed based on the multiple first uncertainty matrices and multiple second uncertainty matrices. Then, multiple sequence risk information corresponding to the multiple candidate task execution sequences is generated based on the multiple uncertainty matrices. Then, the target entropy weight for the current time period is determined based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period, where the initial matrix is ​​determined based on the sequence risk information corresponding to the current time period. Next, the target task execution sequence is selected from the multiple candidate task execution sequences based on the multiple uncertainty matrices, the multiple sequence risk information, and the target entropy weight. Finally, the target task execution sequence is executed to obtain the task execution result.

[0100] This application constructs a robust uncertainty propagation mechanism using a task graph and considers multiple uncertainty factors to obtain a more accurate uncertainty matrix. Furthermore, based on the uncertainty matrix, it considers the path risk, cascade failure probability, and information loss distribution of different task execution sequences, thus providing a clear basis for proactive detection and verification, significantly improving the targeting and accuracy of risk assessment, and avoiding misjudgments. In addition, this application employs a sliding time window technique to achieve adaptive entropy weighting. The time decay coefficient and window length can be dynamically adjusted according to the task execution status. Combined with metacognitive state machine-driven rearrangement, this enables real-time response to various sudden anomalies during execution, rapid adjustment of the solution, and improved adaptability and stability.

[0101] Based on the method described in the above embodiments, the following examples will provide further detailed explanations.

[0102] In this embodiment, the task orchestration device will be specifically integrated into an electronic device, which will be a server, as an example for explanation. Figure 3Another flowchart illustrating the task orchestration method provided in an embodiment of this application is shown. Figure 3 As shown, a task orchestration method is described below: 301. The server receives a task orchestration request and obtains input information based on the task orchestration request.

[0103] For example, after receiving a task orchestration request, the server can obtain input information based on the task orchestration request. The input information may include multiple or one of the following: task objectives, constraint information, resource information, time limit information, compliance information, etc.

[0104] 302. The server constructs a task graph based on the input information.

[0105] For example, a task graph includes multiple nodes and multiple edges. Multiple nodes represent multiple subtasks, which are obtained by breaking down the task objective. Multiple edges are used to connect multiple nodes to represent the dependencies between subtasks.

[0106] 303. The server generates multiple candidate solutions based on the task graph.

[0107] For example, multiple candidate solutions can be generated through rule templates, constraint solving, and heuristic / sampling methods.

[0108] 304. The server performs multi-source uncertainty quantification on multiple candidate schemes to obtain multiple uncertainty matrices.

[0109] For example, for any candidate solution, multi-source uncertainty index vectors are constructed at the node level and the edge level respectively, in order to quantify the impact of various uncertainties on task execution. These uncertainty indices can be generated by the estimator plugin.

[0110] 305. The server generates multiple path risk information corresponding to multiple candidate solutions based on multiple uncertainty matrices.

[0111] For example, an uncertainty propagation operator J can be defined on the task graph, and the multi-source uncertainty vector of a node can be propagated to its successor nodes based on the topological order of the task graph. Simultaneously, risk aggregation can be performed on the propagation results along the critical path, ultimately yielding multiple path risk information corresponding to multiple candidate solutions. Path risk information can include critical path risk, cascade failure probability, and information gap distribution.

[0112] 306. The server determines the target entropy weight corresponding to the current time period by adaptively assigning entropy weights.

[0113] For example, by using a sliding time window, the entropy weights of the multi-source uncertainty vector and the uncertainty propagation results can be adaptively assigned to obtain the dynamic target entropy weights of each uncertainty index, which can then be used for subsequent comprehensive credibility scoring.

[0114] 307. The server determines the evaluation results corresponding to multiple candidate solutions based on the target entropy weight, multiple uncertainty matrices, and multiple path risk information.

[0115] 308. Based on the solution evaluation results, the server selects the target solution from multiple candidate solutions.

[0116] 309. The server monitors, evaluates, and corrects the target solution, and continuously iterates and optimizes the generated target solution.

[0117] For example, multi-dimensional feedback information can be collected during the execution of the target plan, and then the deviation of the target plan's execution status can be monitored in real time through a metacognitive controller. If an anomaly warning is detected, the root cause and level of the anomaly are diagnosed, thereby clarifying the scope of the anomaly's impact on task execution. Next, analysis conclusions can be generated based on the anomaly level to determine the cause of the anomaly. Then, the plan can be dynamically adjusted based on the analysis conclusions to determine whether it needs to be rearranged, whether parameters need to be adjusted, etc.

[0118] 310. The server executes the generated parameters according to the scheme to build and update the database.

[0119] As described above, this embodiment of the application can receive task orchestration requests through a server and obtain input information based on the task orchestration requests. Then, a task graph is constructed based on the input information, and multiple candidate solutions are generated based on the task graph. Next, multi-source uncertainty quantification is performed on the multiple candidate solutions to obtain multiple uncertainty matrices. Then, based on the multiple uncertainty matrices, multiple path risk information corresponding to the multiple candidate solutions is generated. Then, through adaptive entropy weighting, the target entropy weight corresponding to the current time period is determined. Then, based on the target entropy weight, multiple uncertainty matrices, and multiple path risk information, the evaluation results corresponding to the multiple candidate solutions are determined. Next, based on the solution evaluation results, a target solution is selected from the multiple candidate solutions. Then, the target solution is monitored, evaluated, and corrected, and the generated target solution is continuously iterated and optimized. Finally, the database is constructed and updated based on the parameters generated by the solution execution.

[0120] This application constructs a robust uncertainty propagation mechanism using a task graph and considers multiple uncertainty factors to obtain a more accurate uncertainty matrix. Furthermore, based on the uncertainty matrix, it considers the path risk, cascade failure probability, and information loss distribution of different task execution sequences, thus providing a clear basis for proactive detection and verification, significantly improving the targeting and accuracy of risk assessment, and avoiding misjudgments. In addition, this application employs a sliding time window technique to achieve adaptive entropy weighting. The time decay coefficient and window length can be dynamically adjusted according to the task execution status. Combined with metacognitive state machine-driven rearrangement, this enables real-time response to various sudden anomalies during execution, rapid adjustment of the solution, and improved adaptability and stability.

[0121] To better implement the above methods, this application also provides a task orchestration device, which can be integrated into network devices, such as servers or terminals. The terminal may include tablet computers, laptop computers, and / or personal computers.

[0122] Figure 5 A schematic diagram of the task orchestration apparatus provided in an embodiment of this application is shown. Figure 5 As shown, the task orchestration device may include an acquisition unit 501, a sequence generation unit 502, a matrix determination unit 503, an information generation unit 504, a weight determination unit 505, a filtering unit 506, and an execution unit 507, as follows: (1) Obtain unit 501; The acquisition unit 501 is used to acquire multiple subtasks and the relationships between them, wherein the multiple subtasks are determined based on the task objectives.

[0123] (2) Sequence generation unit 502; The sequence generation unit 502 is used to generate multiple candidate task execution sequences corresponding to the task target based on multiple sub-tasks and multiple relationships.

[0124] (3) Matrix determination unit 503; The matrix determination unit 503 is used to determine multiple first uncertainty matrices corresponding to multiple subtasks in multiple candidate task execution sequences based on multiple subtasks and multiple relationships; determine multiple second uncertainty matrices corresponding to multiple relationships in multiple candidate task execution sequences based on multiple subtasks and multiple relationships; and construct multiple uncertainty matrices corresponding to multiple candidate task execution sequences based on multiple first uncertainty matrices and multiple second uncertainty matrices.

[0125] For example, an uncertainty matrix includes multiple dimensions, each corresponding to a different type of uncertainty factor.

[0126] (4) Information generation unit 504; The information generation unit 504 is used to generate multiple sequence risk information corresponding to multiple candidate task execution sequences based on multiple uncertainty matrices.

[0127] For example, the information generation unit 504 can be used to determine multiple preset propagation matrices and multiple first weights corresponding to multiple sub-tasks; generate multiple propagation matrices corresponding to multiple sub-tasks based on multiple first uncertainty matrices, multiple preset propagation matrices and multiple first weights corresponding to multiple sub-tasks; determine multiple second weights corresponding to multiple relationships; and generate multiple propagation risk information corresponding to multiple candidate task execution sequences based on multiple second weights and multiple propagation matrices.

[0128] For example, the information generation unit 504 can be used to determine multiple execution failure probabilities corresponding to multiple subtasks based on multiple first uncertainty matrices corresponding to multiple subtasks; determine multiple connection failure probabilities corresponding to multiple associations based on multiple second uncertainty matrices corresponding to multiple associations; and generate multiple cascading failure probabilities corresponding to multiple candidate task execution sequences based on multiple execution failure probabilities and multiple connection failure probabilities.

[0129] For example, the information generation unit 504 can be used to determine multiple third weights corresponding to multiple dimensions in the first uncertainty matrix corresponding to multiple sub-tasks; generate multiple information missing degrees corresponding to multiple sub-tasks based on the multiple third weights and multiple first uncertainty matrices; construct multiple information missing distributions corresponding to multiple candidate task execution sequences based on the multiple information missing degrees; in response to the information missing distribution including risky sub-tasks with information missing degrees exceeding a preset threshold, add a verification task to the candidate task execution sequence corresponding to the information missing distribution; obtain the state information corresponding to the risky sub-tasks; update the information missing distribution based on the state information until the information missing distribution no longer includes risky sub-tasks with information missing degrees exceeding a preset threshold.

[0130] (5) Weight determination unit 505; The weight determination unit 505 is used to determine the target entropy weight for the current time period based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period. The initial matrix is ​​determined based on the sequence risk information corresponding to the current time period.

[0131] For example, the weight determination unit 505 can be used to determine the initial matrix corresponding to the current time period based on the sequence risk information corresponding to the current time period; determine the initial entropy weight corresponding to the initial matrix based on the information entropy of the initial matrix; obtain the historical entropy weight corresponding to the previous time period; and generate the target entropy weight corresponding to the current time period based on the historical entropy weight and the initial entropy weight.

[0132] (6) Filtering unit 506; The filtering unit 506 is used to filter the target task execution sequence from multiple candidate task execution sequences based on multiple uncertainty matrices, multiple sequence risk information and target entropy weight.

[0133] For example, the screening unit 506 can be used to generate global uncertainty information based on multiple uncertainty matrices and target entropy weights; obtain multiple task orchestration cost information corresponding to multiple candidate task execution sequences; generate multiple sequence evaluation results corresponding to multiple candidate task execution sequences based on global uncertainty information, multiple sequence risk information and multiple task orchestration cost information; and screen the target task execution sequence from multiple candidate task execution sequences based on multiple sequence evaluation results.

[0134] (7) Execution unit 507; Execution unit 507 is used to execute the target task execution sequence and obtain the task execution result.

[0135] For example, execution unit 507 can be specifically used to obtain the set of feedback information corresponding to the execution sequence of the target task; determine the execution state deviation corresponding to the execution sequence of the target task based on the set of feedback information and the set of predicted feedback information; in response to the execution state deviation meeting the warning condition, determine the cause of the anomaly based on the uncertainty matrix and sequence risk information corresponding to the execution sequence of the target task; generate the anomaly analysis conclusion corresponding to the execution sequence of the target task based on the cause of the anomaly; in response to the anomaly analysis conclusion meeting the rearrangement condition, return to the step of generating multiple candidate task execution sequences corresponding to the task target based on multiple sub-tasks and multiple relationships, until the anomaly analysis conclusion corresponding to the updated execution sequence of the target task no longer meets the rearrangement condition.

[0136] In practice, each of the above units can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units, please refer to the previous method embodiments, which will not be repeated here.

[0137] As can be seen from the above, in this embodiment of the application, the acquisition unit 501 can acquire multiple subtasks and the relationships between them, where the multiple subtasks are determined based on the task objective. Then, the sequence generation unit 502 generates multiple candidate task execution sequences corresponding to the task objective based on the multiple subtasks and multiple relationships. Next, the matrix determination unit 503 determines multiple first uncertainty matrices corresponding to the multiple subtasks in the multiple candidate task execution sequences based on the multiple subtasks and multiple relationships; it also determines multiple second uncertainty matrices corresponding to the multiple relationships in the multiple candidate task execution sequences based on the multiple subtasks and multiple relationships; and constructs multiple uncertainty matrices corresponding to the multiple candidate task execution sequences based on the multiple first uncertainty matrices and multiple second uncertainty matrices. Then, the information generation unit 504 generates multiple sequence risk information corresponding to the multiple candidate task execution sequences based on the multiple uncertainty matrices. Then, the weight determination unit 505 determines the target entropy weight corresponding to the current time period based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period, where the initial matrix is ​​determined based on the sequence risk information corresponding to the current time period. Finally, the filtering unit 506 filters the target task execution sequence from the multiple candidate task execution sequences based on the multiple uncertainty matrices, the multiple sequence risk information, and the target entropy weight. The target task execution sequence is then executed by the execution unit 507 to obtain the task execution result.

[0138] This application constructs a robust uncertainty propagation mechanism using a task graph and considers multiple uncertainty factors to obtain a more accurate uncertainty matrix. Furthermore, based on the uncertainty matrix, it considers the path risk, cascade failure probability, and information loss distribution of different task execution sequences, thus providing a clear basis for proactive detection and verification, significantly improving the targeting and accuracy of risk assessment, and avoiding misjudgments. In addition, this application employs a sliding time window technique to achieve adaptive entropy weighting. The time decay coefficient and window length can be dynamically adjusted according to the task execution status. Combined with metacognitive state machine-driven rearrangement, this enables real-time response to various sudden anomalies during execution, rapid adjustment of the solution, and improved adaptability and stability.

[0139] This application also provides an electronic device, such as... Figure 6 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically: The electronic device may include components such as a processor 601 with one or more processing cores, a memory 602 with one or more computer-readable storage media, a power supply 603, and an input unit 604. Those skilled in the art will understand that... Figure 6The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein: The processor 601 is the control center of the electronic device, connecting various parts of the device via various interfaces and lines. It executes various functions and processes data by running or executing software programs and / or modules stored in the memory 602, and by calling data stored in the memory 602. Optionally, the processor 601 may include one or more processing cores; preferably, the processor 601 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 601.

[0140] The memory 602 can be used to store software programs and modules. The processor 601 executes various functional applications and data processing by running the software programs and modules stored in the memory 602. The memory 602 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 602 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 602 may also include a memory controller to provide the processor 601 with access to the memory 602.

[0141] The electronic device also includes a power supply 603 that supplies power to the various components. Preferably, the power supply 603 can be logically connected to the processor 601 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 603 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0142] The electronic device may also include an input unit 604, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0143] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 601 in the electronic device loads the executable files corresponding to the processes of one or more applications into the memory 602 according to the following instructions, and the processor 601 runs the applications stored in the memory 602 to realize various functions, as follows: The process involves: acquiring multiple subtasks and their interrelationships; generating multiple candidate task execution sequences corresponding to the task objective based on the subtasks and their interrelationships; determining multiple first uncertainty matrices corresponding to the subtasks in the candidate task execution sequences based on the subtasks and their interrelationships; determining multiple second uncertainty matrices corresponding to the interrelationships in the candidate task execution sequences based on the subtasks and their interrelationships; constructing multiple uncertainty matrices corresponding to the candidate task execution sequences based on the first and second uncertainty matrices; generating multiple sequence risk information corresponding to the candidate task execution sequences based on the multiple uncertainty matrices; determining the target entropy weight for the current time period based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period; selecting the target task execution sequence from the multiple candidate task execution sequences based on the multiple uncertainty matrices, the multiple sequence risk information, and the target entropy weight; and executing the target task execution sequence to obtain the task execution result.

[0144] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0145] As described above, this embodiment of the application can obtain multiple subtasks and the relationships between them, where the multiple subtasks are determined based on the task objective. Then, multiple candidate task execution sequences corresponding to the task objective are generated based on the multiple subtasks and multiple relationships. Next, multiple first uncertainty matrices are determined based on the multiple subtasks and multiple relationships within the multiple candidate task execution sequences; multiple second uncertainty matrices are determined based on the multiple relationships within the multiple candidate task execution sequences; and multiple uncertainty matrices corresponding to the multiple candidate task execution sequences are constructed based on the multiple first uncertainty matrices and multiple second uncertainty matrices. Then, multiple sequence risk information corresponding to the multiple candidate task execution sequences is generated based on the multiple uncertainty matrices. Then, the target entropy weight for the current time period is determined based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period, where the initial matrix is ​​determined based on the sequence risk information corresponding to the current time period. Next, the target task execution sequence is selected from the multiple candidate task execution sequences based on the multiple uncertainty matrices, the multiple sequence risk information, and the target entropy weight. Finally, the target task execution sequence is executed to obtain the task execution result.

[0146] This application constructs a robust uncertainty propagation mechanism using a task graph and considers multiple uncertainty factors to obtain a more accurate uncertainty matrix. Furthermore, based on the uncertainty matrix, it considers the path risk, cascade failure probability, and information loss distribution of different task execution sequences, thus providing a clear basis for proactive detection and verification, significantly improving the targeting and accuracy of risk assessment, and avoiding misjudgments. In addition, this application employs a sliding time window technique to achieve adaptive entropy weighting. The time decay coefficient and window length can be dynamically adjusted according to the task execution status. Combined with metacognitive state machine-driven rearrangement, this enables real-time response to various sudden anomalies during execution, rapid adjustment of the solution, and improved adaptability and stability.

[0147] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0148] Therefore, embodiments of this application provide a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute steps in any of the task orchestration methods provided in embodiments of this application. For example, the instructions can execute the following steps: The process involves: acquiring multiple subtasks and their interrelationships; generating multiple candidate task execution sequences corresponding to the task objective based on the subtasks and their interrelationships; determining multiple first uncertainty matrices corresponding to the subtasks in the candidate task execution sequences based on the subtasks and their interrelationships; determining multiple second uncertainty matrices corresponding to the interrelationships in the candidate task execution sequences based on the subtasks and their interrelationships; constructing multiple uncertainty matrices corresponding to the candidate task execution sequences based on the first and second uncertainty matrices; generating multiple sequence risk information corresponding to the candidate task execution sequences based on the multiple uncertainty matrices; determining the target entropy weight for the current time period based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period; selecting the target task execution sequence from the multiple candidate task execution sequences based on the multiple uncertainty matrices, the multiple sequence risk information, and the target entropy weight; and executing the target task execution sequence to obtain the task execution result.

[0149] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0150] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0151] Since the instructions stored in the computer-readable storage medium can execute the steps in any of the task orchestration methods provided in the embodiments of this application, the beneficial effects that any of the task orchestration methods provided in the embodiments of this application can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0152] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the methods provided in various alternative implementations of the task orchestration method described above.

[0153] The above provides a detailed description of a task orchestration method, apparatus, electronic device, and storage medium provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A task orchestration method, characterized in that, include: Obtain multiple subtasks and the relationships between them, wherein the multiple subtasks are determined based on the task objective; Based on the multiple sub-tasks and multiple relationships, generate multiple candidate task execution sequences corresponding to the task objective; Based on the multiple subtasks and the multiple relationships, determine multiple first uncertainty matrices corresponding to multiple subtasks in the multiple candidate task execution sequences; Based on the multiple sub-tasks and the multiple relationships, determine multiple second uncertainty matrices corresponding to multiple relationships in the multiple candidate task execution sequences; Based on the plurality of first uncertainty matrices and the plurality of second uncertainty matrices, construct a plurality of uncertainty matrices corresponding to the plurality of candidate task execution sequences; Based on the multiple uncertainty matrices, multiple sequence risk information corresponding to the multiple candidate task execution sequences is generated; Based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period, the target entropy weight corresponding to the current time period is determined. The initial matrix is ​​determined based on the sequence risk information corresponding to the current time period. Based on the multiple uncertainty matrices, the multiple sequence risk information, and the target entropy weight, a target task execution sequence is selected from the multiple candidate task execution sequences; The target task execution sequence is executed to obtain the task execution result.

2. The method according to claim 1, characterized in that, The uncertainty matrix includes multiple dimensions, which correspond to multiple types of uncertainty factors.

3. The method according to claim 1, characterized in that, The sequence risk information includes propagation risk information. The step of generating multiple sequence risk information corresponding to the multiple candidate task execution sequences based on the multiple uncertainty matrices includes: Determine multiple preset propagation matrices and multiple first weights corresponding to the multiple sub-tasks; Based on the multiple first uncertainty matrices corresponding to the multiple sub-tasks, the multiple preset propagation matrices, and the multiple first weights, multiple propagated matrices corresponding to the multiple sub-tasks are generated. Determine multiple second weights corresponding to the multiple association relationships; Based on the multiple second weights and the multiple propagation matrices, multiple propagation risk information corresponding to the multiple candidate task execution sequences is generated.

4. The method according to claim 1, characterized in that, The sequence risk information includes the cascade failure probability. The generation of multiple sequence risk information corresponding to the multiple candidate task execution sequences based on the multiple uncertainty matrices includes: Based on the multiple first uncertainty matrices corresponding to the multiple sub-tasks, determine the multiple execution failure probabilities corresponding to the multiple sub-tasks; Based on the multiple second uncertainty matrices corresponding to the multiple association relationships, determine the multiple connection failure probabilities corresponding to the multiple association relationships; Based on the multiple execution failure probabilities and the multiple connection failure probabilities, multiple cascading failure probabilities are generated corresponding to the multiple candidate task execution sequences.

5. The method according to claim 1, characterized in that, The sequence risk information includes an information missing distribution. The generation of multiple sequence risk information corresponding to the multiple candidate task execution sequences based on the multiple uncertainty matrices includes: Determine multiple third weights corresponding to multiple dimensions in the first uncertainty matrix corresponding to the multiple sub-tasks; Based on the multiple third weights and multiple first uncertainty matrices, multiple information missing degrees are generated for multiple sub-tasks; Based on the multiple information missing degrees, construct multiple information missing distributions corresponding to the multiple candidate task execution sequences.

6. The method according to claim 5, characterized in that, The method further includes: In response to the fact that the information missing distribution includes risky subtasks with information missing degree exceeding a preset threshold, a verification task is added to the candidate task execution sequence corresponding to the information missing distribution; Obtain the status information corresponding to the risk subtask; The information missing distribution is updated based on the state information until the information missing distribution does not include risky subtasks with an information missing degree exceeding a preset threshold.

7. The method according to claim 1, characterized in that, The step of determining the target entropy weight for the current time period based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period includes: Based on the sequence risk information corresponding to the current time period, determine the initial matrix corresponding to the current time period; Based on the information entropy of the initial matrix, determine the initial entropy weight corresponding to the initial matrix; Obtain the historical entropy weight corresponding to the previous time period; Based on the historical entropy weight and the initial entropy weight, the target entropy weight corresponding to the current time period is generated.

8. The method according to claim 1, characterized in that, The step of filtering the target task execution sequence from the multiple candidate task execution sequences based on the multiple uncertainty matrices, the multiple sequence risk information, and the target entropy weight includes: Global uncertainty information is generated based on the multiple uncertainty matrices and the target entropy weight; Obtain the task orchestration cost information corresponding to the multiple candidate task execution sequences; Based on the global uncertainty information, the multiple sequence risk information, and the multiple task orchestration cost information, multiple sequence evaluation results corresponding to the multiple candidate task execution sequences are generated. Based on the evaluation results of the multiple sequences, the target task execution sequence is selected from the multiple candidate task execution sequences.

9. The method according to claim 1, characterized in that, The method further includes: Obtain the set of feedback information corresponding to the execution sequence of the target task; Based on the feedback information set and the predicted feedback information set, the execution state deviation corresponding to the target task execution sequence is determined; In response to the execution state deviation meeting the warning conditions, the cause of the anomaly is determined based on the uncertainty matrix and sequence risk information corresponding to the target task execution sequence; Based on the cause of the anomaly, generate an anomaly analysis conclusion corresponding to the execution sequence of the target task; In response to the anomaly analysis conclusion satisfying the rearrangement condition, the process returns to the step of generating multiple candidate task execution sequences corresponding to the task target based on the multiple sub-tasks and multiple relationships, until the anomaly analysis conclusion corresponding to the updated target task execution sequence no longer satisfies the rearrangement condition.

10. A task orchestration device, characterized in that, include: An acquisition unit is used to acquire multiple subtasks and the relationships between the multiple subtasks, wherein the multiple subtasks are determined based on the task objective; A sequence generation unit is used to generate multiple candidate task execution sequences corresponding to the task target based on the multiple sub-tasks and multiple relationships; A matrix determination unit is used to determine multiple first uncertainty matrices corresponding to multiple subtasks in the multiple candidate task execution sequences based on the multiple subtasks and the multiple relationships; Based on the multiple sub-tasks and the multiple relationships, determine multiple second uncertainty matrices corresponding to multiple relationships in the multiple candidate task execution sequences; Based on the plurality of first uncertainty matrices and the plurality of second uncertainty matrices, construct a plurality of uncertainty matrices corresponding to the plurality of candidate task execution sequences; An information generation unit is used to generate multiple sequence risk information corresponding to the multiple candidate task execution sequences based on the multiple uncertainty matrices; The weight determination unit is used to determine the target entropy weight corresponding to the current time period based on the initial matrix corresponding to the current time period and the historical entropy weight corresponding to the previous time period. The initial matrix is ​​determined based on the sequence risk information corresponding to the current time period. A filtering unit is configured to filter a target task execution sequence from the plurality of candidate task execution sequences based on the plurality of uncertainty matrices, the plurality of sequence risk information and the target entropy weight; An execution unit is used to execute the target task execution sequence and obtain the task execution result.

11. An electronic device, characterized in that, It includes a processor and a memory, the memory storing an application program, and the processor running the application program in the memory to perform the steps of the task orchestration method according to any one of claims 1 to 9.

12. A computer program product comprising computer instructions, characterized in that, When executed by a processor, the computer instructions implement the steps of the task orchestration method according to any one of claims 1 to 9.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to perform the steps of the task orchestration method according to any one of claims 1 to 9.

Images