Task dynamic adjustment system and method based on intelligent decision model

Abstract

The present application provides a task dynamic adjustment system and method based on an intelligent decision model, which relates to the field of intelligent task management. The system first verifies and parses the received task request, extracts task information such as task description, expression form, priority, deadline and task amount, and stores it in a task management database. Then, after extracting the target task from the task management database, the target task is allocated to multiple resource nodes according to the task information and system load status. During the task execution process, the resource utilization of multiple resource nodes and the remaining execution information of the target task are monitored in real time, and these data are input into the intelligent decision model. Through model analysis, the resource allocation optimization and adjustment of the remaining execution part of the target task are performed. In this way, the task can be efficiently executed under the condition of meeting its needs, and the system resources can be fully utilized.

Description

Technical Field

[0001] The present application relates to the field of intelligent task management, and more specifically, to a task dynamic adjustment system and method based on an intelligent decision-making model. Background Art

[0002] With the rapid development of information technology, distributed computing systems and cloud computing platforms are increasingly being used in various industries. These systems usually need to handle a large number of concurrent tasks and require efficient management and allocation of limited computing resources (such as CPU, memory, storage, and network bandwidth) to ensure that all tasks can be completed on time and maximize resource utilization.

[0003] However, in actual operation, due to the uncertainty and dynamic changes of task requirements, as well as the fluctuations in the status of resource nodes, traditional static or semi-static task adjustment methods often fail to meet the needs of modern complex environments. Specifically, once the initial scheduling plan is made, it is difficult to adjust it according to the actual situation during task execution. This means that if a new high-priority task appears or the requirements of an existing task change (such as early completion, delay, or failure), the system may not be able to respond in time, making it difficult to flexibly respond to emergencies. In addition, due to the inability to adjust in real time, some resource nodes may be idle for a long time, while other nodes may be overloaded, resulting in low overall resource utilization.

[0004] Therefore, a dynamic task adjustment scheme based on an intelligent decision-making model is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a task dynamic adjustment system and method based on an intelligent decision model.

[0006] According to one aspect of the present application, a task dynamic adjustment system based on an intelligent decision model is provided, which includes: A task request processing module, used to verify and parse the received task request to obtain task information, wherein the task information includes task description, expression form, priority, deadline and task amount; A task request storage module, used for storing the task information in a task management database; A target task allocation module, used for extracting target tasks from the task management database, and allocating the target tasks to multiple resource nodes according to the task information of the target tasks and the system load status; A resource utilization task information monitoring module is used to monitor the resource utilization of the multiple resource nodes and the remaining execution task information of the target task in real time during the execution of the target task; The resource allocation adjustment module is used to input the resource utilization rates of the multiple resource nodes and the remaining execution task information of the target task into the intelligent decision model to adjust the resource allocation of the remaining execution tasks in the target task.

[0007] According to another aspect of the present application, a method for dynamic adjustment of tasks based on an intelligent decision model is provided, which includes: Verify and parse the received task request to obtain task information, wherein the task information includes task description, expression form, priority, deadline and task amount; Storing the task information in a task management database; Extracting a target task from the task management database, and allocating the target task to a plurality of resource nodes according to task information of the target task and a system load status; During the execution of the target task, real-time monitoring of resource utilization of the plurality of resource nodes and remaining execution task information of the target task; The resource utilization rates of the multiple resource nodes and the remaining execution task information of the target task are input into the intelligent decision model to adjust the resource allocation of the remaining execution tasks in the target task.

[0008] Compared with the prior art, the task dynamic adjustment system and method based on the intelligent decision model provided by the present application first verifies and parses the received task request, extracts task information such as task description, expression form, priority, deadline and task amount, and stores it in the task management database, then extracts the target task from the task management database, and distributes the target task to multiple resource nodes according to the task information and system load status. During the task execution process, the resource utilization rate of multiple resource nodes and the remaining execution information of the target task are monitored in real time, and these data are input into the intelligent decision model, and the resource allocation optimization and adjustment of the remaining execution part of the target task are performed through model analysis. In this way, the task can be efficiently executed under the condition of meeting its needs, and the system resources can be fully utilized. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings: Figure 1 A system block diagram of a task dynamic adjustment system based on an intelligent decision-making model according to an embodiment of the present application.

[0010] Figure 2 A block diagram of a resource allocation adjustment module in a task dynamic adjustment system based on an intelligent decision-making model according to an embodiment of the present application.

[0011] Figure 3 A block diagram of a resource utilization task information fusion encoding unit in a task dynamic adjustment system based on an intelligent decision-making model according to an embodiment of the present application.

[0012] Figure 4 The present invention is a flowchart of a method for dynamically adjusting tasks based on an intelligent decision-making model according to an embodiment of the present application. DETAILED DESCRIPTION

[0013] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0014] Based on the problems in the above background technology, this application proposes a task dynamic adjustment system based on an intelligent decision-making model. Figure 1 FIG. 1 is a system block diagram of a task dynamic adjustment system based on an intelligent decision model according to an embodiment of the present application. Figure 1 As shown, in the task dynamic adjustment system 100 based on the intelligent decision-making model, it includes: a task request processing module 110, which is used to verify and parse the received task request to obtain task information, wherein the task information includes task description, expression form, priority, deadline and task amount; a task request storage module 120, which is used to store the task information in the task management database; a target task allocation module 130, which is used to extract the target task from the task management database, and allocate the target task to multiple resource nodes according to the task information of the target task and the system load status; a resource utilization task information monitoring module 140, which is used to monitor the resource utilization of the multiple resource nodes and the remaining execution task information of the target task in real time during the execution of the target task; a resource allocation adjustment module 150, which is used to input the resource utilization of the multiple resource nodes and the remaining execution task information of the target task into the intelligent decision-making model to adjust the resource allocation of the remaining execution tasks in the target task.

[0015] In an embodiment of the present application, the task request processing module 110 is used to verify and parse the received task request to obtain task information, wherein the task information includes task description, expression form, priority, deadline and task amount. Specifically, in an embodiment of the present application, the task request processing module is used to: verify the format, parameters and permission information of the received task request to obtain a verified task request; extract task request information from the verified task request, and parse the task request information to obtain the task information, wherein the task request information includes task name, parameters, priority requirements and resource requirements.

[0016] It should be understood that in distributed computing systems and cloud computing platforms, the sources of tasks are extensive and complex. In order to ensure the stability and efficient operation of the system and data security, the format, parameters and permission information of the received task request must be verified first. Specifically, the system has specific requirements for the format of the task request, such as a specific field order, data type, etc. By verifying the format of the received task request, it can be ensured that only requests that meet the format requirements can be correctly identified and processed by the system, avoiding the system's inability to understand the content of the task request due to format errors, thereby ensuring the legitimacy of the task request. The task parameters are verified to ensure that the task request contains the necessary information. The execution of the task depends on some specific parameters. If the parameters are missing or incorrect, the task may not be executed correctly. By verifying the parameters, the integrity of the task request can be ensured, so that the system can obtain all the information required to execute the task. The verification of permission information is to ensure that only users or system components with corresponding permissions can submit task requests, which can prevent illegal or malicious requests from invading the system and protect the security of system resources and data. For example, some sensitive tasks may only be initiated by specific users or roles. If permission verification is not performed, data leakage or malicious use of the system may occur. In general, by verifying the format, parameters, and permission information of task requests, the system can eliminate failures and errors caused by incorrect or illegal requests. Only task requests that pass the verification will be further processed, which can reduce the burden of the system to handle invalid or erroneous requests, thereby improving the reliability and stability of the system.

[0017] Accordingly, it is considered that the task name is the key identifier to distinguish different tasks. In a complex distributed computing system or cloud computing platform, multiple types of tasks are processed simultaneously, such as data processing tasks, image recognition tasks, order processing tasks, etc. By extracting the task name in the task request, the system can quickly identify the task type and then allocate the corresponding processing flow and resources for it. For example, for image recognition tasks, the system can call a special image recognition algorithm library and related computing resources; for data processing tasks, it can allocate computing nodes and storage resources suitable for data processing. The parameter information in the task request provides the specific details of the task execution. Different tasks may have different parameter requirements, which directly affect the execution mode and resource requirements of the task. By extracting the parameter information in the task request, it helps the system prepare a suitable computing environment for the task and ensure that the task can be executed correctly. The priority requirement in the task request determines the order of the task in the execution queue. High-priority tasks usually need to be processed first to meet business-critical requirements. By extracting the priority requirement information, the system can understand the urgency of the task and process it in time to ensure that the critical task is completed in time. Resource requirement information is crucial for the system to allocate resources reasonably. Tasks may require a specific number of CPU cores, memory, storage capacity, network bandwidth and other resources. By extracting the resource requirements in the task request, the system can evaluate the current resource availability during the task allocation phase and avoid allocating tasks to nodes that cannot meet their resource requirements, thereby improving resource utilization and preventing task delays caused by resource competition. For example, a task that requires a large amount of memory should not be assigned to a node with tight memory resources. In other words, by extracting the task request information from the verified task request, different tasks can be identified and classified, which helps the system to plan resources and determine priorities.

[0018] Next, the task request information including the task name, parameters, priority requirements and resource requirements needs to be parsed to obtain the task information including the task description, expression form, priority, deadline and task amount. It should be understood that the task description usually contains a detailed explanation of the specific business logic or goal of the task. Obtaining the task description can help the system to deeply understand the nature of the task, so as to better plan the task execution steps and resource allocation strategies. The expression form of the task determines how the system handles the task request. Different expressions may require different parsing methods and processing procedures. Obtaining the expression form of the task enables the system to correctly interpret the task information and convert it into an internally processable format, thereby laying the foundation for subsequent task scheduling and execution. Priority plays a key role in task scheduling. As mentioned above, it determines the execution order of tasks. The accurate priority data obtained by parsing the task request information enables the system to adjust the execution order of tasks in a timely manner according to the resource status and the priority of other tasks during the dynamic adjustment of tasks. When new high-priority tasks appear or resources are tight, the system can suspend or postpone low-priority tasks and give priority to the execution of high-priority tasks to ensure that critical tasks are completed on time. The deadline sets a clear time limit for the task. The system can calculate the remaining time of the task according to the deadline, and reasonably arrange the execution progress of the task by combining factors such as resource utilization and task volume. During the execution of the task, the progress of the task will be monitored in real time. When it is found that the task may not be completed on time, the system can try to meet the deadline requirements by dynamically adjusting resource allocation (such as increasing computing resources or adjusting task allocation strategies). The task volume is an important basis for evaluating the resource requirements and execution time of the task. By obtaining the task volume information, the system can preliminarily estimate the computing resources and execution time required for the task according to the size of the task volume. In the task allocation stage, the system can match the task volume with the processing capacity of the resource node and select the appropriate resource node for task allocation. During the task execution process, as the task volume gradually decreases, the system can dynamically adjust the resource allocation according to the remaining task volume, allocate the released resources to other tasks waiting to be executed, and improve resource utilization. For example, in large-scale data processing tasks, the task volume can be measured by the size of the data file or the number of records. The system can reasonably allocate memory, disk I / O and CPU resources according to the task volume, thereby ensuring efficient execution of the task.

[0019] In an embodiment of the present application, the task request storage module 120 is used to store the task information in a task management database. It should be understood that by storing the verified and parsed task information in a task management database, the system can better manage the task information data. For example, during task allocation and dynamic adjustment, the system can quickly obtain task-related information from the task management database, which reduces the time for information search and processing, and is conducive to making decisions more quickly, thereby helping to improve the overall operating efficiency of the system. Moreover, the task management database provides a stable storage environment for task information, which can avoid task execution errors or system failures caused by data loss or confusion. At the same time, the transaction processing mechanism of the database can ensure the consistency and integrity of the data, which is conducive to further enhancing the stability of the system.

[0020] In an embodiment of the present application, the target task allocation module 130 is used to extract the target task from the task management database, and allocate the target task to multiple resource nodes according to the task information of the target task and the system load status. It should be understood that the system load status refers to the usage and busyness of each resource node in a distributed computing system or cloud computing platform at a specific time. Different tasks have different characteristics and requirements. For example, the task description determines the nature of the task, the task amount affects the amount of resources and execution time required for the task, and the priority and deadline specify the urgency and time limit of the task. For example, a real-time video processing task (as indicated by the task description) may require higher CPU and memory resources (determined by the task amount and the nature of the task), and has a higher priority and a strict deadline due to real-time requirements; while a background data backup task has a relatively lower priority and can be processed when the system load is lighter. Therefore, assigning tasks based only on task information cannot fully consider the actual availability of system resources, and needs to be combined with the system load status. Considering that system resources are limited and the load status is changing all the time. At a certain moment, some resource nodes may be idle, while others may be close to full load. If the task is assigned without considering the system load status, the task may be assigned to an overloaded resource node, resulting in slow task execution or even failure; at the same time, idle resource nodes are not fully utilized, resulting in resource waste. By comprehensively considering the task information of the target task and the system load status, the task can be assigned to the most appropriate resource node, so that the system resources can be fully utilized and the target task can be carried out efficiently.

[0021] In an embodiment of the present application, the resource utilization task information monitoring module 140 is used to monitor the resource utilization of the multiple resource nodes and the remaining execution task information of the target task in real time during the execution of the target task. It should be understood that the resource utilization of a resource node refers to the proportional relationship between the degree to which various resources (such as CPU, memory, storage and network bandwidth, etc.) in the resource node are used and the total amount of resources of the resource node at a specific moment. For example, CPU utilization indicates the percentage of the time that the CPU is busy (executing task instructions) per unit time to the total time; memory utilization is the ratio of the amount of used memory to the total memory capacity. This utilization index intuitively reflects the busyness of the resource node and the availability of the remaining resources. In the system, considering that the task execution environment is dynamically changing, the arrival and end of tasks, fluctuations in resource demand, etc. may occur at any time. By monitoring resource utilization in real time, the real-time status of the resource node can be timely understood so as to make dynamic adjustments according to the usage of resources. And by monitoring resource utilization, the system can discover the imbalance of resource use. For example, some nodes may be in a high load state for a long time, while other nodes may have a low utilization rate. Real-time monitoring of resource utilization data helps the system identify these resource bottlenecks and idle resources, thereby optimizing the task allocation strategy. The task description, expression form, priority, deadline, and task volume of the remaining tasks will change as the tasks progress during the execution of the tasks. The task description and expression form determine the logic and method of task execution to a certain extent. The remaining task volume reflects the progress of task completion, and the priority and deadline continue to affect the execution order and time urgency of the tasks. For example, for a data analysis task, the task description may include the source of the data and the analysis target. As the task is executed, the remaining task volume gradually decreases, while the priority and deadline remain unchanged or may be dynamically adjusted according to business conditions (such as the department's urgent demand for certain data increases, resulting in an increase in the priority of the task). By monitoring the remaining execution task information in real time, the system can perform task scheduling and resource allocation more accurately. For example, when the task requirements change (such as task volume, priority, etc.), the allocation of tasks on various resource nodes is dynamically adjusted to further improve resource utilization efficiency.

[0022] In an embodiment of the present application, the resource allocation adjustment module 150 is used to input the resource utilization rates of the multiple resource nodes and the remaining execution task information of the target task into the intelligent decision model to adjust the resource allocation of the remaining execution tasks in the target task.

[0023] Correspondingly, in the resource allocation adjustment module, the technical concept of the present application is to obtain the time queue of resource utilization of the multiple resource nodes before the time node to be allocated and the remaining execution task information of the target task, and use AI-based data analysis and encoding methods to perform time series encoding on the time queue of resource utilization of each resource node, and embed the remaining execution task information of the target task at the same time. Encode, based on the cross-modal prompt optimization representation between the timing coding features of each resource utilization rate and the embedded features of the remaining execution task information to intelligently obtain the remaining task allocation ratio of multiple resource nodes, and automatically adjust the resource allocation of the remaining execution tasks in the target task. In this way, it is possible to respond immediately when a new high-priority task appears or the existing task requirements change. At the same time, it is possible to dynamically optimize resource allocation, ensure that key tasks can be processed quickly, and identify and balance the load of each resource node to reduce resource waste and improve the overall efficiency of resource utilization.

[0024] Specifically, Figure 2 FIG. 1 is a block diagram of a resource allocation adjustment module in a task dynamic adjustment system based on an intelligent decision model according to an embodiment of the present application. Figure 2 As shown, the resource allocation adjustment module 150 includes: a resource utilization rate time series data acquisition unit 151, which is used to acquire the time queue of resource utilization rates of the multiple resource nodes before the time node to be allocated; a remaining execution task information data acquisition unit 152, which is used to acquire the remaining execution task information of the target task of the time node to be allocated; a resource utilization rate time series encoding unit 153, which is used to perform time series encoding based on resource utilization on the time queue of resource utilization rates of the multiple resource nodes to obtain a set of resource utilization rate time series encoding feature vectors; a remaining execution task encoding unit 154, which is used to encode the target task of the time node to be allocated; The remaining execution task information of the task is embedded and encoded to obtain a remaining execution task information embedded coding vector; the resource utilization task information fusion coding unit 155 is used to perform cross-modal prompt gate optimization coding on the set of resource utilization time series coding feature vectors and the remaining execution task information embedded coding vector to obtain a sequence of resource utilization status time series representation vectors under the semantic guidance of the remaining execution task information; the remaining execution resource allocation adjustment unit 156 is used to adjust the resource allocation of the remaining execution tasks in the target task according to the sequence of resource utilization status time series representation vectors under the semantic guidance of the remaining execution task information.

[0025] In an embodiment of the present application, the resource utilization time series data acquisition unit 151 is used to obtain the time queue of resource utilization of the multiple resource nodes before the time node to be allocated. It should be understood that the time queue of resource utilization of the multiple resource nodes before the time node to be allocated contains the resource utilization data of each resource node at a series of continuous time points. Specifically, it records the changes over time of indicators such as CPU utilization, memory utilization, storage I / O load, and network bandwidth occupancy. By obtaining the time queue data of resource utilization of the multiple resource nodes before the time node to be allocated, the system can analyze the resource utilization trend of each resource node, and then make more reasonable decisions when allocating tasks. For example, before the load of a resource node is about to reach a peak, some tasks are migrated to other nodes with lower resource utilization to avoid node overload.

[0026] Obtaining the time queue of resource utilization of the plurality of resource nodes before the time node to be allocated is a critical process involving multiple operations, and its effective implementation is of key significance for the normal operation and performance optimization of the task dynamic adjustment system based on the intelligent decision-making model. The following is a detailed description of its implementation process: First, in the data collection and recording phase, professional monitoring tools need to be deployed on each resource node of the distributed computing system or cloud computing platform. These tools continuously collect various resource usage data in real time, covering core indicators such as CPU usage, memory occupancy, storage I / O load, and network bandwidth occupancy. Taking CPU usage as an example, the monitoring tool will accurately obtain the percentage of the total time that the CPU is busy (executing task instructions) during the period at a specific frequency (such as once per second), and store this data in order. The collected data will be stored in the local database in chronological order, and will be carefully sorted and marked to clearly define the corresponding timestamp for each data point, thus laying a solid foundation for the subsequent construction of the time queue.

[0027] Next is the time queue construction phase. This process requires careful determination of an appropriate time window based on the actual needs and performance requirements of the system to build a time queue for resource utilization. The selection of this time window must take into account both the timeliness and volume of data, for example, it can be set to the past 5 minutes, 10 minutes, and other durations. After determining the time window, the data in the time window before the time node to be allocated is accurately extracted from the stored resource utilization data in chronological order. For example, if the time node to be allocated is the current moment and the time window is set to the past 10 minutes, all resource utilization data points within 10 minutes from the current moment are traced back, and these data are strictly sorted in chronological order, thus forming a resource utilization time queue for each resource node. For multiple resource nodes, independent and complete time queues will be built respectively, so as to fully and meticulously grasp the resource utilization status of the entire system.

[0028] In terms of data update and maintenance, in order to ensure that the time queue can truly reflect the actual utilization of current system resources, the monitoring tool will continuously collect new resource utilization data and update the time queue in real time. Whenever a new data point is generated, it will be immediately added to the end of the time queue, and the earliest data point will be deleted, so that the length of the time queue is always maintained within the set time window. In this way, the time queue can always contain the latest historical information on resource utilization, providing timely and accurate data support for the system when making task allocation decisions. In addition, during the data update process, it is inevitable to encounter some abnormal conditions, such as data collection errors, data abnormal values ​​caused by short-term system fluctuations, etc. In order to ensure the quality of time queue data, it is necessary to establish a complete data cleaning and exception handling mechanism. By reasonably setting data thresholds, abnormal data points can be effectively identified and filtered. For example, if a CPU usage data point is far beyond the normal range (such as more than 100%), it is determined to be abnormal data and corrected or deleted. For missing data, interpolation algorithms and other methods can be used to supplement it to ensure the integrity and continuity of time queue data.

[0029] In summary, accurate and reliable resource utilization time queue data of multiple resource nodes can be obtained.

[0030] In an embodiment of the present application, the remaining execution task information data acquisition unit 152 is used to obtain the remaining execution task information of the target task at the time node to be allocated. It should be understood that the acquisition of the task description and expression form of the remaining execution task at the time node to be allocated helps the system to deeply understand the logic and method of task execution. Different task descriptions may correspond to different resource demand modes. For example, image processing tasks may require a large amount of CPU computing resources for image algorithm calculations, while data storage tasks focus more on the I / O performance of storage devices. By obtaining this information, the intelligent decision model can more accurately evaluate the type and quantity of resources required for the remaining tasks according to the specific nature of the task, so as to more accurately match the resource status of the resource nodes in the current system. The remaining task amount information directly reflects the completion progress of the task. By obtaining this information at the time node to be allocated, the system can dynamically adjust resource allocation according to the amount of tasks. If the remaining task amount is large, it may be necessary to increase resource investment to speed up the task execution and ensure that the task is completed before the deadline; conversely, if the remaining task amount is small, the system can appropriately reduce the allocated resources and transfer them to other tasks that need more resources to improve the overall resource utilization. Obtaining the priority information of the remaining tasks of the target task at the time node to be assigned enables the system to adjust the task scheduling strategy in time according to the dynamic changes in business needs. In actual business scenarios, task priorities may change with external factors, such as changes in the urgency of customer needs, adjustments to business processes, etc. By obtaining the latest priority information, the intelligent decision-making model can give priority to meeting the resource requirements of high-priority tasks when allocating resources, ensuring that key tasks can be processed in a timely manner, and improving the system's ability to respond to dynamic changes in task priorities. Deadline information plays a key role in time constraints in task scheduling. Obtaining the deadline of the target task at the time node to be assigned, the system can evaluate the urgency of the task based on the remaining time and adjust resource allocation accordingly. For tasks approaching the deadline, the system can increase resource investment to ensure that the task is completed on time and avoid business losses caused by task delays. At the same time, combined with historical data on resource utilization (such as resource utilization time queues), the system can better plan resource allocation to avoid the inability to complete tasks due to insufficient resources when the deadline is approaching. This helps to improve the timeliness and reliability of task execution. In general, obtaining the remaining execution task information of the target task at the time node to be assigned enables the system to fully understand the current status of the task, so as to make more adaptive resource adjustment decisions. When the task changes during execution (such as an increase in task volume, priority adjustment, etc.), the system can adjust resource allocation in a timely manner based on the latest information to ensure that the task can continue to be executed smoothly under the new circumstances. This dynamic adjustment capability enhances the system's ability to cope with uncertainty in task requirements and can avoid unreasonable system resource allocation due to task changes.

[0031] In an embodiment of the present application, the resource utilization time series encoding unit 153 is used to perform time series encoding based on resource utilization on the time series of resource utilization of the multiple resource nodes to obtain a set of resource utilization time series encoding feature vectors. Specifically, in an embodiment of the present application, the resource utilization time series encoding unit is used to: pass the time series of resource utilization of each resource node in the time series of resource utilization of the multiple resource nodes through a resource utilization time series encoder based on a one-dimensional causal convolutional network model to obtain a set of resource utilization time series encoding feature vectors. It should be understood that the resource utilization of each resource node fluctuates over time, showing a trend and time series pattern of time dependence. This means that resource utilization is not static, but changes continuously with the evolution of time, showing obvious time series dynamic characteristics. Considering that the one-dimensional causal convolutional network is a deep learning model particularly suitable for processing time series data, it can effectively extract long-term and short-term time-dependent features from historical data, which is crucial to understanding the changing trend of resource utilization. Based on this, in the technical solution of the present application, the time queues of resource utilization of each resource node in the time queues of resource utilization of the multiple resource nodes are respectively passed through a resource utilization time series encoder based on a one-dimensional causal convolutional network model to utilize dilated convolution to learn and capture the change pattern of resource utilization on different time scales, thereby obtaining a set of resource utilization time series coding feature vectors. In particular, in the one-dimensional causal convolutional network model, causal convolution ensures that the model only relies on the input before a certain moment when predicting the output at that moment, which is consistent with the causal relationship in the real world. For the prediction of resource utilization, this means that the dynamic characteristics of resource usage in the actual environment can be better simulated.

[0032] In an embodiment of the present application, the remaining execution task encoding unit 154 is used to embed the remaining execution task information of the target task to obtain the remaining execution task information embedded coding vector. Specifically, in an embodiment of the present application, the remaining execution task encoding unit is used to: embed the remaining execution task information of the target task using a task embedding matrix to obtain the remaining execution task information embedded coding vector. It should be understood that the original remaining execution task information is usually in text form, but it also contains important and critical data information. Therefore, in order to map these complex information into a continuous vector space, reduce the dimension of the original data while retaining important task information, in the technical solution of the present application, the remaining execution task information of the target task is embedded and encoded using a task embedding matrix to capture and extract the intrinsic connection and key semantics between tasks, and obtain the remaining execution task information embedded coding vector. It can be understood that the task embedding matrix is ​​a tool that maps discrete task identifiers or attributes to a continuous vector space, which can capture the semantic relationship and similarity between tasks. Specifically, the task embedding matrix learns a low-dimensional, dense vector representation (i.e., an embedding vector) so that each task has a corresponding point in the vector space, and the distance between these points reflects the similarity between tasks: similar tasks are closer in the embedding space, so that the semantic information in the task description can be accurately represented. In particular, the task embedding matrix is ​​obtained by training a machine learning or deep learning model based on a large amount of task data.

[0033] In an embodiment of the present application, the resource utilization task information fusion coding unit 155 is used to perform cross-modal prompt gate optimization coding on the set of resource utilization time series coding feature vectors and the remaining execution task information embedded coding vector to obtain a sequence of resource utilization status time series representation vectors under the guidance of the remaining execution task information semantics. It should be understood that the resource utilization time series coding feature describes the usage of each resource node in the system over time, while the remaining execution task information embedded feature captures the specific requirements and characteristics of the target task. Therefore, in order to help identify which resources are most suitable for the current task requirements, thereby achieving the best match between resources and task requirements, in the technical solution of the present application, the set of resource utilization time series coding feature vectors and the remaining execution task information embedded coding vector are deeply interactively encoded by introducing cross-modal prompt gate optimization coding to obtain a sequence of resource utilization status time series representation vectors under the guidance of the remaining execution task information semantics. In this way, the interrelationships between different data modalities of resource utilization and remaining execution task information can be identified more finely, so that the weight of each modal information can be flexibly adjusted and the complementary information and effects between modalities can be strengthened, thereby obtaining a more complete and accurate feature representation to improve the system's response speed and decision-making quality.

[0034] Specifically, Figure 3 FIG. 1 is a block diagram of a resource utilization task information fusion encoding unit in a task dynamic adjustment system based on an intelligent decision model according to an embodiment of the present application. Figure 3 As shown, the resource utilization task information fusion coding unit 155 includes: a semantic interaction attention weight calculation subunit 1551, which is used to calculate the weight value of each resource utilization time series coding feature vector in the set of resource utilization time series coding feature vectors based on the remaining execution task information embedded coding vector to obtain a set of remaining execution task information-resource utilization cross-modal semantic interaction attention weights; a remaining execution task information resource utilization cross-modal interaction subunit 1552, which is used to perform semantic interaction on the set of resource utilization time series coding feature vectors and the remaining execution task information embedded coding vector. Cross-modal fine-grained local interaction is performed to obtain a set of remaining execution task information-resource utilization cross-modal interaction local feature vectors; a remaining execution task information-resource utilization interaction local feature weighting subunit 1553 is used to optimize each remaining execution task information-resource utilization cross-modal interaction local feature vector in the set of remaining execution task information-resource utilization cross-modal interaction local feature vectors based on the set of remaining execution task information-resource utilization cross-modal semantic interaction attention weights to obtain a sequence of resource utilization state time series representation vectors under the guidance of the remaining execution task information semantics.

[0035] Specifically, in the embodiment of the present application, the semantic interaction attention weight calculation subunit is used to: linearly transform the remaining execution task information embedding coding vector to obtain the remaining execution task information query coding vector and the remaining execution task information value coding vector. The process can be expressed as: ;in, is the remaining execution task information embedding coding vector, is the query embedding matrix, is matrix multiplication, is the remaining execution task information query encoding vector, is the value embedding matrix, is the encoding vector of the remaining execution task information value; Transformer cross-modal prompt encoding is performed on each resource utilization time series coding feature vector in the set of the remaining execution task information query coding vector, the remaining execution task information value coding vector, and the resource utilization time series coding feature vector to obtain a set of remaining execution task information-resource utilization cross-modal prompt semantic coding vectors. The process can be expressed as: ;in, is a set of resource utilization time series encoding feature vectors, and They are the first and the second in the set of resource utilization temporal coding feature vectors. and resource utilization temporal encoding feature vector, yes The transposed vector of is matrix multiplication, is the remaining execution task information query encoding vector, is the encoding vector of the remaining execution task information value, yes Length, yes function, yes and The remaining execution task information between them - resource utilization cross-modal cue semantic encoding vector; The set of the remaining execution task information-resource utilization cross-modal prompt semantic encoding vectors is decoded and gated to obtain the set of the remaining execution task information-resource utilization cross-modal semantic interaction attention weights. The process can be expressed as: ;in, yes function, yes and The remaining execution task information between the resource utilization cross-modal hint semantic encoding vector, is the decoding weight matrix, It is a decoder. yes Corresponding remaining execution task information-resource utilization cross-modal semantic interaction attention weights.

[0036] In detail, the remaining execution task information embedding coding vector is first linearly transformed to obtain the remaining execution task information query coding vector and the remaining execution task information value coding vector. In this way, the original remaining execution task information embedding coding vector can be converted into a form suitable for subsequent attention mechanism calculations by linear mapping. The remaining execution task information query coding vector obtained by conversion is mainly used to find the correlation with the resource utilization time series characteristics, while the remaining execution task information value coding vector is mainly used to preserve the important information of the original remaining execution task information semantic features.

[0037] Next, each resource utilization time series coding feature vector in the set of the remaining execution task information query coding vector, the remaining execution task information value coding vector, and the resource utilization time series coding feature vector is respectively subjected to Transformer cross-modal prompt coding, so as to utilize the self-attention mechanism at the core of the Transformer structure to effectively capture the deep semantic interaction information between the remaining execution task information semantic features and the resource utilization time series features. The set of remaining execution task information-resource utilization cross-modal prompt semantic coding vectors obtained by this coding method not only retains the original resource utilization time series information, but also integrates the supplementary information from the remaining execution task information, which is conducive to improving the richness and accuracy of the resource utilization time series feature representation.

[0038] Subsequently, the set of the remaining execution task information-resource utilization cross-modal prompt semantic encoding vectors is decoded and gated to obtain the set of the remaining execution task information-resource utilization cross-modal semantic interaction attention weights. Here, by calculating the remaining execution task information-resource utilization cross-modal semantic interaction attention weights, it is possible to determine which information is more critical, and then allocate attention resources based on the importance of information during calculation, avoiding over-reliance on the remaining execution task information or over-reliance on resource utilization information and ignoring other useful information. This helps the model more accurately capture the complementarity and synergy between the remaining execution task information and the resource utilization information, thereby ensuring the effectiveness and rationality of the results obtained from the final interaction.

[0039] Next, the set of resource utilization temporal coding feature vectors and the remaining execution task information embedded coding vectors are subjected to cross-modal fine-grained local interaction to obtain a set of remaining execution task information-resource utilization cross-modal interaction local feature vectors, and based on the set of remaining execution task information-resource utilization cross-modal semantic interaction attention weights, each remaining execution task information-resource utilization cross-modal interaction local feature vector in the set of remaining execution task information-resource utilization cross-modal interaction local feature vectors is optimized to obtain a sequence of resource utilization state temporal representation vectors under the guidance of the remaining execution task information semantics. The above process can be expressed by the formula: ;in, and They are the first and the second in the set of resource utilization temporal coding feature vectors. and resource utilization temporal encoding feature vector, is the remaining execution task information embedding coding vector, and They are and The corresponding remaining execution task information-resource utilization cross-modal semantic interaction attention weight, It is the point multiplication by position. It is a sequence of time series representation vectors of resource utilization status under the guidance of the semantics of the remaining execution task information.

[0040] It should be understood that the cross-modal fine-grained local interaction between the set of resource utilization temporal coding feature vectors and the remaining execution task information embedded coding vector can be achieved by multiplying by position. This calculation method can effectively capture the complementary information of resource utilization information and remaining execution task information at the same spatial position, and can maximize the interaction effect between the two different modal features. That is, while maintaining the independence of the local features of the two modalities, it is possible to discover and strengthen the common feature patterns of the two information when they interact cross-modally. Finally, based on the set of attention weights of the remaining execution task information-resource utilization cross-modal semantic interaction, each remaining execution task information-resource utilization cross-modal interaction local feature vector in the set of the remaining execution task information-resource utilization cross-modal interaction local feature vector is optimized. The optimization operation here combines the attention mechanism and the interactive local features. By weighted fusion of the resource utilization temporal features and the remaining execution task information semantic features, a sequence of resource utilization state temporal representation vectors guided by the remaining execution task information semantics with stronger expression ability and less redundant information can be generated.

[0041] In an embodiment of the present application, the remaining execution resource allocation adjustment unit 156 is used to adjust the resource allocation of the remaining execution tasks in the target task according to the sequence of resource utilization status timing representation vectors guided by the semantics of the remaining execution task information. Specifically, in an embodiment of the present application, the remaining execution resource allocation adjustment unit is used to: pass the sequence of resource utilization status timing representation vectors guided by the semantics of the remaining execution task information through a decoder-based task allocation adjuster to obtain an adjustment result, and the adjustment result is used to represent the remaining task allocation ratio of the multiple resource nodes; based on the remaining task allocation ratio of the multiple resource nodes and the remaining execution task information of the target task, adjust the resource allocation of the remaining execution tasks in the target task.

[0042] It should be understood that the sequence of resource utilization state time series representation vectors under the guidance of the remaining execution task information semantics is obtained after a series of encoding operations are performed on the resource utilization rate data and the remaining execution task information. They represent the state of the system execution resources under the guidance of the remaining execution task information in a specific high-dimensional data space. In order to understand the actual task allocation decision based on the key information in this vector sequence, it is necessary to input the sequence of resource utilization state time series representation vectors under the guidance of the remaining execution task information semantics into the decoder-based task allocation adjuster for decoding. The decoder is a neural network model that can decode the sequence of input resource utilization state time series representation vectors, extract information related to task allocation according to its internal algorithm logic, and convert it into a form representing the remaining task allocation ratio of multiple resource nodes. By comprehensively considering the remaining task allocation ratio of multiple resource nodes obtained by decoding and the remaining execution task information of the target task, the resource allocation of the remaining execution tasks in the target task can be adjusted. Specifically, the remaining execution task information of the target task (such as task volume, priority, deadline, etc.) reflects the demand characteristics of the task itself, while the remaining task allocation ratio of multiple resource nodes reflects the current allocation tendency and availability of system resources. By combining the remaining task allocation ratio of resource nodes and the remaining execution task information of the target task, the resources of the resource nodes can be fully utilized while meeting the urgency of the task, thereby achieving accurate matching of resource allocation and task requirements.

[0043] Adjusting the resource allocation of the remaining execution tasks in the target task based on the remaining task allocation ratios of the multiple resource nodes and the remaining execution task information of the target task is a systematic and dynamic process. The following is a detailed description of its implementation process: First, in the preliminary preparation stage of resource allocation adjustment, the system needs to accurately obtain the remaining task allocation ratio of multiple resource nodes and the remaining execution task information of the target task. Then enter the specific operation link of resource allocation adjustment. In terms of task volume allocation strategy, the system will accurately calculate the task volume to be allocated to each resource node based on the remaining task volume of the target task and the remaining task allocation ratio of each resource node. Assuming that the target task has 500 units of task volume remaining, the remaining task allocation ratio of resource node A is 20%, the remaining task allocation ratio of resource node B is 40%, and the remaining task allocation ratio of resource node C is 40%. Then, according to the proportion calculation, resource node A should be allocated 100 units of task volume, and resource nodes B and C should be allocated 200 units of task volume respectively. This task volume allocation method based on the remaining task allocation ratio can make the task volume reasonably distributed among resource nodes, maximize the idleness of resource nodes, and effectively improve resource utilization.

[0044] However, task allocation is not based solely on the proportion of remaining tasks, but also needs to fully consider the priority and deadline of the target task. For those tasks with high priority or close to the deadline, the system will give priority to allocating them to resource nodes with a high proportion of remaining tasks and superior performance to ensure that these critical tasks can be successfully completed within the specified time. For example, if the target task is an urgent order processing task, and resource node B not only has a considerable proportion of remaining tasks, but also has strong computing and processing capabilities, the system will appropriately increase the amount of tasks allocated to resource node B. Even if the originally assigned task volume is already at a high level, it must give priority to the needs of urgent tasks to ensure the normal operation of the business.

[0045] During the task execution process, the system will continuously monitor the actual operating status of the resource nodes and the progress of the target tasks. Once it is found that the actual execution efficiency of a resource node exceeds expectations, or the remaining task volume of the target task decreases faster than expected, the system will dynamically adjust the allocation of the remaining tasks in a timely manner. For example, if the task volume originally assigned to resource node A is 100 units, but because its execution efficiency is significantly higher than other nodes and a large number of tasks are completed in a shorter period of time, the system will consider transferring part of the tasks originally assigned to other nodes to resource node A, further speeding up the overall execution speed of the task and improving the system's responsiveness.

[0046] In the subsequent optimization phase of resource allocation adjustment, the system will conduct a comprehensive evaluation of the performance of resource nodes and the effect of task execution. By collecting key indicators such as resource utilization, task completion time, and task execution success rate, the effectiveness of the resource allocation adjustment strategy is deeply analyzed. For example, if it is found that the resource utilization of a resource node is still low after the allocation adjustment, while other nodes are overloaded, this indicates that the allocation strategy may need to be further optimized to achieve a more balanced use of resources.

[0047] Finally, based on the results of the performance evaluation, the system will continuously optimize the resource allocation adjustment strategy. The system can learn the optimal allocation mode under different task types, resource node states, and task requirements, and continuously improve the allocation algorithm. For example, if it is found that for a certain type of task, allocating resources according to a specific remaining task allocation ratio can achieve a more ideal effect, then when encountering similar tasks in the future, the system can draw on this experience for allocation. Through continuous learning and optimization, the system gradually improves its ability to adapt to various complex situations, realizes efficient coordination of resource allocation and task execution, and ensures the stable and efficient operation of the entire system.

[0048] In particular, considering that the set of resource utilization temporal coding feature vectors and the remaining execution task information embedded coding vector respectively represent the one-dimensional temporal correlation characteristics of the resource utilization of the resource node and the embedded coding characteristics of the remaining execution task information of the target task, when performing deep interactive coding guided by a cross-modal prompt gate, due to the cross-modal prompt gate guidance differences caused by the differences in different coding mode features corresponding to heterogeneous data, the sequence of resource utilization state temporal representation vectors under the semantic guidance of the remaining execution task information will also have probabilistic convergence and divergence based on different coding interactivity, thereby affecting the accuracy of the adjustment results obtained by the decoder-based task allocation adjuster.

[0049] Preferably, passing the sequence of resource utilization state time series representation vectors guided by the semantics of the remaining execution task information through a decoder-based task allocation adjuster to obtain an adjustment result includes: Cascading the sequence of resource utilization state time series representation vectors under the guidance of the remaining execution task information semantics into a resource utilization state time series cascade representation vector under the guidance of the remaining execution task information semantics; Determine the median eigenvalue, maximum eigenvalue and minimum eigenvalue in the temporal cascade representation vector of the resource utilization state under the guidance of the semantics of the remaining execution task information, and divide the median eigenvalue by the difference between the maximum eigenvalue and the minimum eigenvalue to obtain the temporal cascade distribution probability value of the resource utilization state under the guidance of the semantics of the remaining execution task information. The process can be expressed as: ;in, and They represent the median eigenvalue, maximum eigenvalue, and minimum eigenvalue in the temporal concatenation representation vector of resource utilization status guided by the semantics of the remaining execution task information, It represents the probability value of the time series cascade distribution of resource utilization status under the guidance of the semantics of the remaining execution task information; The resource utilization state time series cascade representation vector under the guidance of the remaining execution task information semantics is normalized to the maximum value to obtain the resource utilization state time series cascade probability feature vector under the guidance of the remaining execution task information semantics ; The power function of each eigenvalue of the resource utilization state time series cascade probability feature vector guided by the semantics of the remaining execution task information is calculated with one minus the distribution probability value of the resource utilization state time series cascade under the semantics of the remaining execution task information as the exponent to obtain the first resource utilization state time series cascade convergence feature vector guided by the semantics of the remaining execution task information. The process can be expressed as: ;in, represents the temporal cascade probability feature vector of resource utilization status guided by the semantics of the remaining execution task information, It represents the probability value of the time series cascade distribution of resource utilization status under the guidance of the semantics of the remaining execution task information. It means point multiplication by position. Indicates minus one Calculate for index The power function of each eigenvalue in , Represents the time series cascade convergence feature vector of resource utilization state guided by the semantics of the first remaining execution task information; The power function of each eigenvalue of the point difference feature vector between the unit feature vector and the resource utilization state time series cascade probability feature vector guided by the remaining execution task information semantics is calculated with the probability value of the resource utilization state time series cascade distribution under the remaining execution task information semantics as the exponent to obtain the second resource utilization state time series cascade convergence feature vector under the remaining execution task information semantics. The process can be expressed as: ;in, represents the unit eigenvector, represents the temporal cascade probability feature vector of resource utilization status guided by the semantics of the remaining execution task information, They represent point-by-point subtraction and point-by-point multiplication, respectively. It represents the probability value of the time series cascade distribution of resource utilization status under the guidance of the semantics of the remaining execution task information. Indicates Calculate for index Point reduction The power function of each eigenvalue in the obtained eigenvector is Represents the time series cascade convergence feature vector of resource utilization state guided by the semantics of the second remaining execution task information; The resource utilization state temporal cascade probability feature vector under the guidance of the remaining execution task information semantics is multiplied by the difference between one minus the resource utilization state temporal cascade distribution probability value under the guidance of the remaining execution task information semantics to obtain the first resource utilization state temporal cascade qualification feature vector under the guidance of the remaining execution task information semantics. This process can be expressed as: ;in, represents the temporal cascade probability feature vector of resource utilization status guided by the semantics of the remaining execution task information, They represent point multiplication by position, It represents the probability value of the time series cascade distribution of resource utilization status under the guidance of the semantics of the remaining execution task information. Represents a time series cascade qualified feature vector of resource utilization state guided by the semantics of the first remaining execution task information; The point difference feature vector is multiplied by the probability value of the resource utilization state time series cascade distribution under the guidance of the remaining execution task information semantics to obtain the second resource utilization state time series cascade qualification feature vector under the guidance of the remaining execution task information semantics. The process can be expressed as: ;in, represents the unit eigenvector, represents the temporal cascade probability feature vector of resource utilization status guided by the semantics of the remaining execution task information, They represent point-by-point subtraction and point-by-point multiplication, respectively. It represents the probability value of the time series cascade distribution of resource utilization status under the guidance of the semantics of the remaining execution task information. Represents a time series cascade qualified feature vector of resource utilization state guided by the semantics of the second remaining execution task information; After dot-multiplying the first remaining task information semantically guided resource utilization state temporal cascade convergence feature vector and the second remaining task information semantically guided resource utilization state temporal cascade convergence feature vector, dot-multiplying the first remaining task information semantically guided resource utilization state temporal cascade restriction feature vector and the second remaining task information semantically guided resource utilization state temporal cascade restriction feature vector to obtain an optimized remaining task information semantically guided resource utilization state temporal cascade representation vector. This process can be expressed as: ;in, represents the time series cascade convergence feature vector of resource utilization state guided by the semantics of the first remaining execution task information, represents the time series cascade convergence feature vector of resource utilization status under the guidance of the semantics of the second remaining execution task information, represents the time series cascade qualified feature vector of resource utilization state guided by the semantics of the first remaining execution task information, represents the time series cascade qualified feature vector of resource utilization status guided by the semantics of the second remaining execution task information, and They represent point-by-point multiplication and point-by-point addition, respectively. A time-series cascade representation vector of resource utilization status guided by the semantics of the remaining execution task information of the optimization; The resource utilization state time series cascade representation vector guided by the semantics of the optimized remaining execution task information is passed through the decoder-based task allocation adjuster to obtain the adjustment result.

[0050] That is, by using the cross entropy form power series of the distribution analytic probability of the sequence cascade of the resource utilization status time series representation vectors guided by the semantics of the remaining execution task information as the probability distribution convergence limit, the regression probability convergence approximation is performed on the basis of the combination of the feature set distribution and the probability density distribution of the resource utilization status time series cascade representation vectors guided by the semantics of the remaining execution task information, so as to avoid the conservative convergence control of the fatal potential divergence of the resource utilization status time series cascade representation vectors guided by the semantics of the remaining execution task information, further use the probability distribution cross entropy as the objective function to guide the limited recovery strategy, so as to achieve the common agility of the convergence of the probability convergent and divergent feature set to the probability density distribution space, and improve the accuracy of the adjustment result obtained by the decoder-based task allocation adjuster of the sequence of the resource utilization status time series representation vectors guided by the semantics of the remaining execution task information.

[0051] In summary, the resource allocation adjustment module 150 is clearly explained, which uses AI-based data analysis and encoding to perform time series encoding on the time queue of resource utilization of each resource node before the time node to be allocated, and embeds the remaining execution task information of the target task, thereby intelligently obtaining the remaining task allocation ratio of multiple resource nodes based on the cross-modal prompt optimization representation between the timing encoding features of each resource utilization and the embedded features of the remaining execution task information, and automatically adjusting the resource allocation of the remaining execution tasks in the target task. In this way, it is possible to respond immediately when a new high-priority task appears or the requirements of an existing task change. At the same time, it is possible to dynamically optimize resource allocation, ensure that critical tasks can be processed quickly, identify and balance the load of each resource node, so as to reduce resource waste and improve the overall efficiency of resource utilization.

[0052] In summary, the task dynamic adjustment system 100 based on the intelligent decision-making model of the embodiment of the present application is explained, which first verifies and parses the received task request, extracts task information such as task description, expression form, priority, deadline and task amount, and stores it in the task management database, and then extracts the target task from the task management database. According to the task information and system load status, the target task is assigned to multiple resource nodes, and during the task execution process, the resource utilization of multiple resource nodes and the remaining execution information of the target task are monitored in real time, and these data are input into the intelligent decision-making model, and the resource allocation optimization and adjustment are performed on the remaining execution part of the target task through model analysis. In this way, the task can be efficiently executed under the condition of meeting its needs, and the system resources can also be fully utilized.

[0053] Figure 4FIG. 1 is a flowchart of a method for dynamically adjusting tasks based on an intelligent decision-making model according to an embodiment of the present application. Figure 4 As shown, in the task dynamic adjustment method based on the intelligent decision-making model, it includes: S110, verifying and parsing the received task request to obtain task information, wherein the task information includes task description, expression form, priority, deadline and task amount; S120, storing the task information in a task management database; S130, extracting the target task from the task management database, and allocating the target task to multiple resource nodes according to the task information of the target task and the system load status; S140, in the process of the target task being executed, real-time monitoring of the resource utilization of the multiple resource nodes and the remaining execution task information of the target task; S150, inputting the resource utilization of the multiple resource nodes and the remaining execution task information of the target task into the intelligent decision-making model to adjust the resource allocation of the remaining execution tasks in the target task.

[0054] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned task dynamic adjustment method based on the intelligent decision model have been referred to above. Figures 1 to 3 The description of the task dynamic adjustment system based on the intelligent decision-making model has been introduced in detail, and therefore, its repeated description will be omitted.

[0055] In summary, the task dynamic adjustment method based on the intelligent decision-making model of the embodiment of the present application is explained, which first verifies and parses the received task request, extracts task information such as task description, expression form, priority, deadline and task amount, and stores it in the task management database, and then extracts the target task from the task management database. According to the task information and system load status, the target task is allocated to multiple resource nodes, and during the task execution process, the resource utilization of multiple resource nodes and the remaining execution information of the target task are monitored in real time, and these data are input into the intelligent decision-making model, and the resource allocation optimization and adjustment are performed on the remaining execution part of the target task through model analysis. In this way, the task can be efficiently executed under the condition of meeting its needs, and the system resources can also be fully utilized.

Claims

1. A task dynamic adjustment system based on an intelligent decision-making model, characterized in that: include: A task request processing module, used to verify and parse the received task request to obtain task information, wherein the task information includes task description, expression form, priority, deadline and task amount; A task request storage module, used for storing the task information in a task management database; A target task allocation module, used for extracting target tasks from the task management database, and allocating the target tasks to multiple resource nodes according to the task information of the target tasks and the system load status; A resource utilization task information monitoring module is used to monitor the resource utilization of the multiple resource nodes and the remaining execution task information of the target task in real time during the execution of the target task; The resource allocation adjustment module is used to input the resource utilization rates of the multiple resource nodes and the remaining execution task information of the target task into the intelligent decision model to adjust the resource allocation of the remaining execution tasks in the target task.

2. The task dynamic adjustment system based on the intelligent decision model according to claim 1 is characterized in that: The task request processing module is used to: Verifying the format, parameters and permission information of the received task request to obtain a verified task request; Task request information is extracted from the verified task request, and the task request information is parsed to obtain the task information, wherein the task request information includes a task name, parameters, priority requirements, and resource requirements.

3. The task dynamic adjustment system based on the intelligent decision model according to claim 2 is characterized in that: The resource allocation adjustment module comprises: A resource utilization time series data acquisition unit, used to acquire a time queue of resource utilization of the plurality of resource nodes before a time node to be allocated; The remaining execution task information data acquisition unit is used to acquire the remaining execution task information of the target task to be assigned at the time node; A resource utilization rate time series coding unit, configured to perform time series coding based on resource utilization rate on the time series of resource utilization rate of the plurality of resource nodes to obtain a set of resource utilization rate time series coding feature vectors; The remaining execution task encoding unit is used to embed the remaining execution task information of the target task to obtain the remaining execution task information embedding encoding vector; A resource utilization task information fusion encoding unit, used for performing cross-modal prompt gate optimization encoding on the set of resource utilization time series encoding feature vectors and the remaining execution task information embedded encoding vectors to obtain a sequence of resource utilization state time series representation vectors under the guidance of the remaining execution task information semantics; The remaining execution resource allocation adjustment unit is used to adjust the resource allocation of the remaining execution tasks in the target task according to the sequence of resource utilization state time series representation vectors under the guidance of the semantics of the remaining execution task information.

4. The task dynamic adjustment system based on the intelligent decision model according to claim 3 is characterized in that: The resource utilization timing encoding unit is used to: respectively pass the time queue of resource utilization of each resource node in the time queue of resource utilization of the multiple resource nodes through a resource utilization timing encoder based on a one-dimensional causal convolutional network model to obtain a set of resource utilization timing encoding feature vectors.

5. The task dynamic adjustment system based on the intelligent decision model according to claim 4 is characterized in that: The remaining execution task encoding unit is used to: use the task embedding matrix to embed the remaining execution task information of the target task to obtain the remaining execution task information embedding encoding vector.

6. The task dynamic adjustment system based on the intelligent decision model according to claim 5 is characterized in that: The resource utilization task information fusion encoding unit includes: A semantic interaction attention weight calculation subunit, used to calculate the weight value of each resource utilization time series coding feature vector in the set of resource utilization time series coding feature vectors based on the remaining execution task information embedded coding vector to obtain a set of remaining execution task information-resource utilization cross-modal semantic interaction attention weights; The remaining execution task information resource utilization cross-modal interaction subunit is used to perform cross-modal fine-grained local interaction on the set of resource utilization time series encoding feature vectors and the remaining execution task information embedded encoding vector to obtain a set of remaining execution task information-resource utilization cross-modal interaction local feature vectors; The remaining execution task information resource utilization interaction local feature weighting sub-unit is used to optimize each remaining execution task information-resource utilization cross-modal interaction local feature vector in the set of remaining execution task information-resource utilization cross-modal interaction local feature vectors based on the set of remaining execution task information-resource utilization cross-modal semantic interaction attention weights to obtain a sequence of resource utilization state time series representation vectors under the guidance of the remaining execution task information semantics.

7. The task dynamic adjustment system based on the intelligent decision model according to claim 6 is characterized in that: The semantic interaction attention weight calculation subunit is used to: Performing a linear transformation on the remaining execution task information embedding coding vector to obtain a remaining execution task information query coding vector and a remaining execution task information value coding vector; Performing Transformer cross-modal prompt encoding on each resource utilization time series coding feature vector in the set of the remaining execution task information query coding vector, the remaining execution task information value coding vector, and the resource utilization time series coding feature vector to obtain a set of remaining execution task information-resource utilization cross-modal prompt semantic coding vectors; The set of remaining execution task information-resource utilization cross-modal prompt semantic encoding vectors is decoded and gated modulated to obtain a set of remaining execution task information-resource utilization cross-modal semantic interaction attention weights.

8. The task dynamic adjustment system based on the intelligent decision model according to claim 7 is characterized in that: The remaining execution resource allocation adjustment unit is used to: The sequence of resource utilization state time series representation vectors guided by the semantics of the remaining execution task information is passed through a decoder-based task allocation adjuster to obtain an adjustment result, wherein the adjustment result is used to represent the remaining task allocation ratio of the multiple resource nodes; Based on the remaining task allocation ratios of the multiple resource nodes and the remaining execution task information of the target task, resource allocation adjustment is performed on the remaining execution tasks in the target task.

9. A task dynamic adjustment method based on an intelligent decision model, characterized in that: include: Verify and parse the received task request to obtain task information, including task description, expression form, priority, deadline and task amount; Storing the task information in a task management database; Extracting a target task from the task management database, and allocating the target task to a plurality of resource nodes according to task information of the target task and a system load status; During the execution of the target task, real-time monitoring of resource utilization of the plurality of resource nodes and remaining execution task information of the target task; The resource utilization rates of the multiple resource nodes and the remaining execution task information of the target task are input into the intelligent decision model to adjust the resource allocation of the remaining execution tasks in the target task.

10. The task dynamic adjustment method based on the intelligent decision model according to claim 9 is characterized in that: Verify and parse the received task request to obtain task information, including task description, expression form, priority, deadline and task amount, including: Verifying the format, parameters and permission information of the received task request to obtain a verified task request; Task request information is extracted from the verified task request, and the task request information is parsed to obtain the task information, wherein the task request information includes a task name, parameters, priority requirements, and resource requirements.

Images