Self-adaptive task priority management method based on dynamic entropy weight and time decay factor
By adopting an adaptive task priority management method based on dynamic entropy weights and time decay factors, the problems of weight allocation and scheduling fairness in high-concurrency task flows are solved, achieving the adaptability and stability of task scheduling, preventing task starvation, and improving the system's processing capacity.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-11
- Publication Date
- 2026-04-21
AI Technical Summary
Existing task scheduling technologies struggle to achieve objective and adaptive weight allocation and scheduling fairness when faced with high-concurrency task flows with dynamically changing characteristics. They also suffer from high computational complexity or poor robustness, failing to effectively address the task starvation problem.
An adaptive task priority management method based on dynamic entropy weight and time decay factor is adopted. By constructing a task feature matrix, dynamic weights are calculated using information entropy, and a nonlinear time decay factor is introduced to dynamically adjust task priorities and prevent low-priority tasks from being stuck for a long time.
It improves the adaptive task scheduling capability in complex and dynamic environments, prevents task starvation, ensures the stability and accuracy of the system when processing heterogeneous data, and avoids calculation deviations caused by differences in data units.
Smart Images
Figure CN121900893A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer task scheduling and resource allocation technology, and in particular to an adaptive task priority management method based on dynamic entropy weight and time decay factor. Background Technology
[0002] With the rapid development of IoT, edge computing, and industrial automation technologies, modern information processing systems face the challenge of concurrent processing of massive, multi-source, and heterogeneous tasks. In these complex distributed systems, how to efficiently and rationally allocate limited computing resources to the queue of tasks to be processed is a key factor determining the overall system throughput and quality of service. Traditional task management typically employs basic scheduling strategies such as first-come, first-served, static priority queues, or weighted round-robin. These methods can maintain basic system operation in simple scenarios with uniform task attributes and stable loads.
[0003] However, when faced with high-concurrency task flows with dynamically changing characteristics, the aforementioned existing technologies reveal significant shortcomings. First, regarding weight allocation: existing priority algorithms largely rely on manually preset fixed weights. This static assignment method is highly subjective and unpredictable, unable to adaptively adjust according to the dispersion of real-time data, leading to scheduling strategy failure when data characteristics fluctuate. Second, regarding anti-starvation mechanisms: to prevent low-priority tasks from remaining unexecuted for extended periods, existing technologies typically employ linear time aging strategies. However, in high-load congestion scenarios, the linear growth rate often lags behind the arrival rate of new high-priority tasks, causing deadlocks or starvation in long-tail tasks. Furthermore, while deep learning-based intelligent scheduling algorithms can optimize decision-making, their enormous computational overhead and black-box nature make them difficult to deploy in resource-constrained industrial edge devices.
[0004] In summary, existing task scheduling technologies struggle to simultaneously balance the objective adaptability of weight allocation with scheduling fairness under high concurrency, and generally suffer from high computational complexity or poor robustness. Therefore, there is an urgent need to develop a lightweight adaptive task priority management method that can dynamically assign weights using the statistical characteristics of the data itself, combined with a nonlinear time compensation mechanism to completely solve the task starvation problem, in order to meet the real-time and reliable scheduling requirements in complex environments. Summary of the Invention
[0005] To address the above problems, this invention proposes an adaptive task priority management method based on dynamic entropy weight and time decay factor. The specific steps are as follows:
[0006] Step 1: Construct a task feature matrix and quantify attributes. When the management system receives a task flow to be processed, it first defines a task set, then extracts the key feature indicators of each task, and constructs the original decision matrix.
[0007] Step 2: Standardization of feature data. In order to eliminate the influence of inconsistent dimensions of different feature indicators, the matrix needs to be standardized to obtain a standardized matrix.
[0008] Step 3: Calculate the dynamic weights of features based on information entropy. Using information entropy theory, automatically determine which feature can better reflect the differences of the task in the current environment, avoiding the rigidity of manually fixing weights.
[0009] Step 4: Calculate the static composite score of the task. Based on the standardized matrix and dynamic weights, calculate the initial static score for each task.
[0010] Step 5: Introduce a nonlinear time decay factor for dynamic compensation. In order to prevent low-priority tasks from staying in the queue for a long time, a nonlinear compensation mechanism based on waiting time is introduced to calculate the final dynamic priority index.
[0011] Step 6: Queue reordering and resource allocation. The task set is sorted in descending order based on the calculated final priority index.
[0012] This invention provides an adaptive task priority management method based on dynamic entropy weight and time decay factor. The beneficial effects of this invention are as follows:
[0013] 1. This invention employs a dynamic weighting mechanism based on information entropy theory, capable of calculating the data dispersion of each feature index in real time. When the differences of a certain feature in the current task flow are small, the algorithm automatically reduces its weight; conversely, it increases the weight. This enables the system to objectively and sensitively perceive changes in task flow distribution without human intervention, significantly improving the adaptability of the scheduling strategy in complex dynamic environments.
[0014] 2. The exponential model of this invention endows delayed tasks with non-linear acceleration and priority enhancement capabilities. This means that as the waiting time increases, the priority of any low-value task will eventually surpass that of newly arrived high-value tasks. Mathematically, this guarantees absolute fairness in scheduling and effectively prevents the starvation phenomenon where long-tail tasks remain unexecuted for extended periods.
[0015] 3. This invention effectively solves the problem of direct comparison between different physical units. In particular, the reverse flipping process for cost-type indicators ensures the consistency of all data in the logical direction, avoiding calculation deviations caused by huge differences in the order of magnitude of data units, thereby guaranteeing the stability and accuracy of the system when processing heterogeneous data. Attached Figure Description
[0016] Figure 1 This is a flowchart of the present invention.
[0017] Figure 2 This is the flowchart for step 6. Detailed Implementation
[0018] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:
[0019] This invention proposes an adaptive task priority management method based on dynamic entropy weights and time decay factors. This method utilizes information entropy theory to dynamically assign weights to task characteristics, eliminating human subjectivity; simultaneously, it introduces an exponential time decay model to provide nonlinear compensation for delayed tasks. This invention requires no model training, has low computational overhead, and effectively solves the task deadlock and starvation problems under high concurrency, achieving efficient and fair real-time scheduling. The invention flowchart is shown below. Figure 1 As shown, the steps of the present invention will be described in detail below.
[0020] Step 1: Construct a task feature matrix and quantify attributes. When the management system receives a task flow to be processed, it first defines a task set, then extracts the key feature indicators of each task, and constructs the original decision matrix.
[0021] The system transforms the multi-source heterogeneous task flow into a computer-processable mathematical model. Instead of treating tasks as single pending items, the system parses them into a set of multi-dimensional vectors with specific numerical attributes.
[0022] Step 1.1 Task Set Definition and Buffering
[0023] Set in the current time window When the management system receives a batch of tasks to be processed, it will process this batch of tasks. Defined as a set ,in, For the i-th task, This indicates the total number of tasks currently existing in the buffer pool.
[0024] To ensure the system's real-time performance, the management system sets up a dynamic buffer for real-time monitoring. The value of . When When this happens, the subsequent feature extraction process is triggered.
[0025] Step 1.2 Multidimensional Key Feature Extraction
[0026] For each task The management system extracts four key feature indicators based on preset management objectives. To ensure the effectiveness of the entropy weight method calculation, the selected features cover both cost-related and benefit-related dimensions:
[0027] Step 1.2.1 Define resource consumption class features
[0028] Expected execution time : The estimated time slice length for the task.
[0029] Memory / bandwidth utilization Peak hardware resources required for task execution.
[0030] The smaller the value of resource consumption indicators, the easier it is for the system to process the task, and the more likely it is to be prioritized in scheduling.
[0031] Step 1.2.2 Define business value class features
[0032] User / Customer Level The priority of the source that initiates the task is defined as visitor, user, administrator, and system.
[0033] Mission urgency The urgency of the deadline defined in the business logic.
[0034] The higher the value of a business value metric, the more important the task is and the more priority it should be given.
[0035] Step 1.3 Quantization mapping of non-numerical features
[0036] In real-world management scenarios, user / customer levels and mission urgency This is a qualitative textual description. To construct a mathematical matrix, a quantization mapping mechanism is established:
[0037] For user level characteristics : Set up the mapping table .
[0038] For urgency characteristics : Set up the mapping table .
[0039] The system has a built-in lookup table that automatically converts non-numeric tags into corresponding scalar values when reading task metadata.
[0040] Step 1.4 Construct the original decision matrix
[0041] After the above extraction and quantification, the first Task Represented as a 3D feature vector n=4. All Stack the vectors of each task to construct the original decision matrix. :
[0042]
[0043] in, This represents the original value of the i-th task on the j-th feature. Before constructing the matrix, the system performs a null value check. If a null value is detected... If a feature is missing, it is filled in based on the historical average value of that feature to ensure the matrix is intact. The density of the data prevents subsequent calculations from crashing.
[0044] Step 2: Standardization of feature data. To eliminate the influence of inconsistent dimensions of different feature indicators, the original decision matrix needs to be standardized. Perform standardization processing to obtain a standardized matrix;
[0045] For benefit-oriented indicators, higher values take precedence, and positive standardization is used.
[0046]
[0047] Where i is the task index variable and j is the feature index variable. For the minimum value in the column, The maximum value in the column.
[0048] For cost-related indicators, smaller values are given higher priority, and inverse standardization is used:
[0049]
[0050] After processing, the standardized matrix is obtained. .
[0051] Step 3: Calculate the dynamic weights of features based on information entropy. Using information entropy theory, automatically determine which feature can better reflect the differences of the task in the current environment, avoiding the rigidity of manually fixing weights.
[0052] Calculate the first The first feature The proportion of each task :
[0053]
[0054] Introducing the minimum value This is to prevent Logarithmic calculations are meaningless at this time.
[0055] Calculate the first Information entropy of each feature :
[0056]
[0057] Where, constant .
[0058] Calculate the first The dynamic difference coefficient of each feature And determine the final weights. :
[0059]
[0060] If the execution time of all tasks is roughly the same at a certain moment, the entropy value of that feature is large, and its weight will automatically decrease; if the user levels differ greatly, the weight allocation will automatically tilt towards them.
[0061] Step 4: Calculate the static composite score of the task. Based on the standardized matrix and dynamic weights, calculate the initial static score for each task.
[0062]
[0063] in, For each task The initial static score only reflects the attribute value of the task itself.
[0064] Step 5: Introduce a nonlinear time decay factor for dynamic compensation. In order to prevent low-priority tasks from staying in the queue for a long time, a nonlinear compensation mechanism based on waiting time is introduced to calculate the final dynamic priority index.
[0065]
[0066] in, This is the final dynamic priority index; Queue dwell time; It is a natural constant; The time sensitivity coefficient is 0.5; exponential function. This means that the longer a task waits, the faster its priority is increased, thus forcing the system to process the delayed task within a specific time.
[0067] Step 6: Queue reordering and resource allocation. The task set is sorted in descending order based on the calculated final priority index.
[0068] Step 6.1 Select The corresponding tasks are given priority and enter the execution queue. It is the highest criterion for determining the order of task scheduling.
[0069] Step 6.2 Map the task to computing resources for processing.
[0070] Step 6.3 The system periodically repeats steps 3 to 6, or upon task completion, to accommodate the impact of newly added tasks on the overall weight distribution. The flowchart for step 6 is as follows: Figure 2 As shown.
[0071] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.
Claims
1. An adaptive task priority management method based on dynamic entropy weight and time decay factor, the specific steps of which are as follows, characterized by: Step 1: Construct a task feature matrix and quantify attributes. When the management system receives a task flow to be processed, it first defines a task set, then extracts the key feature indicators of each task, and constructs the original decision matrix. Step 2: Standardization of feature data. To eliminate the influence of inconsistent dimensions of different feature indicators, the original decision matrix needs to be standardized. Perform standardization processing to obtain a standardized matrix; Step 3: Calculate the dynamic weights of features based on information entropy. Using information entropy theory, automatically determine which feature can better reflect the differences of the task in the current environment, avoiding the rigidity of manually fixing weights. Step 4: Calculate the static composite score of the task. Based on the standardized matrix and dynamic weights, calculate the initial static score for each task. Step 5: Introduce a nonlinear time decay factor for dynamic compensation. In order to prevent low-priority tasks from staying in the queue for a long time, a nonlinear compensation mechanism based on waiting time is introduced to calculate the final dynamic priority index. Step 6: Queue reordering and resource allocation. The task set is sorted in descending order based on the calculated final priority index.
2. The adaptive task priority management method based on dynamic entropy weight and time decay factor according to claim 1, characterized in that: The construction of the task feature matrix and attribute quantization in step 1 can be represented as follows: The system transforms the multi-source heterogeneous task flow to be processed into a mathematical model that can be processed by a computer; instead of treating tasks as a single to-do item, the system parses them into a set of multi-dimensional vectors with specific numerical attributes. Step 1.1 Task Set Definition and Buffering Set in the current time window When the management system receives a batch of tasks to be processed, it will process the batch of tasks... Defined as a set ,in, For the i-th task, This indicates the total number of tasks currently existing in the buffer pool; To ensure the system's real-time performance, the management system sets up a dynamic buffer for real-time monitoring. The value; when When this occurs, the subsequent feature extraction process is triggered. Step 1.2 Multidimensional Key Feature Extraction For each task The management system extracts four key feature indicators based on preset management objectives; to ensure the effectiveness of the entropy weight method calculation, the selected features cover both cost-related and benefit-related dimensions. Step 1.2.1 Define resource consumption class features Expected execution time : Estimated time slice length for the task; Memory / bandwidth utilization Peak hardware resource requirements for task execution; The smaller the value of resource consumption indicators, the easier it is for the system to process the task, and the more likely it is to be prioritized in scheduling. Step 1.2.2 Define business value class features User / Customer Level The priority of the source for initiating a task is defined as visitor, user, administrator, and system. Mission urgency The urgency of the deadline defined in the business logic; The higher the value of a business value indicator, the more important the task is and the more priority it should be given. Step 1.3 Quantization mapping of non-numerical features In real-world management scenarios, user / customer levels and mission urgency It is a qualitative textual description; in order to construct a mathematical matrix, a quantitative mapping mechanism is established: For user level characteristics : Set up the mapping table ; For urgency characteristics : Set up the mapping table ; The system has a built-in lookup table that automatically converts non-numeric tags into corresponding scalar values when reading task metadata. Step 1.4 Construct the original decision matrix ; After the above extraction and quantification, the first Task Represented as a 3D feature vector n=4; put all Stack the vectors of each task to construct the original decision matrix. : ； in, This represents the original value of the i-th task on the j-th feature; before constructing the matrix, the system performs a null value check; if a null value is detected... If a feature is missing, it is filled in based on the historical average value of that feature to ensure the matrix is intact. The density of the data prevents subsequent calculations from crashing.
3. The adaptive task priority management method based on dynamic entropy weight and time decay factor according to claim 1, characterized in that: The standardization of feature data in step 2 can be represented as follows: For benefit-oriented indicators, higher values take precedence, and positive standardization is used. ； Where i is the task index variable and j is the feature index variable. For the minimum value in the column, Maximum value in the column; For cost-related indicators, smaller values are given higher priority, and inverse standardization is used: ； After processing, the standardized matrix is obtained. .
4. The adaptive task priority management method based on dynamic entropy weight and time decay factor according to claim 1, characterized in that: The dynamic weights of the features calculated based on information entropy in step 3 are represented as follows: Calculate the first The first feature The proportion of each task : ； Introducing the minimum value This is to prevent Logarithmic calculations are meaningless at this time. Calculate the first Information entropy of each feature : ； Where, constant ; Calculate the first The dynamic difference coefficient of each feature And determine the final weights. : ； If the execution time of all tasks is roughly the same at a certain moment, the entropy value of that feature is large, and its weight will automatically decrease; if the user levels differ greatly, the weight allocation will automatically tilt towards them.
5. The adaptive task priority management method based on dynamic entropy weight and time decay factor according to claim 1, characterized in that: The static composite score of the task calculated in step 4 is represented as follows: Calculate the initial static score for each task; ； in, For each task The initial static score only reflects the attribute value of the task itself.
6. The adaptive task priority management method based on dynamic entropy weight and time decay factor according to claim 1, characterized in that: Step 5 introduces a nonlinear time decay factor for dynamic compensation, as shown below: Calculate the final dynamic priority index; ； in, This is the final dynamic priority index; Queue dwell time; It is a natural constant; The time sensitivity coefficient is 0.5; exponential function. This means that the longer a task waits, the faster its priority is increased, thus forcing the system to process the delayed task within a specific time.
7. The adaptive task priority management method based on dynamic entropy weight and time decay factor according to claim 1, characterized in that: The queue reordering and resource allocation in step 6 are represented as follows: Step 6.1 Select The corresponding tasks are given priority and enter the execution queue. The highest criterion for determining the order of task scheduling; Step 6.2 Map the task to computing resources for processing; Step 6.3 The system periodically or when a task is completed repeats steps 3 to 6 to accommodate the impact of newly added tasks on the overall weight distribution.