Search

CN-122022415-A - Resource allocation method meeting maximization of number of parallel underwater sound tasks

CN122022415ACN 122022415 ACN122022415 ACN 122022415ACN-122022415-A

Abstract

The invention discloses a resource allocation method meeting the maximization of the number of parallel underwater acoustic tasks, and belongs to the technical field of underwater acoustic signal processing and task scheduling. The method comprises the steps of S1, task modeling and resource classification, namely establishing a resource demand model for each underwater sound task to be executed, dividing required resources into an irreplaceable resource subset and an alternative resource subset by the model, giving a comprehensive value weight to each task, S2, dynamically predicting, topologically analyzing and elastically reserving, namely predicting dynamic shortage degree of various irreplaceable resources in different future time periods based on historical task execution data, identifying key resource competition clusters based on historical data, calculating cluster comprehensive shortage indexes of the key resource competition clusters, and calculating a model by reservation comprising a smooth activation function according to the dynamic shortage degree and the cluster comprehensive shortage indexes. According to the technical scheme, intelligent, elastic and prospective management of system resources is realized through a closed-loop framework of prediction, optimization, feedback and learning.

Inventors

  • RUAN YUXING
  • QIN HANQIN
  • XIAO ZHUAN
  • LI ZENGXIN

Assignees

  • 中国舰船研究设计中心

Dates

Publication Date
20260512
Application Date
20260415

Claims (10)

  1. 1. The resource allocation method meeting the maximization of the number of the parallel underwater sound tasks is characterized by comprising the following steps of: Step S1, task modeling and resource classification, namely, performing underwater sound tasks to be executed Establishing a resource demand model that divides its required resources into non-replaceable resource subsets And alternative resource subsets And for each task Giving a weight to the comprehensive value ; Step S2, dynamic prediction, topology analysis and elastic reservation, namely predicting various irreplaceable resources in different time periods in the future based on historical task execution data Dynamic shortage of (2) Identifying key resource contention clusters based on historical data And calculates the cluster comprehensive tension index According to the dynamic shortage degree Comprehensive stress index of cluster Dynamically determining and managing an elastic reserved resource pool and a strategic buffer resource pool of each time period in the future through a reserved calculation model comprising a smooth activation function; Step S3, multi-objective optimization and dynamic preemption decision is carried out, namely taking the total value and total quantity of the tasks to be scheduled as the comprehensive optimization objective to maximize the total value and minimize the benefit loss caused by using alternative resources or insufficient resources at the scheduling moment, and integrating the dynamic real-time execution urgency of the tasks as an optimization factor; step S4, feedback tuning and experience learning according to the task Benefit results after actual execution And expected benefit Dynamically adjusting the integrated value weight Parameters of the reservation calculation model and benefit loss coefficients related to resource adaptation And simultaneously, storing the historical scheduling decision cases into a policy experience library, and carrying out case matching and policy multiplexing when making a new decision so as to continuously optimize the subsequent scheduling performance.
  2. 2. The resource allocation method for maximizing the number of parallel underwater acoustic tasks according to claim 1, wherein in said step S2, said identifying key resource competition clusters Calculating cluster comprehensive tension index And the specific steps of managing the strategic buffer resource pool include: S210, analyzing historical task data, identifying resource item combinations frequently co-occurring in non-replaceable resource subsets of a plurality of high-value tasks, defined as key resource competition clusters ; S211, aiming at each key resource competition cluster Calculate each time slot in the future Cluster integrated stress index of (2) The index comprehensively considers each resource in the cluster Individual dynamic shortage degree of (2) The weight of the associated edges between resources; S212, when a critical resource competition cluster is monitored Specific future time slots Cluster integrated stress index of (2) When the preset system level threshold value is exceeded, the system automatically triggers a cooperative reservation strategy; S213, in accordance with the dynamic shortage degree Elastic reservation determined by reservation calculation model Based on the method, a part of resources are additionally allocated from the global idle resources of the system as the key resource competition cluster Dedicated strategic buffer resources; s214, implementing access control on the strategic buffer resources to only allow those applying for the critical resource competition cluster at the same time All or most of the resources in the network, and the overall value weight thereof High priority tasks above another set threshold are used in scheduling.
  3. 3. The method for allocating resources to meet the maximization of the number of parallel underwater sound tasks according to claim 1, wherein in the step S3, the specific steps of dynamically executing urgency of real-time and executing resource preemption of the incorporated tasks include: s310 for each task Calculating dynamic real-time execution urgency, which is task remaining deadline, task waiting time and task comprehensive value weight The value of which increases dynamically as the deadline approaches and the latency increases; S311, when multi-objective optimization decision is made, the system not only considers static comprehensive value weight The dynamic real-time execution urgency is used as a key factor to be integrated into the calculation of an objective function or used as an additional optimization sub-objective; s312, the system maintains a monitoring list of allocated resources, and when a new high-urgency task cannot be immediately scheduled due to insufficient resources, a resource preemption assessment flow is started; s313, the resource preemption assessment flow evaluates the low-urgency tasks which are currently executing or are scheduled, calculates global benefit gain brought by interrupting or deferring the low-urgency tasks to release the occupied key resources, and compares the gain with system overhead and task restarting cost caused by preemption operation; And S314, if the evaluation result shows that the global net benefit is positive and exceeds a preset preemption threshold, the system executes resource preemption operation, forcibly recovers the resources occupied by part of the low-urgency tasks, reassigns the resources to the high-urgency tasks, and then performs optimal scheduling again.
  4. 4. The method for resource allocation to maximize the number of parallel underwater sound tasks according to claim 1, wherein in the step S4, the specific step of constructing the policy experience library for multiplexing with case matching includes: S410, the system not only records the microcosmic benefit deviation of task execution for parameter tuning, but also stores each complete scheduling decision period, the corresponding system state context, the adopted scheduling strategy and the finally generated macroscopic performance index into a strategy experience library as a complete scheduling case, wherein the context of each scheduling case comprises the characteristic distribution of a task queue and the resource shortage degree Pattern of (c), and critical resource contention cluster State of (2); s411, when a new scheduling decision point is encountered later, the system firstly carries out similarity matching on the current system state context and the historical cases stored in the strategy experience library; s412, if the historical cases with high similarity are matched, the system loads the historical scheduling strategy associated with the cases preferentially as an initial strategy or a reference strategy of the multi-objective optimization solution in the current step S3, and performs quick optimization fine adjustment on the basis; And S413, simultaneously, carrying out offline analysis and mining on cases in the system continuous strategy experience library, summarizing high-efficiency scheduling strategy templates under different system state modes through a machine learning method, and using the strategy templates as meta-knowledge to guide initial strategy generation under a new scene.
  5. 5. The resource allocation method for maximizing the number of parallel underwater acoustic tasks according to claim 1, wherein in said step S2, the calculation resources are calculated In the future Dynamic shortage of individual timeslots The formula of (2) is: Wherein, the For predicted future time slots For resources Is not an alternative demand total; Is a resource Is a total available amount of (2); And Respectively of historically the same period of time resources Average and standard deviation of the occupancy; For the task Is a comprehensive value weight of (1); To indicate the function, when the resource Belonging to the task Is not a subset of the alternative resources of (a) The value is 1 when the time is taken, otherwise, the value is 0; For the task In time slot Probability estimation of starting execution; is normalized weight coefficient and satisfies 。
  6. 6. The resource allocation method for maximizing the number of parallel underwater acoustic tasks according to claim 5, wherein in said step S2, the dynamic shortage is determined Computing resources In time slot Elastic reserve of (a) The formula of (2) is: Wherein, the , Is a resource And the coefficient satisfies the maximum reserve ratio coefficient of ; Is a resource A shortage threshold of (2); For the task For resources Is not an alternative demand for (a); and (5) a task set which is submitted and waits for scheduling to be executed.
  7. 7. The resource allocation method for maximizing the number of parallel underwater acoustic tasks according to claim 1, wherein in said step S3, said integrated optimization objective is obtained by the following function Expression: Wherein, the Representing tasks as binary decision variables Whether or not to be scheduled; For continuous decision variables, representing allocation to tasks Alternative resources of (a) Is the actual number of (3); For the task For alternative resources Is a reference demand amount of (a); For the task Using alternative resources Is used for adapting the benefit loss coefficient; for the task Dynamic real-time execution urgency of (2); are weight coefficients for balancing the sub-targets.
  8. 8. The resource allocation method for maximizing the number of parallel underwater sound tasks according to claim 7, wherein the constraint conditions to be satisfied in the solving process of step S3 include: First, for all scheduled tasks, i.e Task of (2) And non-replaceable resource subsets thereof Each resource in (a) Must meet And all tasks pair resources The total amount of such allocation must not exceed the total amount of the resource in the currently active elastic reserved resource pool and strategically buffered resource pool; Second, for all resources in the system The sum of the allocation amounts of all tasks must not exceed the total available amount of the resource ; Third, for alternative resources, there are 。
  9. 9. The resource allocation method for maximizing the number of parallel underwater sound tasks according to claim 1, wherein in the step S4, the adaptive benefit loss coefficient is dynamically updated based on the following formula : Wherein, the And The coefficient values after and before updating respectively; Is a task Benchmark total demand for alternative resources; is the learning rate; And Respectively tasks Expected and actual benefit metrics of (a); For the task Resources actually allocated and used Is the number of (3); is a very small positive constant used to ensure numerical stability.
  10. 10. The resource allocation method for maximizing the number of parallel underwater acoustic tasks according to claim 2, wherein in said step S211, a critical resource competition cluster is calculated In time slot Cluster integrated stress index of (2) The method comprises constructing a weighted undirected graph with intra-cluster resources as nodes and historic co-occurrence strength among the resources as edges And its stress is quantified by the following formula: Wherein, the Representing clusters The number of resources contained therein; Representation of the drawings Is a set of edges of (a); Representing resources And (3) with Historical co-occurrence intensity weights between; Is used for adjusting the relative importance of average degree of tightness and degree of tightness difference dispersion in clusters.

Description

Resource allocation method meeting maximization of number of parallel underwater sound tasks Technical Field The invention belongs to the technical field of underwater sound signal processing and task scheduling, and particularly relates to a resource allocation method capable of meeting the maximization of the number of parallel underwater sound tasks. Background In a complex underwater acoustic task environment, achieving maximization of the number of parallel tasks is a very challenging core goal. Conventional resource allocation methods, such as strategies based on simple rules or static programming, are difficult to effectively address this challenge due to their inherent drawbacks, with limitations mainly manifested in the following aspects. Firstly, modeling of resources and tasks by the traditional method is too simplified, and the real dependency relationship of the tasks on the resources cannot be accurately described. They typically allocate the various computing, communication and sensing resources required for a task as homogenous or arbitrarily replaceable units. In practice, however, the performance of many underwater acoustic tasks relies strictly on a specific set of functionally complementary and non-replaceable "core resource combinations" (e.g. a specific model of sonar array and associated processor), while allowing some alternative flexibility to other auxiliary resources. The traditional model ignores the two-way characteristic of the 'non-replaceable' and 'replaceable' resources, so that a scheduling scheme cannot start high-value tasks due to core resource conflict, or the whole energy efficiency of a system is seriously damaged due to overuse of high-cost alternative resources, and finally the effective improvement of the number of parallel tasks is limited. Secondly, the traditional scheduling mechanism lacks systematic look-ahead and pre-planning capabilities, and cannot actively avoid future resource bottlenecks. Such methods, which are typically "greedy" allocations, make decisions based on the current instantaneous availability of resources. The underwater sound task has long execution period, the resource occupation has space-time continuity, and the task usually presents regularity or burstiness. Because dynamic shortage situations of various key resources in future time periods cannot be predicted, and centralized contention risks of a plurality of tasks on the same group of tightly coupled resources (which can be called as a resource competition cluster), the system is extremely easy to sink into local optimum, core resources are exhausted prematurely, access of subsequent key tasks is blocked, and the throughput of parallel tasks is reduced from the global and long-term viewpoints. Furthermore, the conventional method has insufficient optimization criteria and dynamic adjustment capability. They tend to seek immediate maximization of a single index (e.g., number of tasks), failing to make comprehensive trade-offs and dynamic optimizations between task value, urgency, resource replacement costs, and long-term system benefits. When high-value urgent tasks arrive, there is a lack of efficient dynamic preemption and rescheduling mechanisms based on priority and benefit loss assessment, response delays. Meanwhile, the preset static parameters cannot adapt to the change of the environment and the task mode, so that the scheduling strategy is gradually invalid, and the robustness is poor. Therefore, in order to break through the bottleneck, a new resource allocation method is needed. Disclosure of Invention In view of the above, the present invention aims to provide a resource allocation method for maximizing the number of parallel underwater sound tasks, which is used for realizing intelligent, elastic and prospective management of system resources through a closed-loop framework of prediction-optimization-feedback-learning. In order to achieve the above purpose, the present invention provides the following technical solutions: A resource allocation method meeting the maximization of the number of parallel underwater sound tasks comprises the following steps: Step S1, task modeling and resource classification, namely, performing underwater sound tasks to be executed Establishing a resource demand model that divides its required resources into non-replaceable resource subsetsAnd alternative resource subsetsAnd for each taskGiving a weight to the comprehensive value; Step S2, dynamic prediction, topology analysis and elastic reservation, namely predicting various irreplaceable resources in different time periods in the future based on historical task execution dataDynamic shortage of (2)Identifying key resource contention clusters based on historical dataAnd calculates the cluster comprehensive tension indexAccording to the dynamic shortage degreeComprehensive stress index of clusterDynamically determining and managing an elastic reserved resource pool and a strategic buffer reso