A resource allocation method for satisfying maximum number of parallel underwater acoustic tasks

Abstract

This invention discloses a resource allocation method that maximizes the number of parallel underwater acoustic tasks, belonging to the field of underwater acoustic signal processing and task scheduling technology. It includes the following steps: Step S1, Task Modeling and Resource Classification: A resource requirement model is established for each underwater acoustic task to be executed. This model divides the required resources into a subset of irreplaceable resources and a subset of replaceable resources; and a comprehensive value weight is assigned to each task. Step S2, Dynamic Prediction, Topology Analysis, and Flexible Reservation: Based on historical task execution data, the dynamic scarcity of various irreplaceable resources in different future time periods is predicted; key resource competition clusters are identified based on historical data, and their cluster comprehensive tension index is calculated; based on the dynamic scarcity and cluster comprehensive tension index, a reservation calculation model containing a smooth activation function is used. This technical solution achieves intelligent, flexible, and forward-looking management of system resources through a closed-loop framework of "prediction-optimization-feedback-learning".

Description

Technical Field

[0001] This invention belongs to the field of underwater acoustic signal processing and task scheduling technology, and specifically relates to a resource allocation method that maximizes the number of parallel underwater acoustic tasks. Background Technology

[0002] In complex underwater acoustic mission environments, maximizing the number of parallel tasks is a highly challenging core objective. Traditional resource allocation methods, such as strategies based on simple rules or static programming, are inherently flawed and unable to effectively address this challenge. Their limitations are mainly reflected in the following aspects.

[0003] First, traditional methods oversimplify resource and task modeling, failing to accurately depict the true resource dependencies of tasks. They typically allocate the various computing, communication, and sensing resources required by a task as homogeneous or arbitrarily substitutable units. However, in reality, the execution of many underwater acoustic tasks strictly relies on a specific set of complementary and irreplaceable "core resource combinations" (e.g., a specific sonar array and its accompanying processor), while allowing for a degree of substitutability for other auxiliary resources. Traditional models ignore this dichotomy between "irreplaceable" and "substitutable" resources, leading to scheduling schemes either failing to initiate high-value tasks due to core resource conflicts or severely compromising overall system energy efficiency due to the overuse of high-cost alternative resources, ultimately limiting the effective increase in the number of parallel tasks.

[0004] Secondly, traditional scheduling mechanisms lack systematic foresight and pre-planning capabilities, failing to proactively avoid future resource bottlenecks. These methods often make decisions based on the immediate availability of resources, representing a typical "greedy" allocation. Underwater acoustic tasks have long execution cycles, resource consumption exhibits spatiotemporal continuity, and task flows often display regularity or suddenness. Due to the inability to predict the dynamic scarcity of various key resources in the future, and the risk of concentrated contention among multiple tasks for the same tightly coupled set of resources (which can be called a "resource contention cluster"), the system is highly susceptible to getting trapped in local optima, prematurely exhausting core resources, and thus blocking the access of subsequent key tasks. From a global and long-term perspective, this actually reduces the throughput of parallel tasks.

[0005] Furthermore, traditional methods lack sufficient optimization criteria and dynamic adjustment capabilities. They often pursue the immediate maximization of a single indicator (such as the number of tasks), failing to comprehensively weigh and dynamically optimize between task value, urgency, resource substitution 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, resulting in sluggish response. Simultaneously, pre-set static parameters cannot adapt to changes in the environment and task patterns, causing scheduling strategies to gradually become ineffective and exhibiting poor robustness. Therefore, to overcome these bottlenecks, a novel resource allocation method is urgently needed. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a resource allocation method that maximizes the number of parallel underwater acoustic tasks, so as to achieve intelligent, flexible and forward-looking management of system resources through a closed-loop framework of "prediction-optimization-feedback-learning".

[0007] To achieve the above objectives, the present invention provides the following technical solution: A resource allocation method that maximizes the number of parallel underwater acoustic tasks includes the following steps: Step S1, Task Modeling and Resource Classification: For each underwater acoustic task to be performed... Establish a resource demand model that divides the required resources into a subset of non-substitutable resources. and alternative resource subsets ; and for each task Assign a comprehensive value weight ; Step S2, Dynamic Prediction, Topology Analysis, and Elastic Reservation: Based on historical task execution data, predict various irreplaceable resources in different future time periods. Dynamic scarcity Identifying key resource competition clusters based on historical data And calculate its cluster comprehensive tension index. According to the dynamic scarcity level Cluster Comprehensive Tension Index By using a reserved computation model that includes a smooth activation function, the elastic reserved resource pool and strategic buffer resource pool for each future time period are dynamically determined and managed. Step S3, Multi-objective optimization and dynamic preemption decision-making: At the scheduling time, the comprehensive optimization objective is to maximize the total value and total number of scheduled tasks and minimize the benefit loss caused by the use of alternative resources or insufficient resources. At the same time, the dynamic real-time execution urgency of the tasks is incorporated as an optimization factor. Under the hard constraint that all tasks must obtain all irreplaceable resources from the corresponding time period's elastic reserved resource pool and strategic buffer resource pool, the current optimal task execution set and resource allocation scheme are solved. When high-urgency tasks cannot be scheduled due to insufficient resources, resource preemption for low-urgency tasks is evaluated and executed. Step S4, Feedback Tuning and Experience Learning: Based on the task The actual benefits after implementation Compared with expected benefits The deviation is dynamically adjusted to determine the comprehensive value weight. The parameters of the reserved calculation model and the benefit loss coefficients related to resource adaptation. Meanwhile, historical scheduling decision cases are stored in the strategy experience base, and case matching and strategy reuse are performed when making new decisions to continuously optimize subsequent scheduling performance.

[0008] Furthermore, in step S2, the identification of key resource competition clusters... Calculate the cluster comprehensive tension index The specific steps for managing the strategic buffer resource pool include: S210: Analyze historical task data to identify combinations of resource items that frequently co-occur in irreplaceable resource subsets across multiple high-value tasks, defining these as critical resource competition clusters. ; S211: For each critical resource contention cluster Calculate its future time slots Cluster comprehensive tension index This index comprehensively considers the resources within the cluster. Individual dynamic scarcity And the weights of the edges connecting resources; S212: When a cluster of critical resources is detected to be in contention. In a specific future time slot Cluster comprehensive tension index When the preset system-level threshold is exceeded, the system automatically triggers the collaborative reservation strategy; S213: Based on the aforementioned dynamic scarcity level The elastic reserve amount determined by the reserved calculation model In addition, a portion of resources is allocated from the system's global idle resources for this critical resource contention cluster. Dedicated strategic buffer resources; S214: Implement access control on the strategic buffer resource, allowing only those concurrently requesting the critical resource cluster to access it. All or most of the resources, and their overall value weight High-priority tasks that exceed another set threshold are used during scheduling.

[0009] Furthermore, in step S3, the specific steps for integrating the dynamic real-time execution urgency of the task and the preemption of execution resources include: S310: For each task Calculate the dynamic real-time execution urgency, which is determined by the remaining deadline, the time already elapsed, and the overall value weight of the task. The function's value increases dynamically as the deadline approaches and the waiting time lengthens; S311: When making multi-objective optimization decisions, the system considers not only the static comprehensive value weights. Furthermore, this dynamic real-time execution urgency will be incorporated as a key factor into the calculation of the objective function or as an additional optimization sub-objective; S312: The system maintains a monitoring list of allocated resources. When a new high-urgency task cannot be scheduled immediately due to insufficient resources, a resource preemption assessment process is initiated. S313: The resource preemption assessment process assesses the low-urgency tasks that are currently being executed or have been scheduled, calculates the global benefit gain that could be brought about by interrupting or postponing these low-urgency tasks to release the critical resources they occupy, and compares this gain with the system overhead and task restart cost caused by the preemption operation. S314: If the evaluation result shows that the global net benefit is positive and exceeds the preset preemption threshold, the system performs a resource preemption operation, forcibly reclaims some of the resources occupied by low-urgency tasks, redistributes them to the high-urgency tasks, and then re-optimizes the scheduling.

[0010] Furthermore, in step S4, the specific steps for constructing the strategy experience base and reusing case matching include: S410: The system not only records the micro-benefit deviations of task execution for parameter tuning, but also stores each complete scheduling decision cycle, along with its corresponding system state context, the scheduling strategy adopted, and the final macro-performance indicators, as a complete scheduling case in a strategy experience base. The context of each scheduling case includes the characteristic distribution of the task queue and resource scarcity. The model and the cluster of key resource competition The state; S411: When encountering a new scheduling decision point, the system first performs a similarity match between the current system state context and the historical cases stored in the policy experience base. S412: If a highly similar historical case is matched, the system will first load the historical scheduling strategy associated with the case as the initial strategy or reference strategy for the current step S3 multi-objective optimization solution, and then perform rapid optimization and fine-tuning based on this strategy. S413: At the same time, the system continuously performs offline analysis and mining of cases in the strategy experience base, summarizes efficient scheduling strategy templates under different system state modes through machine learning methods, and uses these strategy templates as meta-knowledge to guide the generation of initial strategies in new scenarios.

[0011] Furthermore, in step S2, computing resources In the future Dynamic scarcity of each time slot The formula is: in, For predicted future time slots Resources The total irreplaceable demand; For resources Total available quantity; and Resources from the same historical period Average and standard deviation of occupancy; For the task The overall value weight; For indicator functions, when resources It belongs to the task Irreplaceable subset of resources The value is 1 if it is true, and 0 otherwise. For the task In the time slot Probability estimate of when execution will begin; These are normalized weight coefficients, and satisfy... .

[0012] Furthermore, in step S2, based on the dynamic scarcity level... Computing resources In the time slot Flexible allowance The formula is: in, , For resources The maximum reserve ratio coefficient, and this coefficient satisfies ; For resources The scarcity threshold; For the task Resources The irreplaceable demand; This is the set of tasks that have been submitted and are waiting to be scheduled for execution.

[0013] Furthermore, in step S3, the comprehensive optimization objective is determined by the following function. Express: in, Let be a binary decision variable, representing the task. Whether it is scheduled; For continuous decision variables, it represents the variables assigned to the task. Alternative resources The actual quantity; For the task For alternative resources The baseline demand; For the task Use alternative resources The adaptation benefit loss coefficient; For the task Dynamic real-time execution urgency; These are the weighting coefficients used to balance the various sub-objectives.

[0014] Furthermore, the constraints that the solution process in step S3 must satisfy include: First, for all scheduled tasks, i.e. Task and its irreplaceable resource subset Each resource in Must meet And all tasks require resources The total amount of such allocations shall not exceed the total amount of such resources in the currently active elastic reserved resource pool and strategic buffer resource pool; Second, for all resources in the system The sum of the allocations to all tasks must not exceed the total available amount of that resource. ; Third, regarding alternative resources, there are .

[0015] Furthermore, in step S4, the adaptation benefit loss coefficient is dynamically updated based on the following formula. : in, and These are the coefficient values ​​before and after the update, respectively. For the task The baseline total demand for alternative resources; The learning rate; and Tasks Expected and actual benefits metrics; For the task Resources actually allocated and used Quantity; It is a very small positive number used to ensure numerical stability.

[0016] Furthermore, in step S211, the critical resource contention cluster is calculated. In the time slot Cluster comprehensive tension index The method is as follows: Construct a weighted undirected graph with intra-cluster resources as nodes and historical co-occurrence strength between resources as edges. And its level of tension is quantified using the following formula: in, Cluster The number of resources contained therein; Representation diagram The set of edges; Representing resources and Historical co-occurrence intensity weights between them; It is a balancing coefficient used to adjust the relative importance of the average scarcity level within a cluster and the dispersion of scarcity differences.

[0017] The beneficial effects of this invention are as follows: Foresight and resilience: By using dynamic scarcity forecasting and flexible reservation mechanisms, we transform passive response into proactive planning, smooth out peak resource demand, and improve the system's ability to cope with unexpected tasks; Collaborative Resource Management: It innovatively proposes the concept of "critical resource competition cluster" and a tension index model based on weighted undirected graphs, which can accurately identify and prevent the "weakest link effect" and "imbalance risk" of resource combinations. Through strategic buffer resource pools, it ensures the resource supply for high-value composite tasks and greatly improves the availability of combined resources. Multi-objective intelligent optimization: A comprehensive objective function integrating task value, quantity, resource substitution cost and dynamic urgency was constructed, and a fast solution algorithm linked with the experience base was designed to achieve real-time optimization of the system's comprehensive benefits under the premise of satisfying complex constraints; Strong adaptability and learning ability: Through parameter tuning based on benefit deviation and case experience learning, a complete "decision-execution-feedback-improvement" closed loop is formed, enabling the system to continuously adapt to environmental changes and task mode evolution, and the scheduling performance to continuously optimize itself. Dynamic preemption and rescheduling: A resource preemption mechanism based on global net benefit assessment is introduced, which can flexibly respond to the emergency insertion of high-urgency tasks while ensuring the optimal overall system benefits, thereby enhancing the robustness and responsiveness of the system.

[0018] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0019] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1 This is a schematic diagram of the overall process of the resource allocation method of the present invention; Figure 2 This is a flowchart illustrating step S2 in the resource allocation method of the present invention; Figure 3 This is a flowchart illustrating step S3 in the resource allocation method of the present invention. Figure 4 This is a flowchart illustrating step S4 in the resource allocation method of the present invention. Detailed Implementation

[0020] like Figures 1-4 As shown, the present invention provides a resource allocation method that maximizes the number of parallel underwater acoustic tasks, comprising the following steps: Step S1, Task Modeling and Resource Classification: For each underwater acoustic task to be performed... Establish a resource demand model that divides the required resources into a subset of non-substitutable resources. and alternative resource subsets ; and for each task Assign a comprehensive value weight ; This step forms the data foundation of the entire method, and its core lies in providing data for each underwater acoustic task. Construct a structured resource requirement model that not only lists the required resources, but more importantly, divides the resource set into "irreplaceable resource subsets" based on the irreplaceability of resources in relation to the quality of task completion. "and alternative resource subsets" At the same time, each task is assigned a comprehensive value weight. This weight can be calculated comprehensively based on dimensions such as mission priority, the military or scientific value of the expected results, and the authority of the mission's source, and is used to quantify the importance of the mission.

[0021] Step S2, Dynamic Prediction, Topology Analysis, and Elastic Reservation: Based on historical task execution data, predict various irreplaceable resources in different future time periods. Dynamic scarcity Identifying key resource competition clusters based on historical data And calculate its cluster comprehensive tension index. According to the dynamic scarcity level Cluster Comprehensive Tension Index By using a reserved computation model that includes a smooth activation function, the elastic reserved resource pool and strategic buffer resource pool for each future time period are dynamically determined and managed. This step is a proactive resource security measure. It first uses historical data to predict future time slots. For each irreplaceable resource Dynamic scarcity More importantly, by analyzing historical task execution records, we can identify resource combinations that are frequently demanded by multiple high-value tasks simultaneously, namely, "critical resource competition clusters." "and calculate an index reflecting the overall tension of the cluster." Subsequently, using a reserved computational model that includes a smooth activation function, based on... and The system dynamically calculates and sets up two types of resource pools for future periods: an "elastic reserved resource pool" to cope with general resource shortages, and a "strategic buffer resource pool" specifically to resolve competitive crises in key resource combinations.

[0022] Step S3, Multi-objective optimization and dynamic preemption decision-making: At the scheduling time, the comprehensive optimization objective is to maximize the total value and total number of scheduled tasks and minimize the benefit loss caused by the use of alternative resources or insufficient resources. At the same time, the dynamic real-time execution urgency of the tasks is incorporated as an optimization factor. Under the hard constraint that all tasks must obtain all irreplaceable resources from the corresponding time period's elastic reserved resource pool and strategic buffer resource pool, the current optimal task execution set and resource allocation scheme are solved. When high-urgency tasks cannot be scheduled due to insufficient resources, resource preemption for low-urgency tasks is evaluated and executed. This step involves the real-time scheduling decision engine. At each scheduling moment, the system constructs a multi-objective optimization problem in the form of mathematical programming. Its objective function Z aims to maximize the weighted total value of the selected tasks in the current waiting queue. With total quantity At the same time, minimize the loss of benefits caused by the inability to allocate sufficient alternative resources. ,in, The dynamic real-time execution urgency of tasks is used to assign higher value weights to tasks with high urgency in the optimization process. Furthermore, the dynamic real-time execution urgency of tasks is incorporated as a key factor into the optimization process. Solving this optimization problem requires satisfying a core hard constraint: all irreplaceable resources required by all tasks must be fully and accurately obtained from the "elastic reserved resource pool" and "strategic buffer resource pool" pre-established for their corresponding time periods in step S2. If, after optimization, high-urgency tasks still cannot be scheduled due to resource constraints, an evaluation process is triggered to calculate the global net benefit of releasing resources by preempting (interrupting or postponing) scheduled low-urgency tasks. If the net benefit is significant, preemption and rescheduling are executed.

[0023] Step S4, Feedback Tuning and Experience Learning: Based on the task The actual benefits after implementation Compared with expected benefits The deviation is dynamically adjusted to determine the comprehensive value weight. The parameters of the reserved calculation model and the benefit loss coefficients related to resource adaptation. Meanwhile, historical scheduling decision cases are stored in the strategy experience base, and case matching and strategy reuse are performed when making new decisions to continuously optimize subsequent scheduling performance.

[0024] This step is crucial for enabling the system's self-evolution, and it involves two levels: first, micro-parameter tuning, which is based on the actual performance benefits of the task. Compared with expected benefits The deviation will be used to adjust the overall value weight of the task in reverse. Reserved model parameters and resource adaptation benefit loss coefficient First, it makes the model more realistic; second, it involves macro-level strategy learning, where each complete scheduling decision (including system state, strategy used, and execution result) is stored as a "case" in a strategy experience base. When facing new scheduling problems in the future, successful strategies can be directly reused or quickly adjusted by matching similar historical cases, thereby achieving continuous accumulation and optimization of decision-making capabilities.

[0025] The technical effects achieved by the above technical solution are as follows: Forming a complete intelligent closed loop: Through a closed-loop architecture of "modeling -> prediction -> optimization -> feedback," static and passive scheduling is transformed into dynamic, proactive, and adaptive intelligent resource governance, systematically solving scheduling challenges in complex environments; Achieving refined resource management: By distinguishing between irreplaceable and substitutable resources and coordinating the management of key resource combinations, the failure of high-value tasks due to resource fragmentation and "combination shortcomings" is greatly reduced, improving the overall utilization efficiency of scarce resources; Ensuring optimal comprehensive decision-making: Integrating a multi-objective optimization model that considers value, quantity, urgency, and substitution costs, scheduling decisions can achieve real-time optimization of the system's overall benefits while meeting complex constraints; Enhancing system resilience and response speed: The dynamic preemption mechanism provides emergency measures in resource stalemates, while the case-based learning mechanism accelerates future decisions, enabling the system to possess both the flexibility to cope with emergencies and the learning ability to improve long-term operational efficiency.

[0026] Furthermore, in step S2, the identification of key resource competition clusters... Calculate the cluster comprehensive tension index The specific steps for managing the strategic buffer resource pool include: S210: Analyze historical task data to identify combinations of resource items that frequently co-occur in irreplaceable resource subsets across multiple high-value tasks, defining these as critical resource competition clusters. ; By analyzing long-term historical data using data mining techniques, we can identify those data points that frequently appear simultaneously in multiple high-value tasks. (Higher) Non-substitutable resource subset The combination of resources. For example, resources (Dedicated processor) and resources (High-speed data links) may be frequently demanded, then Identified as a cluster of critical resource competition .

[0027] S211: For each critical resource contention cluster Calculate its future time slots Cluster comprehensive tension index This index comprehensively considers the resources within the cluster. Individual dynamic scarcity And the weights of the edges connecting resources; For quantization clusters The overall tension is used to construct a system based on the resources within the cluster. For nodes, based on the historical co-occurrence strength between resource pairs Weighted undirected graph with edge weights Tension Index The calculation formula contains two terms: the first term is the dynamic scarcity of all resources within the cluster. The first term is the arithmetic mean, reflecting overall demand pressure; the second term is the key point, which calculates the arithmetic mean of each pair of connected resources within the cluster. , The absolute value of the scarcity difference And use its co-occurrence strength weight A weighted average is used, which specifically measures the risk of mismatch or "imbalance" in demand among resources within a cluster. Weights This ensures that the closer the historical links between resource pairs are, the greater the "contribution" of their demand differences to overall tension.

[0028] S212: When a cluster of critical resources is detected to be in contention. In a specific future time slot Cluster comprehensive tension index When the preset system-level threshold is exceeded, the system automatically triggers the collaborative reservation strategy; The system continuously monitors each cluster. When the index of a cluster exceeds a preset threshold based on system capacity and task success rate targets, it indicates that the resource combination faces high risk, and a higher-level collaborative reservation strategy is automatically triggered.

[0029] S213: Based on the aforementioned dynamic scarcity level The elastic reserve amount determined by the reserved calculation model In addition, a portion of resources is allocated from the system's global idle resources for this critical resource contention cluster. Dedicated strategic buffer resources; S214: Implement access control on the strategic buffer resource, allowing only those concurrently requesting the critical resource cluster to access it. All or most of the resources, and their overall value weight High-priority tasks that exceed another set threshold are used during scheduling.

[0030] Once triggered, the system allocates resources from globally unreserved idle resources to this specific critical resource cluster. An additional portion of resources will be allocated to establish a "strategic buffer resource pool." These resources will be strictly controlled and will only be available to tasks that meet two conditions: first, the irreplaceable resources requested by the task must cover the competing cluster of the critical resource. The first factor is the total or most of the resources in the task; the second is the overall value weight of the task itself. It must be above another, higher threshold. This ensures that strategic buffer resources are only used to "rescue" critical tasks that are extremely valuable to the system and are indeed subject to competition for resources within that cluster.

[0031] Step S2 specifies the management process for the "critical resource competition cluster," revealing an innovative mechanism for quantifying and responding to resource coordination tension. This mechanism achieves the following technical effects: Accurately identifying and preventing the "weakest link" effect: Traditional resource management cannot anticipate the structural risk of "sufficient resource A but insufficient resource B, causing the entire task group to be unable to execute." This solution, by calculating the weighted imbalance, can accurately quantify this risk and intervene in advance when the risk exceeds the limit; Significantly improving the guarantee rate of high-value composite tasks: By establishing dedicated and controlled strategic buffers for identified critical resource combinations, it is equivalent to establishing a "resource green channel" for the most important task flows, directly improving the success rate and timeliness of the system in completing high-end, composite tasks; Upgrading resource reservation from "statistics" to "topology": Introducing graph theory into resource management upgrades the reservation strategy from statistical prediction for individual resources to structural protection of the network of relationships between resources, making resource allocation more forward-looking and intelligent.

[0032] Furthermore, in step S3, the specific steps for integrating the dynamic real-time execution urgency of the task and the preemption of execution resources include: S310: For each task Calculate the dynamic real-time execution urgency, which is determined by the remaining deadline, the time already elapsed, and the overall value weight of the task. The function's value increases dynamically as the deadline approaches and the waiting time lengthens; For each task in the queue Define a dynamic, real-time execution urgency function. This function typically uses the remaining deadline of the task as the urgency level. The primary variable is defined by negative correlations, forming the base component, and is then superimposed with the waiting time. Positively correlated penalty terms, multiplied by the task's value weight. As a coefficient, i.e. This causes the urgency of a task to increase non-linearly and rapidly over time, and it can reflect the differences in the urgency of tasks with different values.

[0033] S311: When making multi-objective optimization decisions, the system considers not only the static comprehensive value weights. Furthermore, this dynamic real-time execution urgency will be incorporated as a key factor into the calculation of the objective function or as an additional optimization sub-objective; When making multi-objective optimization decisions, this dynamic urgency should be considered. As a key optimization factor, it can be implemented by adding it as an additional sub-objective to the objective function. (like Alternatively, it can be used as a coefficient to adjust the original target term.

[0034] S312: The system maintains a monitoring list of allocated resources. When a new high-urgency task cannot be scheduled immediately due to insufficient resources, a resource preemption assessment process is initiated. S313: The resource preemption assessment process assesses the low-urgency tasks that are currently being executed or have been scheduled, calculates the global benefit gain that could be brought about by interrupting or postponing these low-urgency tasks to release the critical resources they occupy, and compares this gain with the system overhead and task restart cost caused by the preemption operation. S314: If the evaluation result shows that the global net benefit is positive and exceeds the preset preemption threshold, the system performs a resource preemption operation, forcibly reclaims some of the resources occupied by low-urgency tasks, redistributes them to the high-urgency tasks, and then re-optimizes the scheduling.

[0035] The system maintains a real-time status list of all allocated resources. When a new, high-urgency task cannot be accommodated by the current optimal scheduling scheme due to insufficient resources, the system initiates a preemption evaluation process. This process iterates through currently executing or resource-allocated low-urgency tasks, simulating calculations: if one or more low-urgency tasks are interrupted or postponed, releasing their occupied critical resources and reassigning them to high-urgency tasks, the entire system's performance in the objective function... The system calculates the global benefit gain; at the same time, it calculates the "overhead" and "cost" of the preemption operation itself. The system only approves the execution of the preemption operation when "global benefit gain - overhead - cost > preset preemption threshold". After execution, the system immediately re-runs the optimization process of step S3 based on the new resource state to generate an updated scheduling scheme.

[0036] Step S3 defines how to enhance the responsiveness and decision-making flexibility of the scheduling system through quantified urgency and controlled preemption mechanisms, achieving the following technical effects: It significantly enhances the responsiveness to urgent tasks: Through a dynamic urgency model and preemption mechanism, the system can effectively handle common sudden high-priority instructions or emergency detection tasks in underwater acoustic environments, ensuring that critical moments are not delayed; it achieves "rational flexibility" in decision-making: preemption is not forced or blind, but based on a rigorous global net benefit assessment; this allows the system to proactively break local optima and seek a better global solution when resources are in absolute conflict, while avoiding system oscillations and the "starvation" of low-priority tasks caused by frequent or improper preemption; and it balances short-term emergency response with long-term stability: by setting a "preemption threshold," system administrators can find an adjustable balance between "responding to urgent needs as much as possible" and "maintaining the stability of the scheduling plan and protecting the rights of ordinary tasks."

[0037] Furthermore, in step S4, the specific steps for constructing the strategy experience base and reusing case matching include: S410: The system not only records the micro-benefit deviations of task execution for parameter tuning, but also stores each complete scheduling decision cycle, along with its corresponding system state context, the scheduling strategy adopted, and the final macro-performance indicators, as a complete scheduling case in a strategy experience base. The context of each scheduling case includes the characteristic distribution of the task queue and resource scarcity. The model and the cluster of key resource competition The state; The system not only records the micro-data of each task execution, but more importantly, it encapsulates each complete scheduling decision cycle into a structured "scheduling case". A case contains at least three parts of information: (1) System state context: the characteristics of the task queue at the time of decision, the scarcity of all resources. Map, competition clusters of key resources state ( (1) Value); (2) The scheduling strategy adopted; (3) Macro performance indicators: the overall performance achieved by the system after the end of the scheduling cycle.

[0038] S411: When encountering a new scheduling decision point, the system first performs a similarity match between the current system state context and the historical cases stored in the policy experience base. S412: If a highly similar historical case is matched, the system will first load the historical scheduling strategy associated with the case as the initial strategy or reference strategy for the current step S3 multi-objective optimization solution, and then perform rapid optimization and fine-tuning based on this strategy. S411-S412 (Online Matching and Reuse): When faced with a new scheduling decision point, the system extracts the current "system state context" and performs similarity matching with the contexts of all historical cases in the experience base. If a highly similar historical case is matched, the system prioritizes loading the historical scheduling strategy corresponding to that case. In the subsequent optimization solution (step S3), this strategy can serve as a high-quality initial solution for the hybrid heuristic optimization algorithm or as a reference solution for the constraint satisfaction problem, thereby greatly reducing the search space.

[0039] S413: At the same time, the system continuously performs offline analysis and mining of cases in the strategy experience base, summarizes efficient scheduling strategy templates under different system state modes through machine learning methods, and uses these strategy templates as meta-knowledge to guide the generation of initial strategies in new scenarios.

[0040] During system idle periods, a machine learning algorithm runs in the background to perform offline analysis on a large number of cases in the experience base. The goal is to summarize a "strategy template" with generalization capabilities, such as: "If the system state satisfies: task queue has a high density of type A tasks and resources..." scarcity And key resource competition clusters Cluster comprehensive tension index If the value is at a high level, the recommended scheduling strategy is to appropriately increase the value weight coefficient in the multi-objective optimization function. Prioritize tasks containing resources Y or Z, and activate the strategic buffer resource pool corresponding to that cluster; these templates, as "meta-knowledge," can be used to guide the generation of initial strategies in entirely new or poorly matched scenarios.

[0041] Step S4 explains how to build and utilize a "strategy experience base" to enable the system to learn from history and accelerate future decision-making. The technical effects achieved are as follows: Significantly accelerates real-time decision-making: By reusing successful strategies from similar historical scenarios as the starting point for optimization, large-scale searches from scratch are avoided for each decision, greatly shortening the solution time and meeting the stringent real-time requirements of underwater acoustic task scheduling; Enables the accumulation and inheritance of decision-making wisdom: Transforming the experience of individuals or teams into storable and reusable digital assets for the system avoids experience loss, allowing the system's scheduling level to be stably maintained at a high level and continuously improved with the accumulation of cases; Enhances adaptability to new scenarios: Through strategy templates generated by offline learning, the system is provided with the ability to "reason" for dealing with new scenarios that have not been fully encountered before, enhancing the generalization and robustness of the method.

[0042] Furthermore, in step S2, computing resources In the future Dynamic scarcity of each time slot The formula is: in, For predicted future time slots Resources The total irreplaceable demand; For resources Total available quantity; and Resources from the same historical period Average and standard deviation of occupancy; For the task The overall value weight; For indicator functions, when resources It belongs to the task Irreplaceable subset of resources The value is 1 if it is true, and 0 otherwise. For the task In the time slot Probability estimate of when execution will begin; These are normalized weight coefficients, and satisfy... .

[0043] First item: This reflects the pressure on absolute future demand. It is a resource In the time slot The total forecast of irreplaceable demand. It is a resource Total physical availability. Second item: This reflects the uncertainty or volatility risk of historical demand. and These are resources from the same historical period. The average and standard deviation of occupancy This is the coefficient of variation (CV). Third term: This term reflects the weighted demand probability of high-value tasks. The numerator is the "weighted sum of the demand probabilities of all high-value tasks for this resource," and the denominator is used for normalization. Weighting coefficients. It must meet the normalization condition (sum of 1).

[0044] The technical effects achieved by this mathematical model are as follows: Comprehensive and accurate perception of scarcity: This model surpasses simple methods based solely on demand forecasting, integrating the three dimensions of "quantity," "stability," and "quality," thus more realistically reflecting the complex pressures faced by resources; Risk perception guidance: The second aspect introduces volatility risk, making the system more conservative during periods of historically unstable demand, reserving more resources to mitigate uncertainty; Value alignment: The third aspect directly links scarcity calculation with the system's core value objectives, ensuring that resource scarcity assessment serves the overall value maximization strategy and guiding reservation strategies towards ensuring high-value tasks.

[0045] Furthermore, in step S2, based on the dynamic scarcity level... Computing resources In the time slot Flexible allowance The formula is: in, , For resources The maximum reserve ratio coefficient, and this coefficient satisfies ; For resources The scarcity threshold; For the task Resources The irreplaceable demand; This is the set of tasks that have been submitted and are waiting to be scheduled for execution.

[0046] Calculate the scarcity offset: ,here It is for resources Preset scarcity threshold; Sigmoid response: Input the offset into the Sigmoid function to obtain a response factor between 0 and 1; Calculate theoretical reserve: Compare this response factor with resources Maximum Reserve Ratio Coefficient ( ) and all currently pending scheduling tasks ( Total non-substitutable demand for this resource Multiplication; Upper limit truncation: through The function ensures the final determined flexibility reserve. No more than resources Total physical availability .

[0047] The technical effects achieved by this model are as follows: It achieves smooth, adaptive reserved adjustment: using the Sigmoid function avoids issues at the threshold. Nearby due to Minor fluctuations lead to reserved amounts The dramatic changes in resource reservation make the changes continuous and smooth, enhancing system stability; the reservation strategy is clearly defined; the model parameters have clear meanings. The trigger point for significantly increasing the reserved space was defined; A maximum reservation ratio is defined. This allows administrators to intuitively configure and optimize based on resource characteristics and business needs; and prevents over-reservation. The product of the sum and total demand, and the final... The function sets multiple safety limits for the reserved amount, effectively preventing excessive reservation of a single resource from crowding out other resource space.

[0048] Furthermore, in step S3, the comprehensive optimization objective is determined by the following function. Express: in, Let be a binary decision variable, representing the task. Whether it is scheduled; For continuous decision variables, it represents the variables assigned to the task. Alternative resources The actual quantity; For the task For alternative resources The baseline demand; For the task Use alternative resources The adaptation benefit loss coefficient; For the task The dynamic real-time execution urgency (obtained from step 310); These are the weighting coefficients used to balance the various sub-objectives.

[0049] : Binary decision variable (0 or 1), representing the task Whether it is selected to be executed in the current scheduled batch; Continuous decision variables (≥0) represent those actually assigned to tasks. Alternative resources The quantity.

[0050] The objective function Z (usually maximized), the first term: This aims to maximize the weighted total value of scheduled tasks, including urgency. As a value-weighted factor; the second item: This item aims to maximize the total number of scheduled tasks; the third item: This aims to minimize the loss of total benefits due to the lack of alternative resources.

[0051] The technical effects achieved by this computational model are as follows: It quantifies and solves multi-objective decisions: unifying multiple originally qualitative and potentially conflicting business objectives into a single, mathematically optimizeable objective function, making the optimal scheduling scheme calculable; and it provides a flexible strategic configuration interface: by adjusting... System administrators can easily change the preferences of scheduling strategies, enabling the same system to adapt to different task modes and command intentions; the cost of resource substitution is clearly defined: by introducing... The third point explicitly acknowledges in the optimization model that using non-ideal alternative resources will lead to performance losses, and requires that while pursuing value and quantity, this loss must be weighed to obtain a scheduling scheme that better reflects actual business experience.

[0052] Furthermore, the constraints that the solution process in step S3 must satisfy include: First, for all scheduled tasks, i.e. Task and its irreplaceable resource subset Each resource in Must meet And all tasks require resources The total amount of such allocations shall not exceed the total amount of such resources in the currently active elastic reserved resource pool and strategic buffer resource pool; Sufficient allocation requirement: For any task scheduled for execution (i.e. (for its irreplaceable resource subset) Each resource in Must meet .

[0053] Source and total limits: All scheduled tasks require a specific, non-replaceable resource. The sum of such full allocations must not exceed the future time slot corresponding to the resource at the current scheduling moment. The sum of the total amount of this resource in the established "elastic reserved resource pool" and "strategic buffer resource pool".

[0054] For all resources in the system The sum of the allocations to all tasks must not exceed the total available amount of that resource. ; For all resources in the system All tasks The actual amount allocated to it The sum of these must not exceed the total physical availability of the resource. Third, regarding alternative resources, there are .

[0055] Non-negativity and upper limit: For substitute resources, Logical dependencies: This means that only when the task... Scheduled ( Only when this condition is met can alternative resources be allocated to it.

[0056] The solution is absolutely feasible: These three types of constraints together ensure that any set of constraints found by the optimization algorithm is feasible. Each value corresponds to a resource allocation scheme that is physically executable and logically self-consistent; solidifying the "prediction-reservation-allocation" workflow: Constraint 1 (especially its second part) is the "rivet" of the logical chain of the method of this invention; it forces that allocation must be based on reservation, making the prediction and reservation work of S2 an insurmountable prerequisite for the optimization decision of S3; providing a foundation for accurate modeling: Constraint 3 clearly describes the dependency relationship between decision variables, which is the basis for correctly constructing mixed integer programming (MIP) or other optimization models.

[0057] Furthermore, in step S4, the adaptation benefit loss coefficient is dynamically updated based on the following formula. : in, and These are the coefficient values ​​before and after the update, respectively. For the task The baseline total demand for alternative resources; The learning rate (dimensionless); and Tasks Expected and actual benefits metrics; For the task Resources actually allocated and used Quantity; It is a very small positive number used to ensure numerical stability (not zero 0).

[0058] Update driver signals: This refers to the deviation between the expected and actual benefits of a task; the error allocation mechanism is as follows: the deviation value is determined based on each available alternative resource. The degree of "unmet needs" Proportional allocation; Normalized learning step size: by introducing... The learning rate is normalized to ensure that the update magnitude matches the task resource scale and maintains consistency in dimensions; the update direction is as follows: when the benefits are lower than expected and resources are insufficient, This increases the resource allocation in subsequent scheduling, leading to more efficient allocation of that resource. This mechanism enables automatic calibration of model parameters, making the resource substitution cost metric in the optimization objective more aligned with actual business impact, thus forming a closed loop for continuous performance improvement.

[0059] Automatic calibration of model parameters: The system can automatically correct for inaccurate initial settings or settings that have changed over time, based on real-world execution feedback. The coefficients make the cost metrics in the optimization model increasingly closer to the actual business impact, forming a continuous performance improvement loop: this update mechanism and the objective function Direct linkage; The updates directly affect the results of the next optimization calculation, thus forming a reinforcement learning loop from "decision-making -> execution -> evaluation -> model correction -> re-decision-making"; achieving fine-grained and targeted feedback: by allocating errors according to the proportion of resource gaps, fine and targeted adjustments to model parameters are achieved, making the learning process more efficient and accurate.

[0060] Furthermore, in step S211, the critical resource contention cluster is calculated. In the time slot Cluster comprehensive tension index The method is as follows: Construct a weighted undirected graph with intra-cluster resources as nodes and historical co-occurrence strength between resources as edges. And its level of tension is quantified using the following formula: in, Cluster The number of resources contained therein; Representation diagram The set of edges; Representing resources and Historical co-occurrence intensity weights between them; It is a balancing coefficient used to adjust the relative importance of the average scarcity level within a cluster and the dispersion of scarcity differences.

[0061] First item (average scarcity): Calculate clusters Dynamic scarcity of all internal resources The arithmetic mean reflects the overall demand pressure.

[0062] The second term (weighted imbalance): This is the key core of this model. It first calculates the weighted imbalance for each pair of related resources within a cluster. The absolute value of the difference in scarcity Then use its historical co-occurrence strength weight Perform a weighted summation and normalization. This term is specifically used to quantify the risk of mismatch, incoordination, or "imbalance" in demand pressure among resources within a cluster. Weights This ensures that resource pairs that are historically more closely related and frequently shared in task requirements will have a greater "contribution" to overall risk from their current scarcity differences. Coefficient balancing: By balancing coefficients... Adjust the relative importance of the above two items in the final index.

[0063] Precisely quantifying the risk of "weakest link": Traditional methods struggle to predict the risk of "extreme shortages of some resources leading to the failure of the entire resource portfolio." This model directly and quantitatively assesses this structural imbalance risk through the second term, enabling the system to identify "high-risk" resource clusters with low average pressure but extreme internal incoordination; Achieving intelligent and precise strategic buffer triggering: This index upgrades the triggering mechanism of the strategic buffer resource pool from simple threshold judgment to intelligent decision-making based on composite risk perception. The system can not only respond to high average demand but also proactively deploy buffer resources in advance for resource portfolios with high imbalance risk, thereby more effectively preventing high-value composite tasks from failing to schedule due to individual resource weaknesses; Deeply integrating graph theory and statistics: By combining the relationships between resources (graph theory) with real-time state differences (statistics), this model achieves a leap from isolated resource management to collaborative resource network management.

[0064] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A resource allocation method for maximizing the number of parallel underwater acoustic tasks, characterized in that, Includes the following steps: Step S1, Task Modeling and Resource Classification: For each underwater acoustic task to be performed... Establish a resource demand model that divides the required resources into a subset of non-substitutable resources. and alternative resource subsets ; and for each task Assign a comprehensive value weight ; Step S2, dynamic prediction, topology analysis and elastic reservation: based on historical task execution data, predict the dynamic shortage degree of various irreplaceable resources in different time periods in the future ;​ Identifying key resource competition clusters based on historical data And calculate its cluster comprehensive tension index. According to the dynamic scarcity level Cluster Comprehensive Tension Index By using a reserved computation model that includes a smooth activation function, the elastic reserved resource pool and strategic buffer resource pool for each future time period are dynamically determined and managed. Step S3, Multi-objective optimization and dynamic preemption decision: At the scheduling time, the comprehensive optimization objective is to maximize the total value and total number of scheduled tasks and minimize the benefit loss caused by the use of alternative resources or insufficient resources. At the same time, the dynamic real-time execution urgency of the tasks is incorporated as an optimization factor. Under the hard constraint that all tasks must obtain all irreplaceable resources from the corresponding time period's elastic reserved resource pool and strategic buffer resource pool, the current optimal task execution set and resource allocation scheme are solved. When high-urgency tasks cannot be scheduled due to insufficient resources, resource preemption for low-urgency tasks is evaluated and executed. Step S4, Feedback Tuning and Experience Learning: Based on the task The actual benefits after implementation Compared with expected benefits The deviation is dynamically adjusted to determine the comprehensive value weight. The parameters of the reserved calculation model and the benefit loss coefficients related to resource adaptation. Meanwhile, historical scheduling decision cases are stored in the strategy experience base, and case matching and strategy reuse are performed when making new decisions to continuously optimize subsequent scheduling performance. In step S2, the identification of key resource competition clusters Calculate the cluster comprehensive tension index The specific steps for managing the strategic buffer resource pool include: S210. Analyze historical task data to identify combinations of resource items that frequently co-occur in multiple high-value task subsets of irreplaceable resources, and define them as critical resource competition clusters. S211, For each critical resource contention cluster Calculate its future time slots Cluster comprehensive tension index This index comprehensively considers all irreplaceable resources within the cluster. Individual dynamic scarcity And the weights of the edges connecting resources; S212, when a cluster of competition for a key resource is detected. In a specific future time slot Cluster comprehensive tension index When the preset system-level threshold is exceeded, the system automatically triggers the collaborative reservation strategy; S213, based on dynamic scarcity level The elastic reserve amount determined by the reserved calculation model In addition, a portion of resources is allocated from the system's global idle resources for this critical resource contention cluster. Dedicated strategic buffer resources; S214. Implement access control on strategic buffer resources, allowing only those concurrently requesting this critical resource cluster to compete for it. All or most of the resources, and their overall value weight High-priority tasks that exceed another set threshold are used during scheduling.

2. The resource allocation method for maximizing the number of parallel underwater acoustic tasks according to claim 1, characterized in that, In step S3, the specific steps for incorporating the dynamic real-time execution urgency of the task and the preemption of execution resources include: S310: For each task Calculate the dynamic real-time execution urgency, which is determined by the remaining deadline, the time already elapsed, and the overall value weight of the task. The value of this function increases dynamically as the deadline approaches and the waiting time lengthens; S311: When making multi-objective optimization decisions, the system considers not only the static comprehensive value weights. Furthermore, this dynamic real-time execution urgency will be incorporated as a key factor into the calculation of the objective function or as an additional optimization sub-objective; S312: The system maintains a monitoring list of allocated resources. When a new high-urgency task cannot be scheduled immediately due to insufficient resources, a resource preemption assessment process is initiated. S313: The resource preemption assessment process assesses the low-urgency tasks that are currently being executed or have been scheduled, calculates the global benefit gain that could be brought about by interrupting or postponing these low-urgency tasks to release the critical resources they occupy, and compares this gain with the system overhead and task restart cost caused by the preemption operation. S314: If the evaluation result shows that the global net benefit is positive and exceeds the preset preemption threshold, the system performs a resource preemption operation, forcibly reclaims some of the resources occupied by low-urgency tasks, redistributes them to the high-urgency tasks, and then re-optimizes the scheduling.

3. The resource allocation method for maximizing the number of parallel underwater acoustic tasks according to claim 2, characterized in that, In step S4, the specific steps for constructing the strategy experience base and reusing case matching include: S410: The system not only records the micro-benefit deviations of task execution for parameter tuning, but also stores each complete scheduling decision cycle, along with its corresponding system state context, the scheduling strategy adopted, and the final macro-performance indicators, as a complete scheduling case in a strategy experience base. The context of each scheduling case includes the characteristic distribution of the task queue and its dynamic scarcity. The model and key resource competition cluster The state; S411: When encountering a new scheduling decision point, the system first performs a similarity match between the current system state context and the historical cases stored in the policy experience base. S412: If a highly similar historical case is matched, the system will first load the historical scheduling strategy associated with the case as the initial strategy or reference strategy for the current step S3 multi-objective optimization solution, and then perform rapid optimization and fine-tuning based on this strategy. S413: At the same time, the system continuously performs offline analysis and mining of cases in the strategy experience base, summarizes efficient scheduling strategy templates under different system state modes through machine learning methods, and uses these strategy templates as meta-knowledge to guide the generation of initial strategies in new scenarios.

4. The resource allocation method for maximizing the number of parallel underwater acoustic tasks according to claim 3, characterized in that, In step S2, the irreplaceable resources are calculated. In the future Dynamic scarcity of each time slot The formula is: in, For predicted future time slots irreplaceable resources The total irreplaceable demand; Irreplaceable resource Total available quantity; and These are irreplaceable resources from the same historical period. Average and standard deviation of occupancy; For the task The overall value weight; For indicator functions, when non-replaceable resources It belongs to the task Irreplaceable subset of resources The value is 1 if the condition is met, and 0 otherwise. For the task In the time slot Probability estimate of when execution will begin; These are normalized weight coefficients, and satisfy... .

5. A resource allocation method for maximizing the number of parallel underwater acoustic tasks according to claim 4, characterized in that, In step S2, based on the dynamic scarcity level Computational irreplaceable resources In the time slot Flexible allowance The formula is: in, , Irreplaceable resource The maximum reserve ratio coefficient, and this coefficient satisfies ; Irreplaceable resource The scarcity threshold; For the task irreplaceable resources The irreplaceable demand; This is the set of tasks that have been submitted and are waiting to be scheduled for execution.

6. A resource allocation method for maximizing the number of parallel underwater acoustic tasks according to claim 5, characterized in that, In step S3, the comprehensive optimization objective is expressed by the following function. Express: in, Let be a binary decision variable, representing the task. Whether it is scheduled; For continuous decision variables, it represents the variables assigned to the task. Alternative resources The actual quantity; For the task For alternative resources The baseline demand; For the task Use alternative resources The adaptation benefit loss coefficient; For the task Dynamic real-time execution urgency; To balance the weighting coefficients of each sub-objective.

7. A resource allocation method for maximizing the number of parallel underwater acoustic tasks according to claim 6, characterized in that, In step S4, the adaptation benefit loss coefficient is dynamically updated based on the following formula. : in, , These are the coefficient values ​​before and after the update, respectively. For the task The baseline total demand for alternative resources; The learning rate; , Tasks The measure of expected and actual benefits; For the task Alternative resources actually allocated and used quantity; For the task For alternative resources The baseline demand; As a replaceable resource; It is a very small positive number used to ensure numerical stability.

8. A resource allocation method for maximizing the number of parallel underwater acoustic tasks according to claim 7, characterized in that, In step S211, the critical resource competition cluster is calculated. In the time slot Cluster comprehensive tension index The method is as follows: Construct a weighted undirected graph with intra-cluster resources as nodes and historical co-occurrence strength between resources as edges. And its level of tension is quantified using the following formula: in, Cluster The number of resources contained therein; Representation diagram The set of edges; Representing resources and Historical co-occurrence intensity weights between them; It is a balancing coefficient used to adjust the relative importance of the average scarcity level within a cluster and the dispersion of scarcity differences.

Images