A collaborative task planning method oriented towards intrinsic security and availability

Abstract

This invention provides a collaborative task planning method for intrinsic security and availability, including: irregular redundancy decomposition of complex tasks, collaborative matching of differences between meta-tasks and execution starting points, a multi-dimensional heterogeneous planning mechanism for meta-tasks, and a dynamic dual feedback strategy. This invention organically combines an intrinsic security and availability collaborative architecture with task planning, focusing on node and task differences within the architecture. While ensuring system availability, it achieves optimal planning of intrinsic security collaborative tasks, extending the system's lifespan.

Description

Technical Field

[0001] This invention belongs to the field of intrinsic security technology in cyberspace, and relates to intrinsic security network availability technology, specifically a collaborative task planning method for intrinsic security availability. Background Technology

[0002] Intrinsic security refers to the security functions or attributes obtained by utilizing the inherent security effects of a system's own architecture, functions, and operational planning mechanisms. It has the basic characteristics of being both inherently constructed and subsequently developed. There are currently three mainstream architectures for intrinsically secure systems: dynamic heterogeneous redundancy architecture, neural network-like architecture, and elastic heterogeneous architecture.

[0003] Endogenous security network design has four important evaluation metrics: availability, confidentiality, integrity, and non-repudiation. Currently, most research on the three mainstream architectures focuses on collaborative task planning mechanisms for endogenous security availability (ESA). In an endogenous security architecture, network resources are collaboratively scheduled to perform tasks such as node fault repair, node performance enhancement, and node external services. This achieves optimal system task planning while ensuring endogenous security availability is met. For example, when a target node is suspected of failing, the core network distributes fault repair tasks to several task-initiating nodes (e.g., clusters). These initiating nodes then schedule executors (e.g., the main device or virtual machine executing the task) or relay nodes to collaboratively forward the fault repair task, quickly transmitting it to the target node. After the target node executes the task, it feeds back the execution status to the core network and the initiating nodes for task correction. This process repeats until the system returns to normal.

[0004] However, collaborative task planning mechanisms for ESA still have a series of problems:

[0005] (1) The scheduling equivalent execution entity suffers from serious biases in system evaluation. The equivalent execution entity, which does not consider the differences in task and node functions, has biases in adjudicating transmission results, thus leading to misjudgments of attacks and even attack escape problems, as well as biases in the evaluation of system availability and overhead. In addition, in order to execute multiple types of tasks within the system, the equivalent execution entity must have the ability to execute all tasks, resulting in an extremely complex structure, high cost, and serious waste of resources.

[0006] (2) Complex task decomposition patterns do not consider redundant tasks and are not suitable for endogenous architectures. The existence of redundant tasks leads to poor parallelism and long task completion cycles, which severely limits the efficient use of resources.

[0007] (3) A single feedback system is commonly used. Using only positive feedback mechanisms can easily lead to fixed online execution entities (or nodes) and require a long period of time to correct the judgment bias of their own feedback learning, which is not conducive to the high availability of the system. Using only negative feedback mechanisms combined with mimicry voting results in a high false alarm rate when the system is vulnerable to attack, which reduces the reasonable online time of execution entities, increases the frequency of system cleaning, and the frequent cleaning significantly reduces the system's ability to provide services.

[0008] The aforementioned problems have limited the development of research on collaborative planning for endogenous security tasks. Summary of the Invention

[0009] To address the aforementioned issues, this invention provides a collaborative task planning method oriented towards intrinsic security and availability (ESA). This method achieves optimal task planning while ensuring system availability, thereby extending the system's lifespan. The steps of this invention for ESA-oriented collaborative task planning and execution are summarized as follows: The central server decomposes complex tasks into several irregular redundant meta-tasks, i.e., small tasks with different redundancies; then, it achieves collaborative matching between meta-tasks and execution starting points; subsequently, the execution starting point transmits meta-tasks to the target node through routing nodes; the system provides feedback calibration to the central server's allocation method based on the task execution status at the target node, thereby achieving system self-evolution.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] A collaborative task planning method oriented towards intrinsic security and availability includes the following steps:

[0012] Step 1: Irregular Redundancy Decomposition of Complex Tasks

[0013] When a complex task is input, if a matching case can be directly searched from the case library, the current task is decomposed according to the previous decomposition method of the matching case; otherwise, the task is added to the case library.

[0014] Step 2: Collaborative matching of differences between meta-tasks and execution starting points

[0015] Based on the information from the irregular redundancy decomposition of complex tasks, we construct a difference matching model using the obtained meta-tasks and the information they carry, and design a hybrid mathematical heuristic FISA algorithm to solve this problem.

[0016] Step 3: Employ a meta-task multi-dimensional heterogeneous planning mechanism

[0017] First, a multidimensional heterogeneous programming problem model is constructed, and then a hybrid local search algorithm is designed to solve the problem.

[0018] Step 4: Implement dynamic dual feedback

[0019] A dual feedback strategy based on a hybrid algorithm is adopted to design a reasonable system continuous service time and achieve an integrated and efficient planning scheme.

[0020] Furthermore, in step one, the case library is collected through the following steps: surveying and collecting application scenarios and decomposition schemes for complex tasks that can reflect the characteristics of the problem, organizing and classifying the data to form an initial case library, and continuously revising the case library and cases during the process of case matching and task planning; the decomposition of complex tasks is implemented using the K-nearest neighbor algorithm, and the irregular redundant meta-tasks after decomposition are output according to the matching results.

[0021] Furthermore, the process of continuously revising the case library and cases is as follows: when the similarity between the current solution and the most similar case in the case library is less than a threshold, a new solution with a similarity balance between the current solution and the most similar case is generated, the current solution is revised into the new solution, and the revised solution is added to the case library.

[0022] Furthermore, step one also includes a meta-task standardization step:

[0023] The necessary information for a standardized metatask includes: metatask execution time, metatask end time, metatask target node, metatask resource types, and conflicts or coupling relationships between the metatask and other metatasks.

[0024] Furthermore, step two specifically includes the following steps:

[0025] (1) Modeling and analysis of differential collaborative matching

[0026] In the difference matching model, meta-tasks correspond to vertices in the topological graph of graph theory. Two meta-tasks that conflict in execution time are adjacent to each other by an edge. The execution starting point corresponds to a color that can be used for coloring. The model is described as follows: Given a graph G = (V, E), where V = {1, 2, ..., v} is the set of meta-tasks, E is the set of conflicts between meta-tasks, and C = {1, 2, ..., c} is the set of execution starting points; each meta-task is matched with an execution starting point such that the number of conflicting meta-tasks is minimized, and the total number of meta-tasks assigned to each execution starting point is as close as possible to its capacity.

[0027] The objective function of the difference matching model first considers reducing system resource overhead and minimizing the total number of execution starting points of the calls, and its expression is:

[0028]

[0029] Where w i Indicates whether the execution start point i is assigned a meta-task, and its expression is: Where x ij This represents the matching relationship between meta-tasks and execution start points, expressed as:

[0030]

[0031] Another objective function is to maximize the availability of the system, i.e.

[0032]

[0033] Where μ i 1≤i≤c represents the availability of the execution starting point i; Formula (3) and the system limit should select the execution starting point with the highest availability as much as possible;

[0034] The model problem has the following constraints:

[0035] a) Node availability

[0036] While considering the system's availability objective function, the availability metric of a node is set to be greater than a threshold ε, i.e.

[0037] μ i ≥ε (4)

[0038] b) Task conflict relationships

[0039] Conflicting meta-tasks are not matched to the same execution starting point, and their expression is:

[0040]

[0041] Here, j and u are two conflicting meta-tasks;

[0042] c) Node capacity constraints

[0043] The total memory required for each metatask allocated to each execution starting point must not exceed the fixed memory capacity B of that execution starting point. i That is, weight balance, its expression is:

[0044]

[0045] Where b ij This indicates the memory required by the metatask j to be allocated to the execution start point i;

[0046] d) Task and node matching relationship

[0047] Metatasks can only be matched with the execution starting point that handles this type of task, and their expression is:

[0048]

[0049]

[0050]

[0051] Where d represents the resource type, r jd This indicates whether task j requires resource d during execution, while r * id This indicates whether the execution starting point i has resource d; if and only if r jd and r * id Task j and execution start point i can only be successfully matched when both values ​​are 1:

[0052] e) Task and node coupling relationship

[0053] Due to the various differences in execution starting points, the coupling benefits of matching meta-tasks with different execution starting points vary, and the benefit expression is as follows:

[0054]

[0055] Where, p ij R represents the coupling benefit between execution start point i and meta-task j, where β1, β2, and β3 are benefit parameters greater than 0. i The amount of resources M for execution starting point i i For the data processing time starting from point i, P i The throughput of execution starting point i;

[0056] f) Coupling between tasks

[0057] Matching similar meta-tasks derived from the same complex task with the same execution starting point yields additional secondary benefits, expressed as follows:

[0058]

[0059] Where, q ij δ represents the additional benefit at execution starting point i. jk Used to indicate whether task j and task k are of the same type of task, β is a reward parameter greater than 0;

[0060] (2) Design of a hybrid algorithm for difference-based collaborative matching

[0061] A mixed integer programming FISA algorithm is designed to solve the model in step (1). First, an integer programming model is constructed to generate the objective function and constraints. Then, a greedy algorithm is used to construct the initial matching of the meta-task and the execution starting point. Next, the matching scheme is optimized and upgraded using the FISA algorithm. Then, a heuristic strategy is used to repair the matching scheme that does not meet the constraints. Finally, the excellent matching scheme is recorded.

[0062] Furthermore, the FISA algorithm for mixed integer programming includes the following specific steps:

[0063] Generate initial matching scheme:

[0064] 1) Select c execution starting points with the highest availability;

[0065] 2) A greedy method is used to select the meta-task j with the largest gain for each execution starting point i in turn. The formula for selecting the meta-task by the greedy algorithm is as follows:

[0066]

[0067] Where, p ij The coupling benefit value when matching the execution starting point i to the meta-task j is obtained through formula (10), q ij The secondary benefit value of the execution at the starting point i of the matching meta-task j is obtained by formula (11);

[0068] FISA algorithm optimizes matching schemes: The FISA algorithm is used to optimize and upgrade matching schemes. If a better matching scheme cannot be found within a certain period of time, some hard constraints are relaxed to soft constraints based on the quality of the current matching scheme and the mechanism of reinforcement learning, until the evaluation quality of the relaxed matching scheme can no longer be improved.

[0069] Heuristic repair matching scheme: Forcefully transfer meta-tasks that exceed the capacity of the execution start point and conflicting meta-tasks to other execution start points in a greedy manner; and after forced repair, further adopt FISA to optimize the matching scheme.

[0070] Furthermore, step three specifically includes the following steps:

[0071] (1) Establish a multidimensional heterogeneous programming problem model

[0072] The integrated design incorporates multi-dimensional heterogeneous planning based on path, spectrum, and time dimensions, specifically including:

[0073] The path heterogeneous programming problem is transformed into the weighted shortest path problem; the spectrum band heterogeneous programming problem is equivalent to the classic bandwidth graph coloring problem; the time heterogeneous programming problem is equivalent to the classic graph coloring problem in graph theory.

[0074] Let G′=(V′,E′) be the physical topology of the network executing the meta-task, where V′ is the set of routing nodes in the network and E′ is the set of bidirectional edges. An integer programming mathematical model for the multidimensional heterogeneous problem is given. Taking three dimensions as an example, its objective function is to minimize the maximum number of spectral bands traversed on the physical edges of the network at the same time.

[0075] min F max (13)

[0076]

[0077] Among the variables This represents the number of network segments (n,m) that connect the source node s and the destination node d at time t, with the spectrum band labeled f.

[0078] For G′, the availability of paths in the network is also defined as U. z , 1≤z≤v, where v is the number of meta-tasks and also the number of requirement paths; the metric for maximizing the availability of each path is defined as...

[0079]

[0080] The previous spectral band spacing constraint of two-dimensional heterogeneity was expressed as

[0081]

[0082] Where φ represents the number of subcarriers, GC represents the size of the guard carrier, and N represents a constant greater than the number of meta-tasks v;

[0083] (2) Hybrid Algorithm for Solving Multidimensional Heterogeneous Programming Problems

[0084] A super-large neighborhood search algorithm with a hybrid reinforcement learning mechanism is designed for the multidimensional heterogeneous programming problem. The algorithm includes: first, constructing an initial multidimensional heterogeneous programming scheme; then, quickly searching for a high-quality programming scheme in the super-large neighborhood; designing a heuristic receiving mechanism to determine whether to replace the current programming scheme; then, using a hybrid reinforcement learning mechanism to slightly adjust the programming scheme; and repeating this process until the programming scheme meets the system requirements.

[0085] Furthermore, the ultra-large neighborhood search algorithm includes the following steps:

[0086] Initialize the multidimensional heterogeneous planning: A greedy construction algorithm based on Dijkstra is used to generate the initial planning scheme, which includes: First, based on the difference matching results between the meta-task and the execution starting point, the initial shortest execution path is assigned to each meta-task; then, some path segments are modified to improve the path heterogeneity; then, the spectral band heterogeneity and temporal domain heterogeneity are improved in the same way, and finally the initial planning scheme of multidimensional heterogeneous meta-task planning is obtained.

[0087] VLSN enhances heterogeneity by making gradual adjustments to the current planning scheme. The multi-dimensional transformation method of gradual adjustment is as follows: extract special parts of the planning from the current planning scheme, and change the path, spectrum band and time of the part of the planning to improve the overall heterogeneity.

[0088] In the VLSN algorithm, reinforcement learning is used to guide neighborhood selection: based on the execution status of the current meta-task, reinforcement learning is designed to determine the variable selection method for each dimension; when determining the selection method, a reward or penalty value is given first, and then the reward or penalty mechanism is combined with the roulette wheel selection strategy to select variables for each dimension.

[0089] Heuristic reception mechanism design: Combine tabu and simulated annealing reception strategies, and dynamically adjust the reception method of the new scheme based on the results of operation;

[0090] Reinforcement learning optimizes heterogeneity: Reinforcement learning is used to adjust the planning scheme, delete and insert η variables, and enhance the global search capability of VLSN.

[0091] Furthermore, the dynamic adjustment method is as follows: when the current planning method is the same as the previous one, the probability of using the same planning method in the next one is increased; otherwise, the probability of using the same planning method in the next one is decreased.

[0092] Furthermore, step four specifically includes the following steps:

[0093] (1) Positive feedback mechanism based on statistical learning

[0094] During network operation, the system employs superior task decomposition, matching, and execution methods that offer high storage availability and low system overhead. It also constructs an efficient statistical probability model to learn these superior planning methods, thereby reducing the computational load for the next allocation. Furthermore, it reduces the probability of planning methods that exceed the continuous service time threshold θ1 being selected.

[0095] (2) Negative feedback mechanism based on tabu algorithm

[0096] Planning methods with continuous service times exceeding θ2 are cleaned up; based on the execution results and combined with a multi-mode adjudication mechanism, abnormal network nodes are forcibly taken offline, and abnormal nodes are cleaned up in a timely manner to ensure system availability.

[0097] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0098] 1. By decomposing complex tasks into irregular redundancies, the long-term occupation of specific devices and the impact on the execution of other tasks can be reduced. Parallel execution of meta-tasks can reduce device idle time, increase resource utilization, and improve system availability. Furthermore, decomposing complex tasks into meta-tasks with varying redundancy levels creates a cognitive dilemma for attackers regarding the target system's structure and operating mechanisms, thereby disrupting or undermining the stability of the attack chain, increasing the cost of system damage for attackers, and further enhancing the availability of the intrinsically secure system. To ensure compatibility with existing networks, this invention considers the redundancy of non-redundant meta-tasks to be 1.

[0099] 2. This invention achieves collaborative matching between meta-tasks and execution starting points. In real-world network models, the performance differences between execution starting points and meta-tasks cannot be ignored; otherwise, it will lead to serious deviations in system evaluation. This invention fully considers the differences in external service capabilities of execution starting points, the coupling between meta-tasks, and the coupling between meta-tasks and nodes, and performs modeling analysis to further improve the efficiency of system task collaboration while ensuring system availability.

[0100] 3. Employing a meta-task multi-dimensional heterogeneous planning mechanism. Heterogeneity is a crucial characteristic for ensuring system availability, significantly reducing the probability of the same vulnerability causing network anomalies. This invention models and analyzes multi-dimensional heterogeneous architectures, narrowing the intersection of attacks from multiple dimensions, enhancing system availability, and providing a theoretical basis for improving system security and performance.

[0101] 4. A dynamic dual feedback strategy is adopted. Based on the task execution feedback information of the target node, this invention designs a positive feedback strategy, providing a theoretical basis and guidance for the optimization and improvement of the above modules; it also designs a negative feedback strategy combined with multi-mode adjudication, forcibly taking offline abnormal paths and forcibly prohibiting tasks from using the same execution method for a long time, so that the system exhibits uncertainty effects externally, thereby ensuring the dynamism and availability of the system.

[0102] 5. This invention, through a fusion architecture and task collaborative planning method, enables the system's security capabilities and system tasks to be tightly coupled, providing flexibility at multiple levels and improving resource utilization, thereby forming a security system with a good balance of endogenous effects in terms of availability, operational efficiency, and system overhead. Attached Figure Description

[0103] Figure 1 This is a schematic diagram of a collaborative task planning method that combines an intrinsic security architecture, as provided by the present invention.

[0104] Figure 2 This is an example diagram of irregular redundancy decomposition of complex tasks provided by the present invention.

[0105] Figure 3 This is a schematic diagram of the basic process of complex task decomposition provided by the present invention.

[0106] Figure 4 This is a schematic diagram of the graph equilibrium coloring variant problem model provided by the present invention, which is equivalent to difference-cooperative matching.

[0107] Figure 5 This is a schematic diagram of the hybrid mathematical heuristic algorithm for solving differential collaborative matching provided by the present invention.

[0108] Figure 6This is a schematic diagram of the routing and heterogeneous spectrum allocation of network G′ provided by the present invention.

[0109] Figure 7 This is a schematic diagram of the hybrid local search algorithm for solving multidimensional heterogeneous programming problems provided by the present invention. Detailed Implementation

[0110] The technical solutions provided by the present invention will be described in detail below with reference to specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0111] The collaborative task planning method for intrinsic security availability proposed in this invention is an intrinsic security network availability technology used to realize the planning function of intrinsic security collaborative tasks. Its schematic diagram is shown below. Figure 1 As shown, in order to more clearly express the logical relationship, Figure 1 The feedback relationship is plotted separately. The method of this invention specifically includes the following steps:

[0112] (I) Irregular Redundancy Decomposition of Complex Tasks

[0113] Decomposing complex tasks into meta-tasks can significantly reduce system completion time and enhance system availability; irregular decomposition makes it difficult for attackers to grasp the system's changing patterns, and reduces redundancy when the system security level is high, minimizing system overhead. To explain the above conclusions, Figure 2 A schematic diagram of irregular redundancy decomposition for complex tasks is provided. Figure 2 To simplify the model, the following assumptions are made: each task can be completed using only one execution node, A and B are both redundant tasks, and task B can only begin execution after its corresponding task A has been completed. Figure 2 (a) represents the case where complex and redundant tasks A and B are not decomposed; Figure 2 (b) indicates that the complex task rules are redundantly decomposed, and the task completion time is in the 6th hour. Figure 2 (c) represents irregular redundancy decomposition. In a relatively safe environment, reducing the redundancy of A2 and B2 will reduce the running time of execution nodes 2 and 3 by 2 hours.

[0114] In summary, the irregular and redundant decomposition of complex tasks is crucial, directly impacting network availability and system overhead, requiring a trade-off analysis based on specific scenario examples. Therefore, this invention proposes a case-base matching method for rapid decomposition of complex tasks. The specific method is as follows: When a complex task is input, if a matching case can be directly searched from the case base, the task is decomposed according to the previous decomposition methods of the matching case; otherwise, the task is added to the case base. The process is as follows: Figure 3 As shown.

[0115] The specific details of the case library generation, case correction, and case matching technologies are as follows:

[0116] Case library generation: This invention surveys and collects application scenarios and decomposition schemes for complex tasks that can reflect the characteristics of the problem. After organizing and classifying the data, an initial case library is formed. During the process of case matching and task planning, the case library and cases are continuously revised.

[0117] Case matching: Based on the database collected above, this invention uses the K-Nearest Neighbor algorithm (KNN) to decompose complex tasks in a scenario-oriented manner, matches the input complex task with standard cases in the case library, and finally outputs the decomposed irregular redundant meta-tasks based on the matching results.

[0118] Case correction technique: When the similarity between the current solution and the most similar case in the case library is less than a threshold H, a new solution is generated that has a balanced similarity with the current solution and the most similar case (i.e., the similarity between the current solution and the most similar case is equal or the difference between them is less than a small threshold). The current solution is then corrected to the new solution, and the corrected solution is added to the case library. The similarity is defined as the number of identical decomposition meta-tasks in the two solutions.

[0119] Meta-task standardization: This provides a reference for extracting constraints in subsequent differential collaborative planning. The decomposed meta-tasks need to carry important information, such as their execution time and conflicts or coupling relationships with other meta-tasks. In this invention, the necessary information for standardizing meta-tasks includes: execution time, end time, target node, required resource type, and conflicts or coupling relationships with other meta-tasks.

[0120] (II) Collaborative Matching Based on Differences Between Meta-Tasks and Execution Starting Points

[0121] Based on the information from the irregular redundancy decomposition of complex tasks, a difference matching model is constructed using the obtained meta-task and its carried information. A hybrid mathematical heuristic FISA algorithm is then designed to solve this problem. The following describes the establishment and solution of the difference matching model:

[0122] (1) Modeling and analysis of differential collaborative matching

[0123] The difference matching problem in this invention is a variant of the classic graph balanced coloring problem, and it is also an NP-hard problem, such as... Figure 4As shown in the difference matching model, in this model, meta-tasks correspond to vertices in the topological graph of graph theory. Two meta-tasks that conflict in execution time are adjacent to each other by an edge, and the starting point of execution corresponds to a color that can be used for coloring. The model can be described as follows: Given a graph G = (V, E), where V = {1, 2, ..., v} is the set of vertices (meta-tasks), E is the set of edges (conflicts between meta-tasks), and C = {1, 2, ..., c} is the set of colors (starting points of execution). Each vertex (meta-task) is matched with a color (an execution starting point) such that the number of conflicting meta-tasks is minimized, and the total number of meta-tasks assigned to each execution starting point is as close as possible to its capacity.

[0124] The objective function of the difference matching model first considers reducing system resource overhead, such as minimizing the total number of execution starting points of the calls, and its expression is:

[0125]

[0126] Where w i Indicates whether the execution start point i is assigned a meta-task, and its expression is: Where x ij This represents the matching relationship between meta-tasks and execution start points, expressed as:

[0127]

[0128] This invention studies a collaborative matching scheme for tasks with different execution starting points, considering both inherent security availability and system overhead. Another objective function of this invention is to maximize system availability.

[0129]

[0130] Where, μ i 1≤i≤c represents the availability of execution starting point i. Formula (3) and the constraint that the system should select execution starting points with higher availability as much as possible further guarantee the availability of the system. As the system runs, the objective function of system overhead (Formula (1)) and the objective function of availability (Formula (3)) will be modified. In addition, the problem of the present invention has the following constraints:

[0131] a) Node availability. Furthermore, to ensure system availability, this invention, while considering the objective function (Formula (3)), sets the node availability index to be greater than a threshold ε, i.e.

[0132] μ i ≥ε (4)

[0133] b) Task Conflict Relationships. In the balanced graph coloring problem, conflicting vertices (meta-tasks) are not matched with the same color (execution start point). Similarly, when two meta-tasks that overlap in time are matched with the same execution start point, it will inevitably lead to one or both meta-tasks failing to execute normally, thus reducing system availability. Therefore, this invention will continue to use this constraint, the expression of which is:

[0134]

[0135] Here, j and u are two conflicting meta-tasks.

[0136] c) Node capacity constraint. Unlike the balanced graph coloring problem, the difference matching problem in this invention does not require a balanced number of meta-tasks allocated to each execution starting point. Instead, it requires that the total memory required for the meta-tasks allocated to each execution starting point does not exceed the fixed memory capacity B of that execution starting point. i This refers to weight balancing. Therefore, during the experiment, this invention will make corresponding modifications to the meta-task constraint capacity, the expression of which is:

[0137]

[0138] Among them, b ij This indicates the memory required by the metatask j to be allocated to the execution start point i.

[0139] d) Task-node matching relationship. In this invention, meta-tasks have different types (e.g., download tasks, command tasks, and detection tasks), and each type of meta-task has specific resource requirements. Since the resources of different execution starting points are different, meta-tasks can only be matched with the execution starting points that handle this type of task.

[0140]

[0141]

[0142]

[0143] Where d represents the resource type, r jd This indicates whether task j requires resource d during execution, while r * id This indicates whether the execution starting point i has resource d. If and only if r jd and r * id Task j and execution start point i can only be successfully matched when both values ​​are 1.

[0144] e) Task and node coupling relationship. Meta-tasks have different benefits when matched with different execution starting points. That is, when a meta-task is matched with an execution starting point that has abundant resources, fast data processing speed, and high throughput, the system's response rate, availability, and benefit value are high, and the completion time is short. Therefore, based on the above differences in execution starting points, the coupling benefit between different meta-tasks and execution starting points is calculated, and its expression is:

[0145]

[0146] Where, p ij R represents the coupling benefit between execution start point i and meta-task j, where β1, β2, and β3 are benefit parameters greater than 0. i The amount of resources M for execution starting point i i For the data processing time starting from point i, P i The throughput of execution starting point i.

[0147] f) Inter-task coupling. For similar meta-tasks decomposed from the same complex task, if they have similar resource requirements, matching them to the same execution starting point can eliminate the need for reliable verification of the execution starting point. Furthermore, according to the principle of locality of reference, nodes executing similar tasks require similar types of resources. Therefore, when similar meta-tasks decomposed from the same complex task are matched with the same execution starting point, there is an additional secondary benefit, expressed as follows:

[0148]

[0149] Where, q ij δ represents the additional benefit at execution starting point i. jk This is used to indicate whether task j and task k are of the same type, and β is a reward parameter greater than 0.

[0150] Since differences between nodes and tasks are unavoidable, when differences exist, this invention can plan the matching model based on the above-mentioned difference-based collaborative matching scheme, which has a certain degree of robustness.

[0151] (2) Design of a hybrid algorithm for difference-based collaborative matching

[0152] Since the difference-matching problem is NP-hard, direct solutions are very difficult. A common approach is to use hybrid mathematical heuristics to solve this problem. Therefore, this invention designs a hybrid integer programming FISA algorithm to solve this problem, and the algorithm flowchart is shown below. Figure 5 As shown.

[0153] First, an integer programming model is constructed to generate the objective function and constraints. Next, a greedy algorithm is used to construct the initial matching of the meta-task and the execution starting point. Then, the FISA algorithm is used to optimize and upgrade the matching scheme. Following this, a heuristic strategy is employed to repair matching schemes that do not meet the constraints. Finally, excellent matching schemes are recorded to accelerate the matching rate in the next iteration. The specific steps are as follows:

[0154] Generating an initial matching scheme: A good initial matching scheme can effectively reduce the number of searches and help the search process quickly find the optimal match, thereby improving the efficiency of the algorithm. The specific steps are as follows: 1) Select c execution starting points with the highest availability; 2) Use a greedy method to select the meta-task j with the largest gain for each execution starting point i in turn. Based on the analysis of the above constraint df, the formula for selecting meta-tasks by the greedy algorithm in this invention is as follows:

[0155]

[0156] Where p ij The coupling benefit value when matching the execution starting point i to the meta-task j (Formula (10)), q ij The secondary benefit value of the execution at the starting point i of the matching meta-task j is given by formula (11).

[0157] Optimizing matching schemes in the FISA algorithm: The design of the conversion mechanism between soft and hard constraints in the model directly affects the algorithm's efficiency. The conversion between soft and hard constraints should be an easily implemented mechanism with a learning strategy. To improve search efficiency, when solving the problem of collaborative matching between the meta-task and the starting point of execution, if a better matching scheme cannot be found within a certain timeframe, some hard constraints are relaxed to soft constraints based on the quality of the current matching scheme and the reinforcement learning mechanism, until the evaluation quality of the relaxed matching scheme cannot be further improved. This invention employs the relaxation of hard constraint condition bc.

[0158] Heuristic Repair Matching Scheme: After the FISA search is completed, some execution starting points may have been matched with unreasonable tasks, such as meta-tasks that are assigned beyond their inherent capacity or conflicting meta-tasks. This invention designs a heuristic greedy repair method, with the following specific steps: forcibly transferring meta-tasks exceeding the execution starting point's capacity and conflicting meta-tasks to other execution starting points in a greedy manner; and after the forced repair, further employing the FISA optimized matching scheme.

[0159] (III) Meta-task Multidimensional Heterogeneous Planning Mechanism

[0160] This invention, based on the difference matching between the meta-task and the execution starting point, further considers the specific multi-dimensional heterogeneous execution mode of the meta-task among nodes in the network, that is, considering the heterogeneity of multiple attributes in the time domain, frequency domain, and spatial domain during execution. Because the more heterogeneous attributes the execution entity has, the higher the attack difficulty. In complex networks and resource-constrained situations, ensuring heterogeneity across various attributes or different levels as much as possible can prevent the simultaneous occurrence of the same vulnerability. Furthermore, to ensure system availability, the multi-dimensional heterogeneous programming of the meta-task, when constructing the objective function, further considers system availability based on the node availability mode of difference matching between the meta-task and the execution starting point. Based on the above ideas, this research scheme first constructs a multi-dimensional heterogeneous programming problem model, and then designs a hybrid local search algorithm to solve the problem.

[0161] (1) Establish a multidimensional heterogeneous programming problem model

[0162] This invention integrates multi-dimensional heterogeneous planning based on design path, spectrum, and time, specifically:

[0163] Heterogeneous Path Planning: To reduce system overhead, heterogeneous path planning should be considered first. To enhance the system's routing heterogeneity, the number of overlapping routing nodes in the execution paths of different meta-tasks should be minimized. Simultaneously, to conserve system resources and avoid impacting other communication paths, the number of relay nodes in the execution paths of meta-tasks should be minimized. Furthermore, considering the connectivity between relay nodes, this problem can be transformed into a weighted shortest path problem.

[0164] Heterogeneous spectrum planning: To improve the heterogeneity of the system spectrum, when different meta-tasks pass through overlapping segments, the spectrum bands used should be as non-overlapping as possible, and the number of spectrum bands used should be as small as possible. This problem can be equivalent to the classic bandwidth graph coloring problem.

[0165] Time-based and other heterogeneous planning: Since time-based planning is similar to other planning methods, this invention uses time-based planning as an example to describe all other heterogeneous planning methods. Based on the adjustable execution time of meta-tasks, when different meta-tasks pass through overlapping segments and use the same spectrum, their execution times should be staggered as much as possible, and the total time required for each meta-task to complete should be minimized. This problem is equivalent to the classic graph coloring problem in graph theory. Because different systems have different resource constraints—for example, some systems cannot use heterogeneous planning due to short timeframes—other planning methods such as software are employed.

[0166] To illustrate the multidimensional heterogeneous programming problem, Figure 6 A multidimensional heterogeneous programming process is given for a given network structure graph G′. Let G′=(V′,E′) be the physical topology graph of the network for the execution of the meta-task, where V′ is the set of routing nodes in the network and E′ is the set of bidirectional edges. Figure 6 (a) represents the network topology G′. For this topology, there are four sets of execution start and target nodes for the meta-tasks, namely {1,2},{1,3},{2,4},{2,5}. Figure 6 (b) illustrates a three-dimensional heterogeneous allocation scheme. Different colors in the diagram represent different spectral bands, and dashed and solid lines represent different times. Different meta-tasks execute via different routes. In summary, the multidimensional heterogeneous planning problem is a multi-level NP-hard problem, which incorporates characteristics of classic combinatorial optimization problems.

[0167] This invention first presents an integer programming mathematical model for multidimensional heterogeneous problems. Taking three dimensions as an example, the objective function is to minimize the maximum number of spectral bands traversed on a network physical edge at the same time, i.e.

[0168] min F max (13)

[0169]

[0170] Among the variables This represents the number of network segments (n,m) connecting source node s and destination node d with spectral band label f, passing through endpoints n and m respectively, at time t. Currently, there is limited research on multidimensional heterogeneity. This invention will add new objective functions, such as the total length of the task execution path and the availability of the execution path, to comprehensively plan the system's security and resource utilization.

[0171] For G′, the availability of paths in the network is also defined as U. z , 1≤z≤v, where v is the number of meta-tasks and also the number of requirement paths. This invention defines the maximization of the availability index for each path as:

[0172]

[0173] The spectral band spacing constraint of two-dimensional heterogeneous structures can be expressed as follows:

[0174]

[0175] Where φ represents the number of subcarriers, GC represents the size of the guard carrier, and N represents a constant greater than the number of meta-tasks v. It should be noted that previous constraints only considered the case where the interval bandwidth was the same, resulting in a huge waste of bandwidth. Therefore, this problem will first adjust the interval bandwidth to be dynamic to reduce bandwidth utilization. Secondly, other constraints such as time domain constraints will be added, that is, for any time and any software, the formula (16) will be satisfied, and the complexity of the problem will also increase significantly.

[0176] (2) Hybrid Algorithm for Solving Multidimensional Heterogeneous Programming Problems

[0177] This invention addresses the multidimensional heterogeneous programming problem, a multidimensional NP-hard problem with more complex objective functions and constraints, many of which are nonlinear. Therefore, the choice of algorithm directly affects the efficiency of solving the system programming problem. This invention designs a Very Large Neighborhood Search (VLSN) algorithm with a hybrid reinforcement learning mechanism for the multidimensional heterogeneous programming problem. The algorithm process is as follows: First, an initial multidimensional heterogeneous programming scheme is constructed; then, a high-quality planning scheme is quickly searched in the very large neighborhood; a heuristic receiving mechanism is designed to determine whether to replace the current planning scheme; subsequently, a hybrid reinforcement learning mechanism is used to slightly adjust the planning scheme. This process is repeated until the planning scheme meets the system requirements. The flowchart of the hybrid VLSN algorithm designed in this invention is as follows: Figure 7 As shown.

[0178] Several key steps in this process need to be designed specifically for the characteristics of this problem. The steps are as follows:

[0179] Initialization of multidimensional heterogeneous planning: This invention uses a greedy construction algorithm based on Dijkstra to generate an initial planning scheme. The specific method is as follows: First, based on the result of matching the difference between the meta-task and the execution starting point, an initial shortest execution path is assigned to each meta-task; then, some path segments are changed to improve path heterogeneity; then, the same approach is used to improve spectral band heterogeneity and temporal heterogeneity, and finally the initial planning scheme of multidimensional heterogeneous meta-task planning is obtained.

[0180] VLSN Improves Heterogeneity: The stepwise adjustment method for the current planning scheme is a key step in the VLSN algorithm to improve system heterogeneity, and the computational complexity of the stepwise adjustment method directly affects the system's computational overhead. A common approach to stepwise multidimensional transformation is as follows: extract a specific part of the plan from the current planning scheme, and simultaneously change the path, spectral band, and time of this part of the plan to improve overall heterogeneity. For example, considering a three-dimensional spatiotemporal system, changing the variables in each of the three dimensions requires computation... There are several selectable adjustment methods, among which The maximum number of available spectrum bands is given by , v is the total number of paths, and T is the total number of paths in time period T. The search range of this method grows exponentially.

[0181] To reduce computational complexity and system overhead, this invention employs reinforcement learning to guide neighborhood selection in the VLSN algorithm. The steps are as follows: Based on the execution status of the current meta-task, reinforcement learning is designed to determine the variable selection method for each dimension. When determining the selection method, reward and penalty values ​​are first given, and then the reward and penalty mechanism is combined with a roulette wheel selection strategy to select variables for each dimension.

[0182] Heuristic Acceptance Mechanism Design: The acceptance mechanism refers to the decision-making mechanism for whether to replace the current scheme with a newly generated heterogeneous programming scheme. This decision-making mechanism greatly affects the performance of the VLSN algorithm. In current research, the following two heuristic acceptance methods are commonly used: a) Simulated annealing strategy: If the newly generated scheme improves heterogeneity, it is accepted; otherwise, it is accepted according to the probability of the simulated annealing algorithm based on the situation; b) Tag strategy: If the newly generated scheme is taboo, it is not accepted; otherwise, it is accepted. This invention combines the tab and simulated annealing acceptance strategies and dynamically adjusts the acceptance method of new schemes based on the running effect. Specifically, in the early stages of the search, in order to ensure the diversity of the algorithm, the acceptance strategy tends to favor simulated annealing, and even if the generated new scheme does not significantly improve heterogeneity, it will be accepted with a high probability. As the search progresses, in order to prevent the algorithm from getting trapped in local optima, the acceptance strategy tends to favor tab, that is, only accepting new schemes that can significantly improve search efficiency and quality.

[0183] Reinforcement learning optimizes heterogeneity: This invention employs a reinforcement learning adjustment scheme, which enhances the global search capability of VLSN by deleting and inserting η variables; and dynamically adjusts the value of η based on the current heterogeneity and historical search information.

[0184] (iv) Dynamic dual feedback strategy

[0185] This invention proposes a dual-feedback strategy based on a hybrid algorithm. To further reduce system computational overhead, enhance the system's self-learning capability, and lower the false alarm rate, a reasonable system continuous service time and an integrated, efficient planning scheme are designed. The specific details are as follows:

[0186] (1) Positive feedback mechanism based on statistical learning

[0187] In the initial planning phase, the complex task decomposition process, task and execution node matching process, and the search for multi-dimensional heterogeneous execution methods all require significant system computational overhead. To adapt to future network demands, reduce the complexity and cost of system planning, avoid redundant searches during the hybrid algorithm solution process, and consider the reasonable online time of some nodes in the system, this invention proposes a positive feedback mechanism based on statistical learning. The main content is as follows: During network operation, excellent task decomposition methods, matching methods, and execution schemes with high availability and low system overhead are stored; an efficient statistical probability model is constructed to learn the aforementioned excellent planning methods, thereby reducing the computational load for the next allocation. Specifically, when the current planning method is the same as the previous one, the probability of using that planning method in the next allocation is increased; conversely, the probability of using that planning method in the next allocation is decreased. Furthermore, the probability of planning methods exceeding the continuous service time threshold θ1 being selected is reduced to ensure the system's dynamism. By continuously utilizing effective information from previous planning, the system can quickly obtain high-quality solutions.

[0188] This invention will integrate the superior allocation components of each module based on previous statistical learning methods, design a positive feedback mechanism for task planning, improve the system planning speed, and thus realize an efficient planning scheme for the integrated intrinsic safety system.

[0189] (2) Negative feedback mechanism based on tabu algorithm

[0190] Meanwhile, to address the drawbacks of positive feedback on intrinsically safe planning and avoid the long-term adoption of the same execution strategy, this invention designs a negative feedback system based on the tabu algorithm. This system cleans up planning methods with continuous service times exceeding θ², making the system exhibit uncertainty externally. Simultaneously, based on the execution results and combined with a multi-mode adjudication mechanism, abnormal network nodes are forcibly taken offline, promptly cleaning up abnormal nodes to ensure system availability. The basic idea of ​​the tabu algorithm is to prohibit certain task allocation schemes from reappearing in subsequent iterations; therefore, the tabu algorithm is naturally suitable for intrinsically safe negative feedback systems. The key to the tabu algorithm lies in the design of the tabu component and the tabu length. This invention comprehensively weighs the network environment, execution overhead, replacement cost, and the security and heterogeneity of this meta-task execution method to design the tabu component in the negative feedback and optimize the tabu length design.

[0191] This invention organically combines an inherently secure and available collaborative architecture with task planning, focusing on differences in nodes and tasks within the architecture. While ensuring system availability, it achieves optimal planning of inherently secure collaborative tasks, thereby extending the system's lifespan.

[0192] The technical means disclosed in this invention are not limited to those disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications are also considered within the scope of protection of this invention.

Claims

1. A method for collaborative mission planning for endogenous security of availability, characterized in that, Includes the following steps: Step 1: Irregular Redundancy Decomposition of Complex Tasks When a complex task is input, if a matching case can be directly searched from the case library, the current task is decomposed according to the previous decomposition method of the matching case; otherwise, the task is added to the case library. Step 2: Collaborative matching of differences between meta-tasks and execution starting points Based on the information from the irregular redundancy decomposition of complex tasks, we construct a difference matching model using the obtained meta-tasks and the information they carry, and design a hybrid mathematical heuristic FISA algorithm to solve this problem. Step two specifically includes the following steps: (1) Modeling and analysis of differential co-matching In the difference matching model, the meta task corresponds to the vertex of the topological structure graph in graph theory, and there is an edge between two meta tasks with conflict in execution time. The execution starting point corresponds to the color available for coloring. The model is described as follows: given a graph wherein is the set of meta tasks, is the set of conflicts between meta tasks, is the set of execution starting points; each meta task is matched with an execution starting point, so that the conflicting meta tasks are as few as possible, and the total amount of meta tasks for each execution starting point is as close to its capacity as possible; The objective function of the difference matching model first considers reducing system resource overhead and minimizing the total number of execution starting points of the calls, and its expression is: max (1) in Indicates the start point of execution. Whether a meta-task is assigned is expressed as follows: ,in This represents the matching relationship between meta-tasks and execution start points, expressed as: （2） Another objective function is to maximize the availability of the system, i.e. max (3) in, As the starting point of execution Availability; Formula (3) and the system are limited to selecting execution starting points with high availability as much as possible; The model problem has the following constraints: a) Node availability While considering the system's availability objective function, the availability metric of a node is set to be greater than a threshold. ,Right now （4） b) Task conflict relationships Conflicting meta-tasks are not matched to the same execution starting point, and their expression is: （5） in, and For two conflicting meta-tasks; c) Node capacity constraints The total memory required for the meta-tasks allocated to each execution starting point must not exceed the fixed memory capacity of that execution starting point. That is, weight balance, its expression is: （6） in Representing meta-tasks Assigned to the execution start point Required memory; d) Task and node matching relationship Metatasks can only be matched with the execution starting point that handles this type of task, and their expression is: （7） （8） （9） in, Representing resource types, Representative task Are resources required during execution? ,and This represents the starting point of execution. Does it have resources? If and only if and When all values ​​are 1, the task and execution start point Only then can a match be successfully made; e) Task and node coupling relationship Due to the various differences in execution starting points, the coupling benefits of matching meta-tasks with different execution starting points vary, and the benefit expression is as follows: （10） in, Represents the starting point of execution He Yuan Mission The coupling benefits, , , For parameters that are greater than 0, As the starting point of execution The amount of resources, As the starting point of execution Data processing time, As the starting point of execution throughput; f) Coupling between tasks Matching similar meta-tasks derived from the same complex task with the same execution starting point yields additional secondary benefits, expressed as follows: （11） in, For meta-task Matching the start point of execution The secondary profit value executed above, Used to represent a task and tasks Are they similar tasks? A return parameter that is greater than 0; (2) Design of a hybrid algorithm for differential collaborative matching A mixed integer programming FISA algorithm was designed to solve the model in step (1). First, an integer programming model was constructed to generate the objective function and constraints. Then, a greedy algorithm was used to construct the initial matching of the meta-task and the execution starting point. Then, the matching scheme was optimized and upgraded using the FISA algorithm. Subsequently, a heuristic strategy was used to repair the matching scheme that did not meet the constraints. Finally, the excellent matching scheme was recorded. The FISA algorithm for mixed integer programming includes the following specific steps: Generate initial matching scheme: 1) Select The execution starting point with the highest availability; 2) Use a greedy approach to sequentially select each execution starting point. Select the meta-task with the greatest gain. The formula for selecting the meta-task using the greedy algorithm is as follows: （12） in, For meta-task Matching the start point of execution The coupling benefit value at that time is obtained through formula (10). For meta-task Matching the start point of execution The secondary profit value obtained from the above execution is obtained through formula (11); FISA algorithm optimizes matching schemes: The FISA algorithm is used to optimize and upgrade matching schemes. If a better matching scheme cannot be found within a certain period of time, some hard constraints are relaxed to soft constraints based on the quality of the current matching scheme and the mechanism of reinforcement learning, until the evaluation quality of the relaxed matching scheme can no longer be improved. Heuristic repair matching scheme: Forcefully transfer meta-tasks that exceed the execution start point capacity and conflicting meta-tasks to other execution start points in a greedy manner; and after forced repair, further adopt FISA to optimize the matching scheme; Step 3: Employ a meta-task multi-dimensional heterogeneous planning mechanism First, a multidimensional heterogeneous programming problem model is constructed, and then a hybrid local search algorithm is designed to solve the problem. Step 4: Implement dynamic dual feedback A dual feedback strategy based on a hybrid algorithm is adopted to design a reasonable system continuous service time and achieve an integrated and efficient planning scheme.

2. The collaborative task planning method for intrinsic security and availability according to claim 1, characterized in that, In step one, the case library is collected through the following steps: surveying and collecting application scenarios and decomposition schemes that can reflect the characteristics of the problem, organizing and classifying the data to form an initial case library, and continuously revising the case library and cases during the process of case matching and task planning; the decomposition of complex tasks is implemented using the K-nearest neighbor algorithm, and the irregular redundant meta-tasks after decomposition are output according to the matching results.

3. The collaborative task planning method for intrinsic security and availability according to claim 2, characterized in that, The process of continuously revising the case library and cases is as follows: when the similarity between the current solution and the most similar case in the case library is less than a threshold, a new solution with a similarity balance between the current solution and the most similar case is generated, the current solution is revised into the new solution, and the revised solution is added to the case library.

4. The collaborative task planning method for intrinsic security and availability according to claim 1, characterized in that, Step one also includes a meta-task standardization step: The necessary information for a standardized metatask includes: metatask execution time, metatask end time, metatask target node, metatask resource types, and conflicts or coupling relationships between the metatask and other metatasks.

5. The collaborative task planning method for intrinsic security and availability according to claim 1, characterized in that, Step three specifically includes the following steps: (1) Establish a multidimensional heterogeneous programming problem model The integrated design incorporates multi-dimensional heterogeneous planning based on path, spectrum, and time dimensions, specifically including: The path heterogeneous programming problem is transformed into the weighted shortest path problem; the spectrum band heterogeneous programming problem is equivalent to the classic bandwidth graph coloring problem; the time heterogeneous programming problem is equivalent to the classic graph coloring problem in graph theory. remember The physical topology diagram of the network for executing meta-tasks, where A set of routing nodes in the network. Given a bidirectional edge set, we present an integer programming mathematical model for a multidimensional heterogeneous problem. Taking three dimensions as an example, the objective function is to minimize the maximum number of spectral bands traversed by the physical edges of the network at the same time. min (13) （14） Among the variables Indicates the first The time intervals are as follows: and network segments Connect to source node and destination node And the spectral band number is The number of; against Similarly, the availability of paths in the network is defined as... ,in This refers to the number of meta-tasks and also the number of requirement paths; the metric for maximizing the availability of each path is defined as follows. max (15) The previous spectral band spacing constraint of two-dimensional heterogeneity was expressed as （16） in Indicates the number of subcarriers. Indicates the size of the guard carrier. This represents a number greater than the number of metatasks. The constant; (2) Hybrid algorithm for solving multidimensional heterogeneous programming problems A super-large neighborhood search algorithm with a hybrid reinforcement learning mechanism is designed for the multidimensional heterogeneous programming problem. The algorithm includes: first, constructing an initial multidimensional heterogeneous programming scheme; then, quickly searching for a high-quality programming scheme in the super-large neighborhood; designing a heuristic receiving mechanism to determine whether to replace the current programming scheme; then, using a hybrid reinforcement learning mechanism to slightly adjust the programming scheme; and repeating this process until the programming scheme meets the system requirements.

6. The collaborative task planning method for intrinsic security and availability according to claim 5, characterized in that, The ultra-large neighborhood search algorithm includes the following steps: Initialize the multidimensional heterogeneous planning: A greedy construction algorithm based on Dijkstra is used to generate the initial planning scheme, which includes: First, based on the difference matching results between the meta-task and the execution starting point, the initial shortest execution path is assigned to each meta-task; then, some path segments are modified to improve the path heterogeneity; then, the spectral band heterogeneity and temporal domain heterogeneity are improved in the same way, and finally the initial planning scheme of multidimensional heterogeneous meta-task planning is obtained. VLSN enhances heterogeneity by making gradual adjustments to the current planning scheme. The multi-dimensional transformation method of gradual adjustment is as follows: extract a portion of the planning from the current planning scheme and change the path, spectrum band and time of that portion of the planning to improve the overall heterogeneity. In the VLSN algorithm, reinforcement learning is used to guide neighborhood selection: based on the execution status of the current meta-task, reinforcement learning is designed to determine the variable selection method for each dimension; when determining the selection method, a reward or penalty value is given first, and then the reward or penalty mechanism is combined with the roulette wheel selection strategy to select variables for each dimension. Heuristic reception mechanism design: Combine tabu and simulated annealing reception strategies, and dynamically adjust the reception method of the new scheme based on the results of operation; Reinforcement learning optimizes heterogeneity: Reinforcement learning is used to adjust the planning scheme, including deletion and insertion. This adds variables to enhance the global search capability of VLSN.

7. The collaborative task planning method for intrinsic security and availability according to claim 1, characterized in that, The dynamic adjustment method is as follows: when the current planning method is the same as the previous one, the probability of using the same planning method in the next one is increased; otherwise, the probability of using the same planning method in the next one is decreased.

8. The collaborative task planning method for intrinsic security and availability according to claim 1, characterized in that, Step four specifically includes the following steps: (1) Positive feedback mechanism based on statistical learning During network operation, we utilize superior task decomposition, matching, and execution schemes that offer high storage availability and low system overhead; we construct efficient statistical probability models and learn from these superior planning methods to reduce the computational load of subsequent allocations. In addition, reduce the threshold for continuous service time. The probability that the planning method is selected; (2) Negative feedback mechanism based on tabu algorithm For continuous service time exceeding The system employs a planned approach to clean up abnormal network nodes; based on the execution results and combined with a multi-mode adjudication mechanism, abnormal network nodes are forcibly taken offline, and abnormal nodes are cleaned up in a timely manner to ensure system availability.

Images