A test assurance system process flow resource matching and conflict identification method

Abstract

The present application relates to the technical field of intelligent planning and resource scheduling of test support system, and particularly relates to a test support system process flow resource matching and conflict identification method, comprising: constructing a structured analysis model of test demand and test plan, using a method combining sequence pattern mining algorithm and frequent item set mining algorithm to perform association rule mining, constructing a multi-tester process system resource configuration conflict matrix, constructing a task-resource adaptation model based on graph embedding, constructing a resource dynamic matching model based on multi-objective Markov decision, establishing a global optimization scheduling model under the condition of multiple test task series and parallel, constructing a power load and circulating water load collaborative matching method, constructing a valve-pipeline state conflict identification and path planning method, and generating an intelligent planning and scheduling scheme; the present application can realize automatic planning of test support process flow, dynamic allocation of resources, early identification of conflicts and intelligent generation of scheduling scheme.

Description

Technical Field

[0001] This invention relates to the field of intelligent planning and resource scheduling technology for test support systems, and in particular to a method for matching process resources and identifying conflicts in test support systems. Background Technology

[0002] The test support system is a collection of infrastructure that supports aero-turbine engines and various major test missions. It typically covers multiple types of resources, including gas supply, air extraction, air pretreatment, circulating water, fuel and natural gas, electricity, valves and pipelines, and measurement and control equipment.

[0003] Currently, test support systems have gradually shifted from a traditional manual experience-driven, static planning and local resource coordination model to a new intelligent support model driven by demand, model-driven, dynamic matching, intelligent scheduling, conflict early warning, and closed-loop optimization. With the increasing variety of test objects, the growing complexity of test conditions, and the enhanced parallelism of tasks, test support systems are highly coupled with various resource types, including gas supply, extraction, air pretreatment, circulating water, fuel oil and natural gas, electricity, valves and pipelines, and measurement and control equipment. Different test tasks exhibit significant temporal, exclusive, capacity-constrained, and safety boundary constraints in their resource usage. Any unreasonable allocation of a single resource can lead to test delays, resource waste, equipment conflicts, and even safety risks. Therefore, existing technologies suffer from capacity conflicts, state conflicts, path conflicts, temporal conflicts, and safety boundary conflicts during the parallel execution of multiple test tasks.

[0004] For the reasons mentioned above, there is an urgent need for a unified modeling method for complex test support scenarios under conditions of multiple testers, multiple parallel tasks, and multiple resource coupling. This method should be used to construct a unified modeling method for test requirements, test plans, process flows, equipment resources, and operating status, and to form a technology system for dynamic resource matching and intelligent conflict identification. This would enable automatic planning of test support process flows, dynamic allocation of resources, early identification of conflicts, and intelligent generation of scheduling schemes. Summary of the Invention

[0005] The purpose of this invention is to provide a method for matching process resources and identifying conflicts in a test support system. This method is designed for complex test support scenarios involving multiple testers, multiple parallel tasks, and multiple resource coupling. It constructs a unified modeling method for test requirements, test plans, process flows, equipment resources, and operating status, forming a dynamic resource matching and intelligent conflict identification technology system. This enables automatic planning of the test support process, dynamic allocation of resources, early identification of conflicts, and intelligent generation of scheduling schemes.

[0006] To achieve the above objectives, the present invention provides a method for matching process resources and identifying conflicts in a test support system, comprising:

[0007] S1 constructs a structured analytical model for test requirements and test plans: Based on the expert rule base, information is extracted and integrated from the test requirements and test plan data. After screening and correction, digital coding is carried out according to the different state expression rules of test requirements and test equipment. Systematic and standardized data modeling is performed on complex and diverse test requirements to generate standardized requirement information including test objects, test subjects, operating conditions, time constraints, explicit resource requirements, safety boundary constraints, and priorities.

[0008] S2 employs a combination of sequence pattern mining and frequent itemset mining algorithms for association rule mining: Based on previously aggregated test requirements, test plans, equipment ledgers, valve-pipeline topology, historical test records, real-time operating status, and expert knowledge, a unified test support knowledge data foundation is constructed. The SPADE sequence pattern mining algorithm and the Apriori frequent itemset mining algorithm are combined to generate association rules between resources such as gas supply, air extraction, air pretreatment, circulating water, electricity, and valves-pipelines and test requirements.

[0009] S3 constructs a resource allocation conflict matrix for a multi-testing device process system: Based on graph theory and state-space theory, the physical entity system is abstracted into a hierarchical dependent logical topological manifold. Through tensor coding and dynamic graph search algorithm, valve-pipeline state conflict matrix, test equipment state conflict matrix, power load conflict matrix, circulating water load conflict matrix and comprehensive resource allocation conflict matrix are constructed to realize the explicit expression of resource conflict relationships. The conflict matrix is ​​dynamically updated in a real-time data-driven manner.

[0010] S4 constructs a task-resource adaptation model based on graph embedding: test tasks, test resources, test equipment, process nodes, valves and pipelines, operating status and expert rules are uniformly modeled as a heterogeneous graph structure. The deep adaptation relationship between tasks and resources is mined through graph representation learning. The adaptation degree between tasks and resources is calculated based on node vector similarity and comprehensive adaptation scoring function to generate a candidate resource set.

[0011] S5 constructs a resource dynamic matching model based on multi-objective Markov decision-making: the resource matching problem of complex process experiment is modeled as a multi-objective Markov decision process, defining the state space, action space and multi-objective reward function, and adopting a hybrid solution approach that combines rule constraints and reinforcement learning, and realizing dynamic resource matching through state perception, action selection, reward feedback and policy update;

[0012] S6 establishes a global optimization scheduling model under the condition of multiple experimental tasks running in parallel: taking into account the logical relationship of tasks, time window, resource capacity, equipment status, conflict matrix, security isolation and task priority, with the optimization objectives of the shortest total completion time, balanced resource load, minimum conflict risk, minimum task delay and minimum resource usage cost, and adopting a solution strategy that combines mixed integer programming and heuristic algorithm to generate an initial scheduling scheme.

[0013] S7 constructs a method for coordinating matching of power load and circulating water load: Incorporate the power system and circulating water system into a unified scheduling model, define power-circulating water coupling risk indicators, and identify load peaks in advance during the test plan generation stage and perform peak shaving and valley filling in real time during the execution process through load forecasting, coupling risk assessment and dynamic scheduling.

[0014] S8 constructs a valve-pipeline state conflict identification and path planning method: The valve-pipeline system is modeled as a directed graph with attributes, valve state conflict function and pipeline conflict function are defined, and process path planning is performed based on multi-constraint shortest path algorithm. Under the premise of satisfying connectivity, media compatibility, pressure constraints, flow constraints and valve state constraints, the optimal path with the minimum comprehensive cost is selected.

[0015] S9 generates intelligent planning and scheduling solutions: It transforms the technical achievements of demand analysis, association rule mining, conflict matrix identification, dynamic resource matching, and global optimization scheduling into executable plans, generating intelligent planning solutions that include task execution sequences, task time arrangements, resource allocation schemes, valve-pipeline routes, equipment status sequences, conflict risk assessments, emergency adjustment plans, and execution monitoring indicators. Through simulation, it performs load assessment, conflict risk verification, and task completion time verification of the solutions, and iteratively optimizes them based on feedback results.

[0016] The present invention provides a method for matching process resources and identifying conflicts in a test support system. This method is designed for complex test support scenarios involving multiple testers, multiple parallel tasks, and multiple resource coupling. It constructs a unified modeling method for test requirements, test plans, process flows, equipment resources, and operating status, forming a dynamic resource matching and intelligent conflict identification technology system. This enables automatic planning of the test support process, dynamic allocation of resources, early identification of conflicts, and intelligent generation of scheduling schemes. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0018] Figure 1 This is a flowchart of steps S1-S5 of the process flow resource matching and conflict identification method of the test support system of the present invention.

[0019] Figure 2 This is a flowchart of steps S6-S9 of the process flow resource matching and conflict identification method of the test support system of the present invention.

[0020] Figure 3 This is an architecture diagram of the process flow resource matching and conflict identification method of the test support system of the present invention.

[0021] Figure 4 This is a schematic diagram of the heterogeneous configuration of the task-resource-equipment-valve-pipeline of the present invention.

[0022] Figure 5 This is a flowchart illustrating the dynamic resource matching process under changing requirements and equipment conditions according to the present invention.

[0023] Figure 6 This is a schematic diagram comparing the power and circulating water load before and after the optimization of the present invention. Detailed Implementation

[0024] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, but should not be construed as limiting the present invention.

[0025] Please see Figures 1-6 , Figure 1 This is a flowchart of steps S1-S5 of the process flow resource matching and conflict identification method for the test support system of the present invention; Figure 2 This is a flowchart of steps S6-S9 of the process flow resource matching and conflict identification method for the test support system of the present invention; Figure 3 This is an architecture diagram of the process flow resource matching and conflict identification method of the test support system of the present invention; Figure 4 This is a schematic diagram of the heterogeneous configuration of the task-resource-equipment-valve-pipeline of the present invention; Figure 5 This is a flowchart illustrating the dynamic resource matching process under changing requirements and equipment conditions according to the present invention. Figure 6 This is a schematic diagram comparing the power and circulating water load synergistic optimization before and after the present invention. Figure 6 In the diagram, the vertical axis “load percentage (%)” represents the ratio of the actual system load to the maximum allowable load at the current moment; the horizontal axis “time (moment)” represents the typical operating period within a day.

[0026] This invention provides a method for matching process resources and identifying conflicts in a test support system, including:

[0027] S1 constructs a structured analytical model for test requirements and test plans: Based on the expert rule base, information is extracted and integrated from the test requirements and test plan data. After screening and correction, digital coding is carried out according to the different state expression rules of test requirements and test equipment. Systematic and standardized data modeling is performed on complex and diverse test requirements to generate standardized requirement information including test objects, test subjects, operating conditions, time constraints, explicit resource requirements, safety boundary constraints, and priorities.

[0028] S2 employs a combination of sequence pattern mining and frequent itemset mining algorithms for association rule mining: Based on previously aggregated test requirements, test plans, equipment ledgers, valve-pipeline topology, historical test records, real-time operating status, and expert knowledge, a unified test support knowledge data foundation is constructed. The SPADE sequence pattern mining algorithm and the Apriori frequent itemset mining algorithm are combined to generate association rules between resources such as gas supply, air extraction, air pretreatment, circulating water, electricity, and valves-pipelines and test requirements.

[0029] S3 constructs a resource allocation conflict matrix for a multi-testing device process system: Based on graph theory and state-space theory, the physical entity system is abstracted into a hierarchical dependent logical topological manifold. Through tensor coding and dynamic graph search algorithm, valve-pipeline state conflict matrix, test equipment state conflict matrix, power load conflict matrix, circulating water load conflict matrix and comprehensive resource allocation conflict matrix are constructed to realize the explicit expression of resource conflict relationships. The conflict matrix is ​​dynamically updated in a real-time data-driven manner.

[0030] Step S3, constructing the valve-pipeline state conflict matrix specifically includes:

[0031] Using tensor framework thinking, a node-link dynamic association model is established, which highly abstracts the complex test pipeline system into a weighted directed graph with time constraints. In this graph, nodes represent control terminals in the logical space, and edges represent fluid transmission paths that are dynamically connected according to the valve status.

[0032] Construct a state topology conflict matrix. This matrix is ​​a two-dimensional square matrix. The matrix element assignment logic is as follows: the diagonal elements represent the physical state of the node itself, and the off-diagonal elements are assigned a conflict flag when there is a physical connection between two nodes and the expert Boolean function based on the process rules determines that the combination of their states triggers the risk of mutual exclusion of flow. Otherwise, they are assigned a no-conflict flag.

[0033] The system recursively accesses the valve nodes starting from the initial valve node using a depth-first search algorithm. It comprehensively determines the high-dimensional nonlinear relationship between valves and performs conflict detection based on the current global state vector and the target test path sequence. When it detects that two nodes are both in the open state and the process rules determine that they are mutually exclusive, it determines that there is a logical-level resource access violation and records the conflict association pair.

[0034] For the initial settings when starting up multiple testers, the local optimal state vector that satisfies the target pressure or target flow rate is solved based on the conflict matrix, so that the system can quickly enter a feasible initial operating state;

[0035] Given the highly time-varying characteristics of aerodynamic systems, when the underlying control system detects valve lag or sudden changes in equipment state, it uses a sliding time window update equation to dynamically reshape the conflict matrix, ensuring that the matrix undergoes accurate state evolution as the system operates. This provides a highly real-time basic dataset and a rigorous spatiotemporal safety envelope for subsequent high-dimensional optimization.

[0036] S4 constructs a task-resource adaptation model based on graph embedding: test tasks, test resources, test equipment, process nodes, valves and pipelines, operating status and expert rules are uniformly modeled as a heterogeneous graph structure. The deep adaptation relationship between tasks and resources is mined through graph representation learning. The adaptation degree between tasks and resources is calculated based on node vector similarity and comprehensive adaptation scoring function to generate a candidate resource set.

[0037] In step S4, constructing the task-resource adaptation model based on graph embedding specifically includes:

[0038] The test tasks, resources, equipment, process flow, valve-pipeline, and operating status are uniformly modeled as a heterogeneous graph structure. The set of nodes in the graph is divided into test task nodes, resource nodes, equipment nodes, process flow nodes, valve-pipeline nodes, and operating status nodes. The set of edges includes task-resource requirement edges, task-equipment occupation edges, resource-equipment support edges, equipment-state constraint edges, task-process dependency edges, process-valve-pipeline control edges, resource-resource conflict edges, and resource-resource substitution edges. Through unified modeling, the information scattered in the test plan, equipment ledger, pipeline diagram, expert rule base, and historical test records is transformed into a computable network structure.

[0039] The Node2vec graph embedding algorithm is used to learn the low-dimensional vector representation of nodes in the graph. By adjusting the return parameters and input / output parameters, a balance is achieved between breadth-first search and depth-first search. At the same time, the local similarity and structural similarity of nodes are captured. For the experimental resource matching problem, when it is necessary to identify direct resource adaptation relationships, a larger input / output parameter is appropriately used to enhance local neighborhood learning. When it is necessary to discover alternative resources or potential resource combinations, the input / output parameter is appropriately reduced to enhance cross-regional exploration capabilities.

[0040] The goal of node embedding learning is to bring closely related or semantically similar nodes in the graph closer together in the low-dimensional vector space. This is achieved through approximate optimization using negative sampling to obtain the vector representation of each task node, resource node, and device node.

[0041] For task nodes and resource nodes, cosine similarity is used to calculate their compatibility, and a comprehensive compatibility scoring function is further constructed. This score comprehensively considers vector similarity, resource availability at the current moment, the degree to which resource capacity meets task requirements, the degree of conflict between resources and other assigned tasks, resource usage cost, and resource security risk. The weights of each item are dynamically adjusted according to the task type.

[0042] The system employs a four-stage matching process: candidate resource screening, comprehensive scoring and ranking, conflict matrix verification, and alternative resource recommendation. The first stage filters the candidate resource set based on the vector similarity between task nodes and resource nodes. The second stage performs comprehensive scoring and ranking on the candidate resources. The third stage inputs the candidate matching results into the resource configuration conflict matrix for verification. In the fourth stage, if the optimal resource is unavailable, alternative resource combinations are recommended based on alternative edges and similarity ranking. This system supports rapid resource rematching in dynamic scenarios such as equipment failure, resource deduction, and task insertion.

[0043] S5 constructs a resource dynamic matching model based on multi-objective Markov decision-making: the resource matching problem of complex process experiment is modeled as a multi-objective Markov decision process, defining the state space, action space and multi-objective reward function, and adopting a hybrid solution approach that combines rule constraints and reinforcement learning, and realizing dynamic resource matching through state perception, action selection, reward feedback and policy update;

[0044] In step S5, the resource dynamic matching model based on multi-objective Markov decision specifically includes:

[0045] The resource matching problem in complex process flow experiments is modeled as a multi-objective Markov decision process. The model includes a state space, an action space, state transition probabilities, a multi-objective reward vector, a discount factor, and a set of objective preference weights. At each decision moment, the system observes the current state, selects an action, the environment transitions to a new state, and provides feedback on multi-objective rewards. Through continuous iteration, the system learns resource allocation strategies under different experimental scenarios.

[0046] The state space is defined as a combination of test task state, resource state, test equipment state, pipeline state, system load state, current resource conflict matrix, and historical decision information. The task execution state includes whether the task has started, the completion rate, the remaining time, priority, and time window. The resource state includes the available capacity of gas supply, extraction, air pretreatment, circulating water, fuel oil and natural gas, and power resources. The test equipment state includes available, occupied, faulty, under maintenance, and derated operation. The valve-pipeline state includes open / closed status, pressure, flow rate, medium, and path connectivity.

[0047] The action space includes allocating resources to tasks, delaying low-priority tasks, executing tasks that meet the conditions in advance, switching resource paths, enabling alternative resources, adjusting load, isolating resources or tasks with risks, and triggering process replanning.

[0048] To ensure that the action output meets the engineering safety requirements, rule constraints are superimposed on the action space. Actions that do not meet the safety constraints, such as pressure not exceeding limits, valve states not mutually exclusive, equipment not overloaded, power load not exceeding limits, and circulating water capacity not exceeding limits, are eliminated. The reinforcement learning agent selects actions only from the set of safe actions to avoid generating unacceptable dangerous strategies during the exploration process.

[0049] The multi-objective reward function takes into account task efficiency, resource efficiency, cost, and security. The reward vector includes task completion reward, resource utilization reward, cost penalty, conflict penalty, security penalty, and strategy stability reward. Each objective is scalarized and weighted by a preference weight vector, which can be dynamically adjusted in different scenarios.

[0050] A hybrid solution approach combining rule constraints and reinforcement learning is adopted: First, an expert rule base and conflict matrix are used to filter out unsuitable actions to ensure policy safety. Second, an Actor-Critic framework is used to learn resource matching strategies, where the Actor network outputs action probabilities and the Critic network evaluates the value of the current strategy. Third, a rolling temporal optimization method is used to regenerate the local optimal strategy based on the latest state within each decision window. At the same time, an experience replay mechanism, a priority sampling mechanism, and an online fine-tuning mechanism are introduced to improve the model's adaptability in non-stationary environments.

[0051] S6 establishes a global optimization scheduling model under the condition of multiple experimental tasks running in parallel: taking into account the logical relationship of tasks, time window, resource capacity, equipment status, conflict matrix, security isolation and task priority, with the optimization objectives of the shortest total completion time, balanced resource load, minimum conflict risk, minimum task delay and minimum resource usage cost, and adopting a solution strategy that combines mixed integer programming and heuristic algorithm to generate an initial scheduling scheme.

[0052] In step S6, the global optimization scheduling model under the condition of multiple experimental tasks running in parallel specifically includes:

[0053] Assume that the set of experimental tasks and the set of resources are determined. Each task has attributes such as duration, earliest start time, latest finish time, priority, resource requirements, and security level. The start and end times of the tasks satisfy their respective time window constraints.

[0054] If one task is a prerequisite task for another task, the start time of the subsequent task must not be earlier than the end time of the prerequisite task plus the required safety interval or process transition time between the two. The depressurization, purging and confirmation time when switching from a high-pressure gas supply task to a pipeline maintenance task, as well as the cleaning and purging time when switching tasks, are all included in the scheduling model as safety intervals.

[0055] The resource capacity constraint is that the total amount of the same resource occupied by all tasks being executed at any given time shall not exceed the available capacity of that resource at that time. For power resources, the sum of the power demand of all tasks being executed at any given time shall not exceed the system power limit. For circulating water resources, the sum of the circulating water demand of all tasks being executed at any given time shall not exceed the system circulating water limit.

[0056] For two tasks sharing the same valve, if their valve status requirements are mutually exclusive, the two tasks must not be executed in parallel at the same time. If they share the same pipeline, the total flow rate must not exceed the pipeline capacity. If the two media are incompatible, task isolation or cleaning and purging time must be set.

[0057] The overall optimization objectives are to minimize the total completion time, minimize the deviation of each resource load from the expected utilization rate, minimize the risk of conflicts between tasks, minimize the resource usage cost, and minimize the task delay penalty. The task delay penalty is related to the task priority and delay duration, and a higher delay penalty is set for high priority tasks.

[0058] The solution strategy adopted is a hierarchical partitioning-rolling time domain-heuristic optimization: First, the experimental task is hierarchically divided according to the process stage and resource type, and the global problem is broken down into three sub-problems: task scheduling, resource allocation, and control path planning. Second, the rolling time domain method is used to divide the long-cycle scheduling into multiple decision windows and perform local optimization in each window. Finally, the initial scheme is perturbed and optimized using genetic algorithms, simulated annealing, or tabu search to gradually reduce resource conflicts and load peaks.

[0059] Task priority is determined by a combination of factors, including the task's importance, urgency, dependence on subsequent tasks, security risk level, and cost of task interruption. The higher the priority, the more the scheduling model tends to guarantee its resource supply and time window.

[0060] S7 constructs a method for coordinating matching of power load and circulating water load: Incorporate the power system and circulating water system into a unified scheduling model, define power-circulating water coupling risk indicators, and identify load peaks in advance during the test plan generation stage and perform peak shaving and valley filling in real time during the execution process through load forecasting, coupling risk assessment and dynamic scheduling.

[0061] In step S7, the method for coordinated matching of electrical load and circulating water load specifically includes:

[0062] The system calculates the total power load and total circulating water load of all tasks currently in operation in real time, and both must not exceed their respective system limits.

[0063] The power-circulating water coupling risk index is defined as the weighted product of the power load rate and the circulating water load rate. The weight is used to adjust the sensitivity of power and circulating water risks. In actual operation, even if neither of them exceeds their respective upper limits, if they are close to the upper limits at the same time, it will lead to insufficient system redundancy and reduced anti-disturbance capability. Therefore, when the coupling risk index exceeds the preset threshold, it is determined that there is a high load coupling risk and the cooperative scheduling strategy is triggered.

[0064] Set constraints on the rate of change of power load and the rate of change of circulating water load to avoid the risk of equipment start-up shock or cooling response lag caused by the simultaneous start-up of multiple high-power equipment or the simultaneous entry of multiple high heat load tasks into the steady state phase.

[0065] The coordinated scheduling strategy includes strategies for peak power load and strategies for peak circulating water load: On the power side, measures include staggered startup of high-power equipment, postponement of low-priority auxiliary tasks, activation of backup power supply paths or zoned power supply, and setting soft-start curves according to equipment operating characteristics to reduce startup impact; On the circulating water side, measures include switching or paralleling backup cooling branches, pre-starting the circulating water system for pre-cooling, reducing the heat load of non-critical equipment, adjusting the steady-state operating time of high heat load tasks to avoid the superposition of multiple high heat tasks, and optimizing the water flow path to reduce local pipeline congestion when pipeline conditions permit.

[0066] By using load forecasting, coupled risk assessment, and dynamic scheduling, load peaks are identified in advance during the test plan generation phase, and peak shaving and valley filling are carried out in real time during execution. This integrates two originally independently managed resource systems into a unified scheduling model, avoiding local optimization problems such as power availability but insufficient cooling or sufficient cooling but power supply exceeding limits.

[0067] S8 constructs a valve-pipeline state conflict identification and path planning method: The valve-pipeline system is modeled as a directed graph with attributes, valve state conflict function and pipeline conflict function are defined, and process path planning is performed based on multi-constraint shortest path algorithm. Under the premise of satisfying connectivity, media compatibility, pressure constraints, flow constraints and valve state constraints, the optimal path with the minimum comprehensive cost is selected.

[0068] In step S8, the valve-pipeline state conflict identification and path planning method specifically includes:

[0069] The valve-pipeline system is modeled as a directed graph with attributes. The graph contains valve nodes, pipe nodes, branch nodes, confluence nodes, interface nodes, and equipment inlet nodes, as well as edges representing pipes. Nodes have attributes such as valve number, open / closed status, control authority, response time, reliability, and safety level. Edges have attributes such as length, pipe diameter, medium type, pressure rating, flow limit, temperature range, flow direction restriction, and cleaning requirements.

[0070] The process path of the test task is represented as an alternating sequence of nodes and edges from the source node to the target node. The path must satisfy connectivity constraints, media compatibility constraints, pressure constraints, flow constraints, and valve status constraints. That is, all pipelines on the path are allowed to transport the media type required by the task, the pressure required by the task does not exceed the pressure limit of all pipelines on the path, the sum of the flow rates of multiple tasks sharing the same pipeline does not exceed the capacity of the pipeline, and all valves on the path remain in the state required by the task during the execution of the task.

[0071] If multiple tasks share the same valve, and the opening and closing requirements of each task for the valve are inconsistent and mutually exclusive, then a valve state conflict is determined. If multiple tasks share the same pipeline, the pipeline conflict risk is comprehensively determined from five dimensions: media type compatibility, pressure level matching, whether the flow superposition exceeds the limit, whether the flow direction is consistent, and whether cleaning and purging are required during switching. The comprehensive pipeline conflict value is obtained by weighted summation of each dimension.

[0072] If the combined conflict value of two tasks sharing a path exceeds a preset threshold, the two tasks must not be executed in parallel, or the path of one of the tasks must be replanned. Traditional process planning may only consider whether resources exist and ignore whether resources can be delivered to the target equipment through reachable paths. Through valve-pipeline diagram modeling and path conflict identification, the system can determine whether the process path is feasible before scheduling.

[0073] Under the premise of satisfying all process constraints, the optimal path is selected as the path that minimizes the combined costs of pipeline length, valve operation, path risk, path conflict, and switching cost. For real-time path planning, an improved shortest path algorithm or a multi-constraint shortest path algorithm is used. For simultaneous path planning of multiple tasks, a multi-device flow model or a conflict-aware graph search method is used.

[0074] S9 generates intelligent planning and scheduling solutions: It transforms the technical achievements of demand analysis, association rule mining, conflict matrix identification, dynamic resource matching, and global optimization scheduling into executable plans, generating intelligent planning solutions that include task execution sequences, task time arrangements, resource allocation schemes, valve-pipeline routes, equipment status sequences, conflict risk assessments, emergency adjustment plans, and execution monitoring indicators. Through simulation, it performs load assessment, conflict risk verification, and task completion time verification of the solutions, and iteratively optimizes them based on feedback results.

[0075] In step S9, the intelligent planning and scheduling scheme generation method specifically includes:

[0076] The intelligent planning scheme is defined as a comprehensive planning object that includes task execution sequence, task time arrangement, resource allocation scheme, valve-pipeline path, equipment and control node status sequence, conflict risk assessment, emergency adjustment plan and execution monitoring indicators;

[0077] The scheme generation process is divided into eight consecutive steps: The first step is to analyze the test requirements, which involves converting natural language, task forms, test outlines, scheduling instructions, or manually entered information into a structured requirement vector that the system can recognize and calculate; the second step is to generate candidate process flow nodes based on the demand-planning mapping model; the third step is to call the association rule base to complete the implicit resource requirements; the fourth step is to construct a candidate resource set and calculate the task-resource suitability through a graph embedding model; the fifth step is to input the candidate resources into the conflict matrix for verification and eliminate schemes with hard conflicts; the sixth step is to generate an initial schedule using mixed integer programming and heuristic algorithms; the seventh step is to evaluate resource load, conflict risk, and task completion time based on simulation; and the eighth step is to output a visualized schedule. Iterative optimization is achieved between each step through a feedback mechanism.

[0078] The simulation is based on the timeline and simulates the execution process of each task node according to the planned sequence. It calculates the occupancy, load level, conflict risk and safety margin of various resources in each time slice. The simulation evaluation indicators include task completion time, resource utilization rate, load peak rate, cumulative conflict risk value and task delay rate. The above indicators are used to determine whether there are resource bottlenecks and risk concentration areas in the initial schedule. If the power load and circulating water load are found to be close to the upper limit at the same time, peak shifting is carried out. If the critical valves are frequently switched, the valve state sequence is optimized to reduce operational risks. If the equipment runs continuously for too long, cooling or inspection nodes are added.

[0079] The intelligent scheduling output has multi-view expression capabilities, including a task view showing the start time, end time, execution stage and dependencies of each test task; a resource view showing the occupancy of various resources at different times and load curves; a conflict view showing the degree of conflict between tasks, resources and paths; a control view showing the valve opening and closing sequence and pipeline path; and a risk view showing the risk levels of power, circulating water, equipment and safety boundary. Through multi-view linkage, test personnel can intuitively understand the system-generated scheme and manually confirm and adjust it as needed.

[0080] To better understand this invention, the entire implementation process and principle of the method of this invention are described below. The specific steps of the experimental support system process resource matching and conflict identification method are as follows:

[0081] The S1 expert rule base, developed by testing experts based on years of experience, covers common test requirement expression standards, key information judgment criteria, and test equipment status identification rules. For the extraction of specific test equipment parameters, the rule base specifies standard formats for parameter names, units, and value ranges, ensuring the accuracy and compliance of extracted test requirement information with industry standards. To optimize test system scheduling and maximize resource utilization, the primary task is to extract and integrate test requirement and test plan data based on the expert rule base. After screening and correction, digital coding is carried out according to different status expression rules for test requirements and test equipment, resulting in systematic and standardized data modeling of complex and diverse test requirements. For the front-end air supply and air handling system of the test apparatus, based on the analysis of the test process, test requirements are extracted according to three factors: flow rate, temperature, and pressure. Taking high-altitude chamber testing as an example, the test types are mainly divided into five categories: low-pressure direct air supply test, medium-pressure direct air supply test, dry air supply test, negative temperature air supply test, and high temperature air supply test.

[0082] S2 This invention, based on the prior aggregation of test requirements, test plans, equipment ledgers, valve-pipeline topology, historical test records, real-time operating status, and expert knowledge, explores the construction of a unified test support knowledge data foundation. Based on the mapping relationship between test requirements and test plans, it adopts a method combining the SPADE sequence pattern mining algorithm and the Apriori frequent itemset mining algorithm to generate association rules between resources such as gas supply, air extraction, air pretreatment, circulating water, electricity, and valves-pipelines and test requirements.

[0083] In the S3 aero-turbine engine air test system, facing complex scenarios with multiple test subjects and concurrent operating conditions, the strong coupling and spatiotemporal competition of internal pipeline resources are the core bottlenecks restricting the efficiency of automatic planning. Traditional static resource checking paradigms assume fixed topological relationships and cannot cope with sudden changes in airflow direction caused by dynamic switching of valve group states. Therefore, this project innovatively proposes a method for constructing a multidimensional heterogeneous resource allocation conflict matrix based on graph theory and state-space theory. This method abstracts the physical entity system into a hierarchically dependent logical topological manifold. Through tensor coding and dynamic graph search algorithms, it achieves accurate mathematical quantization and real-time control of all elements and processes of conflicts related to control nodes (valves), pipeline boundaries, and the states of key compressor units. Specifically, the valve-pipeline flow direction conflict matrix is ​​constructed based on graph theory topological analysis:

[0084] In air duct networks, the opening and closing combinations of valves at control nodes directly define the physical flow direction of the airflow medium and the system boundary. This project utilizes tensor framework thinking to establish a node-link dynamic relationship model, highly abstracting the complex test pipeline system into a time-constrained weighted directed graph:

[0085]

[0086] Among them, node set Edge set represents the control terminal in the logical space. This represents the fluid transport path that is dynamically open depending on the valve state. To quickly detect mutual exclusions in multi-channel concurrent flow, the system is designed with... State topology conflict matrix Matrix elements The mapping assignment logic is defined as a piecewise constraint function:

[0087]

[0088] in, This represents the element in the i-th row and j-th column of the state topology conflict matrix; i and j represent the indices of the valve or control node, respectively. , These represent the physical states of node i and node j at the current moment, respectively. E represents the directed edge connecting node i to node j; E represents the set of edges in the valve-pipe topology. This represents an expert Boolean decision function established based on process rules, used to determine whether the state combination of node i and node j satisfies safety logic constraints.

[0089] in, This is an expert Boolean function compiled based on process rules. When positive conflicts exist (such as valves)... During gas supply, the valve acts as a bypass. When absolute locking is required to prevent pressure loss, matrix elements are forced to be written with -1.

[0090] Using depth-first search (DFS) and other traversal algorithms, the system recursively visits the starting valve node, comprehensively and systematically determining the high-dimensional nonlinear relationships between valves. For the initial settings during the startup of multiple test chambers, the system solves for the target pressure based on the aforementioned matrix. or traffic Local optimal state vector This information is then recorded in a matrix, enabling the system to quickly enter a feasible initial operating state.

[0091] In this embodiment, a depth-first search method is used to detect state conflicts in the dynamic topology of valves. First, a dynamically weighted directed graph is constructed using valves and control nodes as nodes, and the pipeline connection relationships and logical control relationships between valves as directed edges. The current global state vector for recording the real-time opening and closing states of each valve, as well as the sequence of target test paths to be detected, are obtained. A conflict matrix is ​​established based on physical pipeline connectivity and process Boolean constraints. This conflict matrix records the mutual exclusion relationships, interlocking relationships, and prohibition of simultaneous opening relationships between different valve nodes. The target test path is mapped to the dynamically weighted directed graph, and the corresponding source node is determined. A depth-first traversal is then performed starting from this source node. During the traversal, each forward neighbor of the current node is checked sequentially. When the conflict matrix indicates a mutual exclusion relationship between the current node and its neighbor, and the current global state vector indicates that the neighbor is in an open state, a conflict is determined in the current valve state combination. The current node and its neighbor are recorded as a conflict pair, and a global conflict flag is output. When no conflict is detected and the neighbor has not yet been visited, the recursive traversal of that neighbor continues. If a conflict is detected during the traversal, the valve state is adjusted or the process path is reconstructed based on the conflict association; if no conflict is detected after traversing the relevant nodes, the detection result that no valve dynamic topology conflict was found in the current target test path is output.

[0092] Dynamic evolution of conflict superposition in key gas source equipment and state matrix:

[0093] Beyond the underlying valve topology mutual exclusion, the core conflict of multiple testers operating concurrently is more deeply manifested in the competition between total pressure fluctuation and total mass flow rate in large-scale pipeline networks. Therefore, based on Boolean logic topology, the system introduces parameter sensitivity modeling, superimposing continuity constraints into the conflict matrix system.

[0094] Based on a comprehensive evaluation of test conditions and valve status, a locally optimal strategy is employed to quickly determine the valve configuration, enabling the system to reach a feasible operational state as soon as possible. When considering the impact of multiple valves on different testers, priority is given to configuring the valves that have the greatest impact on the startup of critical testers, determining the corresponding valve on / off state based on their required pressure or flow conditions. During the configuration process, valve status information is recorded in real-time in the conflict matrix for subsequent status updates and conflict detection.

[0095] Conflict Detection and Resolution: Utilizing the constructed conflict matrix, the Depth-First Search (DFS) algorithm is employed to detect the existence of conflicts in the current valve state. Starting from a primary valve node, the DFS recursively explores along the valve connections, marking visited valves until a conflict is found or all related valves have been traversed. During the exploration, the relationships between valve pairs are determined based on the element values ​​in the conflict matrix. If a conflict is found, meaning two mutually constraining valves are simultaneously in an unreasonable state, adjustments are made according to a pre-defined conflict resolution strategy. This strategy is based on a priority ranking algorithm, comprehensively considering factors such as the importance of the test equipment, the urgency of the task, and the reliability of the valves. Priority is given to ensuring the stability of valves on critical test equipment, and conflicts are eliminated by adjusting the states of other related valves. Valves on critical test paths are given higher priority, and when a conflict occurs, the states of valves on non-critical paths are adjusted first. During the adjustment process, a step-by-step trial-and-error method is used. Based on the dynamic characteristic curves of the valves, the impact of different adjustment amounts on the system state is calculated, and the adjustment scheme with the least impact on the overall system is selected.

[0096] Status Update and Feedback: After each valve status adjustment, the valve status information in the conflict matrix is ​​updated promptly to ensure that the matrix always reflects the actual state of the current test system. Through a message queue mechanism, using a RabbitMQ or ActiveMQ-based message queue system, the adjustment results are fed back to the intelligent test plan scheduling system. The message queue ensures the asynchronous and reliable transmission of data, avoiding test plan chaos caused by data transmission delays or losses. The feedback information includes not only the updated valve status but also detailed information such as the reason for the adjustment, the adjustment time, and the potential impact on the test plan, enabling the scheduling system to fully understand system changes and make corresponding optimizations and adjustments.

[0097] Based on the information coding table of test equipment, the operating status parameters of key test equipment such as compressor units are comprehensively analyzed. A test equipment status conflict matrix is ​​constructed using different combinations of key equipment statuses as rows and columns of a matrix to quantify the conflict between equipment statuses. Based on the association rules between the constructed process flow and test requirements, a feasibility analysis model for key equipment based on the conflict matrix is ​​established. This model breaks through the traditional equipment feasibility analysis method and can more comprehensively and accurately judge the feasibility of key equipment such as compressor units under different test conditions, providing a more scientific basis for test path planning.

[0098] Since the operating status of the test equipment is dynamically changing, the test equipment state conflict matrix also needs to be dynamically updated. A real-time data processing and analysis module is established to compare the equipment state parameters with the state combinations in the matrix. When the equipment state changes, the corresponding element values ​​in the matrix are updated in a timely manner to reflect the current conflict situation of the equipment. At the same time, with the continuous accumulation of test data and in-depth exploration of equipment operating patterns, the matrix's dimensional division and element assignment rules are periodically optimized and adjusted to improve the matrix's accuracy and practicality.

[0099]

[0100] in, This represents the set of concurrent trials that are currently active; k represents... The kth trial in; This represents the mass flow rate used or consumed by the k-th experiment at time t; This represents the maximum allowable mass flow rate of the pipeline network under the conditions of total pressure P and total temperature T; P represents the total pressure of the pipeline network, T represents the total temperature of the pipeline network, and t represents the operating time.

[0101] In the formula, For the first The mass flow rate consumed by each concurrent test, with a boundary upper limit. It's about the total pressure of the pipeline network. With temperature The state equations are used. By quantifying the nonlinear superposition effect of flow rate and temperature, the extended matrix can accurately identify the risks of local shock waves or main pipe depressurization caused by simultaneous large-scale gas intake or venting from multiple compartments, thereby ensuring the absolute safety margin of process planning at the underlying logic level.

[0102] Meanwhile, given the highly time-varying characteristics of aerodynamic systems, the conflict matrix must be capable of real-time updates. This project constructs a dynamic feedback flow control mechanism based on a high-performance message queue. When the underlying control system detects valve lag or sudden changes in unit state, the system uses a sliding time window update equation to reshape the matrix in real time:

[0103] Where C(t) represents the resource allocation conflict matrix at time t; Δt represents the update step size of the sliding time window; C(t+Δt) represents the conflict matrix at the next update time; ⊕ represents the update operation of superimposing, replacing or reconstructing the original conflict matrix based on incremental information; Represents a set of sensor events The increment of the triggered conflict matrix; This represents the set of real-time sensed events collected within the current time window, including valve status, equipment status, pressure, and flow rate. This purely data-driven and logic-based modeling mechanism ensures that the multi-dimensional conflict matrix can accurately evolve its state as the system operates.

[0104] By constructing the detailed conflict matrix of control node valve-pipeline status and the conflict matrix of test equipment status, the process of constructing the conflict matrix of resource allocation for multi-test instrument process system is completed. This process can comprehensively and accurately reflect the conflict between the status of key equipment and between the status of control node valve-pipeline in the test system, providing solid and reliable basic data support for subsequent feasibility analysis of key equipment and selection of test paths based on the conflict matrix.

[0105] In the S4 test support system, the relationship between test tasks and resources is not a simple binary one of task requiring resources and resources satisfying tasks. Instead, it is a complex network relationship comprised of task conditions, equipment capabilities, resource status, process paths, safety boundaries, historical adaptation experience, and alternative support schemes. Traditional resource matching methods typically employ manual rules, static forms, or single-attribute filtering. While these methods are practical in simple task scenarios, they often fail to address the challenges of complex environments with multiple testers, parallel tasks, coupled resources, and dynamically changing equipment statuses. These problems include insufficient expressive power of matching relationships, difficulty in discovering implicit associations, challenges in identifying alternative resource paths, and inadequate conflict propagation analysis. Therefore, this invention introduces a graph embedding method, unifying test tasks, test resources, test equipment, process nodes, valves and pipelines, operating status, and expert rules into a heterogeneous graph structure. Through graph representation learning, deep adaptation relationships between tasks and resources are mined, providing a high-dimensional foundation for subsequent dynamic resource matching, conflict identification, and intelligent scheduling.

[0106] The task-resource heterogeneity graph constructed by this invention can be represented as:

[0107]

[0108] in, Represents a set of nodes. Denotes the set of edges. Represents the node type mapping function, This represents the edge type mapping function. The node set is further divided into:

[0109]

[0110] in, Indicates the test task node, Represents resource nodes, Indicates a device node. Indicates process flow nodes, Indicates valve-pipeline node, This represents the running status node. The edge set includes various types such as task-resource requirement edges, task-equipment occupation edges, resource-equipment support edges, equipment-state constraint edges, task-process dependency edges, process-valve-pipeline control edges, resource-resource conflict edges, and resource-resource substitution edges. Through this unified modeling approach, the system can transform information originally scattered in test plans, equipment ledgers, pipeline diagrams, expert rule bases, and historical test records into a computable network structure.

[0111] In this heterogeneous graph, the attributes of task nodes include not only task number and task type, but also information such as test object, test subject, test stage, operating parameters, priority, time window, safety level, and resource demand intensity. Resource nodes' attributes include resource type, rated capacity, current available capacity, load limit, operating cost, substitutability, and safety boundary. Equipment nodes' attributes include equipment capacity, operating status, maintenance plan, failure probability, remaining life, and derating factor. Valve-pipeline nodes' attributes include open / closed status, flow limit, pressure rating, media type, path connectivity, and control authority. Through this attribute modeling, the graph embedding algorithm can learn topological relationships while integrating node attribute information, improving the accuracy of task-resource matching judgments.

[0112] To learn the low-dimensional vector representations of nodes in a graph, this invention employs the Node2vec graph embedding algorithm. The core idea of ​​Node2vec is to perform biased random walks on the graph and adjust the return parameters... and input / output parameters This approach balances breadth-first search and depth-first search to capture both local and structural similarities of nodes. For the current node... If the previous access node was The next candidate node is The transition probability of a random walk is related to the following weights:

[0113]

[0114] in, This represents the unnormalized transition probability from the current node v to the candidate node x; represents the search bias coefficient determined by the return parameter p and the input / output parameter q; t represents the previous node before node v visited in the random walk; x represents the next candidate node; This represents the weight of the edge between node v and node x; p is used to adjust the tendency to return to the previous node, and q is used to adjust the tendency to explore the local neighborhood or distant nodes.

[0115] in, Representing an edge The weight, Defined as:

[0116]

[0117] in, Represents a node With nodes The shortest path distance between them. When When the number of nodes is small, random walks tend to return to the previous node, making them suitable for uncovering local close relationships; when... A smaller value indicates a greater tendency for random walks to explore outwards, making them suitable for discovering structurally similar but geographically distant nodes. In experimental resource matching problems, if it's necessary to identify direct resource fit relationships, a larger value can be appropriately used. To enhance local neighborhood learning; if it is necessary to discover alternative resources or potential resource combinations, the threshold can be appropriately reduced. Enhance cross-regional exploration capabilities.

[0118] The goal of node embedding learning is to bring nodes that are closely related or semantically similar in a graph closer together in a low-dimensional vector space. For nodes... Its embedding vector is The set of neighboring nodes is The optimization objective is:

[0119]

[0120] Where V represents the set of all nodes in the graph; u represents the current center node; and f(u) represents the low-dimensional embedding vector of node u. Let represent the set of neighboring nodes of node u obtained through random walk sampling; This represents the observed neighborhood set of a given node u, given its embedding representation. The probability of the above condition is denoted by ; max represents maximizing the conditional probability by optimizing the embedding mapping f.

[0121] Under the conditional independence assumption, it can be expanded as follows:

[0122]

[0123] in:

[0124]

[0125] in, represents the conditional probability of generating the set of its neighboring nodes given the embedding vector of the center node u; v represents Neighboring nodes in; represents the conditional probability that node v appears as a neighbor of node u; f(v) and f(u) represent the embedding vectors of nodes v and u, respectively; "·" represents the vector inner product; n represents any node in the graph used for Softmax normalization; V represents the set of all nodes; exp(·) represents the exponential function.

[0126] Since the computational complexity of the full Softmax algorithm is high, this invention uses a negative sampling method for approximate optimization, and its loss function is:

[0127]

[0128] Where L represents the loss function when using negative sampling to train the graph embedding model; E+ represents the set of positive sample node pairs, and E− represents the set of negative sample node pairs; u and v constitute positive sample node pairs, and u and v′ constitute negative sample node pairs; , , Let u, v, and v′ represent the embedding vectors of nodes u, v, and v′, respectively; T represents the vector transpose; σ(·) represents the Sigmoid function; and log represents the natural logarithm.

[0129] This method yields vector representations of each task node, resource node, and device node. Subsequently, for each task... and resources The degree of fit between the two is calculated using cosine similarity:

[0130]

[0131] in, Indicates task With resources Cosine similarity between them; This represents the i-th experimental task. This represents the j-th candidate resource; , represents the embedding vectors of the task node and the resource node, respectively; "·" represents the vector inner product; ||·|| represents the vector L2 norm.

[0132] Resource matching based solely on similarity is insufficient for engineering applications, as the availability of experimental resources depends not only on historical compatibility but also on current operational status, resource capacity, conflict severity, usage costs, and safety margins. Therefore, this invention further constructs a comprehensive compatibility scoring function:

[0133]

[0134] in, Indicates the resource at time Availability This indicates the degree to which resource capacity meets task requirements. Indicates the degree of conflict between resources and other assigned tasks. Indicates the cost of resource usage. This indicates resource security risks. (Each weight...) It can be dynamically adjusted according to the type of task. For example, for safety-sensitive experiments, A larger value should be taken; for urgent insertion tasks, and A larger value should be taken; for regular long-cycle tasks, It can be appropriately increased to reduce the cost of resource use.

[0135] To demonstrate the role of graph embedding models in practical resource matching, this invention designs a four-stage matching process: candidate resource screening, comprehensive scoring and ranking, conflict matrix verification, and alternative resource recommendation. In the first stage, the system screens the candidate resource set based on the vector similarity between task nodes and resource nodes.

[0136]

[0137] in, Indicates task The corresponding set of candidate resources; This represents the j-th resource; Indicates task With resources The similarity between the two is used to determine the matching similarity; θ represents the candidate resource selection threshold, and the resource is included in the candidate set when the similarity is not lower than θ.

[0138] The second stage involves comprehensively scoring and ranking the candidate resources:

[0139]

[0140] in, Represented as a task The selected optimal resource; Indicates task The set of candidate resources; This represents the j-th resource in the candidate set; Indicates task With resources The overall fit score; argmax represents selecting the resource with the highest overall score from the candidate resource set.

[0141] In the third stage, the candidate matching results are input into the resource configuration conflict matrix to determine whether there are capacity conflicts, state conflicts, path conflicts, or security boundary conflicts. In the fourth stage, if the optimal resource is unavailable, alternative resource combinations are recommended according to the alternative edges in the graph and similarity ranking. Through this process, the graph embedding model can not only complete resource adaptation recommendation, but also support rapid resource rematching in dynamic scenarios such as device failure, resource de-capacity, and task insertion.

[0142] Table 1

[0143] Modeling phase Input data Core Algorithm Output Engineering function Heterogeneous graph construction Task, resource, equipment, pipeline, status data Graph modeling Task - Resource Map Unified expression of complex relationships Random walk Graph topology and edge weights Node2vec Node sequence Capturing local and global structures Indicates learning Node sequence Skip-gram + negative sampling Node embedding vector Quantify the similarity between tasks and resources Compatibility rating Embedding vector, resource state, conflict matrix Comprehensive scoring function Candidate resource ranking Supports automatic matching Conflict checking Candidate resources, conflict matrix Rule constraints + matrix verification Executable resource composition Ensure project safety Alternative Recommendations Alternate edges, historical rules Similarity ranking Alternative resource plan Support emergency dispatch

[0144] Referring to Table 1, which shows the input and output results of the core algorithms for each model stage, the graph embedding model has three advantages compared to resource matching methods based on manual rules. First, it can learn implicit relationships from historical data. For example, although a certain type of experimental task may not explicitly specify a backup resource, similar resource combinations may appear multiple times in historical tasks. The graph embedding model can identify the potential suitability of this resource based on structural similarity. Second, it can support resource substitution recommendations. When the primary resource is unavailable, the model can quickly find alternative resources based on substitution relationships and vector similarity in the graph. Third, it can improve the proactiveness of conflict identification. When a resource has strong connections with multiple tasks, the system can predict in advance that it may become a bottleneck resource and perform peak-shifting processing during the scheduling stage.

[0145] S5: A Dynamic Resource Matching Model Based on Multi-Objective Markov Decision Making: The resource matching process in a test support system exhibits significant dynamic, uncertain, and multi-objective characteristics. From a dynamic perspective, the execution status of test tasks, equipment availability, valve and pipeline status, power load, circulating water load, and resource occupancy continuously change over time. From an uncertain perspective, unexpected situations may arise during the test, such as equipment failure, demand changes, operating condition adjustments, task insertion, resource depreciation, and approaching safety boundaries. From a multi-objective perspective, the system needs to simultaneously optimize task completion time, resource utilization efficiency, operating costs, safety risks, and scheduling stability, and these objectives often have interdependent relationships. For example, increasing task parallelism to shorten the test cycle may lead to increased peak power load and valve / pipeline conflict risks; increasing task intervals to improve safety margins may result in decreased resource utilization and extended test cycles. Therefore, this invention models the resource matching problem for complex process flow tests as a multi-objective Markov decision process, achieving dynamic resource matching through state awareness, action selection, reward feedback, and strategy updates.

[0146] A multi-objective Markov decision model can be represented as:

[0147]

[0148] in, Representing the state space, Represents the action space. Represents the state transition probability. Represents a multi-objective reward vector. Indicates the discount factor. This represents the set of target preference weights. The system at each decision time... Observe the current state Select Action The environment shifts to a new state and provide feedback on multi-objective rewards. Through continuous iteration, the model learns resource allocation strategies under different experimental scenarios.

[0149] State space is crucial for model design. This invention defines a state as a combination of experimental task state, resource state, equipment state, pipeline state, and risk state.

[0150]

[0151] in, It indicates the task execution status, including whether the task has started, the completion rate, the remaining time, the priority, and the time window; It indicates the status of resources, including the available capacity of gas supply, extraction, air pretreatment, circulating water, fuel oil and natural gas, and electricity resources; It indicates the status of the test equipment, including available, occupied, faulty, under maintenance, and derated operation; It indicates the valve-pipeline status, including open / closed status, pressure, flow rate, medium, and path connectivity; It indicates the system load status, including power load, circulating water load, and critical equipment load; This represents the current resource conflict matrix; It represents historical decision-making information, which is used to improve strategy stability and avoid execution risks caused by frequent resource switching.

[0152] The action space includes resource allocation actions, task timing adjustment actions, resource switching actions, load regulation actions, and emergency response actions. Specifically, it can be represented as:

[0153]

[0154] in, This indicates the allocation of resources for the task. This indicates that low-priority tasks are delayed. This indicates that tasks that meet the conditions will be executed ahead of schedule. This indicates a switch to a different resource path. This indicates that an alternative resource is being used. Indicates adjusting the load. This indicates that resources or tasks pose a risk when isolated. This indicates a trigger for process replanning. To ensure that the action output meets engineering safety requirements, this invention superimposes rule constraints on the action space:

[0155]

[0156] in, Representing state The set of actions that satisfy the safety constraints; A represents the set of all selectable actions; a represents any candidate action; This represents the system state at time t; Let K represent the k-th safety constraint function; K represents the total number of safety constraints; when all Only when action 'a' is performed will action 'a' be retained in the set of safe actions.

[0157] in, This represents a safety constraint function, such as pressure not exceeding limits, valve states not being mutually exclusive, equipment not being overloaded, electrical load not exceeding limits, and circulating water capacity not exceeding limits. The reinforcement learning agent can only operate within the set of safe actions. Internal selection of actions can help avoid unacceptable and dangerous strategies during the exploration process.

[0158] The design of a multi-objective reward function needs to balance task efficiency, resource efficiency, cost, and security. This invention defines the reward vector as follows:

[0159]

[0160] The task completion reward can be defined as follows:

[0161]

[0162] Resource utilization rewards can be defined as:

[0163]

[0164] in, Representing resources Expected utilization rate. Cost penalty:

[0165]

[0166] The penalty for conflict is:

[0167]

[0168] The safety penalty is:

[0169]

[0170] The strategy stability reward is:

[0171]

[0172] in, Representing state Next action The obtained multi-objective reward vector; This indicates the reward for completing the task. Indicates a reward for resource utilization. This indicates a penalty for the cost of resource usage. Indicates a penalty for resource or path conflicts. Indicates a penalty at the security boundary. This indicates a reward for scheduling stability. This represents the utilization rate or allocation amount of resource j at time t. This represents the value at the previous time step. Used to measure the magnitude of changes in resource allocation between adjacent time points.

[0173] in, Representing resources At any moment The allocation status is crucial; frequent changes in resource allocation introduce additional operational risks, therefore, penalties must be imposed for frequent switching. Ultimately, this is determined through a preference weight vector. Standardize the rewards:

[0174]

[0175] in, Representing state Next action Standardized comprehensive reward; These represent the preference weights for six categories of objectives: task completion, resource utilization, cost, conflict, security, and stability. , , , , , These represent the corresponding sub-rewards or penalties, and the weights of each can be dynamically adjusted based on the task type and operating scenario.

[0176] The weight vector can be dynamically adjusted in different scenarios. For example, during the normal test preparation phase, the weights of resource utilization and cost can be increased; during the steady-state operation phase of the test, the weights of safety and stability should be increased; and during the emergency failure handling phase, the weights of conflict resolution and mission continuity should be increased.

[0177] In terms of algorithm implementation, this invention adopts a hybrid solution approach of rule constraints, reinforcement learning, and rolling optimization. First, an expert rule base and conflict matrix are used to filter out ineffective actions, ensuring policy safety. Second, an Actor-Critic framework is used to learn resource matching strategies, where the Actor network outputs action probabilities and the Critic network evaluates the value of the current strategy. Third, a rolling temporal optimization method is employed to regenerate a locally optimal strategy based on the latest state within each decision window. Its value function can be defined as:

[0178]

[0179] The action value function is:

[0180]

[0181] The strategy optimization objective is:

[0182]

[0183] in, The value function of state s under policy π; γ represents the expected state transition and action selection according to strategy π; γ represents the discount factor; t represents the decision time; T represents the termination time of the current rolling decision window; Indicates an immediate reward; This represents the initial state. Indicates the optimal strategy; This represents the state access distribution induced by strategy π; This indicates that actions are sampled according to strategy π; Let represent the state-action value function under policy π; arg max is used to select the policy that maximizes the expected long-term return.

[0184] To improve the model's adaptability in non-stationary environments, this invention introduces an experience replay mechanism, a priority sampling mechanism, and an online fine-tuning mechanism. When equipment failure or sudden changes in task requirements occur, the system adds this scenario as a high-priority sample to the experience pool, enabling the model to strengthen its response to such abnormal scenarios in subsequent training. For safety-critical scenarios, an expert policy demonstration learning approach is adopted, transforming manual scheduling experience into the model's initial strategy and reducing the risk of reinforcement learning exploration.

[0185] Furthermore, this invention includes a dynamic resource matching method oriented towards changes in test requirements and equipment. Specifically, during the implementation of major test tasks, test requirements and equipment status are not static but constantly changing with test progress, task instructions, equipment health status, environmental conditions, and safety assessment results. Changes in test requirements may manifest as adjustments to test operating parameters, extended test duration, increased measurement items, increased task priority, temporary insertion of new tasks, or cancellation of existing tasks. Changes in equipment status may manifest as equipment failure, equipment maintenance, de-rated equipment operation, abnormal valve operation, abnormal pipeline pressure, leakage in circulating water branches, or decreased power supply capacity. These changes directly affect the original resource matching relationships and scheduling schemes. If they are not identified and dynamically adjusted in a timely manner, they may lead to task delays, resource conflicts, test interruptions, or even safety accidents. Therefore, this invention proposes a dynamic resource matching method oriented towards changes in test requirements and equipment, forming a complete process of "change perception - impact analysis - rule update - conflict reconstruction - resource re-matching - scheduling correction - closed-loop feedback".

[0186] First, the system needs to establish a change perception mechanism. Data sources for change perception include task change instructions issued by the test command system, health status data collected by the equipment condition monitoring system, execution status data uploaded by the valve and pipeline control system, load data uploaded by the power and circulating water monitoring system, and manually confirmed anomaly information. Change events can be defined as:

[0187]

[0188] in, Indicates the type of change. Indicates the object of change. Indicates the time of occurrence. Indicates the amount of change. Indicates the level of impact. This indicates the data source. Based on the affected objects, change events can be categorized into demand change events, equipment change events, resource change events, process change events, and safety change events. Upon receiving a change event, the system first classifies the event and determines its impact level to ascertain whether the change requires triggering resource reallocation.

[0189] Secondly, an impact analysis of the changes needs to be conducted. For changes in requirements, the system recalculates the task resource requirement vector:

[0190]

[0191] And perform a difference analysis with the original demand vector:

[0192]

[0193] In response to equipment changes, the system updates the equipment availability coefficient:

[0194]

[0195] in, This indicates that the equipment is fully usable. This indicates that the equipment is operating at reduced capacity. This indicates that the device is unavailable. The available resource capacity has been adjusted to:

[0196]

[0197] If a device changes from an available state to a faulty state, then the feasibility of all tasks that depend on that device needs to be reassessed.

[0198]

[0199] in, This represents the updated requirement vector after a change in requirements for task i. This indicates the updated information about the test subjects or task objects. This indicates the updated operating conditions / performance parameters. This indicates the updated constraints. This indicates the updated resource requirements. This indicates the updated time constraint. This indicates the updated security constraints. Indicates task The results of the resource feasibility assessment; Indicates task Required resource set; Represents the changed resources Effective available capacity at time t; Indicates task Resources Demand; when the capacity of all resources is not less than the demand. Otherwise, it is 0.

[0200] When a task becomes unavailable, the system needs to find alternative resources or adjust the task timing.

[0201] Secondly, the system needs to re-evaluate the association rules and conflict matrix. Changes in requirements may render the original rules inapplicable. For example, if the original plan was for a medium-load test, but it is changed to a high-load, long-term steady-state test, the demand for circulating water, electricity, and air extraction resources will increase significantly. In this case, the system will re-match the association rules according to the new task attributes:

[0202]

[0203] in, Representation and update requirements A set of matching association rules; This represents the k-th rule in the rule base; Representation rules Triggering conditions; Demand vector The matching function for the rule conditions takes the value 1 if the match is true, and 0 otherwise.

[0204] Equipment changes will alter the conflict matrix. For example, if a leak occurs in a circulating water branch, the corresponding resource node should be marked as unavailable, and the conflict values ​​of all tasks related to that branch will increase. The updated overall conflict matrix is ​​as follows:

[0205]

[0206] in, This represents the updated comprehensive resource conflict matrix at time t; , , , , , These represent the updated resource capacity, valve-pipeline, equipment status, power load, circulating water load, and safety boundary conflict matrix, respectively. ~ The fusion weights are used for various conflict matrices.

[0207] After updating the conflict matrix, the system performs a feasibility check on the current scheduling plan. If a hard conflict exists, replanning must be triggered immediately; if a soft conflict exists, a partial adjustment is selected based on priority and risk level.

[0208] During resource reallocation, the system employs a hierarchical decision-making mechanism. The first layer is the safety constraint layer, where any scheme that violates the safety boundaries is directly eliminated. The second layer is the feasibility layer, which determines whether resource capacity, equipment status, and valve / pipeline routes meet the task requirements. The third layer is the optimization layer, which selects the scheme with the least task delay, highest resource utilization, lowest cost, and lowest conflict risk from multiple feasible options. Its optimization objective can be expressed as:

[0209]

[0210] Where J represents the overall optimization goal of the resource rematching scheme; Delay represents the task delay; Conflict represents the risk of conflict after adjustment; Cost represents the resource usage cost; Switch represents the cost of switching resources or paths; and Risk represents the security risk. ~ For the weights of the corresponding objective items, min indicates the selection of the rematching scheme that minimizes the overall cost.

[0211] in, This indicates a task delay caused by re-matching. This indicates the adjusted risk of conflict. Indicates the cost of resource usage. Indicates the cost of resource switching. This indicates a security risk. Through this objective function, the system can avoid plan instability caused by frequent resource switching, while ensuring security and task continuity under sudden changes.

[0212] Taking a circulating water pipeline leak as an example, after the system detects an abnormal drop in pressure in a certain cooling branch, it marks the branch as unavailable, updates the corresponding resource capacity to zero, and adds all tasks that depend on this branch to the affected set. Subsequently, the system searches for backup circulating water branches and alternative cooling paths, calculating their capacity, distance, switching time, and conflict severity. If the backup branch capacity meets the task requirements, the system generates a resource switching plan; if the backup branch capacity is insufficient, the system adjusts the task timing, staggering the execution of high-heat-load tasks and reducing the number of parallel tasks. Similarly, when the power load approaches its limit, the system can postpone low-priority tasks, reduce auxiliary equipment power, or switch some equipment to backup power supply paths.

[0213] The core advantage of this method lies in its transformation of resource matching from static planning to dynamic closed-loop control. The system can not only generate resource matching schemes during the planning phase, but also continuously adjust these schemes based on real-time conditions during test execution, ensuring that resource allocation always adapts to current task requirements and equipment status, thus guaranteeing the stable operation of complex test tasks.

[0214] As the scale of test tasks expands and the degree of test resource sharing increases, the parallel execution of multiple test tasks has become the norm in test support systems. Sequential tasks emphasize the order of processes and the connection between stages; for example, pipeline purging must precede gas supply pressurization, and equipment precooling must precede high-heat-load testing. Parallel tasks, on the other hand, emphasize resource sharing and conflict avoidance; for example, multiple test devices can operate simultaneously, but cannot simultaneously exceed the power capacity, circulating water capacity, or the flow limit of shared pipelines. The multi-task parallel scheduling problem requires simultaneous consideration of task logical relationships, time windows, resource capacity, equipment status, conflict matrices, safety isolation, and task priorities; essentially, it is a combinatorial optimization problem with complex constraints.

[0215] Let the set of experimental tasks be:

[0216]

[0217] The resource set is as follows:

[0218]

[0219] Where T represents the set of all experimental tasks, Let R represent the nth experimental task, where n represents the total number of tasks; R represents the total set of available resources. Let m represent the m-th resource, where m represents the total number of resources.

[0220] Each task With duration Earliest start time Latest completion time Priority Resource requirements and security level Define the task start time as... The end time is ,but:

[0221]

[0222] And satisfy the time window constraint:

[0223]

[0224] If task It is a task The prerequisite task is:

[0225]

[0226] in, Indicates the prerequisite task End time; Indicates a follow-up task The start time; Indicates task With the task A safety interval or process transition time must be maintained between them; the inequality indicates that the subsequent task can only be started after the preceding task is completed and the safety interval is met.

[0227] For example, when switching from a high-pressure gas supply task to a pipeline maintenance task, it is necessary to set time for depressurization, purging, and confirmation. This time can be used as... Incorporate into the scheduling model.

[0228] To describe whether a task is executed at a certain time, a binary variable is defined:

[0229]

[0230] Define resource allocation variables:

[0231]

[0232] Resource capacity constraints are:

[0233]

[0234] Regarding electricity resources:

[0235]

[0236] Regarding circulating water resources:

[0237]

[0238] in, This is a binary variable representing the task execution status. The value is 1 if the operation is executed at time t, otherwise it is 0. Indicates task The circulating water demand at time t; This indicates the maximum allowable water supply capacity of the circulating water system at time t; This represents the total circulating water demand at time t for all tasks currently in progress.

[0239] For valve-pipeline resources, both state compatibility constraints and capacity constraints must be satisfied simultaneously. If the task... and For a certain valve If the state requirements are mutually exclusive, then the two cannot run concurrently:

[0240]

[0241] If sharing a pipeline The total flow rate must not exceed the pipeline capacity.

[0242]

[0243] If the two media are incompatible, task isolation or setting cleaning and purging times is necessary.

[0244]

[0245] The comprehensive optimization objective function is designed as follows:

[0246]

[0247] in, , Representing tasks , Whether to execute at time t; when there is a hard conflict between two tasks, such as mutual exclusion of valve states, through Both must not be executed simultaneously. In the combined objective function, Z represents the overall optimization objective; Indicates the total completion time; This represents the actual load of resource j at time t. This represents the available capacity of resource j at time t. Indicates the expected utilization rate; This indicates the risk of conflict between task i and task k at time t; This represents the cost of resource j used by task i; Represents task-resource allocation variables; This represents the delay penalty for task i; ~ This indicates the weight of each objective item.

[0248] in, The first term represents the total completion time; the second term represents the degree to which resource load deviates from expected utilization; the third term represents the conflict risk; the fourth term represents the resource usage cost; and the fifth term represents the task delay penalty. For high-priority tasks, the delay penalty can be set as follows:

[0249]

[0250] in, Indicates task Delay penalties; Indicates task Overall priority; Indicates task The actual completion time; Indicates task The planned expiration date or the latest expected completion date; This indicates the duration of the overdue period; if the period has not expired, the value is 0.

[0251] In engineering implementation, relying solely on exact mixed-integer programming to solve large-scale problems can be computationally time-consuming. Therefore, this invention employs a solution strategy of "hierarchical partitioning + rolling time domain + heuristic optimization". First, the problem is hierarchically divided according to the process stage and resource type of the experimental task, breaking down the global problem into three sub-problems: task scheduling, resource allocation, and control path planning. Second, a rolling time domain method is used to divide the long-cycle scheduling into multiple decision windows, performing local optimization within each window. Finally, genetic algorithms, simulated annealing, or tabu search are used to perturb and optimize the initial solution, gradually reducing resource conflicts and load peaks.

[0252] In multi-task parallel scenarios, task priority is a crucial input to the scheduling model. Priority is determined not only by task importance but also by a comprehensive consideration of task urgency, impact on subsequent tasks, resource consumption time, task interruption costs, and safety level. A comprehensive priority function can be constructed as follows:

[0253]

[0254] in, Indicates task Overall priority; Indicates the importance of the task. Indicates the urgency of the task. This indicates the dependency and impact on subsequent tasks. Indicates the mission's safety risk level. Indicates the cost of task interruption; ~ The weights of each evaluation factor.

[0255] in, Indicates the importance of the task. Indicates the level of urgency. This indicates the dependency and impact on subsequent tasks. Indicates the level of safety risk. This represents the cost of task interruption. The higher the priority, the more the scheduling model tends to guarantee its resource supply and time window.

[0256] Table 2

[0257] Constraint Categories Constraints Mathematical expression or judgment method Processing strategy Time constraints Earliest start, latest finish Adjust start time Preconditions Process sequence Topological sorting Capacity constraints Resource load not exceeded Peak shaving and staggered peak State constraints Valve equipment status compatibility Parallelism is prohibited Path constraints Pipeline path available Alternative Path Safety constraints Pressure and temperature do not exceed limits Forced protection Priority constraints High-priority task assurance Prioritize resource allocation

[0258] The main symbols used in the mathematical expressions in Table 2 have the following meanings: , Indicates the i-th and k-th trial tasks; , Representing tasks The start and end times; , These represent the earliest allowed start time and the latest allowed finish time, respectively. Indicates the safety interval between tasks or the process changeover time; Indicates task Whether to execute at time t; Indicates task Resources Demand; Representing resources Available capacity at time t; This represents the j-th resource; Indicates task The corresponding process path; P and T represent pressure and temperature, respectively; Indicate the corresponding security boundary; Indicates task Overall priority.

[0259] Referring to Table 2, which lists the constraints and corresponding handling strategies, this model enables the system to perform global coordination when multiple tasks compete for resources in parallel, avoiding overall conflict caused by local optima. For example, if a low-priority task prematurely occupies critical cooling resources, it may prevent subsequent high-priority tasks from starting on time. The global optimization scheduling model can identify this risk in advance and postpone low-priority tasks, thereby ensuring the overall experimental plan is optimal.

[0260] A Global Optimization Scheduling Model for Multiple Simultaneous Test Tasks in S7: Power load and circulating water load are the two most critical, continuous, and bottleneck-prone basic resources in the test support system. The power system supplies power to test equipment, extraction equipment, compressors, pump sets, control systems, measurement systems, and auxiliary systems; the circulating water system handles tasks such as cooling test equipment, air pretreatment cooling, extraction system cooling, thermal management of fuel oil and natural gas systems, and safety protection. These two systems often exhibit strong coupling during operation; the startup of high-power equipment leads to increased power load and consequently higher heat load, resulting in increased circulating water demand. Therefore, optimizing only power or circulating water may lead to implicit overload of the other resource. To address this issue, this invention proposes a collaborative matching method for power load and circulating water load.

[0261] Set a task At any moment The electricity demand is The circulating water demand is The total power load and circulating water load of the system are as follows:

[0262]

[0263] The fundamental constraints for electricity and circulating water are:

[0264]

[0265] Where P(t) represents the total electrical load of the system at time t, and W(t) represents the total circulating water load of the system at time t; Indicates task Whether to execute at time t; Representing tasks Electricity demand and circulating water demand at time t; These represent the maximum allowable loads of the power system and the circulating water system at time t, respectively.

[0266] However, in actual operation, even if neither exceeds the upper limit, if both are close to the upper limit simultaneously, it will lead to insufficient system redundancy and reduced disturbance rejection capability. Therefore, this invention defines a power-circulating water coupling risk index:

[0267]

[0268] in, This represents the power-circulating water coupling risk index at time t; Indicates the electricity load factor. Indicates the circulating water load rate; These represent sensitivity indices for power risk and circulating water risk, respectively, and are used to adjust the degree of risk amplification when the two types of loads approach their upper limits.

[0269] in, Used to regulate the risk sensitivity of electricity and circulating water. When Exceeding the threshold When the system determines that there is a risk of high load coupling during that period, it is necessary to trigger a coordinated scheduling strategy.

[0270] In collaborative matching, the system considers not only the current load but also the rate of load change. If the load increases too rapidly, it may cause equipment startup shock or delayed cooling response. Therefore, the rate of load change is defined as follows:

[0271]

[0272] And set a rate of change constraint:

[0273]

[0274] Where ΔP(t)=P(t)−P(t−1) represents the change in power load between two adjacent moments, and ΔW(t)=W(t)−W(t−1) represents the change in circulating water load between two adjacent moments; These represent the maximum permissible change in electrical load and the maximum permissible change in circulating water load, respectively; |·| indicates taking the absolute value.

[0275] This constraint can prevent multiple high-power devices from starting up simultaneously or multiple high-heat-load tasks from entering the steady-state phase at the same time.

[0276] To address peak power loads, the system can adopt the following strategies: First, stagger the start-up of high-power equipment to avoid simultaneous startup of extraction equipment, compressors, and main test equipment; second, postpone the execution of low-priority auxiliary tasks to free up power capacity for critical tasks; third, activate backup power supply paths or zoned power supply strategies to improve power redundancy; and fourth, set soft-start curves based on equipment operating characteristics to reduce startup impact. To address peak circulating water loads, the system can adopt the following strategies: First, switch or parallel backup cooling branches; second, start the circulating water system in advance for pre-cooling; third, reduce the heat load on non-critical equipment; fourth, adjust the steady-state operating time of high-heat-load tasks to avoid the overlap of multiple high-heat-load tasks; and fifth, optimize the water flow path where pipeline conditions permit to reduce local pipeline congestion.

[0277] Table 3

[0278] Scene Power load status Circulating water load status Risk assessment Cooperative scheduling strategy Normal operation Below 60% Below 60% Low risk Execute as planned High power Above 85% Below 70% Medium risk Delay high-power auxiliary tasks High water load Below 70% Above 85% Medium risk Activate backup cooling branch High load Above 85% Above 85% High risk High-load tasks staggered rapid rise Rate of change exceeds limit Rate of change exceeds limit Shock risk Phased start-up of equipment Single resource failure Power supply or water branch abnormality Load redistribution Emergency Risk Enable backup resources and rearrange

[0279] Referring to Table 3, which presents risk assessment and coordination strategies for electricity and water loads under different scenarios, the engineering value of the electricity and circulating water coordinated matching method lies in its ability to incorporate two originally independently managed resource systems into a unified scheduling model. This avoids local optimization problems such as available electricity but insufficient cooling, or sufficient cooling but excessive power supply. Through load forecasting, coupled risk assessment, and dynamic scheduling, the system can identify load peaks in advance during the test plan generation phase and perform peak shaving and valley filling in real time during execution, thereby improving the overall stability of the system.

[0280] S8 Valve-Pipeline State Conflict Identification and Path Planning Method: The valve-pipeline system is a critical channel connecting test resources and test equipment in the test support system. Its state directly determines whether the test process is reachable, whether resources can be delivered as required, and whether tasks can be executed in parallel. Unlike continuous capacity resources such as electricity and circulating water, valve-pipeline resources have obvious topological constraints, state constraints, medium constraints, pressure constraints, and directional constraints. A pipeline may be required to transport different media in different tasks, a valve may be required to be in different open / closed states at the same time, and a junction node may exceed its limits due to the superposition of flow rates from multiple tasks. Therefore, valve-pipeline conflict identification is one of the most complex and critical aspects of process resource matching in the test support system.

[0281] This invention models the valve-pipeline system as a directed graph with attributes:

[0282]

[0283] in, This represents valves, pipe nodes, branch nodes, junction nodes, interface nodes, and equipment inlet nodes; Indicates the edge of the pipe; It represents node attributes, including valve number, open / closed status, control authority, response time, reliability, and security level; This indicates pipe side attributes, including length, pipe diameter, medium type, pressure rating, upper limit of flow rate, temperature range, flow direction restriction, and cleaning requirements.

[0284] Task The process path is represented as follows:

[0285]

[0286] The path must satisfy connectivity constraints:

[0287]

[0288] Media compatibility constraints:

[0289]

[0290] Pressure constraints:

[0291]

[0292] Flow constraints:

[0293]

[0294] Valve status constraints:

[0295]

[0296] in, Indicates task The process path; ,… indicates valves, branch / merge, interfaces, or device inlet nodes on the path; ,… represent the pipe edges between adjacent nodes; k represents the length of the path node sequence. State(v,t) represents the actual state of node v at time t; Indicates task The target state required for node v; This means that all nodes in the path should meet the corresponding state requirements.

[0297] When multiple tasks share the same valve or pipeline, it is necessary to determine whether their states are compatible. Define a state conflict function:

[0298]

[0299] The pipeline conflict function is defined as follows:

[0300]

[0301] Among them, media conflict, pressure conflict, flow conflict, direction conflict, and cleaning conflict describe the risks of shared pipelines from different dimensions. The comprehensive path conflict value is:

[0302]

[0303] like If the threshold is exceeded, the two tasks cannot be executed in parallel, or the path of one of the tasks must be replanned.

[0304] The goal of path planning is to select the path that is low in cost, low in risk, and has few conflicts while meeting process requirements. A path cost function can be defined as follows:

[0305]

[0306] The optimal path is:

[0307]

[0308] in, Indicates task With the task Conflicting state values ​​at shared valve v; These represent the states required by the two tasks for valve v; 0 indicates that the states are consistent and there is no conflict, 1 indicates that the states are mutually exclusive and form a hard conflict, and α represents the soft conflict coefficient when they are partially compatible but there is a risk. Indicates task The optimal path; Indicates task The set of feasible candidate paths; This represents the overall cost of the path. For real-time path planning, improved Dijkstra's algorithm, A* search, or multi-constraint shortest path algorithm can be used. For simultaneous path planning of multiple tasks, multi-device flow models or conflict-aware graph search methods can be used. arg min represents the path with the minimum overall cost selected from the candidate paths.

[0309] Table 4

[0310] Conflict Types Typical manifestations Judgment basis Processing strategy Valve status conflict The same valve needs to be opened / closed simultaneously. Mutual exclusion of states Parallel processing or process adjustments are prohibited. Pipeline media conflict Different media sharing non-mixed pipelines Media compatibility table Change path or purge Pressure level conflict High-pressure task and low-pressure pipeline mismatch Pressure limit Replace high pressure pipeline Traffic capacity conflict Multi-task traffic exceeding limits Pipeline capacity Diversion or staggered peak hours Flow to conflict Incompatible bidirectional requirements Pipe direction attribute Replan the route Cleaning and Isolation Conflict Task switching requires purging Cleaning Time Rules Increase process interval

[0311] Referring to Table 4, which outlines the situation determination and handling strategies for different conflict types, this method significantly improves the executability of automated process planning. Traditional process planning may only consider the existence of resources, ignoring whether those resources can be delivered to the target equipment via reachable paths. Through valve-pipeline diagram modeling and path conflict identification, the system can determine the feasibility of process paths before scheduling, avoiding the problem of discovering valve status conflicts or unreachable pipeline paths only after the plan has been issued.

[0312] The S9 Intelligent Planning and Scheduling Solution Generation Method: The intelligent planning and scheduling solution generation for the process flow of the test support system is a key step in transforming technical achievements such as demand analysis, association rule mining, conflict matrix identification, dynamic resource matching, and global optimization scheduling into an executable plan. This solution not only needs to clarify which tasks will be executed at what time, but also which resources will support them, which process path will be used, the valve status, how power and circulating water loads will change, how potential conflicts will be handled, and how to switch to a backup plan in abnormal situations. Therefore, this invention defines the intelligent planning and scheduling solution as a comprehensive planning object comprising a task layer, resource layer, control layer, risk layer, and feedback layer.

[0313] The intelligent planning scheme can be represented as:

[0314]

[0315] in, Indicates the sequence of task execution. Indicates the task schedule. Indicates the resource allocation plan, Indicates the valve-pipeline path. This represents the sequence of device and control node states. Indicates conflict risk assessment, This indicates an adjustment to the emergency response plan. This indicates the implementation of monitoring indicators.

[0316] The intelligent planning scheme generation process is a core component of the test support system, moving from resource data management to intelligent decision-making and autonomous scheduling optimization. Essentially, it unifies and integrates test requirements, process flows, resource status, equipment capabilities, valve-pipeline topology, historical experience rules, real-time monitoring data, and safety constraints to form an executable, interpretable, verifiable, and dynamically adjustable test support process flow planning scheme. This process is not simply about generating a timetable based on a task list; rather, it requires establishing a closed-loop reasoning chain from requirement analysis to scheme generation, focusing on key questions such as "whether the task can be executed, whether resources are sufficient, whether the path is reachable, whether the status is compatible, whether the load exceeds limits, whether the risk is controllable, and whether the scheme is optimal." To adapt to the complex scenarios of multiple objects, multiple tasks, multiple resources, multiple constraints, and dynamic changes in major national test missions, this invention divides the intelligent planning scheme generation process into eight consecutive steps: test requirement analysis, candidate process flow node generation, implicit resource requirement completion, candidate resource set construction and suitability calculation, resource conflict matrix verification, initial schedule optimization generation, simulation and evaluation, and final visualized schedule output. Each step has a clear input-output relationship and achieves iterative optimization through a feedback mechanism, thereby ensuring that the generated test support plan not only conforms to the process logic but also meets resource constraints and safety requirements.

[0317] The intelligent scheduling output should have multi-view capabilities. The task view displays the start time, end time, execution stage, and dependencies of each test task; the resource view displays the occupancy and load curves of various resources at different times; the conflict view displays the conflict matrix between tasks, resources, and paths; the control view displays valve opening and closing sequences and pipeline routes; and the risk view displays the risk levels of electricity, circulating water, equipment, and safety boundaries. Through multi-view linkage, test personnel can intuitively understand the system-generated plan and manually confirm and adjust it when necessary.

[0318] The first step is test requirement analysis, which involves converting natural language, task forms, test outlines, scheduling instructions, or manually entered information into a structured requirement vector that the system can recognize and calculate. Test tasks in their raw state are typically characterized by strong descriptiveness, implicit constraints, scattered parameters, and inconsistent semantics. For example, a description like "Conduct a long-term steady-state test on a certain type of test object under high temperature and high pressure, requiring the completion of gas supply, cooling, extraction, and data acquisition support, with the test window from 9:00 to 15:00 on a certain day" contains multiple types of information, including the test object, test subject, operating condition level, resource type, and time window. The system first identifies the core elements of the task through methods such as natural language processing, form field parsing, and rule template matching, and represents them as a unified requirement vector:

[0319]

[0320] in, Indicates the test subject, Indicates the subject of the test. Indicates operating conditions. Indicates time constraints. This indicates explicit resource requirements. This indicates the quality and performance requirements. Indicates safety boundary constraints. This indicates task priority or assurance level. For task inputs from different sources, the system needs to perform field normalization and semantic standardization. For example, expressions like "high-temperature test" may correspond to the same type of operating condition label; "circulating cooling" and "cooling water input" may both map to circulating water resource requirements. By establishing a terminology dictionary, operating condition classification table, and resource coding table, unstructured descriptions can be uniformly transformed into computable objects.

[0321] During the requirements analysis process, it is also necessary to identify the explicit and implicit constraints of the task. Explicit constraints usually appear directly in the task description, such as the test start time, end time, test pressure, target temperature, and duration. Implicit constraints need to be derived by combining expert rules and historical experience. For example, high-temperature conditions usually imply high load requirements for circulating water, long-term steady-state conditions usually imply continuous power supply and high reliability requirements for equipment health, and high-pressure gas supply tasks usually imply valve-pipeline pressure rating matching requirements. To improve the completeness of the analysis results, this invention adopts a combination of rule templates, semantic recognition, and parameter verification to form a credibility score for the requirements analysis results.

[0322]

[0323] in, Indicate demand The overall analytical confidence level; This indicates the confidence level of natural language or form text parsing. This represents the confidence level of the expert rule matching. This indicates the confidence level for verifying the completeness and validity of key parameters; The fusion weights are the three confidence levels.

[0324] in, Indicates the confidence level of text or form recognition. This indicates the confidence level of the rule matching. This indicates the confidence level for verifying the completeness and reasonableness of the parameters. When the confidence level of the requirement analysis falls below a set threshold, the system should prompt the test personnel to supplement key information to avoid subsequent planning errors due to incomplete input.

[0325] The second step is to generate candidate process flow nodes based on the demand-planning mapping model. Test requirements cannot be directly converted into executable actions; instead, they must be mapped to a series of process flow nodes, such as equipment preparation, pipeline inspection, valve presetting, pipeline purging, air supply pressurization, air pretreatment, circulating water input, vacuum start-up, steady-state operation, parameter acquisition, load reduction shutdown, and system recovery. The role of the demand-planning mapping model is to generate a set of candidate nodes that satisfy the process logic based on the test requirement type, operating conditions, and equipment resource status. Let the test plan node be represented as:

[0326]

[0327] in, Indicates process actions, Indicates the required equipment. Indicates the valve status. Indicates the pipeline path, Indicates the execution time. Indicates a prerequisite condition. This indicates a subsequent state. The mapping from demand to process nodes can be represented as:

[0328]

[0329] Where F represents the mapping relationship from structured requirements to a set of candidate process flows; This represents the requirement for the i-th structured test; Indicates and The corresponding set of candidate process flows; Let represent the first to the kth candidate process flows corresponding to the i-th requirement, where k represents the number of candidate processes.

[0330] This mapping is not a single path, but may generate multiple candidate process flows. For example, the same gas supply task can use the main gas supply path or the backup gas supply path; the same cooling task can use the main circulating water branch or the backup cooling branch; the same extraction task can select different combinations of extraction units based on the equipment status. Therefore, when generating candidate process nodes, the system needs to retain multiple feasible solutions to provide space for subsequent resource matching and conflict resolution.

[0331] There are usually strict dependencies between candidate process nodes, which can be modeled as a directed acyclic graph:

[0332]

[0333] in, For the set of process nodes, For nodes, there are dependent edges. If a node... Must be on the node If it is completed before execution, then there is an edge. For example, "pipeline purging" must precede "gas supply pressurization," "circulating water commissioning" must precede "high heat load steady-state operation," and "pressure relief confirmation" must precede "pipeline restoration." Through directed graph modeling, the system can identify serial nodes, parallel nodes, mutually exclusive nodes, and synchronous nodes, providing a structural basis for subsequent scheduling.

[0334] The third step is to call the association rule base to complete the implicit resource requirements. Experimental task descriptions typically only provide the main resource requirements, but complex processes often involve numerous auxiliary resources and support conditions. For example, the task description might only state "conduct a high-temperature gas supply test," but actual execution would require air pretreatment, circulating water cooling, power supply, valve control, measurement and control systems, pipeline purging media, backup gas supply paths, and safety monitoring systems, etc. To avoid generating incomplete solutions based solely on explicit requirements, this invention introduces an association rule base to automatically complete the implicit resource requirements. Association rules can be represented as:

[0335]

[0336] in, Indicates task conditions or operating characteristics. This represents the resource requirements or process actions inferred from this condition. Rule quality is evaluated using support, confidence, and lift:

[0337]

[0338] Where X represents the antecedent of the association rule, i.e., the set of task conditions or working condition features; Y represents the consequent of the association rule, i.e., the set of resource requirements or process actions inferred from X; count(X∪Y) represents the number of historical samples that simultaneously contain X and Y; count(X) represents the number of samples that contain X; N represents the total number of historical samples; support represents the support, confidence represents the confidence level, and lift represents the lift level; support(Y) represents the proportion of the consequent Y in all samples.

[0339] By completing the association rules, the system can expand the demand vector into a complete set of resource demands:

[0340]

[0341] in, This indicates explicit resource requirements. This step represents the implicit resource requirements derived from rule-based reasoning. It is crucial for improving the completeness of planning solutions and can effectively reduce common problems in manual scheduling, such as omissions of auxiliary resources, state conditions, and safety constraints.

[0342] The fourth step is to construct a candidate resource set and calculate the task-resource fit using a graph embedding model. After completing the resource requirement completion, the system needs to filter candidate resources from the resource pool that can meet the task requirements. The resource pool includes experimental equipment, gas supply systems, extraction systems, air pretreatment systems, circulating water systems, fuel oil and natural gas systems, power systems, valve and pipeline systems, measurement and control systems, and safety monitoring systems, etc. Candidate resource screening first involves an initial selection based on resource type, capability parameters, and current status.

[0343]

[0344] in, Indicates task The set of candidate resources; This represents the j-th resource; Representing resources Type; Indicates task The required resource type; Representing resources Available capacity or capacity; Indicates task The corresponding demand for resources; Representing resources The current running state; Available represents the set of available states that are allowed to participate in the matching.

[0345] However, filtering by resource type and capacity alone is insufficient to determine whether resources are truly suitable for the task, as different resources may have differences in historical suitability, path reachability, operating costs, and substitution relationships. Therefore, this invention employs a graph embedding model to further calculate the deep suitability between tasks and resources. The system constructs a heterogeneous graph from tasks, resources, equipment, process nodes, valves and pipelines, and operating states:

[0346]

[0347] Wherein, G represents a heterogeneous graph consisting of tasks, resources, equipment, and process nodes; V represents the set of nodes in the graph; and E represents the set of edges in the graph, used to describe relationships such as task requirements, resource support, equipment occupancy, process dependence, state constraints, conflicts, and substitutions.

[0348] Obtain task node vectors through Node2vec, GraphSAGE, or other graph representation learning methods. and resource node vector And use cosine similarity to calculate the fit:

[0349]

[0350] To adapt to actual engineering conditions, resource availability, capacity matching, conflict risk, usage cost, and safety margin should also be included in the comprehensive score.

[0351]

[0352] in, Indicates task With resources Cosine similarity; Let |||||| represent the embedding vectors of the task and the resource, respectively; ||·|| represents the vector L2 norm. This indicates the overall compatibility score between the two. Representing resources Availability at time t; Indicates capacity matching degree; Indicates the current level of conflict; Indicates the cost of resource usage; Indicates security risk; ~ The weights of each indicator are given.

[0353] in, Indicates the current availability of resources. This indicates the degree of matching between resource capabilities and task requirements. Indicates the current level of resource conflict. Indicates the cost of resource usage. This indicates resource security risks. The system ranks candidate resources based on a comprehensive score and generates a primary resource plan and alternative resource plans.

[0354] The fifth step is to input the candidate resources into the conflict matrix for verification, eliminating solutions with hard conflicts. High resource suitability does not necessarily mean the resource is available, as it may already be occupied by other tasks, or there may be issues such as valve status mutual exclusion, pipeline path conflicts, excessive power load, insufficient circulating water capacity, or incompatible equipment states with other tasks. Therefore, the system needs to input the candidate resource solutions into a comprehensive conflict matrix for verification. The comprehensive conflict matrix can be represented as:

[0355]

[0356] Where M represents the comprehensive resource conflict matrix. , , , , , These represent the resource capacity conflict matrix, equipment status conflict matrix, valve-pipeline conflict matrix, power load conflict matrix, circulating water load conflict matrix, and safety boundary conflict matrix, respectively. ~ This represents the fusion weights of the corresponding matrix.

[0357] For candidate solutions Its conflict risk can be calculated as follows:

[0358]

[0359] in, Indicates candidate planning schemes The overall conflict risk; i and l represent the task or resource indexes involved in the risk accumulation calculation; This represents the conflict intensity in the i-th row and l-th column of the comprehensive conflict matrix; This is an indicator variable that indicates whether task i and task l overlap in time or share resources. It takes the value 1 if the association condition is met, and 0 otherwise.

[0360] in, Indicates task With the task There are overlaps in time or shared resources among candidate solutions. If a candidate solution has a hard conflict, such as the same valve being required to open and close simultaneously, the same pipeline transporting incompatible media, the power load exceeding the limit, or the safety pressure boundary being breached, then the solution should be directly eliminated. For soft conflicts, such as resource load approaching the limit, high utilization rate of circulating water branches, or increased health risks due to prolonged continuous operation of a certain equipment, the system can retain the solution, but the risk level needs to be marked, and it needs to be handled during the scheduling optimization phase through peak shifting, alternative resources, or load balancing.

[0361] The sixth step is to generate an initial schedule using mixed-integer programming and heuristic algorithms. After resource adaptation and conflict verification, the system obtains a batch of feasible or basically feasible task-resource combinations. The next step is to determine the start time, end time, resource occupancy time, and execution order of each process node. This problem is essentially a combinatorial optimization problem under multiple tasks, multiple resources, and multiple constraints. This invention uses a mixed-integer programming model to describe its core constraints. Let the tasks... The start time is The end time is The duration is ,but:

[0362]

[0363] If task It is a task The prerequisite task is:

[0364]

[0365] in, This indicates the process changeover or safety interval time. Resource capacity constraints are:

[0366]

[0367] The constraints for electricity and circulating water are as follows:

[0368]

[0369] The optimization objective can be designed as follows:

[0370]

[0371] in, Indicates task End time, Indicates subsequent tasks The start time, This represents the safety interval or process changeover time between tasks. In the comprehensive objective function, Z represents the global scheduling objective. Indicates the total completion time; This represents the load of resource j at time t. This indicates the resource capacity at that moment. Indicates the expected utilization rate; This indicates the risk of conflict between task i and task k; This indicates the execution status of the task at time t; This indicates a task delay; This represents the cost of resource j used by task i; Represents resource allocation variables; The weights for each objective item.

[0372] in, The first term represents the total completion time, the second term represents the resource load balancing level, the third term represents the conflict risk, the fourth term represents the task delay, and the fifth term represents the resource usage cost. Due to the large scale of actual experimental tasks, simply using exact mixed-integer programming may result in long solution times. Therefore, the system adopts a combined approach of "mixed-integer programming + heuristic algorithms." For smaller-scale or critical task scheduling, mixed-integer programming is prioritized to find the global optimum; for large-scale tasks or real-time adjustment scenarios, heuristic methods such as genetic algorithms, simulated annealing, tabu search, greedy insertion, or rolling time-domain optimization are used to quickly generate approximate optimal solutions.

[0373] The seventh step is to evaluate resource load, conflict risk, and task completion time based on simulation. The initial schedule is not the same as the final plan because there may be discrepancies between the constraints in the mathematical model and the actual experimental process. For example, there may be fluctuations in equipment start-up time, valve response delays, changes in pipeline purging time, instantaneous power load surges, and lag in circulating water cooling response. Therefore, before officially outputting the plan, a simulation of the initial schedule is required. The simulation uses the time axis as the main thread, simulating the execution process of each task node according to the planned sequence, and calculating the occupancy, load level, conflict risk, and safety margin of various resources in each time slice.

[0374] Simulation evaluation metrics mainly include task completion time, resource utilization, peak load, conflict risk, resource switching frequency, and scheme stability. Resource utilization can be expressed as:

[0375]

[0376] Peak load rate can be expressed as:

[0377]

[0378] The cumulative value of conflict risk can be expressed as:

[0379]

[0380] Task delay rate can be expressed as:

[0381]

[0382] in, This represents the average utilization rate of resource j over the entire statistical time domain; This represents the actual load of resource j at time t; This represents the available capacity of resource j at time t; This represents the summation over all discrete time slices. DR represents the task delay rate; Indicates task The actual completion time; Indicates the planned expiration date; Indicates task Priority or delay weight; This indicates the duration the task has exceeded its deadline.

[0383] Based on the above indicators, the system can determine whether there are resource bottlenecks and risk concentration areas in the initial scheduling. For example, if the simulation finds that the power load and circulating water load are both close to their upper limits during the 10:00-11:00 time period, the system needs to stagger the tasks during that time period; if a critical valve is found to be switching frequently in a short period of time, the valve state sequence needs to be optimized to reduce operational risks; if a piece of equipment is found to be running continuously for too long, cooling or inspection nodes need to be added.

[0384] Table 5

[0385]

[0386] Referring to Table 5, which is a resource configuration conflict matrix table for a multi-tester process system, 1 in Table 5 indicates that the resource is occupied or needs to be invested, 0 indicates that it is not occupied, and −1 indicates that the task conflicts with the corresponding resource or is prohibited from being combined.

[0387] In summary, the intelligent planning solution generation process is a complex decision-making process involving multi-model collaboration, multi-round verification, multi-objective optimization, and closed-loop feedback. Through requirement analysis, the system addresses the question of "what is the task?"; through requirement-plan mapping, it addresses the question of "which process actions should be performed?"; through association rule completion, it addresses the question of "what implicit resources are needed?"; through graph embedding adaptation, it addresses the question of "which resources are most suitable?"; through conflict matrix verification, it addresses the question of "which solutions are infeasible or risky?"; through mixed-integer programming and heuristic algorithms, it addresses the question of "how to schedule tasks and allocate resources?"; through simulation, it addresses the question of whether the system solution is reliable in actual operation; and through visualization output, it addresses the question of how the system solution can be understood, executed, and monitored. This process can significantly improve the automation, intelligence, and refinement of the experimental support system's process flow planning, providing efficient, safe, and reliable end-to-end support for complex experimental tasks.

[0388] This invention discloses a method for resource matching and conflict identification in a test support system. Based on the prior aggregation of test requirements, test plans, equipment ledgers, valve-pipeline topology, historical test records, real-time operating status, and expert knowledge, it explores the construction of a unified test support knowledge data foundation. Based on the mapping relationship between test requirements and test plans, it employs a combination of the SPADE sequence pattern mining algorithm and the Apriori frequent itemset mining algorithm to generate association rules between resources such as air supply, air extraction, air pretreatment, circulating water, electricity, and valves / pipelines and test requirements. Using matrix modeling, it constructs valve-pipeline state conflict matrices, test equipment state conflict matrices, power load conflict matrices, circulating water load conflict matrices, and a comprehensive resource allocation conflict matrix, achieving an explicit expression of resource conflict relationships. Based on multi-objective Markov decision processes, mixed integer programming, graph embedding, and reinforcement learning, it constructs a dynamic resource matching model to achieve online resource optimization under changing test requirements or equipment status conditions. The resource matching results are integrated with the automatic process flow planning results to generate an intelligent scheduling scheme for test task execution, and continuous iteration is achieved through simulation and practical application feedback. This invention addresses complex test support scenarios involving multiple testers, parallel multitasking, and coupled multi-resource conditions. It constructs a unified modeling method for test requirements, test plans, process flows, equipment resources, and operational status, forming a resource dynamic matching and conflict intelligent identification technology system. This enables automatic planning of test support processes, dynamic allocation of resources, early identification of conflicts, and intelligent generation of scheduling schemes.

[0389] The above-disclosed embodiments are merely one or more preferred embodiments of this application and should not be construed as limiting the scope of this application. Those skilled in the art can understand that all or part of the processes for implementing the above embodiments and equivalent changes made in accordance with the claims of this application still fall within the scope of this application.

Claims

1. A method for matching and identifying process resources and conflicts in a test support system, characterized in that, include: S1 constructs a structured analytical model for test requirements and test plans: Based on the expert rule base, information is extracted and integrated from the test requirements and test plan data. After screening and correction, digital coding is carried out according to the different state expression rules of test requirements and test equipment. Systematic and standardized data modeling is performed on complex and diverse test requirements to generate standardized requirement information including test objects, test subjects, operating conditions, time constraints, explicit resource requirements, safety boundary constraints, and priorities. S2 employs a combination of sequence pattern mining and frequent itemset mining algorithms for association rule mining: Based on previously aggregated test requirements, test plans, equipment ledgers, valve-pipeline topology, historical test records, real-time operating status, and expert knowledge, a unified test support knowledge data foundation is constructed. The SPADE sequence pattern mining algorithm and the Apriori frequent itemset mining algorithm are combined to generate association rules between resources such as gas supply, air extraction, air pretreatment, circulating water, electricity, and valves-pipelines and test requirements. S3 constructs a resource allocation conflict matrix for a multi-testing device process system: Based on graph theory and state-space theory, the physical entity system is abstracted into a hierarchical dependent logical topological manifold. Through tensor coding and dynamic graph search algorithm, valve-pipeline state conflict matrix, test equipment state conflict matrix, power load conflict matrix, circulating water load conflict matrix and comprehensive resource allocation conflict matrix are constructed to realize the explicit expression of resource conflict relationships. The conflict matrix is ​​dynamically updated in a real-time data-driven manner. S4 constructs a task-resource adaptation model based on graph embedding: test tasks, test resources, test equipment, process nodes, valves and pipelines, operating status and expert rules are uniformly modeled as a heterogeneous graph structure. The deep adaptation relationship between tasks and resources is mined through graph representation learning. The adaptation degree between tasks and resources is calculated based on node vector similarity and comprehensive adaptation scoring function to generate a candidate resource set. S5 constructs a resource dynamic matching model based on multi-objective Markov decision-making: the resource matching problem of complex process experiment is modeled as a multi-objective Markov decision process, defining the state space, action space and multi-objective reward function, and adopting a hybrid solution approach that combines rule constraints and reinforcement learning, and realizing dynamic resource matching through state perception, action selection, reward feedback and policy update; S6 establishes a global optimization scheduling model under the condition of multiple experimental tasks running in parallel: taking into account the logical relationship of tasks, time window, resource capacity, equipment status, conflict matrix, security isolation and task priority, with the optimization objectives of the shortest total completion time, balanced resource load, minimum conflict risk, minimum task delay and minimum resource usage cost, and adopting a solution strategy that combines mixed integer programming and heuristic algorithm to generate an initial scheduling scheme. S7 constructs a method for coordinating matching of power load and circulating water load: Incorporate the power system and circulating water system into a unified scheduling model, define power-circulating water coupling risk indicators, and identify load peaks in advance during the test plan generation stage and perform peak shaving and valley filling in real time during the execution process through load forecasting, coupling risk assessment and dynamic scheduling. S8 constructs a valve-pipeline state conflict identification and path planning method: The valve-pipeline system is modeled as a directed graph with attributes, valve state conflict function and pipeline conflict function are defined, and process path planning is performed based on multi-constraint shortest path algorithm. Under the premise of satisfying connectivity, media compatibility, pressure constraints, flow constraints and valve state constraints, the optimal path with the minimum comprehensive cost is selected. S9 generates intelligent planning and scheduling solutions: It transforms the technical achievements of demand analysis, association rule mining, conflict matrix identification, dynamic resource matching, and global optimization scheduling into executable plans, generating intelligent planning solutions that include task execution sequences, task time arrangements, resource allocation schemes, valve-pipeline routes, equipment status sequences, conflict risk assessments, emergency adjustment plans, and execution monitoring indicators. Through simulation, it performs load assessment, conflict risk verification, and task completion time verification of the solutions, and iteratively optimizes them based on feedback results.

2. The method for matching and identifying process resources and conflicts in a test support system as described in claim 1, characterized in that, Step S3, constructing the valve-pipeline state conflict matrix specifically includes: Using tensor framework thinking, a node-link dynamic association model is established, which highly abstracts the complex test pipeline system into a weighted directed graph with time constraints. In this graph, nodes represent control terminals in the logical space, and edges represent fluid transmission paths that are dynamically connected according to the valve status. Construct a state topology conflict matrix. This matrix is ​​a two-dimensional square matrix. The matrix element assignment logic is as follows: the diagonal elements represent the physical state of the node itself, and the off-diagonal elements are assigned a conflict flag when there is a physical connection between two nodes and the expert Boolean function based on the process rules determines that the combination of their states triggers the risk of mutual exclusion of flow. Otherwise, they are assigned a no-conflict flag. The system recursively accesses the valve nodes starting from the initial valve node using a depth-first search algorithm. It comprehensively determines the high-dimensional nonlinear relationship between valves and performs conflict detection based on the current global state vector and the target test path sequence. When it detects that two nodes are both in the open state and the process rules determine that they are mutually exclusive, it determines that there is a logical-level resource access violation and records the conflict association pair. For the initial settings when starting up multiple testers, the local optimal state vector that satisfies the target pressure or target flow rate is solved based on the conflict matrix, so that the system can quickly enter a feasible initial operating state; Given the highly time-varying characteristics of aerodynamic systems, when the underlying control system detects valve lag or sudden changes in equipment state, it uses a sliding time window update equation to dynamically reshape the conflict matrix, ensuring that the matrix undergoes accurate state evolution as the system operates. This provides a highly real-time basic dataset and a rigorous spatiotemporal safety envelope for subsequent high-dimensional optimization.

3. The method for matching and identifying process resources and conflicts in a test support system as described in claim 2, characterized in that, In step S4, constructing the task-resource adaptation model based on graph embedding specifically includes: The test tasks, resources, equipment, process flow, valve-pipeline, and operating status are uniformly modeled as a heterogeneous graph structure. The set of nodes in the graph is divided into test task nodes, resource nodes, equipment nodes, process flow nodes, valve-pipeline nodes, and operating status nodes. The set of edges includes task-resource requirement edges, task-equipment occupation edges, resource-equipment support edges, equipment-state constraint edges, task-process dependency edges, process-valve-pipeline control edges, resource-resource conflict edges, and resource-resource substitution edges. Through unified modeling, the information scattered in the test plan, equipment ledger, pipeline diagram, expert rule base, and historical test records is transformed into a computable network structure. The Node2vec graph embedding algorithm is used to learn the low-dimensional vector representation of nodes in the graph. By adjusting the return parameters and input / output parameters, a balance is achieved between breadth-first search and depth-first search. At the same time, the local similarity and structural similarity of nodes are captured. For the experimental resource matching problem, when it is necessary to identify direct resource adaptation relationships, a larger input / output parameter is appropriately used to enhance local neighborhood learning. When it is necessary to discover alternative resources or potential resource combinations, the input / output parameter is appropriately reduced to enhance cross-regional exploration capabilities. The goal of node embedding learning is to bring closely related or semantically similar nodes in the graph closer together in the low-dimensional vector space. This is achieved through approximate optimization using negative sampling to obtain the vector representation of each task node, resource node, and device node. For task nodes and resource nodes, cosine similarity is used to calculate their compatibility, and a comprehensive compatibility scoring function is further constructed. This score comprehensively considers vector similarity, resource availability at the current moment, the degree to which resource capacity meets task requirements, the degree of conflict between resources and other assigned tasks, resource usage cost, and resource security risk. The weights of each item are dynamically adjusted according to the task type. The system employs a four-stage matching process: candidate resource screening, comprehensive scoring and ranking, conflict matrix verification, and alternative resource recommendation. The first stage filters the candidate resource set based on the vector similarity between task nodes and resource nodes. The second stage performs comprehensive scoring and ranking on the candidate resources. The third stage inputs the candidate matching results into the resource configuration conflict matrix for verification. In the fourth stage, if the optimal resource is unavailable, alternative resource combinations are recommended based on alternative edges and similarity ranking. This system supports rapid resource rematching in dynamic scenarios such as equipment failure, resource deduction, and task insertion.

4. The method for matching process resources and identifying conflicts in the test support system as described in claim 3, characterized in that, In step S5, the resource dynamic matching model based on multi-objective Markov decision specifically includes: The resource matching problem in complex process flow experiments is modeled as a multi-objective Markov decision process. The model includes a state space, an action space, state transition probabilities, a multi-objective reward vector, a discount factor, and a set of objective preference weights. At each decision moment, the system observes the current state, selects an action, the environment transitions to a new state, and provides feedback on multi-objective rewards. Through continuous iteration, the system learns resource allocation strategies under different experimental scenarios. The state space is defined as a combination of test task state, resource state, test equipment state, pipeline state, system load state, current resource conflict matrix, and historical decision information. The task execution state includes whether the task has started, the completion rate, the remaining time, priority, and time window. The resource state includes the available capacity of gas supply, extraction, air pretreatment, circulating water, fuel oil and natural gas, and power resources. The test equipment state includes available, occupied, faulty, under maintenance, and derated operation. The valve-pipeline state includes open / closed status, pressure, flow rate, medium, and path connectivity. The action space includes allocating resources to tasks, delaying low-priority tasks, executing tasks that meet the conditions in advance, switching resource paths, enabling alternative resources, adjusting load, isolating resources or tasks with risks, and triggering process replanning. To ensure that the action output meets the engineering safety requirements, rule constraints are superimposed on the action space. Actions that do not meet the safety constraints, such as pressure not exceeding limits, valve states not being mutually exclusive, equipment not being overloaded, power load not exceeding limits, and circulating water capacity not exceeding limits, are eliminated. The reinforcement learning agent selects actions only from the set of safe actions to avoid generating unacceptable dangerous strategies during the exploration process. The multi-objective reward function takes into account task efficiency, resource efficiency, cost, and security. The reward vector includes task completion reward, resource utilization reward, cost penalty, conflict penalty, security penalty, and strategy stability reward. Each objective is scalarized and weighted by a preference weight vector, which can be dynamically adjusted in different scenarios. A hybrid solution approach combining rule constraints and reinforcement learning is adopted: First, an expert rule base and conflict matrix are used to filter out unsuitable actions to ensure policy safety. Second, an Actor-Critic framework is used to learn resource matching strategies, where the Actor network outputs action probabilities and the Critic network evaluates the value of the current strategy. Third, a rolling temporal optimization method is used to regenerate the local optimal strategy based on the latest state within each decision window. At the same time, an experience replay mechanism, a priority sampling mechanism, and an online fine-tuning mechanism are introduced to improve the model's adaptability in non-stationary environments.

5. The method for matching and identifying process resources and conflicts in a test support system as described in claim 4, characterized in that, In step S6, the global optimization scheduling model under the condition of multiple experimental tasks running in parallel specifically includes: Assume that the set of experimental tasks and the set of resources are determined. Each task has attributes such as duration, earliest start time, latest finish time, priority, resource requirements, and security level. The start and end times of the tasks satisfy their respective time window constraints. If one task is a prerequisite task for another task, the start time of the subsequent task must not be earlier than the end time of the prerequisite task plus the required safety interval or process transition time between the two. The depressurization, purging and confirmation time when switching from a high-pressure gas supply task to a pipeline maintenance task, as well as the cleaning and purging time when switching tasks, are all included in the scheduling model as safety intervals. The resource capacity constraint is that the total amount of the same resource occupied by all tasks being executed at any given time shall not exceed the available capacity of that resource at that time. For power resources, the sum of the power demand of all tasks being executed at any given time shall not exceed the system power limit. For circulating water resources, the sum of the circulating water demand of all tasks being executed at any given time shall not exceed the system circulating water limit. For two tasks sharing the same valve, if their valve status requirements are mutually exclusive, the two tasks must not be executed in parallel at the same time. If they share the same pipeline, the total flow rate must not exceed the pipeline capacity. If the two media are incompatible, task isolation or cleaning and purging time must be set. The overall optimization objectives are to minimize the total completion time, minimize the deviation of each resource load from the expected utilization rate, minimize the risk of conflicts between tasks, minimize the resource usage cost, and minimize the task delay penalty. The task delay penalty is related to the task priority and delay duration, and a higher delay penalty is set for high priority tasks. The solution strategy adopted is a hierarchical partitioning-rolling time domain-heuristic optimization: First, the experimental task is hierarchically divided according to the process stage and resource type, and the global problem is broken down into three sub-problems: task scheduling, resource allocation, and control path planning. Second, the rolling time domain method is used to divide the long-cycle scheduling into multiple decision windows and perform local optimization in each window. Finally, the initial scheme is perturbed and optimized using genetic algorithms, simulated annealing, or tabu search to gradually reduce resource conflicts and load peaks. Task priority is determined by a combination of factors, including the task's importance, urgency, dependence on subsequent tasks, security risk level, and cost of task interruption. The higher the priority, the more the scheduling model tends to guarantee its resource supply and time window.

6. The method for matching and identifying process resources and conflicts in a test support system as described in claim 5, characterized in that, In step S7, the method for coordinated matching of electrical load and circulating water load specifically includes: The system calculates the total power load and total circulating water load of all tasks currently in operation in real time, and both must not exceed their respective system limits. The power-circulating water coupling risk index is defined as the weighted product of the power load rate and the circulating water load rate. The weight is used to adjust the sensitivity of power and circulating water risks. In actual operation, even if neither of them exceeds their respective upper limits, if they are close to the upper limits at the same time, it will lead to insufficient system redundancy and reduced anti-disturbance capability. Therefore, when the coupling risk index exceeds the preset threshold, it is determined that there is a high load coupling risk and the cooperative scheduling strategy is triggered. Set constraints on the rate of change of power load and the rate of change of circulating water load to avoid the risk of equipment start-up shock or cooling response lag caused by the simultaneous start-up of multiple high-power equipment or the simultaneous entry of multiple high heat load tasks into the steady state phase. The coordinated scheduling strategy includes strategies for peak power load and strategies for peak circulating water load: On the power side, measures include staggered startup of high-power equipment, postponement of low-priority auxiliary tasks, activation of backup power supply paths or zoned power supply, and setting soft-start curves according to equipment operating characteristics to reduce startup impact; On the circulating water side, measures include switching or paralleling backup cooling branches, pre-starting the circulating water system for pre-cooling, reducing the heat load of non-critical equipment, adjusting the steady-state operating time of high heat load tasks to avoid the superposition of multiple high heat tasks, and optimizing the water flow path to reduce local pipeline congestion when pipeline conditions permit. By using load forecasting, coupled risk assessment, and dynamic scheduling, load peaks are identified in advance during the test plan generation phase, and peak shaving and valley filling are carried out in real time during execution. This integrates two originally independently managed resource systems into a unified scheduling model, avoiding local optimization problems such as power availability but insufficient cooling or sufficient cooling but power supply exceeding limits.

7. The method for matching process resources and identifying conflicts in a test support system as described in claim 6, characterized in that, In step S8, the valve-pipeline state conflict identification and path planning method specifically includes: The valve-pipeline system is modeled as a directed graph with attributes. The graph contains valve nodes, pipe nodes, branch nodes, confluence nodes, interface nodes, and equipment inlet nodes, as well as edges representing pipes. Nodes have attributes such as valve number, open / closed status, control authority, response time, reliability, and safety level. Edges have attributes such as length, pipe diameter, medium type, pressure rating, flow limit, temperature range, flow direction restriction, and cleaning requirements. The process path of the test task is represented as an alternating sequence of nodes and edges from the source node to the target node. The path must satisfy connectivity constraints, media compatibility constraints, pressure constraints, flow constraints, and valve status constraints. That is, all pipelines on the path are allowed to transport the media type required by the task, the pressure required by the task does not exceed the pressure limit of all pipelines on the path, the sum of the flow rates of multiple tasks sharing the same pipeline does not exceed the capacity of the pipeline, and all valves on the path remain in the state required by the task during the execution of the task. If multiple tasks share the same valve, and the opening and closing requirements of each task for the valve are inconsistent and mutually exclusive, then a valve state conflict is determined. If multiple tasks share the same pipeline, the pipeline conflict risk is comprehensively determined from five dimensions: media type compatibility, pressure level matching, whether the flow superposition exceeds the limit, whether the flow direction is consistent, and whether cleaning and purging are required during switching. The comprehensive pipeline conflict value is obtained by weighted summation of each dimension. If the combined conflict value of two tasks sharing a path exceeds a preset threshold, the two tasks must not be executed in parallel, or the path of one of the tasks must be replanned. Traditional process planning may only consider whether resources exist and ignore whether resources can be delivered to the target equipment through reachable paths. Through valve-pipeline diagram modeling and path conflict identification, the system can determine whether the process path is feasible before scheduling. Under the premise of satisfying all process constraints, the optimal path is selected as the path that minimizes the combined costs of pipeline length, valve operation, path risk, path conflict, and switching cost. For real-time path planning, an improved shortest path algorithm or a multi-constraint shortest path algorithm is used. For simultaneous path planning of multiple tasks, a multi-device flow model or a conflict-aware graph search method is used.

8. The method for matching and identifying process resources and conflicts in a test support system as described in claim 7, characterized in that, In step S9, the intelligent planning and scheduling scheme generation method specifically includes: The intelligent planning scheme is defined as a comprehensive planning object that includes task execution sequence, task time arrangement, resource allocation scheme, valve-pipeline path, equipment and control node status sequence, conflict risk assessment, emergency adjustment plan and execution monitoring indicators; The scheme generation process is divided into eight consecutive steps: The first step is to analyze the test requirements, which involves converting natural language, task forms, test outlines, scheduling instructions, or manually entered information into a structured requirement vector that the system can recognize and calculate; the second step is to generate candidate process flow nodes based on the demand-planning mapping model; the third step is to call the association rule base to complete the implicit resource requirements; the fourth step is to construct a candidate resource set and calculate the task-resource suitability through a graph embedding model; the fifth step is to input the candidate resources into the conflict matrix for verification and eliminate schemes with hard conflicts; the sixth step is to generate an initial schedule using mixed integer programming and heuristic algorithms; the seventh step is to evaluate resource load, conflict risk, and task completion time based on simulation; and the eighth step is to output a visualized schedule. Iterative optimization is achieved between each step through a feedback mechanism. The simulation is based on the timeline and simulates the execution process of each task node according to the planned sequence. It calculates the occupancy, load level, conflict risk and safety margin of various resources in each time slice. The simulation evaluation indicators include task completion time, resource utilization rate, load peak rate, cumulative conflict risk value and task delay rate. The above indicators are used to determine whether there are resource bottlenecks and risk concentration areas in the initial schedule. If the power load and circulating water load are found to be close to the upper limit at the same time, peak shifting is carried out. If the critical valves are frequently switched, the valve state sequence is optimized to reduce operational risks. If the equipment runs continuously for too long, cooling or inspection nodes are added. The intelligent scheduling output has multi-view expression capabilities, including a task view showing the start time, end time, execution stage and dependencies of each test task; a resource view showing the occupancy of various resources at different times and load curves; a conflict view showing the degree of conflict between tasks, resources and paths; a control view showing the valve opening and closing sequence and pipeline path; and a risk view showing the risk levels of power, circulating water, equipment and safety boundary. Through multi-view linkage, test personnel can intuitively understand the system-generated scheme and manually confirm and adjust it as needed.

Images