Test task scheduling method based on phase adaptive context weight modeling

Abstract

The invention discloses a test task scheduling method based on phase adaptive context weight representation. The test task scheduling method comprises the following steps: data preprocessing: converting an original test demand into uniform task vector representation; context modeling: performing dynamic coding on different task features through construction of anchor point topology and a self-adaptive weight mechanism; scheduling generation: constructing a task list under the driving of a global phase evolution rule; and the verification feedback is used for carrying out step-by-step verification on the task list through a hierarchical certificate mechanism and realizing feedback correction. According to the method, a weight modeling mechanism driven by context features is introduced, and a scheduling scheme which not only can ensure the performability, but also has the dynamic self-adaptive capability is constructed; performing structured modeling on the key threshold condition based on the anchor point topological graph, and generating a hierarchical feasibility certificate on the basis; layer-by-layer verification and automatic correction of the task list can be realized, and the security, the performability and the resource utilization rate of the finally output daily scheduling list are ensured.

Description

Experimental Task Scheduling Method Based on Phase Adaptive Context Weight Evolution Technical Field

[0001] This invention relates to the field of aero-engines, and more particularly to a test mission scheduling method based on phase adaptive context weight evolution. Background Technology

[0002] In the field of aero-engines, to verify the performance and reliability of engines under different environments, various ground simulation tests are typically conducted in specialized facilities such as high-altitude test benches. These tests include in-flight start-up tests, high-altitude performance tests, thrust transient tests, and control system adjustment tests. Each type of test often consists of multiple environmental conditions, each involving multi-dimensional parameters such as temperature, pressure, altitude, Mach number, speed, test duration, and equipment resources. Test benches usually have a list of environmental conditions for various tests, along with clear operational descriptions for each condition (e.g., specifying that when the temperature is 20°C, the altitude is 0km, and the pressure reaches 5kPa, specific equipment needs to be activated and corresponding steps need to be performed; when the conditions are switched to high-altitude conditions, different preparations and adjustments are required). The core objective of scheduling is to automatically generate a reasonable daily work list based on this data.

[0003] In actual scheduling, experimental tasks often need to consider multiple constraints across parameter dimensions. These include, but are not limited to, strict anchor thresholds, such as extreme temperature points, critical height points, or available windows for critical equipment. Different tasks cannot be placed in the same list when crossing these anchor conditions. Furthermore, the arrangement of the experimental list must adhere to the daily working time principle; for example, the total daily task duration should be controlled within 6-7 hours to avoid significant overtime, and should be as close as possible to the target duration to improve equipment utilization and personnel efficiency. Within the same list, tasks typically need to be arranged in a monotonic order based on height or other parameters to avoid additional losses caused by frequent switching. These constraints collectively determine the complexity of the scheduling problem, making manual scheduling time-consuming and labor-intensive, and conventional solutions often struggle to generate a feasible and optimized list.

[0004] Current experimental task scheduling methods can be mainly divided into three categories: The first category is scheduling methods based on manual rules or expert experience, which create a list by manually setting priorities and constraints. This type of method is simple to implement and can quickly obtain preliminary solutions for small-scale tasks, but due to the lack of dynamic adjustment capabilities, it is difficult to maintain effectiveness when there are diverse parameters and complex constraints.

[0005] The second category is modeling methods based on mathematical optimization or constraint programming, such as solving integer programming or constraint satisfaction problems (CSP). These methods can theoretically obtain better solutions, but their computational complexity is extremely high, especially when the task scale is large, the time cost of generating solutions is too high, and they lack flexibility for temporary insertions or urgent tasks.

[0006] The third category consists of data-driven methods that have been increasingly adopted in recent years, including machine learning and reinforcement learning. These methods infer and optimize scheduling strategies by training on historical mission data. While these methods have certain advantages in adaptability and self-learning, they often lack interpretability and robustness when faced with strict threshold constraints and multidimensional coupling relationships in aerospace test missions, and are prone to producing solutions that are not feasible in engineering.

[0007] The shortcomings and deficiencies of existing technologies are as follows: 1. Insufficient dynamic adaptability. Many methods use fixed weights or priorities, which cannot be adjusted according to scheduling stages and context changes, making it difficult for the generated inventory to cope with the differences in the importance of tasks at different stages.

[0008] 2. The anchor point threshold is handled in a coarse manner. Common methods usually treat it as a hard boundary, ignoring the buffer zone characteristics that should exist near the critical condition. This can easily lead to cross-boundary tasks near the threshold, making the actual inventory unexecutable.

[0009] 3. The lack of a systematic verification and correction mechanism means that the results generated by existing methods often require manual inspection and revision, which is inefficient and cannot accumulate optimization experience through automatic feedback.

[0010] 4. Limited scalability: When the task scale expands or the parameter dimensions increase, manual rule methods are difficult to maintain, mathematical optimization methods have too high computational cost, and learning methods have insufficient transferability in different experimental environments, making it difficult to guarantee long-term stability and universality. Summary of the Invention

[0011] The purpose of this invention is to overcome the shortcomings of the prior art and provide a test task scheduling method based on phase adaptive context weight evolution.

[0012] The objective of this invention is achieved through the following technical solution: In its first aspect, this invention provides a method for scheduling experimental tasks based on phase adaptive context weight evolution, comprising the following steps: data preprocessing: used to transform the original experimental requirements into a unified task vector representation; context modeling: used to dynamically encode different task features through the construction of anchor topology and an adaptive weight mechanism; scheduling generation: used to construct a task list driven by global phase evolution rules; and verification feedback: used to perform step-by-step verification of the task list and implement feedback correction through a hierarchical certificate mechanism.

[0013] Furthermore, in data preprocessing, the transformation of the original experimental requirements into a unified task vector representation includes: transforming each task into a... The feature vector is represented in dimensional form, with each dimension representing one of the elements corresponding to the task. Continuous features retain their original values ​​or are normalized. Categorical features are processed through one-hot encoding or embedding vectorization. Textual step descriptions are transformed into low-dimensional representations through keyword extraction or predefined mapping tables. When a feature vector of a certain dimension meets specific conditions, constraint labels are added to the metadata of the entire feature vector.

[0014] Furthermore, in the context modeling step, the construction of the anchor topology includes: the anchor point is a critical threshold in the task parameters, and a certain buffer is defined around it to determine the potential coupling strength between tasks; all tasks constitute a node set, where each node corresponds to a task, and the task attributes are vectorized; for any two nodes, firstly, the global basic similarity is calculated, and then the global basic similarity is modulated using a directional anchor modulation factor to obtain the edge weight; the edge set is sparsified, retaining only the edges with edge weights greater than the threshold, and the final anchor topology graph contains the node set and the edge set.

[0015] Furthermore, in the context modeling step, the adaptive weighting mechanism dynamically encodes different task features, including: establishing a dynamic weighting mechanism for the task set, generating context-related offset weights, i.e., context weights, for each task using attention mapping to characterize the actual priority of the task under the current environment and constraints; wherein the information sources of the context include: the task grouping results of the previous stage, the constraint check status of the hierarchical feasibility certificate feedback, the execution data of the historical list, and the status of currently available resources and remaining capacity; the entire scheduling process is divided into multiple stages, including the initial grouping stage, the task packaging stage, and the sequence sorting stage, each stage corresponding to a global phase weight, the elements of which reflect the importance of different feature dimensions in the corresponding stage; the comprehensive weight of each task in the corresponding stage is a weighted combination of context weights and global phase weights based on a mixed ratio.

[0016] Furthermore, in the scheduling generation step, the construction of the task list driven by the global phase evolution rule includes: weight initialization, before entering scheduling, firstly calculating the initial context weight vector of each task based on the task set and anchor topology graph; phase partitioning, dividing the entire scheduling process into multiple stages, including task grouping stage, task packaging stage, and sequence sorting stage; each stage presets a global phase weight vector and mixing ratio according to different target characteristics to control the fusion ratio between global phase weight and context weight; task selection, in each stage, calculating the comprehensive weight vector of all unscheduled tasks, and sorting the tasks using a priority queue or heuristic search; subsequently, candidate groups are gradually formed by seed expansion: firstly, the task with the highest weight is selected as the group seed, and then based on the edge weights and coupling relationships of the anchor topology graph, multiple strongly related tasks are absorbed to form local groups. Grouping; during this process, only the remaining tasks that share resources or are strongly coupled with the current group are subject to local weight updates to reduce computational complexity; list updates and context corrections: when a group is committed to be added to the list, the context environment is immediately updated, including the resource occupancy table, anchor buffer status, and task chain progress; for tasks affected by this group, including tasks that share resources or are in the same anchor buffer, the context weights are recalculated, and local corrections are triggered if necessary; if the resource occupancy rate or number of conflicts reaches a set threshold, a global context recalculation is triggered to ensure overall feasibility; weight evolution and phase switching: as the scheduling enters the next phase, the comprehensive weights of all remaining tasks are recalculated based on the new global phase weights and mixing ratios; in this phase, the task selection and list updates and context corrections are repeated until all tasks are assigned, forming a complete scheduling list.

[0017] Furthermore, in the verification feedback step, the layered certificate mechanism includes: a layered feasibility certificate, used to perform multi-dimensional verification of the daily task list to ensure the executability of the list in terms of parameters, resources, timing, and global topology; the feasibility certificate includes: a parameter layer certificate (PFC), used to check whether each task meets the key parameter anchor point constraints; for tasks exceeding the threshold, it records deviation values ​​or Boolean violation flags for subsequent correction; a resource layer certificate (RFC), used to check the device resource allocation of tasks within the list; it marks conflicting tasks and can provide reallocation suggestions; a timing layer certificate (SFC), used to verify whether the execution order of tasks meets the dependency relationship; it inserts interval tasks or adjusts the order of sequences that do not meet the requirement to ensure the integrity of the chain; a global layer certificate (GFC), used to verify the global connectivity based on the anchor point topology graph; and a global layer certificate used to synthesize the results of each layer to generate an overall executability score for the list.

[0018] Furthermore, in the verification feedback step, the step-by-step verification of the task list includes: inputting the candidate task list and anchor point topology diagram; verifying layer by layer, generating parameter layer certificate PFC, resource layer certificate RFC, and time sequence layer certificate SFC in sequence, with each layer independently verifying the corresponding constraints; synthesizing, summarizing the results of each layer to generate a global layer certificate GFC, and calculating the overall compliance score; outputting, if all layer constraints are satisfied, the list is marked as executable; if any layer fails, a local correction mechanism is triggered.

[0019] Further, in the verification feedback step, the implementation feedback correction includes: dynamic list correction, used to process tasks that fail the parameter layer certificate PFC, adjust tasks that fail the resource layer certificate RFC, adjust task chains that fail the time sequence certificate SFC, and optimize the topology for cases where the global layer certificate GFC fails; the dynamic list correction adopts an iterative approach, with the following process: initial list input, inputting candidate lists; certificate verification, performing four-layer certificate verification (parameter layer certificate PFC, resource layer certificate RFC, time sequence certificate SFC, and global layer certificate GFC) on the candidate lists, generating verification feedback; local correction, calling the corresponding correction strategy to locally adjust the task set based on the verification results of each layer; list update, generating a new list version and updating the anchor topology relationship and context weight; iterative judgment, if... If all layers of feasibility certificates are verified, the final list is output; otherwise, the process proceeds to the next iteration.

[0020] Furthermore, in the verification feedback step, the implementation of feedback correction also includes: a convergence mechanism, which, to ensure that the correction process converges within a limited number of rounds, specifically includes: priority convergence control: in multiple iterations, the task context weights are gradually solidified to avoid large fluctuations and ensure that the list sorting and grouping gradually stabilize; local optimum freezing: sub-lists that have passed the hierarchical feasibility certificate verification are marked as "frozen areas" and will not be adjusted in subsequent iterations to prevent repeated modifications from causing oscillations; convergence criterion: if the difference between the lists is less than a set threshold in two consecutive iterations, the lists are considered to have converged, and the final executable version is output.

[0021] The beneficial effects of this invention are as follows: In an exemplary embodiment of this invention, by performing structured vectorized modeling of experimental tasks, introducing a context-feature-driven weight evolution mechanism, and combining a multi-stage phase scheduling strategy and a hierarchical feasibility verification mechanism, a scheduling scheme that ensures both executability and dynamic adaptability is constructed. Compared with existing methods, this exemplary embodiment can comprehensively characterize the complex coupling relationships between multi-dimensional experimental parameters and support dynamic adjustment of the importance weights of different parameters during scheduling, avoiding the rigidity problem of traditional fixed-weight schemes. Simultaneously, by performing structured modeling of key threshold conditions based on anchor point topology graphs and generating hierarchical feasibility certificates, this exemplary embodiment can achieve layer-by-layer verification and automatic correction of the task list, ensuring that the final daily scheduling list is significantly superior to existing manual rule, mathematical optimization, or learning methods in terms of safety, executability, resource utilization, and cross-scenario adaptability. Attached Figure Description

[0022] Figure 1 is a flowchart of an experimental task scheduling method based on phase adaptive context weight evolution provided by an exemplary embodiment of the present invention; Figure 2 is a schematic diagram of an anchor topology graph (ATG) provided by an exemplary embodiment of the present invention; Figure 3 is a schematic diagram of the phase adaptive context weight evolution process provided by an exemplary embodiment of the present invention; Figure 4 is a schematic diagram of a dynamic correction and convergence mechanism for the inventory provided by an exemplary embodiment of the present invention. Detailed Implementation

[0023] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0024] Referring to Figure 1, Figure 1 shows a flowchart of an experimental task scheduling method based on phase adaptive context weight evolution provided by an exemplary embodiment of the present invention, including the following steps: data preprocessing: used to transform the original experimental requirements into a unified task vector representation; context modeling: used to dynamically encode different task features through the construction of anchor topology and adaptive weight mechanism; scheduling generation: used to construct a task list under the drive of global phase evolution rules; verification feedback: used to perform step-by-step verification of the task list through a hierarchical certificate mechanism and realize feedback correction.

[0025] Specifically, in this exemplary embodiment, data preprocessing enables task element parsing and vectorization, context modeling enables anchor point extraction and weight evolution, scheduling generation enables list construction based on phase evolution, and verification feedback enables hierarchical feasibility certificates and list correction.

[0026] By constructing a structured, vectorized model of experimental tasks, introducing a context-feature-driven weight evolution mechanism, and combining a multi-stage phase scheduling strategy with a hierarchical feasibility verification mechanism, a scheduling scheme that guarantees both executability and dynamic adaptability is constructed. Compared with existing methods, this exemplary embodiment can comprehensively characterize the complex coupling relationships between multi-dimensional experimental parameters and support dynamic adjustment of the importance weights of different parameters during scheduling, avoiding the rigidity problem of traditional fixed-weight schemes. Simultaneously, by structurally modeling key threshold conditions based on anchor point topology graphs and generating hierarchical feasibility certificates, this exemplary embodiment can achieve layer-by-layer verification and automatic correction of the task list, ensuring that the final daily scheduling list significantly outperforms existing manual rule, mathematical optimization, or learning-based methods in terms of safety, executability, resource utilization, and cross-scenario adaptability.

[0027] The following will elaborate on the preferred exemplary embodiments: More preferably, in one exemplary embodiment, in data preprocessing, the transformation of the original experimental requirements into a unified task vector representation includes: transforming each task into a... The feature vector is represented in dimensional form, with each dimension representing one of the elements corresponding to the task. Continuous features retain their original values ​​or are normalized. Categorical features are processed through one-hot encoding or embedding vectorization. Textual step descriptions are transformed into low-dimensional representations through keyword extraction or predefined mapping tables. When a feature vector of a certain dimension meets specific conditions, constraint labels are added to the metadata of the entire feature vector.

[0028] Specifically, in this exemplary embodiment, the existing test task data is first structured and vectorized into a model. The test task data originates from pre-collected test conditions, presented in tabular form (Excel, CSV, or database format), containing several records, each corresponding to a specific test task. In a specific exemplary embodiment, taking an aero-engine test scheduling scenario as an example, each task typically includes the following elements: temperature (T), pressure (P), altitude (H), Mach number (M), speed (V), estimated execution time (L), and corresponding equipment requirements, task priority, and operational procedure descriptions. For applications in different fields, additional attributes such as test risk level, task dependencies, and test phase labels can also be added.

[0029] To facilitate subsequent processing, this exemplary embodiment converts each task into a single task. 3D eigenvector representation: ;in, Indicates the first Feature vectors of each task This indicates that the task is in the... Values ​​can be taken in each dimension. For example, when When a vector is used, it can be specifically represented as: During vectorization, continuous features (such as temperature, pressure, altitude, speed, etc.) retain their original values ​​or undergo normalization to ensure comparability of features under different dimensions; categorical features (such as task type, equipment number) can be processed through one-hot encoding or embedding vectorization; textual step descriptions can optionally be converted into low-dimensional representations through keyword extraction or predefined mapping tables.

[0030] Meanwhile, to accommodate subsequent constraint modeling, this exemplary embodiment introduces constraint labels during vector modeling. For example, when the temperature is greater than 130°C, the corresponding temperature dimension is labeled as "high temperature condition"; when the altitude is greater than 15km, the corresponding altitude dimension is labeled as "high altitude condition". These labels do not replace the original values, but serve as auxiliary identifiers and anchor point conditions for subsequent anchor point topology map construction and partitioning.

[0031] For example: Suppose an Excel spreadsheet records an experimental task with the following parameters: temperature 145℃, pressure 0.9 atm, altitude 16km, Mach number 2.3, speed 800km / h, and estimated execution time 2.0 hours. Its vectorized representation would be: Simultaneously, two constraint labels are added to the metadata of this vector: "High Temperature Conditions" and "High Altitude Conditions". Through the above steps, the original task table is uniformly transformed into a set of vectorized tasks: ;in, The set represents the number of tasks. This set retains the numerical characteristics of the original conditions while adding constraint-related label information, thus providing standardized input for subsequent context weight evolution, anchor point protection, and inventory generation.

[0032] More preferably, in an exemplary embodiment, the construction of the anchor topology in the context modeling step includes: the anchor point is a critical threshold in the task parameters, and a certain buffer is defined around it to determine the potential coupling strength between tasks; all tasks constitute a node set, where each node corresponds to a task, and the task attributes are vectorized; for any two nodes, firstly, the global basic similarity is calculated, and then the global basic similarity is modulated using a directional anchor modulation factor to obtain the edge weight; the edge set is sparsified, and only edges with edge weights greater than the threshold are retained, and the final anchor topology graph includes the node set and the edge set.

[0033] Specifically, after completing the task vectorization modeling, this exemplary embodiment further proposes a method for constructing an Anchor Topology Graph (ATG) to explicitly characterize the relative positional relationships and constraint conflicts of each experimental task across key parameter dimensions. By establishing the ATG, the anchor coupling relationships between tasks can be clearly represented globally, thus providing a structured constraint basis for subsequent context weight modeling and phase evolution mechanisms.

[0034] Anchor points are defined based on critical thresholds in task parameters. For example, in high-altitude test missions, dimensions such as temperature, pressure, altitude, Mach number, and speed all have preset physical or engineering critical points. When the parameter value of a task approaches or crosses this critical point, its execution stability and resource constraints in the task sequence will change significantly. Therefore, this exemplary embodiment defines these critical points as "anchor points" and delineates a certain buffer zone around them to determine the potential coupling strength between tasks. In ATG, anchor points are not only reference points in the parameter space but also important benchmarks for characterizing task constraint conflicts and coupling relationships.

[0035] In ATG, all tasks constitute a set of nodes. Each node Corresponding task Its attribute is vectorized representation. For any two nodes and First, calculate their global basic similarity: ;in, This is a scale parameter used to control the degree of decay in global similarity between tasks.

[0036] To more accurately represent the interrelationships of tasks along the anchor point dimension, this exemplary embodiment introduces a directional anchor point modulation factor. Let the anchor point threshold along parameter dimension k be... ,Task , The parameter values ​​in this dimension are respectively , Then its signed distance relative to the anchor point is defined as: The sign indicates whether the task is to the left or right of the anchor point.

[0037] Based on this definition, this exemplary embodiment proposes the following anchor modulation factor: ;in, This is a sigmoid buffer function used to achieve smooth decay within the neighborhood of the anchor point; The buffer radius represents the range of flexible task coupling allowed near the anchor point. This is the same-side reward coefficient, used to strengthen the coupling between tasks on the same side of the anchor point; This is an indicator function that takes the value 1 when two tasks are on the same side of the anchor point, and 0 otherwise.

[0038] This approach ensures that when a task crosses an anchor point, even if the absolute distance between them is equal, the edge weight still decays significantly, thus reflecting the switching barrier brought about by crossing the critical point; conversely, tasks on the same side and close in distance receive additional enhancement.

[0039] Combining the basic similarity with the anchor modulation factor across all parameter dimensions, the final edge weight is defined as follows: Where d is the total number of parameter dimensions. This is the normalized anchor modulation factor to ensure numerical stability across different dimensions.

[0040] To avoid an overly dense graph structure, this exemplary embodiment further sparsifies the edge set E, retaining only edges with weights greater than a threshold. The edge, that is: The final anchor point topology map As shown in Figure 2, Represents a set of nodes. The topological graph, which represents the set of edges, can not only reflect the global similarity of tasks in the parameter space, but also characterize the critical segmentation effect brought about by the directional constraints of anchor points, thereby more realistically simulating the coupling mode of experimental tasks in the process of resource scheduling and execution.

[0041] In summary, in this exemplary embodiment, by constructing the anchor point topology graph and modeling the buffer, the key threshold conditions and the coupling relationship between tasks can be accurately characterized in the multi-dimensional parameter space, avoiding the problem of cross-boundary error in task arrangement near the critical conditions, and greatly improving the executability and stability of the scheduling scheme.

[0042] More preferably, in an exemplary embodiment, in the context modeling step, the adaptive weighting mechanism dynamically encodes different task features, including: establishing a dynamic weighting mechanism for the task set, generating context-related offset weights, i.e., context weights, for each task using attention mapping, to characterize the actual priority of the task under the current environment and constraints; wherein the information sources of the context include: the task grouping results of the previous stage, the constraint check status of the hierarchical feasibility certificate feedback, the execution data of the historical list, and the current available resources and remaining capacity status; dividing the entire scheduling process into multiple stages, including the initial grouping stage, the task packaging stage, and the sequence sorting stage, each stage corresponding to a global phase weight, the elements of which reflect the importance of different feature dimensions in the corresponding stage; the comprehensive weight of each task in the corresponding stage is a weighted combination of the context weight and the global phase weight based on a mixed ratio.

[0043] Specifically, in this exemplary embodiment, after completing task vectorization modeling and anchor point topology graph (ATG) construction, this exemplary embodiment further proposes a phase-adaptive contextual weight evolution (PACWE) method to dynamically generate comprehensive weight vectors for tasks during scheduling, guiding task grouping, packaging, and inventory generation. Unlike traditional fixed-weight scheduling methods, PACWE can gradually evolve weights according to changes in the context environment at different stages, thereby significantly improving the flexibility and robustness of scheduling.

[0044] First, in the context weight modeling phase, this exemplary embodiment models the task set... Establish a dynamic weighting mechanism to generate context-dependent offset weight vectors for each task. This is used to characterize the actual priority of tasks under the current environment and constraints. Contextual information comes from multiple sources, including: the task grouping results of the previous stage; the constraint check status feedback from the Hierarchical Feasibility Certificate (HFC); execution data from historical lists (such as timeout records, default frequency, etc.); and the status of currently available resources and remaining capacity.

[0045] For the first For each task, its context offset weight vector can be represented as: ;in, For task vectors, For the stage The context indicates that The attention mapping function can be implemented using weighted averaging, attention networks, or shallow neural networks. This modeling process automatically identifies the importance of a task under different constraints. For example, when a task has multiple timeout records, its context weight in the "time" dimension will be significantly increased, thus guiding subsequent scheduling to prioritize this task.

[0046] Secondly, in the phase evolution mechanism, this exemplary embodiment divides the entire scheduling process into multiple phases, such as the initial grouping phase, the task packaging phase, and the sequence sorting phase. Each phase corresponds to a global phase weight vector. Its elements reflect the importance of different feature dimensions at each stage. For example, in the initial grouping stage, temperature and altitude parameters have a greater weight; while in the final sorting stage, task duration and order constraints have a higher weight. The formula is defined as follows: ;in, Indicates the first stage The importance coefficients of each feature dimension. For example, in the initial grouping stage, temperature and altitude have higher weights; in the sorting stage, task duration and order constraints have more important weights.

[0047] In the phase evolution mechanism, each task is in a phase The overall weight vector is a weighted combination of the context weights and the global phase weights: ;in, This is a hybrid scaling factor for phase weighting and context correction. This is a normalization function that makes the sum of all components of the weight vector equal to 1.

[0048] In summary, in this exemplary embodiment, the phase-adaptive context weight evolution mechanism enables dynamic adjustment of task feature weights based on environmental constraints and context states at different scheduling stages, overcoming the rigidity of traditional fixed weight methods and thus significantly improving the flexibility and adaptability of scheduling.

[0049] More preferably, in an exemplary embodiment, in the scheduling generation step, the construction of the task list under the drive of the global phase evolution rule, as shown in Figure 3, includes: Step 1 Weight initialization, before entering the scheduling, firstly based on the task set... With anchor point topology graph Calculate the initial context weight vector for each task. 0).

[0050] This calculation comprehensively considers the distance between task parameters and anchor thresholds, historical execution performance (such as timeout frequency), resource availability, and the coupling strength between tasks. The result forms a set of dynamic weight vectors reflecting the initial priority of tasks in the current environment, providing input for subsequent stages.

[0051] Step 2, Phase Partitioning, divides the entire scheduling process into multiple stages, including task grouping, task packaging, and sequence sorting. For each stage, a global phase weight vector and mixing ratio are preset based on different target characteristics to control the global phase weight. Fusion ratio with context weights Step 3: Task Selection. In each stage, calculate the combined weight vector of all unscheduled tasks. The process involves prioritizing tasks using priority queues or heuristic search, then gradually forming candidate groups using a seed expansion approach: first, the task with the highest weight is selected as the group seed, and then, based on the edge weights and coupling relationships of the anchor topology graph, multiple strongly related tasks are incorporated to form local groups. During this process, only the remaining tasks that share resources or are strongly coupled with the current group undergo local weight updates to reduce computational complexity. Step 4 involves list updates and context correction: once a group is committed to the list, the context environment is immediately updated, including the resource occupancy table, anchor buffer status, and task chain progress. For tasks affected by this group... (e.g., tasks sharing resources or within the same anchor buffer) recalculate context weights. And trigger local corrections when necessary; if resource utilization or number of conflicts reaches a set threshold, trigger a global context recalculation to ensure overall feasibility; Weight evolution and phase switching occur as the schedule moves to the next phase, based on the new global phase weights. and mixing ratio Recalculate the overall weights for all remaining tasks. During this stage, the Step 3 task selection and the above steps are repeated. The list is updated and the context is corrected until all tasks have been assigned, forming a complete scheduling list.

[0052] In summary, the PACWE mechanism allows the weight of each task to evolve dynamically with the stage and context, effectively avoiding the rigidity of fixed-weight schemes and improving the flexibility, feasibility, and overall coordination of the scheduling scheme.

[0053] More preferably, in an exemplary embodiment, the layered certificate mechanism in the verification feedback step includes: a layered feasibility certificate, used to perform multi-dimensional verification on the daily task list to ensure the executability of the list in terms of parameters, resources, timing, and global topology; the feasibility certificate includes: a parameter layer certificate PFC, used to check whether each task meets the key parameter anchor point constraints; for tasks exceeding the threshold, a deviation value or a Boolean violation mark is recorded for subsequent correction; a resource layer certificate RFC, used to check the device resource allocation of tasks in the list; conflicting tasks are marked and a reallocation suggestion can be provided; a timing layer certificate SFC, used to verify whether the execution order of tasks meets the dependency relationship; for sequences that do not meet the dependency relationship, an interval task is inserted or the order is adjusted to ensure the integrity of the chain; a global layer certificate GFC, used to verify the global connectivity based on the anchor point topology graph; the global layer certificate is used to integrate the results of each layer to generate an overall executability score for the list.

[0054] Specifically, to achieve systematic verification and automatic correction of the task list, this exemplary embodiment further proposes a hierarchical feasibility certificate (HFC) for multi-dimensional verification of the daily task list, ensuring its executability in terms of parameters, resources, timing, and global topology. The feasibility certificate is divided into four layers as follows: Parameter-level Feasibility Certificate (PFC): This primarily checks whether each task meets key parameter anchor point constraints, including whether the temperature exceeds high or low temperature thresholds; whether the pressure exceeds safe limits; and whether the altitude meets task zoning requirements. For tasks exceeding the thresholds, the deviation value is recorded. Alternatively, a Boolean violation flag can be used for subsequent corrections.

[0055] Resource-level Feasibility Certificate (RFC): Checks the device resource allocation for tasks in the inventory, including whether the same device is repeatedly used within the same time period; the probability of resource conflicts between tasks; and device load balancing. It marks conflicting tasks and provides reallocation suggestions.

[0056] Sequence-level Feasibility Certificate (SFC): Verifies whether the execution order of tasks satisfies dependencies, whether preceding tasks have been completed, whether cooldown intervals meet security requirements, and whether critical tasks are sequential. For sequences that do not meet these requirements, it inserts interval tasks or adjusts the order to ensure the chain remains intact.

[0057] Global Feasibility Certificate (GFC): Verifies the global connectivity of the anchor point topology graph, checking for loop conflicts or isolated nodes in the task chain; checking whether edges crossing anchor point barriers are correctly cut or decayed; and verifying the reasonable allocation of tasks within the buffer. The GFC is used to synthesize the results from each layer to generate an overall executability score for the manifest.

[0058] More preferably, in an exemplary embodiment, the step of verifying the task list step by step in the verification feedback step includes: inputting a candidate task list and an anchor point topology map. The process involves layer-by-layer verification, generating parameter-layer certificates (PFC), resource-layer certificates (RFC), and sequence-layer certificates (SFC) sequentially, with each layer independently verifying its corresponding constraints. Finally, the results from each layer are combined to generate a global certificate (GFC), and the overall compliance score is calculated. Output: If all layer constraints are satisfied, the list is marked as executable; if any layer fails, a local correction mechanism is triggered.

[0059] It should be noted that HFC supports tree-structured representation and can output in JSON, XML, or tabular format. For example, the certificate for a certain list can be represented as: {"list_id": "Day_01","global_score": 0.85,"checks": {"parameter_constraints": {"temperature_rule": {"pass": true},"altitude_rule": {"pass": true},"pressure_rule": {"pass": false, "deviation": +0.5}},"resource_constraints": {"device_conflict": {"pass": true},"load_balance": {"score": 0.92}},"sequence_constraints": {"task_order": {"pass": true},"cooling_interval": {"pass": false, "violations": 1}},"topology_checks": {"connectivity": {"status": "ok"},"anchor_crossings": {"count": In the example above, the pressure in the parameter layer exceeds the threshold by 0.5 units; there is one cooling interval violation in the time series layer; the task connectivity is normal in the topology check, but there is one cross-anchor task; the overall score is 0.85, which can be used for automatic optimization and manual review.

[0060] More preferably, in an exemplary embodiment, the implementation feedback correction in the verification feedback step, as shown in FIG4, includes: dynamic inventory correction, used to process tasks that fail the parameter layer certificate PFC, adjust tasks that fail the resource layer certificate RFC, adjust task chains that fail the sequence layer certificate SFC, and perform topology optimization for cases where the global layer certificate GFC fails; the dynamic inventory correction adopts an iterative approach, and the process is as follows: initial inventory input, input candidate inventory. Certificate verification: Performs four-layer certificate verification on the candidate list: parameter layer certificate PFC, resource layer certificate RFC, timing layer certificate SFC, and global layer certificate GFC, generating verification feedback; Local correction: Based on the verification results of each layer, calls the corresponding correction strategy to make local adjustments to the task set; List update: Generates a new list version. It updates the anchor point topology and context weights; iteratively checks if the new list version... If all layers of feasibility certificates are verified, the final list is output; if not, the next iteration begins, and the corresponding formula is shown in Equation 3-13. ;in, Indicates the first Round list, )) indicates the results of the hierarchical feasibility certificate verification for the list. This represents a local correction operator based on verification feedback.

[0061] Specifically, in the scheduling of aero-engine test missions, even with the integration of Context Weighted Modeling (PACWE), Anchor Topology Graph (ATG) partitioning, and Hierarchical Feasibility Certificate (HFC) verification, the initially generated daily inventory may still exhibit local infeasibility or insufficient optimization, such as mission parameters exceeding thresholds, resource conflicts, or incomplete time-series chains. Therefore, in this exemplary embodiment, a dynamic inventory correction and convergence mechanism (the next exemplary embodiment) is proposed. Through iterative correction strategies and convergence control, the global feasibility and stability of the daily inventory are achieved. Specifically: Parameter Adjustment: Tasks that fail the Parameter Layer Certificate (PFC) are processed. For example, if the mission temperature or pressure exceeds the anchor threshold, the context weight of the task is reduced or its execution order is postponed. For tasks with excessively large continuous parameter changes, the order of adjacent tasks can be rearranged to improve smoothness.

[0062] Resource Reallocation: Adjustments are made to tasks whose Resource Layer Certificates (RFCs) have failed. Resource conflict task sets are identified in the anchor topology graph. Resources are reallocated according to task priority, and task order is swapped if necessary. Ensure that the same resource is occupied by only a single task at the same time.

[0063] Sequence Adjustment: Adjusts task chains that do not meet the Sequence Certificate (SFC) requirements by inserting cooling or interval tasks to ensure the completion of preceding tasks; performs partial rearrangement of task chains to satisfy dependency order and safe interval.

[0064] Global Topology Adjustment: Performs topology optimization for cases where the Global Certificate of Creation (GFC) fails. It uses the topology cut set algorithm to identify conflicting subgraphs, minimizes edge breaks to resolve circular dependencies or broken task chains, and preserves the overall inventory structure and local connectivity.

[0065] The correction and iteration process is as shown above and will not be elaborated here.

[0066] In summary, in this exemplary embodiment, by combining the hierarchical feasibility certificate verification mechanism with the dynamic list correction strategy, the generated task list can be automatically verified and iteratively optimized at multiple levels, ensuring that the final output solution meets the feasibility requirements in terms of parameters, resources, timing, and global topology, thereby improving the reliability and engineering applicability of the scheduling process.

[0067] More preferably, in an exemplary embodiment, the implementation of feedback correction in the verification feedback step further includes: a convergence mechanism, specifically including: priority convergence control: in multiple iterations, the task context weights are gradually solidified to avoid large fluctuations and ensure that the list sorting and grouping gradually stabilize; local optimum freezing: sub-lists that have passed the hierarchical feasibility certificate verification (HFC verification) are marked as "frozen areas" and will not be adjusted in subsequent iterations to prevent repeated modifications from causing oscillations; convergence criterion: if the difference between the lists in two consecutive iterations is less than the expected value, the implementation of feedback correction will be further optimized. Less than the set threshold If the list has converged, the final executable version will be output: in, Indicates task In the list The position or allocation status within.

[0068] Through the above mechanism, this exemplary embodiment can automatically correct locally infeasible or conflicting tasks in the initial list during multiple iterations; maintain the stability of the overall list structure and avoid repeated large-scale adjustments; and finally generate a daily execution list that is globally feasible, has smooth parameters, is reasonably ordered, and meets resource constraints.

[0069] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A test task scheduling method based on phase adaptive context weight evolution, characterized in that: Includes the following steps: Data preprocessing: used to transform the original experimental requirements into a unified task vector representation; Context modeling: used to dynamically encode different task features through the construction of anchor topology and adaptive weighting mechanism; Schedule generation: Used to construct a task list driven by global phase evolution rules; Verification feedback: Used to verify the task list step by step through a hierarchical certificate mechanism and provide feedback for correction.

2. The experimental task scheduling method based on phase adaptive context weight evolution according to claim 1, characterized in that: In data preprocessing, the transformation of the original experimental requirements into a unified task vector representation includes: converting each task into a... The feature vector is represented in dimensional form, with each dimension representing one of the elements corresponding to the task. Continuous features retain their original values ​​or are normalized. Categorical features are processed through one-hot encoding or embedding vectorization. Textual step descriptions are transformed into low-dimensional representations through keyword extraction or predefined mapping tables. When a feature vector of a certain dimension meets specific conditions, constraint labels are added to the metadata of the entire feature vector.

3. The experimental task scheduling method based on phase adaptive context weight evolution according to claim 2, characterized in that: In the context modeling step, the construction of the anchor topology includes: the anchor point is a critical threshold in the task parameters, and a certain buffer is defined around it to determine the potential coupling strength between tasks; all tasks constitute a node set, where each node corresponds to a task, and the task attributes are vectorized; for any two nodes, the global basic similarity is first calculated, and then the global basic similarity is modulated using a directional anchor modulation factor to obtain the edge weight; the edge set is sparsified, and only edges with edge weights greater than the threshold are retained, and the final anchor topology graph contains the node set and the edge set.

4. The experimental task scheduling method based on phase adaptive context weight evolution according to claim 3, characterized in that: In the context modeling step, the adaptive weighting mechanism dynamically encodes different task features, including: establishing a dynamic weighting mechanism for the task set, generating context-related offset weights, i.e., context weights, for each task using attention mapping to characterize the actual priority of the task under the current environment and constraints; wherein the information sources of the context include: the task grouping results of the previous stage, the constraint check status of the hierarchical feasibility certificate feedback, the execution data of the historical list, and the current available resources and remaining capacity status; the entire scheduling process is divided into multiple stages, including the initial grouping stage, the task packaging stage, and the sequence sorting stage, each stage corresponding to a global phase weight, the elements of which reflect the importance of different feature dimensions in the corresponding stage; the comprehensive weight of each task in the corresponding stage is a weighted combination of the context weight and the global phase weight based on a mixed ratio.

5. The experimental task scheduling method based on phase adaptive context weight evolution according to claim 4, characterized in that: In the scheduling generation step, the construction of the task list driven by the global phase evolution rule includes: weight initialization, before entering the scheduling process, firstly calculating the initial context weight vector of each task based on the task set and the anchor topology graph; phase partitioning, dividing the entire scheduling process into multiple stages, including the task grouping stage, the task packaging stage, and the sequence sorting stage; each stage presets a global phase weight vector and a mixing ratio according to different target characteristics to control the fusion ratio between the global phase weight and the context weight; task selection, in each stage, calculating the comprehensive weight vector of all unscheduled tasks and sorting the tasks using a priority queue or heuristic search; then, a seed expansion method is used to gradually form candidate groups: firstly, the task with the highest weight is selected as the group seed, and then based on the edge weights and coupling relationships of the anchor topology graph, multiple strongly related tasks are absorbed to form a local group; During this process, only the remaining tasks that share resources or are strongly coupled with the current group are subject to local weight updates to reduce computational complexity. List updates and context corrections are performed: once a group is committed to the list, the context environment is immediately updated, including the resource occupancy table, anchor buffer status, and task chain progress. For tasks affected by this group, including those sharing resources or located within the same anchor buffer, the context weights are recalculated, and local corrections are triggered if necessary. If the resource occupancy rate or conflict count reaches a set threshold, a global context recalculation is triggered to ensure overall feasibility. Weight evolution and phase switching occur as the scheduling progresses to the next phase, and the comprehensive weights for all remaining tasks are recalculated based on the new global phase weights and mixing ratios. The task selection and list updates / context corrections are repeated during this phase until all tasks are allocated, forming a complete scheduling list.

6. The experimental task scheduling method based on phase adaptive context weight evolution according to claim 5, characterized in that: In the verification feedback step, the layered certificate mechanism includes: a layered feasibility certificate, used to perform multi-dimensional verification of the daily task list to ensure the executability of the list in terms of parameters, resources, timing, and global topology; the feasibility certificate includes: a parameter layer certificate (PFC), used to check whether each task meets the key parameter anchor point constraints; for tasks exceeding the threshold, it records the deviation value or Boolean violation mark for subsequent correction; a resource layer certificate (RFC), used to check the device resource allocation of tasks in the list; it marks conflicting tasks and can provide reallocation suggestions; a timing layer certificate (SFC), used to verify whether the execution order of tasks meets the dependency relationship; for sequences that do not meet the requirement, it inserts interval tasks or adjusts the order to ensure the integrity of the chain; a global layer certificate (GFC), used to verify the global connectivity based on the anchor point topology graph; the global layer certificate is used to integrate the results of each layer to generate an overall executability score for the list.

7. The experimental task scheduling method based on phase adaptive context weight evolution according to claim 6, characterized in that: In the verification feedback step, the step-by-step verification of the task list includes: inputting the candidate task list and anchor point topology diagram; verifying layer by layer, generating parameter layer certificate PFC, resource layer certificate RFC, and time sequence layer certificate SFC in sequence, with each layer independently verifying the corresponding constraints; synthesizing, summarizing the results of each layer to generate a global layer certificate GFC, and calculating the overall compliance score; outputting, if all layer constraints are satisfied, the list is marked as executable; if any layer fails, a local correction mechanism is triggered.

8. The experimental task scheduling method based on phase adaptive context weight evolution according to claim 7, characterized in that: In the verification feedback step, the implementation feedback correction includes: dynamic list correction, used to process tasks that fail the parameter layer certificate PFC, adjust tasks that fail the resource layer certificate RFC, adjust task chains that fail the time-series certificate SFC, and optimize the topology for cases where the global layer certificate GFC fails. The dynamic list correction adopts an iterative approach, with the following process: initial list input, inputting candidate lists; certificate verification, performing four-layer certificate verification (parameter layer certificate PFC, resource layer certificate RFC, time-series certificate SFC, and global layer certificate GFC) on the candidate lists, generating verification feedback; local correction, calling the corresponding correction strategy to locally adjust the task set based on the verification results of each layer; list update, generating a new list version and updating the anchor topology relationship and context weight; iterative judgment, if the new list version passes all layered feasibility certificate layer verifications, the final list is output; if not, the next iteration begins.

9. The experimental task scheduling method based on phase adaptive context weight evolution according to claim 8, characterized in that: In the verification feedback step, the implementation of feedback correction also includes: a convergence mechanism, which, to ensure that the correction process converges within a limited number of rounds, specifically includes: priority convergence control: in multiple iterations, the task context weights are gradually solidified to avoid large fluctuations and ensure that the list sorting and grouping gradually stabilize; local optimum freezing: sub-lists that have passed the hierarchical feasibility certificate verification are marked as "frozen areas" and will not be adjusted in subsequent iterations to prevent repeated modifications from causing oscillations; convergence criterion: if the difference between the lists is less than a set threshold in two consecutive iterations, the lists are considered to have converged, and the final executable version is output.

Images