Method and system for dynamic scheduling of personnel and heterogeneous technical resources based on coupling graph

Abstract

The present application belongs to the technical field of computer resource scheduling, and particularly relates to a personnel and heterogeneous technology resource dynamic scheduling method and system based on a coupling graph, wherein personnel and resource scheduling data are acquired to construct a coupling graph, a scheduling state change is perceived and a change node is located relying on the coupling graph, an affected subgraph is extracted around the change node, an original scheduling decision outside the frozen subgraph is extracted, candidate scheduling schemes are generated in the affected subgraph in multiple ways, a target scheduling scheme is selected after feasibility verification and conflict resolution, the scheme is executed and feedback is collected, the coupling graph and scheduling parameters are incrementally updated using the feedback, and local incremental dynamic scheduling of heterogeneous resources is realized; thus, the problems in the prior art that multi-source heterogeneous states are difficult to model uniformly, personnel and technology resource collaborative constraints are difficult to express simultaneously, full amount rescheduling overhead and scheme disturbance are large under dynamic events, scheduling experience is difficult to continuously deposit and reuse, and the like are solved.

Description

Technical Field

[0001] This invention belongs to the field of computer resource scheduling technology, specifically relating to a method and system for dynamic scheduling of personnel and heterogeneous technical resources based on a coupling graph. Background Technology

[0002] In solution delivery scenarios, FDE (Frontline Deployment Engineer) personnel scheduling solutions are mainly divided into two categories: human experience and spreadsheet-driven scheduling solutions, and fixed rule, single optimization or single recommendation-driven scheduling solutions.

[0003] The drawbacks of manual experience-driven and spreadsheet-based scheduling solutions are as follows: fragmented status information, with customer needs, project progress, FDE personnel skills and working hours, and technical resources such as testing environments and licenses recorded in different business systems, making it difficult to ensure data consistency and timeliness during manual aggregation; insufficient consideration of collaborative constraints, as manual scheduling typically prioritizes personnel selection before supplementing technical resources, making it difficult to simultaneously verify constraints such as skills, working hours, resource capacity, network windows, task precedence, and geographical permissions, easily leading to schedules that cannot be actually executed; slow dynamic response speed, requiring project managers to re-verify multiple projects after the addition of urgent tasks, temporary unavailability of personnel, or failure of critical resources, resulting in a large scope of changes and a high risk of overlooking affected relationships; and difficulty in structured reuse of experience, as personnel matching patterns, resource substitution paths, and conflict resolution processes in historical projects are mostly based on personal experience, making it difficult to form calculable and traceable scheduling knowledge.

[0004] For fixed-rule, single-optimization, or single-recommendation scheduling schemes, the drawbacks are as follows: insufficient scenario adaptability; fixed priority, idle time, or region matching rules cannot fully reflect the real-time changes in demand urgency, technical solution differences, personnel skill levels, and heterogeneous resource availability. Full recalculation causes significant disturbances; single centralized optimization methods re-solve all tasks after any state change, increasing computational overhead and potentially unnecessarily adjusting a large number of unaffected tasks, leading to frequent switching. Recommendation results are not equivalent to executable solutions; single machine learning or case recommendation methods typically only provide personnel or resource rankings, lacking joint constraint verification of personnel, resources, tasks, and time slices, requiring manual re-arrangement of recommendation results. Multiple model outputs are difficult to directly fuse; operations research optimization, strategy models, and knowledge reasoning outputs are mostly discrete allocation schemes or action sequences, and direct numerical weighting easily yields intermediate results that are non-existent or violate constraints.

[0005] In summary, existing FDE scheduling systems generally suffer from the following core problems: a lack of dynamic data structures that can uniformly represent project tasks, FDE personnel, skills, heterogeneous technical resources, time slices, solution components, and risk events; a lack of local incremental rescheduling mechanisms that can accurately locate the affected scope and freeze irrelevant decisions when changing events occur; a lack of a full-chain mechanism for complete scheduling solutions, from multi-channel candidate generation to verification and repair to feasible solution selection; and a lack of a closed-loop mechanism that uses execution deviations, manual intervention, resource failures, and task completion results to update scheduling knowledge and model parameters. Summary of the Invention

[0006] This application provides a dynamic scheduling method and system for personnel and heterogeneous technical resources based on coupled graphs, in order to solve the problems in the prior art such as the difficulty in uniformly modeling multi-source heterogeneous states, the difficulty in simultaneously expressing the collaborative constraints of personnel and technical resources, the large overhead and scheme disturbance of full rescheduling under dynamic events, and the difficulty in continuously accumulating and reusing scheduling experience.

[0007] The first aspect of this application provides a method for dynamic scheduling of personnel and heterogeneous technical resources based on a coupling graph, including: Acquire personnel scheduling data and resource scheduling data; A coupling graph is constructed based on the personnel scheduling data and resource scheduling data. The coupling graph is used to uniformly represent multiple heterogeneous elements and their interrelationships within the scheduling domain as a graph structure with correlation relationships. The state changes within the scheduling domain are perceived based on the coupling graph. When the state change causes an existing scheduling scheme to become unexecutable, the changed node corresponding to the state change in the coupling graph is located. Centered on the changed node, the affected range is dynamically determined by expanding along the path according to the association represented by the coupling graph. The affected subgraph is extracted from the coupling graph, and the existing scheduling decisions outside the affected subgraph are frozen. Within the affected subgraph, candidate scheduling schemes are generated through multiple candidate generation methods. Each candidate scheduling scheme contains multiple allocation units, and each allocation unit associates and binds personnel, tasks, resources, and time at least once. For each candidate scheduling scheme, a feasibility check is performed; for candidate schemes with conflicts, conflict resolution is performed; and the target scheduling scheme is selected from the feasible candidate scheduling schemes after check and conflict resolution. The target scheduling scheme is executed, execution feedback is collected, and the coupling graph and scheduling parameters are incrementally updated based on the execution feedback.

[0008] Preferably, taking the changed node as the center, the affected range is dynamically determined by extending along the path according to the association represented by the coupling graph, the affected subgraph is extracted from the coupling graph, and existing scheduling decisions outside the affected subgraph are frozen, including: Starting from the changed node, determine the initial expansion step number according to the type of the changed node; Starting from the initial expansion step number, expand outward layer by layer along the association relationships in the coupling graph; After each layer is expanded, it is determined whether there is a valid influence relationship between the newly included node and the changed node. If a valid influence relationship exists, the expansion continues; otherwise, the expansion stops. The affected range is defined as the area involved in the final expansion, and the nodes and edges within the affected range are extracted from the coupling graph to form the affected subgraph; A freeze flag is set for existing scheduling decisions associated with nodes in the coupling graph that do not belong to the affected subgraph, so as to keep their original allocation unchanged.

[0009] Preferably, the effective influence relationship includes one or more of the following: There is a personnel or resource sharing relationship between the newly included node and the changed node; There is a task pre- or post-task dependency between the newly included nodes and the changed nodes; There is a time window conflict between the newly included nodes and the changed nodes; There is a resource substitution relationship between the newly included nodes and the changed nodes; There is a capacity competition relationship between the newly added nodes and the changed nodes; There is a hard constraint transitive relationship between the newly included node and the changed node.

[0010] Preferably, the state change renders an existing scheduling scheme unexecutable, including: The changes in task requirements, personnel availability, and resource availability between the current state and the previous state are calculated. The weighted sum of these changes is used as the degree of change. If the degree of change exceeds a preset threshold, it is determined that the state change renders the existing scheduling scheme unexecutable. Alternatively, if the state change results in personnel conflicts, resource capacity exceeding limits, or task time window violations in the existing scheduling scheme, it is determined that the state change renders the existing scheduling scheme unexecutable. In particular, if the state change results in personnel conflicts, resource capacity exceeding limits, or task time window violations in the existing scheduling scheme, it is determined that the state change renders the existing scheduling scheme unexecutable regardless of whether the degree of change exceeds the preset threshold.

[0011] Preferably, the multiple candidate generation methods include at least two of the following methods: Constraint optimization method: Establish the objective function and constraints, solve them, and generate candidate scheduling schemes; Strategy decision-making method: Based on the historical scheduling status and action output, the scheduling action is output, and the action expander generates candidate scheduling schemes; Case-based reasoning: Retrieve successful cases that match the currently affected subgraph from historical scheduling cases, perform variable substitution and time alignment, and generate candidate scheduling schemes.

[0012] Preferably, the conflict resolution includes: Construct a conflict graph, where nodes represent allocation actions that conflict, and edges represent mutual exclusion or resource competition relationships between allocation actions. The repair order is determined based on task priority, availability of alternative resources, switching costs, and remaining time slack. According to the repair order, perform one or more of the following operations on the allocation actions corresponding to the conflict nodes in the conflict graph: personnel replacement, resource replacement, time slice shifting, low-priority task postponement, local re-solution, until the candidate scheduling scheme meets all feasibility conditions.

[0013] Preferably, the target scheduling scheme is selected from the feasible candidate scheduling schemes after verification and conflict resolution, including: Based on the characteristics of the current scheduling scenario and the historical performance of each candidate generation method, an evaluation weight is generated for each feasible candidate scheduling scheme. Calculate the comprehensive score for each feasible candidate scheduling scheme based on the evaluation weights; The candidate scheduling scheme with the highest comprehensive score is determined as the target scheduling scheme.

[0014] Preferably, collecting execution feedback and incrementally updating the coupling graph and scheduling parameters based on the execution feedback includes: One or more of the following data are used as execution feedback: task completion status, actual working hours of personnel, resource utilization rate, number of scheme switching times, and results of manual intervention. The feedback signal is calculated based on the execution feedback, wherein the feedback signal is calculated by weighting at least two of the following: project progress improvement, personnel and resource utilization, delay, switchover cost, and hard constraint violation penalty. Update one or more of the following based on the feedback signal: the weight of the edges between nodes in the coupled graph, the internal parameters of the candidate generation method, the scheduling triggering condition, and the conflict resolution strategy parameters.

[0015] Preferably, before generating candidate scheduling schemes through multiple candidate generation methods, a preliminary risk assessment is also included: Obtain historical failure records, including records of candidate solution generation failure, feasibility verification failure, conflict resolution failure, manual rejection, or rollback. The risk of each candidate generation path is assessed by matching the characteristics of the currently affected subgraph with the historical failure records, and the assessment result is obtained. Candidate paths deemed high-risk by the evaluation results are eliminated, and no complete candidate scheduling schemes are generated for them.

[0016] A second aspect of this application provides a dynamic scheduling system for personnel and heterogeneous technical resources based on a coupling graph, comprising: The acquisition module is used to acquire personnel scheduling data and resource scheduling data; The positioning module is used to construct a coupling graph based on the personnel scheduling data and resource scheduling data. The coupling graph is used to uniformly represent multiple heterogeneous elements and their interrelationships within the scheduling domain as a graph structure with association relationships. Based on the coupling graph, the module senses state changes within the scheduling domain. When the state change causes an existing scheduling scheme to become unexecutable, the module locates the changed node in the coupling graph corresponding to the state change. The extraction module is used to dynamically determine the affected range by expanding along the path based on the association represented by the coupling graph, with the changed node as the center, extracting the affected subgraph from the coupling graph, and freezing the existing scheduling decisions outside the affected subgraph. The generation module is used to generate candidate scheduling schemes through multiple candidate generation methods within the affected subgraph range. Each candidate scheduling scheme contains multiple allocation units, and each allocation unit associates and binds personnel, tasks, resources, and time at least. The determination module is used to perform feasibility verification on each of the candidate scheduling schemes, perform conflict resolution on candidate schemes with conflicts, and select the target scheduling scheme from the feasible candidate scheduling schemes after verification and conflict resolution. The update module is used to execute the target scheduling scheme, collect execution feedback, and incrementally update the coupling graph and scheduling parameters based on the execution feedback.

[0017] Therefore, this application has the following beneficial effects: (1) This invention uses a time-varying coupled graph to uniformly represent skills, capacity, time window, dependency, conflict, substitution and historical success rate as nodes, edges and their attributes, so that scheduling calculation can simultaneously consider whether FDE personnel and technical resources have common execution conditions, avoid the situation of only allocating according to personnel idleness and ignoring test environment, license, cloud resources or network window, and support local incremental updates of new tasks, personnel changes and resource failures, maintain the real-time performance of state expression, and provide a unified data foundation for accurate location of subsequent affected subgraphs and local rescheduling, thereby solving the problems of difficulty in unified modeling of multi-source heterogeneous states and difficulty in simultaneously expressing the collaborative constraints of personnel and technical resources in the prior art; (2) At the same time, based on the coupling graph, the present invention regenerates and adjusts the scheme only for the necessary range by changing node positioning, extracting the affected subgraph and freezing the decision outside the subgraph, reducing the calculation range of full recalculation and reducing irrelevant switching of unaffected items. It can control the scope of influence according to the relationship type, effective time interval and number of hops, and can merge the local scheme with the frozen original scheme after the local scheme is generated to maintain the integrity of the global scheme, providing a limited solution space for subsequent multi-channel candidate generation, thereby solving the problems of full rescheduling overhead and large scheme disturbance under dynamic events in the prior art; (3) In addition, within the affected subgraph, the constraint optimization, strategy decision and case reasoning channels provide structured solution, fast response and historical experience reuse capabilities respectively. Each channel first generates complete candidate solutions, then performs unified hard constraint verification and conflict repair, and finally selects feasible candidates by scoring. This avoids directly weighting the results of discrete personnel or resource allocation to produce unexecutable solutions, and provides an executable scheduling scheme for subsequent execution feedback. This solves the problem in the prior art that the recommended results are not equal to the executable solutions and that the outputs of multiple models are difficult to directly fuse. (4) This application also collects feedback after execution and transforms negative feedback such as incomplete candidate generation, hard constraint verification failure, conflict repair failure, manual rejection and execution rollback into reusable historical failure modes. In the new round of scheduling, risk prediction is made for candidate generation paths, and high-risk candidate paths exceeding the pruning threshold are stopped in advance. This reduces the computational overhead of combination search, scheme completion, normalization coding and conflict repair generated by invalid candidates. The updated coupling graph and scheduling parameters are fed back to the next round of scheduling, solving the problem that scheduling experience is difficult to continuously accumulate and reuse in the prior art.

[0018] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0019] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a dynamic scheduling method for personnel and heterogeneous technical resources based on a coupling graph, according to an embodiment of this application. Figure 2 This is a schematic diagram of incremental rescheduling of the affected subgraph triggered by an event, according to an embodiment of this application. Figure 3 This is a schematic diagram illustrating historical feedback-driven candidate prediction, multi-channel generation, and adaptive selection based on embodiments of this application. Figure 4 This is a flowchart of a method for dynamic scheduling of personnel and heterogeneous technical resources based on a coupling graph, according to an embodiment of this application. Figure 5 This is a schematic diagram of the resource-task coupling graph structure provided according to an embodiment of this application; Figure 6 This is a schematic diagram of the closed loop of scheduling execution feedback and continuous knowledge update provided according to the embodiments of this application; Figure 7 This is a structural diagram of a dynamic scheduling system for personnel and heterogeneous technical resources based on a coupling graph, according to an embodiment of this application. Figure 8 This is a structural diagram of a dynamic scheduling system for personnel and heterogeneous technical resources based on a coupling graph, according to an embodiment of this application. Detailed Implementation

[0020] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0021] The following description, with reference to the accompanying drawings, illustrates a method and system for dynamic scheduling of personnel and heterogeneous technical resources based on coupled graphs, according to embodiments of this application. Addressing the problem mentioned in the background art that it is difficult to simultaneously express the collaborative constraints of personnel and technical resources, this application provides a method for dynamic scheduling of personnel and heterogeneous technical resources based on coupled graphs. In this method, time-varying coupled graphs are used to graphically represent skills, capacity, time windows, dependency conflicts, substitution relationships, and historical success rates, taking into account the collaborative execution conditions of personnel and technical resources, supporting local incremental updates, and solving the problems of difficult unified modeling of multi-source heterogeneous states and the inability to synchronously express the collaborative constraints of personnel and technical resources. Based on the coupled graph, local incremental rescheduling is achieved through changing node location, affected subgraph extraction, and external decision freezing, controlling... The system expands its impact scope and integrates local and original solutions to reduce the solution space, overcoming the drawbacks of high overhead and severe perturbation of full rescheduling under dynamic events. Within the affected subgraph, it generates complete candidate solutions through constraint optimization, strategy decision-making, and case reasoning via multiple channels. After unified hard constraint verification and conflict resolution, the optimal solution is selected to obtain an executable scheduling plan, solving the problems of difficulty in merging multiple model outputs and non-executable output results. Simultaneously, it collects various scheduling negative feedbacks to construct historical failure modes, performs risk prediction and pruning on candidate paths to reduce invalid computations, and iteratively feeds the coupled graph and scheduling parameters back into the scheduling process, enabling the accumulation and reuse of scheduling experience. This solves the problems in existing technologies such as difficulty in unified modeling of multi-source heterogeneous states, difficulty in simultaneously expressing the collaborative constraints of personnel and technical resources, high overhead and perturbation of full rescheduling under dynamic events, and difficulty in continuously accumulating and reusing scheduling experience.

[0022] Specifically, Figure 1 This is a flowchart illustrating a dynamic scheduling method for personnel and heterogeneous technical resources based on a coupling graph, provided in an embodiment of this application.

[0023] like Figure 1 As shown, this dynamic scheduling method for personnel and heterogeneous technical resources based on a coupling graph includes the following steps: In step S101, personnel scheduling data and resource scheduling data are obtained.

[0024] Personnel scheduling data refers to a multi-dimensional set of information related to scheduling personnel, including at least: FDE personnel identifier, skill tag, skill level, region, available working hours, current task allocation, and historical delivery results. Personnel scheduling data provides basic information for the construction of personnel nodes in the coupled graph and provides data support for the verification of skill matching constraints and working hour capacity constraints. Resource scheduling data refers to a multi-dimensional set of information related to heterogeneous technical resources. Heterogeneous technical resources include computing devices, testing environments, software licenses, cloud resource quotas, network maintenance windows, special tools, and specialized technical support capabilities. Resource scheduling data includes at least: technical resource identifier, resource type, capacity, availability time, location, substitution relationship, and fault status. Resource scheduling data provides basic information for the construction of resource nodes in the coupling graph and provides data support for the verification of resource capacity constraints and mutually exclusive resource constraints.

[0025] In step S102, a coupling graph is constructed based on personnel scheduling data and resource scheduling data. The coupling graph is used to uniformly represent multiple heterogeneous elements and their interrelationships within the scheduling domain as a graph structure with related relationships. The state changes within the scheduling domain are perceived based on the coupling graph. When a state change causes an existing scheduling scheme to become unexecutable, the changed node corresponding to the state change in the coupling graph is located.

[0026] The scheduling domain refers to the entire business scope and object set that requires collaborative scheduling of personnel and heterogeneous technical resources. The scheduling domain includes at least: FDE personnel set, project task set, skill set, heterogeneous technical resource set, time slice set, solution component set, and risk event set. State change refers to any change that occurs in any element within the scheduling domain, including but not limited to: adding tasks, changing task requirements, changing personnel availability, updating personnel skills or working hours, changing technical resource capacity, resource failure, time window adjustment, occurrence of risk events, execution feedback, and manual modification instructions. Change node refers to the node in the coupling diagram that corresponds to the state change within the scheduling domain.

[0027] It is understood that, by constructing a coupling graph, this application unifies the representation of multiple heterogeneous elements and their interrelationships within the scheduling domain into a graph structure with associated relationships. This provides a unified data foundation and traceable associated paths for subsequent state perception, change node location, and affected subgraph extraction. This allows any change in any type of element within the scheduling domain to be tracked and located through the graph structure. At the same time, the associated relationships represented by the graph structure are used to determine whether a state change disrupts the feasibility of an existing scheduling scheme. This enables accurate location of changed nodes and structured determination of scheduling trigger conditions, solving the problems of dispersed states of heterogeneous scheduling elements and difficulty in uniformly capturing and identifying dynamic events.

[0028] Specifically, a time-varying resource-task coupling graph is constructed. , is represented as: ; in, Indicates time The set of nodes includes at least project task nodes, FDE personnel nodes, skill nodes, technical resource nodes, and time slice nodes, and may further include solution component nodes, risk event nodes, regional nodes, and organizational nodes. It represents a set of relationships between nodes, including at least the following relationships: "requires skills", "possesses skills", "assigned to tasks", "occupies resources", "depends on resources", "planned in time slices", "preceded", "conflicts with", "substitutes", and "affected by events". It represents the set of node attributes and edge attributes, including capacity, matching degree, confidence, valid time interval, substitution priority, and historical success rate.

[0029] The system senses state changes within the scheduling domain based on the coupling graph, and generates the current state vector based on standardized data and the resource-task coupling graph. : ; in, This indicates the task requirements, priority, and schedule characteristics. This indicates FDE skills, load, and availability characteristics. This indicates the capacity and availability of heterogeneous technology resources. Indicates time window and region constraints. Indicates the characteristics of events and risks. This represents the structural features calculated from the coupling diagram.

[0030] In this embodiment of the application, the state change causing the existing scheduling scheme to become unexecutable includes: calculating the changes in task requirements, personnel availability, and resource availability between the current state and the previous state, using the weighted sum of the changes as the degree of change, and determining that the state change causes the existing scheduling scheme to become unexecutable when the degree of change exceeds a preset threshold; or determining that the state change causes the existing scheduling scheme to become unexecutable when the state change causes personnel conflicts, resource capacity exceeding limits, or task time window violations in the existing scheduling scheme.

[0031] Specifically, when a state change causes personnel conflicts, resource capacity exceeding limits, or task time window violations in an existing scheduling scheme, regardless of whether the degree of change exceeds a preset threshold, it is determined that the state change has rendered the existing scheduling scheme unexecutable.

[0032] Among them, change refers to the degree of difference of each element in the scheduling domain in a specific dimension between the current time and the previous time; the degree of change is a comprehensive indicator used to quantify the degree of difference between the current state and the state at the previous time; the preset threshold is a preset numerical limit used to judge whether the state change has reached the level of triggering scheduling, which can be preset based on historical scheduling logs and business experience, and can be dynamically adjusted based on execution feedback; personnel conflict refers to the situation where the same FDE personnel are assigned to two or more tasks in the same time slice; resource capacity overrun refers to the situation where the task demand of a certain technical resource in the same time slice exceeds its available capacity, where technical resources include computing equipment, testing environment, software license, cloud resource quota, network maintenance window, special tools and special technical support capabilities; task time window violation refers to the situation where the planned execution time of a task conflicts with its allowed time window, such as: the planned start time of the task is earlier than its earliest start time, the planned completion time of the task is later than its latest completion time, the planned execution interval of the task and the time window do not intersect, etc.

[0033] It is understood that the embodiments of this application provide two ways to determine that an existing scheduling scheme is not executable. On the one hand, the degree of change is obtained by calculating the changes in task requirements, personnel availability, and resource availability and weighting them. When the degree of change exceeds a preset threshold, a determination is triggered. On the other hand, the determination is completed by directly identifying constraint violations such as personnel conflicts, resource capacity exceeding limits, and task time window violations that occur within the scheme. This achieves a deterministic judgment that the scheme is not executable. The two paths together ensure the comprehensiveness and accuracy of state change triggering and effectively avoid missed or false triggering caused by relying on only a single rule.

[0034] Specifically, the degree of change is calculated by comparing the demand, personnel and resource availability, risks, and current constraints at adjacent time points: ; in, This is the normalized task requirement vector. This is the normalized task requirement vector from the previous time step. For personnel and resource availability vectors, This is the normalized personnel and resource availability vector from the previous time step. As for the intensity of the risk event, The current scheme's constraint violation indicators, to The weights are respectively assigned to changes in task requirements, changes in personnel and resource availability, the intensity of risk events, and the degree of constraint violation, to satisfy... and .

[0035] The weights are determined based on the degree of impact of various state changes on historical scheduling results. This application embodiment utilizes historical scheduling logs to construct calibration samples. These samples include at least the amount of state change, whether rescheduling was actually triggered, task delays, resource conflicts, results of manual intervention, and the cost of switching schemes. The calibration objectives are to improve the effective trigger recognition rate, reduce the number of ineffective reschedulings, and minimize scheduling disturbances. Weight combinations are determined using grid search, heuristic search, Bayesian optimization, or parameter optimization methods. In the initial stage where historical samples are lacking, each weight can be set to the same value or preset based on business experience. As execution feedback accumulates, the weights are recalibrated according to a preset period or rolling time window.

[0036] When the periodic scheduling point is reached, Scheduling is initiated when any of the following conditions are met: exceeding the preset trigger threshold, a hard constraint violation occurs, or a manual trigger command is received.

[0037] It should be noted that a hard constraint violation refers to a situation where the scheduling plan violates mandatory constraints that must be met unconditionally. Hard constraints include at least the following: personnel cannot be assigned to two or more tasks within the same time slice; the total demand for technical resources occupied within any time slice cannot exceed its available capacity; and the planned execution time of tasks must be within the allowed time window. Hard constraint violations will directly render the scheduling plan unexecutable. Hard constraint violations trigger hard-trigger decision paths and are not limited by preset thresholds; that is, regardless of the magnitude of the change, scheduling is triggered as soon as a hard constraint violation occurs.

[0038] In step S103, with the changed node as the center, the affected range is dynamically determined by expanding along the path according to the association represented by the coupling graph. The affected subgraph is extracted from the coupling graph, and the existing scheduling decisions outside the affected subgraph are frozen.

[0039] It is understood that the embodiments of this application dynamically determine the affected range by expanding outward layer by layer along the association relationship represented by the coupling graph with the changed node as the center, and extract the affected subgraph, so that the boundary of the affected range matches the actual impact propagation situation, avoiding the problem of the range being too large or too small due to fixed range expansion; at the same time, by freezing the existing scheduling decisions outside the affected subgraph, the unaffected scheduling decisions are kept unchanged, transforming the global rescheduling problem into a local incremental rescheduling problem, reducing the computational overhead and scheme disturbance while ensuring the integrity of the scheme.

[0040] like Figure 2As shown, the change event 31 can be an emergency task, personnel unavailability, resource failure, or requirement change. The change event is input to the location change node 32, which locates the change node in the coupling graph and calculates the degree of change and associated edges. Then, by extracting the h-hop neighborhood 33, the h-hop neighborhood is extracted with the change node as the center to form the affected subgraph. In the original global coupling graph and the original scheduling scheme, the nodes inside the dashed box constitute the affected subgraph, and the nodes outside the dashed box are nodes that are not affected by the event. The original allocation decision 34 and historical allocation results of the unaffected nodes are frozen, and only the solution variables inside the affected subgraph are released. Local candidate schemes 35 are generated, repaired, and selected on the affected subgraph. Candidate generation, constraint repair, and optimal scheme selection are completed only within the affected subgraph. Finally, by merging the local schemes with the frozen part 36, the local solution results are merged with the original allocation decisions that are frozen and retained, and the updated global scheduling scheme is output.

[0041] In this embodiment, the affected area is dynamically determined by expanding along the path based on the association relationships represented by the coupling graph, with the changed node as the center. The affected subgraph is extracted from the coupling graph, and existing scheduling decisions outside the affected subgraph are frozen. This includes: starting from the changed node, determining the initial expansion step number based on the type of the changed node; starting from the initial expansion step number, expanding outward layer by layer along the association relationships in the coupling graph; after each expansion layer, determining whether there is an effective influence relationship between the newly included node and the changed node. If there is an effective influence relationship, the expansion continues; otherwise, the expansion stops; the area involved in the final expansion is determined as the affected area, and the nodes and edges within the affected area are extracted from the coupling graph to form the affected subgraph; and existing scheduling decisions associated with nodes in the coupling graph that do not belong to the affected subgraph are set with a freeze flag to keep their original allocation unchanged.

[0042] The initial expansion step count refers to the initial expansion layer when expanding the path starting from the changed node. When the changed node is a project task node, the initial expansion step count corresponds to the task pre- / post-task relationship layer and resource occupancy relationship layer directly associated with it; when the changed node is an FDE personnel node, the initial expansion step count corresponds to the task allocation relationship layer and skill relationship layer directly associated with it; when the changed node is a technical resource node, the initial expansion step count corresponds to the resource occupancy relationship layer and substitution relationship layer directly associated with it; when the changed node is a risk event node, the initial expansion step count corresponds to the event impact relationship layer directly associated with it. Effective influence relationship refers to the substantial association between a newly included node and a changed node that may lead to scheduling conflict or inoperability. Expansion step count refers to the number of edges traversed along the association from the changed node. Freeze flag is a protective mark set on the existing scheduling decisions associated with nodes in the coupling graph that do not belong to the affected subgraph.

[0043] It is understood that the embodiments of this application set the initial expansion step number according to the type of the changed node, traverse and expand outward layer by layer along the coupling graph association relationship, and verify the effective influence relationship between the newly added node and the changed node at each layer, thereby dynamically terminating the expansion process, accurately determining the affected range and generating the affected subgraph, and setting a freeze flag for scheduling decisions outside the subgraph to maintain the original allocation result. It can adaptively control the diffusion boundary according to the actual influence relationship between elements, avoid excessive expansion of the affected range or incomplete truncation, and further compress the solution scale while ensuring that the rescheduling covers all related elements, suppressing the modification of irrelevant scheduling objects, and reducing the computational burden and scheme disturbance of rescheduling.

[0044] It should be noted that an effective influence relationship includes one or more of the following: There are personnel or resource sharing relationships between the newly included nodes and the changed nodes; There is a task dependency relationship between the newly added nodes and the changed nodes; There is a time window conflict between the newly included nodes and the changed nodes; There is a resource substitution relationship between the newly added nodes and the changed nodes; There is a capacity competition relationship between newly added nodes and changed nodes; There is a hard constraint propagation relationship between the newly included nodes and the changed nodes.

[0045] Among these, task pre- and post-dependent relationships refer to the execution order constraints between the tasks associated with the newly added node and the tasks associated with the changed node, meaning the execution of one task requires the completion of the other. Time window conflict relationships refer to the time window conflicts between the tasks associated with the newly added node and the tasks associated with the changed node, meaning the planned execution times of the two tasks overlap or exceed the allowed time range, making it impossible to simultaneously meet their respective time constraints. Capacity competition relationships refer to the resource capacity competition between the tasks associated with the newly added node and the tasks associated with the changed node, meaning multiple tasks simultaneously require the same resource capacity, and when resource capacity is insufficient, capacity competition occurs between related tasks. Hard constraint transitivity relationships refer to the existence of a path formed by multiple hard constraint edges between the newly added node and the changed node, allowing the influence of the changed node to be passed down layer by layer to the original node through hard constraint relationships. The existence of hard constraint transitivity relationships indicates that the node is indirectly affected by the changed node and needs to be included in the affected scope.

[0046] Specifically, the first step is to locate the changed tasks, personnel, technical resources, or risk events, and define these as the set of changed nodes for this round of scheduling. The number of hops used for extracting the affected subgraph is then determined. Instead of being set to a fixed global value, the initial hop count is dynamically determined based on the changing node type, the impact level of the event, and the propagation range of constraints. First, the initial hop count is determined according to a preset node type-initial hop count mapping rule. When the change node is a project task node, it expands along the relationships of "required skills", "preceded", "planned", and "occupied resources"; when the change node is an FDE personnel node, it expands along the relationships of "possessing skills", "assigned to tasks", "planned", and "conflicts with"; when the change node is a technical resource node, it expands along the relationships of "occupied resources", "dependent on resources", "substitutable", and "conflicts with"; when the change node is a risk event node, it expands along the relationships of "affected by the event", "conflicts with", and "dependent on resources".

[0047] From the initial jump number The process begins by expanding the neighborhood layer by layer, and after each layer is completed, it is determined whether the newly added boundary node still has a valid influence relationship with the changed node. If the newly added boundary node and the affected task have relationships such as personnel or resource sharing, task prioritization, time window conflict, resource substitution, capacity contention, or hard constraint propagation, the hop count is increased by one layer and expansion continues. If the newly added boundary node only has a weak relationship below a preset association threshold, or if further expansion does not introduce new affected tasks, personnel, technical resources, or time slices, expansion stops. The hop count... No more than the preset maximum number of jumps This limits the size of the affected subgraph and the computational scope of local rescheduling.

[0048] Based on the final determined number of jumps The allowed relationship types and the time validity intervals of edges form the affected subgraph. A freeze flag is set for nodes not belonging to the affected subgraph and their original allocation decisions. Only the necessary decision variables within the affected subgraph and those connected to its boundary nodes are opened, thus transforming the global rescheduling problem into a local incremental rescheduling problem.

[0049] In step S104, within the affected subgraph, candidate scheduling schemes are generated through multiple candidate generation methods. Each candidate scheduling scheme contains multiple allocation units, and each allocation unit associates and binds personnel, tasks, resources, and time.

[0050] The candidate scheduling scheme is a complete set of schemes composed of multiple allocation units, covering all tasks to be scheduled in the affected subgraph, and specifying the executor, required resources and execution time for each task.

[0051] It is understood that the embodiments of this application are limited to the local solution scope of the affected subgraph, and multiple candidate generation methods are used to output candidate scheduling schemes in parallel. The allocation unit is used as the smallest scheduling carrier to realize the association and binding of personnel, tasks, resources and time, and output complete and feasible allocation combinations, rather than discrete single allocation results. This not only makes full use of the solution space of local rescheduling, but also integrates the advantages of different generation methods, providing multiple complete candidate samples for subsequent feasibility verification, conflict resolution and scheme selection.

[0052] In the embodiments of this application, the multiple candidate generation methods include at least two of the following methods: constraint optimization method: establish an objective function and constraint conditions for solving, and generate candidate scheduling schemes; strategy decision method: output scheduling actions based on historical scheduling states and actions, and generate candidate scheduling schemes by the action expander; case reasoning method: retrieve successful cases that match the current affected subgraph from historical scheduling cases, perform variable replacement and time alignment, and generate candidate scheduling schemes.

[0053] Among them, the constrained optimization method refers to the candidate generation method of solving the scheduling problem by establishing a mathematical programming model. Specifically, using project tasks, FDE personnel, technical resources, and time slices as indices, personnel allocation variables and resource allocation variables are defined, an optimization model containing objective functions and constraints is established, and the model is solved by a solver or optimization algorithm to generate candidate scheduling schemes that satisfy all constraints. The strategy decision-making method refers to the method of generating candidate scheduling actions output by the strategy model. Specifically, the strategy model outputs the probability or ranking of actions such as personnel allocation, resource replacement, and time window shifting based on historical scheduling status and action execution results. Then, the action expander combines these actions into complete candidate scheduling schemes. Case-based reasoning refers to generating candidate scheduling schemes by retrieving and reusing historical scheduling cases. Specifically, successful cases similar to the current affected subgraph structure are retrieved from the historical scheduling case library. Personnel identifiers, resource identifiers, and time parameters in the historical schemes are replaced with currently available personnel, resources, and time slices, and candidate scheduling schemes are generated based on the mapping relationship of task nodes.

[0054] It is understood that the embodiments of this application integrate at least two candidate generation methods: constraint optimization, strategy decision-making, and case reasoning. Constraint optimization relies on the objective function and constraint conditions to achieve an accurate solution. Strategy decision-making relies on the output of scheduling actions based on historical scheduling states and then expands to obtain a solution. Case reasoning generates candidate solutions by retrieving and matching historical successful cases, and generating them through variable substitution and time alignment. Different generation methods leverage their respective advantages of accurate solution, fast output, and reuse of historical experience, producing candidate scheduling solutions through multiple paths, enriching the diversity of candidate samples, avoiding the limitations of a single algorithm, and providing a sufficient basis for subsequent feasibility verification and optimal solution selection.

[0055] Specifically, each channel outputs a complete candidate solution consisting of a personnel-task-resource-time slice quadruple, rather than just a recommendation for a single personnel or resource.

[0056] Constraint optimization channel based on project tasks FDE personnel Technical resources and time slices Define personnel assignment variables for the index. Resource allocation variables Task delay slack variables and personnel overload slack variables An example objective function is: ; in, This represents the switching disturbance relative to the original scheme. Indicates the cost of consuming technical resources. Indicates the benefits of skill matching. This indicates that high-priority tasks satisfy the reward. to Weights that are configurable or can be calibrated by historical feedback.

[0057] Skills matching between personnel and tasks Determined by skill vector similarity and graph path association: ; in, For FDE personnel Skill vectors, For the task The skill demand vector This represents the shortest effective path length between personnel nodes and the required solution components in the coupling graph. and For parameters.

[0058] Constraints should include at least the following: personnel manpower capacity, minimum skill matching, task requirement fulfillment, technical resource capacity, simultaneous availability of personnel and resources, time window, task pre- and post-task, geographic or security permissions, mutually exclusive resources, and freeze decision retention constraints. An example is shown below: ; in, This represents the time requirement of task i for FDE personnel j in time slice τ. Let τ be the upper limit of available working hours for FDE personnel j in time slice τ. This represents the minimum skill matching threshold for task i. This represents the time requirement of task i in time slice τ. This represents a 0-1 decision variable indicating whether FDE personnel j is assigned to task i in time slice τ. This represents the value of decision variable u in the previous scheduling scheme. A 0-1 indicator parameter indicating whether task i requires technical resource k. This represents the affected subgraph extracted from the coupled graph at time t.

[0059] The strategy decision-making channel outputs the probability or ranking of actions such as personnel allocation, resource replacement, and time window shifting based on historical scheduling status, actions, and execution results. The action expander then combines these with constraint templates to generate complete candidate solutions. The graph reasoning or case study channel retrieves historical subgraphs and successful solutions with structures similar to the currently affected subgraph from the coupled graph. It then performs variable substitution and time alignment based on current personnel, resource, and time conditions to form knowledge-driven candidate solutions.

[0060] To standardize the output format across different channels, each candidate scheduling scheme is converted into a set of candidate schemes consisting of several personnel-task-resource-time slice quadruples. Each quadruple is represented as: ; in, Indicates the assigned FDE personnel identifier. Indicates the project task identifier. This represents the technical resource identifier assigned to this task. A standardized time slice representing the execution of a task, including start and end times.

[0061] Each quadruple is also associated with the candidate scheme identifier, generation channel identifier, original output identifier, and sub-configuration confidence level to preserve its generation source and transformation basis.

[0062] For the constrained optimization channel, based on the personnel allocation variables, resource allocation variables, and time variables that are in an effective allocation state in the solution results, and using the task identifier as the association key, the personnel, technical resources, and execution time corresponding to the same task are combined into a quadruple. For the strategy decision channel, actions such as personnel allocation, resource replacement, task postponement, and time window shifting are first applied to the current scheduling scheme. Then, the action expander completes the task, personnel, resource, and time slice information corresponding to the actions based on the resource-task coupling graph and converts it into a quadruple. For the graph reasoning or case channel, the personnel identifier, resource identifier, and time parameters in the historical scheme are replaced with the available personnel, available resources, and executable time slices in the current affected subgraph, and quadruples are generated based on the mapping relationship of task nodes.

[0063] When a task requires multiple technical resources simultaneously, it is split into multiple quadruples with the same personnel identifier, task identifier, and time slice, but different resource identifiers, according to the rule of "one resource corresponds to one quadruple". These quadruples are then associated with the same allocation group identifier. When a task does not require independent technical resources, the resource identifier is set to a preset empty resource identifier. The quadruples are deduplicated using a composite key consisting of "personnel identifier, task identifier, resource identifier, start time, and end time". If there are records for the same task with inconsistent personnel, resources, or time arrangements, they are retained as different candidate allocations, and their conflict is determined during subsequent feasibility checks.

[0064] In this embodiment of the application, before generating candidate scheduling schemes through various candidate generation methods, a preliminary risk assessment is also included: obtaining historical failure records, which include records of candidate scheme generation failure, feasibility verification failure, conflict resolution failure, manual rejection, or execution rollback; matching the characteristics of the currently affected subgraph with the historical failure records to assess the risk of each candidate generation path and obtain the assessment result; and eliminating candidate generation paths with high risk assessment results, without generating complete candidate scheduling schemes for them.

[0065] Among them, candidate solution generation failure refers to the situation where the candidate solution cannot cover all the scheduled tasks in the affected subgraph during the generation process, or cannot supplement the required personnel, technical resources and execution time for each task, resulting in the inability to form a complete candidate solution; feasibility verification failure refers to the situation where the candidate solution fails to pass the hard constraint checks such as personnel unavailability, resource capacity exceeding the limit, time window conflict, violation of task precedence relationship or change of freeze decision in hard constraint verification; conflict resolution failure refers to the situation where, after performing repair operations such as personnel replacement, resource replacement, time slice shifting, low priority task postponement or partial re-solution for candidate solutions with conflicts, it is still impossible for the candidate solution to meet all hard constraints; being manually rejected refers to the situation where the candidate solution is explicitly rejected or returned for regeneration by the scheduler when output for manual selection; execution rollback refers to the situation where, during the execution of the candidate solution, due to reasons such as interface failure, inconsistency between the actual state of resources and the plan, or unconfirmed key tasks, the relevant allocation action is canceled and rolled back to the original solution or the most recently feasible state.

[0066] It is understood that the embodiments of this application, by adding a pre-risk prediction step before generating a complete candidate scheduling scheme, transform historical failure records such as candidate scheme generation failure, feasibility verification failure, conflict resolution failure, manual rejection, or execution rollback into reusable structured knowledge, forming a set of historical failure modes. Based on the characteristics of the currently affected subgraph, the historical failure modes are matched, and high-risk candidate generation paths are identified and eliminated in advance before the complete scheme is expanded and accurately verified. This ensures that high-risk paths no longer undergo subsequent calculation processes such as combination expansion, scheme completion, normalization coding, and conflict repair, thereby reducing the computational overhead caused by obviously invalid candidates. At the same time, the retained low-risk paths still undergo subsequent deterministic hard constraint verification and conflict repair, ensuring the executability of the final scheme and achieving an effective balance between computational efficiency and scheme quality.

[0067] Specifically, regarding the candidate generation path Calculate and predict risks: ; in, Indicates the candidate generation path Risks of incompleteness or infeasibility This indicates the task, personnel, technical resources, time window, and graph structure features corresponding to the current candidate generation path. This represents the set of historical negative feedback records, consisting of incomplete candidate records, hard constraint verification failures, repair failures, manual rejections, and execution rollback records. This represents a risk prediction function based on rules, statistical models, classification models, case similarity models, or a combination thereof. This indicates the corresponding parameter.

[0068] When satisfied When the candidate generation path is deemed to have a high risk of failure, subsequent personnel-task-resource-time slice combinations and complete solution generation for that path are halted; when If this happens, the candidate path is allowed to continue into the process of generating a complete candidate solution.

[0069] in, This serves as the candidate pruning threshold and is dynamically updated based on subsequent actual verification results and execution feedback. The preliminary judgment is only used to eliminate high-risk candidate generation paths in advance and does not replace subsequent deterministic hard constraint verification.

[0070] It should be noted that candidate generation paths retained after preliminary risk assessment and pruning will continue to generate complete solutions. A channel output is only considered a complete candidate solution if it covers all scheduled tasks in the affected subgraph and completes the required personnel, technical resources, and execution time for each task. Outputs that were not pruned during the preliminary assessment stage but still cannot complete the necessary fields or cover all scheduled tasks during the actual generation process will not be included in the subsequent candidate solution set.

[0071] In step S105, a feasibility check is performed on each candidate scheduling scheme, conflict resolution is performed on candidate schemes with conflicts, and the target scheduling scheme is selected from the feasible candidate scheduling schemes after check and conflict resolution.

[0072] It is understood that the embodiments of this application uniformly conduct feasibility verification on each candidate scheduling scheme, perform conflict resolution processing on candidate schemes that are found to have conflicts, and select the best target scheduling scheme from all candidate schemes that have been guaranteed to be feasible, thereby avoiding the direct issuance and execution of unverified candidate schemes, eliminating internal constraint contradictions of the schemes, ensuring that the final output target scheduling scheme meets various hard scheduling constraints, and improving the executability and reliability of the scheduling results.

[0073] It should be noted that the final solution is converted into an executable personnel task list, technical resource allocation list, and time window plan. Each allocation record must include at least the project task identifier, FDE identifier, skill matching basis, resource identifier, planned time slice, pre- and post-dependencies, differences from the original solution, and execution status. The project management platform, resource management platform, and notification system are updated through standard interfaces. When the execution interface fails, the actual resource status is inconsistent with the plan, or critical tasks are not confirmed, the relevant actions are rolled back to the original solution or the most recent feasible state, and the failed actions, related tasks, and resource nodes are marked as new changed nodes, re-triggering the extraction of affected subgraphs and local rescheduling.

[0074] In this embodiment, conflict resolution includes: constructing a conflict graph, where nodes represent conflicting allocation actions and edges represent mutual exclusion or resource contention relationships between allocation actions; determining a repair order based on task priority, availability of alternative resources, switching cost, and remaining time slack; and performing one or more of the following operations sequentially on the allocation actions corresponding to conflicting nodes in the conflict graph according to the repair order: personnel replacement, resource replacement, time slice shifting, postponement of low-priority tasks, and local re-solution, until the candidate scheduling scheme satisfies all feasibility conditions.

[0075] Among them, the conflict diagram is a graph structure used to characterize the conflict relationships between allocation actions in candidate scheduling schemes. Mutual exclusion refers to the relationship that two allocation actions cannot coexist in the same time slice because they involve the same personnel or the same resources. Specifically, this includes: the same FDE personnel being assigned to two or more tasks in the same time slice (personnel mutual exclusion); the same technical resource being assigned to two or more tasks in the same time slice (resource mutual exclusion); resource competition refers to the relationship that multiple allocation actions require the capacity of the same technical resource in the same time slice, and the total demand exceeds the available capacity of that resource. The repair order refers to the sequence in which the assigned actions corresponding to the conflict nodes in the conflict diagram are repaired. The repair order is determined based on the following four dimensions: task priority, availability of alternative resources, switching cost, and remaining time slack. Among them, the remaining time slack refers to the adjustable margin between the current planned execution time of the task and its latest allowed completion time. Personnel replacement refers to the conflict resolution operation of replacing the current FDE personnel in a conflicting allocation action with another FDE personnel with the same or similar skills who are available in the same time slice; resource replacement refers to the conflict resolution operation of replacing the current technical resource in a conflicting allocation action with another technical resource with the same or similar functions that are available in the same time slice; time slice shifting refers to the conflict resolution operation of moving the execution time slice of the task in a conflicting allocation action forward or backward to avoid the conflicting time period of personnel or resources; low-priority task postponement refers to the conflict resolution operation of moving the execution time slice of lower-priority tasks backward to free up critical time slice resources for high-priority tasks; local re-solution refers to the conflict resolution operation of extracting the allocation actions involved in the conflict graph from the current candidate solutions and re-performing local optimization for situations where the conflict cannot be resolved by a single operation such as personnel replacement, resource replacement, time slice shifting, or low-priority task postponement.

[0076] It is understood that the embodiments of this application formally express conflict allocation actions and mutual exclusion and resource competition relationships by constructing a conflict graph. The conflict resolution order is determined by combining task priority, availability of alternative resources, switching cost, and remaining time slack. Conflict nodes are iteratively processed according to this order using methods such as personnel replacement, resource replacement, time slice shifting, delaying low-priority tasks, and local re-solution until the candidate scheduling scheme meets all feasibility conditions. This achieves structured positioning and orderly resolution of conflict problems, avoids secondary conflicts caused by blind adjustments, improves conflict resolution efficiency and scheme repair success rate, and ensures the constraint compliance of candidate scheduling schemes.

[0077] Specifically, when a candidate solution has a remediable conflict, a conflict graph is constructed. Nodes in the conflict graph represent conflicting allocation actions, and edges represent mutual exclusion or resource contention between actions. The remediation order is determined based on task priority, availability of alternative resources, switching costs, and remaining relaxation time. Personnel replacement, resource replacement, time slice shifting, postponement of low-priority tasks, or partial re-solution are executed sequentially until all hard constraints are satisfied; irremediable candidates are removed from the candidate set.

[0078] In this embodiment of the application, the selection of the target scheduling scheme from the feasible candidate scheduling schemes after verification and conflict resolution includes: generating an evaluation weight for each feasible candidate scheduling scheme based on the characteristics of the current scheduling scenario and the historical performance of each candidate generation method; calculating the comprehensive score of each feasible candidate scheduling scheme based on the evaluation weight; and determining the candidate scheduling scheme with the highest comprehensive score as the target scheduling scheme.

[0079] It is understood that the embodiments of this application combine the characteristics of the current scheduling scenario and the historical performance of various candidate generation methods to configure evaluation weights for feasible candidate scheduling schemes, calculate the comprehensive score of each candidate scheme based on the evaluation weights, and select the scheme with the highest comprehensive score as the target scheduling scheme; thereby realizing the quantitative evaluation of multi-channel output candidate schemes, the weights can be adapted to different scheduling scenarios and refer to the past effects of each generation method, avoiding subjective selection, and making the selected target scheduling scheme more in line with the actual needs of the current business.

[0080] This invention does not directly perform vector-weighted averaging on discrete personnel or resource allocation results. Instead, it first converts the outputs from different channels into complete candidate solutions and performs feasibility repair before evaluating each complete feasible solution. This is based on demand determinism. rate of change Knowledge sufficiency Remaining calculation time limit and historical channel reliability Generate evaluation weights : ; in, Let m be the bias term for the m-th candidate generation method. Let r be the bias term for the r-th candidate generation method. This refers to the Softmax temperature parameter. Let be the learnable weight vector for the m-th candidate generation method. Let r be the learnable weight vector of the r-th candidate generation method; For feasible candidates Calculate the overall score: ; in, This represents a comprehensive objective encompassing task completion rate, priority satisfaction, and resource utilization. This indicates robustness to personnel absence, resource fluctuations, and schedule changes. This indicates consistency with historical success stories and graph relationships. Indicates execution risk, Let X be the switching disturbance cost of scheme X relative to the original scheme. The final scheme is: ; When the difference between the highest-scoring candidate and the second-highest-scoring candidate is less than a preset confidence threshold, or when a candidate contains a high-risk change that requires manual confirmation, the candidate ranking, conflict description, and differences from the original plan are output for manual selection; otherwise, the highest-scoring candidate is automatically output.

[0081] like Figure 3 As shown, the current input state and the affected subgraph are used as inputs, and are respectively fed into the constraint optimization channel 401, the strategy decision channel 402, and the graph reasoning / case channel 403; the constraint optimization channel 401 outputs a candidate set, the strategy decision channel 402 outputs scheduling actions and expands them to form a candidate set, and the graph reasoning / case channel 403 outputs a candidate set; the three candidate sets are fed into the candidate merging and deduplication 503 to obtain the union of the candidate sets C; the merged candidate set is input into the hard constraint verification and repair 500, and after constraint verification and conflict resolution, a feasible candidate set is obtained; the scene weight calculation 601 calculates the scene evaluation weight based on scene determinism, dynamism, knowledge sufficiency, and task time limit; the candidate scoring and selection 602 uses the scoring function Q(X) to score the feasible candidates, according to... The candidate with the highest comprehensive score is selected as the target scheduling scheme. Candidate schemes are generated through multi-path generation, and after merging, deduplication, and constraint repair, quantitative selection is completed based on the scenario dynamic weights. This takes into account the solution quality, reuse of historical experience, and ensures the feasibility and scenario adaptability of the output scheme.

[0082] In step S106, the target scheduling scheme is executed, execution feedback is collected, and the coupling graph and scheduling parameters are incrementally updated based on the execution feedback.

[0083] It is understood that the embodiments of this application execute the target scheduling scheme and collect the corresponding execution feedback. Based on the feedback results, incremental updates are performed on the coupling graph and scheduling parameters. There is no need to fully rebuild the graph model and reset the parameters. This realizes the return flow of scheduling execution results to the underlying scheduling model, and the positive and negative business information in the scheduling process is deposited into the system. The basic data and control parameters of subsequent scheduling operations are continuously optimized, and the adaptive capability of the scheduling system is continuously improved.

[0084] In this embodiment, execution feedback is collected, and the coupling graph and scheduling parameters are incrementally updated based on the execution feedback. This includes collecting one or more of the following as execution feedback: task completion status, actual working hours of personnel, resource utilization rate, number of scheme switching times, and results of manual intervention; calculating feedback signals based on the execution feedback, wherein the feedback signals are calculated by weighting at least two of the following: project progress improvement, personnel and resource utilization rate, delay amount, switching cost, and hard constraint violation penalty; and updating one or more of the following based on the feedback signals: the weight of the edges between nodes in the coupling graph, the internal parameters of the candidate generation method, the scheduling triggering condition, and the conflict resolution strategy parameters.

[0085] Among them, execution feedback refers to the effect data generated during the actual execution of the target scheduling scheme, including at least one or more of the following: task completion status, actual working hours of personnel, resource utilization rate, number of scheme switching times, and results of manual intervention. The weight of the edge between nodes in the coupling graph refers to the correlation strength value corresponding to the edge connecting two nodes in the coupling graph, which is used to quantify the importance or credibility of the correlation between the two nodes. The internal parameters of the candidate generation method refer to the model parameters or algorithm parameters within each of the constraint optimization method, strategy decision method, and case reasoning method. The scheduling triggering condition refers to the conditions and thresholds used to determine whether to start a new round of scheduling, including at least the preset threshold for change, the interval of periodic scheduling time points, and the trigger sensitivity. The conflict resolution strategy parameters refer to the priority weight, threshold, or strategy selection parameters corresponding to each operation in the conflict resolution process, including at least the personnel replacement priority weight, resource replacement priority weight, upper limit of time slice shift amplitude, and maximum delay time of low priority tasks.

[0086] Feedback signal refers to a comprehensive evaluation value calculated based on collected execution feedback, used to quantify the execution effect of the scheduling scheme; in this embodiment, feedback signal Calculate using the following formula: ; in, As a measure of improvement in project schedule, For personnel and resource utilization, For the delayed amount, For the cost of switching, To enforce strict penalties for violations, to These are the weighting coefficients for each item.

[0087] It should be noted that, based on feedback, the strategy parameters, coupled graph edge weights, case confidence, scheduling trigger threshold, candidate risk prediction parameters, and candidate pruning threshold are updated, transforming failure information generated during candidate generation and execution into historical negative feedback samples. For candidate solutions that cannot form a complete solution, fail hard constraint verification, remain infeasible after conflict repair, are manually rejected, or experience rollback during actual execution, the corresponding affected subgraph structure, candidate generation channel, task coverage status, personnel and resource conditions, time constraints, failure type, and failure reason are recorded, and the historical failure mode set is updated. For candidate solutions that successfully pass the verification and complete the execution, positive feedback samples are generated, and the risk prediction function is updated based on the newly added positive and negative feedback samples. and candidate pruning threshold This enables subsequent scheduling to identify high-risk candidate paths similar to historical failure patterns in advance before the generation of complete candidate solutions, and to simultaneously save the input status, triggering reasons, affected subgraphs, candidates for each channel, constraint check results, repair actions, scoring basis, final solution, execution status and parameters before and after the update, forming a complete decision chain.

[0088] This application provides a dynamic scheduling method for personnel and heterogeneous technical resources based on coupled graphs. It utilizes time-varying coupled graphs to graphically represent skills, capacity, time windows, dependency conflicts, substitution relationships, and historical success rates, taking into account the collaborative execution conditions of personnel and technical resources. It supports local incremental updates, addressing the challenges of unified modeling of multi-source heterogeneous states and the inability to synchronously express collaborative constraints of personnel and technical resources. Based on the coupled graph, it achieves local incremental rescheduling by locating changing nodes, extracting affected subgraphs, and freezing external decisions. This controls the scope of influence and merges local and original solutions, reducing the solution space and overcoming the drawbacks of high overhead and severe perturbation of full rescheduling under dynamic events. Within the affected subgraph, it generates complete candidate solutions through constraint optimization, strategy decision-making, and case reasoning. After unified hard constraint verification and conflict repair, the optimal solution is selected for output, resulting in an executable scheduling plan. This solves the problems of difficulty in merging multiple model outputs and non-executable output results. Simultaneously, it collects various scheduling negative feedbacks to construct historical failure modes, performs risk prediction and pruning on candidate paths to reduce invalid computations, and iteratively feeds the coupled graph and scheduling parameters back into the scheduling process, enabling the accumulation and reuse of scheduling experience. This solves the problems in existing technologies, such as the difficulty in uniformly modeling multi-source heterogeneous states, the difficulty in simultaneously expressing the collaborative constraints of personnel and technical resources, the large overhead and scheme disturbance of full rescheduling under dynamic events, and the difficulty in continuously accumulating and reusing scheduling experience.

[0089] The following will illustrate a dynamic scheduling method for personnel and heterogeneous technical resources based on a coupling graph through a specific embodiment, such as... Figure 4 As shown, taking a solution delivery scenario where multiple customer projects share FDE personnel, testing environment, software licenses, and network maintenance windows as an example, the content is as follows: S1. Establish basic data on project tasks, personnel, and technical resources.

[0090] Each customer project is broken down into several schedulable tasks. Each task records the required skills, planned working hours, earliest start time, latest finish time, prerequisites, geographic or security permissions, as well as the required computing environment, testing equipment, software licenses, and network change windows. FDE personnel record skill tags, skill levels, regions, available working hours, current tasks, and historical delivery results. Technical resources record resource types, capacity, available time, location, substitution relationships, and fault status, completing the collection of multi-source business data and field standardization processing.

[0091] S2. Construct / incrementally update the resource-task coupling graph and form an initial solution.

[0092] The system generates corresponding nodes for tasks, FDE personnel, skills, technical resources, time slices, and solution components, and establishes relationships such as "Required Skills (Skill Requirements)", "Possessing Skills", "Occupied Resources (Occupied / Dependencies)", "Planned On", "Preceded", "Replaceable", "Assigned / Assigned", "Resource Dependencies", "Affected by Unavailability", "Included", and "Available / Occupied". Based on the daily plan window, the system generates an initial state and an initial scheduling plan on the global coupling graph. It also writes each personnel-task-resource-time slice allocation record to the project management platform and the resource management platform. When business changes occur in the scheduling domain, the system performs incremental updates to the coupling graph.

[0093] like Figure 5 As shown, the resource-task coupling diagram includes project task nodes 2101, FDE personnel nodes 2102, skill nodes 2103, technical resource nodes 2104, time slice nodes 2105, solution impact nodes 2106, and risk event nodes 2107. Relationships are established between nodes, including allocation / assignment, availability, skill requirements, occupancy / dependency, resource dependency, unavailability due to disruption, inclusion, and availability / occupancy. Project task nodes are associated with FDE personnel nodes through allocation / assignment relationships, FDE personnel nodes are associated with skill nodes through availability relationships, and project task nodes are associated with skill nodes through skill requirements. Project task nodes are associated with technical resource nodes through occupancy / dependency relationships. Project task nodes include solution impact nodes, which establish resource dependency relationships with technical resource nodes and availability / occupancy relationships with time slice nodes and risk event nodes, respectively. A causation-unavailability relationship is established between technical resource nodes and time slice nodes. This coupling diagram integrates multiple heterogeneous elements in the scheduling domain—tasks, personnel, skills, technical resources, time, scheduling allocation, and risk events—into a unified model, providing underlying data support for locating change nodes and extracting affected subgraphs.

[0094] S3. Generate the current state vector and calculate the degree of change.

[0095] During the execution of the original scheduling scheme, taking the newly added emergency fault handling task of Project A as an example, the emergency task requires FDE personnel with specific skills to complete the task within a limited time and using a dedicated test environment. The newly added task, corresponding skills, Project B task currently occupying the test environment, replaceable FDE personnel and related time slices are updated to the coupling graph to generate the current state vector of the scheduling domain and calculate the degree of state change.

[0096] S4. Determine the rescheduling trigger condition.

[0097] Determine whether the conditions for periodic triggering, event triggering, or constraint inverse triggering are met. If the degree of change exceeds the preset threshold or an emergency task identifier is detected, the triggering condition is determined to be met, and the local rescheduling process is initiated. If the triggering condition is not met, the status monitoring of the scheduling domain continues, and S1 is returned to continuously monitor multi-source scheduling data.

[0098] S5. Extract the affected subgraph and freeze irrelevant tasks.

[0099] Starting with the emergency task and dedicated test environment node of Project A as the change node, the neighborhood is expanded outward according to the preset number of hops and the "need, occupy, conflict, substitute, and precedence" relationship to obtain the affected subgraph. The affected subgraph includes the emergency task of Project A, the Project B task that competes with it for the test environment, FDE personnel with relevant skills, and the corresponding time slices. The scheduling tasks corresponding to Project C and Project D, which have no effective influence relationship with the above nodes, remain frozen and do not participate in this round of rescheduling and reallocation.

[0100] S6. Perform historical feedback prediction and candidate path pruning.

[0101] Based on the current project A's urgent task, dedicated testing environment, candidate FDE personnel, and corresponding time slices, candidate predictive features are constructed and matched with pre-stored historical failure patterns. For example, historical records show that when the urgent task time window is short, the remaining working hours of candidate FDE personnel are insufficient, and the availability confidence of the backup testing environment in the corresponding time period is low, the probability of failure in the generation or constraint verification of such candidate subsequent solutions is high. Based on this, the predicted risk of each candidate generation path is calculated. When the risk exceeds the candidate pruning threshold, the time slice expansion and complete solution construction of the personnel-resource combination are no longer carried out, and the path is directly discarded, and other personnel or technical resource combinations are turned to.

[0102] S7. Multi-channel generation of candidate scheduling schemes.

[0103] Within the affected subgraph, candidate scheduling schemes are generated through multiple channels: the constraint optimization channel generates several local candidate schemes and outputs a joint allocation result including FDE personnel, test environment, license, and time slice; the strategy decision channel outputs scheduling actions such as "postpone low-priority tasks", "call alternative FDE", and "switch to backup test environment", which are then expanded by the action expander to form a complete candidate scheduling scheme; the case reasoning channel retrieves successful scheduling cases with the same skill requirements and resource competition structure in history, and generates knowledge-driven candidate scheduling schemes after personnel variable replacement and time alignment.

[0104] S8, Perform constraint verification, conflict resolution, and candidate repair.

[0105] Feasibility verification is conducted on all candidate scheduling schemes to check for constraints such as personnel double occupation, resource capacity exceeding limits, insufficient skill thresholds, time window conflicts, and violations of pre- and post-requirement relationships. For candidates with conflicts, a conflict graph is constructed, where nodes represent conflict allocation actions and edges represent mutual exclusion or resource competition relationships. The order of repair is determined by combining project priority and remaining time slack. For example, if a dedicated testing environment in a candidate scheme conflicts with project B task, a corresponding mutual exclusion edge is established in the conflict graph, and conflict resolution operations such as shifting project B task, replacing the backup testing environment, or local re-solving are executed in sequence until the candidate scheme meets all constraints.

[0106] S9. Dynamically score and select the final solution and issue it to handle execution exceptions.

[0107] For feasible candidate scheduling schemes after conflict resolution, a comprehensive score is calculated based on the completion rate of emergency tasks, personnel overload, resource occupancy, disturbance magnitude of the original scheme, and historical success consistency. The candidate with the highest comprehensive score is selected as the target scheduling scheme. The local scheme of the target scheduling scheme is merged with the original scheduling decisions that are frozen and retained to obtain a new global scheduling scheme. Personnel affiliation, resource reservation, and time window are updated through the interface. If an operational anomaly is detected during execution, such as a failure of the backup test environment, the allocation action corresponding to that environment is canceled, the remaining valid scheduling actions are retained, the failure environment and affected tasks are marked as new change nodes, and a new round of local rescheduling is triggered in S4.

[0108] S10. Collect feedback and update graph relationships and model parameters.

[0109] Within a preset observation window, execution feedback is collected, including actual task completion time, planned-actual time deviation, resource utilization, manual reassignment records, and number of abnormal rollbacks. If the personnel-task combination performs well multiple times, the historical success weight of the corresponding "available / suitable" edge in the coupling graph is increased. If a resource frequently fails during a specific period, the availability confidence of that resource is reduced. If a candidate channel or candidate generation path repeatedly fails to complete a solution, fails to verify hard constraints, fails to resolve conflicts, is manually rejected, or is rolled back, the reliability of that channel is reduced, and the failure characteristics are included in the historical failure mode set to increase the prediction risk of similar candidate paths. For candidate paths that consistently execute successfully, their prediction risk is reduced. Candidate risk prediction parameters and pruning thresholds are updated synchronously. The updated coupling graph, historical failure modes, and scheduling model parameters are written back to S2 in a closed loop for use in the next round of scheduling calculations.

[0110] like Figure 6As shown, the final scheduling scheme is input to the interface for distribution and execution, completing the deployment and execution of the scheme at the personnel, resource, and time window levels. The effect collection unit collects actual operational data such as task delivery progress, load, faults, and switching costs. The quality evaluation conducts scheduling quality evaluation based on the collected operational data, completing reward and deviation calculations. The update module updates strategy parameters, coupling graph edge weights, case confidence, and trigger thresholds according to the quality evaluation results. The next-moment state and resource-task coupling graph unit outputs the updated system state and coupling graph, and enters the next round of scheduling. This application forms a complete closed loop of scheduling solution, implementation, effect evaluation, and parameter iteration, continuously optimizing scheduling-related parameters and coupling graph models using real-world operational feedback, and constantly improving the adaptability of subsequent scheduling output schemes.

[0111] Next, referring to the accompanying drawings, a dynamic scheduling system for personnel and heterogeneous technical resources based on a coupling graph is described according to an embodiment of this application.

[0112] Figure 7 This is a schematic diagram of the structure of a dynamic scheduling system for personnel and heterogeneous technical resources based on a coupling graph, according to an embodiment of this application.

[0113] like Figure 7 As shown, the dynamic scheduling system 10 for personnel and heterogeneous technical resources based on coupling graph includes: acquisition module 1000, positioning module 2000, extraction module 3000, generation module 4000, determination module 5000 and update module 6000.

[0114] The acquisition module 1000 is used to acquire personnel scheduling data and resource scheduling data; the positioning module 2000 is used to construct a coupling graph based on the personnel scheduling data and resource scheduling data. The coupling graph is used to uniformly represent multiple heterogeneous elements and their interrelationships within the scheduling domain as a graph structure with relationships. Based on the coupling graph, it senses state changes within the scheduling domain. When a state change renders an existing scheduling plan unexecutable, it locates the changed node in the coupling graph corresponding to the state change; the extraction module 3000 is used to dynamically determine the affected range by expanding along the path represented by the coupling graph, centered on the changed node, and extracting the affected subgraph from the coupling graph. The existing scheduling decisions outside the affected subgraph are frozen; the generation module 4000 is used to generate candidate scheduling schemes within the affected subgraph using multiple candidate generation methods. Each candidate scheduling scheme contains multiple allocation units, and each allocation unit associates and binds at least personnel, tasks, resources, and time; the determination module 5000 is used to verify the feasibility of each candidate scheduling scheme, resolve conflicts for candidate schemes with conflicts, and select the target scheduling scheme from the feasible candidate scheduling schemes after verification and conflict resolution; the update module 6000 is used to execute according to the target scheduling scheme, collect execution feedback, and incrementally update the coupling graph and scheduling parameters based on the execution feedback.

[0115] It should be noted that the foregoing explanation of an embodiment of a dynamic scheduling method for personnel and heterogeneous technical resources based on a coupling graph also applies to this embodiment of a dynamic scheduling system for personnel and heterogeneous technical resources based on a coupling graph, and will not be repeated here.

[0116] The following specific embodiment illustrates a dynamic scheduling system for personnel and heterogeneous technical resources based on a coupling graph, such as... Figure 8 As shown, the content is as follows: The customer demand / work order system, project management platform, FDE work hours and skills library, technical resource management platform, and operation logs and feedback data are connected as external data sources to the data acquisition and standardization module 100. The data acquisition and standardization module 100 includes data collection 101, field mapping 102, and time series alignment 103, completing multi-source data collection, field mapping, and time series alignment processing. The processed data is sent to the resource-task coupling graph module 200, which includes graph construction 201, state update 202, and association weight maintenance 203, used to construct and maintain the resource-task coupling graph. The state perception and triggering module 300 includes state vector generation 301, change degree calculation 302, and scheduling trigger judgment 303, which perceives system state changes based on the coupling graph and completes rescheduling trigger judgment. The affected subgraph extraction and candidate prediction generation module 400 includes constraint optimization channel 401, strategy decision channel 402, graph reasoning channel 403, historical feedback prediction and pruning. 404. Complete seed generation 405, completing the extraction of affected subgraphs and the generation of multi-path candidate solutions; Feasibility verification and repair module 500 includes fast graph traversal 501, constraint repair 502, and candidate deduplication 503, performing feasibility verification, conflict repair, and deduplication processing on candidates; Adaptive evaluation and selection module 600 includes scene weight calculation 601, candidate scoring 602, and optimal solution selection 603, combining scene dynamic weights to complete candidate scoring and output the optimal scheduling solution; Solution generation and execution module 700 includes plan splitting 701, interface distribution 702, and exception rollback 703, realizing solution splitting and distribution and exception rollback; Feedback learning module 800 includes effect collection 801, model / graph weight update 802, and threshold calibration 803, collecting execution results to complete the iterative update of model, graph weights, and thresholds; Audit and interpretation module 900 includes decision chain recording 901 and constraint basis 902, recording the scheduling decision chain and constraint basis, realizing that the scheduling process is auditable and interpretable. This system drives local incremental rescheduling through coupled graphs, integrates multi-channel candidate generation and feedback iterative learning, balances scheduling efficiency and solution quality, and achieves full traceability of scheduling decisions.

[0117] In summary, this application provides a dynamic scheduling system for personnel and heterogeneous technical resources based on a coupled graph. By connecting with multi-source business systems to complete data standardization processing, a resource-task coupled graph is constructed to uniformly model the relationships between tasks, personnel, and various heterogeneous technical resources. When business disturbances occur, the system senses event changes and extracts the affected subgraph, transforming global scheduling into local incremental rescheduling, reducing the solution scale and disturbances to the original scheduling scheme. A multi-channel parallel approach is used to generate candidate scheduling schemes, which are then screened for feasibility verification, constraint repair, and scenario adaptive evaluation to obtain the optimal scheduling scheme. After the scheduling scheme is issued and executed, the actual operating effect is collected, and the graph weights, strategy parameters, and scheduling trigger thresholds are continuously updated based on feedback, forming a closed-loop iterative optimization. Simultaneously, the scheduling decision-making chain and constraint basis are fully retained, making the scheduling process auditable and explainable. This application can adapt to dynamically changing business scenarios and effectively reduces computational overhead while ensuring the quality of the scheduling scheme.

[0118] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium. When executed, the program includes one or a combination of the steps of the method embodiments.

[0119] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for dynamic scheduling of personnel and heterogeneous technical resources based on a coupling graph, characterized in that, include: Acquire personnel scheduling data and resource scheduling data; A coupling graph is constructed based on the personnel scheduling data and resource scheduling data. The coupling graph is used to uniformly represent multiple heterogeneous elements and their interrelationships within the scheduling domain as a graph structure with correlation relationships. The state changes within the scheduling domain are perceived based on the coupling graph. When the state change causes an existing scheduling scheme to become unexecutable, the changed node corresponding to the state change in the coupling graph is located. Centered on the changed node, the affected range is dynamically determined by expanding along the path according to the association represented by the coupling graph. The affected subgraph is extracted from the coupling graph, and the existing scheduling decisions outside the affected subgraph are frozen. Within the affected subgraph, candidate scheduling schemes are generated through multiple candidate generation methods. Each candidate scheduling scheme contains multiple allocation units, and each allocation unit associates and binds personnel, tasks, resources, and time at least once. For each candidate scheduling scheme, a feasibility check is performed; for candidate schemes with conflicts, conflict resolution is performed; and the target scheduling scheme is selected from the feasible candidate scheduling schemes after check and conflict resolution. The target scheduling scheme is executed, execution feedback is collected, and the coupling graph and scheduling parameters are incrementally updated based on the execution feedback.

2. The method according to claim 1, characterized in that, Centered on the changed node, the affected range is dynamically determined by extending along the path according to the association represented by the coupling graph. The affected subgraph is extracted from the coupling graph, and existing scheduling decisions outside the affected subgraph are frozen, including: Starting from the changed node, determine the initial expansion step number according to the type of the changed node; Starting from the initial expansion step number, expand outward layer by layer along the association relationships in the coupling graph; After each layer is expanded, it is determined whether there is a valid influence relationship between the newly included node and the changed node. If a valid influence relationship exists, the expansion continues; otherwise, the expansion stops. The affected range is defined as the area involved in the final expansion, and the nodes and edges within the affected range are extracted from the coupling graph to form the affected subgraph; A freeze flag is set for existing scheduling decisions associated with nodes in the coupling graph that do not belong to the affected subgraph, so as to keep their original allocation unchanged.

3. The method according to claim 2, characterized in that, The effective influence relationship includes one or more of the following: There is a personnel or resource sharing relationship between the newly included node and the changed node; There is a task pre- or post-task dependency between the newly included nodes and the changed nodes; There is a time window conflict between the newly included nodes and the changed nodes; There is a resource substitution relationship between the newly included nodes and the changed nodes; There is a capacity competition relationship between the newly added nodes and the changed nodes; There is a hard constraint transitive relationship between the newly included node and the changed node.

4. The method according to claim 1, characterized in that, The state change renders existing scheduling schemes unexecutable, including: Calculate the changes in task requirements, personnel availability, and resource availability between the current state and the previous state. Use the weighted sum of these changes as the degree of change. When the degree of change exceeds a preset threshold, determine that the state change renders the existing scheduling scheme unexecutable; or When the state change causes personnel conflicts, resource capacity exceeding limits, or task time window violations in the existing scheduling scheme, it is determined that the state change renders the existing scheduling scheme unexecutable; wherein, when the state change causes personnel conflicts, resource capacity exceeding limits, or task time window violations in the existing scheduling scheme, regardless of whether the degree of change exceeds the preset threshold, it is determined that the state change renders the existing scheduling scheme unexecutable.

5. The method according to claim 1, characterized in that, The multiple candidate generation methods include at least two of the following: Constraint optimization method: Establish the objective function and constraints, solve them, and generate candidate scheduling schemes; Strategy decision-making method: Based on the historical scheduling status and action output, the scheduling action is output, and the action expander generates candidate scheduling schemes; Case-based reasoning: Retrieve successful cases that match the currently affected subgraph from historical scheduling cases, perform variable substitution and time alignment, and generate candidate scheduling schemes.

6. The method according to claim 1, characterized in that, The conflict resolution includes: Construct a conflict graph, where nodes represent allocation actions that conflict, and edges represent mutual exclusion or resource competition relationships between allocation actions. The repair order is determined based on task priority, availability of alternative resources, switching costs, and remaining time slack. According to the repair order, perform one or more of the following operations on the allocation actions corresponding to the conflict nodes in the conflict graph: personnel replacement, resource replacement, time slice shifting, low-priority task postponement, local re-solution, until the candidate scheduling scheme meets all feasibility conditions.

7. The method according to claim 1, characterized in that, The target scheduling scheme is selected from the feasible candidate scheduling schemes after verification and conflict resolution, including: Based on the characteristics of the current scheduling scenario and the historical performance of each candidate generation method, an evaluation weight is generated for each feasible candidate scheduling scheme. Calculate the comprehensive score for each feasible candidate scheduling scheme based on the evaluation weights; The candidate scheduling scheme with the highest comprehensive score is determined as the target scheduling scheme.

8. The method according to claim 1, characterized in that, Collect execution feedback, and incrementally update the coupling graph and scheduling parameters based on the execution feedback, including: One or more of the following data are used as execution feedback: task completion status, actual working hours of personnel, resource utilization rate, number of scheme switching times, and results of manual intervention. The feedback signal is calculated based on the execution feedback, wherein the feedback signal is calculated by weighting at least two of the following: project progress improvement, personnel and resource utilization, delay, switchover cost, and hard constraint violation penalty. Update one or more of the following based on the feedback signal: the weight of the edges between nodes in the coupled graph, the internal parameters of the candidate generation method, the scheduling triggering condition, and the conflict resolution strategy parameters.

9. The method according to claim 1, characterized in that, Before generating candidate scheduling schemes using multiple candidate generation methods, a preliminary risk assessment is also included: Obtain historical failure records, including records of candidate solution generation failure, feasibility verification failure, conflict resolution failure, manual rejection, or rollback. The risk of each candidate generation path is assessed by matching the characteristics of the currently affected subgraph with the historical failure records, and the assessment result is obtained. Candidate paths deemed high-risk by the evaluation results are eliminated, and no complete candidate scheduling schemes are generated for them.

10. A dynamic scheduling system for personnel and heterogeneous technical resources based on a coupled graph, characterized in that, include: The acquisition module is used to acquire personnel scheduling data and resource scheduling data; The positioning module is used to construct a coupling graph based on the personnel scheduling data and resource scheduling data. The coupling graph is used to uniformly represent multiple heterogeneous elements and their interrelationships within the scheduling domain as a graph structure with association relationships. Based on the coupling graph, the module senses state changes within the scheduling domain. When the state change causes an existing scheduling scheme to become unexecutable, the module locates the changed node in the coupling graph corresponding to the state change. The extraction module is used to dynamically determine the affected range by expanding along the path based on the association represented by the coupling graph, with the changed node as the center, extracting the affected subgraph from the coupling graph, and freezing the existing scheduling decisions outside the affected subgraph. The generation module is used to generate candidate scheduling schemes through multiple candidate generation methods within the affected subgraph range. Each candidate scheduling scheme contains multiple allocation units, and each allocation unit associates and binds personnel, tasks, resources, and time at least. The determination module is used to perform feasibility verification on each of the candidate scheduling schemes, perform conflict resolution on candidate schemes with conflicts, and select the target scheduling scheme from the feasible candidate scheduling schemes after verification and conflict resolution. The update module is used to execute the target scheduling scheme, collect execution feedback, and incrementally update the coupling graph and scheduling parameters based on the execution feedback.

Images