A peak-constrained collaborative resource repair scheduling method and scheduling system

Abstract

This application provides a peak-constraint collaborative resource repair scheduling method and system. The method includes the following steps: Step 1, data input, which collects three types of information: task data, constraint data, and parameter configuration; Step 2, initial scheduling generation, which generates an initial scheduling result based on the task data, constraint data, and parameter configuration; Step 3, collaborative resource repair, which repairs the initial scheduling result, including peak constraint landing point selection and post-processing smoothing repair; Step 4, constraint consistency correction, which performs post-rule consistency correction on the scheduling result after collaborative resource repair and performs capacity feasibility verification; Step 5, output, which outputs the final scheduling result after the above repair and outputs process records or logs. This PCSR method achieves peak load control, resource over-limit suppression, scheduling feasibility maintenance, and load balancing improvement without global re-scheduling, and is suitable for multi-task and multi-resource scheduling scenarios under complex constraints.

Description

Technical Field

[0001] This application belongs to the field of scheduling technology, specifically a peak-constrained collaborative resource repair scheduling method and scheduling system. Background Technology

[0002] In the monthly task allocation and execution management, testing and inspection organizations need to coordinate tasks from multiple departments. Tasks typically have clear time windows and are constrained by resources such as personnel, equipment, and facilities, as well as resource exclusivity. Collaborative tasks also involve simultaneous use of multiple resources and cross-departmental coordination, increasing scheduling complexity. In current engineering practices, to quickly generate an executable plan, an initial schedule is usually generated using priority ranking, heuristics, or greedy strategies, and then adjusted based on resource conflicts and schedule deviations during execution. While this approach can quickly yield feasible solutions, it lacks a dedicated mechanism for handling the problem of "dense clustering on the timeline."

[0003] When tasks occur concentrated within a few time periods, the initial scheduling is prone to localized congestion, manifesting as excessively high peak loads, increased queuing, and escalating conflicts in collaborative tasks, further increasing the risk of overdue tasks. These problems are exacerbated when the proportion of collaborative tasks is high, resource capacity varies significantly, or time windows are tight. Existing improvement methods often face two types of difficulties: one is that expanding the adjustment scope can disrupt plan stability, leading to large-scale rescheduling and increased implementation costs; the other is that in the absence of a "movable task set" constraint, localized movements can easily disrupt prior dependencies or introduce new mutual exclusion conflicts, resulting in a chain reaction of adjustments that are difficult to control. Therefore, a post-processing repair technique for the initial scheduling is needed. This technique, while maintaining the feasibility of hard constraints, should implement controlled adjustments to key congested segments to selectively reduce peak loads and simultaneously suppress overdue risks. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application provides a peak-constraint collaborative resource repair scheduling method and system. This method repairs the schedule based on a feasible initial schedule output by an existing scheduling algorithm, without replacing the original algorithm's generation process. Instead, it targets and compresses local peak loads while suppressing overdue risks, thus improving the stability, executability, and resource load balance of the scheduling results, while maintaining the integrity of hard constraints. The hard constraints include at least time window constraints, prerequisite dependencies, departmental capacity limits, priority rules for urgent tasks, rules for postponing destructive tasks, and collaborative and mutual exclusion rules defined by business requirements. The PCSR employs a collaborative mechanism of "peak shaving at the landing stage + smooth repair at the result stage," and ensures feasibility throughout the process through consistency correction.

[0005] This technology is implemented using the following technical solution: A peak-constrained collaborative resource repair scheduling method, comprising the following steps: Step 1: Data Input. Input the three types of information: task data, constraint data, and parameter configuration. Step 2: Initial schedule generation. The initial schedule result is generated based on task data, constraint data, and parameter configuration. Step 3 is collaborative resource repair, which repairs the results of the initial scheduling. Collaborative resource repair includes peak constraint landing point selection and post-processing smoothing repair. Step 4: Constraint Consistency Correction. Perform post-rule consistency correction on the scheduling results after collaborative resource repair, and perform capacity feasibility verification. Step 5 outputs the final scheduling result after the above repairs, and also outputs the process record or log.

[0006] Furthermore, the specific steps of step 2, the initial scheduling, are as follows: First, a sample-level processing order is established, which can be based on the comprehensive score to rank the samples. Subsequently, priority scheduling for urgent tasks is performed, with the set of urgent tasks occupying resources first and completing capacity feasibility checks; if an urgent task is not feasible, the corresponding set of failed urgent tasks is output; after the urgent task scheduling is successful, the scheduling of non-urgent tasks continues. Finally, the schedule is checked for capacity feasibility on the discrete time axis. The check principle is that the occupancy of any department in any discrete time slice does not exceed its capacity limit, thus obtaining the initial scheduling result.

[0007] Furthermore, the peak constraint landing point selection in step 3 specifically includes three stages: feasible window construction, feasible starting point enumeration and peak prediction, and cost evaluation and optimal landing point determination. The construction of the feasible window first determines the range of selectable starting points based on the mapping relationship between the hard and soft time windows of the task, and then performs a trigger determination. The feasible starting point enumeration and peak prediction process involves enumerating candidate starting points within the selectable starting point range and performing capacity constraint filtering during the enumeration process to retain only feasible starting points. For each feasible starting point, the peak occupancy of the time period after the task is placed is predicted based on a preset time granularity, and the risk amount such as the overdue half-day amount relative to the original end time is calculated. The cost assessment and optimal landing point determination are then carried out. Subsequently, the cost is constructed and the starting point with the optimal cost is selected as the final landing point. Finally, the landing point comparison record is output. The comparison record includes at least the sample identifier, department identifier, window boundary, baseline landing point, optimal landing point, and the corresponding peak value and overdue amount.

[0008] Furthermore, the post-processing smoothing repair in step 3 specifically includes the following steps: S1 takes the initial scheduling results and the sorted sequence of the comprehensive scores as input; S2 performs time-segment discretization modeling, dividing the time axis into discrete time segments, and aggregates task counts for departmental load in the form of "department × time slice" to obtain a comparable load distribution representation. S3 identifies the segment with the highest load; S4 determines the location of the period when the current pressure is most concentrated, and selects a set of candidate samples around this period, forming a protection set based on the candidate set; The S5 protection set is determined based on the scoring ranking. TopK scoring samples are not included in the scoring downweighting process to stabilize the ranking structure of key samples. Trial scoring downweighting is performed on non-protected samples. S6 reorders the trial sequences according to the updated scores; S7 then calls the scheduling calculation based on the trial sequence to generate a new scheduling result; For each trial result, S8 first performs a feasibility assessment. If the trial violates capacity feasibility or other hard constraints, it rolls back and continues to the next round of trials. S9 When the trial results are feasible, further evaluation of indicators should be carried out. The indicators should include at least peak level, failure rate and overdue rate, and the acceptance criteria should be determined based on the allowable incremental threshold and peak reduction requirements. If the acceptance criteria are not met in S10, then backtrack and continue with new trial adjustments; If the acceptance criteria are met in S11, the current scheduling result is updated and used as a new baseline for the next round of identification and exploration. S12 executes S3 through S11 in a loop until the termination condition is met, including reaching the maximum number of iterations, insufficient continuous improvement, or no acceptable solution. S13 finally outputs the repaired scheduling results and the repair process record. The repair process record records at least the peak position identified in each round, the size of the candidate set, the number of attempts, the number of acceptances / backoffs, and the changes in key indicators, for verification and reproduction.

[0009] Furthermore, the constraint consistency correction in step 4 is used to ensure that the repair result meets the consistency requirements of destructive post-rules and capacity constraints. This step uses post-rules to be executed at a half-day granularity, which can realize the constraint requirements of different half-days on the same day or half-days on the next day. When the post-correction causes the original landing point to no longer meet the rules, the feasible starting point is re-determined within the allowable range, and the capacity feasibility check continues to be performed, thereby ensuring that the destructive post-rules and capacity constraints are met at the same time, and avoiding the introduction of infeasibility risks during the repair process.

[0010] Furthermore, the output step described in step 5 is used to output the repaired scheduling results and process records / logs. The scheduling results include at least the task start time, end time, department allocation information, overdue mark, and failure sample set; the process records include at least the landing point comparison log and the post-processing smoothing repair iteration log.

[0011] A peak-constrained collaborative resource repair scheduling system includes: The input module is used to collect three types of information: task data, constraint data, and parameter configuration. Task data includes at least the sample, department, start and end times of the time window, task cycle, priority, number of samples, and destructive flag; constraint data includes at least the department capacity limit, soft time window mapping relationship, and destructive post-rule; parameter configuration includes at least the time granularity, weight parameters, allowable incremental threshold, termination condition, and log switch. The initial scheduling generation module is used to obtain feasible initial scheduling results; The collaborative resource repair module is used to repair feasible initial ranking results, including peak constraint landing point selection and post-processing smoothing repair. The constraint consistency correction module is used to ensure that the repair results meet the consistency requirements of destructive post-rules and capacity constraints; The output module is used to output the repaired scheduling results and process records / logs.

[0012] Furthermore, the collaborative resource repair module includes a peak constraint landing point selection module and a post-processing smoothing repair module. The peak constraint landing point selection module operates in the task landing point determination stage, performing peak shaving and point selection for non-urgent tasks with soft time windows. The post-processing smoothing repair module is used to perform iterative repair after the initial scheduling results are output.

[0013] An electronic device, characterized in that it includes one or more processors, the processors executing the peak constraint collaborative resource repair scheduling method described in any one of the preceding claims.

[0014] A computer storage medium storing computer instructions, the computer instructions being configured to cause a computer to execute the peak-constrained collaborative resource repair scheduling method described above.

[0015] In summary, the technical solutions provided in the embodiments of this application have the following technical effects or advantages: 1. This method is designed for post-processing repair of the initial schedule, implemented without violating hard constraints such as time windows, prerequisite dependencies, and resource mutual exclusion. Its focus is on targeted adjustments during periods of localized congestion to achieve peak load constraints and load balancing.

[0016] 2. This method and system are applicable to the monthly task assignment and execution management of testing and inspection institutions, addressing the business characteristics of multi-department collaboration, varying resource capabilities, and concentrated tasks on the timeline. To address the risks of queuing, collaborative conflicts, and overdue tasks caused by excessively high load peaks, this invention reduces peak loads and simultaneously suppresses risks through remedial scheduling. This technology can be used in conjunction with any initial schedule generation method to improve the executability and stability of the schedule.

[0017] 3. Without global reordering, the PCSR method can achieve consistent comparison results in terms of peak control, resource overrun control, scheduling feasibility maintenance and load balancing improvement. It is suitable for multi-task and multi-resource scheduling scenarios under complex constraints. Attached Figure Description

[0018] Figure 1 This is a flowchart of the overall PCSR process in this application; Figure 2 This is a flowchart of the smooth repair process in this application; Figure 3 This is a diagram illustrating the overall benefit comparison of the PCSR method in this application; Figure 4 This is a diagram showing the peak values ​​of various departments and units in this application; Figure 5 This is a time-series comparison diagram of the company-level half-day peak load in this application; Figure 6 This is a schematic diagram comparing the half-day load timing of department H in this application. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0020] A Peak-Constrained Collaborative Resource Repair (PCSR) scheduling method is proposed for multi-task, multi-resource scheduling scenarios. The following section combines... Figure 1 The overall flowchart shown, and Figure 2 The flowchart shown below illustrates the specific implementation method of the smooth repair process. In this embodiment, PCSR uses "sample-department-time window" as the basic organizational unit for scheduling and repair. Input data includes at least three categories: task data, constraint data, and parameter configuration. Task data includes at least the sample, department, start and end times of the time window, task cycle, priority, number of samples, and destructive flags; constraint data includes at least the department capacity limit, soft time window mapping relationship, and destructive task post-reset rules; parameter configuration includes at least time granularity, weight parameters, allowed incremental threshold, termination conditions, and log switch.

[0021] In this embodiment, the time axis is modeled using discretization. Preferably, half a day is used as the basic time slice, and any continuous time is mapped to a half-day index. The task cycle is converted into duration slices in half-day units. Departmental resource occupancy is maintained using an occupancy count of "department × time slice", and the department capacity limit is used as the feasibility verification benchmark, that is, the occupancy count of any department in any time slice does not exceed the capacity limit. The above discretization modeling method facilitates subsequent peak shaving and smoothing repair to be performed at a uniform granularity. As an alternative implementation, the time granularity can be replaced with hours, days or other fixed-duration time slices, or a variable-granularity time slice that is adaptively divided according to the task density can be used, as long as the mapping from task time windows to discrete indices and the cumulative statistics of department occupancy over time slices can be achieved. During the feasibility verification process, the pre-dependencies between tasks are kept intact to avoid dependency chain reversal or conflict due to the adjustment of the landing point.

[0022] In this embodiment, the initial schedule generation process is executed first. This process preferably employs a sample-level processing order, i.e., first determining the sample processing order, and then determining the landing point for each task within the sample. The sample-level processing order can be obtained by sorting based on a comprehensive score, which can be formed by weighting features such as urgency, rush level, and resource conflict, and can be aggregated by sample for sorting. During the initial schedule generation process, the set of urgent tasks is processed first, and a capacity feasibility check is performed on these urgent tasks. If there are infeasible samples among the urgent tasks, the set of failed urgent tasks is output, and subsequent non-urgent scheduling is stopped. After successful urgent scheduling, the set of non-urgent tasks continues to be processed, and a capacity feasibility check is performed on each task to generate a feasible initial schedule result. As an alternative implementation, the urgency priority strategy can be replaced by "grouping and hierarchical scheduling by priority" or "sorting by due date," as long as it ensures that urgent tasks receive priority in the sorting relationship and complete the feasibility check.

[0023] In this embodiment, the PCSR collaborative resource repair module includes two parallel core repair operators: peak constraint landing point selection and post-processing smoothing repair. Figure 1The two types of operators can be enabled individually or simultaneously. When enabled simultaneously, peak constraint landing point selection is performed first to reduce local peak pressure in the landing point stage, and then post-processing smoothing repair is performed to further compress peaks and suppress risks in the result stage. At the same time, the process record is output for easy review.

[0024] Regarding the selection of the peak constraint landing point, in this embodiment, the triggering condition is preferably set to "non-urgent and with a soft time window". For tasks that meet the triggering condition, a feasible window is first constructed: the starting discrete index of the task's hard time window is used as the earliest starting point, and the ending discrete index of the task's hard time window plus the soft time window expansion amount is used as the upper bound of the optional starting point. The range of enumerable candidate starting points is determined by combining the number of task duration slices. Then, starting points are enumerated within this candidate range, and capacity feasibility checks are performed one by one, retaining only feasible starting points. For each feasible starting point, peak prediction and risk calculation are performed. Peak prediction uses the peak occupancy of the time period under a preset time granularity as the peak index, and the risk is preferably used as the number of overdue time slices relative to the original end time as the risk index. Based on this, a cost function is constructed and the starting point with the minimum cost is selected as the final landing point. The cost function is preferably formed by a linear combination of a peak term, an overdue term, and a starting point preference term, where the peak term is used to suppress local peaks, the overdue term is used to suppress overdue risk, and the starting point preference term is used to avoid meaningless excessive shifting. As an alternative implementation, the peak value can be the maximum occupancy of a department in a single time slice, or the maximum occupancy of two half-day time slices on a daily scale; the overdue value can be the number of overdue half-days, the number of overdue days, or an overdue indicator variable; the starting point preference can be replaced with a slight preference for an earlier starting point, or with a penalty for the deviation of the starting point from the original planned starting point. The weight parameters can be configured as constants or adaptively adjusted according to the department's congestion level, as long as comparable costs can be established among candidate starting points and the optimal starting point can be selected. To enhance traceability, the peak constraint landing point selection can output a landing point comparison record. The record should include at least the sample, department, priority, window boundary, baseline landing point, optimal landing point, and the corresponding peak value and overdue amount, for comparative analysis and verification.

[0025] In this embodiment, after the task landing point within the sample is determined, constraint consistency correction is performed, preferably with post-restriction rule correction for destructive tasks. Specifically, when a destructive task exists within the sample, the latest end position of the non-destructive task is first determined, and the earliest allowed start position of the destructive task is determined accordingly. Post-restriction constraints are implemented at half-day granularity, allowing different half-days on the same day or half-days on the next day. When the current landing point of the destructive task is later than the earliest allowed start position, it is allowed to scan forward and select an earlier feasible starting point without violating capacity constraints to reduce invalid waiting. Finally, a lower bound is applied to the starting point of the destructive task to ensure that it is not earlier than the earliest allowed start position. As an alternative implementation, the destructive post-restriction constraint can be replaced with rules such as requiring an interval of N time slices or requiring execution across days, as long as the feasible starting point of the destructive task can be limited to the range that satisfies the post-restriction relationship and triggers re-selection when necessary.

[0026] After obtaining the initial sorting results (including the landing point correction resulting from peak constraint landing point selection), post-processing smoothing correction is performed to iteratively correct the sorting results. The process is as follows: Figure 2As shown. Post-processing smoothing repair first reads the initial scheduling results and the scoring ranking sequence, performs time-period discretization modeling, and aggregates task counts for each department's load by department × time slice to quantify the load distribution over time. Then, it identifies the segment with the highest load, determines the department and time segment with the most concentrated current pressure, and filters a candidate sample set around this segment. To avoid excessive disturbance to critical samples, this embodiment sets up a protection set mechanism: the TopK samples are selected as the protection set based on the scoring ranking, where TopK represents the set of samples ranked in the top K positions. Samples in the protection set do not participate in score reduction. For non-protected samples, a trial-based score reduction is performed within the candidate set, and a trial sequence is generated by re-sorting according to the updated scores. Based on the trial sequence, a new scheduling result is generated by calling the scheduling calculation, and a feasibility judgment is first performed. If it is not feasible, the trial is rolled back and the next trial continues. For feasible trial results, a set of indicators for acceptance judgment is calculated, including at least peak level, failure rate, and overdue rate. Then, acceptance judgment is made based on the allowable increment threshold and peak reduction requirements. If the acceptance criteria are met, the current scheduling result is updated as a new baseline and the next round of identification and trial is initiated. If the acceptance criteria are not met, the process is rolled back and trial continues. The iterative process continues until the termination conditions are met. Preferred termination conditions include reaching the maximum number of iterations, insufficient continuous improvement, or the absence of an acceptable solution. Finally, the corrected scheduling result and a correction process record are output. The process record includes at least the peak location identified in each round, the size of the candidate set, the size of the protection set, the number of trials, the number of acceptances and rollbacks, and changes in key indicators, for verification and reproduction. As an alternative implementation, the candidate set screening can be replaced by screening based on occupancy contribution during peak periods or by screening congested samples within departments; the protection set can be replaced by generating it based on a scoring threshold, an expiration threshold, or sample importance; scoring weighting can use multiplicative weighting or additive bias, as long as the ranking can be changed during the trial and different results can be formed through rescheduling. Acceptance criteria can be replaced with forms such as the peak value must decrease and the failure rate must not increase, the peak value decrease must be prioritized and the increase in the overdue rate must be limited; termination conditions can be replaced with forms such as the peak value reaching the target threshold, reaching the maximum number of attempts, or no improvement after several rounds.

[0027] In this embodiment, the output module is used to output the scheduling results and process records / logs after PCSR repair. The scheduling results include at least the task start time, end time, overdue flag, and failure sample set; the process records include at least the landing point comparison log and the smooth repair iteration log. The above outputs can be used for subsequent statistical analysis, as well as for runtime auditing and reproduction verification, thereby supporting the engineering deployment and continuous optimization of PCSR in multi-constraint collaborative scheduling scenarios.

[0028] A peak-constrained collaborative resource repair scheduling system includes an input module, an initial schedule generation module, a collaborative resource repair module, a constraint consistency correction module, and an output module. The input module is used to collect three types of information: task data, constraint data, and parameter configuration. Task data includes at least the sample, department, start and end times of the time window, task cycle, priority, number of samples, and destructive flags. Constraint data includes at least the department capacity limit, soft time window mapping relationship, and destructive post-rules. Parameter configuration includes at least the time granularity, weight parameters, allowable incremental threshold, termination conditions, and log switch. The initial scheduling generation module is used to obtain feasible initial scheduling results; The collaborative resource repair module is used to repair feasible initial scheduling results, including peak constraint landing point selection and post-processing smoothing repair. The peak constraint landing point selection module operates in the task landing point determination stage, performing peak shaving and point selection for non-urgent tasks with soft time windows. The post-processing smoothing repair module is used to perform iterative repair on the scheduling results after the initial scheduling results are output.

[0029] The constraint consistency correction module is used to ensure that the repair results meet the consistency requirements of destructive post-rules and capacity constraints; The output module is used to output the repaired scheduling results and process records / logs.

[0030] To verify the effectiveness of the Peak Constraint Collaborative Resource Repair (PCSR) method described in this invention, the repaired scheduling results obtained using PCSR are compared and analyzed with the baseline scheduling results without repair, under the same initial scheduling conditions and constraints. The comparison indicators include resource overrun levels, company-level daily peak load, scheduling feasibility indicators, and load balancing indicators, all of which are statistically analyzed based on the baseline results.

[0031] Regarding key indicators, such as Figure 3 As shown, the resource overrun rate and company-level daily peak load corresponding to PCSR show changes of approximately 47.82% and 63.2% respectively compared to the baseline results. The resource overrun rate characterizes the degree to which resource occupancy exceeds its available capacity limit or available time window; the company-level daily peak load characterizes the maximum load level at the company level on a daily scale. These comparative results indicate that PCSR can regulate the distribution of resource pressure during high-load periods, compressing peak pressure and reducing the occurrence of resource overruns.

[0032] In terms of feasibility and balance, such as Figure 3As shown, the scheduling success rate and worst-case department success rate corresponding to PCSR changed by approximately 1.38 percentage points and 1.40 percentage points respectively compared to the baseline results; the load Gini coefficient changed by approximately 0.61% compared to the baseline results. These results indicate that, while controlling peak load, PCSR did not introduce additional infeasibility risks, and the balance of resource load distribution was further improved.

[0033] Regarding peak control at the departmental level, such as Figure 4 As shown, a comparison of the peak values ​​per unit in each department reveals that different departments exhibit varying degrees of change in peak values ​​after adopting PCSR. Taking department H as an example, its peak value per unit decreased from approximately 19 units / day at baseline to approximately 5 units / day; departments C, D, and E also showed corresponding changes, indicating that PCSR has portability and adaptability in multi-department, multi-resource-constrained environments.

[0034] Regarding the load fluctuation characteristics in the time dimension, such as Figure 5 As shown, the overall fluctuation amplitude of the half-day peak load time-series curve at the company level decreased after the adoption of PCSR, and the peak spikes were suppressed, indicating that PCSR can adjust local congestion in the time axis dimension, making the load change trend more stable.

[0035] Further detailed analysis was conducted by selecting typical departments, such as Figure 6 As shown, department H exhibited multiple peak surges and fluctuations under baseline conditions. However, after implementing PCSR, its half-day load time-series curve became smoother, peak levels were suppressed, and no new abnormal peaks appeared. This comparative result verifies, from the perspective of a typical department, the ability of PCSR to maintain the stability of the scheduling structure when implementing local repairs.

[0036] This application also provides an electronic device, including at least one processor and at least one memory; the memory stores program instructions, which, when executed by the processor, are used to implement the above-described peak constraint collaborative resource repair scheduling method.

[0037] This application also provides a computer-readable storage medium storing program instructions thereon, which, when executed by a processor, are used to implement the above-described peak constraint cooperative resource repair scheduling method.

[0038] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A peak-constrained collaborative resource repair scheduling method, characterized in that the steps include: include: Step 1: Data Input. Input the three types of information: task data, constraint data, and parameter configuration. Step 2: Initial schedule generation. The initial schedule result is generated based on task data, constraint data, and parameter configuration. Step 3 is collaborative resource repair, which repairs the results of the initial scheduling. Collaborative resource repair includes peak constraint landing point selection and post-processing smoothing repair. Step 4: Constraint Consistency Correction. Perform post-rule consistency correction on the scheduling results after collaborative resource repair, and perform capacity feasibility verification. Step 5 outputs the final scheduling result after the above repairs, and also outputs the process record or log.

2. The peak-constrained collaborative resource repair and scheduling method according to claim 1, characterized in that, The specific steps of the initial scheduling in step 2 are as follows: First, a sample-level processing order is established, which can be based on the comprehensive score to rank the samples. Subsequently, priority scheduling for urgent tasks is performed, with the set of urgent tasks occupying resources first and completing capacity feasibility checks; if an urgent task is not feasible, the corresponding set of failed urgent tasks is output; after the urgent task scheduling is successful, the scheduling of non-urgent tasks continues. Finally, the schedule is checked for capacity feasibility on the discrete time axis. The check principle is that the occupancy of any department in any discrete time slice does not exceed its capacity limit, thus obtaining the initial scheduling result.

3. The peak-constrained collaborative resource repair and scheduling method according to claim 2, characterized in that, The peak constraint landing point selection in step 3 specifically includes three stages: feasible window construction, feasible starting point enumeration and peak prediction, and cost evaluation and optimal landing point determination. The construction of the feasible window first determines the range of selectable starting points based on the mapping relationship between the hard and soft time windows of the task, and then performs a trigger determination. The feasible starting point enumeration and peak prediction enumerate candidate starting points within the range of selectable starting points, and perform capacity constraint filtering during the enumeration process to retain only feasible starting points; For each feasible starting point, the peak occupancy of the time period after the task is placed is predicted according to the preset time granularity, and the risk amount such as the half-day overdue amount relative to the original end time is calculated. The cost assessment and optimal landing point determination are then carried out. Subsequently, the cost is constructed and the starting point with the optimal cost is selected as the final landing point. Finally, the landing point comparison record is output. The comparison record includes at least the sample identifier, department identifier, window boundary, baseline landing point, optimal landing point, and the corresponding peak value and overdue amount.

4. The peak-constrained collaborative resource repair and scheduling method according to claim 3, characterized in that, The post-processing smoothing repair in step 3 specifically includes the following steps: S1 takes the initial scheduling results and the sorted sequence of the comprehensive scores as input; S2 performs time-segment discretization modeling, dividing the time axis into discrete time segments, and aggregates task counts for departmental load in the form of "department × time slice" to obtain a comparable load distribution representation. S3 identifies the segment with the highest load; S4 determines the location of the period when the current pressure is most concentrated, and selects a set of candidate samples around this period, forming a protection set based on the candidate set; The S5 protection set is determined based on the scoring ranking. TopK scoring samples are not included in the scoring downweighting process to stabilize the ranking structure of key samples. Trial scoring downweighting is performed on non-protected samples. S6 reorders the trial sequences according to the updated scores; S7 then calls the scheduling calculation based on the trial sequence to generate a new scheduling result; For each trial result, S8 first performs a feasibility assessment. If the trial violates capacity feasibility or other hard constraints, it rolls back and continues to the next round of trials. S9 When the trial results are feasible, further evaluation of indicators should be carried out. The indicators should include at least peak level, failure rate and overdue rate, and the acceptance criteria should be determined based on the allowable incremental threshold and peak reduction requirements. If the acceptance criteria are not met in S10, then backtrack and continue with new trial adjustments; If the acceptance criteria are met in S11, the current scheduling result is updated and used as a new baseline for the next round of identification and exploration. S12 executes S3 through S11 in a loop until the termination condition is met, including reaching the maximum number of iterations, insufficient continuous improvement, or no acceptable solution. S13 finally outputs the repaired scheduling results and the repair process record. The repair process record records at least the peak position identified in each round, the size of the candidate set, the number of attempts, the number of acceptances / backoffs, and the changes in key indicators, for verification and reproduction.

5. The peak-constrained collaborative resource repair scheduling method according to claim 1, characterized in that, The constraint consistency correction in step 4 is used to ensure that the repair results meet the consistency requirements of destructive post-rules and capacity constraints. This step uses post-rules to be executed at a half-day granularity, which can realize the constraint requirements of different half-days on the same day or half-days on the next day. When the post-correction causes the original landing point to no longer meet the rules, a feasible starting point is re-determined within the allowable range, and the capacity feasibility check continues to be performed, thereby ensuring that the destructive post-rules and capacity constraints are met at the same time, and avoiding the introduction of infeasibility risks during the repair process.

6. The peak-constrained collaborative resource repair scheduling method according to claim 1, characterized in that, The output step described in step 5 is used to output the repaired scheduling results and process records / logs. The scheduling results include at least the task start time, end time, department allocation information, overdue mark, and failure sample set; the process records include at least the landing point comparison log and the post-processing smooth repair iteration log.

7. A peak-constrained collaborative resource repair and scheduling system, characterized in that, include: The input module is used to collect three types of information: task data, constraint data, and parameter configuration. Task data includes at least the sample, department, start and end times of the time window, task cycle, priority, number of samples, and destructive flag; constraint data includes at least the department capacity limit, soft time window mapping relationship, and destructive post-rule; parameter configuration includes at least the time granularity, weight parameters, allowable incremental threshold, termination condition, and log switch. The initial scheduling generation module is used to obtain feasible initial scheduling results; The collaborative resource repair module is used to repair feasible initial ranking results, including peak constraint landing point selection and post-processing smoothing repair. The constraint consistency correction module is used to ensure that the repair results meet the consistency requirements of destructive post-rules and capacity constraints; The output module is used to output the repaired scheduling results and process records / logs.

8. A peak-constrained collaborative resource repair and scheduling system according to claim 7, characterized in that, The collaborative resource repair module includes a peak constraint landing point selection module and a post-processing smoothing repair module. The peak constraint landing point selection module operates in the task landing point determination stage and performs peak shaving and point selection for non-urgent tasks with soft time windows. The post-processing smoothing repair module is used to perform iterative repair after the initial scheduling results are output.

9. An electronic device, characterized in that, It includes one or more processors, which execute the peak-constrained collaborative resource repair scheduling method as described in any one of claims 1 to 6.

10. A computer storage medium, characterized in that the computer-readable storage medium stores computer instructions for causing a computer to execute the peak-constrained cooperative resource repair scheduling method as described in any one of claims 1 to 6.

Images