Intelligent scheduling method and system based on deep learning

Abstract

The invention provides an intelligent scheduling method and system based on deep learning, and belongs to the technical field of big data processing. According to the method, the to-be-decided variable employees are accurately determined by adopting the deep learning algorithm, and the candidate shifts corresponding to the highest comprehensive value score are allocated to the to-be-decided variable employees, so that accurate adaptation of employee skills and post requirements is realized, and the balance of employee workloads is fully considered. The scheduling constraint is verified and subjected to conflict resolution by adopting an arc consistency propagation mechanism, so that the constraint problem in the scheduling process can be identified in real time and adjusted in time, the efficient determination of the scheduling search path and the accurate evaluation of the candidate shift value are realized, the scheduling calculation speed and the scheduling scheme quality are improved, and the scheduling efficiency is improved. The scheduling requirements under large-scale personnel and numerous and jumbled constraint conditions are effectively met, and the integrity and feasibility of the whole scheduling scheme are greatly improved.

Description

Technical Field

[0001] This application relates to the field of big data processing technology, specifically to an intelligent scheduling method and system based on deep learning. Background Technology

[0002] Currently, intelligent scheduling plays an indispensable role in modern enterprise management as a crucial means to improve operational efficiency and employee satisfaction. This is particularly true in industries such as healthcare, retail, and logistics, where the rational allocation of employee work hours and tasks directly impacts service quality and cost control. However, existing scheduling methods often struggle to cope with dynamic demands from large-scale workforces and under multiple constraints. When faced with differences in employee skills, unexpected adjustments, and diverse work rules, scheduling efficiency suffers, resulting in poor outcomes. Especially in scheduling environments requiring rapid responses to changes, existing methods often fail to balance calculation speed and scheduling quality, impacting business operations.

[0003] The information provided in the background section of this application is only for enhancing the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] In view of this, this application provides a deep learning-based intelligent scheduling method and system that can improve the efficiency and stability of scheduling.

[0005] In a first aspect, embodiments of this application provide a deep learning-based intelligent scheduling method, which includes: acquiring an employee profile database and a business task plan table; determining the skill tag set and fatigue accumulation value of each employee within a scheduling cycle based on the employee profile database, and determining a daily skill requirement matrix for each position based on the business task plan table; determining the matching similarity score between each employee and each shift within the scheduling cycle based on the skill tag set and the daily skill requirement matrix for each position; determining an employee variable ranking sequence based on the matching similarity score and the fatigue accumulation value, and using a deep learning algorithm to determine the employee to be decided variable based on the employee variable ranking sequence; acquiring the employee to be decided variable... The decision variable employee is assigned a first candidate shift, and the comprehensive value score of each first candidate shift is determined. The first candidate shift with the highest comprehensive value score is assigned to the employee. The second candidate shifts associated with the employee are updated based on the first candidate shifts. An arc consistency propagation mechanism is used to determine the third candidate shifts for employees who are constrained by the current allocation of the second candidate shifts. A global scheduling scheme is generated based on the third candidate shifts. If at least one employee is not assigned a shift for any date within the scheduling period, the decision variable employee is updated again using a deep learning algorithm.

[0006] Optionally, determining the skill tag set and fatigue accumulation value of each employee within the scheduling cycle based on the employee file database includes:

[0007] In the employee file database, the skill tags and historical work records of each employee within the scheduling cycle are obtained, and the skill tag set is determined based on the skill tags; Obtain a preset shift duration table, and determine the cumulative working hours based on the historical work records and the preset shift duration table; The cumulative fatigue value is determined based on the cumulative working time.

[0008] Optionally, after determining the matching similarity score between each employee and each shift within the scheduling cycle based on the skill tag set and the daily skill requirement matrix for the job, the method further includes: If the fatigue accumulation value exceeds the preset fatigue accumulation value threshold, then the fatigue decay coefficient is obtained, and the fatigue decay coefficient is multiplied by the current matching similarity score to adjust the matching similarity score; The number of consecutive working days and rest intervals for each employee are determined based on the aforementioned historical work records; If the current employee's consecutive working days exceed the preset consecutive working days threshold or the rest interval does not exceed the preset rest interval threshold, then the matching similarity score between the current employee and each shift in the scheduling cycle is set to zero.

[0009] Optionally, determining the employee variable ranking sequence based on the matching similarity score and the fatigue accumulation value includes: The matching similarity scores are sorted from the maximum to the minimum to generate the sorted sequence of the employee variables; If the difference between the matching similarity scores of adjacent employees in the employee variable sorting sequence does not exceed the preset similarity difference threshold, then the fatigue accumulation value is sorted from the minimum value to the maximum value to update the employee variable sorting sequence.

[0010] Optionally, the step of determining the employee variable to be decided using a deep learning algorithm based on the employee variable sorting sequence includes: Based on the employee variable sorting sequence, a deep learning algorithm is used to determine the conflict value of each employee, and the employee with the highest conflict value is identified as the conflicting employee. Filter each unassigned employee in the sorted sequence of employee variables, and use a weighted algorithm to determine the closeness between the conflicting employees and each of the unassigned employees; The unassigned employee with the highest affinity to the conflicting employee is selected as the employee to be decided.

[0011] Optionally, determining the comprehensive value score of each of the first candidate train services includes: Based on each of the first candidate shifts, determine the contribution value of each of the first candidate shifts to the slackness of the remaining shifts, the impact value of each of the first candidate shifts on the increase of the cumulative fatigue value of employees, and the estimated value of each of the first candidate shifts on the average number of available shifts on unassigned days. Obtain the weights corresponding to the contribution value, the influence value, and the estimated value, respectively. Based on the weights corresponding to the contribution value, the influence value, and the estimated value, a weighted calculation is performed on the contribution value, the influence value, and the estimated value, and the result of the weighted calculation is used as the comprehensive value score of each first candidate class.

[0012] Optionally, the step of using the arc consistency propagation mechanism to determine each third candidate shift that has constraints on employees assigned to each of the current second candidate shifts includes: The arc consistency propagation mechanism is used to obtain the influence range of employees who are constrained by the current allocation of each second candidate shift. Update the cumulative fatigue value and the number of consecutive working days for each employee according to the scope of the impact. Based on the updated accumulated fatigue value and the updated number of consecutive working days, determine the third candidate shifts for each employee with constraints in the current allocation of each second candidate shift.

[0013] Optionally, after determining the third candidate shifts for each of the current second candidate shifts that have employees with constraints, the method further includes: If the value range of each of the third candidate shifts is empty, the arc consistency propagation mechanism is used again to update the current scope of influence in order to adjust the third candidate shifts.

[0014] Optionally, after generating the global scheduling scheme based on each of the third candidate shifts, the method further includes: If there are employees on leave, then obtain the affected date range and the shift of the employee on leave; Within the affected date range, calculate the matching similarity score between the remaining employees (excluding the employee on leave) and the shift of the employee on leave, and the cumulative fatigue value of the remaining employees; The global scheduling scheme is adjusted based on the matching similarity score between the shifts of the remaining employees and the shifts of the employees on leave, and the cumulative fatigue value of the remaining employees.

[0015] Secondly, embodiments of this application provide a deep learning-based intelligent scheduling system, comprising: an acquisition module, a first determination module, a second determination module, a third determination module, a fourth determination module, an update module, and a generation module. The acquisition module acquires an employee file database and a business task plan table; the first determination module determines the skill tag set and fatigue accumulation value of each employee within the scheduling cycle based on the employee file database, and determines a daily skill requirement matrix for each position based on the business task plan table; the second determination module determines the matching similarity score between each employee and each shift within the scheduling cycle based on the skill tag set and the daily skill requirement matrix for each position; the third determination module determines an employee variable ranking sequence based on the matching similarity score and the fatigue accumulation value, and uses a deep learning algorithm to determine the employee to be decided variable based on the employee variable ranking sequence; the fourth determination module acquires the employee to be decided variable... The system comprises: a first candidate shift for each employee; a comprehensive value score for each first candidate shift; and an allocation of the first candidate shift with the highest comprehensive value score to the employee to be decided. An update module updates the second candidate shifts for employees associated with the employee to be decided based on the first candidate shifts, and uses an arc consistency propagation mechanism to determine the third candidate shifts for employees constrained by the current allocation of the second candidate shifts. A generation module generates a global scheduling scheme based on the third candidate shifts, and in the global scheduling scheme, if at least one employee has not received a shift allocation for any date within the scheduling period, the employee to be decided is updated again using a deep learning algorithm.

[0016] This application provides a deep learning-based intelligent scheduling method and system. By combining employee skill tags, job skill requirement matrices, and employee fatigue accumulation values ​​to construct an employee variable ranking sequence, and employing a deep learning algorithm to accurately determine the employees to be decided, the candidate shifts corresponding to the highest comprehensive value scores are assigned to these employees. This achieves both precise matching of employee skills with job requirements and a balanced workload for employees. By using an arc consistency propagation mechanism to verify and resolve scheduling constraints, constraints in the scheduling process can be identified and adjusted in real time, achieving efficient determination of scheduling search paths and accurate evaluation of candidate shift values, thus improving scheduling calculation speed and scheduling plan quality. Through iterative generation of a global scheduling plan, it ensures that all employees receive reasonable shift allocations on each date within the scheduling cycle, forming a complete scheduling plan that satisfies multiple constraints such as skill matching, fatigue balance, and rest intervals. This effectively addresses scheduling needs under large-scale personnel and complex constraints, significantly improving the completeness and feasibility of the overall scheduling plan. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the conventional technology, the drawings used in the description of the embodiments or the conventional technology will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an exemplary embodiment of the intelligent scheduling method based on deep learning provided in this application. Figure 2 This is a flowchart illustrating a deep learning-based intelligent scheduling method provided in another exemplary embodiment of this application; Figure 3 This is a flowchart illustrating a deep learning-based intelligent scheduling method provided in another exemplary embodiment of this application; Figure 4 This is a flowchart illustrating a deep learning-based intelligent scheduling method provided in yet another exemplary embodiment of this application. Figure 5 This is a flowchart illustrating a deep learning-based intelligent scheduling method provided in yet another exemplary embodiment of this application. Figure 6 This is a flowchart illustrating a deep learning-based intelligent scheduling method provided in yet another exemplary embodiment of this application. Figure 7 This is a flowchart illustrating a deep learning-based intelligent scheduling method provided in yet another exemplary embodiment of this application. Figure 8 This is a flowchart illustrating a deep learning-based intelligent scheduling method provided in another exemplary embodiment of this application. Detailed Implementation

[0019] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that this application will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this application.

[0020] The terms “a,” “one,” and “the” are used to indicate the existence of one or more elements / components / etc.; the terms “including” and “having” are used to indicate an open-ended inclusion and that other elements / components / etc. may exist in addition to those listed. The terms “first” and “second” are used only as markers and are not a limitation on the number of objects.

[0021] Currently, intelligent scheduling plays an indispensable role in modern enterprise management as a crucial means to improve operational efficiency and employee satisfaction. This is particularly true in industries such as healthcare, retail, and logistics, where the rational allocation of employee work hours and tasks directly impacts service quality and cost control. However, existing scheduling methods often struggle to cope with dynamic demands from large-scale workforces and under multiple constraints. When faced with differences in employee skills, unexpected adjustments, and diverse work rules, scheduling efficiency suffers, resulting in poor outcomes. Especially in scheduling environments requiring rapid responses to changes, existing methods often fail to balance calculation speed and scheduling quality, impacting resource allocation effectiveness and employee work experience in actual operations.

[0022] Specifically, the core challenge lies in quickly finding a near-ideal scheduling plan under complex constraints. The selection of candidate shift paths is the primary difficulty, as blindly sifting through numerous possible combinations is time-consuming, while ignoring key candidate paths may lead to missing out on high-quality scheduling solutions. Furthermore, without accurately assessing the potential value of each candidate shift path, it's difficult to focus on the most promising scheduling plan within a limited timeframe. For example, in a large hospital, nurse scheduling needs to consider skill matching, shift fairness, and unforeseen leave adjustments. Failure to quickly identify reasonable candidate shift directions could result in unfair scheduling outcomes and even affect the continuity of medical services.

[0023] Therefore, how to efficiently determine the search scheduling path and accurately evaluate its usability in a complex scheduling environment has become a technical problem that this application urgently needs to solve.

[0024] This application provides a deep learning-based intelligent scheduling method, such as... Figure 1 The diagram illustrates a deep learning-based intelligent scheduling method. This method may include the following steps: Step S110: Obtain the employee file database and business task plan table; Step S120: Determine the skill tag set and fatigue accumulation value of each employee within the scheduling cycle based on the employee file database, and determine the daily skill requirement matrix for each position based on the business task plan. Step S130: Determine the matching similarity score between each employee and each shift within the scheduling cycle based on the skill tag set and the daily skill requirement matrix for the job. Step S140: Determine the employee variable ranking sequence based on the matching similarity score and fatigue accumulation value, and use a deep learning algorithm to determine the employee variable to be decided based on the employee variable ranking sequence; Step S150: Obtain the first candidate shifts corresponding to the employees of the decision variable, determine the comprehensive value score of each first candidate shift, and assign the first candidate shift with the highest comprehensive value score to the employees of the decision variable. Step S160: Update the second candidate shifts for employees associated with the decision variable based on the first candidate shifts, and use the arc consistency propagation mechanism to determine the third candidate shifts for employees who are constrained by the current allocation of the second candidate shifts. Step S170: Generate a global scheduling plan based on each third candidate shift. In the global scheduling plan, if at least one employee is not assigned a shift on any date within the scheduling cycle, the decision variable employee is updated again using a deep learning algorithm.

[0025] The intelligent scheduling method based on deep learning provided in this application constructs a ranking sequence of employee variables by combining employee skill tags, job skill requirement matrices, and employee fatigue accumulation values. It then uses a deep learning algorithm to accurately determine the employees to be decided, assigning the candidate shifts with the highest comprehensive value scores to these employees. This achieves both precise matching of employee skills with job requirements and a balanced workload for all employees. By employing an arc consistency propagation mechanism to verify and resolve scheduling constraints, it can identify and adjust constraints in the scheduling process in real time, achieving efficient determination of scheduling search paths and accurate evaluation of candidate shift values, thus improving scheduling calculation speed and scheduling plan quality. Through iterative generation of a global scheduling plan, it ensures that all employees receive reasonable shift allocations on each date within the scheduling cycle, forming a complete scheduling plan that satisfies multiple constraints such as skill matching, fatigue balance, and rest intervals. This effectively addresses scheduling needs under large-scale personnel and complex constraints, significantly improving the completeness and feasibility of the overall scheduling plan.

[0026] The steps of the intelligent scheduling method based on deep learning provided in the embodiments of this application are described in detail below: In one embodiment of this application, step S110 involves obtaining an employee file database and a business task plan table. Specifically, the employee file database is a structured database used in the human resource management system of an enterprise or hospital to centrally store comprehensive human resource information of employees. When obtaining this database, the intelligent scheduling system can be connected to the general data interface of the human resource management system, and big data processing technology can be used to filter and extract the complete file data of all employees participating in the current scheduling according to the scheduling cycle. This database is a relational structured database, including core fields that can be searched and calculated, such as employee unique identifier ID, employee skill tag set, employee historical continuous work records, employee daily attendance data, and employee job basic suitability information. The business task plan table is a phased production task planning table formulated in the enterprise management system based on actual production factors such as market order demand, enterprise production capacity, and material supply progress. When obtaining this plan table, the intelligent scheduling system is connected to the production planning module data interface of the enterprise production management system to extract the structured data of the business task plan table that perfectly matches the current scheduling cycle. The business task plan table is stored in a tabular, structured format, including core fields such as shift date, job title, daily skill requirements for the job, fixed time slots for each shift, staffing requirements for each shift, and job task execution standards. Furthermore, the business task plan table can also be a phased medical treatment task plan table developed within the hospital's operations management system based on actual medical business elements such as outpatient needs, inpatient treatment needs, surgical scheduling, emergency care requirements, and medical quality control standards. When obtaining this plan table, the structured data of the business task plan table that perfectly matches the current shift schedule is extracted through the data interface of the medical task planning module of the hospital's HIS system, department operations management system, outpatient management system, surgical anesthesia system, and emergency management system. The business task plan table is stored in a tabular, structured format, including core fields such as shift date, medical job title, daily skill requirements for the job, fixed time slots for each shift, staffing requirements for each shift, and job task execution standards.

[0027] In one embodiment of this application, step S120, which determines the skill tag set and fatigue accumulation value of each employee within the scheduling cycle based on the employee file database, further includes the following steps: Figure 2 As shown, the specific content is as follows: Step S210: In the employee file database, obtain the skill tags and historical work records of each employee within the scheduling cycle, and determine the skill tag set based on the skill tags; Step S220: Obtain the preset shift duration table, and determine the cumulative working hours based on historical work records and the preset shift duration table; Step S230: Determine the cumulative fatigue value based on the cumulative working time.

[0028] Specifically, taking the shift schedule of a manufacturing company from April 1st to April 14th as an example, the company's human resource management system's employee file database selects 38 front-line production workshop operators participating in this shift. The original skill tag data and historical work record data from March 1st to March 31st for these 38 employees are extracted in batches. Employee A101's skill tag data includes injection molding machine operation, quality inspection, forklift license, and basic maintenance; therefore, employee A101's skill tag set is {injection molding machine operation, quality inspection, forklift license, basic maintenance}. Employee B203's skill tag data includes packaging, forklift license, and injection molding machine operation; therefore, employee B203's skill tag set is also {packaging, forklift license, injection molding machine operation}. Next, the company's preset shift duration table is retrieved. This preset shift duration table is pre-set according to the company's labor employment system and job hour arrangements, stored in the company's human resource management system, and reflects the standard working hours corresponding to different shifts. The table specifies that the morning shift is from 8:00 to 16:30, with a standard duration of 8.5 hours; the afternoon shift is from 14:00 to 22:00, with a standard duration of 8 hours; and the evening shift is from 22:00 to 6:00 the next day, with a standard duration of 7.5 hours. Overtime is calculated based on the standard duration when there is no overtime; overtime hours are accumulated based on actual attendance data. The starting date of this scheduling cycle is April 1st, therefore the statistical period is determined to be from March 23rd to April 1st. The extracted employee historical work records are filtered within this time range, and the accumulated working hours are calculated by matching them against the preset shift duration table. Employee A101 worked no rest or overtime during the statistical period. Their daily shifts were: March 26 (early shift), March 27 (early shift), March 28 (afternoon shift), March 29 (early shift), March 30 (afternoon shift), March 31 (early shift), and April 1 (early shift). The total working hours were calculated as follows: 8.5 + 8.5 + 8 + 8.5 + 8 + 8.5 + 8.5 = 58.5 hours. Therefore, employee A101's total working hours were 58.5 hours. Employee B203 had 2 days off and no overtime during the statistical period. The daily shifts were as follows: March 26 (rest), March 27 (middle shift), March 28 (middle shift), March 29 (rest), March 30 (night shift), March 31 (middle shift), and April 1 (middle shift). Calculated by adding up the standard working hours, the total is: 8 + 8 + 7.5 + 8 + 8 = 39.5 hours. That is, employee B203's cumulative working hours are 39.5 hours.

[0029] The preset fatigue accumulation value calculation model is: Fatigue accumulation value = Cumulative working hours / (Statistical period days × Average daily standard working hours per person). In this embodiment, the statistical period is 7 days, and the average daily standard working hours per person is 8 hours. Therefore, the average total standard working hours per person within the statistical period is 7 × 8 = 56 hours. The average daily standard working hours per person are preset according to the standard working hour system implemented by the enterprise, for example, determined according to the 8-hour work system. The number of days in the statistical period can be set according to the sensitive period of fatigue accumulation, for example, using the same length as the shift scheduling cycle (such as a 7-day weekly shift schedule), or determined according to the restrictions on continuous working time in labor regulations (such as no more than 40 hours per week). Alternatively, a time window that reflects the trend of employee fatigue accumulation (such as 7 days or 30 days) can be selected based on historical data statistical analysis. Substituting the cumulative working hours obtained in step S220 into the model, the fatigue accumulation value of employee A101 is calculated as follows: Employee A101's fatigue accumulation value = 58.5 / 56 = 1.04. The fatigue cumulative value for employee B203 is calculated to be 39.5 / 56 = 0.7. Similarly, the fatigue cumulative values ​​for all 38 employees are calculated.

[0030] Optionally, taking the bi-weekly shift schedule of the cardiology department of a tertiary hospital as an example, the 12 clinical doctors in the department with independent practice qualifications participate in this schedule. Skill tags and historical work records of all participating doctors are extracted from the employee file database within the scheduling period, and the skill tag sets are standardized and aggregated. Specifically, Dr. Zhang's skill tag set is: {Cardiovascular Medicine Practitioner, Level 4 Coronary Interventional Surgery Authorization, Acute Myocardial Infarction Emergency Treatment Qualification, CCU Intensive Care Qualification}; Dr. Li's skill tag set is: {Cardiovascular Medicine Practitioner, General Outpatient Clinic Qualification, Inpatient Bed Management Qualification}; and Dr. Wang's skill tag set is: {Cardiovascular Medicine Practitioner, Emergency Internal Medicine Clinic Qualification, CCU Intensive Care Qualification}. Simultaneously, the historical work records of these three doctors within the corresponding statistical period are extracted. Subsequently, the hospital retrieved its pre-set shift schedule, which was tailored to industry standards and clinical work characteristics. This schedule specifies a standard working time of 8 hours for general outpatient day shifts and inpatient ward day shifts, 10 hours for emergency room night shifts, and 24 hours for CCU on-call shifts. Using the start date of this shift cycle as a baseline, a 7-day historical work record statistical period was determined. The working hours of the three doctors were calculated daily and accumulated against the pre-set shift schedule. The results showed that Doctor Zhang had accumulated 72 hours, Doctor Li 40 hours, and Doctor Wang 66 hours. Finally, a pre-set fatigue accumulation value calculation model was used, with a 7-day statistical period and an average daily standard working time of 8 hours per doctor as the calculation benchmark. The cumulative working hours of the three doctors were substituted into the model for normalization calculation, resulting in a fatigue accumulation value of approximately 1.29 for Doctor Zhang, approximately 0.71 for Doctor Li, and approximately 1.18 for Doctor Wang.

[0031] In one embodiment of this application, in step S120, a daily skill requirement matrix for each position is determined based on the business task plan. Specifically, for the obtained business task plan, all structured production data for the corresponding time period is filtered out according to a preset scheduling cycle. The extracted fields include the specific date of the scheduling, the type of core production position, the fixed division standard for each shift, the type of skill requirement required for each position to complete the business task, the importance of different skills in the position, and the basic personnel configuration indicators for each position in each shift. Based on the enterprise's production management specifications and the priority of the current production tasks, different skill requirements for each position under each date and shift are quantitatively assigned a value between 0 and 1. The key skills required for the position to complete the core production task are assigned a high value, the supporting skills for auxiliary production are assigned a relatively low value, and the dimension with no skill requirement is assigned a value of 0. This ensures that the quantitative assignment results are consistent with the actual skill requirement intensity of the production positions in each time period, avoiding a disconnect between skill requirements and production tasks. Finally, using the scheduling date + shift as the row dimension of the matrix and the unified set of all employees' skills as the column dimension, the quantified skill requirement values ​​for each date, shift, and position are filled into the corresponding row and column positions of the matrix to form a structured and standardized daily skill requirement matrix for each position.

[0032] For example, taking the shift schedule of a manufacturing company from March 1st to March 14th as an example, production demand data for the corresponding production workshop within the period is extracted from the company's business task plan. The core production positions are injection molding and packaging, with shifts divided into morning shift (8:00-16:30), afternoon shift (14:00-22:00), and evening shift (22:00-6:00 the next day). The extracted job skill requirements include injection molding machine operation, quality inspection, forklift license, and packaging. After aligning the dimensions with the union of all employee skills based on 12 skill tags compiled from the database of 38 employee files, the column dimensions of the daily job skill requirement matrix are determined to be these 12 skill tags. Based on the current production tasks, the job skill requirements for each date and shift are quantified and assigned values. For example, the core production tasks for the injection molding position on the morning shift of March 1st are finished product injection molding and initial inspection. Therefore, injection molding machine operation is assigned a value of 1.0, quality inspection 0.8, forklift license 0.6, and packaging 0.4, while the other 8 auxiliary skills are assigned a value of 0. The production tasks for the injection molding position on the evening shift of March 1st are injection molding of semi-finished products and their packaging. Therefore, injection molding machine operation is assigned a value of 0.9, packaging 1.0, forklift license 0.7, and the remaining skills are assigned a value of 0. The core tasks for the packaging position on the afternoon shift of March 1st are finished product packaging and transfer. Therefore, packaging is assigned a value of 1.0, forklift license 0.8, and the remaining skills are assigned a value of 0. After quantifying and assigning skill requirements for all dates, 3 shifts, and all core production positions within the 14-day scheduling cycle in this manner, a row dimension of 42 rows is formed by "14 days × 3 shifts", and a column dimension of 12 columns is formed by 12 skill tags. All quantified values ​​are then filled into the corresponding positions to generate the daily skill requirement matrix for the production workshop positions of the manufacturing enterprise.

[0033] For example, taking the bi-weekly shift schedule of the internal medicine department of a top-tier hospital as an example, 10 clinical doctors participate in this shift. Skill tags and historical work records for each doctor are extracted from the employee file database to form a skill tag set. For instance, Doctor A's skill tag set is {Internal Medicine Practitioner, General Outpatient Clinic Qualification, Acute Disease Emergency Treatment Qualification}, and Doctor B's skill tag set is {Internal Medicine Practitioner, Inpatient Bed Management Qualification, Intensive Care Qualification}. Then, the preset doctor shift duration table is retrieved, and the cumulative working hours are calculated based on the historical work records of the previous 7 days of the shift cycle. Doctor A's cumulative working hours are calculated to be 70 hours, and Doctor B's to be 45 hours. Then, using the fatigue accumulation value calculation formula, based on a 7-day statistical period and an average daily standard working hour of 8 hours, the fatigue accumulation value is approximately 1.25 for Doctor A and approximately 0.80 for Doctor B. Simultaneously, the core medical positions such as general outpatient, inpatient ward management, and emergency positions for this scheduling are extracted from the business task plan table, along with the shift division information for day and night shifts. Combined with clinical treatment guidelines, the skill requirements for each position on each date and shift are quantitatively assigned in the range of 0-1. For example, the general outpatient day shift assigns a value of 1.0 to the general outpatient reception qualification, and the emergency night shift assigns a value of 1.0 to the emergency treatment qualification for acute illnesses. Finally, using the 28 rows formed by the 14-day × 2-shift scheduling cycle as the row dimension, and the union of the 8 summarized all-staff physician skill tags as the column dimension, all quantitative assignment results are accurately filled into the corresponding positions, ultimately constructing the daily skill requirement matrix for the positions in this scheduling cycle of the department.

[0034] In one embodiment of this application, in step S130, the matching similarity score between each employee and each shift within the scheduling cycle is determined based on the skill tag set and the daily job skill requirement matrix. Specifically, the vectors of the constructed skill tag sets for each employee and the job skill requirement vectors for each date and shift in the generated daily job skill requirement matrix are retrieved, ensuring that the two types of vectors are of the same dimension. Cosine similarity is calculated for each employee's skill tag vector and the corresponding shift's job skill requirement vector, using date and shift as units. This is done in batches through big data processing. The calculation result reflects the basic matching degree between the employee's skills and the job shift skill requirements, with a value ranging from 0 to 1. A higher value indicates a higher matching degree between the employee's skills and the job shift requirements.

[0035] In one embodiment of this application, after determining the matching similarity score between each employee and each shift within the scheduling cycle based on the skill tag set and the daily skill requirement matrix for the job in step S130, the following steps are also included: Figure 3 As shown, the specific content is as follows: Step S310: If the fatigue accumulation value exceeds the preset fatigue accumulation value threshold, obtain the fatigue decay coefficient and multiply the fatigue decay coefficient by the current matching similarity score to adjust the matching similarity score; Step S320: Determine the number of consecutive working days and rest intervals for each employee based on historical work records; Step S330: If the current employee's consecutive working days exceed the preset consecutive working days threshold or the rest interval does not exceed the preset rest interval threshold, then set the matching similarity score between the current employee and each shift in the scheduling cycle to zero.

[0036] Specifically, taking the work schedule of a manufacturing company from March 1st to March 14th as an example, the following parameters were set: a preset fatigue accumulation threshold of 0.8, a fatigue attenuation coefficient of 0.9, a preset continuous working days threshold of 6 days, and a preset rest interval threshold of 12 hours. The matching similarity scores and fatigue accumulation statistics for 38 production workshop employees have been completed. The preset fatigue accumulation threshold is used to determine whether an employee's fatigue level is too high. If it exceeds this value, the matching degree needs to be reduced using the attenuation coefficient to prevent highly fatigued employees from being prioritized for scheduling. This threshold can be determined by the company's human resources department based on employee health monitoring data or expert experience. The preset continuous working days threshold is set according to the company's continuous work system, such as a scheduling rule that requires rest after no more than 6 consecutive days of work. The preset rest interval threshold is set according to the minimum rest time requirements for adjacent shifts in labor regulations, such as a rest interval of no less than 12 hours. Among them, employee A101 has a similarity score of 0.7477 and a cumulative fatigue value of 0.973; employee B203 has a similarity score of 0.7821 and a cumulative fatigue value of 0.679; employee C105 has a similarity score of 0.812 and a cumulative fatigue value of 0.72; and employee E302 has a similarity score of 0.765 and a cumulative fatigue value of 0.75. The cumulative fatigue value of each employee is compared with the preset fatigue cumulative value threshold of 0.8. Employee A101's cumulative fatigue value of 0.973 > 0.8, triggering a decay adjustment. The fatigue decay coefficient is multiplied by the current similarity score, resulting in an adjusted score of 0.7477 × 0.9 = 0.67293. The cumulative fatigue values ​​of employees B203, C105, and E302 do not exceed the preset fatigue cumulative value threshold, so their similarity scores remain unchanged. The fatigue decay coefficient is preset based on the degree of impact of employee fatigue on job competence and is used to proportionally correct the similarity of the matching. According to historical work records: Employee A101 worked 6 consecutive days with a 24-hour rest interval before March 1; Employee B203 worked 2 consecutive days with a 48-hour rest interval; Employee C105 worked 7 consecutive days with no rest interval; Employee E302 worked 3 consecutive days with an 8-hour rest interval.

[0037] Among them, employee A101 worked 6 consecutive days, which is equal to the preset continuous working days threshold of 6 days, and had a rest interval of 24 hours, which exceeds the preset rest interval threshold of 12 hours. Therefore, the matching similarity score was not reset to zero, and the adjusted matching similarity score remained at 0.67293. Employee B203 worked 2 consecutive days, which is less than 6 days, and had a rest interval of 48 hours, which is greater than 12 hours. The matching similarity score remained at 0.7821. Employee C105 worked 7 consecutive days, which is greater than 6 days. Therefore, the matching similarity score for all shifts from March 1st to 14th was reset to zero. Employee E302 had a rest interval of 8 hours, which is less than 12 hours. Therefore, the matching similarity score for all shifts within the scheduling period was reset to zero.

[0038] In the above method, the attenuation processing based on the preset fatigue accumulation value threshold allows the matching similarity score to intuitively reflect the actual workload status of employees, avoids prioritizing employees with high fatigue values ​​in the scheduling, takes into account the balance of employees' work fatigue, and meets the needs of enterprises for humane scheduling.

[0039] In one embodiment of this application, step S140, which determines the employee variable ranking sequence based on the matching similarity score and fatigue accumulation value, further includes the following steps: Figure 4 As shown, the specific content is as follows: Step S410: Sort the matching similarity scores from maximum to minimum to generate a sorted sequence of employee variables; Step S420: If the difference between the matching similarity scores of adjacent employees in the employee variable sorting sequence does not exceed the preset similarity difference threshold, then the fatigue accumulation value is sorted from the minimum value to the maximum value to update the employee variable sorting sequence.

[0040] Specifically, taking the shift schedule of a manufacturing company from March 1st to March 14th as an example, the preset similarity difference threshold is 0.1. Five production workshop employees (Employee A, Employee B, Employee C, Employee D, and Employee E) participate in this shift. The preset similarity difference threshold is used to determine whether the matching similarity scores of two employees are sufficiently close (i.e., "similar"). If the difference is less than this threshold, the two are considered to have comparable skill matching. This threshold can be flexibly set according to the actual application scenario, such as based on historical matching score distribution, expert experience, or simulation optimization. Specifically, Employee A has a matching similarity score of 0.83 and a cumulative fatigue value of 11; Employee B has a matching similarity score of 0.75 and a cumulative fatigue value of 9; Employee C has a matching similarity score of 0.48 and a cumulative fatigue value of 7; Employee D has a matching similarity score of 0.95 and a cumulative fatigue value of 13; and Employee E has a matching similarity score of 0.60 and a cumulative fatigue value of 18. Based on the matching similarity score, employees are sorted in descending order, resulting in the following initial employee variable sorting sequence: Employee D (0.95) → Employee A (0.83) → Employee B (0.75) → Employee E (0.60) → Employee C (0.48). The difference in matching similarity scores between Employee D and Employee A is 0.95 - 0.83 = 0.12, exceeding the preset threshold of 0.1, thus maintaining their original order in the sequence. The difference in matching similarity scores between Employee A and Employee B is 0.83 - 0.75 = 0.08, not exceeding the preset similarity difference threshold of 0.1, thus classifying them as employees with similar scores. The difference in matching similarity scores between Employee B and Employee E is 0.75 - 0.60 = 0.15, exceeding the preset similarity difference threshold of 0.1, thus maintaining their original order in the sequence. The difference in similarity scores between employee E and employee C is 0.60 - 0.48 = 0.12. This difference exceeds the preset similarity difference threshold of 0.1, so the original order of the two employees in the sequence is maintained. Therefore, only the order of employees A and B, who have similarity scores, is adjusted. Their cumulative fatigue values ​​are retrieved and reordered in ascending order from minimum to maximum: Employee B's cumulative fatigue value is 9 < Employee A's cumulative fatigue value is 11. Therefore, the order of this group of employees is adjusted to Employee B → Employee A, and the order of other adjacent employees remains unchanged. The final updated employee variable sorting sequence is: Employee D (0.95, 13) → Employee B (0.75, 9) → Employee A (0.83, 11) → Employee E (0.60, 18) → Employee C (0.48, 7).

[0041] In the above method, the similarity score is used as the core sorting criterion to ensure that the sorting results prioritize the matching degree requirements of the job skills. This allows subsequent scheduling decisions to prioritize employees with high skill suitability, ensuring the quality of skill matching in the scheduling plan. For employees with similarity scores, the order is adjusted in ascending order based on their accumulated fatigue value. This ensures that the employee variable sorting sequence takes into account both skill matching degree and the actual workload status of employees. This effectively avoids prioritizing employees with high fatigue values ​​in subsequent scheduling. The sorting process lays the foundation for the balance of employee workload from the very beginning, meeting the dual requirements of humanized and compliant scheduling.

[0042] In one embodiment of this application, step S140, which involves determining the employee variable to be decided using a deep learning algorithm based on the employee variable sorting sequence, further includes the following steps: Figure 5 As shown, the specific content is as follows: Step S510: Based on the employee variable sorting sequence, use a deep learning algorithm to determine the conflict value of each employee, and take the employee with the highest conflict value as the conflicting employee; Step S520: Filter each unassigned employee in the sorted sequence of employee variables, and use a weighted algorithm to determine the closeness between conflicting employees and each unassigned employee; Step S530: Select the unassigned employee with the highest affinity to the conflicting employee as the employee to be decided.

[0043] Specifically, this can be achieved by retrieving a pre-constructed employee variable ranking sequence, along with data such as each employee's historical continuous work records, current shift allocation status, minimum rest interval duration, and maximum consecutive working days limit. This data can then be used as input features for a pre-defined deep learning model. This deep learning model is trained based on historical scheduling conflict data and industry scheduling constraints. It automatically learns the conflict impact weights under different scheduling constraints, performs quantitative analysis on the input employee features, and calculates the conflict value for each employee. The magnitude of the conflict value directly reflects the severity of the employee's violation of scheduling constraints such as the minimum rest interval and maximum consecutive working days in the current scheduling plan. The conflict value calculation revolves around the number of times an employee violates the minimum rest interval and the number of days they violate the maximum consecutive working days. The deep learning model assigns differentiated weight coefficients to these two types of violations (the weight coefficient for violating the maximum consecutive working days is higher than that for violating the minimum rest interval, conforming to labor regulations and scheduling priorities), and performs comprehensive quantification in conjunction with the employee's current shift allocation status. After the system completes the calculation of the conflict value of all employees in the employee variable sorting sequence, it sorts the conflict values ​​in descending order and selects the employee with the highest conflict value as the conflicting employee. This employee is the one with the most prominent constraint conflict in the current scheduling process and is the core subject that should be given priority in subsequent scheduling decisions.

[0044] For example, the shift allocation status of all employees in the employee variable sorting sequence can be checked, identifying and filtering out employees who have not yet received any shift allocation, forming an unassigned employee set, and excluding employees who have completed shift allocation, thereby narrowing the scope of subsequent closeness calculations and improving decision-making efficiency. Then, a preset weighted algorithm formula is retrieved, which uses the matching similarity score between the employee and the conflicting employee and the reciprocal of the employee's cumulative fatigue value as the core calculation dimensions, and assigns preset weight coefficients to the two dimensions (the matching similarity score is the core dimension with a higher weight, and the reciprocal of the cumulative fatigue value is the auxiliary dimension, taking into account the employee's workload status), where the reciprocal of the cumulative fatigue value is 1 / the employee's cumulative fatigue value, used to inversely transform the numerical value of the cumulative fatigue value into a positive evaluation indicator. For each employee in the set of unassigned employees, the matching similarity score between them and the conflicting employee during the scheduling cycle, along with their own accumulated fatigue value, are retrieved and used in a weighted algorithm formula to calculate the closeness value between each unassigned employee and the conflicting employee. A higher closeness value indicates a stronger compatibility between the unassigned employee and the conflicting employee in terms of scheduling skills and workload balance, making them the optimal potential candidate for resolving the scheduling constraints of conflicting employees. Finally, the unassigned employee with the highest closeness value is selected as the employee to be considered in the decision-making process, and subsequent scheduling procedures will prioritize extracting candidate shifts for this employee.

[0045] For example, the preset deep learning model assigns weights to scheduling conflicts as follows: a weight coefficient of 2 for violating the maximum consecutive working days and a weight coefficient of 1 for violating the minimum rest interval. The weighted algorithm formula for calculating the tightness is: Tightness = Matching Similarity Score × 0.6 + Reciprocal of Fatigue Accumulation Value × 0.4. The preset thresholds for the maximum consecutive working days are 5 days and the minimum rest interval threshold is 2 days. The employee variable sorting sequence is: Employee D (Matching Similarity Score 0.92, Fatigue Accumulation Value 12) → Employee A (0.85, 10) → Employee B (0.78, 15) → Employee E (0.70, 20) → Employee C (0.65, 8). Currently, Employee D and Employee A have completed shift assignments, while Employee B, Employee E, and Employee C are unassigned employees. The historical continuous work records and current shift allocation status of five employees in the sequence were retrieved and input into a deep learning model to calculate the conflict value of each employee. The results showed that employee A violated the maximum continuous work days threshold by working 6 consecutive days, resulting in a conflict value of 2; employee B violated the minimum rest interval threshold by 1 day, resulting in a conflict value of 1; and employees D, E, and C had no violations, resulting in a conflict value of 0. Therefore, employee A, with the highest conflict value, was selected as the conflicting employee. The allocation status of the employee variable sequence was verified, and the set of unassigned employees {employee B, employee E, employee C} was selected. The closeness value between each employee in this set and the conflicting employee A was calculated using the closeness calculation formula. The closeness value of employee B was 0.78×0.6+(1 / 15)×0.4≈0.4947, the closeness value of employee E was 0.70×0.6+(1 / 20)×0.4=0.44, and the closeness value of employee C was 0.65×0.6+(1 / 8)×0.4=0.44. The density values ​​of the unassigned employees are sorted. Since employee B has the highest density value, employee B is identified as the employee to be decided.

[0046] In one embodiment of this application, step S150, which involves obtaining the first candidate shifts corresponding to the employees who are the decision variables, determining the comprehensive value score of each first candidate shift, and assigning the first candidate shift with the highest comprehensive value score to the employees who are the decision variables, further includes the following steps: Figure 6 As shown, the specific content is as follows: Step S610: Based on each first candidate shift, determine the contribution value of each first candidate shift to the task slack of the remaining shifts, the impact value of each first candidate shift on the increase of the cumulative fatigue value of employees, and the estimated value of each first candidate shift on the average number of available shifts for unassigned days. Step S620: Obtain the weights corresponding to the contribution value, the influence value, and the estimated value, respectively; Step S630: Based on the weights corresponding to the contribution value, the influence value, and the estimated value, perform a weighted calculation on the contribution value, influence value, and estimated value, and use the result of the weighted calculation as the comprehensive value score of each first candidate class.

[0047] Specifically, the remaining available shift range for the employee to be decided variable is obtained, and all shifts within this range are identified as the first candidate shifts. Subsequently, data such as the total remaining scheduling tasks, the current cumulative fatigue value of the employee to be decided variable, the preset cumulative fatigue value threshold, the number of unassigned days for the employee to be decided variable, and the scheduling constraint rules for unassigned days are retrieved simultaneously, and three quantitative indicators are calculated for each first candidate shift.

[0048] The contribution value of remaining shift task slack: This refers to the number of remaining shift task demands that can be directly reduced after assigning this first candidate shift. The higher the value, the greater the effect of this shift on alleviating the current tightness of the shift tasks. This value is automatically calculated by a coverage model trained on historical shift data, combined with the job skill matching degree of the candidate shift. The impact value of the fatigue accumulation increment: This refers to the control effect of assigning this first candidate shift on the increment of the employee's fatigue accumulation value, which is a normalized value between 0 and 1. The higher the value, the better the fatigue increment control effect. It is calculated using the formula: Impact value of fatigue accumulation increment = [Preset fatigue accumulation threshold - (Employee's current fatigue accumulation value + Fatigue increment of this shift)] / Preset fatigue accumulation threshold. The fatigue increment for each shift is preset based on the preset shift duration schedule and job workload. The estimated value of the average number of available shifts on unassigned dates: This refers to the average number of valid shifts that employees, the decision variable, can choose on each day during the remaining unassigned dates after the first candidate shift is assigned. The higher the value, the smaller the impact of the shift assignment on the flexibility of the employee's subsequent scheduling. This value is automatically estimated by simulating the propagation process of scheduling constraints and combining it with the job requirements on unassigned dates.

[0049] For example, taking the shift scheduling cycle of a manufacturing company from March 1st to March 14th as an example, the preset threshold for accumulated employee fatigue is 50, and the preset weighting coefficients are: 0.4 for the contribution value of the remaining shift task slack, 0.3 for the impact value of the increase in accumulated fatigue, and 0.3 for the estimated value of the average number of available shifts on unassigned days. The company's fatigue increment is 5 for the morning shift, 10 for the afternoon shift, and 15 for the evening shift. The current decision variable is employee B. The remaining available shift values ​​for employee B are retrieved as morning, afternoon, and evening shifts, and these three shifts are identified as the first candidate shifts. Employee B's current accumulated fatigue value is 30, the total remaining shift task demand is 10 shifts, and employee B has 3 remaining unassigned days. Calculate the three indicator values ​​for the three first candidate shifts respectively: (1) Morning shift: the contribution value of the remaining shift task slack is 2 (which can reduce the demand for 2 remaining shifts), the impact value of the fatigue accumulation value increment (50-(30+5)) / 50=0.3, and the estimated value of the average number of shifts available on unassigned dates is 2.5; (2) Afternoon shift: the contribution value of the remaining shift task slack is 1 (which can reduce the demand for 1 remaining shift), the impact value of the fatigue accumulation value increment (50-(30+10)) / 50=0.2, and the estimated value of the average number of shifts available on unassigned dates is 2.0; (3) Evening shift: the contribution value of the remaining shift task slack is 0.5 (which can reduce the demand for 0.5 remaining shifts), the impact value of the fatigue accumulation value increment (50-(30+15)) / 50=0.1, and the estimated value of the average number of shifts available on unassigned dates is 1.5. The comprehensive value score of each first candidate shift was calculated using a weighted algorithm: (1) Comprehensive value score of the morning shift: 2×0.4+0.3×0.3+2.5×0.3=1.64; (2) Comprehensive value score of the afternoon shift: 1×0.4+0.2×0.3+2.0×0.3=1.06; (3) Comprehensive value score of the evening shift: 0.5×0.4+0.1×0.3+1.5×0.3=0.68. Finally, the comprehensive value scores of the three shifts were associated with the corresponding shifts and stored. The comprehensive value score of the morning shift, 1.64, was the highest. This shift was then assigned as the optimal shift to employee B, the variable to be decided.

[0050] In one embodiment of this application, step S160, which updates the second candidate shifts for employees associated with the decision variable employee based on the first candidate shift, and uses an arc consistency propagation mechanism to determine the third candidate shifts for employees with constraints on the current allocation of each second candidate shift, further includes the following steps: Figure 7 As shown, the specific content is as follows: Step S710: Use the arc consistency propagation mechanism to obtain the influence range of employees who are constrained by the current allocation of each second candidate shift; Step S720: Update the cumulative fatigue value and number of consecutive working days for each employee according to the scope of impact; Step S730: Determine the third candidate shifts for employees with constraints for each of the current second candidate shifts based on the updated fatigue accumulation value and the updated continuous working days.

[0051] Specifically, the arc consistency propagation mechanism can be used to verify and propagate the relationships between scheduling constraints. It uses the first candidate shift assigned to the employee (the decision variable) as the core propagation node, retrieving all preset hard scheduling constraints, including personnel capacity constraints for the same position and shift, mutual exclusion constraints for the same shift, consecutive work days constraints, minimum rest interval constraints, and employee partner shift constraints. Personnel capacity constraints define the maximum number of employees that can be assigned to the same position and shift, which can be provided by the production department based on the number of equipment, workspace, and safety requirements for the position. Employee partner shift constraints can be set according to the company's work processes or safety requirements (e.g., some positions require two people to work together), and the definition of the partner relationship (e.g., fixed pairing or arbitrary shift) must be clearly defined.

[0052] Based on these constraint rules, the system iterates through all related employees in the employee variable ranking sequence who have direct constraint relationships with the employee to be decided (whose second candidate shift has been updated), and further traces and identifies indirectly related employees whose constraints are linked due to directly related employees, accurately defining the scope of all employees whose constraint conflicts may be triggered by the allocation of the second candidate shift; finally, all employees within the scope of influence are exclusively marked to form a set of employees affected by the constraints. This set defines a clear and unique object for subsequent employee status updates, ensuring that the constraint propagation and status update process is not missed or redundant.

[0053] For example, the system can retrieve the current basic status data of each employee within the set of employees affected by the constraints, including the current cumulative fatigue value, the continuous working days counter value, and the assigned shift records. Simultaneously, it can retrieve data such as the preset shift duration table and the fatigue increment calculation rules for each shift. Then, based on the first candidate shift allocation results for the employees to be decided, the system performs a status quantification update for each employee within the affected area: for employees with continuous working constraints and rest interval constraints, the continuous working days counter is incremented or reset according to the correlation between their assigned shifts and the first candidate shift, and the actual rest interval duration is updated simultaneously; for all employees within the affected area, the fatigue cumulative value increment is calculated according to the preset fatigue increment calculation rules based on the shift type and standard duration of their second candidate shift, and this increment is added to the current fatigue cumulative value to obtain the updated fatigue cumulative value.

[0054] For example, the updated cumulative fatigue value, consecutive working days, and corresponding thresholds of each employee within the employee set affected by the constraints can be quantitatively compared one by one. Simultaneously, a compliance screening operation can be performed on the candidate shifts by combining this with the current second candidate shift value range for these employees. If assigning a second candidate shift would cause an employee's consecutive working days to exceed a preset consecutive working days threshold, rest interval to fall below a preset rest interval threshold, or cumulative fatigue value to exceed a preset cumulative fatigue value threshold, then that non-compliant candidate shift would be directly removed from the second candidate shift value range. If, after assigning a second candidate shift, all of the employee's status data conforms to the preset hard constraint rules, then that candidate shift would be retained. The final employee candidate shift value range formed after compliance screening is the employee's third candidate shift.

[0055] For example, taking the scheduling cycle of a manufacturing company from March 1st to March 14th as an example, the AC-3 arc consistency propagation mechanism is adopted. The preset hard constraints and corresponding thresholds for scheduling are: maximum staff capacity of 4 people for the injection molding job in the morning shift, staff capacity constraint for the same job and shift, maximum continuous working days threshold of 6 days, minimum rest interval threshold of 12 hours, fatigue accumulation threshold of 50, and fatigue increment for each shift is 5 for the morning shift, 10 for the afternoon shift, and 7.5 for the evening shift. The employee to be decided is employee B, whose first candidate shift is the morning shift of the injection molding job on March 1st. The related employees with direct constraint relationship with employee B are employees D, E, and F in the same injection molding job (all three of them have second candidate shifts including the morning shift of the injection molding job on March 1st). The indirect related employee with partner and shift constraint relationship with employee F is employee G (whose second candidate shifts include the morning and afternoon shifts of the injection molding job on March 1st). Taking employee B's injection molding morning shift on March 1st as the core propagation node, based on the AC-3 arc consistency algorithm and preset constraint rules, the employees directly affected by the constraints are identified as employees D, E, and F, and the employee indirectly affected by the chain constraints is employee G. Finally, the scope of the constraint influence is defined as {employees D, E, F, and G}, and employees within this set are marked with exclusive tags.

[0056] For example, the current status data of four employees within the employee set affected by the constraints can be retrieved: Employee D has worked 3 consecutive days and has a fatigue accumulation value of 22; Employee E has worked 4 consecutive days and has a fatigue accumulation value of 28; Employee F has worked 5 consecutive days and has a fatigue accumulation value of 35; Employee G has worked 2 consecutive days and has a fatigue accumulation value of 18. Combining Employee B's injection molding morning shift assignment result on March 1st, the status data of the four employees is updated according to preset rules: If Employee D is assigned to this morning shift, their consecutive working days increase by 1 to 4, and their fatigue accumulation value increases by 5 to 27; if Employee E is assigned to this morning shift, their consecutive working days increase by 1 to 5, and their fatigue accumulation value increases by 5 to 33; if Employee F is assigned to this morning shift, their consecutive working days increase by 1 to 6, and their fatigue accumulation value increases by 5 to 40; since Employee G is partnered with F on the same shift, if F is assigned to this morning shift, their consecutive working days increase by 1 to 3, and their fatigue accumulation value increases by 5 to 23.

[0057] For example, preset hard constraint thresholds can be retrieved to screen the updated status data of four employees against the second candidate shifts for compliance: Employee D, after updating, has worked 4-6 consecutive days and has a fatigue accumulation value of 27-50, meeting all constraints, and is retained for the March 1st injection molding morning shift as the third candidate shift; Employee E, after updating, has worked 5-6 consecutive days and has a fatigue accumulation value of 33-50, meeting all constraints, and is retained for the March 1st injection molding morning shift as the third candidate shift; Employee F, after updating, has worked 6 consecutive days, reaching the maximum consecutive working days threshold. If assigned to this morning shift, there would be no room for subsequent adjustments and it would easily trigger constraint conflicts, so this morning shift is removed, and its third candidate shift is the March 1st injection molding middle shift from the original second candidate shifts; Employee G, due to the constraint of being paired with F in the same shift, is also removed from the March 1st injection molding morning shift, and its third candidate shift is the March 1st injection molding middle shift from the original second candidate shifts. The final three candidate shifts for the four employees to be assigned to are: Employee D {March 1st, injection molding morning shift}, Employee E {March 1st, injection molding morning shift}, Employee F {March 1st, injection molding afternoon shift}, and Employee G {March 1st, injection molding afternoon shift}.

[0058] In the above method, the global linkage verification of scheduling constraints is realized by combining the arc consistency propagation mechanism with big data processing. This allows the shift allocation result of a single employee to quickly update the candidate shifts of all related employees, effectively reducing the occurrence of constraint deadlock and empty value range of employee selectable shifts in the subsequent scheduling process, and improving the stability and feasibility of the overall scheduling scheme.

[0059] In one embodiment of this application, after determining that each of the second candidate shifts has a constrained employee's third candidate shift in the current allocation of each second candidate shift, step S160 further includes: if the value range of each third candidate shift has an empty set, then the arc consistency propagation mechanism is used again to update the current scope of influence to adjust the third candidate shifts. Specifically, a full value range verification can be performed on all employees within the scope of the constraint influence, and the value range data of each employee's third candidate shift can be retrieved one by one to determine whether there are valid and compliant shift options within the value range; if there are no valid shifts in the value range of a certain employee's third candidate shift, that is, the value range is an empty set, the employee will be marked as an employee with an empty value range. Next, the arc consistency propagation mechanism will be invoked again, with employees in the empty set of the value domain becoming the new core propagation node, replacing the first candidate shift node of the decision variable employees in the first propagation. At the same time, the link of the first constraint propagation will be traced back in reverse, starting from the constraint relationship of employees in the empty set of the value domain, tracing back to employees with direct constraints, and further expanding to identify potential indirectly constrained employees not covered in the first propagation. Combined with all the hard constraint rules of the enterprise's preset scheduling, a more comprehensive and accurate scope of constraint influence will be redefined. The new scope of constraint influence includes both the original employees in the first propagation and the constraint-related employees newly added due to the empty set of the value domain, ensuring that all constraint links related to the empty set problem are covered, and then the third candidate shift will be dynamically adjusted based on the updated scope of influence.

[0060] In one embodiment of this application, in step S170, a global scheduling plan is generated based on each third candidate shift. Within this global scheduling plan, if at least one employee is not assigned a shift for any date within the scheduling period, the employee variable to be decided is updated again using a deep learning algorithm. Specifically, all third candidate shifts after constraint filtering for all employees, as well as the daily skill requirement matrix for positions within the scheduling period, personnel capacity constraints for each shift, employee fatigue accumulation threshold, maximum consecutive working days threshold, and all other core scheduling rules and basic data can be retrieved. Subsequently, using the date-shift-position dimension within the scheduling cycle as the three-dimensional allocation dimension, and with the core objectives of "maximizing skill matching, balancing employee fatigue accumulation, and fully satisfying constraint rules," the third candidate shifts of all employees are matched and allocated across the entire domain with the skill requirements and personnel capacity of each position on each date. Prioritizing employees with pending decision variables first, followed by ordinary employees, and prioritizing employees with high matching similarity first, followed by employees with low matching similarity, the shift allocation for all employees on each date within the scheduling cycle is completed. Finally, all allocation results are integrated into a structured global scheduling scheme, which includes information such as each employee's specific shift, working hours, and position allocation for each day within the scheduling cycle, while also linking status data such as the employee's accumulated fatigue value and number of consecutive working days.

[0061] Optionally, the system can iterate through the shift allocation records of all employees participating in this scheduling, and check the allocation status of each employee for each day within the scheduling cycle, marking employees who are "not allocated on a single date" and "not allocated for the entire cycle". If the detection result is that all employees have shift allocations for at least one day within the scheduling cycle, and only some employees are not allocated on a single day, then the system attempts to swap the unallocated date with the shifts on adjacent dates, or adjust the shifts of other employees in the same position, supplementing the allocation under the premise of satisfying all constraints, and completing the optimization of the global scheduling scheme. If the detection result is that at least one employee has not received a shift allocation for any day within the scheduling cycle, then the current global scheduling scheme is determined to have an allocation integrity problem, and a deep learning algorithm is used to update the decision variable of employees.

[0062] In one embodiment of this application, after generating the global scheduling scheme based on each third candidate shift in step S170, the following steps are also included: Figure 8 As shown, the specific content is as follows: Step S810: If there are employees on leave, obtain the affected date range and shift of the employees on leave; Step S820: Within the affected date range, calculate the matching similarity score between the remaining employees (excluding those on leave) and the shifts of those on leave, and the cumulative fatigue value of the remaining employees; Step S830: Adjust the global shift scheduling scheme based on the matching similarity score between the shifts of the remaining employees and the shifts of the employees on leave, and the cumulative fatigue value of the remaining employees.

[0063] Specifically, taking the work schedule of a manufacturing company from March 1st to March 14th as an example, the company has 38 production workshop employees participating in this schedule. The preset hard constraints and thresholds for the schedule are: a maximum of 6 consecutive working days, a minimum rest interval of 12 hours, a fatigue accumulation threshold of 0.8, and a fatigue attenuation coefficient of 0.9. The fatigue increment for each shift is 5 for the morning shift, 10 for the afternoon shift, and 7.5 for the evening shift. The positions are divided into injection molding, packaging, and logistics positions, with the core skill requirement for the injection molding position being injection molding machine operation. Employee A101 in the injection molding position submitted a sudden leave application for March 5th to March 7th on March 3rd. At this time, the leave application module can be used to retrieve employee A101's leave start and end time, which is March 5th to March 7th, i.e., the affected date range. At the same time, the vacant shifts already assigned to him within this range are the injection molding morning shift on March 5th, the injection molding afternoon shift on March 6th, and the injection molding morning shift on March 7th. From the remaining 37 employees, 12 employees with injection molding machine operation skills were selected to form the remaining candidate employee set. The skill requirement vectors for the injection molding positions corresponding to the three vacant shifts were retrieved, and the weighted cosine similarity algorithm was used to calculate the matching similarity score. Employee B203 had a matching similarity score of 0.7821 for the morning shift on March 5th, 0.7532 for the afternoon shift on March 6th, and 0.7615 for the morning shift on March 7th. Employee D102 had a matching similarity score of 0.8100 for the morning shift on March 5th, 0.7220 for the afternoon shift on March 6th, and 0.7905 for the morning shift on March 7th. Employee C105 had a matching similarity score of 0.7200 for the morning shift on March 5th, 0.7710 for the afternoon shift on March 6th, and 0.7300 for the morning shift on March 7th. At the same time, the cumulative fatigue values ​​of these 12 employees were recalculated. The cumulative fatigue value of employee B203 was 0.679, the cumulative fatigue value of employee D102 was 0.650, and the cumulative fatigue value of employee C105 was 0.710. None of them exceeded the preset cumulative fatigue value threshold of 0.8.

[0064] Next, a local variable sorting sequence was constructed for each vacant shift. The local variable sorting sequence for the morning shift on March 5th was: D102(0.8100,0.650)→B203(0.7821,0.679)→C105(0.7200,0.710); the local variable sorting sequence for the afternoon shift on March 6th was: C105(0.7710,0.710)→B203(0.7532,0.679)→D102(0.7220,0.650); and the local variable sorting sequence for the morning shift on March 7th was: D102(0.7905,0.650)→B203(0.7615,0.679)→C105(0.7300,0.710). A preliminary shift allocation was then performed: employee D102 was assigned to the morning shift on March 5th, employee C105 to the afternoon shift on March 6th, and employee B203 to the morning shift on March 7th. The arc consistency propagation mechanism was invoked to perform constraint checks. It was found that the three employees had consecutive working days of 4, 2, and 3 days respectively after allocation, all within the 6-day consecutive working day threshold. Rest intervals all met the 12-hour requirement. The cumulative fatigue values, after adding the corresponding shift increments, were 0.700, 0.760, and 0.729 respectively, all within the preset fatigue cumulative value threshold of 0.8, indicating no constraint conflicts. Finally, this allocation result replaced the vacant shifts in the original global scheduling plan, updated the cumulative fatigue values ​​and consecutive working days of employees D102, C105, and B203, locked the shift allocations for the remaining 35 employees and all shifts on unaffected dates, and generated the adjusted global scheduling plan.

[0065] In some embodiments, taking the bi-weekly shift schedule of an internal medicine department in a community hospital as an example, eight clinical doctors participate in the scheduling, with medical positions including general outpatient, emergency, and inpatient ward management, and shifts divided into day and night shifts. First, the system connects to the hospital's human resources management system and departmental operations management system to obtain an employee file database containing the professional qualifications, skill tags, and historical work records of the eight doctors, as well as a business task plan table for the department's bi-weekly treatment tasks, staffing needs, and shift divisions. Next, the skill tag set and fatigue accumulation value for each doctor are determined from the employee file database. For example, Dr. Wang's skill tag set is {Internal Medicine Physician, Emergency Treatment Qualification, General Outpatient Reception Qualification}, and Dr. Li's is {Internal Medicine Physician, Inpatient Ward Management Qualification}. Simultaneously, based on the business task plan table, skill requirements for different positions on each day and night shift are quantitatively assigned within a 0-1 range, constructing a daily skill requirement matrix for each position. Subsequently, the skill tag set of each doctor is compared with the job skill requirements for each date and shift in the matrix using cosine similarity calculation to determine the matching similarity score. For example, the matching similarity score between Dr. Wang, who has emergency treatment qualifications, and the emergency night shift is 0.95. An initial employee variable sorting sequence is generated according to the matching similarity score from high to low. For doctors whose score difference does not reach the preset threshold, the sequence is adjusted according to the fatigue accumulation value from low to high. Then, based on the adjusted sequence, a deep learning algorithm is used to calculate the conflict value of each doctor. Dr. Zhao, the unassigned doctor with the highest conflict value, is identified as the employee to be decided. The remaining optional shifts for Dr. Zhao, the outpatient day shift and the inpatient ward day shift, are extracted as the first candidate shifts. The contribution value of the two shifts to the remaining task slackness and the impact value of the fatigue accumulation increment are calculated respectively, and weighted to obtain a comprehensive value score. The outpatient day shift with the higher score is assigned to Dr. Zhao. Following Dr. Zhao's outpatient day shift assignments, the second candidate shifts for two related doctors in the same outpatient department were updated. Then, an arc consistency propagation mechanism was used to obtain the influence range of doctors with constraints under this assignment, updating the cumulative fatigue value and continuous working days of doctors within that range, and selecting their third candidate shifts. Finally, the third candidate shifts of all eight doctors were integrated, and a global scheduling plan was generated based on departmental staffing, scheduling constraints, and other requirements. After checking the plan, it was confirmed that all doctors had corresponding shift assignments within the bi-weekly scheduling cycle, eliminating the need to re-update the employee variable using deep learning algorithms. If the check revealed that a doctor had not received a shift assignment for any date within the scheduling cycle, the employee variable would be updated again, and the scheduling process iterated.

[0066] This application also provides a deep learning-based intelligent scheduling system, which may include an acquisition module, a first determination module, a second determination module, a third determination module, a fourth determination module, an update module, and a generation module. The acquisition module acquires an employee file database and a business task plan table. The first determination module determines the skill tag set and fatigue accumulation value of each employee within the scheduling cycle based on the employee file database, and determines the daily skill requirement matrix for each position based on the business task plan table. The second determination module determines the matching similarity score between each employee and each shift within the scheduling cycle based on the skill tag set and the daily skill requirement matrix for each position. The third determination module determines the employee variable ranking sequence based on the matching similarity score and fatigue accumulation value, and uses a deep learning algorithm to determine the employee to be decided variable based on the employee variable ranking sequence. The fourth determination module acquires the decision variable... The system firstly identifies the first candidate shifts for each employee in the decision variable, determines the comprehensive value score of each first candidate shift, and assigns the first candidate shift with the highest comprehensive value score to the employee in the decision variable. The second candidate shift update module updates the second candidate shifts for employees associated with the employees in the decision variable based on the first candidate shifts, and uses an arc consistency propagation mechanism to determine the third candidate shifts for employees constrained by the current allocation of second candidate shifts. The third candidate shift generation module generates a global scheduling plan based on the third candidate shifts. In the global scheduling plan, if at least one employee has not received a shift allocation for any date within the scheduling period, the employee in the decision variable is updated again using a deep learning algorithm.

[0067] It should be noted that the embodiments of the deep learning-based intelligent scheduling system provided in this application can be used to execute the processing flow of the embodiments of the deep learning-based intelligent scheduling method in the above embodiments. Its functions will not be repeated here, but can be referred to the detailed description of the above method embodiments.

[0068] This application also provides an electronic device, which includes one or more processors and memory resources represented by memory for storing instructions executable by the processor, such as application programs. The application programs stored in the memory may include one or more modules, each corresponding to a set of instructions. Furthermore, the processor is configured to execute the instructions to perform the aforementioned deep learning-based intelligent scheduling method.

[0069] The electronic device may also include a power supply component configured to perform power management of the electronic device, a wired or wireless network interface configured to connect the electronic device to a network, and an input / output (I / O) interface. The electronic device can be operated based on operating devices stored in memory, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or similar.

[0070] In one embodiment, a computer device, which may be a server, is also provided. The computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device stores data. The I / O interfaces of the computer device are used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a deep learning-based intelligent scheduling method.

[0071] It should be noted that although the steps of the deep learning-based intelligent scheduling method in this application are described in a specific order in the accompanying drawings, this does not require or imply that these steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps, such as omitting certain steps, combining multiple steps into one step, and / or decomposing one step into multiple steps, should all be considered part of this application.

[0072] It should be understood that this application is not limited to the detailed structure and arrangement of the modules in the deep learning-based intelligent scheduling system proposed in this specification. This application can have other implementations and can be implemented and executed in various ways. The foregoing variations and modifications fall within the scope of this application. It should be understood that the application and its definition in this specification extend to all alternative combinations of two or more individual features mentioned or apparent in the text and / or drawings. All these different combinations constitute multiple alternative aspects of this application. The embodiments described in this specification illustrate the best known mode for implementing this application and will enable those skilled in the art to utilize this application.

Claims

1. A deep learning-based intelligent scheduling method, characterized in that, include: Obtain employee profile database and business task plan; Based on the employee file database, determine the skill tag set and fatigue accumulation value of each employee within the scheduling cycle, and determine the daily skill requirement matrix for each position based on the business task plan. Based on the skill tag set and the daily skill requirement matrix for the job, determine the matching similarity score between each employee and each shift within the scheduling cycle; The employee variable ranking sequence is determined based on the matching similarity score and the fatigue accumulation value, and a deep learning algorithm is used to determine the employee variable to be decided based on the employee variable ranking sequence. Obtain each first candidate shift corresponding to the employee to be decided variable, determine the comprehensive value score of each first candidate shift, and assign the first candidate shift corresponding to the highest comprehensive value score to the employee to be decided variable. Update the second candidate shifts for employees associated with the employee to be decided variable based on the first candidate shift, and use the arc consistency propagation mechanism to determine the third candidate shifts for employees who are constrained by the current allocation of each second candidate shift. A global scheduling scheme is generated based on each of the third candidate shifts. In the global scheduling scheme, if at least one employee is not assigned a shift for any date within the scheduling period, the employee, the decision variable, is updated again using a deep learning algorithm.

2. The intelligent scheduling method based on deep learning according to claim 1, characterized in that, The step of determining the skill tag set and fatigue accumulation value of each employee within the scheduling cycle based on the employee file database includes: In the employee file database, the skill tags and historical work records of each employee within the scheduling cycle are obtained, and the skill tag set is determined based on the skill tags; Obtain a preset shift duration table, and determine the cumulative working hours based on the historical work records and the preset shift duration table; The cumulative fatigue value is determined based on the cumulative working time.

3. The intelligent scheduling method based on deep learning according to claim 2, characterized in that, After determining the matching similarity score between each employee and each shift within the scheduling cycle based on the skill tag set and the daily skill requirement matrix for the job, the method further includes: If the fatigue accumulation value exceeds the preset fatigue accumulation value threshold, then the fatigue decay coefficient is obtained, and the fatigue decay coefficient is multiplied by the current matching similarity score to adjust the matching similarity score; The number of consecutive working days and rest intervals for each employee are determined based on the aforementioned historical work records; If the current employee's consecutive working days exceed the preset consecutive working days threshold or the rest interval does not exceed the preset rest interval threshold, then the matching similarity score between the current employee and each shift in the scheduling cycle is set to zero.

4. The intelligent scheduling method based on deep learning according to claim 1, characterized in that, The step of determining the employee variable ranking sequence based on the matching similarity score and the fatigue accumulation value includes: The matching similarity scores are sorted from the maximum to the minimum to generate the sorted sequence of the employee variables; If the difference between the matching similarity scores of adjacent employees in the employee variable sorting sequence does not exceed the preset similarity difference threshold, then the fatigue accumulation value is sorted from the minimum value to the maximum value to update the employee variable sorting sequence.

5. The intelligent scheduling method based on deep learning according to claim 1, characterized in that, The step of determining the employee as the decision variable using a deep learning algorithm based on the employee variable sorting sequence includes: Based on the employee variable sorting sequence, a deep learning algorithm is used to determine the conflict value of each employee, and the employee with the highest conflict value is identified as the conflicting employee. Filter each unassigned employee in the sorted sequence of employee variables, and use a weighted algorithm to determine the closeness between the conflicting employees and each of the unassigned employees; The unassigned employee with the highest affinity to the conflicting employee is selected as the employee to be decided.

6. The intelligent scheduling method based on deep learning according to claim 1, characterized in that, The determination of the comprehensive value score for each of the first candidate classes includes: Based on each of the first candidate shifts, determine the contribution value of each of the first candidate shifts to the slackness of the remaining shifts, the impact value of each of the first candidate shifts on the increase of the cumulative fatigue value of employees, and the estimated value of each of the first candidate shifts on the average number of available shifts on unassigned days. Obtain the weights corresponding to the contribution value, the influence value, and the estimated value, respectively. Based on the weights corresponding to the contribution value, the influence value, and the estimated value, a weighted calculation is performed on the contribution value, the influence value, and the estimated value, and the result of the weighted calculation is used as the comprehensive value score of each first candidate class.

7. The intelligent scheduling method based on deep learning according to claim 3, characterized in that, The method of using arc consistency propagation to determine each third candidate shift that has constraints on employees assigned to each of the current second candidate shifts includes: The arc consistency propagation mechanism is used to obtain the influence range of employees who are constrained by the current allocation of each second candidate shift. Update the cumulative fatigue value and the number of consecutive working days for each employee according to the scope of the impact. Based on the updated accumulated fatigue value and the updated number of consecutive working days, determine the third candidate shifts for each employee with constraints in the current allocation of each second candidate shift.

8. The intelligent scheduling method based on deep learning according to claim 7, characterized in that, After determining the third candidate shifts for each of the current second candidate shifts that have employees with constraints, the method further includes: If the value range of each of the third candidate shifts is empty, the arc consistency propagation mechanism is used again to update the current scope of influence in order to adjust the third candidate shifts.

9. The intelligent scheduling method based on deep learning according to claim 1, characterized in that, After generating the global scheduling plan based on each of the third candidate shifts, the method further includes: If there are employees on leave, then obtain the affected date range and the shift of the employee on leave; Within the affected date range, calculate the matching similarity score between the remaining employees (excluding the employee on leave) and the shift of the employee on leave, and the cumulative fatigue value of the remaining employees; The global scheduling scheme is adjusted based on the matching similarity score between the shifts of the remaining employees and the shifts of the employees on leave, and the cumulative fatigue value of the remaining employees.

10. A deep learning-based intelligent scheduling system, characterized in that, include: The acquisition module is used to acquire employee file database and business task plan table; The first determining module is used to determine the skill tag set and fatigue accumulation value of each employee within the scheduling cycle based on the employee file database, and to determine the daily skill requirement matrix of the position based on the business task plan table. The second determining module is used to determine the matching similarity score between each employee and each shift in the scheduling cycle based on the skill tag set and the daily skill requirement matrix of the job. The third determining module is used to determine the employee variable ranking sequence based on the matching similarity score and the fatigue accumulation value, and to determine the employee variable to be decided based on the employee variable ranking sequence using a deep learning algorithm. The fourth determining module is used to obtain each first candidate shift corresponding to the employee to be decided variable, to determine the comprehensive value score of each first candidate shift, and to allocate the first candidate shift corresponding to the highest comprehensive value score to the employee to be decided variable. The update module is used to update each second candidate shift of employees associated with the employee to be decided variable according to the first candidate shift, and to use the arc consistency propagation mechanism to determine each third candidate shift of employees who have constraints on the current allocation of each second candidate shift. The generation module is used to generate a global scheduling plan based on each of the third candidate shifts, and in the global scheduling plan, if at least one employee has not been assigned a shift for any date within the scheduling period, the decision variable employee is updated again using a deep learning algorithm.

Images