An automated enterprise attendance scheduling system based on big data analysis

Abstract

The application discloses an automatic enterprise attendance scheduling system based on big data analysis, comprising: a data reading and processing module for obtaining business data and preprocessing to generate human resource historical data set; a prediction module for constructing an N-BEATS model and generating an enhanced prediction sequence through counterfactual reconstruction; a scheduling optimization module for constructing a branch pricing algorithm and generating an optimal scheduling result through a learnable heuristic guide and a dynamic constraint activation mechanism; a scheduling output module for mapping the optimal scheduling result into an employee scheduling table; and an intelligent attendance module for collecting attendance and comparing deviations to update the data set. The application realizes the collaborative promotion of accurate human demand prediction and intelligent scheduling optimization, and improves the overall scheduling quality, resource utilization rate and attendance management efficiency.

Description

Technical Field

[0001] This invention relates to the field of human resource management, and in particular to an automated enterprise attendance and scheduling system based on big data analysis. Background Technology

[0002] Attendance management and scheduling are fundamental to ensuring operational continuity and service levels. Currently, many companies still rely on manual experience to create schedules, manually compile attendance records, and adjust staffing levels based on subjective judgment. This approach struggles to handle the manpower needs of multiple positions, shifts, and business scenarios. Existing scheduling systems that incorporate conventional time series forecasting or simple rule-based forecasting have limited ability to identify disturbances such as holiday peaks, ad-hoc events, and attendance anomalies. The forecast results fluctuate significantly, leading to unstable manpower demand inputs and impacting the quality of subsequent scheduling.

[0003] Some solutions employ integer programming or heuristic algorithms to solve scheduling problems. These have fixed constraint structures, making fine-grained adjustments impossible for different positions and time periods. The solution scale expands rapidly with the number of employees and time periods, resulting in significant computational overhead. Branch-based pricing methods often solve pricing sub-problems for a large number of candidate shifts during column generation, lacking a learnable guidance mechanism based on historical solution behavior. Column generation paths lack filtering, resulting in numerous iterations and slow convergence. Attendance systems primarily focus on recording functions, making it difficult to feed attendance deviations back to prediction and scheduling models. This fails to create a closed loop that uses attendance deviations to update historical datasets and continuously refine prediction and scheduling strategies, hindering the adaptive evolution of human resource allocation with business rhythms.

[0004] Therefore, how to provide an automated enterprise attendance and scheduling system based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an automated enterprise attendance and scheduling system based on big data analysis. This invention utilizes the N-BEATS model, counterfactual reconstruction method, and branch pricing optimization algorithm, and introduces learnable pricing heuristics and fine-grained dynamic constraint control to achieve scheduling optimization. It has the advantages of good predictive stability and high solution efficiency.

[0006] An automated enterprise attendance and scheduling system based on big data analysis according to an embodiment of the present invention includes: The data reading and processing module is used to acquire business data and perform preprocessing to generate a historical human resources dataset; The prediction module is used to construct an N-BEATS model based on the historical human resources dataset, initialize the parameters of the forward subnetwork and the backward subnetwork, set the number of basic blocks and the number of stacking layers, perform forward propagation to generate a human resources demand prediction sequence, and generate counterfactual samples based on the historical perturbation sequence through the built-in counterfactual reconstruction structure to reconstruct the human resources demand prediction sequence and output an enhanced prediction sequence. The scheduling optimization module is used to build a branch pricing algorithm based on the enhanced prediction sequence. It sets the objective function and constraints of the main problem, generates an initial candidate solution set for scheduling, sets branch variables and constructs pricing sub-problems, integrates a learnable pricing heuristic structure to select paths for pricing sub-problems, integrates a dynamic activation structure to adaptively activate or freeze some constraint blocks, and generates the optimal scheduling result through branch pricing. The scheduling output module is used to map the optimal scheduling results into an employee schedule, and allocate them to each position and shift in chronological order to form the final scheduling execution plan; The intelligent attendance module is used to collect employee attendance information and generate attendance records. It compares and analyzes actual attendance behavior with employee work schedules, outputs attendance deviation information, and updates the human resources historical dataset.

[0007] Optionally, modules can be integrated using the following methods: S1. Acquire business data, perform preprocessing, and generate a unified time-series historical human resources dataset; S2. Construct the N-BEATS model, initialize the parameters of the forward and backward sub-networks, set the number of basic blocks and stacking layers, input the historical human resources dataset into the N-BEATS model, perform forward propagation, and generate a human resources demand prediction sequence. S3. Embed a counterfactual reconstruction module in the N-BEATS model. Generate counterfactual samples by comparing the human resource demand forecast sequence with the historical disturbance sequence. Perform structural comparison analysis and reconstruction on the human resource demand forecast sequence and output an enhanced forecast sequence. S4. Construct a branch pricing algorithm, set the objective function and constraints of the main problem, determine the human resource demand constraints based on the enhanced prediction sequence, generate an initial scheduling candidate solution set, set branch variables and construct pricing sub-problems; S5. Introduce a learnable pricing sub-problem heuristic guide into the branch pricing algorithm, train it on historical pricing problems, generate a heuristic scoring function, select the optimal sub-problem path, and output the current optimal scheduling solution. S6. Integrate a dynamic adaptive block activation mechanism into the branch pricing algorithm. Based on the structural characteristics and cost function gradient of the current optimal scheduling solution, adaptively activate or freeze some constraint blocks and output the converged optimal scheduling result. S7. Map the optimal scheduling results to the employee scheduling table, allocate them to each position and shift in time order, and generate the final scheduling execution plan.

[0008] Optionally, S2 specifically includes: S21. Set the number of layers in the forward and backward subnetworks, determine the data input and output dimensions used for prediction, and establish the overall structural framework of the N-BEATS model. S22. Initialize the weight and bias parameters of the forward and backward subnetworks, configure the structural parameters of the base blocks and stacked layers, and set the initial state of each computational unit inside the model. S23. The historical human resources dataset is input into the N-BEATS model in chronological order, and the forward propagation process is executed to output the human resources demand prediction sequence.

[0009] Optionally, S3 specifically includes: S31. Perform periodic comparisons on the business volume sequence in the historical human resources dataset, mark the time nodes where the difference from the mean of adjacent periods exceeds a preset threshold, and form a business fluctuation identification sequence. S32. Perform a line-by-line comparison of attendance records in the human resources historical dataset and mark the time nodes with absences, early departures, late arrivals, or continuous work exceeding the limit as attendance abnormality identifier sequences. S33. Perform statistics on the scheduling records in the historical human resources dataset, and mark the time nodes in which the difference between the actual number of people scheduled and the average number of people scheduled in the same period in history exceeds a preset threshold as scheduling deviation identification sequences. S34. Merge the business fluctuation identifier sequence, attendance exception identifier sequence and scheduling deviation identifier sequence according to the time index to form a historical disturbance sequence; S35. Process each time node of the prediction sequence according to the disturbance identifier in the historical disturbance sequence. By finding the business volume deviation value, attendance deviation value or scheduling deviation value recorded in the historical data for the corresponding time node, remove the deviation value from the original value of the prediction sequence to form a counterfactual sample sequence. S36. Compare the counterfactual sample sequence with the predicted sequence point by point, and replace the time nodes where the difference reaches the preset threshold to form an enhanced predicted sequence.

[0010] Optionally, S4 specifically includes: S41. Based on the enhanced forecast sequence, set the human resource demand constraints, take the scheduling requirements of each position and time period as the coverage conditions of the main problem, and establish the objective function and constraint set of the main problem. S42. According to the preset scheduling rules, combine feasible shift patterns for each position and time period to form an initial candidate solution set for scheduling, which serves as the initial column set for the main problem. S43. Select the scheduling variable as the branching condition, divide the feasible solution of scheduling into several sub-problems according to different value methods, and establish the initial node of the branch tree structure. S44. Based on the dual information of the current main problem, construct a pricing subproblem, perform simplified cost calculation on the shift pattern, generate candidate columns with improvement target values, and return to the main problem.

[0011] Optionally, the learnable pricing sub-problem heuristic guide structure specifically includes: A heuristic structure consisting of an input layer, several hidden layers, and an output layer is established. The input layer receives the dual variables corresponding to the pricing sub-problem, shift characteristic information, and shift coverage deviation information. The hidden layers combine and process the input data. By collecting the input features and corresponding simplified cost results of the pricing subproblems through the past solution records of the branch pricing algorithm, historical pricing samples are formed; the historical pricing samples are then input into the heuristic guided structure to execute the training process. The trained heuristic guidance structure is solidified into a heuristic scoring function.

[0012] Optionally, S5 specifically includes: S51. Before performing the pricing sub-problem, input the dual variables, shift characteristic information and shift coverage deviation information corresponding to the candidate columns into the heuristic scoring function to generate the score value of each candidate column, and sort the candidate columns according to the score value to form a candidate column sequence arranged from high to low. S52. Select candidate columns whose scores are above the preset score threshold from the sorted candidate column sequence, and use these candidate columns as the solution objects of the pricing subproblem. Exclude candidate columns whose scores are below the preset threshold from the pricing solution range. S53. In the process of solving the pricing sub-problem, the candidate columns are solved in order of score value to generate candidate columns, and the generated candidate columns are returned to the main problem for solving.

[0013] Optionally, S6 specifically includes: S61. According to the job category and scheduling time granularity, the constraint set is divided into several sub-constraint blocks, so that each sub-constraint block corresponds to the constraint conditions of a single job and a single time period, forming a fine-grained constraint block set. S62. During the iteration of the branch pricing algorithm, for each sub-constraint block corresponding to the job and time period, calculate the unmet coverage, overtime hours and job matching deviation under the sub-constraint block to form a set of local structural feature indicators. S63. Calculate the difference between the demand values ​​of the enhanced forecast sequence and the counterfactual sample sequence for the corresponding job and time period. When the difference of any sub-constraint block exceeds the preset deviation threshold, the sub-constraint block is marked as a demand fluctuation sub-block; when the difference does not exceed the deviation threshold, the sub-constraint block is marked as a demand stable sub-block. S64. For sub-constraint blocks whose local structural characteristic indicators exceed the set threshold or are marked as demand fluctuation sub-blocks, the sub-constraint blocks are written into the main problem for solution; for sub-constraint blocks whose local structural characteristic indicators do not exceed the threshold and are marked as demand stable sub-blocks, the sub-constraint blocks are not written into the main problem for solution. S65. Generate a dynamically updated fine-grained constraint structure based on the activation state of each sub-constraint block, and use the current constraint structure for the next solution of the main problem.

[0014] Optionally, S7 specifically includes: Write the optimal scheduling results into the scheduling output table according to job positions and time periods. Generate an employee scheduling table based on the employee number, job category, and shift time recorded in the scheduling output table. Arrange the generated employee scheduling table in chronological order and organize the job allocation of each employee in each time period into an executable scheduling execution plan.

[0015] Optionally, the intelligent attendance module is specifically used to collect employees' entry and exit records, attendance duration records, and on-duty status at various time periods. The collected attendance records are compared with the employee schedule generated by the scheduling output module for each time period to form attendance deviation information. The human resources historical dataset is updated based on the attendance deviation information.

[0016] The beneficial effects of this invention are: (1) By constructing a demand forecasting model based on deep structure and introducing a counterfactual reconstruction mechanism, the influence of business fluctuations, attendance abnormalities and scheduling deviations in historical data on the forecast results can be effectively removed, resulting in a more stable input of human resources demand and improving the accuracy and reliability of scheduling optimization.

[0017] (2) By introducing a learnable pricing subproblem heuristic structure and a fine-grained dynamic constraint activation mechanism through the branch pricing algorithm, the column generation path can be pre-screened and the importance of the constraint block can be adaptively adjusted, reducing the number of times the pricing subproblem is solved, improving the solution efficiency, and enabling the scheduling results to converge to the optimal solution more quickly.

[0018] (3) The intelligent attendance module compares the scheduling execution in real time and feeds the attendance deviation information back to the historical dataset, forming a closed-loop optimization process of prediction and scheduling. This enables the system to continuously correct the trend of human resource demand and scheduling strategy during continuous use, thereby improving the adaptability and long-term operation effect of the overall human resource allocation. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural diagram of an automated enterprise attendance and scheduling system based on big data analysis proposed in this invention. Figure 2 This is a data flow diagram of an automated enterprise attendance and scheduling system based on big data analysis proposed in this invention. Figure 3 This is a scheduling optimization structure diagram for an automated enterprise attendance scheduling system based on big data analysis proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figure 1-3 An automated enterprise attendance and scheduling system based on big data analytics includes: The data reading and processing module is used to acquire business data and perform preprocessing to generate a historical human resources dataset; The prediction module is used to construct an N-BEATS model based on the historical human resources dataset, initialize the parameters of the forward subnetwork and the backward subnetwork, set the number of basic blocks and the number of stacking layers, perform forward propagation to generate a human resources demand prediction sequence, and generate counterfactual samples based on the historical perturbation sequence through the built-in counterfactual reconstruction structure to reconstruct the human resources demand prediction sequence and output an enhanced prediction sequence. The scheduling optimization module is used to build a branch pricing algorithm based on the enhanced prediction sequence. It sets the objective function and constraints of the main problem, generates an initial candidate solution set for scheduling, sets branch variables and constructs pricing sub-problems, integrates a learnable pricing heuristic structure to select paths for pricing sub-problems, integrates a dynamic activation structure to adaptively activate or freeze some constraint blocks, and generates the optimal scheduling result through branch pricing. The scheduling output module is used to map the optimal scheduling results into an employee schedule, and allocate them to each position and shift in chronological order to form the final scheduling execution plan; The intelligent attendance module is used to collect employee attendance information and generate attendance records. It compares and analyzes actual attendance behavior with employee work schedules, outputs attendance deviation information, and updates the human resources historical dataset.

[0022] In this embodiment, the modules are interconnected using the following method: S1. Acquire business data, perform preprocessing, and generate a unified time-series historical human resources dataset; S2. Construct the N-BEATS model, initialize the parameters of the forward and backward sub-networks, set the number of basic blocks and stacking layers, input the historical human resources dataset into the N-BEATS model, perform forward propagation, and generate a human resources demand prediction sequence. S3. Embed a counterfactual reconstruction module in the N-BEATS model. Generate counterfactual samples by comparing the human resource demand forecast sequence with the historical disturbance sequence. Perform structural comparison analysis and reconstruction on the human resource demand forecast sequence and output an enhanced forecast sequence. S4. Construct a branch pricing algorithm, set the objective function and constraints of the main problem, determine the human resource demand constraints based on the enhanced prediction sequence, generate an initial scheduling candidate solution set, set branch variables and construct pricing sub-problems; S5. Introduce a learnable pricing sub-problem heuristic guide into the branch pricing algorithm, train it on historical pricing problems, generate a heuristic scoring function, select the optimal sub-problem path, and output the current optimal scheduling solution. S6. Integrate a dynamic adaptive block activation mechanism into the branch pricing algorithm. Based on the structural characteristics and cost function gradient of the current optimal scheduling solution, adaptively activate or freeze some constraint blocks and output the converged optimal scheduling result. S7. Map the optimal scheduling results to the employee scheduling table, allocate them to each position and shift in time order, and generate the final scheduling execution plan.

[0023] In this embodiment, S1 specifically includes: S11. Collect employee attendance records, historical shift records, employee attribute information, business volume data and date attribute information to form a raw business data set; S12. Perform data cleaning on the original business data set, remove invalid records, standardize the fields of records with inconsistent formats, and unify the coding of employee identifiers, job identifiers and time identifiers from different data sources. S13. Perform time alignment processing on the cleaned business data according to the preset time granularity, align the attendance records, shift records and business volume data on the same time axis, and fill in missing values ​​according to the conventional interpolation rules to form a complete and continuous data sequence. S14. The data sequences that have been time-aligned and missing data filled in are summarized and organized in chronological order, and attendance information, workload information and employee attribute information are combined to generate a human resources historical dataset.

[0024] In this embodiment, S2 specifically includes: S21. Set the number of layers in the forward and backward subnetworks, determine the data input and output dimensions used for prediction, and establish the overall structural framework of the N-BEATS model. S22. Initialize the weight and bias parameters of the forward and backward subnetworks, configure the structural parameters of the base blocks and stacked layers, and set the initial state of each computational unit inside the model. S23. The historical human resources dataset is input into the N-BEATS model in chronological order, and the forward propagation process is executed to output the human resources demand prediction sequence.

[0025] In this embodiment, the initialization process of the N-BEATS model specifically includes: The N-BEATS model has an input sequence length of 48 time steps and a prediction sequence length of 8 time steps. It has 6 base blocks and 3 stacked layers. Each base block contains 2 forward subnetworks and 2 backward subnetworks, with 128 neurons per layer. The activation function is ReLU. The weight parameters are initialized using the Xavier method, and the bias is initialized to 0. During forward propagation, the corresponding 8-time-step prediction sequence is generated according to the order in which the input sequence is processed by the base blocks and stacked layers.

[0026] In this embodiment, S3 specifically includes: S31. Perform periodic comparisons on the business volume sequence in the historical human resources dataset, and mark time nodes where the difference from the average of adjacent periods exceeds a preset threshold to form a business fluctuation identification sequence; the periodic comparison adopts a weekly cycle, and the difference between the current time node's business volume and the average business volume of the same period in the previous week is compared; when the difference exceeds a set threshold of 15%, it is marked as a business fluctuation node; S32. Perform a line-by-line comparison of attendance records in the human resources historical dataset, and mark the time nodes with absences, early departures, late arrivals, or continuous work exceeding the limit as attendance abnormality identifier sequences; when late arrivals exceed 10 minutes, early departures exceed 10 minutes, continuous work exceeds 3 shifts, or there is an absence of the entire shift, mark them as attendance abnormality nodes. S33. Statistically analyze the scheduling records in the historical human resources dataset, and mark the time nodes where the difference between the actual number of scheduled employees and the average number of scheduled employees in the same historical period exceeds a preset threshold as scheduling deviation identification sequences; when the difference between the actual number of scheduled employees and the average number of scheduled employees in the same historical period exceeds 10% of the actual number of scheduled employees, mark it as a scheduling deviation node. S34. Merge the business fluctuation identifier sequence, attendance exception identifier sequence and scheduling deviation identifier sequence according to the time index to form a historical disturbance sequence; S35. Process each time node of the predicted sequence according to the disturbance identifier in the historical disturbance sequence. By finding the business volume deviation value, attendance deviation value, or scheduling deviation value recorded in the historical data for the corresponding time node, remove the deviation value from the original value of the predicted sequence to form a counterfactual sample sequence. The business volume deviation value is obtained by the difference between the business volume of the current period and the average business volume of the same period in the previous period. The attendance deviation value is obtained by the difference between the actual attendance duration and the standard attendance duration of the current period. The scheduling deviation value is obtained by the difference between the actual number of people scheduled for the current period and the average number of people scheduled for the same period in history. When performing disturbance stripping processing, for each time node with a disturbance identifier, perform a subtraction operation on the corresponding value of the predicted sequence according to the business volume deviation value, attendance deviation value, or scheduling deviation value to obtain a counterfactual sample without abnormal fluctuations. S36. Compare the counterfactual sample sequence with the predicted sequence point by point, and perform replacement at time points where the difference reaches a preset threshold to form an enhanced predicted sequence. When the difference between the counterfactual sample sequence and the predicted sequence exceeds the set threshold of 10%, replace the predicted sequence value with the counterfactual sample value; if the difference does not exceed the threshold, keep the predicted sequence unchanged.

[0027] In this embodiment, S4 specifically includes: S41. Based on the enhanced forecast sequence, set the human resource demand constraints, take the scheduling requirements of each position and time period as the coverage conditions of the main problem, and establish the objective function and constraint set of the main problem. S42. According to the preset scheduling rules, combine feasible shift patterns for each position and time period to form an initial candidate solution set for scheduling, which serves as the initial column set for the main problem. S43. Select the scheduling variable as the branching condition, divide the feasible solution of scheduling into several sub-problems according to different value methods, and establish the initial node of the branch tree structure. S44. Based on the dual information of the current main problem, construct a pricing subproblem, perform simplified cost calculation on the shift pattern, generate candidate columns with improvement target values, and return to the main problem.

[0028] In this embodiment, the construction of the branch pricing algorithm specifically includes: The main problem adopts a scheduling coverage model, which obtains the minimum personnel demand for each position and each time period through enhanced forecast sequences, and uses the minimum personnel demand as the scheduling coverage constraint; the boundary of the position demand constraint is the rounded-up value of the enhanced forecast; the main problem constraints are as follows: The manpower coverage constraint is that the number of selected shifts for a certain position and a certain time period in the shift schedule is greater than or equal to the demand value corresponding to the enhanced forecast sequence, and the number is rounded up. Each employee has a maximum daily working time limit. If a shift is longer than 4 hours, it cannot be added to the initial column. If the interval between two shifts is less than 1 hour, they cannot be undertaken by the same employee. Job qualification constraints: Employees who have not obtained the corresponding job qualifications will not have their corresponding job schedules generated. The initial scheduling candidate solution, i.e. the candidate column of the branch pricing algorithm, needs to be constructed according to the combination of job and time period. The initial shift set consists of all feasible shifts that meet the constraints of shift length and job qualification. When constructing the branch tree, a binary decision variable is used as the branch variable, which takes the value of 0 or 1, indicating whether to include a certain shift column in the scheduling solution. The pricing subproblem is constructed based on dual information, and the goal is to find columns with negative simplification costs. When the cost of a shift minus the weighted value corresponding to the dual variable is less than 0, the current shift is considered to be included in the master problem. When the simplification cost of all generable shift columns is greater than or equal to 0, the main problem converges, and the current shift schedule is obtained.

[0029] In this embodiment, the learnable pricing sub-problem heuristic guide structure specifically includes: A heuristic structure consisting of an input layer, several hidden layers, and an output layer is established. The input layer receives the dual variables corresponding to the pricing sub-problem, shift characteristic information, and shift coverage deviation information. The hidden layers combine and process the input data. By collecting the input features and corresponding simplified cost results of the pricing subproblems through the past solution records of the branch pricing algorithm, historical pricing samples are formed; the historical pricing samples are then input into the heuristic guided structure to execute the training process. The trained heuristic guidance structure is solidified into a heuristic scoring function.

[0030] In this embodiment, S5 specifically includes: S51. Before performing the pricing sub-problem, input the dual variables, shift characteristic information and shift coverage deviation information corresponding to the candidate columns into the heuristic scoring function to generate the score value of each candidate column, and sort the candidate columns according to the score value to form a candidate column sequence arranged from high to low. S52. Select candidate columns whose scores are above a preset score threshold from the sorted candidate column sequence, and use these candidate columns as the solution objects for the pricing subproblem. Exclude candidate columns whose scores are below the preset threshold from the pricing solution range. In this embodiment, the score values ​​are normalized, and the preset threshold is 0.2. S53. In the process of solving the pricing sub-problem, the candidate columns are solved in order of score value to generate candidate columns, and the generated candidate columns are returned to the main problem for solving.

[0031] In this embodiment, S6 specifically includes: S61. According to the job category and scheduling time granularity, the constraint set is divided into several sub-constraint blocks, so that each sub-constraint block corresponds to the constraint conditions of a single job and a single time period, forming a fine-grained constraint block set. S62. During the iteration of the branch pricing algorithm, for each sub-constraint block corresponding to the job and time period, calculate the unmet coverage, overtime hours and job matching deviation under the sub-constraint block to form a set of local structural feature indicators. S63. Calculate the difference between the demand values ​​of the enhanced forecast sequence and the counterfactual sample sequence for the corresponding job and time period. When the difference of any sub-constraint block exceeds the preset deviation threshold, the sub-constraint block is marked as a demand fluctuation sub-block; when the difference does not exceed the deviation threshold, the sub-constraint block is marked as a demand stable sub-block. S64. For sub-constraint blocks whose local structural characteristic indicators exceed the set threshold or are marked as demand fluctuation sub-blocks, the sub-constraint blocks are written into the main problem for solution; for sub-constraint blocks whose local structural characteristic indicators do not exceed the threshold and are marked as demand stable sub-blocks, the sub-constraint blocks are not written into the main problem for solution. S65. Generate a dynamically updated fine-grained constraint structure based on the activation state of each sub-constraint block, and use the current constraint structure for the next solution of the main problem.

[0032] In this embodiment, the calculation process of the local structural feature index set specifically includes: Read the number of employees required for the corresponding position in the corresponding time period from the sub-constraint block in the enhanced prediction sequence, and then count the number of employees already assigned to the same position and the same time period from the current scheduling solution; if the number of employees required is greater than the number of employees already assigned, the difference between the two is taken as the unmet coverage; if the number of employees required is less than or equal to the number of employees already assigned, the unmet coverage is recorded as zero; set the threshold for unmet coverage to 1. In the process of calculating the amount of overtime, for each employee assigned to the sub-constraint block, the cumulative working hours of the day in the current scheduling solution are counted; the daily working hour limit is predetermined by the system; in this implementation, it is 8 hours. When the employee's cumulative working hours of the day exceed the daily working hour limit, the excess part is the employee's overtime. The overtime of all employees in the sub-constraint block is summed, and the total amount is the amount of overtime of the sub-constraint block. Determine the range of job qualifications for each employee based on employee information; check each assigned employee to see if they have the required job qualifications; if an employee in a certain scheduling record does not have the job qualifications, record the current scheduling record as a job matching deviation; if the employee has the qualifications, record it as zero; sum the deviation values ​​corresponding to all scheduling records in the sub-constraint block to form the job matching deviation amount. When any indicator exceeds its respective threshold, the unmet coverage reaches or exceeds 1, the overtime exceeds 1, or the job matching deviation is greater than 0, the structural characteristic indicator of the sub-constraint block is considered to exceed the set threshold; the sub-constraint block will be included in the priority activation range.

[0033] In this embodiment, S7 specifically includes: Write the optimal scheduling results into the scheduling output table according to job positions and time periods. Generate an employee scheduling table based on the employee number, job category, and shift time recorded in the scheduling output table. Arrange the generated employee scheduling table in chronological order and organize the job allocation of each employee in each time period into an executable scheduling execution plan.

[0034] In this embodiment, the intelligent attendance module is specifically used to collect employees' entry and exit records, attendance duration records, and on-duty status at various time periods. The collected attendance records are compared with the employee schedule generated by the scheduling output module for each time period to form attendance deviation information. The human resources historical dataset is updated based on the attendance deviation information.

[0035] Example 1: To verify the feasibility of this invention in practice, it was applied to a daily operational scenario. This scenario involves 35 temporary employees participating in shift scheduling, with each employee having a maximum daily shift duration of 6 hours. Each employee can handle different job types, and job demands fluctuate significantly throughout the day. Business volume data, attendance records, historical shift scheduling data, employee attribute information, and date characteristics were collected over 30 consecutive days. Through data reading and cleaning steps, a historical human resources dataset suitable for direct modeling was generated. Historical data shows that there is typically a shortage of staff during the peak business period from 9:00 AM to 11:00 AM, while there is often a surplus of staff from 2:00 PM to 4:00 PM. Some employees have accumulated instances of working consecutive hours exceeding the stipulated duration in historical shifts, leading to an increased probability of subsequent absences.

[0036] After inputting the aforementioned historical data into the system of this invention, the prediction module first generates a basic prediction sequence, and then constructs counterfactual samples based on historical disturbances to eliminate abnormal business fluctuations caused by occasional factors such as holidays, making the prediction sequence more stable. The reconstructed enhanced prediction sequence becomes the input for scheduling optimization. The scheduling optimization module uses a branch pricing algorithm to establish a solution framework and utilizes a learnable heuristic method to filter candidate columns, reducing the number of solutions. The dynamic constraint activation mechanism determines whether to activate the corresponding constraint block based on the manpower deviation of different positions and time periods, making the scheduling solution process more targeted.

[0037] After the system generates the optimal scheduling result based on the enhanced prediction sequence and constraint structure, the scheduling output module organizes the scheduling result into an employee scheduling table according to job position and shift. In this embodiment, the scheduling time for each employee in the scheduling solution is controlled within 6 hours, and there are no cases of continuous work exceeding the limit. After collecting the actual attendance data on the second day, the intelligent attendance module compares the attendance data with the scheduling table and generates 4 deviation information entries to update the historical dataset and provide input for the next round of prediction.

[0038] To compare the data differences between manual scheduling and the method of this invention in practical applications, the traditional scheduling scheme and the scheduling scheme of this invention were compared on the same day. In the traditional scheme, the average shortage during peak hours was 1.6 people, the average redundancy during off-peak hours was 3.2 people, there were 19 instances of job matching deviation, 7 instances of continuous overwork exceeding limits, the average scheduling time per employee was 5.4 hours, there were 13 attendance deviation records, and the average prediction error was 16.5%. Under the scheduling scheme of this invention, the peak shortage was reduced to 0 people, the off-peak redundancy was 0.9 people, the job matching deviation was 3 times, the continuous overwork exceeding limits was reduced to 0 times, the average scheduling time was 4.9 hours, there were 4 attendance deviation records, and the average prediction error was 6.3%. These data demonstrate the system's comprehensive performance in terms of manpower prediction stability, job allocation rationality, scheduling accuracy, and attendance feedback loop.

[0039] The following comparison table illustrates the differences between manual scheduling and the scheduling scheme of this invention.

[0040] Table 1: Comparison of Scheduling Effectiveness

[0041] As shown in Table 1, the method of this invention can eliminate manpower shortages during peak hours and reduce redundant personnel during off-peak hours under the same number of employees and the same scheduling time constraints, while achieving a lower number of deviations in job matching. The scheduling results generated by the system show a more balanced distribution of working hours, fewer attendance deviation records, and a lower mean prediction error, indicating that the system has high stability and applicability in handling business fluctuations, planning job requirements, and integrating attendance feedback. By continuously updating the historical dataset, the method of this invention can gradually improve the quality of scheduling in subsequent operations, ensuring a closer alignment between actual scheduling needs and business models.

[0042] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An automated enterprise attendance and scheduling system based on big data analysis, characterized in that, include: The data reading and processing module is used to acquire business data and perform preprocessing to generate a historical human resources dataset; The prediction module is used to construct an N-BEATS model based on the historical human resources dataset, initialize the parameters of the forward subnetwork and the backward subnetwork, set the number of basic blocks and the number of stacking layers, perform forward propagation to generate a human resources demand prediction sequence, and generate counterfactual samples based on the historical perturbation sequence through the built-in counterfactual reconstruction structure to reconstruct the human resources demand prediction sequence and output an enhanced prediction sequence. The scheduling optimization module is used to build a branch pricing algorithm based on the enhanced prediction sequence. It sets the objective function and constraints of the main problem, generates an initial candidate solution set for scheduling, sets branch variables and constructs pricing sub-problems, integrates a learnable pricing heuristic structure to select paths for pricing sub-problems, integrates a dynamic activation structure to adaptively activate or freeze some constraint blocks, and generates the optimal scheduling result through branch pricing. The scheduling output module is used to map the optimal scheduling results into an employee schedule, and allocate them to each position and shift in chronological order to form the final scheduling execution plan; The intelligent attendance module is used to collect employee attendance information and generate attendance records. It compares and analyzes actual attendance behavior with employee work schedules, outputs attendance deviation information, and updates the human resources historical dataset.

2. The automated enterprise attendance and scheduling system based on big data analysis according to claim 1, characterized in that, The modules are connected in the following way: S1. Acquire business data, perform preprocessing, and generate a unified time-series historical human resources dataset; S2. Construct the N-BEATS model, initialize the parameters of the forward and backward sub-networks, set the number of basic blocks and stacking layers, input the historical human resources dataset into the N-BEATS model, perform forward propagation, and generate a human resources demand prediction sequence. S3. Embed a counterfactual reconstruction module in the N-BEATS model. Generate counterfactual samples by comparing the human resource demand forecast sequence with the historical disturbance sequence. Perform structural comparison analysis and reconstruction on the human resource demand forecast sequence and output an enhanced forecast sequence. S4. Construct a branch pricing algorithm, set the objective function and constraints of the main problem, determine the human resource demand constraints based on the enhanced prediction sequence, generate an initial scheduling candidate solution set, set branch variables and construct pricing sub-problems; S5. Introduce a learnable pricing sub-problem heuristic guide into the branch pricing algorithm, train it on historical pricing problems, generate a heuristic scoring function, select the optimal sub-problem path, and output the current optimal scheduling solution. S6. Integrate a dynamic adaptive block activation mechanism into the branch pricing algorithm. Based on the structural characteristics and cost function gradient of the current optimal scheduling solution, adaptively activate or freeze some constraint blocks and output the converged optimal scheduling result. S7. Map the optimal scheduling results to the employee scheduling table, allocate them to each position and shift in time order, and generate the final scheduling execution plan.

3. The automated enterprise attendance and scheduling system based on big data analysis according to claim 2, characterized in that, S2 specifically includes: S21. Set the number of layers in the forward and backward subnetworks, determine the data input and output dimensions used for prediction, and establish the overall structural framework of the N-BEATS model. S22. Initialize the weight and bias parameters of the forward and backward subnetworks, configure the structural parameters of the base blocks and stacked layers, and set the initial state of each computational unit inside the model. S23. The historical human resources dataset is input into the N-BEATS model in chronological order, and the forward propagation process is executed to output the human resources demand prediction sequence.

4. The automated enterprise attendance and scheduling system based on big data analysis according to claim 3, characterized in that, S3 specifically includes: S31. Perform periodic comparisons on the business volume sequence in the historical human resources dataset, mark the time nodes where the difference from the mean of adjacent periods exceeds a preset threshold, and form a business fluctuation identification sequence. S32. Perform a line-by-line comparison of attendance records in the human resources historical dataset and mark the time nodes with absences, early departures, late arrivals, or continuous work exceeding the limit as attendance abnormality identifier sequences. S33. Perform statistics on the scheduling records in the historical human resources dataset, and mark the time nodes in which the difference between the actual number of people scheduled and the average number of people scheduled in the same period in history exceeds a preset threshold as scheduling deviation identification sequences. S34. Merge the business fluctuation identifier sequence, attendance exception identifier sequence and scheduling deviation identifier sequence according to the time index to form a historical disturbance sequence; S35. Process each time node of the prediction sequence according to the disturbance identifier in the historical disturbance sequence. By finding the business volume deviation value, attendance deviation value or scheduling deviation value recorded in the historical data for the corresponding time node, remove the deviation value from the original value of the prediction sequence to form a counterfactual sample sequence. S36. Compare the counterfactual sample sequence with the predicted sequence point by point, and replace the time nodes where the difference reaches the preset threshold to form an enhanced predicted sequence.

5. The automated enterprise attendance and scheduling system based on big data analysis according to claim 4, characterized in that, S4 specifically includes: S41. Based on the enhanced forecast sequence, set the human resource demand constraints, take the scheduling requirements of each position and time period as the coverage conditions of the main problem, and establish the objective function and constraint set of the main problem. S42. According to the preset scheduling rules, combine feasible shift patterns for each position and time period to form an initial candidate solution set for scheduling, which serves as the initial column set for the main problem. S43. Select the scheduling variable as the branching condition, divide the feasible solution of scheduling into several sub-problems according to different value methods, and establish the initial node of the branch tree structure. S44. Based on the dual information of the current main problem, construct a pricing subproblem, perform simplified cost calculation on the shift pattern, generate candidate columns with improvement target values, and return to the main problem.

6. The automated enterprise attendance and scheduling system based on big data analysis according to claim 5, characterized in that, The learnable pricing sub-problem heuristic guide structure specifically includes: A heuristic structure consisting of an input layer, several hidden layers, and an output layer is established. The input layer receives the dual variables corresponding to the pricing sub-problem, shift characteristic information, and shift coverage deviation information. The hidden layers combine and process the input data. By collecting the input features and corresponding simplified cost results of the pricing subproblems through the past solution records of the branch pricing algorithm, historical pricing samples are formed; the historical pricing samples are then input into the heuristic guided structure to execute the training process. The trained heuristic guidance structure is solidified into a heuristic scoring function.

7. The automated enterprise attendance and scheduling system based on big data analysis according to claim 6, characterized in that, S5 specifically includes: S51. Before performing the pricing sub-problem, input the dual variables, shift characteristic information and shift coverage deviation information corresponding to the candidate columns into the heuristic scoring function to generate the score value of each candidate column, and sort the candidate columns according to the score value to form a candidate column sequence arranged from high to low. S52. Select candidate columns whose scores are above the preset score threshold from the sorted candidate column sequence, and use these candidate columns as the solution objects of the pricing subproblem. Exclude candidate columns whose scores are below the preset threshold from the pricing solution range. S53. In the process of solving the pricing sub-problem, the candidate columns are solved in order of score value to generate candidate columns, and the generated candidate columns are returned to the main problem for solving.

8. The automated enterprise attendance and scheduling system based on big data analysis according to claim 7, characterized in that, S6 specifically includes: S61. According to the job category and scheduling time granularity, the constraint set is divided into several sub-constraint blocks, so that each sub-constraint block corresponds to the constraint conditions of a single job and a single time period, forming a fine-grained constraint block set. S62. During the iteration of the branch pricing algorithm, for each sub-constraint block corresponding to the job and time period, calculate the unmet coverage, overtime hours and job matching deviation under the sub-constraint block to form a set of local structural feature indicators. S63. Calculate the difference between the demand values ​​of the enhanced forecast sequence and the counterfactual sample sequence for the corresponding job and time period. When the difference of any sub-constraint block exceeds the preset deviation threshold, the sub-constraint block is marked as a demand fluctuation sub-block; when the difference does not exceed the deviation threshold, the sub-constraint block is marked as a demand stable sub-block. S64. For sub-constraint blocks whose local structural characteristic indicators exceed the set threshold or are marked as demand fluctuation sub-blocks, the sub-constraint blocks are written into the main problem for solution; for sub-constraint blocks whose local structural characteristic indicators do not exceed the threshold and are marked as demand stable sub-blocks, the sub-constraint blocks are not written into the main problem for solution. S65. Generate a dynamically updated fine-grained constraint structure based on the activation state of each sub-constraint block, and use the current constraint structure for the next solution of the main problem.

9. The automated enterprise attendance and scheduling system based on big data analysis according to claim 8, characterized in that, Specifically, S7 includes: Write the optimal scheduling results into the scheduling output table according to job positions and time periods. Generate an employee scheduling table based on the employee number, job category, and shift time recorded in the scheduling output table. Arrange the generated employee scheduling table in chronological order and organize the job allocation of each employee in each time period into an executable scheduling execution plan.

10. An automated enterprise attendance and scheduling system based on big data analysis according to claim 9, characterized in that, The intelligent attendance module is specifically used to collect employees' entry and exit records, attendance duration records, and on-duty status at various time periods. It compares the collected attendance records with the employee schedule generated by the scheduling output module for each time period to form attendance deviation information; and updates the human resources historical dataset based on the attendance deviation information.

Images