A data-driven service personnel scheduling optimization method and system

Abstract

The present application relates to artificial service operation management technical field, especially to a kind of service personnel clock optimization method and system based on data driving.The method includes the following steps: obtaining historical service personnel clock-in data;Clock-in feature extraction is carried out according to historical service personnel clock-in data, and clock-in feature data is obtained;Clock-out same frequency processing is carried out according to clock-in feature data, and clock-out same frequency data is obtained;Clock conflict processing is carried out according to clock-out same frequency data, and clock conflict data is obtained;Clock sequence disturbance is carried out according to clock conflict data, and clock disturbance data is obtained;Cooling window control is carried out according to clock disturbance data, and cooling window data is obtained;Dispersed degree processing is carried out on clock disturbance data according to cooling window data, and clock optimization data is obtained.The present application realizes dynamic perception and intelligent optimization to service personnel clock process.Compared with traditional static scheduling mode, resource conflict can be effectively relieved, and service continuity and fairness can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial service operation management, and particularly relates to a service personnel clocking optimization method and system based on data driving. BACKGROUND

[0002] In the artificial service industry such as foot massage, body beautifying physiotherapy and the like, the clocking scheduling of service personnel generally adopts artificial experience or simple shift mechanism, and common practices include shift sequence rotation, clocking according to check-in sequence and the like. The traditional way has certain feasibility in actual operation, but it is often difficult to cope with dynamic scenarios such as service intensity fluctuation and customer demand intensive change, and is prone to cause low service resource scheduling efficiency, and to show problems such as uneven distribution, frequent clocking conflicts and the like in peak period or multi-project concurrent service. For example, without rhythm control or conflict identification mechanism, service arrangement between different rooms is prone to resource overlap, thereby affecting the continuity and stability of overall operation. SUMMARY

[0003] The present application provides a service personnel clocking optimization method and system based on data driving to solve at least one of the above technical problems.

[0004] The present application provides a service personnel clocking optimization method based on data driving, which comprises the following steps:

[0005] S1, obtaining historical service personnel clocking data; performing clocking feature extraction according to the historical service personnel clocking data to obtain clocking feature data;

[0006] S2, performing clocking same-frequency processing according to the clocking feature data to obtain clocking same-frequency data; performing clocking conflict processing according to the clocking same-frequency data to obtain clocking conflict data;

[0007] S3, performing clocking sequence disturbance according to the clocking conflict data to obtain clocking disturbance data;

[0008] S4, performing cooling window control according to the clocking disturbance data to obtain cooling window data; performing dispersion degree processing on the clocking disturbance data according to the cooling window data to obtain clocking optimization data.

[0009] The application can effectively alleviate the phenomena of "clock collision", "crowding", "unfairness" and the like in the high-concurrency clock arrangement scene of service personnel, and realize intelligent clock arrangement control with better rhythm perception and resource balance. By modeling the characteristics of historical clock data, multi-dimensional indicators such as service rhythm, clock connection structure and idle residual error are extracted, and the system can accurately identify potential clock synchronization groups and conflict risks; combined with the rhythm disturbance mechanism and the cooling window strategy, the high-overlapping clock sequence is effectively dispersed, and the spatial-time dispersion of the service sequence is improved; the application has feedback regulation capability, supports disturbance correction according to real-time rhythm density, and thus realizes high-robustness clock arrangement optimization control for multiple rooms, multiple seats and multiple service objects.

[0010] Optionally, S1 comprises:

[0011] The historical service personnel clock-in data is acquired;

[0012] Service cycle calculation, average service time calculation, clock connection length calculation and idle residual time calculation are performed according to the historical service personnel clock-in data, and service cycle data, average time data, clock connection length data and idle residual time data are obtained respectively;

[0013] Idle cycle rebound index calculation is performed according to the service cycle data and the average service time data, and idle cycle rebound index data is obtained;

[0014] Clock connection attenuation residual index calculation is performed according to the clock connection length data and the idle residual time data, and clock connection attenuation residual index data is obtained;

[0015] Clock-in item extraction is performed according to the historical service personnel clock-in data, and clock-in item data is obtained;

[0016] Project adaptation degree processing is performed on the clock-in item data according to the idle cycle rebound index data and the clock connection attenuation residual index data, and clock-in feature data is obtained.

[0017] In the application, basic data such as service cycle, average service time, clock connection behavior and idle residual time are extracted, and a quantitative index system of service rhythm and load state is constructed. Through the "idle cycle rebound index" and the "clock connection attenuation residual index", two behavior measurement indicators with dynamic perception ability, the ability of the service personnel to recover from continuous work state to stable rhythm and the degree of rhythm distortion caused by clock connection can be reflected, thereby avoiding the static and fragmented understanding of service state in traditional feature modeling. Through the project adaptation degree processing mechanism, each clock-in task is matched with the current rhythm state of the service personnel, the clock arrangement rationality and individual scheduling sensitivity are effectively improved, and a high-resolution data foundation support is provided for subsequent rhythm intervention and conflict avoidance.

[0018] Optionally, the idle cycle rebound index calculation comprises:

[0019] According to the service cycle data and the average service time length data, time residual calculation is performed to obtain time residual data;

[0020] According to the time residual data, cycle rebound trend calculation is performed to obtain cycle rebound trend data;

[0021] According to the cycle rebound trend data and the time residual data, idle cycle rebound index calculation is performed to obtain idle cycle rebound index data.

[0022] In the present application, the service cycle data and the average service time length data of the service personnel are subjected to differential residual modeling, and time residual data is constructed, so as to capture the fluctuation characteristics of individuals under different service rhythms. Through cycle rebound trend calculation, the recovery mode of the service rhythm under multiple time windows can be dynamically tracked, and the trend law of the service personnel from the high load state to the regular rhythm is identified. The idle cycle rebound index generated by combining the time residual and the rebound trend can quantify the rhythm recovery elasticity of individuals to the clock disturbance, and can be used as an index representing the clock tolerance and the scheduling rhythm matching ability.

[0023] Optionally, the project adaptation degree processing includes:

[0024] According to the idle cycle rebound index data and the clock-continuous attenuation residual index data, project load matrix construction is performed on the clock-on project data to obtain project load matrix data;

[0025] According to the historical service personnel clock-on data, service personnel state extraction is performed to obtain service personnel state data;

[0026] According to the project load matrix data and the service personnel state data, project adaptation risk calculation is performed to obtain clock-on feature data.

[0027] In the present application, the idle cycle rebound index and the clock-continuous attenuation residual index are fused to construct a project load matrix reflecting the disturbance degree of different service projects on individual rhythm, and the influence degree of various clock-on tasks on service personnel resource consumption and rhythm stability is described. The current state of the service personnel (such as clock-continuous history, current fatigue degree or rhythm stability) is extracted by combining historical service data, so as to realize dynamic state coupling between the service supply side and the project demand side. Through project adaptation risk calculation, unreasonable clock matching combinations are identified, so as to warn potential high rhythm impact risk and service imbalance risk in the clock scheduling stage. The present application breaks through the traditional coarse-grained clock scheduling mode depending only on availability or queuing order, and effectively improves the rhythm adaptation and load balance of service personnel clock scheduling distribution through fine-grained rhythm load modeling and individual recovery ability perception.

[0028] Optionally, the clock-off same frequency processing includes:

[0029] According to historical service personnel clock-in data and clock-in feature data, a clock-in reply point data is obtained by identifying the clock-in reply point;

[0030] According to the clock-in reply point data, a rhythm regression trajectory data is obtained by constructing the rhythm regression trajectory;

[0031] According to the rhythm regression trajectory data, a preliminary clock-synchronous data is obtained by sliding window clustering;

[0032] According to the historical service personnel clock-in data, a room service record data is obtained by extracting the room service record;

[0033] According to the room service record data, a room clock-out data is obtained by mapping the room clock-out;

[0034] According to the room clock-out data, a concurrent clock-out block data is obtained by extracting the convolution kernel;

[0035] According to the concurrent clock-out block data, a clock-out synchronous cluster data is obtained by extracting the service personnel synchronous cluster from the preliminary clock-synchronous data.

[0036] In the present application, a time-space joint perception clock-out synchronous identification mechanism is constructed by a two-dimensional modeling strategy of rhythm regression trajectory and space concurrent clock-out block. Based on historical clock-in data and clock-in feature data, the rhythm recovery nodes of clock-in are identified, the rhythm regression trajectory of individual service personnel is established, and the potential rhythm synchronization group is extracted by sliding window clustering to avoid misjudgment caused by only relying on static time points. Starting from the physical space dimension, the room service record is extracted, the clock-out density atlas is constructed by extracting the convolution kernel, and the service concurrent end group in the same room or adjacent area is effectively identified. Based on the fusion of concurrent clock-out block and rhythm trajectory, the high-confidence service personnel synchronous cluster is extracted to realize the deep perception and labeling of implicit synchronous risk.

[0037] Optionally, the service personnel synchronous cluster extraction includes:

[0038] According to the concurrent clock-out block data, a concurrent block marker data is obtained by marking and mapping the preliminary clock-synchronous data;

[0039] According to the concurrent block marker data, a trajectory phase difference data is obtained by calculating the phase difference of the rhythm regression trajectory;

[0040] According to the trajectory phase difference data, a synchronization graph data is obtained by constructing the synchronization graph from the historical service personnel clock-in data;

[0041] According to the synchronization graph data, a local movement graph data is obtained by local movement processing;

[0042] The local movement graph data is subjected to fine-grained sub-cluster reconstruction to obtain sub-cluster reconstruction data;

[0043] According to the sub-group reconstruction data, the aggregated graph reconstruction is performed, and the same frequency data of the next bell is obtained.

[0044] In the application, the rhythm phase difference driven service personnel synchronization graph is constructed, and the high-precision identification of potential implicit synchronization clusters is realized by combining the graph structure optimization technology, so that the fine granularity and interpretability of the same frequency identification of the next bell are significantly improved. The system firstly performs concurrent block label mapping on the preliminary same frequency data of the next bell, introduces the rhythm regression trajectory phase difference as a synchronization feature measurement, and quantifies the rhythm convergence between service personnel. Then, a weighted synchronization graph is constructed, the service personnel are represented by graph nodes, and the rhythm similarity is reflected by edge weight, so that the graph expression of the synchronization structure is realized. Through local movement processing and sub-group reconstruction operation, the weak connection structure in the original group is broken, the local optimal synchronization sub-group is identified, and the stability and continuity of the graph division are improved. The same frequency cluster extraction of the next bell with more organizational structure is realized through graph aggregation reconstruction. The application not only breaks through the adaptability limitation of traditional static time period clustering to rhythm sliding, but also can accurately restore the real service synchronization relationship in the complex scene of spatial synchronization and rhythm disturbance intersection.

[0045] Optionally, the bell arrangement conflict processing includes:

[0046] According to the same frequency data of the next bell, view angle extraction is performed to obtain view angle data, wherein the view angle extraction includes project view angle extraction, region view angle extraction and space view angle extraction;

[0047] According to the view angle data, a cross-room conflict matrix is constructed to obtain cross-room conflict data;

[0048] According to the cross-room conflict data, conflict intensity propagation processing is performed to obtain conflict propagation data;

[0049] According to the conflict propagation data, conflict grouping is performed to obtain bell arrangement conflict data.

[0050] In the application, the system is based on the same frequency data of the next bell, and fuses multiple dimensions such as project type, service region and physical space to construct behavior slice data under multiple perspectives to represent potential conflict factors of service personnel in task attributes, geographical arrangement and spatial overlap. By constructing a cross-room conflict matrix, local room conflicts and cross-region resource congestion problems are uniformly included in the graph structure expression, key indicators such as service object sharing rate, service project overlap degree and target area conflict coefficient are quantified, and a high-dimensional conflict atlas is formed. The conflict intensity propagation processing is adopted, the weight propagation path is introduced to simulate the diffusion trend and overlap influence of potential conflicts, and the dynamic description of conflict chains and cross bottlenecks is realized. The conflict grouping is performed by graph division or community clustering, and the bell arrangement conflict data set with scheduling intervention priority is output.

[0051] Optionally, S3 includes:

[0052] According to the clock conflict data, a clock conflict association graph is constructed, and clock conflict association graph data is obtained;

[0053] The clock conflict association graph data is subjected to high conflict cluster identification, and high conflict cluster data is obtained;

[0054] The high conflict cluster data is subjected to disturbance priority division, and disturbance priority data is obtained;

[0055] The disturbance priority data is subjected to jump disturbance, and jump disturbance data is obtained;

[0056] According to the jump disturbance data, disturbance constraint verification is performed, and clock disturbance data is obtained.

[0057] In the present application, the system first constructs a conflict association graph with the clock conflict relationship between service personnel as the edge and the personnel node as the vertex. On this basis, a high conflict cluster is identified by using a graph clustering algorithm to accurately locate the personnel group with high-frequency competition in the same period or adjacent period. Through a disturbance priority division mechanism, combined with conflict intensity, historical clock load and rhythm rebound ability, a differentiated adjustment weight is allocated to each conflict cluster member to realize hierarchical disturbance decision. The jump disturbance strategy breaks through the limitations of traditional sequential fine-tuning, breaks the conflict concentrated structure through local segment jump rearrangement, maintains rhythm continuity, and enhances scheduling flexibility. To ensure the implementability of the results, disturbance constraint verification is performed, and multiple rounds of screening and back compensation are performed according to hard constraints such as working time limit, personnel skill matching and room resource capacity, so as to output clock disturbance data that meets the system execution standard.

[0058] Optionally, S4 comprises:

[0059] According to the clock disturbance data, disturbance node intensive area data is extracted, and disturbance node intensive area data is obtained;

[0060] The disturbance node intensive area data and the preset cooling window template data are subjected to cooling marking, and cooling window data is obtained;

[0061] According to the cooling window data, the clock disturbance data is subjected to disturbance delay extraction, and disturbance delay data is obtained;

[0062] The disturbance delay data and the preset service combination reconstruction table are subjected to service combination reconstruction, and clock optimization data is obtained.

[0063] In the present application, by extracting the data of the dense area of disturbance nodes, the system can identify the hotspot area with the risk of time and space aggregation after the clock adjustment; then, the preset cooling template is used for cooling marking to realize the rhythm peak cutting processing of the service intensity dense section; the disturbance delay extraction mechanism can dynamically delay the scheduling time of part of the nodes to form the rhythmical peak-shifting and position-shifting arrangement; combined with the service combination reconstruction table, the service project is reorganized and the personnel task is recombined to improve the space utilization and load balancing in the service period.

[0064] Optionally, the present application also provides a data-driven service personnel clock optimization system for executing the data-driven service personnel clock optimization method as described above, and the data-driven service personnel clock optimization system comprises:

[0065] a clock-on feature extraction module for obtaining historical service personnel clock-on data; performing clock-on feature extraction according to the historical service personnel clock-on data to obtain clock-on feature data;

[0066] a clock conflict identification module for performing clock-on same frequency processing according to the clock-on feature data to obtain clock-on same frequency data; performing clock conflict processing according to the clock-on same frequency data to obtain clock conflict data;

[0067] a clock sequence disturbance module for performing clock sequence disturbance according to the clock conflict data to obtain clock disturbance data;

[0068] a rhythm cooling dispersion module for performing cooling window control according to the clock disturbance data to obtain cooling window data; performing dispersion degree processing on the clock disturbance data according to the cooling window data to obtain clock optimization data.

[0069] The purpose of the present application is to quantitatively depict the rhythm recovery ability of service personnel and the residual fatigue after the continuous bell by the idle cycle rebound index and the continuous bell attenuation residual index in addition to the traditional statistical characteristics, output the bell feature data with enhanced project adaptation, and provide fine parameter basis for determining who can arrange, arrange, and suitable arrangement. The rhythm regression trajectory and sliding window clustering time sequence method are combined with the spatial method of room bell convolution kernel extraction to identify the same frequency of the bell, and then the service personnel synchronous cluster extraction is used to avoid the misjudgment caused by time proximity, realize the high confidence same frequency detection of time-space-behavior multi-perspective, and significantly reduce the probability of implicit bell collision in subsequent scheduling. The conflict correlation graph is constructed by the conflict data, and the executability is guaranteed by disturbance constraint checking, the key conflict source is fixed and dispersed, the ordering overlap and conflict degree are significantly reduced under the premise of ensuring availability and skill matching, and the fairness and ordering stability are improved. Based on the disturbance result, a cooling window is generated, and the queue is re-optimized by dispersion degree (such as ordering dispersion entropy and local clustering degree LCI), forming a feedback closed loop of disturbance-cooling-dispersion, which can still maintain rhythm balance and optimal resource utilization in the scene of high-concurrency passenger flow and multi-project mixed arrangement across rooms. BRIEF DESCRIPTION OF DRAWINGS

[0070] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings:

[0071] Figure 1 A step flow chart of a data-driven service personnel bell arrangement optimization method of an embodiment is shown;

[0072] Figure 2 A step flow chart of a bell feature extraction method of an embodiment is shown;

[0073] Figure 3 A step flow chart of a bell same frequency processing method of an embodiment is shown;

[0074] Figure 4 A step flow chart of a bell arrangement sequence disturbance method of an embodiment is shown;

[0075] Figure 5 A step flow chart of a rhythm cooling dispersion method of an embodiment is shown;

[0076] The implementation of the purpose of the present application, functional characteristics and advantages will be further described with reference to the accompanying drawings. DETAILED DESCRIPTION

[0077] The technical method of the present application will be described clearly and completely in combination with the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0078] In addition, the drawings are only schematic illustrations of the present application, and are not necessarily drawn to scale. The functional entities can be implemented in software, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0079] It should be understood that although the terms "first", "second" and the like can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the example embodiments, a first element can be referred to as a second element, and similarly a second element can be referred to as a first element. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0080] Referring to Figures 1 to 5 The present application provides a data-driven service personnel clock-in optimization method, which comprises:

[0081] S1, obtaining historical service personnel clock-in data; performing clock-in feature extraction according to the historical service personnel clock-in data to obtain clock-in feature data;

[0082] Specifically, the system calls the service records in the database in the past 30 days, including the fields of the clock-on time of each service staff, service items, service duration, and the room where the service staff is located. The average and variance of the service cycle of each service staff are calculated (such as the average service cycle of A service staff is 65 minutes); the length of the clock-on sequence is extracted (such as the continuous clock-on sequence with an idle time less than a preset threshold (for example, 10 minutes) between adjacent two clock-on times is regarded as a clock-on, and three continuous clock-ons are regarded as one clock-on unit); the idle recovery time residual after each clock-on is extracted (such as the theoretical rest time is 30 minutes, but the actual rest time is only 15 minutes); the idle cycle rebound index and the clock-on attenuation residual index are calculated based on the foregoing, such as the idle cycle rebound index is used to represent whether the service cycle of the service staff is pulled back to the stable rhythm of the service staff after experiencing a service by using the idle period. The clock-on attenuation residual index is used to represent whether the remaining fatigue caused by continuous clock-on is fully released, which is constructed based on the deviation degree of the idle residual time after each clock-on from the standard recovery time, for example, the average value of the absolute value of the residual, the residual ratio, or the attenuation trend of the residual in the time sequence is calculated. The statistical quantities obtained above are combined with the dynamic indicators to form the clock-on feature data used for subsequent clocking calculation. The clock-on feature data at least includes the average and variance of the service cycle, the average service duration, the average (and maximum) clock-on length, the idle residual time statistics, the idle cycle rebound index, the clock-on attenuation residual index, and the service item adaptation degree derived therefrom.

[0083] S2, according to the clock-on feature data, performing the same frequency processing of the next clock to obtain the same frequency data of the next clock; according to the same frequency data of the next clock, performing the clocking conflict processing to obtain the clocking conflict data;

[0084] Specifically, the system predicts the time period of the next clock-on (i.e., the back clock) of each service staff based on the extracted clock-on feature data, such as the service cycle, the idle cycle rebound index, and the clock-on attenuation residual index, and the end time of the service of each service staff in the current clocking period. For example, if the average service cycle of a service staff is 65 minutes, and the last clock-on time is 9:35, the system can calculate that the next clocking of the service staff is concentrated in the time window between 10:35 and 10:50, forming the back clock rhythm time period. The system clusters and identifies the back clock rhythm of all service staff in the predicted window, performs time bucket processing on the back clock time point by using the sliding time window method, and clusters the back clock time point into several clock-on synchronous clusters based on the window overlap degree and the time density distribution.

[0085] The system extracts room service trajectory data of each service staff from historical service records, including the room number and time period served. For service staff currently at the end of service, the system extracts the room location and combines the bell prediction time to construct a room-time two-dimensional distribution map. On this two-dimensional distribution map, the system uses a sliding convolution kernel operation to calculate the spatio-temporal density of the room location, identifying areas where multiple service staff are close in space and close in bell time. This operation is defined as room bell convolution. For example, if service staff A and B both end their service in rooms 301 and 302 around 10:45, the system marks them as a spatio-temporal bell synchronization block and records its spatial center, number of participants, and time overlap.

[0086] The system performs dimensional decomposition on the current clustering results based on the extracted bell item information and bell target area information, including item perspective extraction, which identifies the skill overlap between clustered service staff in service items, such as service staff A and B both being good at "shoulder and neck essential oil" items; and area perspective extraction, which extracts the expected bell target room or area attributes, such as both being located in the "core dynamic line area" or high-traffic areas. The above information can be matched by constructing a service staff-item bipartite graph and a service staff-area bipartite graph. Matching the data extracted based on the perspectives, the system obtains conflict service staff group identification, conflict time period, conflict room cluster, and service item label.

[0087] S3, bell sequence perturbation based on bell conflict data to obtain bell perturbation data;

[0088] Specifically, for the high-conflict pair identified in the previous step (e.g., service staff A and B), the system constructs a conflict graph, representing the rhythm overlap, item overlap, and room intersection between service staff nodes with edge weights. The system identifies the group with the highest conflict intensity in the conflict cluster, i.e., taking service staff as nodes, and connecting edges between any two people if there is rhythm overlap, service item overlap, room use intersection, etc., and assigning conflict weights. Through edge weight threshold pruning and connectivity analysis, the system extracts strongly connected components in the conflict graph as candidate conflict clusters. For each conflict cluster, the system calculates the average edge weight between all members as the conflict intensity indicator of the cluster. Selecting the cluster with the highest average conflict intensity as the group of personnel that needs to be dispersed and optimized the most in the current bell scheme, such as A-B-C forming a complete bell synchronization group; the system sets the perturbation priority based on the historical service characteristics of each service staff in the conflict cluster (e.g., prioritizing long, bouncy service staff), and performs a jump operation, i.e., delaying the order of service staff C by 2 positions. After perturbation, it checks whether it meets the conditions of continuous bell limit, customer demand matching, cooling window, etc., and generates executable bell perturbation data.

[0089] S4, cooling window control according to the clock disturbance data, to obtain cooling window data; and dispersion degree processing on the clock disturbance data according to the cooling window data, to obtain clock optimization data.

[0090] Specifically, the system generates a personalized cooling window for each service staff according to the obtained clock disturbance result. The cooling window of the service staff is denoted as . The system can set it as the product of the minimum rebound index (i.e. the minimum value or lower quantile value of the idle period rebound index) of the staff in the current period and the expected service interval (preset value or input value): , wherein is the cooling window of the th service staff, is a scaling coefficient configurable by the system, used to adjust the cooling strength in different stores / at different times, is the minimum value, is the idle period rebound index of the th service staff, is the expected service interval. The system generates cooling window data accordingly, including the window start and end time of each service staff, the corresponding restriction strength (soft / hard constraint), the trigger reason (such as high-density disturbance, low rebound index, and large clock residual) and the like.

[0091] The system reviews the disturbed clock sequence one by one. If the clock time of any two service staffs (for example, service staff B and service staff C) is still in the high-sensitivity period covered by each other's cooling window, and the time difference is less than the preset threshold (for example, only 5 minutes apart), the system further adjusts the order of at least one of them so that the clock-on time difference of the two people is not less than the cooling window length of each of them or the minimum interval threshold set by the system (the larger one). The adjustment can be completed in the form of forward shift, backward shift, insertion of low-conflict service staff or replacement of room, etc. After performing the peak-shifting operation, the system counts the "time-personnel distribution" of the global clock queue, and quantitatively represents the dispersion degree and fairness of the queue according to the following indexes, such as calculating the sorting dispersion entropy, dividing the entire clock period into several time periods (such as fixed-width sliding window), and calculating the proportion of the number of clock-on staff in each time period to the total number of staff, then the sorting dispersion entropy is: , wherein is the clock sorting dispersion entropy, is the time period (window) index, is the proportion of the number of clock-on staff in the th time period to the total number of staff. The fairness index is calculated as , wherein is the fairness index, is the service staff sequence, is the number of service staffs, for service staff At a certain resource allocation amount under the current optimization scheme (such as the number of clock-in times obtained within an observation period, the effective service time length, or the project benefit weight), the system forms dispersion and fairness evaluation data based on the above-mentioned threshold judgment, which is used to judge whether the current scheme meets the preset dispersion and fairness constraints.

[0092] If the system determines that the current scheme still has obvious concentration according to the evaluation results (for example, the proportion of clock-in people in a certain time period exceeds the threshold or the fairness index is lower than the set lower bound of fairness), the system performs lightweight fine-tuning again according to the perturbation constraint boundary. The perturbation constraint boundary includes but is not limited to not breaking the confirmed customer appointment period and project type matching relationship; not violating the clock-in upper limit, working time upper limit, and cooling window limit of service staff; not allowing any service staff's fairness index (such as the proportion of clock-in times) to break through the preset range. The system performs iterative fine-tuning within this constraint framework, aiming to minimize the newly added delay time, minimize the conflict intensity increment, maximize the dispersion entropy, or maximize the fairness index. When any (or multiple) target reaches the threshold value or reaches the maximum number of iterations, the fine-tuning is stopped, and the clock optimization data is output.

[0093] Optionally, S1 includes:

[0094] S11, obtaining historical service staff clock-in data;

[0095] Specifically, the historical records of the past 30 days are extracted from the service system of the store, including service staff ID, clock-in time, clock-out time, service project type, room number, and customer ID.

[0096] S12, performing service period calculation, average service time length calculation, clock length calculation, and idle residual time calculation according to the historical service staff clock-in data, respectively obtaining service period data, average time length data, clock length data, and idle residual time data;

[0097] Specifically, for each service staff, sort all clock-in records by time; calculate the interval between adjacent clock-in times (such as 10:00→11:10 for 70min), obtain the service period; take the mean of all service time lengths (such as each service time length is 45min), obtain the average service time length; count the number of clock-in segments with continuous service and interval <10min (such as three consecutive services with a clock length of 3), obtain the clock length; take the expected / preset recovery time of 30min as the basis, calculate the actual recovery time deviation (such as actually only resting for 12min, residual deviation is -18min), obtain the idle residual time.

[0098] S13, calculate the idle cycle rebound index according to the service cycle data and the average service duration data, and obtain idle cycle rebound index data;

[0099] Specifically, a service cycle sequence (e.g., [65, 70, 62, 68, 90] min) is constructed for each service personnel; a time residual sequence Δt = cycle - average length is constructed by comparing the average service duration (e.g., 45 min); the variability and recovery trend of the cycle residual (e.g., standard deviation, regression slope) are calculated; and the idle cycle rebound index is calculated: , wherein is the idle cycle rebound index of the kth service personnel, is the standard deviation of the time residual sequence, is the regression slope of the time residual sequence, is the average service duration, is the index of the service time sequence.

[0100] S14, calculate the continuous bell attenuation residual index according to the continuous bell length data and the idle residual time data, and obtain continuous bell attenuation residual index data;

[0101] Specifically, for each group of continuous bells (e.g., service personnel A continuously serving 3 units), the recovery time after each service is calculated; if the recovery time continues to shorten (e.g., 30 min → 18 min → 6 min), it indicates that there is excessive overdraft; the residual attenuation rate is calculated and the index is constructed: , wherein is the continuous bell attenuation residual index, indicating the recovery abnormality degree of the kth service personnel, is the number of continuous bell services (i.e., the number of units served in the continuous bell group), is the index number in the continuous bell, is the kth recovery time, is the standard rest time.

[0102] S15, extract the on-duty item according to the historical service personnel on-duty data, and obtain on-duty item data;

[0103] ​​Specifically, extract the service item types each service staff has served in the historical clock records; take the service staff as the row and the service item as the column, and form a sparse two-dimensional frequency matrix, where the matrix element represents the historical frequency of the service staff performing the item, to obtain a service staff-item matrix (for example, service staff A has served "essential oil massage" 45 times and "cupping" 20 times); for each item, based on historical data or expert evaluation, maintain a set of service load parameters such as average duration, physical strength score, physiological consumption coefficient, etc., and construct an item load dictionary; associate the service staff-item matrix with the average service load of the item (for example, "cupping" has a short average duration but high physiological consumption), that is, through the docking of the matrix and the load dictionary.

[0104] S16, according to the idle period rebound index data and the continuous clock attenuation residual index data, performing item adaptation degree processing on the on-clock item data to obtain on-clock feature data.

[0105] Specifically, in combination with the IRRI (rebound index) and DRDI (decay index) of each service staff, an "item-service staff adaptation load matrix" is constructed: if the IRRI of service staff A is high and the DRDI is low, it can adapt to intensive items; if the IRRI is low and the DRDI is high, it tends to arrange recovery type items or low load services; by calculation (such as weighted calculation with a preset weight), an item adaptation risk value is generated, such as the risk of item X to service staff A is 0.23 (low) and to B is 0.78 (high).

[0106] Optionally, the idle period rebound index calculation comprises:

[0107] According to the service period data and the average service duration data, time residual data is calculated to obtain the time residual data;

[0108] Specifically, input the service period data (such as the service interval of service staff A after continuous service in a week: [65, 70, 62, 90] minutes); average service duration data (such as the historical average service duration of service staff A is 45 minutes); for each service period, calculate the difference residual between it and the average service duration: , is the difference residual, is the service period, is the average service duration.

[0109] According to the time residual data, the period rebound trend is calculated to obtain the period rebound trend data;

[0110] Specifically, take the sequence time as the horizontal axis and the residual as the vertical axis, and fit the trend curve, such as linear regression, polynomial fitting, Savitzky-Golay filtering; for example, linear fitting is adopted: wherein For the first Time residual value for the next service cycle The slope of the rebound trend. For service periodic sequence index, The trend intercept term is used to fit the slope of the rebound trend. If a < 0, it means the rest rhythm is shortening (the rebound speed is increasing); if a > 0, it means the rhythm is lengthening (the rhythm is disordered).

[0111] The idle cycle rebound index is calculated based on the cycle rebound trend data and time residual data to obtain the idle cycle rebound index data.

[0112] Specifically, the idle cycle rebound index is calculated as follows: ,in This refers to the idle period rebound index. The standard deviation of the residuals. It is the hyperbolic tangent function. This represents the rhythm trend offset. A higher value indicates that the service staff are regularly restoring the rhythm, which is suitable for a fast-paced clocking schedule; a lower value indicates a disordered rhythm, and buffer items or cooling-off periods should be arranged appropriately.

[0113] Optionally, the project adaptation processing includes:

[0114] Based on the idle cycle rebound index data and the continuous clock attenuation residual index data, the project load matrix is ​​constructed from the clocking project data to obtain the project load matrix data.

[0115] Specifically, based on the data from the service project, static parameters of the service project side are extracted (obtainable through historical statistics or manual annotation), such as the average service duration of the project, physical strength / intensity sensitivity (obtained by extracting indicators from a project characteristic library based on a preset expert engine or expert knowledge), and the recovery difficulty coefficient of the project (the average time spent by the service personnel from completing one project to starting the next, divided by the overall average time spent between projects). The system can process any service personnel... With any service item The combination defines its load value. This value represents the execution pressure of the project under the current status of the service personnel. A higher value indicates a heavier burden on the service personnel currently assigned to the project, making it less suitable for priority allocation. The project load cost can be weighted and composed of the following three parts: ,in The project workload-based value represents the expected burden and pressure on service personnel when performing the project. For service duration load weighting coefficient, The average service time for the project, The service personnel's rhythm rebound index. is a stable small value, such as 0.01, is a project intensity sensitivity weight coefficient, is a project physical strength / intensity sensitivity, is a service staff clock attenuation residual error index, is a project recovery difficulty weight coefficient, is a project recovery difficulty coefficient. After the system performs the formula calculation on all service staff-project combinations, a complete two-dimensional project load matrix data is formed.

[0116] According to the historical service staff clock data, the service staff state extraction is performed to obtain service staff state data;

[0117] Specifically, the service staff state data includes behavior rhythm data, load / fatigue data, and skill / preference data, wherein the behavior rhythm data includes average service period, period variance, idle residual mean, IRRI, and CDRI; the load / fatigue data includes average clock length, longest continuous clock record, and number of cooling window triggers in the past 7 days (i.e., the task is delayed, changed room, or reduced load (project replacement / splitting) due to cooling mask, which is considered as a trigger); and the skill / preference data includes project proficiency coefficient (such as project service time / average employee project service time) and average completion time residual.

[0118] According to the project load matrix data and the service staff state data, the project adaptation risk calculation is performed to obtain clock-on feature data.

[0119] Specifically, the system extracts a group of state adjustment factors from the fatigue and rhythm characteristics in the service staff state vector to represent the state degradation degree of the current service staff, which specifically includes idle recovery ability, represented by , the smaller the IRRI, the worse the recovery elasticity; clock residual fatigue intensity, represented by , the larger the value, the more fatigue; and clock continuous working intensity, represented by , which is the historical average number of consecutive clock-on times, the larger the value, the heavier the load. The state adjustment factor is formed by linear weighting, and the calculation formula is as follows: , wherein is an idle recovery ability weight coefficient, taking a value of 0.4, is a clock continuous working intensity weight coefficient, taking a value of 0.4, is a clock continuous working intensity weight coefficient, taking a value of 0.2, and the weight coefficient is a preset value, which can also be set according to platform experience or training results.

[0120] The system fuses the current state of the service personnel with the project load matrix to form the adaptation risk value of each service personnel for each service project. The adaptation risk value (Risk) represents the load intensity of performing the project in the current state, and the higher the value, the less suitable it is. The specific calculation formula is as follows: wherein is the adaptation risk value of the service personnel performing the project , is the load cost of the service personnel performing the project , is the current state adjustment factor. The system compresses and maps the above risk value to obtain the adaptation degree wherein is the adaptation degree of the service personnel performing the project , is a natural exponential function, is a compression coefficient, which controls the score attenuation slope and can be set in the range of [0.01, 0.05] or dynamically determined through online learning, is the adaptation risk value of the service personnel performing the project .

[0121] Optionally, the off-duty same frequency processing includes:

[0122] S21, identifying the off-duty reply point according to historical service personnel on-duty data and on-duty feature data to obtain off-duty reply point data;

[0123] Specifically, for each service personnel after the last off-duty time , the off-duty reply time is calculated according to wherein is the off-duty reply time, is the end time point of the service personnel completing the last service task, is the recovery time interval, is the quick recovery item coefficient, is the instant recovery index, is the cumulative fatigue item coefficient, is the cumulative fatigue index. The system predicts the off-duty reply point based on the foregoing calculation. For example, service personnel A (IRRI=0.6, CDRI=0.2, ), the last off-duty time is 10:20, then minutes, so the off-duty reply point is approximately 10:38.

[0124] S22, constructing a rhythm regression track according to the off-duty reply point data to obtain rhythm regression track data;

[0125] Specifically, a time series trajectory is constructed for each service staff , where is the end time of the service task, is the recoverable time after the service, is the sequence index, is the total number of valid service recovery segments. In the time axis, it is represented by , ]→recovery vector, and multiple periods are spliced into a rhythm recovery trajectory. For example, service staff A records the following three times: 09:05→09:22, 09:55→10:12, and 10:20→10:38 (this time), and the trajectory [09:22, 10:12, 10:38] is obtained.

[0126] S23, sliding window clustering is performed according to the rhythm recovery trajectory data to obtain preliminary clock synchronization frequency data;

[0127] Specifically, the sliding window width W=5 minutes and the step size 1 minute are set; the of all service staff is projected onto the time axis, and the number of service staff in each window is counted; or the is directly subjected to density clustering (neighbor maximum time difference eps=5min, minimum cluster member number min_samples=2). For example, A (10:38), B (10:41), and C (10:36) are clustered into the same cluster (center 10:39), and the preliminary synchronization frequency cluster T-1={A,B,C} is obtained.

[0128] S24, room service record extraction is performed according to historical service staff clock-in data to obtain room service record data;

[0129] Specifically, (service staff ID, room ID, clock-in / clock-out time) is extracted from historical clock-in data; a room-time-clock-out event list is generated.

[0130] S25, room clock-out mapping is performed according to the room service record data to obtain room clock-out data;

[0131] Specifically, a two-dimensional matrix of room time is constructed, and the element is the number of clock-out staff in the room in that minute.

[0132] S26, convolution kernel extraction is performed according to the room clock-out data to obtain concurrent clock-out block data;

[0133] Specifically, a 3 3 or 1 5 is convolved (room x time) sliding, statistics of the number of events in the clock. If the convolution result > set threshold (such as ≥ 2), it is considered that the region is concurrent clock block. For example, in (room 301, time 10:18-10:22) kernel aggregation to 2 people (A, D), mark concurrent block CCB-301-1; in (room 302, 10:18) single point density is 1, does not constitute a concurrent block.

[0134] S27, according to the concurrent clock block data, the preliminary clock same frequency data is extracted, and the clock same frequency data is obtained.

[0135] Specifically, the preliminary same frequency cluster (based on time / rhythm) is crossed with the concurrent clock block (based on spatial convolution): if the historical clock event of the same frequency cluster member falls into the same concurrent block, the synchronization confidence is improved, that is, the system performs confidence accumulation for each preliminary same frequency cluster member: the initial confidence is set to 1, and the confidence is increased every time the same cluster member falls into the same concurrent clock block in history; the synchronization graph of service personnel-service personnel (edge weight = time synchronization degree spatial concurrency degree); the time synchronization degree is the closeness of the clock time between two service personnel. The system extracts the clock recovery time points of any two service personnel in the same day or the same clock period, matches the time difference value, and calculates the average time interval. The system normalizes the time interval, converts it into a synchronization score, and the higher the score indicates that the clock time of two people is closer, with strong rhythm synchronization; the system divides the historical clock events of all service personnel according to time period, and constructs a service occupation matrix of "time x room". For any two service personnel A and B, if their clock events fall into the same time period (such as 5-minute sliding window), and the service location is in the same room or adjacent room, it is considered as a spatial concurrent event. Spatial concurrency degree is the proportion of all clock events of two people that meet the concurrency condition.

[0136] The system performs high threshold screening based on the synchronization edge weight data between service personnel (which can be generated by time synchronization degree and space concurrency degree weighting), and retains edges with edge weight higher than a preset seed threshold. The system constructs multiple connected subgraphs according to these high-intensity edges, and each subgraph represents a group of service personnel with high consistency in rhythm and potential coupling in space, which is called an initial seed cluster. For each seed cluster, the system identifies its boundary nodes, that is, external nodes that have connections with members in the cluster but have not yet joined the cluster. The system evaluates the consistency gain brought by the incorporation of each boundary node into the cluster in turn. The gain consists of two parts: one is the average rhythm-space consistency between the node and existing cluster members, and the other is the degree of spatial overlap between the node and cluster members. If a node is highly consistent in time rhythm but has too high a degree of spatial overlap with existing cluster members, the system will assign a negative penalty score to the node. When the gain value of a node exceeds the minimum threshold set by the system, the system will include it in the current cluster, and repeat the process until the cluster can no longer be expanded, and output the synchronization cluster.

[0137] Optionally, the service personnel synchronization cluster extraction comprises:

[0138] According to the concurrent bell block data, the preliminary bell synchronization frequency data is marked and mapped in concurrent blocks to obtain concurrent block label data;

[0139] Specifically, the preliminary bell synchronization frequency data (such as service personnel A / B / C belonging to cluster T-1) is input; the concurrent bell block data (such as block name CCB-301-1, containing service personnel A and D, time 10:18-10:22); the bell-off event of each preliminary cluster member is mapped to a concurrent block; if the bell-off time falls within the block time window, a concurrent label is assigned; for example, A belongs to both T-1 and CCB-301-1→marked as (T-1, CCB-301-1); B is not in any concurrent block→marked as (T-1, NULL).

[0140] According to the concurrent block label data, rhythm regression trajectory phase difference calculation is performed to obtain trajectory phase difference data;

[0141] Specifically, the system takes the historical continuous recovery time sequence of each service personnel (i.e. the idle time point or recovery window start time point after its continuous bell-on) as input to construct the rhythm regression trajectory of the service personnel. Each trajectory can be represented as a sequence of time stamps: , wherein is the rhythm trajectory of service personnel i, is the th recovery start time of service personnel i, is the th recovery start time of service personnel i, is the th recovery start time of service personnel i, the first recovery start time of service personnel i, the second recovery start time of service personnel i, the number of time points within the trajectory. Similarly, the trajectory of service personnel j is which can be extracted based on the system recorded clock-out time and cooling window distribution. For the service personnel pairs marked by the system as concurrent blocks (i.e., there are time overlaps or service interval overlaps between the service personnel groups), the system performs sliding window alignment (in this method, the system compares the time difference of the two service personnel time series item by item in a window alignment manner, and calculates the average absolute difference value) or periodic model alignment (the system can fit the trajectory of each service personnel as a periodic function (such as a cosine function or a Fourier series), and extract the phase offset between the two functions through a phase item comparison method) on their rhythm trajectories, and calculates the average phase difference between them. The system outputs the trajectory phase difference between all service personnel pairs marked as concurrent blocks, forming a symmetric matrix.

[0142] According to the trajectory phase difference data, a synchronization graph is constructed based on the historical service personnel clock-in data, and synchronization graph data is obtained;

[0143] Specifically, the system constructs an undirected weighted graph G=(V,E), where V represents a node set consisting of all service personnel; E represents a synchronization edge between any two service personnel. For any pair of nodes i and j in the graph, the corresponding edge weight is calculated according to the following method: wherein is the synchronization edge weight, is a natural exponential term, is a phase difference compression coefficient for controlling the amplitude of exponential decay, and the value is 0.1, is the rhythm trajectory phase difference between service personnel and service personnel , in minutes, and the smaller the value, the higher the synchronization degree, is a concurrent score weighting coefficient, and the value range is 0.5-1.0, is a concurrent block marking variable, which takes a value of 1 if there is a spatial concurrent relationship between the two, and 0 otherwise; the aforementioned coefficients are not fixed and can be adjusted according to actual conditions.

[0144] According to the synchronization graph data, local movement processing is performed to obtain local movement graph data;

[0145] Specifically, the synchronization graph is input into a graph neural network model (such as GCN or GAT), with service personnel as graph nodes and edge weights representing their synchronization strength (such as time overlap rate of jointly completing services, room sharing frequency, etc.). The initial input vector includes service rhythm vectors (such as project frequency per unit time, task average period); spatial activity (such as the number of rooms, room switching frequency); synchronization frequency features (such as the number of times of appearing together with adjacent service personnel); the graph neural network updates the embedding representation of each service personnel through multiple rounds of adjacent convolution operations to capture the synchronization relationship pattern with adjacent service personnel. In the initial stage, each service personnel is a separate cluster; in each iteration, for each node, the system checks the highest synchronization edge weight of the node among its first-order adjacent nodes, and attempts to move the node to the cluster where the highest synchronization edge weight of the node is located; if the total modularity improves by more than a certain threshold after moving, the movement is accepted; the maximum number of movements and the minimum cluster size (to prevent micro-cluster proliferation) are set, and the process is terminated when a stable state is reached or early stopping conditions are met. The output results include the service personnel set of each cluster, the average synchronization strength (average edge weight) within the cluster, and the embedded vector mean square error (reflecting the consistency within the cluster) of each cluster, which constitute the local mobile graph data.

[0146] Refine the local mobile graph data to obtain sub-group reconstruction data.

[0147] Specifically, the system extracts the rhythm trajectory vector of each service personnel (such as consecutive service period, clock-on time interval); based on the normalized trajectory data, the K-means clustering algorithm is applied within the same local mobile graph data; the elbow rule or silhouette coefficient is used to determine the optimal number of sub-groups K; service personnel with similar trajectory characteristics are assigned to the same sub-group.

[0148] According to the sub-group reconstruction data, the aggregated graph is reconstructed to obtain the clock-on same frequency data.

[0149] Specifically, a non-directed graph is constructed between sub-groups where each node represents a sub-group; for each pair of sub-groups, the edge weight is calculated as follows: where is the edge weight between sub-groups, is the rhythm consistency weight, is the rhythm trajectory correlation (such as the inverse of the average phase difference, trajectory correlation coefficient, etc.) between sub-groups and , is the spatial conflict penalty weight, is the spatial overlap degree between two sub-group members (such as the overlap probability of service rooms).

[0150] For the constructed sub-group graph, the system constructs the attraction degree matrix and the repulsion degree matrix between the service sub-groups. The attraction degree matrix is the rhythm consistency between the service sub-groups, i.e., the rhythm similarity of the service personnel, such as the similarity of the bell time or the bell item. The repulsion degree matrix is the spatial intersection degree of the service personnel, i.e., extracting the room occupancy records (such as room ID and service period) of each sub-group in the historical service process. For any sub-group pair, it is counted whether the bell time falls into the same time window (such as a 5-minute sliding window) and whether the rooms coincide. The spatial conflict degree is set as the proportion of these conflict events in the total service events. The corresponding degree matrix of the above two matrices is calculated, i.e., the total connection strength of each sub-group in the attraction or repulsion. Based on these matrices, the system further constructs a normalized Laplacian matrix with a repulsion regularization term, wherein the main part is used to maintain the aggregation of the rhythm similar sub-groups, and the regularization term guides the sub-groups with serious spatial coincidence away by adjusting the repulsion weight parameter. The system performs eigenvalue decomposition on the normalized Laplacian matrix, extracts the eigenvectors corresponding to the first k smallest eigenvalues, maps each service sub-group to a k-dimensional vector, and forms a low-dimensional expression in the embedding space. In the low-dimensional embedding space, the system performs clustering operation with non-same cluster constraint. For the sub-group pair with spatial coincidence degree exceeding the set threshold, the system sets that it cannot be assigned to the same cluster in the clustering process. Such limitation is realized by introducing a penalty term into the objective function or directly shielding the corresponding pairing relationship in the cluster assignment process. The clustering process is realized by using K-Means with constraint or restricted spectral clustering. After clustering, the system performs consistency and spatial safety review operation on each synchronization cluster. The average rhythm correlation within each cluster is calculated, and if it is lower than the rhythm consistency threshold, the cluster is determined to be out of rhythm. Then the maximum spatial coincidence degree within each cluster is calculated, and if it is higher than the spatial safety threshold, the cluster is determined to exist spatial conflict. For the cluster that does not meet the above standards, the system marks it as a weak cluster and re-executes the feature embedding and clustering process within the cluster range until all clusters meet the dual standards of rhythm synchronization and spatial separation, or reach the maximum refinement iteration number set by the system.

[0151] Optionally, the bell conflict processing includes:

[0152] According to the bell time same frequency data, the view angle is extracted to obtain view angle data, wherein the view angle extraction includes item view angle extraction, region view angle extraction and space view angle extraction.

[0153] Specifically, project perspective extraction involves inputting service project information (such as massage, foot massage, aromatherapy, etc.); service personnel offering the same service within the same time window are grouped into project perspective groups; for example, service personnel A, B, and C performing foot massage simultaneously from 10:20 to 0:30 form a project subset. Regional perspective extraction involves inputting the operating region to which the service personnel belong (such as the East Zone, VIP Zone); services within the same operating region are grouped into regional perspective groups; for example, service personnel D and E both belong to the East Zone and are grouped into the same perspective. Spatial perspective extraction involves inputting room location number, floor, etc.; physically adjacent rooms (such as room numbers 101 and 102) or rooms sharing a common passageway are considered spatially adjacent; the output room perspective groups are (e.g., {A in 101, B in 102} → belonging to the same spatially adjacent group).

[0154] A cross-room conflict matrix is ​​constructed based on the split-view data to obtain cross-room conflict data;

[0155] Specifically, constructing a two-dimensional conflict matrix , Assign room number or service personnel number; if two service personnel belong to rooms in the same spatial perspective group and the service end time difference is <5min, mark it as a potential conflict; ,in For the first The end time of service for each service staff member. For the first The end time of service for each service staff member.

[0156] Conflict intensity propagation processing is performed on cross-room conflict data to obtain conflict propagation data;

[0157] Specifically, a conflict graph is constructed, where nodes represent service personnel and edge weights represent conflict intensity. Based on this service personnel conflict graph, the initial conflict intensity of each service personnel node is used as a priori score, which is propagated in a directed manner along the edge weights in the conflict network to gradually adjust the conflict importance value of each node. Let the conflict index of service personnel i at the t-th iteration be... Its value in the next iteration Calculated using the following formula: ,in For service personnel i in the first The conflict index at the next iteration The diffusion coefficient is 0.85. For service personnel The basic conflict degree can be set as a normalized value of the sum of the weights of its conflicting edges, or a uniform constant, such as the reciprocal of the number of nodes. For the index of adjacent service personnel that have conflicting edges with service personnel i, For nodes There exists a set of adjacent service personnel with incoming edges. For service personnel To service personnel Conflict of border rights, For the neighboring node index of service personnel j, It is the weighted sum of the out-degrees of node j (i.e. its ability to output conflict information to other nodes). Let be the conflict index of service personnel j in the t-th iteration. The convergence condition is set as the 1-norm of the difference between two rounds of scores being lower than a threshold. (like ),Right now Simultaneously, a maximum number of iterations (e.g., 100) is set to prevent infinite loops in extreme structures. If A strongly conflicts with B, and B weakly conflicts with C, then the indirect conflict from A to C is partially inherited; alternatively, an adjacency-weighted iteration rule can be used. ,in The updated indirect propagation conflict strength between i and k. To perform the maximum value operation, The direct conflict strength between i and j The direct conflict strength between j and k This represents the known conflict strength between i and k before the update. For example, A B Conflict intensity 1.0, B C Conflict intensity 0.8 → A C yields a propagation conflict intensity of 0.8.

[0158] Conflicts are clustered based on conflict propagation data to obtain clockwork conflict data.

[0159] Specifically, the system constructs a weighted adjacency matrix based on the set of nodes V (representing service personnel) and the set of edges E (representing the conflict propagation relationships and their strengths among service personnel) in the propagation graph G=(V,E). ={ },in Let the conflict edge weights between service personnel i and j be represented. Next, the degree matrix of the graph is calculated. diagonal elements Then, construct the normalized Laplace matrix based on this. The system's Laplace matrix Eigenvalue decomposition is performed, the eigenvectors corresponding to the smallest k eigenvalues are extracted, and each service personnel node is mapped into a k-dimensional embedding space, k is 32, 64 or 256. The geometric distribution structure of the embedding space can effectively represent the potential conflict propagation relationship between service personnel. In the above low-dimensional embedding space, the system uses a clustering algorithm (such as a clustering method based on centroid partition) to group the service personnel nodes, thereby obtaining a plurality of conflict clusters. Each conflict cluster corresponds to a group of dense conflict nodes; output the member list and conflict density of each cluster; support setting a minimum cluster conflict value threshold to avoid too small clusters without optimization value; for example, cluster 1={A, B, C}, the average conflict intensity in the cluster is 0.92; cluster 2={D, E}, the intensity is 0.84.

[0160] Optionally, S3 comprises:

[0161] S31, constructing a clock conflict association graph according to the clock conflict data to obtain clock conflict association graph data;

[0162] Specifically, a conflict association graph G=(V,E) is constructed, where V is the node set of the graph, corresponding to all service personnel participating in the clock plan; E is the edge set of the graph, representing the conflict relationship between any two service personnel. For any pair of service personnel i and j, the edge weight of the conflict edge is: wherein is the edge weight of the conflict edge, is a time conflict degree weighting coefficient, is the historical service end time overlap degree of service personnel i and j, which can be calculated based on the intersection proportion of the time window, is a spatial adjacency weighting coefficient, is the spatial proximity of the service rooms, for example, based on the physical distance or sharing frequency of the house number. If any two people in a conflict cluster have a non-zero edge weight as described above, the system will create a corresponding edge in the graph and record the calculated edge weight.

[0163] S32, high conflict cluster identification is performed on the clock conflict association graph data to obtain high conflict cluster data;

[0164] Specifically, the system applies a predetermined graph partitioning algorithm to the graph G to extract communities (also known as clusters) with high structural tightness. The graph partitioning algorithm includes but is not limited to any of the following: one, Louvain algorithm based on modularity optimization, which maximizes the modularity index of the graph to obtain a stable community partitioning result; two, label propagation algorithm based on label propagation mechanism, which initializes node labels and iteratively propagates label states to form a stable partition; the first method is used by default. For each community / cluster obtained by graph partitioning, the system calculates its internal average conflict intensity, and the calculation method is: wherein is the intra-cluster average conflict strength, is the intra-cluster edge number, is the node sequence term, is the intra-cluster adjacent node sequence term, is the intra-cluster edge set, is the node and conflict edge weight, based on is normalized to a 0-1 range. The system sets a conflict density threshold (e.g. 0.7), and marks all clusters with an average conflict strength higher than the threshold as high conflict clusters, which is set based on historical clock-out data statistical analysis, covering more than 85% of high conflict scenarios.

[0165] S33, performing perturbation priority division on the high conflict cluster data to obtain perturbation priority data;

[0166] Specifically, the centrality of the service personnel node in each high conflict cluster is analyzed to measure its conflict dominance in the cluster. The centrality index can include degree centrality, specifically the number of connected edges or the total edge weight of the node; betweenness centrality, specifically the frequency of the node appearing in the shortest path, indicating its intermediary position in the conflict propagation path; and the perturbation index is calculated as follows: wherein is the perturbation index, is the degree centrality weight data, with a value of 0.6, is the degree centrality, is the betweenness centrality weight data, with a value of 0.4, is the betweenness centrality, and the aforementioned weights can be pre-set or fitted based on empirical data. The system sorts all service personnel in the high conflict cluster in descending order according to the perturbation index of each node to obtain a perturbation priority queue.

[0167] S34, performing skip perturbation on the perturbation priority data to obtain skip perturbation data;

[0168] Specifically, the system performs the jump intervention in turn according to the service personnel in the disturbance priority queue. The clock position of the high-priority service personnel is adjusted to reduce the conflict intensity between the high-priority service personnel and other personnel. Assuming that the current clock order is: service personnel A→service personnel B→service personnel C; if it is detected that there is a high time or space conflict intensity between A and C, and there is no conflict between B and C, B can be skipped, and the positions of A and C are exchanged or adjusted in priority, so as to realize the jump disturbance of “non-adjacent position”. The jump intervention includes fine-tuning the clock time of the target service personnel by a certain time interval (such as ± 15 minutes) forward or backward; arranging the service personnel to other available rooms; inserting a cooling buffer time before and after the clock of the service personnel; directly exchanging the clock time period or resource allocation of two service personnel. The system tries multiple sets of disturbance actions for each high-priority service personnel, forms a set of candidate clock schemes, and records the change of conflict degree before and after the disturbance.

[0169] S35, disturbance constraint verification is performed according to the jump disturbance data to obtain clock disturbance data.

[0170] Specifically, the verification rules are performed for each candidate disturbance scheme, including performing the following constraint verification for each candidate clock scheme, including performing service continuity constraint, verifying whether the continuous working time of each service personnel exceeds the preset upper limit (for example, the clock of each service personnel cannot be arranged for more than 4 hours continuously); if the continuous service time period of a service personnel in the current candidate sequence exceeds the threshold, it is determined that the scheme is not compliant. The resource exclusivity constraint is performed, and it is checked whether the room is repeatedly allocated to multiple service personnel in the same time period; it is checked whether there is an overlapping conflict between the clock time periods of the service personnel; if there is a space or time overlapping conflict, the scheme is considered invalid. The user reservation reservation constraint is performed, and if part of the clock time period has been explicitly reserved by the user, the candidate disturbance scheme cannot be adjusted in the time period; if the “unchangeable node” is adjusted in the candidate sequence, it is considered that the clock freezing node rule is violated. The verification failure processing is performed, and if the candidate scheme does not satisfy any verification rule, the disturbance attempt is discarded; the system can set an upper limit for the disturbance attempt, such as a maximum of 10 attempts; if it fails continuously, a rollback strategy can be performed, such as restoring to the original clock sequence or adjusting the disturbance intensity based on the preset parameters, and rolling back to the constraint verification stage.

[0171] Optionally, S4 includes:

[0172] S41, disturbance node dense area extraction is performed according to the clock disturbance data to obtain disturbance node dense area data;

[0173] Specifically, the start time of each disturbance node is mapped to the corresponding time discrete interval; for each disturbance task, its service room number and time slice (the start time of the disturbance node is rounded down to the nearest time granularity window) are extracted to form a two-dimensional coordinate pair; a two-dimensional sparse matrix is constructed to represent the number of disturbances that occur in a specific room and time period; the value of each matrix element is the total number of disturbance events that occur in that room during that time period. Perform a two-dimensional sliding window operation (such as a 3x3 room-time block) on the above density matrix; if the cumulative number of disturbances in the window exceeds a certain threshold, it is marked as a hotspot area. The identified hotspot windows are aggregated according to time continuity and spatial adjacency; overlapping or adjacent windows are merged to form stable disturbance dense blocks; each aggregation result defines a hotspot area.

[0174] S42, cooling mark on disturbance node dense area data and preset cooling window template data, to get cooling window data;

[0175] Specifically, the cooling template data is a preset cooling parameter configuration library, where each record is a key-value pair with the following fields: the key is the project type, the service personnel immediate recovery index IRRI hierarchical interval, and the cumulative fatigue index CDRI hierarchical interval; the value includes the basic cooling time (in minutes), the IRRI adjustment coefficient, the CDRI adjustment coefficient, and the hard block flag (Boolean value, indicating whether a mandatory empty window period is required). For each disturbance task record, extract its project type, IRRI value and CDRI value of the service personnel, and map them to the segmented intervals in the preset template. For each disturbance task, calculate the cooling window length according to the matched template item, the calculation formula is as follows: wherein is the cooling window length of service personnel No. , is the basic cooling time (in minutes), is the IRRI correction coefficient, is the idle period resilience index data of service personnel i, is the CDRI correction coefficient, is the cumulative fatigue index of service personnel , wherein is the cumulative fatigue index of service personnel , is the task sequence item, is the number of tasks completed by service personnel i in the selected time window, is the load level weighting coefficient (such as high load weighting 1.5, medium load 1.2, and low load 1.0), is the load level of task j, is the interval recovery time (in minutes) between task j and the previous task, is a tiny constant, set to 0.01, is the duration of task j (in minutes), is a natural exponential term, is a time decay coefficient (e.g. 0.01-0.05), adjustable, is the current time, is the execution time of task j. If the hard-block flag in the matched template is true, then no high-load or medium-high-load project task is allowed to be scheduled in this window. For each disturbance task, a cooling time mask is generated according to its start time and the calculated cooling window length.

[0176] S43, disturb delay extraction is performed on the clock disturbance data according to the cooling window data, to obtain disturbance delay data;

[0177] Specifically, for each record in the disturbance task, it is judged whether its start time falls into the cooling mask area of the corresponding service personnel: if it falls into the cooling window (i.e. the cooling mask value is 1), it is determined that the task needs to be delayed; otherwise, the task time is legal and no change is needed. Taking the start time of the current task as a reference, the time slots are scanned in sequence; it is judged whether the following conditions are met at the same time on each candidate slot: the service personnel in the cooling mask is not restricted in this time period (the mask value is 0); the selected room is available in the resource matrix and does not conflict with the existing appointment of the customer; the alternative time does not introduce significant new conflicts, or the conflict score is lower than the acceptable threshold; once the earliest time slot that meets all the conditions is found, it is selected as the delayed start time.

[0178] S44, service combination reconstruction is performed on the disturbance delay data and the preset service combination reconstruction table, to obtain clock optimization data.

[0179] In particular, the service combination reconfiguration table is a predefined set of service combination reconfiguration rules, including triggering conditions for reconfiguration operations (such as task density exceeding a threshold within a time period, belonging to the same project cluster or the same room cluster), optional transformation actions (such as task order interchanging, room changing, splitting long tasks into multiple short tasks, merging short tasks into a long task), optimization objective functions (such as minimizing peak period density, minimizing waiting time), and constraint conditions (such as skill suitability, project exclusivity, room capacity limit, etc.). The system filters the task combination set that meets the conditions from the disturbance delay data according to the triggering conditions in the combination reconfiguration table, including finding multiple tasks belonging to the same project type (such as meridian massage) in the same time window; filtering the task set that overlaps in spatial position (the same room or area) and has a task density exceeding a threshold. Through the above screening, a local combination candidate set is constructed. For each local combination, the system generates a feasible scheduling transformation based on its associated reconfiguration operation list. Specific ways include but are not limited to enumerating task order interchanging arrangements; reassigning tasks to currently unsaturated idle rooms; splitting long tasks into consecutive schedulable periods; merging multiple short tasks with close intervals into a single long task; considering combination replacement strategies to make task-to-task replacement under the premise of ensuring service quality. Each combination transformation generates a candidate solution. The candidate solutions are sorted according to the task aggregation degree in high-density time periods and the total delay sum of the actual scheduling time of the tasks relative to the original time. For each local combination, the corresponding optimal solution is selected. All optimal combination reconfiguration solutions are merged into the current scheduling overall plan to construct a new optimized scheduling result table. After the merging operation, the system performs consistency checking to ensure that all service personnel and room resources do not have time conflicts; the cooling window is not violated; the original appointment time point or range of the customer is not arbitrarily changed; the service combination transformation does not destroy the service process logic and project independence.

[0180] Optionally, the application also provides a data-driven service personnel scheduling optimization system for executing the data-driven service personnel scheduling optimization method as described above, the data-driven service personnel scheduling optimization system comprising:

[0181] The on-duty feature extraction module is configured to obtain historical service personnel on-duty data; and perform on-duty feature extraction based on the historical service personnel on-duty data to obtain on-duty feature data.

[0182] The scheduling conflict identification module is configured to perform on-duty same-frequency processing based on the on-duty feature data to obtain on-duty same-frequency data; and perform scheduling conflict processing based on the on-duty same-frequency data to obtain scheduling conflict data.

[0183] The scheduling sequence disturbance module is configured to perform scheduling sequence disturbance based on the scheduling conflict data to obtain scheduling disturbance data.

[0184] The rhythm cooling dispersion module is used for cooling window control according to the clock disturbance data to obtain cooling window data; and the cooling window data is used for dispersion degree processing on the clock disturbance data to obtain clock optimization data.

[0185] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, the scope of the application being defined by the appended application file and not by the above description, therefore all variations falling within the meaning and scope of the equivalent requirements of the application file are intended to be included within the present application.

[0186] The above description is merely a specific implementation of the present application, enabling those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A data-driven based service personnel scheduling optimization method, characterized in that, The method comprises: S1, obtaining historical service personnel clock-in data; performing clock-in feature extraction according to the historical service personnel clock-in data to obtain clock-in feature data; S2, performing clock-repliable point identification according to the historical service personnel clock-in data and the clock-in feature data to obtain clock-repliable point data; performing rhythm regression trajectory construction according to the clock-repliable point data to obtain rhythm regression trajectory data; performing sliding window clustering according to the rhythm regression trajectory data to obtain preliminary clock-same-frequency data; performing room service record extraction according to the historical service personnel clock-in data to obtain room service record data; performing room clock-out mapping according to the room service record data to obtain room clock-out data; performing convolution kernel extraction according to the room clock-out data to obtain concurrent clock-out block data; performing service personnel synchronization cluster extraction on the preliminary clock-same-frequency data according to the concurrent clock-out block data to obtain clock-out same-frequency data; performing clock conflict processing according to the clock-out same-frequency data to obtain clock conflict data; S3, performing clock order disturbance according to the clock conflict data to obtain clock disturbance data; S4, performing cooling window control according to the clock disturbance data to obtain cooling window data; performing dispersion degree processing on the clock disturbance data according to the cooling window data to obtain clock optimization data; The service personnel synchronization cluster extraction comprises: performing concurrent block label mapping on the preliminary clock-same-frequency data according to the concurrent clock-out block data to obtain concurrent block label data; performing rhythm regression trajectory phase difference calculation according to the concurrent block label data to obtain trajectory phase difference data; constructing a synchronization graph according to the historical service personnel clock-in data according to the trajectory phase difference data to obtain synchronization graph data; performing local movement processing according to the synchronization graph data to obtain local movement graph data; performing fine-grained subgroup reconstruction on the local movement graph data to obtain subgroup reconstruction data; performing aggregated graph reconstruction according to the subgroup reconstruction data to obtain clock-out same-frequency data.

2. The method of claim 1, wherein, S1 comprises: obtaining historical service personnel clock-in data; performing service period calculation, average service duration calculation, clock-continuous length calculation, and idle residual time calculation according to the historical service personnel clock-in data to obtain service period data, average duration data, clock-continuous length data, and idle residual time data, respectively; performing idle period rebound index calculation according to the service period data and the average service duration data to obtain idle period rebound index data; performing clock-continuous attenuation residual index calculation according to the clock-continuous length data and the idle residual time data to obtain clock-continuous attenuation residual index data; performing clock-in item extraction according to the historical service personnel clock-in data to obtain clock-in item data; performing item adaptation degree processing on the clock-in item data according to the idle period rebound index data and the clock-continuous attenuation residual index data to obtain clock-in feature data.

3. The method of claim 2, wherein, The idle period rebound index calculation comprises: performing time residual calculation according to the service period data and the average service duration data to obtain time residual data; performing period rebound trend calculation according to the time residual data to obtain period rebound trend data; performing idle period rebound index calculation according to the period rebound trend data and the time residual data to obtain idle period rebound index data.

4. The method of claim 2, wherein, The project adaptation processing includes: According to the idle cycle rebound index data and the bell attenuation residual index data, project load matrix construction is performed on the bell-on project data to obtain project load matrix data; According to the historical service personnel bell-on data, service personnel state extraction is performed to obtain service personnel state data; According to the project load matrix data and the service personnel state data, project adaptation risk calculation is performed to obtain bell-on feature data.

5. The method of claim 1, wherein, The bell-off conflict processing includes: According to the bell-off same frequency data, a visual angle extraction is performed to obtain visual angle data, wherein the visual angle extraction includes project visual angle extraction, regional visual angle extraction, and spatial visual angle extraction; According to the visual angle data, a cross-room conflict matrix construction is performed to obtain cross-room conflict data; According to the cross-room conflict data, a conflict intensity propagation processing is performed to obtain conflict propagation data; According to the conflict propagation data, a conflict grouping is performed to obtain bell-off conflict data.

6. The method of claim 1, wherein S3 It includes: According to the bell-off conflict data, a bell-off conflict correlation graph construction is performed to obtain bell-off conflict correlation graph data; High conflict cluster data is obtained by performing high conflict cluster identification on the bell-off conflict correlation graph data; The perturbation priority data is obtained by performing perturbation priority division on the high conflict cluster data; The skip perturbation data is obtained by performing skip perturbation on the perturbation priority data; According to the skip perturbation data, perturbation constraint verification is performed to obtain bell-off perturbation data.

7. The method according to claim 1, characterized in that S4 It includes: According to the bell-off perturbation data, a perturbation node dense area extraction is performed to obtain perturbation node dense area data; Cooling window data is obtained by performing cooling marking on the perturbation node dense area data and the preset cooling window template data; The perturbation delay data is obtained by performing perturbation delay extraction on the bell-off perturbation data according to the cooling window data; The bell-off optimization data is obtained by performing service combination reconstruction on the perturbation delay data and the preset service combination reconstruction table.

8. A data-driven based service personnel scheduling optimization system, characterized in that, The data-driven service personnel bell-off optimization system for executing the data-driven service personnel bell-off optimization method of claim 1 includes: A bell-on feature extraction module is configured to obtain historical service personnel bell-on data, and perform bell-on feature extraction according to the historical service personnel bell-on data to obtain bell-on feature data; A bell-off conflict identification module is configured to perform bell-off same frequency processing according to the bell-on feature data to obtain bell-off same frequency data, and perform bell-off conflict processing according to the bell-off same frequency data to obtain bell-off conflict data; A bell-off sequence perturbation module is configured to perform bell-off sequence perturbation according to the bell-off conflict data to obtain bell-off perturbation data; A rhythm cooling dispersion module is configured to perform cooling window control according to the bell-off perturbation data to obtain cooling window data, and perform dispersion degree processing on the bell-off perturbation data according to the cooling window data to obtain bell-off optimization data.

Images