A heuristic logic adjusted elevator maintenance holiday scheduling method and system

Abstract

The application discloses an elevator maintenance holiday scheduling method and system based on heuristic logic adjustment, and the method comprises the following steps: obtaining an elevator team queue based on the calculation of elevator data by using the Euclidean distance aggregation; extracting a night shift elevator team set from the elevator team queue, constructing a Monte Carlo search tree, and generating an initial scheduling strategy; performing global scheduling based on the initial scheduling strategy by using a dynamic selection decision tree, and screening to obtain a candidate scheduling strategy; and performing heuristic logic adjustment on the candidate scheduling strategy, and constraining an output scheduling sequence. Therefore, by means of the variance reward of the Monte Carlo tree search and the logic adjustment, the accumulated work orders can be smoothly dispersed to the compliance time window, and according to various constraint conditions such as night shift priority and route comparison, it can be ensured that the maintenance personnel can stably rest the next day, the workload overload and work order accumulation can be effectively avoided, the global scheduling load is balanced, and the human resource scheduling is reasonable.

Description

Technical Field

[0001] This invention relates to the field of scheduling planning technology, and in particular to a method and system for scheduling elevator maintenance on holidays using heuristic logic adjustment. Background Technology

[0002] The elevator maintenance industry is characterized by strict cyclical nature, with each elevator typically requiring cleaning and maintenance every 15 days. Furthermore, elevator maintenance work is geographically dispersed, with elevators maintained by the same company often scattered across various locations. This necessitates that maintenance companies allocate human resources to manage the irregularly dispersed elevator projects within their jurisdiction within fixed scheduling cycles. This process involves an extremely complex set of constraints. On one hand, they must adhere to hard regulatory constraints (such as a maximum maintenance interval of 15 days, mandatory rest for night shifts, and user-specified maintenance dates), while on the other hand, they must also consider soft management constraints (such as prioritizing weekdays, ensuring a balanced number of elevators maintained, and addressing uneven geographical distribution).

[0003] In addition, due to various types of holidays disrupting normal work schedules, and the need for make-up shifts and rescheduling for some holidays, some maintenance projects will inevitably encounter holidays and need to be moved forward or backward. This will cause a surge in maintenance work orders before and after holidays, resulting in the problem of night shifts not being able to rest the next day, further exacerbating the difficulty of human resource scheduling, and potentially leading to a complete depletion of available human resources, resulting in maintenance exceeding the time limit. Summary of the Invention

[0004] The first aspect of this embodiment discloses a heuristic logic-based method for scheduling elevator maintenance during holidays, specifically including: Elevator team column is obtained by aggregating elevator data based on Euclidean distance; Extract the set of night shift elevator teams involved in night shift maintenance tasks from the elevator team column, construct a Monte Carlo search tree, and generate an initial scheduling strategy; Based on the initial scheduling strategy, a dynamic selection decision tree is used for global scheduling to filter and obtain candidate scheduling strategies. Heuristic logic is applied to the candidate scheduling strategy to constrain the output scheduling sequence.

[0005] As an optional implementation, the method further includes: Several initial clusters were built based on the maintenance project; In each of the initial clusters, the number of elevators is less than 5, and the maintenance dates for all elevators in the same initial cluster are the same. Elevators that are not included in any initial cluster in the elevator data are placed into a discrete queue.

[0006] As an optional implementation, obtaining the elevator team column based on Euclidean distance aggregation includes: Calculate the Euclidean distance for each discrete elevator in the discrete queue; Elevators whose Euclidean distance is less than the distance threshold are aggregated into several distance clusters; The initial clusters and the distance clusters constitute the elevator team column.

[0007] As an optional implementation, the step of extracting the set of night shift elevator teams involved in night shift maintenance tasks from the elevator team column, constructing a Monte Carlo search tree, and generating an initial scheduling strategy includes: A hierarchical strategy is set to prioritize night shifts and subordinate day shifts, extracting only the night shift elevator group set that is designed for night shift maintenance tasks; A Monte Carlo search tree is constructed based on the night shift elevator group set, and the Monte Carlo tree search algorithm is executed. Set a reward function based on the night shift maintenance task the night before a holiday; Set up discrete penalties based on the night shift rest day arrangement after the night shift maintenance task; The reward function and the grouped discrete penalty are introduced into the Monte Carlo tree search algorithm to generate the initial scheduling strategy.

[0008] As an optional implementation, the step of using a dynamic selection decision tree to perform global scheduling based on the initial scheduling strategy and filtering to obtain candidate scheduling strategies includes: In the dynamic selection decision tree, the search space is intelligently pruned according to the holiday schedule; For scheduling chains that avoid all statutory holidays and night shift rest days, no branch selection is made, and the original solution is reused; For elevator groups that interfere with holiday schedules, a multi-branch tree is constructed, and a legality checker is built in when generating child nodes to filter out those with 7 days or more and 15 days or less, while removing the scheduleable intervals for rest days. Set a capacity penalty counter to reduce the UCB value for paths that reach the maximum daily work order volume.

[0009] As an optional implementation, the method further includes: We employ fast approximate simulation, abandoning full TSP path planning; The nearest neighbor × 2 algorithm is used to estimate the marginal path cost of the newly added node. If the estimated marginal path cost exceeds the cost threshold, pruning is triggered and the simulation of that branch is stopped. Set a convex penalty reward function with an exponentially increasing penalty weight to force infeasible solutions to be eliminated first.

[0010] As an optional implementation, the step of performing heuristic logic adjustments on the candidate scheduling strategy to constrain the output scheduling sequence includes: Scan for idle dates in the candidate scheduling strategies where no maintenance tasks are available; Search for maintenance work orders associated with dates adjacent to the idle date; Under the premise that the maximum daily work order volume and mileage increase do not exceed the limit, maintenance work orders associated with adjacent dates of the idle date will be partially migrated and reassigned to the idle date.

[0011] As an optional implementation, the method further includes: A convergent local search is employed to continuously migrate some maintenance work orders associated with the date with the longest distance to other valid dates until the scheduling sequence is optimized.

[0012] The second aspect of this embodiment discloses a heuristic logic-based elevator maintenance holiday scheduling system, specifically including: The filtering unit is used to analyze cluster data based on maintenance projects and obtain discrete queues; Clustering units are used to iteratively cluster the discrete queue using an iterative self-organizing data analysis algorithm to obtain elevator team queues; The search generation unit is used to extract the set of night shift elevator groups involving night shift maintenance tasks from the elevator team column, construct a Monte Carlo search tree, and generate an initial scheduling strategy. The global scheduling unit is used to perform global scheduling based on the initial scheduling strategy using a dynamic selection decision tree, and to filter out candidate scheduling strategies. The heuristic adjustment unit is used to perform heuristic logic adjustments on the candidate scheduling strategy and constrain the output scheduling sequence.

[0013] Compared with the prior art, this embodiment has the following beneficial effects: By using the variance reward and logical adjustment of Monte Carlo tree search, accumulated work orders can be smoothly distributed to compliant time windows. Furthermore, based on various constraints such as night shift priority and route comparison, it can be ensured that maintenance personnel can get stable rest the day after their night shift, effectively avoiding workload overload and work order accumulation, achieving balanced global shift load, and reasonable human resource allocation. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in this embodiment, the accompanying drawings used in the embodiment will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1This is a schematic diagram of the workflow of a heuristic logic adjustment method for elevator maintenance holiday scheduling disclosed in this embodiment; Figure 2 This is a schematic diagram of the system structure of a heuristic logic-based elevator maintenance holiday scheduling system disclosed in this embodiment. Detailed Implementation

[0016] The technical solutions in this embodiment will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1 Please see Figure 1 This embodiment discloses a heuristic logic-based method for scheduling elevator maintenance during holidays, including: 101. Elevator team column obtained by aggregate calculation of elevator data based on Euclidean distance.

[0018] In this embodiment, elevator clusters can not only more reasonably represent the actual distance of maintenance work orders in a geographical area, but also significantly reduce the number of terminals in the optimization calculation process (e.g., 1000 elevators are reduced to 200 elevator clusters), thereby improving the efficiency of optimization calculation.

[0019] As an optional implementation method, several initial clusters are built based on the maintenance project; In this case, the number of elevators in each initial cluster is less than 5, and the maintenance dates of all elevators in the same initial cluster are the same. Elevators that are not included in any initial cluster in the elevator data are placed into a discrete queue.

[0020] Here, elevators that belong to the unified maintenance project and were originally scheduled to be maintained on the same day will be forcibly bound together.

[0021] Furthermore, discrete elevators that fail to achieve homogeneous aggregation are placed in a discrete queue, awaiting further aggregation processing through different methods.

[0022] Understandably, the limit of 5 elevators for the initial cluster is set based on maintenance time and the difficulty of temporary scheduling. If more than 5 elevators are installed, the excessively long working time will significantly compress the schedulable time in the cluster, drastically reducing scheduling flexibility.

[0023] As an optional implementation, elevator team columns are obtained based on Euclidean distance aggregation, including: Calculate the Euclidean distance for each discrete elevator in the discrete queue; Elevators whose Euclidean distance is less than the distance threshold are aggregated into several distance clusters; Several initial clusters and several distance clusters constitute an elevator team.

[0024] Here, by calculating the Euclidean distance between each discrete elevator, discrete elevators that are close to each other in the commuting area are placed into the same distance cluster to achieve the initial division of discrete elevators.

[0025] That is, for discrete elevators that lack homogeneous labels and clear classification standards, a preliminary classification is made based on the convenience of commuting, so as to significantly reduce the number of terminals required for optimization calculation and reduce the computational load.

[0026] In this embodiment, the ladder density of the stationed project is calculated; Stationary projects with a gradient density exceeding the density threshold or a coverage radius exceeding the radius threshold are set as cluster data.

[0027] It is understandable that, because some site projects have a large number of elevator maintenance work orders in a very small area, the location of the site project appears as a significant outlier in the overall data. If it is included in the whole-day optimization calculation, other discrete work orders will be significantly affected, resulting in an unsatisfactory overall optimization effect.

[0028] Therefore, instead of using a single elevator as the dispatch unit, we use elevator clusters as the smallest courtyard unit to strongly bind elevators that are maintained on the same day in the same project.

[0029] 102. Extract the set of night shift elevator teams involved in night shift maintenance tasks from the elevator team column, construct a Monte Carlo search tree, and generate an initial scheduling strategy.

[0030] In this embodiment, "non-continuous operation" and "rest required the day after night shift" are used as strong constraints, and a hierarchical strategy of "night shift priority and day shift subordinate" is adopted to ensure that maintenance personnel can get stable rest the day after night shift.

[0031] As an optional implementation, the set of night shift elevator teams involved in night shift maintenance tasks is extracted from the elevator team column, a Monte Carlo search tree is constructed, and an initial scheduling strategy is generated, including: A hierarchical strategy is set to prioritize night shifts and subordinate day shifts, extracting only the night shift elevator group set that is designed for night shift maintenance tasks; A Monte Carlo search tree is constructed based on the set of night shift elevator groups, and the Monte Carlo tree search algorithm is executed. Set a reward function based on the night shift maintenance task the night before a holiday; Set up discrete penalties based on the night shift rest day arrangement after the night shift maintenance task; The reward function and grouped discrete penalty are introduced into the Monte Carlo tree search algorithm to generate the initial scheduling strategy.

[0032] Here, locking the night shift date is a hard, unchangeable constraint to avoid scheduling maintenance work orders for the following day after a night shift.

[0033] 103. Based on the initial scheduling strategy, a dynamic selection decision tree is used for global scheduling to filter and obtain candidate scheduling strategies.

[0034] In this embodiment, based on the clear night shift schedule, the remaining maintenance work orders are globally scheduled. In order to achieve rapid convergence with limited computing power, a hybrid tree structure and capacity-aware search are adopted in combination.

[0035] As an optional implementation, a dynamic selection decision tree is used for global scheduling based on the initial scheduling strategy to filter and obtain candidate scheduling strategies, including: In the dynamic selection decision tree, the search space is intelligently pruned according to the holiday schedule. For scheduling chains that avoid all statutory holidays and night shift rest days, no branch selection is made, and the original solution is reused; For elevator groups that interfere with holiday schedules, a multi-branch tree is constructed, and a legality checker is built in when generating child nodes to filter out those with 7 days or more and 15 days or less, while removing the scheduleable intervals for rest days. Set a capacity penalty counter to reduce the UCB value for paths that reach the maximum daily work order volume.

[0036] Thus, by reusing the original solution, 30% to 50% of the invalid search space is directly eliminated, significantly improving optimization efficiency and reducing computational load.

[0037] As an optional implementation, a fast approximate simulation is used, abandoning full TSP path planning; The nearest neighbor × 2 algorithm is used to estimate the marginal path cost of the newly added node. If the estimated marginal path cost exceeds the cost threshold, pruning is triggered and the simulation of that branch is stopped. Set a convex penalty reward function with an exponentially increasing penalty weight to force infeasible solutions to be eliminated first.

[0038] Therefore, by introducing a penalty mechanism, extreme situations of a single day's order surge can be effectively avoided, and infeasible solutions can be eliminated in a timely manner.

[0039] 104. Perform heuristic logic adjustments on the candidate scheduling strategies to constrain the output scheduling sequence.

[0040] In this embodiment, guided optimization is also performed for local defects.

[0041] As an optional implementation, heuristic logic adjustments are performed on the candidate scheduling strategy to constrain the output scheduling sequence, including: Scan candidate scheduling strategies for idle dates where maintenance tasks are not available; Search for maintenance work orders associated with dates adjacent to the idle dates; Under the premise that the maximum daily work order volume and mileage increase do not exceed the limit, maintenance work orders associated with adjacent dates of idle dates will be partially migrated and reassigned to idle dates.

[0042] As an optional implementation, a convergent local search is adopted to continuously migrate some of the maintenance work orders associated with the date with the longest distance to other valid dates until the scheduling sequence is optimized.

[0043] Here, maintenance work orders in the scheduling sequence will be continuously optimized, making manpower allocation more reasonable, reducing the probability of high workload in specific situations, and minimizing the standard deviation of the number of maintenance work orders throughout the entire cycle.

[0044] Compared with the prior art, this embodiment has the following beneficial effects: By using the variance reward and logical adjustment of Monte Carlo tree search, accumulated work orders can be smoothly distributed to compliant time windows. Furthermore, based on various constraints such as night shift priority and route comparison, it can be ensured that maintenance personnel can get stable rest the day after their night shift, effectively avoiding workload overload and work order accumulation, achieving balanced global shift load, and reasonable human resource allocation.

[0045] Example 2 Please see Figure 2 This embodiment discloses a heuristic logic-based elevator maintenance holiday scheduling system, including: The filtering unit is used to analyze cluster data based on maintenance projects to obtain discrete queues; Clustering units are used to iteratively cluster discrete queues using an iterative self-organizing data analysis algorithm to obtain elevator team queues; The search generation unit is used to extract the set of night shift elevator groups involving night shift maintenance tasks from the elevator team column, construct a Monte Carlo search tree, and generate an initial scheduling strategy. The global scheduling unit is used to perform global scheduling based on the initial scheduling strategy using a dynamic selection decision tree, and to filter out candidate scheduling strategies. The heuristic adjustment unit is used to perform heuristic logic adjustments on candidate scheduling strategies and constrain the output scheduling sequence.

Claims

1. A heuristic logic-based method for scheduling elevator maintenance shifts during holidays, characterized in that, include: Elevator team column is obtained by aggregating elevator data based on Euclidean distance; Extract the set of night shift elevator teams involved in night shift maintenance tasks from the elevator team column, construct a Monte Carlo search tree, and generate an initial scheduling strategy; Based on the initial scheduling strategy, a dynamic selection decision tree is used for global scheduling to filter and obtain candidate scheduling strategies. Heuristic logic is applied to the candidate scheduling strategy to constrain the output scheduling sequence.

2. The elevator maintenance holiday scheduling method based on heuristic logic adjustment according to claim 1, characterized in that, The method further includes: Several initial clusters were built based on the maintenance project; In each of the initial clusters, the number of elevators is less than 5, and the maintenance dates for all elevators in the same initial cluster are the same. Elevators that are not included in any initial cluster in the elevator data are placed into a discrete queue.

3. The elevator maintenance holiday scheduling method based on heuristic logic adjustment according to claim 2, characterized in that, The elevator team column obtained based on Euclidean distance aggregation includes: Calculate the Euclidean distance for each discrete elevator in the discrete queue; Elevators whose Euclidean distance is less than the distance threshold are aggregated into several distance clusters; The initial clusters and the distance clusters constitute the elevator team column.

4. The elevator maintenance holiday scheduling method based on heuristic logic adjustment according to claim 1, characterized in that, The process involves extracting the set of night shift elevator teams involved in night shift maintenance tasks from the elevator team column, constructing a Monte Carlo search tree, and generating an initial scheduling strategy, including: A hierarchical strategy is set to prioritize night shifts and subordinate day shifts, extracting only the night shift elevator group set that is designed for night shift maintenance tasks; A Monte Carlo search tree is constructed based on the night shift elevator group set, and the Monte Carlo tree search algorithm is executed. Set a reward function based on the night shift maintenance task the night before a holiday; Set up discrete penalties based on the night shift rest day arrangement after the night shift maintenance task; The reward function and the grouped discrete penalty are introduced into the Monte Carlo tree search algorithm to generate the initial scheduling strategy.

5. The elevator maintenance holiday scheduling method based on heuristic logic adjustment according to claim 1, characterized in that, The process of using a dynamic selection decision tree to perform global scheduling based on the initial scheduling strategy, and filtering to obtain candidate scheduling strategies, includes: In the dynamic selection decision tree, the search space is intelligently pruned according to the holiday schedule; For scheduling chains that avoid all statutory holidays and night shift rest days, no branch selection is made, and the original solution is reused; For elevator groups that interfere with holiday schedules, a multi-branch tree is constructed, and a legality checker is built in when generating child nodes to filter out those with 7 days or more and 15 days or less, while removing the scheduleable intervals for rest days. Set a capacity penalty counter to reduce the UCB value for paths that reach the maximum daily work order volume.

6. The elevator maintenance holiday scheduling method based on heuristic logic adjustment according to claim 5, characterized in that, The method further includes: We employ fast approximate simulation, abandoning full TSP path planning; The nearest neighbor × 2 algorithm is used to estimate the marginal path cost of the newly added node. If the estimated marginal path cost exceeds the cost threshold, pruning is triggered and the simulation of that branch is stopped. Set a convex penalty reward function with an exponentially increasing penalty weight to force infeasible solutions to be eliminated first.

7. The elevator maintenance holiday scheduling method based on heuristic logic adjustment according to claim 1, characterized in that, The step of performing heuristic logic adjustments on the candidate scheduling strategy to constrain the output scheduling sequence includes: Scan for idle dates in the candidate scheduling strategies where no maintenance tasks are available; Search for maintenance work orders associated with dates adjacent to the idle date; Under the premise that the maximum daily work order volume and mileage increase do not exceed the limit, maintenance work orders associated with adjacent dates of the idle date will be partially migrated and reassigned to the idle date.

8. The elevator maintenance holiday scheduling method based on heuristic logic adjustment according to claim 7, characterized in that, The method further includes: A convergent local search is employed to continuously migrate some maintenance work orders associated with the date with the longest distance to other valid dates until the scheduling sequence is optimized.

9. A heuristic logic-based elevator maintenance holiday scheduling system, characterized in that, include: Clustering units are used to aggregate elevator data based on Euclidean distance to obtain elevator team columns; The search generation unit is used to extract the set of night shift elevator groups involving night shift maintenance tasks from the elevator team column, construct a Monte Carlo search tree, and generate an initial scheduling strategy. The global scheduling unit is used to perform global scheduling based on the initial scheduling strategy using a dynamic selection decision tree, and to filter out candidate scheduling strategies. The heuristic adjustment unit is used to perform heuristic logic adjustments on the candidate scheduling strategy and constrain the output scheduling sequence.

Images