An elevator dynamic nearest dispatching system fusing traffic flow and historical response rate

Abstract

The application provides an elevator dynamic nearest dispatching system fusing traffic flow and historical response rate, and the system comprises the following steps: a data acquisition module acquires elevator fault information and maintenance personnel state information in real time; a historical portrait module constructs an ability evaluation value of maintenance personnel under different space-time conditions, and establishes a probability mapping of fault codes and spare part types; a traffic feature module screens candidate maintenance personnel, and extracts real-time traffic feature vectors and spare part matching features; a decision module inputs the above features into a prediction model to obtain a predicted effective arrival time consumption quantile estimation value, and determines target maintenance personnel to execute dispatching according to the estimation value; and the application further introduces a dynamic rescheduling mechanism, continuously tracks the dispatched work order in the way, triggers secondary optimization when the predicted time consumption is abnormal, and supports dynamic handover of the work order. The application improves the dispatching accuracy and emergency response efficiency.

Description

Technical Field

[0001] This invention relates to the field of elevator maintenance and dispatching technology, specifically a dynamic, proximity-based elevator dispatching system that integrates traffic flow and historical response rates. Background Technology

[0002] Existing elevator dispatching systems generally adopt the principle of proximity dispatching, primarily based on the spatial distance or simple road network distance between the maintenance personnel's real-time location and the fault point, combined with skill matching and idle status. This dispatching method ignores the significant differences in historical response efficiency among different maintenance personnel in the same area and at the same time, meaning that the dispatched personnel may not be the fastest to arrive. At the same time, the dispatching decision does not consider the matching relationship between the spare parts required for the fault and the spare parts currently carried by the maintenance personnel, and it also lacks fault spare parts demand prediction based on historical maintenance data. After arriving at the scene, maintenance personnel often need to retrieve parts a second time or wait for delivery due to the lack of critical spare parts, significantly increasing the effective time spent on actually troubleshooting the fault.

[0003] For emergency faults such as personnel being trapped, existing dispatch systems often fail to impose legally mandated arrival times as a hard constraint on dispatch decisions. Predicted arrival times are typically based on a single point estimate, failing to account for forecast uncertainties. The system cannot assess whether repair personnel can arrive on time under worst-case traffic conditions. Furthermore, the dynamic adjustment mechanism after dispatch is inadequate. When en route repair personnel encounter sudden congestion or other situations that significantly extend the estimated arrival time, the system struggles to reassess spare parts resources and other available repair personnel en route and to execute work order handover, leaving the timely handling of emergency faults unreliable. Summary of the Invention

[0004] To address the technical problems mentioned in the background section, this invention proposes a dynamic elevator dispatching system that integrates traffic flow and historical response rate.

[0005] Therefore, the technical solution adopted by the present invention is as follows: A dynamic, proximity-based elevator dispatching system that integrates traffic flow and historical response rates includes: Data acquisition module: collects elevator fault information and maintenance personnel status information in real time; the elevator fault information includes the fault location, fault type, fault code, elevator brand and model, and emergency level; the status information includes the maintenance personnel's real-time location, skill tags, busy status, and current spare parts list; Historical profile module: During system initialization, a personalized profile of each maintenance worker is constructed, as well as a probability mapping between fault codes and required spare parts types; the personalized profile represents the maintenance worker's ability assessment value under different time and space conditions; Traffic Feature Module: Based on preset filtering conditions, candidate maintenance personnel are selected from all maintenance personnel; and real-time traffic feature vectors and spare parts matching features of each candidate maintenance personnel from real-time geographical location to fault geographical location are obtained; Decision module: Input the real-time traffic feature vector, spare parts matching feature, and capability assessment value that matches the current spatiotemporal conditions of each candidate maintenance personnel into the pre-trained prediction model to obtain the quintiles prediction value, quintiles prediction value, and confidence level of the effective arrival time of each candidate maintenance personnel, and confirm the target maintenance personnel based on these values ​​and execute work order dispatch.

[0006] Compared with the prior art, the advantages of the present invention are as follows: 1. This invention constructs a probability mapping between fault codes and required spare parts types, matches the spare parts list currently carried by maintenance personnel with the spare parts demand prediction results, and incorporates spare parts matching features into candidate screening and time prediction, effectively reducing the risk of maintenance personnel having to retrieve parts a second time due to missing parts after arriving on site, and increasing the probability of resolving faults on the first visit.

[0007] 2. This invention uses the ninetieths of the predicted effective arrival time as the basis for determining the feasibility of the time limit for emergency faults. Only maintenance personnel who can still meet the arrival time requirement under the worst-case scenario are included in the dispatchable personnel. When no one can meet the time limit, an alarm is issued proactively to coordinate the handling, thus avoiding the risk of emergency rescue delays caused by optimistic bias in point estimation prediction.

[0008] 3. This invention sets tiered dynamic rescheduling trigger conditions based on emergency level settings. During in-transit tracking, the predicted values ​​of the 90ths or 50ths digits are continuously reviewed. If the value exceeds the limit, a secondary optimization is triggered. During the secondary optimization, the spare parts matching situation is re-evaluated. This enables timely handover of work orders and dynamic rematching of spare parts resources under sudden road conditions, ensuring the reliability of emergency fault handling. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of the system execution flow of the present invention; Figure 2 This is a schematic diagram of the personalized portrait generation process of the present invention; Figure 3 This is a schematic diagram illustrating the composition of real-time traffic feature vectors according to the present invention. Detailed Implementation

[0011] To achieve the above objectives, this invention provides a dynamic, proximity-based elevator dispatching system that integrates traffic flow and historical response rates. Please refer to [link / reference]. Figure 1 ,include: Data acquisition module: collects elevator fault information and maintenance personnel status information in real time; the elevator fault information includes the fault location, fault type, fault code, elevator brand and model, and emergency level; the status information includes the maintenance personnel's real-time location, skill tags, busy status, and current spare parts list; Elevator fault information is obtained in real time through the elevator IoT platform; the fault location includes longitude, latitude or building code, etc.; the fault type includes people trapped, door failure, emergency stop, non-level stop, abnormal noise, lighting failure, etc.; the fault code is the original fault code reported by the elevator control system, such as "E41 door lock circuit abnormal" "E48 brake release failure", etc.; the elevator brand and model are read from the elevator archive database for auxiliary judgment of spare parts compatibility. The emergency level is automatically generated by the system based on the fault type and supplementary information reported in real time by the elevator IoT (such as whether there are people in the car, the status of people in the car, and the building attributes where the elevator is located) according to preset mapping rules. During system initialization, a mapping table between fault types and emergency levels is established, and supplementary information is used as a condition for level adjustment in the mapping. An example rule is as follows: Level I: People trapped; or although not trapped, there are injuries, elderly / pregnant women / children in the elevator car and communication is interrupted, the elevator is located in a critical location such as a hospital / transportation hub and affects emergency evacuation; the time limit is less than or equal to 30 minutes. Level II: The elevator stops at a non-level position but passengers can escape on their own; the door lock circuit has a fault but no one is trapped; the elevator is out of service in critical locations, affecting normal operation; the time limit can be configured, for example, 60 minutes; Level III: Common faults, such as abnormal noises, lighting failure, button malfunction, and decreased leveling accuracy; no hard time limit requirement. Status information is continuously received from the terminals of all registered maintenance personnel; including: Real-time geolocation, such as GPS coordinates; skill tags include elevator brands that can be repaired, fault types, etc.; busy status includes idle, dispatched, and under repair; the current spare parts list is a snapshot of the latest inventory of the spare parts storage units (such as vehicle spare parts cabinets and toolboxes) on the repair personnel's vehicle, and each entry includes spare parts type code, specifications, quantity, and entry update timestamp; the entire list also has a snapshot timestamp, which is the timestamp of the most recent update of any entry; The spare parts list is collected as follows: each time a maintenance worker takes or replenishes a spare parts from the vehicle's spare parts cabinet, they scan the spare parts' QR code or RFID tag with a mobile terminal, and the system automatically updates the spare parts list snapshot for that maintenance worker. At the same time, the system pushes a spare parts list verification reminder to all maintenance workers at a fixed time every day, requiring them to confirm or correct the list to ensure its timeliness.

[0012] Preprocessing is required after elevator fault and status information is collected. The specific steps for elevator fault information preprocessing are as follows: For the same elevator that reports the same fault code multiple times in a very short period of time (such as within 10 seconds), only the first report is retained to avoid generating redundant work orders due to network retransmission or sensor jitter. The reported fault locations are uniformly converted to the standard coordinate system used internally. For example, the coordinate system for China is uniformly converted to the GCJ-02 coordinate system for easy connection with map service providers. Records with missing coordinates or coordinates that are obviously out of bounds are marked and discarded or re-reported. Map elevator fault codes to standard fault types defined within the system; for example, trapped people, doors that cannot be closed, and abnormal emergency stops, eliminating code differences between elevators of different brands.

[0013] The specific operations for state information preprocessing are as follows: Receive continuous positioning data streams from maintenance personnel, smooth the GPS coordinates using Kalman filtering or moving average methods, eliminate obvious drift points (e.g., instantaneous jumps exceeding 500 meters), and convert the coordinate system to the standard coordinate system used internally by the system. Status values ​​(such as 0 / 1 or strings) are uniformly mapped to predefined standard tags within the system, such as idle, dispatching, repairing, etc., and the timeliness of the status is checked: if a repairman fails to report the status within a predetermined time (such as 5 minutes), he / she is automatically marked as unknown or offline and is not included in the subsequent screening of candidate repairmen. The system associates and maps the skill tags registered by maintenance personnel in the system, such as "Brand A gate machine maintenance" and "Brand B control system maintenance", with predefined fault type capabilities to generate a standardized skill set, which makes it easier for the candidate screening module to directly match skills. A timestamp is appended to each status message. When a subsequent module requests the latest status, only data within the valid window (e.g., within 1 minute) of the current time is returned, and expired data is considered invalid.

[0014] The preprocessing rules for the spare parts list are as follows: For each maintenance worker's spare parts list snapshot, if the timestamp of the spare parts list snapshot is more than 24 hours from the current time and the worker has not responded to the verification reminder, the entire list will be marked as time-sensitive. Items whose update timestamp is more than 24 hours from the current time will not be included in the subsequent spare parts matching calculation. If the maintenance personnel have responded to the verification reminder, the list confirmed by the maintenance personnel shall prevail, regardless of the snapshot timestamp, and the snapshot timestamp of the list shall be updated to the verification time. Merge multiple entries with the same spare part type code, using the number corresponding to the most recent update timestamp of the entry as the merging criterion; For items with missing or unidentifiable spare parts type codes, mark them as pending verification. Items marked as pending verification in the entire list will not be included in subsequent spare parts matching calculations.

[0015] Historical profile module: During system initialization, a personalized profile of each maintenance worker is constructed, as well as a probability mapping between fault codes and required spare parts types; the personalized profile represents the maintenance worker's ability assessment value under different time and space conditions; For any maintenance worker, their personalized profile is constructed as a multi-dimensional mapping relationship, which can output their capability assessment value for any given time period and location.

[0016] The process of generating a personalized profile is as follows; please refer to [link / reference]. Figure 2 : The first step is to extract all historical work orders for each repairman. Each historical work order should include at least: the timestamp of the work order and the corresponding time period attributes (such as weekday morning peak, weekday off-peak, weekend, etc.), the geographical location of the fault point, the fault type, the geographical location of the repairman when the order was accepted, the actual time taken to arrive on site, the time taken to obtain the missing parts, and the effective time taken. Actual arrival time refers to the time taken from receiving the order to arriving at the fault site; time for obtaining missing parts refers to the extra time spent by the repair personnel after arriving at the site, having to detour to the spare parts warehouse to retrieve the parts or waiting for the spare parts to be delivered due to the lack of spare parts. This value is zero for work orders without missing parts. Effective time is equal to the sum of actual time spent on site and time spent obtaining missing parts, representing the total time actually spent by maintenance personnel on site and having the conditions for maintenance. If the time taken to acquire missing parts is not recorded in the historical work order, the time taken to acquire missing parts will be determined by comparing the actual spare parts used in the work order with the snapshot of the spare parts list of the maintenance personnel at the time of dispatch: if all the spare parts actually used are included in the spare parts list at the time of dispatch, the time taken to acquire missing parts is 0; otherwise, the time interval between the arrival of the maintenance personnel and the actual start of maintenance will be used as an approximate value of the time taken to acquire missing parts.

[0017] The second step is to divide all the geographic space covered by the system into non-overlapping grid units, such as a 1 km × 1 km square, with each grid unit as an independent spatial area and given a unique identifier. The third step is to map the geographical location of the fault to the corresponding grid cell for each historical work order, and record it as the fault area of ​​that historical work order. The fourth step is to discretize the preset timeline into a finite set of time period attributes. Each time period attribute includes the time period and the corresponding attributes. For example, a week can be divided into: weekday morning peak (7:00-9:00), weekday evening peak (17:00-19:00), weekday off-peak (0:00-7:00, 9:00-17:00, 19:00-24:00), and weekend (Saturday and Sunday as a whole are treated as a separate time period). The fifth step is to iterate through all spatial regions and time period attributes, and then use each spatial region as the target region in turn. A joint weighted algorithm based on time decay factor and spatial clustering weight is used to calculate the maintenance personnel. Attributes in each time period Capability assessment values; capability assessment values ​​include weighted average time taken, weighted time standard deviation, and recent response trend; Weighted average time The calculation formula is as follows:

[0018] in, For maintenance personnel The Effective time spent on each historical work order; For the first Time decay factor for historical work orders; For the first Spatial clustering weights for historical work orders; For maintenance personnel Time period attribute The historical work order index set within; The time decay factor is calculated using an exponential decay method, as shown in the following formula:

[0019] in, The attenuation coefficient; For the current moment and the first The time difference between the completion times of each work order, in days. is a natural constant, approximately 2.71828, and is the base of the natural logarithm function; The formula for calculating the attenuation coefficient is as follows:

[0020] in, This is the number of days it takes for the time decay factor to decay to half of its initial value. It can be flexibly selected based on the business scenario: if maintenance personnel turnover is low and traffic patterns are stable, set it to 30-60 days; if maintenance personnel turnover is frequent or urban road conditions change rapidly, set it to 7-1 days. For example... sky, The system can be adjusted based on actual prediction results after it is running; Spatial clustering weights are determined based on the spatial relationship between the fault area and the target area of ​​historical work orders:

[0021] in, for and The spatial distance between them is usually expressed as the Manhattan distance or road network distance between the center points of the grid cells; It is the attenuation factor; The threshold for the spatial influence radius; This indicates that the fault area of ​​the historical work order falls exactly within the target area being calculated. This indicates that the fault area of ​​the historical work order is inconsistent with the target area being calculated currently; Attenuation factor This indicates the degree of transferability of historical information in neighboring areas, with a value range of (0, 1). Its value is determined by road network density and differences in inter-regional traffic: cities with dense road networks and high similarity in regional response times... It can be set to around 0.9; for cities with sparse road networks and large differences in response between regions, It can be set to 0.7; Spatial influence radius threshold This represents the distance boundary in kilometers where historical work orders become completely irrelevant to the target area. It can be set directly based on the statistical distribution of service distances between the order acceptance location and the fault location in historical work orders: calculate the service distances of all historical work orders and take the 90th percentile (i.e., 90% of the dispatch distances do not exceed this value) as the boundary. The initial value; or it can be set by the operator as the maximum service radius for maintenance personnel (e.g., 20 kilometers). ; The formula for calculating the weighted average time standard deviation is as follows:

[0022] in, The weighted standard deviation of time consumption; The smaller the value, the more stable the response time of the maintenance personnel under the corresponding time and space conditions; conversely, the larger the value, the greater the fluctuation.

[0023] The formula for calculating the recent response trend is as follows:

[0024] in, This reflects recent response trends; For maintenance personnel The average effective time spent on work orders over a recent period (30 days or earlier); For maintenance personnel The average effective time of work orders over a period of time in the early (recent) period; a positive value in the recent response trend indicates that the response time is increasing and the efficiency is decreasing, while a negative value indicates that the time is decreasing and the efficiency is increasing.

[0025] The sixth step is to store the capability assessment values ​​as structured profile data with maintenance personnel, time period attributes, and target area as joint primary keys to generate personalized profiles of maintenance personnel.

[0026] The system maintains a timestamped version snapshot library for each maintenance worker's personalized profile. Before updating the personalized profile each time an incremental work order is completed, the current version of the maintenance worker's profile is completely copied into a timestamped historical version record, and then an overwrite update is performed. Each snapshot record includes the maintenance worker's identifier, time period attribute, target area, capability assessment value, and version generation timestamp. When constructing training samples for the prediction model, the profile status at any historical moment is retrieved by timestamp.

[0027] For new maintenance personnel without historical work orders, the system uses the average of the capability assessment values ​​of all maintenance personnel with similar skill tags and adjacent service areas within the same time period as the default capability assessment value, and updates it gradually as the individual's historical work orders accumulate.

[0028] The specific steps for mapping fault codes to required spare parts types are as follows: The first step is to group historical work orders for elevators of the same brand according to fault codes; The second step is to count the frequency of occurrence of each spare part type code within each group and sort them in descending order of frequency. The third step is to calculate the frequency of occurrence of each spare part under the fault code, which is taken as the required probability for that spare part. If the sample size in a fault code group is lower than a preset lower limit (e.g., less than 5 entries), work orders of adjacent brands or similar models of elevators are merged for supplementary statistics based on the brand-model similarity matrix. Adjacent brands refer to brands whose elevator manufacturers belong to the same group or share the same control system platform. Similar models refer to models with the same door operator system type, traction machine type, control system platform, or interchangeable spare parts. The brand-model similarity matrix is ​​maintained during system initialization to determine whether to merge statistics. The fourth step is to store the spare parts list in each group in descending order of their required probabilities into a probability mapping table. Only the top few spare parts in each group whose cumulative occurrence frequency reaches a preset frequency threshold (e.g., 90%) are retained, and the rest are classified into the "other spare parts" category.

[0029] For new fault codes for which there are no historical work orders, the system constructs an initial probability mapping based on the average spare parts usage frequency of all fault codes under that elevator brand; if brand-level samples are also missing, the system is initialized with the average spare parts usage frequency of all brands within the system.

[0030] When the system is first deployed and there are no historical work orders, the probability mapping table is empty, and the aforementioned brand-level and system-level averages are unavailable. At this time, the system administrator configures a pre-set initial probability mapping based on the elevator maintenance manual, spare parts BOM list, or manufacturer recommendations. Before the initial mapping configuration is complete, the system automatically skips spare parts matching degree filtering, using only skill matching and idle status filtering results as candidate maintenance personnel, and sets the spare parts matching characteristics to default values: sum of required probabilities for each spare part = 0, number of key spare parts hit = 0, identifier of the spare part with the highest missing probability = 1, spare parts completeness = 0. After the first batch of historical work orders accumulates, the system automatically updates the probability mapping according to actual maintenance data and resumes normal spare parts matching degree filtering.

[0031] When a fault occurs, the system uses the fault code and elevator brand and model as keys to query the probability mapping table to obtain the list of required spare parts and the required probability of each spare part, thus forming the spare parts demand prediction result. The spare parts demand forecast results are output as a list of spare parts type codes plus required probabilities.

[0032] After each work order is completed, the system merges the actual spare parts information used in that work order into the probability mapping table as an incremental data entry, and updates the spare parts frequency and required probability under the corresponding fault code group.

[0033] The system maintains a timestamped version snapshot library for the fault code-spare part probability mapping table. Before each merge update is performed upon completion of a work order, the current version of the probability mapping table is completely copied into a timestamped historical version record before the merge update is executed. Each snapshot record contains the generation timestamp and the complete fault code-spare part probability mapping content before the update. When constructing training samples for the prediction model, the probability mapping status at any historical moment is retrieved by timestamp.

[0034] Traffic Feature Module: Based on preset filtering conditions, candidate maintenance personnel are selected from all maintenance personnel; and real-time traffic feature vectors and spare parts matching features of each candidate maintenance personnel from real-time geographical location to fault geographical location are obtained; The selection process for candidate maintenance personnel is as follows: The first step is to match the fault type with the skill tags of all maintenance personnel, and retain only maintenance personnel who are qualified to handle that type of fault. The second step is to screen the maintenance personnel who are currently available in the first step; The third step is to calculate the spare parts matching degree for each maintenance personnel after the first two steps, and retain only individuals whose spare parts matching degree is not lower than the preset matching degree threshold to obtain a set of candidate maintenance personnel.

[0035] The spare parts matching degree is calculated as follows: The three spare parts with the highest probability of being needed from the spare parts demand forecast are selected as the key spare parts set for this maintenance (the top five are selected when the work order urgency level is I, and the top three are selected when it is II and III; spare parts matching degree screening can be skipped for level III ordinary faults). The number of types of the key spare parts set that match the spare parts list currently carried by the candidate maintenance personnel is counted. The number of matching types is divided by the total number of types in the key spare parts set, and the resulting ratio is the spare parts matching degree. The higher the spare parts matching degree, the higher the probability that the maintenance personnel will complete the maintenance in one go after arriving on site.

[0036] If the spare parts demand forecast result is empty and the set of key spare parts cannot be determined, the spare parts matching degree screening is skipped, and all maintenance personnel who pass the skill matching and idle status screening are regarded as candidate maintenance personnel; their spare parts matching characteristics are uniformly processed according to the default value.

[0037] The default threshold for spare parts matching is 0.5, meaning that maintenance personnel must carry at least half of the first three key spare parts to pass the screening. This threshold can be adjusted based on the availability of spare parts and historical statistics on the spare parts carrying rate of maintenance personnel. For Level I emergency work orders, if no one passes the screening after using the default threshold, the threshold can be temporarily lowered to 0.2 to expand the candidate range and avoid excessive compression of emergency work orders in the spare parts screening stage.

[0038] If no maintenance personnel pass the screening according to the above threshold, the following downgrade process will be implemented: All maintenance personnel after skill matching and status screening will be ranked from high to low according to spare parts matching degree, and the top two will be directly selected as candidate maintenance personnel. After identifying the target maintenance personnel and executing the work order, a missing parts alarm is sent to the target maintenance personnel. The alarm content includes the type of missing critical spare parts and their required probability, the location of the nearest spare parts warehouse and the navigation path. At the same time, this downgrade screening event is recorded in the work order log, including the trigger time, work order number, downgrade reason and the spare parts matching degree of each maintenance personnel before downgrade.

[0039] Real-time traffic feature vectors are obtained from real-time traffic flow data from the real-time geographical location of candidate maintenance personnel to the geographical location of the fault.

[0040] Real-time traffic flow data is obtained by calling the real-time traffic interface of a third-party map service provider. This interface returns in real time the passage data of one or more feasible paths from the current location of each candidate maintenance personnel to the fault point. For real-time traffic feature vectors, please refer to [link / reference]. Figure 3 ,include: Shortest estimated travel time: The estimated travel time for the feasible path returned by the map interface, in minutes; Real-time congestion index: The ratio of the actual travel time of a feasible path to the free-flow travel time; where free-flow travel time refers to the travel time required for a vehicle to pass through the path under ideal conditions where the road is completely unobstructed and there are no traffic signal delays. It is provided directly by the map service provider based on the road grade and historical free-flow speed, or estimated by dividing the path distance by the design free-flow speed corresponding to the road type. Total number of traffic lights along the route: The number of traffic lights along the feasible route, derived from map road network data; Path distance: Total driving distance of feasible paths, in kilometers, sourced from map interface; Traffic incident indicator: Whether there are traffic accidents or road closures on the feasible route, derived from a real-time traffic incident list returned by the map interface.

[0041] For each candidate maintenance worker, the spare parts matching characteristics are obtained, including: Sum of required probabilities for each spare part: The sum of the required probabilities of all spare parts that match the spare part demand prediction results among the spare parts currently carried by the candidate maintenance personnel, and normalized to [0, 1]. The larger the value, the higher the degree to which the carried spare parts cover the demand. Key spare parts hit count: The number of elements in the intersection of the key spare parts currently carried by the candidate maintenance personnel and the set of key spare parts (the top three / top five required probabilities) determined in the spare parts matching degree calculation; Highest probability missing spare part identifier: If the spare part with the highest probability of being needed is not carried, it is marked as 1; otherwise, it is marked as 0. Spare parts completeness: The proportion of spare parts that the candidate maintenance personnel are currently carrying that match all the spare parts in the spare parts demand forecast results, with a value range of [0, 1].

[0042] Decision module: Input the real-time traffic feature vector, spare parts matching feature, and capability assessment value that matches the current spatiotemporal conditions of each candidate maintenance personnel into the pre-trained prediction model to obtain the quintiles prediction value, quintiles prediction value, and confidence level of the effective arrival time of each candidate maintenance personnel, and confirm the target maintenance personnel based on these values ​​and execute work order dispatch.

[0043] The prediction models include the quintile model and the decile model.

[0044] The training samples for the prediction model are constructed on a per-historical-work-order basis; for a completed historical-work-order: Based on the time period and spatial area of ​​the historical work order, retrieve the portrait version of the maintenance personnel from the portrait version snapshot library that was generated earlier than the dispatch time of the historical work order and is closest in time, and read the corresponding capability assessment value from it to ensure that future information is not introduced during training. Obtain the real-time traffic feature vector collected and stored at the time of the historical work order dispatch; if the real-time traffic features were not stored at that time, they can be backtracked and filled through the historical traffic data query interface provided by the third-party map service provider; At the same time, add matching features for the spare parts for this historical work order: The spare parts demand prediction result is obtained by taking the snapshot of the most recent probability mapping version generated before the dispatch time of the historical work order. Based on the snapshot of the spare parts list reported by the maintenance personnel at the dispatch time, the spare parts matching characteristics of the work order are recalculated according to the current online logic.

[0045] If it is not possible to retrieve a snapshot of the spare parts list at the time of dispatch, and there is no record of the spare parts list that the maintenance personnel last properly inventoried before that time in the historical database, then this historical work order will not be included in the model training sample set with spare parts matching features for the time being, and will be supplemented after the relevant data of the work order is obtained later. If the historical database retains the spare parts list record of the most recent normal inventory before the time of the maintenance personnel, then the most recent record is used to approximate the snapshot of the dispatch time to construct the spare parts matching feature of the work order; at the same time, a snapshot quality flag feature is added to the sample, with the flag set to 0 for accurate snapshot samples and 1 for approximate snapshot samples; During the online prediction phase, all spare parts list snapshots for candidate maintenance personnel are real-time snapshots, so the quality marker feature of this snapshot is uniformly set to 0; this feature is always used as one of the model input features in training and online prediction to ensure that the input feature structure is completely consistent in the two stages. The aforementioned capability assessment values, real-time traffic feature vectors, spare parts matching features, and snapshot quality label features are integrated into an input feature set, with the effective time consumption of the historical work order record used as the training label.

[0046] All historical work orders were processed into sample data in the manner described above and then fed into the regression model for training. The model can employ gradient boosting tree models (such as LightGBM) or deep cross-networks, which are capable of effectively learning the non-linear mapping relationship between mixed features and response time. In order to output the confidence of the prediction, instead of training a single point value prediction model, we train prediction models responsible for two different target quantiles. The two quantiles selected are the fiftieth quantile and the ninetieth quantile; For each quantile, the input feature set is fed into a gradient boosting tree model (e.g., LightGBM) or a deep cross-network, and the parameters are optimized using the quantile loss function; the quantile loss function is defined as follows:

[0047] in, The loss value. This represents the actual and valid time spent on this historical work order; The model is positioned at the target quantile for this historical work order. The predicted time is as follows; The target quantiles are set to 0.5 and 0.9 respectively. Both models optimize their parameters by minimizing the sum of the target quantile losses for all training samples; after training, the two models are serialized and saved.

[0048] When making predictions, the real-time traffic feature vectors of candidate maintenance personnel, spare parts matching features, and their capability assessment values ​​under the current spatiotemporal conditions are combined as inputs and input into two quantile models respectively to obtain the predicted values ​​of the median and the ninetieths. The median is taken as the effective prediction time, and the 90th percentile prediction value is the 90th percentile prediction value. The confidence level is measured by the difference between the 90th percentile predicted value and the 50th percentile predicted value. The smaller the difference, the lower the uncertainty of the prediction and the higher the confidence level. The process for confirming the target maintenance personnel is as follows: Work orders with an urgency level greater than or equal to a preset threshold (e.g., Level I, Level II): The first step is to determine whether the predicted value of the ninetieths of each candidate maintenance worker is not greater than the time limit threshold corresponding to the work order (for example, 30 minutes for level I and 60 minutes for level II). Only candidate maintenance workers who meet this condition are retained to form a set of feasible dispatch objects. The second step is to determine the target maintenance personnel within the set of feasible dispatch objects according to the rule that the one with the smallest 50ths predicted value is given priority, and the one with the highest confidence level is given priority when the 50ths predicted values ​​are the same. The third step is to select the candidate maintenance personnel whose predicted value of the 90ths place exceeds the time limit threshold. Then, the candidate maintenance personnel with the smallest predicted value of the 90ths place is selected as the target maintenance personnel, and an alarm message is pushed to the emergency collaboration platform. The alarm message includes the spare parts demand prediction result of this work order so that emergency personnel can bring the required spare parts in advance. At the same time, a spare parts shortage alarm and the location of the nearest spare parts warehouse are pushed to the target maintenance personnel. For work orders with an emergency level lower than the preset level threshold (Level III): among all candidate maintenance personnel, the target maintenance personnel are determined according to the rule that the one with the smallest 50ths predicted value is given priority, and if the 50ths predicted values ​​are the same, the one with the highest confidence level is given priority. Once the target repair personnel are identified, a work order is dispatched, and at the same time, the spare parts demand forecast is sent to the target repair personnel. If the spare parts carried by the target repair personnel do not fully cover all the spare parts required for this repair, the dispatched content will include an additional list of missing spare parts, the location of the nearest spare parts warehouse, and a suggested pick-up route, so that the repair personnel can decide whether to pick up the parts along the way based on their own situation.

[0049] If the confidence level and the predicted effective arrival time are both within the same order of magnitude, for example, the difference is less than the preset threshold, then the system will randomly select or the staff will select the target maintenance personnel and execute the work order dispatch.

[0050] In real-world work order dispatching scenarios, maintenance personnel may encounter unforeseen circumstances such as sudden traffic congestion, traffic accidents, or temporary road closures en route to the scene after accepting an order. This can significantly extend the actual arrival time compared to the predicted time at the time of dispatch. For highly urgent work orders, such as those involving trapped personnel, this delay can have serious consequences. To address the uncertainty in this dynamic environment, this system introduces a dynamic rescheduling mechanism after dispatching an order. This mechanism performs continuous en route tracking and conditional re-optimization on dispatched work orders. The specific operation is as follows: After the work order is successfully dispatched, the data acquisition module continues to receive the real-time geographical location report through the maintenance personnel's mobile terminal at a preset frequency (e.g., every 30 seconds). Whenever a new location update is received, the system takes the current location of the maintenance personnel as the new starting point and the geographical location of the fault point as the ending point, and calls the real-time traffic condition interface of a third-party map that is completely consistent with the traffic feature module to re-obtain traffic features such as the estimated travel time and congestion index at the current moment. These real-time traffic features are concatenated with the corresponding capability assessment value of the maintenance worker in the user profile database and fed into the deployed quantile prediction model to obtain the effective arrival time re-predicted at the current moment. The rescheduling trigger conditions for emergency work orders (Level I and Level II) are as follows: The initial ninetieths prediction value of the work order is recorded at the time of dispatch. During each location update in the on-the-go tracking, the system inputs real-time traffic characteristics, spare parts matching characteristics, and capability assessment values ​​under the current spatiotemporal conditions into the prediction model to obtain real-time decimal prediction values. For Level I work orders, at any tracking time, as long as the real-time 90ths digit prediction value exceeds the statutory time limit of 30 minutes, the secondary optimization process is immediately triggered without waiting for the time delay to reach the threshold. For Level II work orders, a secondary optimization is triggered when the real-time 90ths percentile predicted value exceeds its configurable time limit threshold (default 60 minutes) and is outside the preset deviation time range (e.g., 5 minutes) to reduce unnecessary rescheduling costs.

[0051] The rescheduling trigger conditions for Level III work orders are as follows: The system continuously compares the difference between the real-time quintile prediction value and the initial quintile prediction value recorded at the time of dispatch, which is called the time delay. When the time delay exceeds the preset rescheduling trigger threshold (15 minutes by default, but can also be configured), the secondary optimization process is started.

[0052] The specific operations for the second optimization stage are as follows: Include all currently available maintenance personnel with the skills to repair this type of fault in the candidate pool, while retaining maintenance personnel who have already been dispatched and are en route. For the new candidate set, the traffic feature module is requested one by one for the real-time traffic feature vectors starting from the current location of each candidate, and the spare parts matching degree and spare parts matching features of each candidate are recalculated. For maintenance personnel who have been screened and eliminated by spare parts matching degree during the initial dispatch, they are re-evaluated during the second optimization, because they may have replenished spare parts in the middle. The capability assessment value that matches the current time period attributes and target area is extracted from the personalized profile, input into the prediction model, and the latest 50ths digit prediction value, 90ths digit prediction value and confidence level of each candidate are obtained. Finally, in accordance with the decision-making rules for the initial dispatch, the best candidate is selected from all candidates (including the original dispatch personnel who are currently en route). If the work order is an urgent work order that has reached the preset threshold, the second optimization will still perform time limit feasibility screening, that is, the premise is that the predicted value of the ninetieth digit does not exceed the time limit threshold. If the optimal personnel selected in the second search are the maintenance personnel who are already en route, it means that despite the road delays, they are still the fastest choice. The system will not make any changes, but will only update the predicted effective time in the dispatch record and continue to track it.

[0053] If another target maintenance personnel is selected in the second search, the system will perform work order handover: send the work order details and spare parts demand forecast results to the newly selected target maintenance personnel, and send a cancellation notice to the original dispatched maintenance personnel, informing them that they do not need to continue to go to the fault point; The original maintenance personnel's busy status has been restored to idle, and the new maintenance personnel's status has been updated to "dispatched". The system records this rescheduling event in the dispatch log, including the triggering reason, the original maintenance personnel's identifier, the new maintenance personnel's identifier, the predicted time of rescheduling, and the spare parts matching degree at the time of rescheduling.

[0054] When the target maintenance personnel arrive at the site and report their arrival via mobile terminal, the system automatically records the effective time of this work order, the actual time period attribute of the execution, the spatial area where the fault point is located, and the spare parts type code and quantity actually used for the work order after the repair is completed; the above data are then entered into three update links: Profile update process: Using the effective time consumption of the work order, the actual execution time period attribute, and the fault location spatial area as incremental data, the system recalculates the capability assessment value of the maintenance personnel under the time period attribute and spatial area; before performing the overwrite update, the system first copies the current version of the maintenance personnel's profile as a time-stamped historical version snapshot and stores it in the profile version snapshot library; then the recalculated capability assessment value is used to overwrite the original value in the personalized profile, and the update takes effect immediately; Spare parts mapping update link: Using the spare parts type code and quantity actually used in the work order as incremental data, update the spare parts frequency and required probability under the corresponding fault code group in the probability mapping table; before performing the merge update, the system first completely copies the current version of the probability mapping table into a historical version snapshot with a timestamp and stores it in the probability mapping version snapshot library; then the merge update is performed so that the spare parts demand prediction is continuously calibrated with the actual maintenance data. Prediction model update chain: Taking the work order as a new training sample, extract the capability assessment value, real-time traffic feature vector, and spare parts matching feature according to the aforementioned training sample construction method, and use its effective time consumption as a training label. Periodically incorporate it into the incremental retraining of the prediction model so that the quantile prediction model continuously fits the latest spatiotemporal response pattern and spare parts matching pattern.

[0055] If a different target maintenance worker is selected in the second search, the original target maintenance worker's personalized profile will not be updated. Only the profile and corresponding spare parts mapping of the target maintenance worker who successfully executed the work order will be updated. For the original target maintenance worker, the system will record an event of dispatched but not executed in their maintenance worker file as a negative reference item for evaluating the reliability of the worker's response when building the profile in the future. This event will not affect the calculation of their spare parts matching characteristics.

[0056] This invention proposes a dynamic proximity-based elevator dispatching system that integrates traffic flow and historical response rates. By constructing personalized profiles of maintenance personnel that integrate spatiotemporal characteristics and mapping fault and spare parts probabilities, combined with real-time traffic feature vectors and quantile prediction models, it achieves reliable prediction of arrival time and quantification of uncertainty, effectively overcoming the limitations of static distance dispatching and decision-making based on a single prediction value. At the same time, it introduces in-transit tracking and dynamic rescheduling mechanisms to support on-demand work order handover, solving the defect of traditional open-loop dispatching that cannot intervene and adjust under sudden traffic conditions, thereby significantly improving dispatching accuracy and emergency fault response efficiency.

[0057] In summary, this invention improves the accuracy and risk controllability of dispatch decisions by integrating historical response profiles of maintenance personnel with real-time traffic characteristics and combining quantile prediction models to output reliable estimates of arrival time and their confidence levels. Simultaneously, by incorporating spare parts matching into dispatch considerations, it effectively increases the first-time repair rate after maintenance personnel arrive, avoiding secondary delays caused by missing parts. Furthermore, the system introduces in-transit tracking and dynamic rescheduling mechanisms, which can automatically trigger work order handover when road conditions change abruptly, compensating for the lack of mid-course intervention capabilities in traditional open-loop dispatching. This comprehensively enhances the system's emergency response efficiency and dispatch robustness in complex urban environments.

[0058] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A dynamic, proximity-based elevator dispatching system that integrates traffic flow and historical response rate, characterized in that, include: Data acquisition module: Collects elevator fault information and maintenance personnel status information in real time; The elevator fault information includes the fault location, fault type, fault code, elevator brand and model, and emergency level; The status information includes the maintenance personnel's real-time geographical location, skill tags, busy status, and current spare parts list; Historical profile module: During system initialization, a personalized profile of each maintenance worker is built, as well as a probability mapping between fault codes and required spare parts types; The personalized profile represents the maintenance personnel's ability assessment value under different time and space conditions; Traffic Feature Module: Based on preset filtering conditions, candidate maintenance personnel are selected from all maintenance personnel; and real-time traffic feature vectors and spare parts matching features of each candidate maintenance personnel from real-time geographical location to fault geographical location are obtained; Decision module: Input the real-time traffic feature vector, spare parts matching feature, and capability assessment value that matches the current spatiotemporal conditions of each candidate maintenance personnel into the pre-trained prediction model to obtain the quintiles prediction value, quintiles prediction value, and confidence level of the effective arrival time of each candidate maintenance personnel, and confirm the target maintenance personnel based on these values ​​and execute work order dispatch.

2. The system according to claim 1, characterized in that, The personalized profile represents the maintenance personnel's competence assessment value under different spatiotemporal conditions, and the generation process is as follows: The first step is to extract all historical work orders for each maintenance worker; The second step is to divide all the geographic space covered by the system into non-overlapping grid units, with each grid unit serving as an independent spatial region; The third step is to map the geographical location of the fault to the corresponding grid cell for each historical work order, and record it as the fault area of ​​the historical work order. The fourth step is to discretize the preset time axis into a finite set of time period attributes, each of which includes a time period and the corresponding attributes. The fifth step is to traverse all spatial regions and time period attributes, take each of the spatial regions as the target region in turn, and use a joint weighted algorithm based on time decay factor and spatial clustering weight to calculate the capability assessment value of maintenance personnel for each time period attribute. The sixth step is to store the capability assessment values ​​as structured profile data with maintenance personnel, time period attributes, and target area as joint primary keys to generate personalized profiles of maintenance personnel.

3. The system according to claim 2, characterized in that, The capability assessment values ​​include the weighted average time taken, the weighted standard deviation of time taken, and the recent response trend; The weighted average time is used to characterize the weighted average time taken for maintenance personnel to complete a response under specific spatiotemporal conditions; The weighted time standard deviation is used to characterize the stability of the response time of maintenance personnel; The recent response trend is used to characterize the direction of change in the increase or decrease of the response time of maintenance personnel.

4. The system according to claim 3, characterized in that, The specific operation of the probability mapping between the fault code and the required spare part type is as follows: The first step is to group historical work orders for elevators of the same brand according to fault codes; The second step is to count the frequency of occurrence of each spare part type code within each group and sort them in descending order of frequency. The third step is to calculate the frequency of occurrence of each spare part under this fault code, which is taken as the required probability of that spare part. The fourth step is to store the spare parts list in each group in descending order of their required probabilities into a probability mapping table, and only retain the top few spare parts whose cumulative occurrence frequency reaches the preset frequency threshold in each group, and classify the rest into other spare parts categories.

5. The system according to claim 4, characterized in that, The selection process for the candidate maintenance personnel is as follows: The first step is to match the fault type with the skill tags of all maintenance personnel, and retain only maintenance personnel who are qualified to handle the fault type. The second step is to screen the maintenance personnel who are currently available in the first step; The third step is to calculate the spare parts matching degree of each maintenance personnel after the first two steps, and only retain individuals whose spare parts matching degree is not lower than the preset matching degree threshold to obtain candidate maintenance personnel. The spare parts matching degree is calculated as follows: Select the top N spare parts with the highest probability of being needed from the spare parts demand forecast results as the set of critical spare parts. The spare parts demand prediction result is obtained by the system querying the probability mapping table using the fault code and elevator brand and model as keys to obtain the list of required spare parts and the required probability of each spare part. The number of types of the critical spare parts set that match the current spare parts list carried by the candidate maintenance personnel is counted. The ratio of the number of matching types to the total number of types in the critical spare parts set is the spare parts matching degree.

6. The system according to claim 5, characterized in that, The spare parts matching features include: Sum of probabilities required for each spare part: the sum of the probabilities required for all spare parts that match the spare part demand prediction results among the spare parts currently carried by the candidate maintenance personnel, and normalized to the interval [0, 1]. Critical spare parts hit count: The number of elements in the intersection of the spare parts currently carried by the candidate maintenance personnel and the set of critical spare parts; Highest probability missing spare part identifier: Identifier indicating whether the required highest probability spare part is carried; Spare parts completeness: The percentage of spare parts that a candidate maintenance personnel currently carry that match all the spare parts predicted in the spare parts demand forecast.

7. The system according to claim 6, characterized in that, The process for confirming the target maintenance personnel is as follows: When the emergency level is greater than or equal to the preset level threshold, only the candidate maintenance personnel whose 90th percentile predicted value is not greater than the preset time limit threshold are included in the dispatchable objects, and then the target maintenance personnel are determined according to the rule that the one with the smallest 50th percentile predicted value is given priority, and the one with the highest confidence level is given priority when the 50th percentile predicted values ​​are the same. When there are no candidate maintenance personnel whose predicted value of the 90ths place is not greater than the preset time limit threshold, the candidate maintenance personnel with the smallest predicted value of the 90ths place is selected as the target maintenance personnel, and an alarm message is issued. Conversely, the target maintenance personnel are determined according to the rule that the smallest predicted 50ths value is given priority, and the highest confidence level is given priority when the predicted 50ths values ​​are the same. Based on the target maintenance personnel, work orders are dispatched.

8. The system according to claim 7, characterized in that, The system is equipped with a rescheduling mechanism, the triggering mechanism of which is as follows: For the target maintenance personnel whose emergency level is greater than or equal to the level threshold, when their location is updated, the system will input the real-time traffic feature vector, spare parts matching features and the capability assessment value under the current spatiotemporal conditions into the prediction model and output its 190ths predicted value. When the predicted value of the 90th percentile exceeds the time limit threshold, or exceeds the time limit threshold but is outside the preset deviation time range, a secondary optimization process is triggered. For the target maintenance personnel whose emergency level is less than the level threshold, the system inputs the real-time traffic feature vector, spare parts matching features, and capability assessment value under the current spatiotemporal conditions into the prediction model to obtain a real-time quintile prediction value; when the difference between the real-time quintile prediction value and the quintile prediction value recorded at the time of dispatch exceeds the preset rescheduling trigger threshold, a secondary optimization process is triggered.

9. The system according to claim 8, characterized in that, The specific operation of the secondary optimization process is as follows: According to the filtering method in the traffic feature module, re-filter the candidate maintenance personnel and retain the current target maintenance personnel; The new target maintenance personnel will be reconfirmed according to the target maintenance personnel confirmation method in the decision module.

10. The system according to claim 9, characterized in that, In the secondary optimization process, the spare parts matching degree and spare parts matching features need to be recalculated and obtained. If the new target maintenance personnel is not the original target maintenance personnel, the work order will be handed over, and the original target maintenance personnel's busy status will be restored to idle status.

Images