An ai-based elevator update control method and system

Abstract

The application relates to the technical field of elevator module updating, and discloses an elevator updating control method and system based on AI, which comprises the following steps: obtaining an adjustment factor by prediction based on running state data, combining a pre-constructed waiting time model to calculate elevator waiting time, enabling the elevator service state to dynamically reflect running differences, dividing the elevator coverage area of a target elevator group through the elevator waiting time, evaluating a migration load value based on a queuing theory model, constructing a dynamic upgrading path graph, enhancing the identification capability of the task transfer capability difference between elevators, and finally determining an elevator upgrading sequence based on a graph search algorithm, taking minimizing the migration load value as the target, so that the upgrading tasks are distributed in the order of minimizing the overall system service influence when multiple elevator units need to be updated, thereby avoiding the problems of overload, task backlog or service blind area caused by improper scheduling.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of elevator module updating, and more particularly to an AI-based elevator updating control method and system. BACKGROUND

[0002] In intelligent buildings and high-rise buildings, as a core vertical transportation tool, the elevator system needs to be periodically upgraded to ensure its performance and safety as the software complexity and AI scheduling strategy continue to evolve. Unlike the traditional whole-machine shutdown upgrade method, a modular updating method has been gradually developed in recent years, which only upgrades or remotely replaces one or more functional modules in the elevator control system through hot swapping. This modular updating method does not require the entire elevator to be shut down, and only needs to replace the local logic in a short time to achieve function enhancement and system optimization. It has the advantages of high availability, small upgrade window, and no impact on the main process, and has become an important trend of the new generation of elevator control systems.

[0003] However, in actual applications, when multiple elevator units need to perform modular updating operations within the same time window, the existing elevator scheduling system mostly uses a fixed sequence or simple polling strategy to determine the upgrade sequence of each elevator. This strategy does not consider the current system running state, load difference between elevators, and task coverage relationship, and lacks an upgrade decision mechanism targeting "minimum overall system service impact". Especially during user request peak periods, improper scheduling can cause some elevators to be overloaded, tasks to be accumulated, and even service blind spots to appear.

[0004] In view of this, the present application provides an AI-based elevator updating control method and system to solve the above problems. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the present application provides an AI-based elevator updating control method and system.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] In a first aspect, an AI-based elevator updating control method is provided, comprising:

[0008] Obtaining running state data of each elevator in a target elevator group, predicting an adjustment factor based on the running state data, and calculating an elevator waiting time based on a pre-constructed waiting time model and the adjustment factor, the adjustment factor including a speed adjustment factor and a time adjustment factor, the target elevator group referring to the elevator group that needs to be upgraded;

[0009] The elevator coverage area of ​​the target elevator group is divided according to the elevator waiting time. The migration load value between every two elevators in the elevator coverage area is evaluated based on the queuing theory model. With elevators as nodes and migration load values ​​as edges, a dynamic upgrade path graph is constructed.

[0010] The graph search algorithm transforms the dynamic upgrade path graph into a sequential decision-making process. With the goal of minimizing the migration load value, the elevator upgrade sequence is determined, and the preset upgrade tasks are assigned to the corresponding elevators according to the elevator upgrade sequence.

[0011] In some embodiments, the method for predicting adjustment factors based on operational status data includes:

[0012] The running status data is input into the pre-trained deep neural network model to obtain the adjustment factor of the deep neural network model output.

[0013] Training methods for deep neural network models include:

[0014] The input layer of a deep neural network model receives historical running state data, and the output layer of a deep neural network model outputs historical adjustment factors.

[0015] When training a deep neural network model, the cross-entropy loss function is chosen as the loss function, and the loss function is minimized by gradient descent.

[0016] Update the weight parameters of the deep neural network model and obtain the trained deep neural network model through iterative training.

[0017] In some embodiments, the method for calculating elevator waiting time based on a pre-built waiting time model and an adjustment factor includes:

[0018]

[0019] In the formula, EWT′ i Let f be the elevator waiting time calculated after introducing an adjustment factor for elevator i, and let f be the floor number currently requested by the passenger. i Let v be the floor number where elevator i is currently located. i Let f be the standard design operating speed of elevator i, Δv be the speed adjustment factor, n be the number of floors currently stopped in the elevator's task queue, and f be the speed of elevator i. ij For the j-th target floor to be served in the elevator task queue, f i(j-1) Let t be the floor where the elevator is located during the (j-1)th task. s Δt is the standard average dwell time of the elevator for each normal stop at a floor, and Δt is the time adjustment factor.

[0020] In some embodiments, the method for dividing the elevator coverage area of the target elevator group according to the elevator waiting time comprises:

[0021] The elevator waiting time is subjected to sliding differential processing to obtain a waiting time change rate;

[0022] The average waiting time deviation between the elevator waiting time and the historical waiting time is calculated;

[0023] The preset coverage area is divided into Q floor sub-areas, the waiting time change rate and the average waiting time deviation are mapped to a preset two-dimensional state diagram, and an adjustment index is output through the two-dimensional state diagram;

[0024] When the adjustment index is greater than or equal to a preset index threshold, Q-K floor sub-areas are taken as the elevator coverage area, and when the adjustment index is less than the preset index threshold, Q floor sub-areas are taken as the elevator coverage area, wherein Q is greater than K, and K is an integer greater than 0.

[0025] In some embodiments, the method for dividing the elevator coverage area of the target elevator group according to the elevator waiting time comprises:

[0026] The elevator waiting time is divided into a waiting time sequence through a preset time window, the difference between every two adjacent elements in the waiting time sequence is processed, the unit time change rate between every two adjacent elements is calculated, and the average value between a plurality of unit time change rates is taken as the waiting time change rate.

[0027] In some embodiments, the method for constructing the two-dimensional state diagram comprises:

[0028] The waiting time change rate is taken as the horizontal axis of the two-dimensional state diagram, the average waiting time deviation is taken as the vertical axis of the two-dimensional state diagram, the section division boundary in the two-dimensional state diagram is determined based on a clustering algorithm, the two-dimensional state diagram is divided into P state sections according to the section division boundary, and P is an integer greater than 1.

[0029] In some embodiments, the method for evaluating the transfer load value between every two elevators in the elevator coverage area based on the queuing theory model comprises:

[0030] The queue length difference and the processing quantity difference between every two elevators are extracted based on the queuing theory model, and the transfer load value is determined according to the queue length difference and the processing quantity difference.

[0031] In some embodiments, the method for extracting the queue length difference and the processing quantity difference between every two elevators based on the queuing theory model comprises:

[0032] The running state data and the elevator waiting time of each elevator are input into the queuing theory model, and the future queue length and the future processing quantity of each elevator are predicted, the queue length difference between each two elevators is calculated by difference, and the processing quantity difference between each two elevators is calculated by difference.

[0033] In some embodiments, the method for determining the migration load value according to the queue length difference and the processing quantity difference comprises:

[0034] The first weight factor of the queue length difference is determined according to the entropy weight method, the second weight factor of the processing quantity difference is determined according to the entropy weight method, the queue length difference and the processing quantity difference are weighted and summed based on the first weight factor and the second weight factor, and the migration load value is obtained.

[0035] In the second aspect, an AI-based elevator updating control system is provided, which is used to implement the AI-based elevator updating control method described above, and comprises:

[0036] The prediction module is used to obtain the running state data of each elevator in the target elevator group, to obtain an adjustment factor based on the running state data, and to calculate the elevator waiting time based on the pre-constructed waiting time model and the adjustment factor, wherein the adjustment factor comprises a speed adjustment factor and a time adjustment factor, and the target elevator group refers to the elevator group that needs to be upgraded;

[0037] The first processing module is used to divide the elevator coverage area of the target elevator group according to the elevator waiting time, to evaluate the migration load value between each two elevators in the elevator coverage area based on the queuing theory model, to construct a dynamic upgrading path graph with the elevators as nodes and the migration load value as edges, and to determine the elevator upgrading sequence based on the graph search algorithm.

[0038] The second processing module is used to convert the dynamic upgrading path graph into a sequence decision process based on the graph search algorithm, to determine the elevator upgrading sequence with the minimum migration load value as the target, and to assign the preset upgrading task to the corresponding elevator according to the elevator upgrading sequence.

[0039] Compared with the prior art, the present application has the following advantages:

[0040] This invention obtains adjustment factors based on operational status data and calculates elevator waiting times using a pre-built waiting time model, enabling elevator service status to dynamically reflect operational differences. It then divides the elevator coverage area of ​​the target elevator group based on waiting times and evaluates migration load values ​​using a queuing theory model, constructing a dynamic upgrade path graph. This enhances the ability to identify differences in task transfer capabilities between elevators. Finally, it determines the elevator upgrade sequence based on a graph search algorithm with the goal of minimizing migration load values. This allows for the allocation of upgrade tasks in the order that minimizes the overall system service impact when multiple elevator units need updating, thereby avoiding overload, task backlog, or service blind spots caused by improper scheduling. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating an AI-based elevator update control method according to the present invention.

[0042] Figure 2 This is a flowchart illustrating the method for constructing a two-dimensional state diagram in this invention.

[0043] Figure 3 This is a schematic diagram of the structure of an AI-based elevator update control system according to the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. In the following detailed description, many specific details are set forth to provide a thorough understanding of the exemplary embodiments described. However, it will be apparent to those skilled in the art that the described embodiments may be practiced without some or all of these specific details. In other exemplary embodiments, well-known structures have not been described in detail to avoid unnecessarily obscuring the concepts of this disclosure. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention. Furthermore, the various aspects described in the embodiments may be combined arbitrarily without conflict.

[0045] Example 1

[0046] Please see Figure 1 As shown, this embodiment discloses an AI-based elevator update control method, including:

[0047] S10: Obtain the operating status data of each elevator in the target elevator group, predict the adjustment factor based on the operating status data, and calculate the elevator waiting time according to the pre-built waiting time model and the adjustment factor. The adjustment factor includes a speed adjustment factor and a time adjustment factor. The target elevator group refers to the elevator group that needs to be upgraded.

[0048] In this embodiment, the target elevator group refers to the elevator group that needs to be upgraded. The elevator group can be multiple elevator units with mutual coordinated scheduling relationships deployed in the same building or building group, or it can be a set of elevators marked as the same functional area or logical scheduling unit in the scheduling system, such as high-zone elevators, middle-zone elevators and low-zone elevators in a building divided according to the service floors.

[0049] It should be noted that the operating status data includes, but is not limited to, current floor information, real-time running direction, current load rate, and current idle time. Real-time running direction refers to the elevator's movement trend relative to the floor in the current operating cycle, including "up", "down", or "stationary" states, used to determine whether the elevator is responding to passenger requests in a certain direction. Current load rate refers to the ratio between the current number of passengers or load on the elevator and its maximum rated carrying capacity, usually expressed as a percentage, used to reflect the current congestion level or carrying capacity of the elevator. Current idle time refers to the duration during which the elevator is not assigned an effective task or does not respond to user requests in the current time period, used to assess its scheduleability and service availability.

[0050] Methods for predicting adjustment factors based on operational status data include:

[0051] The running status data is input into the pre-trained deep neural network model to obtain the adjustment factor of the deep neural network model output.

[0052] Training methods for deep neural network models include:

[0053] The input layer of a deep neural network model receives historical running state data, and the output layer of a deep neural network model outputs historical adjustment factors.

[0054] When training a deep neural network model, the cross-entropy loss function is chosen as the loss function, and the loss function is minimized by gradient descent.

[0055] Update the weight parameters of the deep neural network model and obtain the trained deep neural network model through iterative training.

[0056] It is understandable that the adjustment factors include speed adjustment factors and time adjustment factors. The speed adjustment factor represents the dynamic correction required relative to the standard design speed under the current elevator operating conditions, while the time adjustment factor represents the degree of correction of the average dwell time of the elevator at each floor under the current scheduling environment relative to the standard dwell time.

[0057] Taking the current load rate and current idle time as examples, the current load rate is one of the key factors affecting the speed adjustment factor. When the elevator load rate is high (e.g., close to full load), the system often reduces the elevator speed to ensure operational safety and stability, so the model predicts a lower speed adjustment factor. At the same time, under high load conditions, passengers take longer to go up and down, and the elevator door opening and closing frequency increases, resulting in a longer stop time at each floor, which directly increases the output value of the time adjustment factor. The current idle time is mainly used to reflect whether the elevator has been idle recently. If the elevator is idle for a long time, it means that its task pressure is low, and it can usually run at close to the design speed. The model tends to output a smaller speed correction value. At the same time, idle elevators often respond to new requests faster and stop more smoothly, so the corresponding time adjustment factor is lower. Conversely, if the idle time is very short, it means that the tasks are frequent and the state is busy, and the time adjustment factor increases.

[0058] Methods for calculating elevator waiting time based on pre-built waiting time models and adjustment factors include:

[0059]

[0060] In the formula, EWT′ i Let f be the elevator waiting time calculated after introducing an adjustment factor for elevator i, and let f be the floor number currently requested by the passenger. i Let v be the floor number where elevator i is currently located. i Let f be the standard design operating speed of elevator i, Δv be the speed adjustment factor, n be the number of floors currently stopped in the elevator's task queue, and f be the speed of elevator i. ij For the j-th target floor to be served in the elevator task queue, f i(j-1) Let t be the floor where the elevator is located during the (j-1)th task. s Δt is the standard average dwell time of the elevator for each normal stop at a floor, and Δt is the time adjustment factor.

[0061] In this embodiment, a deep neural network model is introduced to dynamically correct the elevator design speed and standard stopping time based on adjustment factors predicted from real-time collected operating status data (including current load rate, current idle time, current floor and direction of travel, etc.). This transforms the elevator waiting time from a traditional static estimation mechanism to a dynamic prediction mechanism. For example, when the elevator load rate is high, the model predicts a corresponding trend of decreasing speed and increasing stopping time, which is reflected in the waiting time model as a decrease in overall service capacity. When the elevator is idle for a long time, the model can output a small correction value to restore the waiting time to a level close to the design baseline. This significantly improves the accuracy and robustness of waiting time estimation, making subsequent task migration, coverage area division, and upgrade scheduling more consistent with the actual operating status of the system. It effectively avoids the problem of decreased system service capacity or local congestion caused by scheduling estimation errors, and realizes intelligent protection of service continuity and user response experience during the elevator system's program module upgrade process.

[0062] S20: Divide the elevator coverage area of ​​the target elevator group according to the elevator waiting time, evaluate the migration load value between every two elevators in the elevator coverage area based on the queuing theory model, and construct a dynamic upgrade path graph with elevators as nodes and migration load values ​​as edges.

[0063] It should be noted that the elevator coverage area refers to the set of floors that the elevator can serve or prioritize in the scheduling system. The elevator coverage area can be the set of all floors that the elevator is structurally allowed to stop at, or the set of target floors that the elevator is prioritized to be assigned a task in the scheduling strategy.

[0064] Methods for dividing the elevator coverage area of ​​a target elevator group based on elevator waiting time include:

[0065] The elevator waiting time is processed using the sliding difference method to obtain the rate of change of waiting time;

[0066] Calculate the average waiting time deviation between elevator waiting time and historical waiting time;

[0067] The preset coverage area is divided into Q floor sub-areas. The waiting time change rate and average waiting time deviation are mapped to a preset two-dimensional state diagram. The adjustment index is output through the two-dimensional state diagram.

[0068] When the adjustment index is greater than or equal to the preset index threshold, Q-K floor sub-areas are used as the elevator coverage area. When the adjustment index is less than the preset index threshold, Q floor sub-areas are used as the elevator coverage area. Here, Q is greater than K, and K is an integer greater than 0.

[0069] Understandably, the preset coverage area refers to the set of floors that can be served by each elevator in the elevator group control system, based on its physical structure, stopping capacity, scheduling strategy and service zoning rules. For example, the high-zone elevator serves floors 20-40 and the low-zone elevator serves floors 1-20. The floor sub-area refers to a number of consecutive floor segments obtained by dividing the preset coverage area of ​​the elevator according to a predetermined granularity. Each sub-area consists of a group of adjacent floors and serves as the smallest movable service unit. It can be divided according to a fixed number of floors, such as every 5 floors as a group.

[0070] Methods for obtaining the rate of change of elevator waiting time by performing a sliding difference calculation include:

[0071] The elevator waiting time is divided into a waiting time series by a preset time window. The difference between each two adjacent elements in the waiting time series is processed, and the rate of change per unit time between each two adjacent elements is calculated. The average of multiple rate of change per unit time is taken as the rate of change of waiting time.

[0072] In this embodiment, the preset time window refers to a continuous time period set by the system for statistical analysis when performing dynamic trend analysis on elevator waiting time. The elements in the waiting time sequence represent the average waiting time value of the elevator corresponding to a specific sampling time. This value can be obtained by statistically analyzing or averaging the queuing time of all tasks to be responded to at that time.

[0073] It should be noted that in this embodiment, the historical average waiting time is calculated based on the historical waiting time, and the difference between the elevator waiting time and the historical average waiting time is used as the average waiting time deviation.

[0074] Specifically, the two-dimensional state diagram refers to a two-dimensional feature space region division model constructed using "waiting time change rate" and "average waiting time deviation" as two orthogonal dimensions. It is used to reflect the joint evaluation results of the current service status of the elevator in both trend and stability directions. The horizontal axis of the two-dimensional state diagram is the "waiting time change rate", which is used to depict whether the elevator service trend is rising, falling or stable, and the vertical axis is the "average waiting time deviation", which is used to determine whether the current load of the elevator deviates from its historical operating baseline. The two-dimensional space is divided into multiple state segments (such as low-risk area, critical area, high-risk area, etc.), and each segment corresponds to a preset adjustment index range value.

[0075] This application guides the dynamic release of floor sub-regions by adjusting the output of an index. Specifically, the system first maps the current service status of the elevator to a two-dimensional state diagram based on the elevator's waiting time change rate and average waiting time deviation, and obtains an adjustment index accordingly. This adjustment index is not only used to determine whether to trigger the area release operation, but also linked with the real-time service status of each sub-region within the coverage area. When the release conditions are met, the system prioritizes the release of sub-regions with larger historical waiting time deviations, or releases sub-regions corresponding to high-risk state segments, thereby realizing the dynamic mapping between the adjustment index and the specific floor release strategy.

[0076] like Figure 2 As shown, the methods for constructing a two-dimensional state diagram include:

[0077] Using the rate of change of waiting time as the horizontal axis of the two-dimensional state diagram and the average waiting time deviation as the vertical axis, the segment division boundary in the two-dimensional state diagram is determined based on the clustering algorithm. The two-dimensional state diagram is then divided into P state segments according to the segment division boundary, where P is an integer greater than 1.

[0078] This application introduces two dynamic and trend-aware service status indicators: waiting time change rate and average waiting time deviation, to accurately characterize the elevator's operating trend and load stability during the current period. Based on this, a two-dimensional state diagram model based on a clustering algorithm is constructed, mapping the two indicators to the horizontal and vertical axes of the two-dimensional space, respectively. Through adaptive learning and cluster analysis, the two-dimensional space is divided into multiple state segments with scheduling significance (e.g., low load area, fluctuation area, high risk area, etc.), and a corresponding adjustment index output interval is assigned to each state segment.

[0079] The adjustment index, as a fusion and quantification result of elevator service status, is used to drive a flexible and dynamic division strategy for elevator coverage areas: when the adjustment index is higher than a set threshold, the system automatically triggers the release of some floor sub-areas; when the adjustment index is lower, the service coverage of the elevator is maintained, thereby achieving adaptive adjustment of the area based on service pressure status.

[0080] Through this mechanism, the system not only realizes the visual risk classification of elevator operation status, but also achieves multi-dimensional service risk identification and trend prediction control without relying on fixed rules. This greatly improves the scheduling flexibility, service continuity and overall system response efficiency of the group control system in the event of multiple elevator program upgrades, temporary shutdowns or high load emergencies.

[0081] Methods for evaluating the migration load value between every two elevators in an elevator coverage area based on queuing theory models include:

[0082] Based on the queuing theory model, the difference in queue length and the difference in processing quantity between every two elevators are extracted, and the migration load value is determined based on the difference in queue length and the difference in processing quantity.

[0083] Methods based on queuing theory models for extracting the queue length difference between every two elevators and processing the quantity difference include:

[0084] The operating status data and waiting time of each elevator are input into the queuing theory model to predict the future queue length and future processing quantity of each elevator. The difference between the future queue lengths of every two elevators is calculated to obtain the queue length difference, and the difference between the future processing quantities of every two elevators is calculated to obtain the processing quantity difference.

[0085] It should be noted that queuing theory is a mathematical modeling method based on operations research theory, used to analyze the operation process of a system with characteristics of task arrival, queuing, resource service, and completion departure. In an elevator scheduling system, a passenger's elevator call request can be regarded as a task arrival, and the elevator's response to the request and the transportation process can be regarded as a service process. Through queuing theory, key scheduling performance indicators such as service capacity, queue length, and waiting time of the elevator system in a future time period can be predicted. It is an important evaluation tool in this field for task allocation and scheduling optimization.

[0086] In this application, the future queue length represents the number of tasks accumulated by the elevator within a future preset time window. The number of tasks can correspond to the floor calls, passenger requests, or scheduling instructions that the elevator needs to respond to in the current or future period. The future processing quantity refers to the number of scheduling tasks that the elevator is expected to complete per unit time, which is a key indicator of the elevator's service capacity. The future processing quantity can dynamically reflect the elevator's task clearing speed and service response capability in the future period. The elevator's operating status data (such as the current task queue length, current load rate, and current idle time) and elevator waiting time together constitute the core input variables of the queuing theory model. The task queue length can be used to initialize the current queuing state, the current load rate and current idle time reflect the elevator's service capacity and task saturation, and the elevator waiting time reflects the system's response efficiency. Based on these input data, the queuing theory model can dynamically estimate the elevator's task arrival rate and service rate, and then predict the elevator's accumulated queue length and the number of tasks processed per unit time within a future preset time window, providing accurate support for cost assessment in the task migration path.

[0087] Methods for determining migration load values ​​based on queue length differences and processing quantity differences include:

[0088] The first weighting factor for the queue length difference is determined using the entropy weighting method, and the second weighting factor for the processing quantity difference is determined using the entropy weighting method. The queue length difference and the processing quantity difference are then weighted and summed based on the first and second weighting factors to obtain the migration load value.

[0089] It should be noted that, taking the queue length difference as an example, assuming the standardized distribution of the queue length difference data of the target elevator group is [0.4, 0.3, 0.2, 0.1], and the sample size r = 4, the entropy value is calculated as follows:

[0090]

[0091] w L =1 - 0.85 = 0.15;

[0092] Where E is the entropy value, w L It is the first weighting factor.

[0093] Similarly, the second weighting factor for the difference in the number of processes can be determined according to the entropy weight method, which will not be elaborated on in this embodiment.

[0094] Migration load is used to measure the service pressure required to migrate the task of an elevator to be upgraded to another elevator. It is an important quantitative indicator of the cost of task migration between elevators. This application predicts the future queue length and future processing quantity of each elevator through a queuing theory model, and calculates the queue length difference and processing quantity difference based on the difference between two elevators. These represent the imbalance between task supply and demand. The queue length difference reflects the acceptable space of the target elevator, while the processing quantity difference represents its future processing capacity per unit time. Together, they constitute the "resistance strength" of the elevator to task migration. The system then weights and fuses these two indicators according to the entropy weight method. The output migration load value can be used as the edge weight in the dynamic upgrade path graph, providing a reference for the path with the minimum service burden for subsequent upgrade sequence optimization.

[0095] In this embodiment, by introducing a queuing theory model and combining elevator operating status data with waiting time information, dynamic prediction of the future queue length and future processing capacity of elevators is achieved, replacing the traditional approach of task allocation and scheduling evaluation based on current static parameters.

[0096] Furthermore, to avoid the bias caused by the subjective setting of task migration cost weights in traditional scheduling methods, this application further introduces the entropy weight method, which adaptively calculates the weight coefficients based on the discreteness of the difference in queue length and the difference in the number of processes under different scheduling states and the information entropy, thereby constructing a more objective and reasonable "migration load value" index.

[0097] Unlike existing technologies that calculate migration costs based solely on fixed formulas or empirical rules, the method proposed in this application achieves dynamic modeling, precise quantification, and adaptive state adjustment for task migration cost assessment. This significantly enhances the ability to perceive changes in system load during task migration, effectively reduces the risk of service congestion caused by unreasonable scheduling, and ensures service continuity and response balance of the elevator group control system during program upgrades.

[0098] S30: Based on the graph search algorithm, the dynamic upgrade path graph is transformed into a sequence decision process. With the goal of minimizing the migration load value, the elevator upgrade sequence is determined, and the preset upgrade tasks are assigned to the corresponding elevators according to the elevator upgrade sequence.

[0099] In this embodiment, the graph search algorithm can be the A* algorithm or the Dijkstra algorithm. Graph search algorithms (such as the Dijkstra algorithm or the A* heuristic search algorithm) are a class of classic algorithms specifically used to find the shortest path or the path with the minimum total cost in graph structures. In this application, the graph search algorithm is used to search the dynamic upgrade path graph with the goal of "minimizing the migration load value", which is equivalent to finding an upgrade access sequence with the minimum cumulative task migration cost among multiple elevators.

[0100] Since the edge weights in the graph objectively reflect the intensity of the scheduling impact of the elevators, the graph search algorithm always prioritizes the path extension direction with the "lowest current cost" during node traversal. This allows it to output a globally optimal or near-optimal elevator upgrade sequence, ensuring that the overall service burden of the system is minimized during the upgrade process. This avoids scheduling imbalances or blind spot expansion caused by prioritizing the upgrade of high-load elevators. In this way, the system triggers the upgrade process of each elevator in sequence according to the order of the elevators in the sequence, and schedules the corresponding upgrade tasks (such as AI control module updates, function patch installations, etc.) to the elevator nodes in the current sequence for execution.

[0101] This embodiment obtains adjustment factors based on operational status data and calculates elevator waiting times using a pre-built waiting time model, enabling elevator service status to dynamically reflect operational differences. The elevator coverage area of ​​the target elevator group is then divided based on waiting times, and migration load values ​​are evaluated using a queuing theory model to construct a dynamic upgrade path graph. This enhances the ability to identify differences in task transfer capabilities between elevators. Finally, an elevator upgrade sequence is determined based on a graph search algorithm with the goal of minimizing migration load values. This achieves the allocation of upgrade tasks in the order that minimizes the overall system service impact when multiple elevator units need updating, thereby avoiding overload, task backlog, or service blind spots caused by improper scheduling.

[0102] Example 2

[0103] Please see Figure 3As shown, based on the same inventive concept, this embodiment discloses an AI-based elevator update control system. For details not covered in this embodiment, please refer to the relevant sections of Embodiment 1. The system includes:

[0104] Prediction module: used to acquire the operating status data of each elevator in the target elevator group, make predictions based on the operating status data to obtain adjustment factors, and calculate the elevator waiting time according to the pre-built waiting time model and adjustment factors. The adjustment factors include speed adjustment factors and time adjustment factors. The target elevator group refers to the elevator group that needs to be upgraded.

[0105] In this embodiment, the target elevator group refers to the elevator group that needs to be upgraded. The elevator group can be multiple elevator units with mutual coordinated scheduling relationships deployed in the same building or building group, or it can be a set of elevators marked as the same functional area or logical scheduling unit in the scheduling system, such as high-zone elevators, middle-zone elevators and low-zone elevators in a building divided according to the service floors.

[0106] Methods for predicting adjustment factors based on operational status data include:

[0107] The running status data is input into the pre-trained deep neural network model to obtain the adjustment factor of the deep neural network model output.

[0108] Training methods for deep neural network models include:

[0109] The input layer of a deep neural network model receives historical running state data, and the output layer of a deep neural network model outputs historical adjustment factors.

[0110] When training a deep neural network model, the cross-entropy loss function is chosen as the loss function, and the loss function is minimized by gradient descent.

[0111] Update the weight parameters of the deep neural network model and obtain the trained deep neural network model through iterative training.

[0112] Methods for calculating elevator waiting time based on pre-built waiting time models and adjustment factors include:

[0113]

[0114] In the formula, EWT′ i Let f be the elevator waiting time calculated after introducing an adjustment factor for elevator i, and let f be the floor number currently requested by the passenger. i Let v be the floor number where elevator i is currently located. i Let f be the standard design operating speed of elevator i, Δv be the speed adjustment factor, n be the number of floors currently stopped in the elevator's task queue, and f be the speed of elevator i.ij For the j-th target floor to be served in the elevator task queue, f i(j-1) Let t be the floor where the elevator is located during the (j-1)th task. s Δt is the standard average dwell time of the elevator for each normal stop at a floor, and Δt is the time adjustment factor.

[0115] The first processing module is used to divide the elevator coverage area of ​​the target elevator group according to the elevator waiting time, evaluate the migration load value between every two elevators in the elevator coverage area based on the queuing theory model, and construct a dynamic upgrade path graph with elevators as nodes and migration load values ​​as edges.

[0116] It should be noted that the elevator coverage area refers to the set of floors that the elevator can serve or prioritize in the scheduling system. The elevator coverage area can be the set of all floors that the elevator is structurally allowed to stop at, or the set of target floors that the elevator is prioritized to be assigned a task in the scheduling strategy.

[0117] Methods for dividing the elevator coverage area of ​​a target elevator group based on elevator waiting time include:

[0118] The elevator waiting time is processed using the sliding difference method to obtain the rate of change of waiting time;

[0119] Calculate the average waiting time deviation between elevator waiting time and historical waiting time;

[0120] The preset coverage area is divided into Q floor sub-areas. The waiting time change rate and average waiting time deviation are mapped to a preset two-dimensional state diagram. The adjustment index is output through the two-dimensional state diagram.

[0121] When the adjustment index is greater than or equal to the preset index threshold, Q-K floor sub-areas are used as the elevator coverage area. When the adjustment index is less than the preset index threshold, Q floor sub-areas are used as the elevator coverage area. Here, Q is greater than K, and K is an integer greater than 0.

[0122] Methods for obtaining the rate of change of elevator waiting time by performing a sliding difference calculation include:

[0123] The elevator waiting time is divided into a waiting time series by a preset time window. The difference between each two adjacent elements in the waiting time series is processed, and the rate of change per unit time between each two adjacent elements is calculated. The average of multiple rate of change per unit time is taken as the rate of change of waiting time.

[0124] Methods for constructing two-dimensional state diagrams include:

[0125] Using the rate of change of waiting time as the horizontal axis of the two-dimensional state diagram and the average waiting time deviation as the vertical axis, the segment division boundary in the two-dimensional state diagram is determined based on the clustering algorithm. The two-dimensional state diagram is then divided into P state segments according to the segment division boundary, where P is an integer greater than 1.

[0126] Methods for evaluating the migration load value between every two elevators in an elevator coverage area based on queuing theory models include:

[0127] Based on the queuing theory model, the difference in queue length and the difference in processing quantity between every two elevators are extracted, and the migration load value is determined based on the difference in queue length and the difference in processing quantity.

[0128] Methods based on queuing theory models for extracting the queue length difference between every two elevators and processing the quantity difference include:

[0129] The operating status data and waiting time of each elevator are input into the queuing theory model to predict the future queue length and future processing quantity of each elevator. The difference between the future queue lengths of every two elevators is calculated to obtain the queue length difference, and the difference between the future processing quantities of every two elevators is calculated to obtain the processing quantity difference.

[0130] Methods for determining migration load values ​​based on queue length differences and processing quantity differences include:

[0131] The first weighting factor for the queue length difference is determined using the entropy weighting method, and the second weighting factor for the processing quantity difference is determined using the entropy weighting method. The queue length difference and the processing quantity difference are then weighted and summed based on the first and second weighting factors to obtain the migration load value.

[0132] The second processing module is used to transform the dynamic upgrade path graph into a sequence decision process based on a graph search algorithm. With the goal of minimizing the migration load value, it determines the elevator upgrade sequence and assigns the preset upgrade tasks to the corresponding elevators according to the elevator upgrade sequence.

[0133] The detailed description above, in conjunction with the accompanying drawings, describes examples but does not represent all examples that can be implemented or fall within the scope of the claims. The terms “example” and “exemplary” are used in this specification to mean “serving as an example, instance or illustration” and do not mean “superior to or better than other examples”.

[0134] Throughout this specification, the phrase "an embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment of the invention. Therefore, the use of these phrases may refer to more than one embodiment. Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0135] It should also be noted that these embodiments may be described as processes depicted as flowcharts, structural diagrams, or block diagrams. Although a flowchart may describe the operations as sequential processes, many of these operations can be performed in parallel or concurrently, and the order of these operations may be rearranged.

Claims

1. An AI-based elevator update control method, characterized in that, include: The operation status data of each elevator in the target elevator group is obtained, and the adjustment factor is obtained by prediction based on the operation status data. The elevator waiting time is calculated according to the pre-constructed waiting time model and the adjustment factor. The adjustment factor includes speed adjustment factor and time adjustment factor. The target elevator group refers to the elevator group that needs to be upgraded. The elevator coverage area of ​​the target elevator group is divided according to the elevator waiting time. The migration load value between every two elevators in the elevator coverage area is evaluated based on the queuing theory model. With elevators as nodes and migration load values ​​as edges, a dynamic upgrade path graph is constructed. The graph search algorithm transforms the dynamic upgrade path graph into a sequential decision-making process. With the goal of minimizing the migration load value, the elevator upgrade sequence is determined, and the preset upgrade tasks are assigned to the corresponding elevators according to the elevator upgrade sequence.

2. The AI-based elevator update control method according to claim 1, characterized in that, The method for obtaining adjustment factors based on operational status data includes: The running status data is input into the pre-trained deep neural network model to obtain the adjustment factor output by the deep neural network model; Training methods for deep neural network models include: The input layer of a deep neural network model receives historical running state data, and the output layer of the deep neural network model outputs historical adjustment factors. When training a deep neural network model, the cross-entropy loss function is chosen as the loss function, and the loss function is minimized by gradient descent. Update the weight parameters of the deep neural network model and obtain the trained deep neural network model through iterative training.

3. The AI-based elevator update control method according to claim 2, characterized in that, The method for calculating elevator waiting time based on a pre-built waiting time model and adjustment factors includes: In the formula, EWT′ i Let f be the elevator waiting time calculated after introducing an adjustment factor for elevator i, and let f be the floor number currently requested by the passenger. i Let v be the floor number where elevator i is currently located. i Let f be the standard design operating speed of elevator i, Δv be the speed adjustment factor, n be the number of floors currently stopped in the elevator's task queue, and f be the speed of elevator i. ij For the j-th target floor to be served in the elevator task queue, f i(j-1) Let t be the floor where the elevator is located during the (j-1)th task. s Δt is the standard average dwell time of the elevator for each normal stop at a floor, and Δt is the time adjustment factor.

4. The AI-based elevator update control method according to claim 1, characterized in that, The method for dividing the elevator coverage area of ​​a target elevator group based on elevator waiting time includes: The elevator waiting time is processed using the sliding difference method to obtain the rate of change of waiting time; Calculate the average waiting time deviation between elevator waiting time and historical waiting time; The preset coverage area is divided into Q floor sub-areas. The waiting time change rate and average waiting time deviation are mapped to a preset two-dimensional state diagram. The adjustment index is output through the two-dimensional state diagram. When the adjustment index is greater than or equal to the preset index threshold, Q-K floor sub-areas are used as the elevator coverage area. When the adjustment index is less than the preset index threshold, Q floor sub-areas are used as the elevator coverage area. Here, Q is greater than K, and K is an integer greater than 0.

5. The AI-based elevator update control method according to claim 4, characterized in that, Methods for obtaining the rate of change of elevator waiting time by performing a sliding difference calculation include: The elevator waiting time is divided into a waiting time sequence by using a preset time window; Perform difference processing on every two adjacent elements in the waiting time series, and calculate the rate of change per unit time between every two adjacent elements; The average of multiple unit time change rates is taken as the waiting time change rate.

6. The AI-based elevator update control method according to claim 4, characterized in that, Methods for constructing two-dimensional state diagrams include: The rate of change of waiting time is used as the horizontal axis of the two-dimensional state diagram, and the average waiting time deviation is used as the vertical axis of the two-dimensional state diagram. Based on the clustering algorithm, the segment division boundary in the two-dimensional state diagram is determined, and the two-dimensional state diagram is divided into P state segments according to the segment division boundary, where P is an integer greater than 1.

7. The AI-based elevator update control method according to claim 1, characterized in that, The method for evaluating the migration load value between every two elevators in an elevator coverage area based on a queuing theory model includes: Based on the queuing theory model, the difference in queue length and the difference in processing quantity between every two elevators are extracted, and the migration load value is determined based on the difference in queue length and the difference in processing quantity.

8. The AI-based elevator update control method according to claim 7, characterized in that, The method for extracting the queue length difference between every two elevators and processing the quantity difference based on the queuing theory model includes: The operating status data and waiting time of each elevator are input into the queuing theory model to predict the future queue length and the number of elevators to be processed in the future. The queue length difference is calculated based on the difference between the future queue lengths of every two elevators, and the processing quantity difference is calculated based on the difference between the future processing quantities of every two elevators.

9. The AI-based elevator update control method according to claim 8, characterized in that, The method for determining the migration load value based on the queue length difference and the processing quantity difference includes: The first weighting factor for the difference in queue length is determined using the entropy weighting method, and the second weighting factor for the difference in the number of processes is determined using the entropy weighting method. The migration load value is obtained by weighting and summing the difference in queue length and the difference in the number of processes based on the first and second weighting factors.

10. An AI-based elevator update control system, used to implement the AI-based elevator update control method according to any one of claims 1-9, characterized in that, include: Prediction module: used to acquire the operating status data of each elevator in the target elevator group, make predictions based on the operating status data to obtain adjustment factors, and calculate the elevator waiting time according to the pre-built waiting time model and adjustment factors. The adjustment factors include speed adjustment factors and time adjustment factors. The target elevator group refers to the elevator group that needs to be upgraded. The first processing module is used to divide the elevator coverage area of ​​the target elevator group according to the elevator waiting time, evaluate the migration load value between every two elevators in the elevator coverage area based on the queuing theory model, and construct a dynamic upgrade path graph with elevators as nodes and migration load values ​​as edges. The second processing module is used to transform the dynamic upgrade path graph into a sequence decision process based on a graph search algorithm. With the goal of minimizing the migration load value, it determines the elevator upgrade sequence and assigns the preset upgrade tasks to the corresponding elevators according to the elevator upgrade sequence.

Images