An elevator group control dynamic scheduling method and system based on reinforcement learning, an electronic device, and a storage medium

Abstract

The application aims to provide an elevator group control dynamic scheduling method and system based on reinforcement learning, an electronic device and a storage medium. The method comprises the following steps: S100: acquiring elevator demand data; S200: acquiring a time-space feature of a boarding behavior according to the elevator demand data; and S300: generating an elevator resource dynamic scheduling strategy according to the time-space feature of the boarding behavior. Through the intelligent closed loop of "data collection, time-space modeling, dynamic strategy generation and continuous evolution", the application realizes the autonomous evolution of the dispatching logic according to the building passenger flow law.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method, system, electronic device, and storage medium for dynamic scheduling of elevator groups based on reinforcement learning. Background Technology

[0002] Elevator resource scheduling optimizes the response strategies of multiple elevators through real-time algorithms to minimize passenger waiting time and energy consumption. Its core steps include: Demand perception: capturing passenger requests via floor buttons (up / down) and floor selection signals within the car; Dynamic allocation: assigning new requests to the optimal elevator based on the elevator's current location, direction of travel, load status, and predicted traffic flow (e.g., during peak hours), using scheduling algorithms such as SCAN, LOOK, or reinforcement learning-based strategies; Path planning: adjusting elevator stop sequences, merging requests in the same direction, and avoiding invalid stops, such as through a "carry-on" strategy; Real-time adjustment: recalculating priorities for sudden demands, such as emergency calls or changes in traffic flow, to ensure overall efficiency. This process relies on multi-objective optimization (speed, energy consumption, fairness) and achieves intelligent scheduling through centralized controllers or distributed collaboration.

[0003] Elevator group control systems are the core transportation hubs of modern high-rise buildings, and their scheduling efficiency directly affects passenger experience and building energy consumption. Existing elevator group control technologies mainly rely on preset rules, such as prioritizing the shortest waiting time or simple time-segmentation strategies, such as peak / off-peak modes. These have the following inherent drawbacks: Static rules lack spatiotemporal adaptability: Fixed algorithms cannot perceive the dynamic spatiotemporal patterns of passenger flow within the building, leading to a mismatch between scheduling strategies and actual demand; High cost of manual parameter tuning: The system requires engineers to manually adjust parameters based on experience, resulting in not only delayed response but also difficulty in covering complex building scenarios; Disconnect between prediction and response capabilities: Traditional methods only respond to real-time requests and cannot predict future passenger flow trends based on historical data, missing pre-scheduling optimization windows; Imbalance in multi-objective optimization: Rule engines struggle to coordinate and optimize conflicting objectives such as waiting time, energy consumption, and elevator balance losses, easily falling into local optima, such as shortening waiting time while increasing energy consumption by 30%. Especially with the development of smart buildings, the diversification of building functions leads to increasingly complex passenger flow patterns, such as cross-floor connections and peak fluctuations during holidays. Existing technologies can no longer meet the demands for efficient, flexible, and low-consumption scheduling. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, electronic device and storage medium for dynamic scheduling of elevator groups based on reinforcement learning. This method realizes the autonomous evolution of elevator dispatch logic according to the building passenger flow pattern by constructing an intelligent closed loop of "data acquisition → spatiotemporal modeling → dynamic strategy generation → continuous evolution".

[0005] A dynamic scheduling method for elevator group control based on reinforcement learning includes:

[0006] S100: Obtain real-time elevator demand data;

[0007] S200: Obtain the spatiotemporal characteristics of elevator riding behavior based on the real-time elevator demand data;

[0008] S300: Generate a dynamic elevator resource scheduling strategy based on the spatiotemporal characteristics of the elevator riding behavior, including:

[0009] S310A: When elevator demand exhibits cyclical characteristics, elevator resource planning is performed using a pre-deployed scheduling strategy;

[0010] S310B: When elevator demand does not conform to periodic characteristics, the elevator resource scheduling intelligent agent calculates dynamic scheduling strategies in real time to plan elevator resources.

[0011] Preferably, before acquiring real-time elevator demand data, the method further includes:

[0012] S010: Obtain prior elevator data;

[0013] S020: Calculate the deployed elevator resource scheduling strategy based on prior elevator riding data.

[0014] Preferably, obtaining real-time elevator demand data includes:

[0015] S110: Obtain the timestamp, source floor, and target floor data of the real-time elevator request;

[0016] S120: The acquired real-time data is divided into temporal data and spatial data.

[0017] Preferably, obtaining the spatiotemporal characteristics of elevator riding behavior based on the real-time elevator demand data includes:

[0018] S210: Obtain the passenger flow direction matrix between floors within different time periods based on time and spatial data;

[0019] S220: Identify passenger flow paths during peak and off-peak hours based on the passenger flow direction matrix for different time periods;

[0020] S230: Extract the spatiotemporal features of elevator riding behavior based on the passenger flow paths during peak and off-peak hours.

[0021] Preferably, obtaining the passenger flow direction matrix between floors within different time periods based on time and spatial data includes:

[0022] Define time slices to divide a day into T time slots;

[0023] For time period t, generate an N×N matrix M. tN represents the floor number, rows represent the starting floor, and columns represent the destination floor.

[0024]

[0025] Preferably, the elevator resource scheduling strategy calculated and deployed based on prior elevator riding data includes:

[0026] S021: Use the prior elevator data as a training set to input into the reinforcement learning-based agent model;

[0027] S022: Train and optimize the reinforcement learning-based agent model;

[0028] S023: The trained reinforcement learning-based agent model outputs the deployed elevator resource scheduling strategy.

[0029] Preferably, after generating the elevator resource dynamic scheduling strategy based on the spatiotemporal characteristics of the elevator riding behavior, the method further includes iteratively optimizing the elevator resource dynamic scheduling strategy through a reward function feedback mechanism:

[0030] Set the reward function as follows:

[0031] R=ω1R wait +ω2R energy +ω3R balance +ω4R emergency

[0032] Where ω1, ω2, ω3, and ω4 are adjustable weights, and R wait As a waiting time penalty, R energy As a penalty for energy consumption, R balance As a load balancing reward, R emergency Rewards are given for urgent needs.

[0033] A dynamic scheduling system for elevator group control based on reinforcement learning includes:

[0034] The data acquisition module is used to acquire real-time elevator demand data;

[0035] The feature extraction module is used to obtain the spatiotemporal features of elevator riding behavior based on the real-time elevator demand data;

[0036] The strategy deployment module is used to generate a dynamic elevator resource scheduling strategy based on the spatiotemporal characteristics of the elevator riding behavior, including:

[0037] When elevator demand exhibits cyclical characteristics, elevator resource planning is performed using pre-deployed scheduling strategies.

[0038] When elevator demand does not conform to periodic characteristics, the elevator resource scheduling intelligent agent calculates dynamic scheduling strategies in real time to plan elevator resources.

[0039] An electronic device includes a chip, a processor, and a memory, the memory storing computer program code including computer instructions. When the chip executes the computer instructions, the electronic device executes a reinforcement learning-based dynamic scheduling method for elevator group control.

[0040] A computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor of an electronic device, cause the processor to perform a reinforcement learning-based dynamic scheduling method for elevator group control.

[0041] The beneficial effects of this invention are as follows: The scheduling agent of this invention autonomously learns the spatiotemporal patterns of elevator passenger data, such as morning peak upbound traffic and midday cross-floor dispersed travel, automatically generating and dynamically updating scheduling strategies; it eliminates the need for manually defining complex rules, completely solving the problem of reliance on human experience and significantly reducing system operation and maintenance difficulty and costs; the scheduling agent of this invention analyzes high-dimensional data such as timestamps and source / target floors to identify subtle spatiotemporal patterns, generating spatiotemporally variable strategies, accurately modeling and dynamically responding to high-dimensional spatiotemporal passenger flow patterns within buildings, significantly improving scheduling efficiency; the system of this invention, through real-time data closed-loop (collection → training → deployment), continuously iterates and optimizes scheduling strategies, and the agent automatically detects changes in passenger flow patterns, actively triggering model retraining to ensure the system maintains optimal performance in the long term, extending the technology lifecycle; the multi-objective collaborative optimization mechanism of this invention reduces the overall system operating cost while ensuring a good elevator experience, and reasonable scheduling arrangements can reduce elevator idle operation and energy consumption. Attached Figure Description

[0042] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart of a dynamic scheduling method for elevator group control based on reinforcement learning according to the present invention.

[0045] Figure 2 This is a diagram illustrating the architecture of a dynamic scheduling system for elevator group control based on reinforcement learning, as described in this invention.

[0046] Figure 3This is a schematic diagram of the hardware structure of an electronic device according to the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0048] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0049] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" and "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0050] Existing elevator group control technologies mainly rely on preset rules, such as prioritizing the shortest waiting time or simple time-segmentation strategies, such as peak / off-peak modes. These technologies have the following inherent drawbacks: Static rules lack spatiotemporal adaptability: Fixed algorithms cannot perceive the dynamic spatiotemporal patterns of passenger flow within a building, leading to a mismatch between scheduling strategies and actual demand; High cost of manual parameter tuning: The system requires engineers to manually adjust parameters based on experience, resulting in not only delayed response but also difficulty in covering complex building scenarios; Disconnect between prediction and response capabilities: Traditional methods only respond to real-time requests and cannot predict future passenger flow trends based on historical data, missing pre-scheduling optimization windows; Imbalance in multi-objective optimization: Rule engines struggle to coordinate and optimize conflicting objectives such as waiting time, energy consumption, and elevator balance losses, easily falling into local optima, such as shortening waiting time while increasing energy consumption by 30%. Especially with the development of smart buildings, the diversification of building functions leads to increasingly complex passenger flow patterns, such as cross-floor connections and peak fluctuations during holidays. Existing technologies can no longer meet the demands for efficient, flexible, and low-consumption scheduling.

[0051] To address the aforementioned issues, this invention proposes an adaptive scheduling system for elevator group control based on agent learning. The system executes a dynamic scheduling method for elevator group control based on reinforcement learning. The adaptive scheduling method for elevator group control proposed in this invention specifies the deployment scheme of elevator resources by analyzing the spatiotemporal characteristics in elevator passenger data, thereby greatly improving the utilization rate and efficiency of elevator resources.

[0052] Please see Figure 1 and Figure 2 The present invention proposes an elevator group control adaptive scheduling system based on agent learning, comprising: a data acquisition module, a feature extraction module, and a strategy deployment module, used to convert elevator ride data into spatiotemporal features and construct an elevator resource scheduling agent to generate a dynamic elevator resource scheduling strategy based on the spatiotemporal features of the ride behavior, and to customize the dynamic elevator resource scheduling strategy according to the different spatiotemporal features of each building.

[0053] In practice, before obtaining elevator demand data, the following steps are also included:

[0054] S010: Obtain prior elevator data.

[0055] S020: Calculate the deployed elevator resource scheduling strategy based on prior elevator riding data.

[0056] S021: Input the elevator resource scheduling agent with prior elevator riding data as a training set;

[0057] Training on prior data: The agent is pre-trained using historical elevator usage data to accelerate convergence. Online iterative optimization: The strategy is continuously updated with new data to achieve adaptive scheduling.

[0058] S022: Train and optimize the reinforcement learning-based agent model.

[0059] The parameters of the elevator resource scheduling agent are iteratively optimized through a reward function feedback mechanism, including:

[0060] Set the reward function as follows:

[0061] R=ω1R wait +ω2R energy +ω3R balance +ω4R emergency

[0062] Where ω1, ω2, ω3, and ω4 are adjustable weights, and R wait As a waiting time penalty, R energy As a penalty for energy consumption, R balance As a load balancing reward, R emergency Rewards are given for urgent needs.

[0063] The waiting time penalty is a linear penalty set within a certain waiting time. After the timeout, the penalty increases sharply to avoid extremely long waiting times.

[0064]

[0065] Where N: number of requests awaiting response in the current time period, Tk: cumulative waiting time of the k-th request, α: basic waiting penalty coefficient, β: long waiting penalty intensity, and τ: tolerance threshold.

[0066] The energy consumption penalty is imposed on both the operating distance (positively correlated with energy consumption) and frequent start-stop (mechanical wear).

[0067]

[0068] Where E: total number of elevators, De: running distance of elevator e, We: load weight of elevator e, Se: number of start-stop cycles of elevator e, γ and C: energy consumption coefficients.

[0069] Load balancing rewards are given, with higher rewards for more balanced loads, to prevent some elevators from being overloaded or idle.

[0070]

[0071] Where, Le: the current load rate of elevator e (0~1), σ: the standard deviation of the load rate, μ: the mean of the load rate, ε: the smoothing factor, and δ: the reward intensity.

[0072] Rewards for urgent needs, such as peak hours and emergency room visits.

[0073]

[0074] Where Kemerg: the set of urgent / priority needs, τ emerg Maximum tolerance time for emergency needs, [X] + =max(x,0): ensures non-negativity, η: priority coefficient.

[0075] In this embodiment of the invention, the reinforcement learning-based agent model can be trained and optimized multiple times during use. Before being deployed to the elevator group control system, the reinforcement learning-based agent model has been trained according to the current building environment and has a certain elevator resource scheduling capability. During subsequent use, the reinforcement learning-based agent model can be trained by continuously acquiring real-time elevator data to better adapt to changes in the usage environment. The training time can be selected during off-peak hours. For office buildings, training can be conducted on weekends, while for shopping malls, training can be conducted on weekdays or closed days. The reinforcement learning-based agent model can continuously update the deployed elevator resource dynamic scheduling strategy to adapt to the ever-changing usage environment.

[0076] S023: The trained reinforcement learning-based agent model outputs the deployed elevator resource scheduling strategy.

[0077] In practice, an elevator resource scheduling agent can be used as a trained reinforcement learning-based agent model. Based on the prior elevator demand of the current building, an elevator resource scheduling strategy that can meet most elevator demand can be formulated. With a well-deployed elevator resource scheduling strategy, the system can respond very quickly in most cases, thus improving elevator scheduling efficiency.

[0078] In specific implementation, scenario 1: a high-end office building, the characteristics of which are as follows: the upward peak is from 8:00 to 10:00 on weekday mornings, the scattered cross-floor flow is from 12:00 to 14:00 at noon, and the downward peak is from 18:00 to 19:00 in the afternoon.

[0079] Deploy an intelligent elevator resource scheduling system in the office building's group control system.

[0080] Data was continuously collected and recorded for 4 weeks, including:

[0081] Morning peak: Uplink requests concentrated on floors 1-5 and 20-30 (accounting for 70%).

[0082] At lunchtime: Random requests between different floors (restaurant / meeting room).

[0083] Based on the collected and processed elevator demand data, spatiotemporal feature data is extracted and a "time period-floor traffic heat map" is generated, and the "midday cross-floor request dispersion" index is extracted.

[0084] The DDQN algorithm is used, with average waiting time and energy saving rate as performance metrics for the reward function. After training, an elevator resource deployment scheme is obtained.

[0085] Morning rush hour: Assign empty elevators to be stationed in advance on lower floors, bypassing middle floors to reach the target high area directly.

[0086] At midday: The "regional circulation" mode is activated, with a single elevator serving five adjacent floors.

[0087] The elevator group control system uses the dynamic scheduling strategy generated by the aforementioned scheduling agent to control the elevator main control system to execute elevator scheduling. For example, at 8:30 a.m. on Monday, a new request "1→25F" triggers the strategy, and the system assigns elevator A, which is waiting on the 1st floor, to go directly there (the traditional system would have to wait for elevator B to descend from the 15th floor).

[0088] Automatic retraining every weekend.

[0089] Scene 2:

[0090] A general hospital has the following characteristics: random peak hours throughout the day on the outpatient floors (2-4), and regular meal / examination needs on the inpatient floors (8-12), as well as sudden emergency elevator calls.

[0091] The elevator resource scheduling intelligent agent is integrated into the hospital's elevator group control system.

[0092] Continuously collect and record emergency elevator call data (sudden and high priority), and statistically analyze meal delivery times (peak hours of 11:00 / 17:00 inpatient department).

[0093] Based on the collected and processed elevator demand data, spatiotemporal feature data is extracted, the correlation between "emergency response delay" and elevator location is identified, and a "demand prediction model for inpatient wards during meal delivery hours" is constructed.

[0094] Using PPO reinforcement learning, with emergency response time and general passenger satisfaction as performance metrics for reward functions, an elevator resource deployment scheme was obtained after training.

[0095] Standard practice: One elevator is reserved and permanently stationed on the first floor to cover the emergency room.

[0096] 10 minutes before meal delivery: Arrange the elevator to be on standby in the upper floors of the inpatient department.

[0097] Scene 3:

[0098] A shopping mall exhibits the following characteristics: a surge in family foot traffic on weekends, concentrated demand for food and beverage / cinema services on upper floors, and sparse foot traffic on weekday afternoons.

[0099] Deploy an intelligent elevator dispatching agent in the shopping mall's elevator group control system.

[0100] Comparing weekday / weekend data: During the weekend cinema closing time (21:00-22:00), the number of requests on floors 7-9 → 1 surged by 300%.

[0101] Based on the collected and processed elevator demand data, spatiotemporal feature data is extracted, the "downward request clustering index" is calculated, and the "cross-floor connecting demand" is extracted.

[0102] Using the Actor-Critic framework, with long waiting rate and multiplication matching degree as performance metrics for the reward function, an elevator resource deployment scheme is obtained after training:

[0103] During the weekend downhill peak: "funnel scheduling" is activated - elevators on high floors only go down and not up, while elevators in the middle zone take over to transfer passengers.

[0104] Detect connecting passenger requests: If a passenger presses "4F→7F", assign a service to the same elevator to reduce waiting time for transfers.

[0105] After building and training the elevator resource scheduling intelligent agent, the trained elevator resource scheduling intelligent agent is deployed in the elevator control system of the application scenario. The trained elevator resource scheduling intelligent agent outputs elevator resource scheduling strategy based on the received elevator demand data.

[0106] In practice, the data acquisition module is used to execute step S100 to acquire real-time elevator demand data.

[0107] The data acquisition module includes a data extraction unit and a data classification unit.

[0108] The data extraction unit is used to perform step S110 to obtain the timestamp of the real-time elevator request, the source floor of the elevator request, and the target floor of the elevator request.

[0109] In practice, the timestamp of the elevator request, the source floor of the elevator request, and the target floor of the elevator request are obtained through cameras, elevator buttons, etc. The elevator demand data that can be obtained may also include one or more of the following: call direction, call type (car call or floor call), passenger number information, and current status information of the elevator car (including location, direction of travel, load, and door status).

[0110] The data classification unit is used to perform step S120, which divides the acquired real-time data into temporal data and spatial data.

[0111] The time data includes timestamps of elevator ride requests, and the spatial data includes the source floor and target floor of the elevator ride request.

[0112] The feature extraction module is used to perform step S200, which obtains the spatiotemporal features of elevator riding behavior based on the real-time elevator demand data.

[0113] The data processing module includes a flow direction matrix construction unit, a passenger flow path construction unit, and a spatiotemporal feature extraction unit.

[0114] The matrix construction unit is used to execute step S210 to obtain the passenger flow direction matrix between floors in different time periods based on time data and spatial data.

[0115] In practice, an elevator dispatching intelligent agent is used to obtain the passenger flow direction matrix between floors within different time periods based on the elevator demand data.

[0116] The specific implementation steps include: defining time slices: dividing a day into T time periods, such as each time period being 15 minutes.

[0117] Constructing the passenger flow matrix: For time period t, generate an N×N matrix M. t (N represents the floor number, rows represent the starting floor, and columns represent the destination floor):

[0118]

[0119] After constructing the passenger flow matrix, normalization can be performed.

[0120] Identify core routes through passenger flow matrices; during peak hours, locate high-frequency floor pairs; during off-peak hours, filter low-frequency requests; predict passenger flow patterns and extract periodic features by comparing passenger flow matrices for the same time period on different dates, such as weekday morning peak hours; detect anomalies by triggering real-time scheduling when the deviation between the current matrix and the historical average matrix exceeds a threshold; optimize resource allocation and assign elevators to high-frequency routes, such as from floor 1 to floor 20, pre-allocate dedicated elevators; and perform dynamic route planning by calculating elevator stopping priority based on matrix weights, prioritizing high-weight routes.

[0121] The passenger flow path construction unit is used to perform step S220, which identifies the passenger flow paths during peak and off-peak periods based on the passenger flow direction matrix in different time periods.

[0122] In practice, an elevator dispatching intelligent agent is used to identify passenger flow paths during peak and off-peak hours based on the passenger flow direction matrix for different time periods.

[0123] The spatiotemporal feature extraction unit is used to perform step S230, extracting spatiotemporal features of elevator riding behavior based on the passenger flow paths during peak and off-peak periods.

[0124] In practice, an elevator dispatching intelligent agent is used to extract the spatiotemporal features of elevator riding behavior based on the passenger flow paths during peak and off-peak periods.

[0125] The strategy deployment module is used to execute step S300, which generates a dynamic scheduling strategy for elevator resources based on the spatiotemporal characteristics of the elevator riding behavior.

[0126] In practice, an elevator scheduling intelligent agent is used to generate a dynamic scheduling strategy for elevator resources based on the spatiotemporal characteristics of the elevator riding behavior.

[0127] The strategy deployment module includes a decision-making unit, which executes step S310A. When the elevator demand meets the periodic characteristics, the deployed scheduling strategy is used for elevator resource planning.

[0128] In practice, if it is determined that the current elevator demand conforms to the prior periodic characteristics, the elevator resource scheduling scheme described above can be directly used to schedule elevator resources for each application scenario.

[0129] S310B: When elevator demand does not conform to periodic characteristics, the elevator resource scheduling agent calculates dynamic scheduling strategies in real time to plan elevator resources.

[0130] In practice, if it is determined that the current elevator demand does not conform to the prior periodic characteristics, the elevator resource scheduling intelligent agent calculates the dynamic scheduling strategy in real time to plan elevator resources. Different weight coefficients of the reward function can be adjusted according to different scenario needs. For example, in the emergency scenario, the emergency demand coefficient can be adjusted to 0.8.

[0131] Scene 1:

[0132] The office building group control system uses an elevator dispatching intelligent agent. If there is an emergency request for a meeting or event, the elevator dispatching intelligent agent can extract the spatiotemporal characteristics of the emergency based on the meeting or event information entered by the property management, and then dispatch the current elevator resources in an emergency.

[0133] For example, if a company needs to hold an event, the office building's group control system will proactively arrange idle elevators to the first floor to receive guests when the event is about to start, and will proactively arrange idle elevators to the event floor when the event ends.

[0134] Scene 2:

[0135] The elevator group control system uses the elevator scheduling intelligent agent to deploy a dynamic scheduling strategy for elevator resources and control the elevator main control system to execute elevator scheduling. For example, if an emergency request for "3F→1F" suddenly occurs at 15:08 in the afternoon, the elevator scheduling intelligent agent will immediately assign a reserved standby elevator to respond. The traditional system would need to recall the elevator from the 12th floor, which would delay the elevator by 22 seconds.

[0136] In a hospital setting, request priorities can be set directly. When an emergency request is triggered, the elevator group control system will actively ignore other elevator requests and go directly to the requested floor. After receiving the emergency patient, the system will take them directly to the target floor without stopping along the way.

[0137] When the new inpatient building is put into use, causing a change in traffic patterns, the agent detects the data shift and automatically triggers incremental training to adapt to the new passenger flow distribution within 2 weeks.

[0138] Scene 3:

[0139] The elevator group control system uses the dynamic scheduling strategy generated by the aforementioned scheduling agent to control the elevator main control system to perform elevator scheduling: for example, when adding extra screenings at cinemas on weekends, the elevator scheduling agent can perform emergency scheduling of elevator resources in real time by obtaining the opening and closing information released by the cinemas.

[0140] As the movie was about to start, the elevator group control system proactively dispatched the idle elevators to the first floor for standby.

[0141] When the movie ends, the elevator group control system automatically dispatches the idle elevators from the 1st floor to the 8th floor for standby, whereas the traditional system requires passengers to call for an elevator before it can respond.

[0142] The model was pre-trained using historical data from holidays.

[0143] Based on the same inventive concept, this application also provides an electronic device, including: a chip, a processor, and a memory, wherein the memory is used to store computer program code, the computer program code including computer instructions, and when the chip executes the computer instructions, the electronic device executes a reinforcement learning-based elevator group control dynamic scheduling method.

[0144] refer to Figure 3 The electronic device 2 includes a processor 21, a memory 22, an input device 23, and an output device 24. The processor 21, memory 22, input device 23, and output device 24 are coupled together via connectors, which may include various interfaces, transmission lines, or buses, etc., and are not limited in this embodiment of the invention. It should be understood that in the various embodiments of the invention, coupling refers to mutual connection through a specific method, including direct connection or indirect connection through other devices, such as through various interfaces, transmission lines, buses, etc.

[0145] Processor 21 can be one or more graphics processing units (GPUs). If processor 21 is a GPU, the GPU can be a single-core GPU or a multi-core GPU. Optionally, processor 21 can be a processor group composed of multiple GPUs, with the multiple processors coupled to each other via one or more buses. Optionally, the processor can also be other types of processors, etc., and this embodiment of the invention is not limited thereto.

[0146] The memory 22 can be used to store computer program instructions, as well as various types of computer program code, including program code for executing the present invention. Optionally, the memory includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or compact disc read-only memory (CD-ROM), which is used for related instructions and data.

[0147] Input device 23 is used to input data and / or signals, and output device 24 is used to output data and / or signals. Output device 24 and input device 23 can be independent devices or an integrated device.

[0148] Based on the same inventive concept, this application also provides a computer-readable storage medium corresponding to the aforementioned embodiment of a reinforcement learning-based elevator group control dynamic scheduling method. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the reinforcement learning-based elevator group control dynamic scheduling method described in any of the above embodiments.

[0149] This application may take the form of a computer program product implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing program code. Computer storage media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information may be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to: phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0150] The scheduling agent of this invention learns the spatiotemporal patterns of elevator traffic data autonomously, such as morning peak upbound traffic and midday cross-floor dispersed travel, and automatically generates and dynamically updates scheduling strategies. This eliminates the need for manually defined complex rules, completely resolving reliance on human experience and significantly reducing system operation and maintenance difficulty and costs. By analyzing high-dimensional data such as timestamps and source / target floors, the scheduling agent identifies subtle spatiotemporal patterns and generates spatiotemporally variable strategies. This enables accurate modeling and dynamic response to high-dimensional spatiotemporal passenger flow patterns within buildings, significantly improving scheduling efficiency. Through real-time data closed-loop (collection → training → deployment), the system continuously iterates and optimizes scheduling strategies. The agent automatically detects changes in passenger flow patterns and proactively triggers model retraining, ensuring the system maintains optimal performance over the long term and extending the technology's lifespan. The multi-objective collaborative optimization mechanism of this invention reduces overall system operating costs while ensuring a good elevator experience. Reasonable scheduling arrangements can reduce elevator idle operation and energy consumption.

[0151] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A dynamic scheduling method for elevator group control based on reinforcement learning, characterized in that, include: S100: Acquire real-time elevator demand data and divide the elevator demand data into time data and spatial data; S200: Obtain the spatiotemporal characteristics of elevator riding behavior based on the real-time elevator demand data, including: S210: Obtain the passenger flow direction matrix between floors within different time periods based on time and spatial data; S220: Identify passenger flow paths during peak and off-peak hours based on the passenger flow direction matrix for different time periods; S230: Extract the spatiotemporal features of elevator riding behavior based on the passenger flow paths during peak and off-peak periods; S300: Generate a dynamic elevator resource scheduling strategy based on the spatiotemporal characteristics of the elevator riding behavior, including: S310A: When elevator demand exhibits cyclical characteristics, elevator resource planning is performed using a pre-deployed scheduling strategy; S310B: When elevator demand does not conform to periodic characteristics, the elevator resource scheduling intelligent agent calculates dynamic scheduling strategies in real time to plan elevator resources.

2. The elevator group control dynamic scheduling method based on reinforcement learning according to claim 1, characterized in that, Before obtaining real-time elevator demand data, the process also includes: S010: Obtain prior elevator data; S020: Calculate the deployed elevator resource scheduling strategy based on prior elevator riding data.

3. The elevator group control dynamic scheduling method based on reinforcement learning according to claim 1, characterized in that, The acquisition of real-time elevator demand data includes: S110: Obtain the timestamp, source floor, and target floor data of the real-time elevator request; S120: The acquired real-time data is divided into temporal data and spatial data.

4. The elevator group control dynamic scheduling method based on reinforcement learning according to claim 1, characterized in that, The process of obtaining the passenger flow direction matrix between floors within different time periods based on time and spatial data includes: Define time slices to divide a day into T time slots; For time period t, generate an N×N matrix. N represents the floor number, rows represent the starting floor, and columns represent the destination floor. 。 5. The elevator group control dynamic scheduling method based on reinforcement learning according to claim 2, characterized in that, The elevator resource scheduling strategy calculated and deployed based on prior elevator riding data includes: S021: Use the prior elevator data as a training set to input into the reinforcement learning-based agent model; S022: Train and optimize the reinforcement learning-based agent model; S023: The trained reinforcement learning-based agent model outputs the deployed elevator resource scheduling strategy.

6. The elevator group control dynamic scheduling method based on reinforcement learning according to claim 1, characterized in that, After generating the elevator resource dynamic scheduling strategy based on the spatiotemporal characteristics of the elevator riding behavior, the method further includes iteratively optimizing the elevator resource dynamic scheduling strategy through a reward function feedback mechanism. Set the reward function as follows: in, , , and For adjustable weights, As a penalty for waiting time, As a penalty for energy consumption, As a load balancing reward, Rewards are given for urgent needs.

7. A dynamic scheduling system for elevator group control based on reinforcement learning, characterized in that, include: The data acquisition module is used to acquire real-time elevator demand data, which is divided into time data and spatial data. The feature extraction module is used to obtain spatiotemporal features of elevator riding behavior based on the real-time elevator demand data, including: Based on time and spatial data, obtain a passenger flow direction matrix between floors within different time periods; Identify passenger flow paths during peak and off-peak hours based on the passenger flow direction matrix for different time periods; Based on the passenger flow paths during peak and off-peak hours, the spatiotemporal features of elevator riding behavior are extracted; The strategy deployment module is used to generate a dynamic elevator resource scheduling strategy based on the spatiotemporal characteristics of the elevator riding behavior, including: When elevator demand exhibits cyclical characteristics, elevator resource planning is performed using pre-deployed scheduling strategies. When elevator demand does not conform to periodic characteristics, the elevator resource scheduling intelligent agent calculates dynamic scheduling strategies in real time to plan elevator resources.

8. An electronic device, characterized in that, include: The electronic device comprises a chip, a processor, and a memory, the memory being used to store computer program code, the computer program code including computer instructions, wherein, when the chip executes the computer instructions, the electronic device performs a dynamic scheduling method for elevator group control based on reinforcement learning as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which includes program instructions that, when executed by a processor of an electronic device, cause the processor to perform a dynamic scheduling method for elevator group control based on reinforcement learning, as described in any one of claims 1 to 6.

Images