Cloud-side collaborative elevator dispatching method and equipment and medium

Abstract

The invention discloses a cloud-side collaborative elevator dispatching method and device and a medium, and relates to the field of automation control. The method comprises the steps that multi-mode elevator data in a target building are collected in real time; according to the multi-mode elevator data, through a strategy generation model, operation weight parameters in a future preset time period are generated; constructing a scheduling decision objective function through a scheduling decision engine, and solving the scheduling decision objective function according to the multi-modal elevator data to generate an elevator scheduling instruction; scheduling the elevators in the target building, and collecting running state data of the elevators in the scheduling process; and receiving the operation state data through the remote cloud, training and optimizing the strategy generation model based on historical operation state data in a historical preset time period, and issuing and updating the strategy generation model. Absolute stability and predictability of scheduling behaviors are ensured, and the requirement of an elevator system for operation safety is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automation control technology, in particular to an elevator scheduling method and device based on cloud-edge collaboration and a medium. BACKGROUND

[0002] Elevators are indispensable vertical transportation tools in modern high-rise buildings. With the development of automation control technology, modern elevators have integrated computer control and power electronics technology, and have functions such as intelligent scheduling and fault self-checking. Elevator group control refers to a way of centrally scheduling multiple elevators through an intelligent system, sharing elevator hall buttons and computer programs to run collaboratively, so as to optimize transportation efficiency, shorten waiting time and reduce energy consumption.

[0003] Traditional elevator group control systems usually use fixed scheduling algorithms. After passengers press the call button, the elevator is scheduled in response. It is difficult to adjust and optimize according to the unique and time-varying passenger flow pattern of the building. Especially when encountering "tidal" passenger flow during peak hours such as office building commuting and lunch, it is difficult to predict the peak period in advance and adjust the elevator state in advance to cope with the peak. This results in a lag in elevator response and an extension of passenger waiting time. At the same time, during the period of passenger concentration, some passengers will take the opposite direction to ensure that they can take the elevator. The traditional elevator group control system is difficult to identify and avoid such behavior, reducing the overall efficiency of the elevator.

[0004] In addition, with the development of artificial intelligence technology, especially the maturity of machine learning and deep learning algorithms, existing advanced elevator group control solutions have attempted to introduce these technologies to improve system performance. Through the deployment of a central server in the cloud, online real-time inference is performed based on machine learning or deep learning algorithms to generate specific elevator scheduling control instructions that are sent to local elevator controllers for execution. However, since cloud-based decision-making relies on the real-time performance and stability of the network, network delays, jitter, or interruptions can cause delays in sending instructions, affecting scheduling efficiency. At the same time, since deep learning and reinforcement learning models have "black box" characteristics, their internal decision-making logic is complex and opaque. Directly using them to generate control instructions can make it difficult to predict, verify, and authenticate the behavior of the elevator system, making it difficult to ensure high safety and high reliability of elevator operation. SUMMARY

[0005] To solve the above problems, the present application proposes an elevator scheduling method based on cloud-edge collaboration, comprising: Real-time collection of multi-modal elevator data in a target building; the multi-modal elevator data includes internal sensor data of each elevator and external passenger flow data of each floor; According to the multi-modal elevator data, an operation weight parameter of the target building in a future preset time period is generated through a strategy generation model. By means of the scheduling decision engine, a scheduling decision objective function is constructed based on the operation weight parameter, the scheduling decision objective function is solved according to the multi-modal elevator data, and an elevator scheduling instruction is generated; Based on the elevator scheduling instruction, an elevator control physical signal is generated, each elevator in the target building is scheduled, and operation state data of the elevators in the scheduling process is collected; The remote cloud receives the operation state data, the strategy generation model is trained and optimized based on historical operation state data in a historical preset time period, the optimized strategy generation model is sent to an artificial terminal for auditing, and the strategy generation model is updated and issued.

[0006] On the other hand, the application also proposes a cloud-edge collaborative elevator scheduling device, comprising: at least one processor; and, The memory is in communication connection with the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute a cloud-edge collaborative elevator scheduling method as described in the above examples.

[0007] On the other hand, the application also proposes a non-volatile computer storage medium, which stores computer executable instructions, and the computer executable instructions are configured to execute a cloud-edge collaborative elevator scheduling method as described in the above examples.

[0008] The cloud-edge collaborative elevator scheduling method proposed in the application can bring the following beneficial effects: By placing the complex model training and optimization process of artificial intelligence in the cloud offline, and placing the final scheduling instruction generation and execution in the local online, the core contradiction faced by the application of advanced artificial intelligence in the safety-critical field of elevators is fundamentally solved. At the same time, the scheduling uncertainty and reliability risk caused by network delay and interruption in the traditional single cloud intelligent decision-making mode is effectively overcome, and the security risks of unpredictable behavior and difficult to verify and authenticate caused by directly using black box models for real-time control are avoided. The deterministic decision engine deployed locally ensures the absolute stability and predictability of the scheduling behavior, meets the requirements of the elevator system for operation safety, and realizes the efficient unification of the intelligent level and the operation reliability.

[0009] And, by constantly learning and adapting to the unique and time-varying passenger flow patterns of a specific building in the cloud, a tailor-made optimal scheduling strategy is generated, overcoming the inherent defects of traditional fixed algorithm strategies, such as single strategy and inability to cope with tidal passenger flow. Not only does it significantly shorten passenger waiting time and eliminate anxiety and invalid elevator behavior caused by information opacity, but also reduces elevator invalid stops and empty running through optimized scheduling, thereby reducing overall energy consumption and equipment wear and tear. BRIEF DESCRIPTION OF DRAWINGS

[0010] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings: Figure 1 FIG. 1 is a flowchart of a cloud-edge collaborative elevator dispatching method according to an embodiment of the present application; Figure 2 FIG. 2 is a schematic diagram of an online-offline separated overall architecture of an elevator group control system according to an embodiment of the present application; Figure 3 FIG. 3 is a schematic diagram of a cloud-edge collaborative elevator dispatching device according to an embodiment of the present application. DETAILED DESCRIPTION

[0011] To make the objects, technical solutions, and advantages of the present application clearer, the following will describe the technical solutions of the present application in conjunction with specific embodiments of the present application and corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0012] The following will describe the technical solutions provided by the embodiments of the present application in conjunction with the drawings.

[0013] As shown in FIG. 1, the embodiments of the present application provide a cloud-edge collaborative elevator dispatching method, which comprises: Figure 1 S101, real-time collection of multi-modal elevator data in a target building; the multi-modal elevator data comprises internal sensor data of each elevator and external passenger flow data of each floor. S101, real-time collection of multi-modal elevator data in a target building; the multi-modal elevator data comprises internal sensor data of each elevator and external passenger flow data of each floor.

[0014] Through deployment of infrared sensor devices or 3D stereo vision sensors, ToF sensors, etc. in the elevator hall of each floor in the target building, the waiting area is scanned and monitored in real time. The sensor obtains the heat source or contour information of the waiting crowd in an anonymous manner, and generates the time sequence passenger flow density characteristics of the current uplink and downlink directions of each floor through built-in algorithm processing, including but not limited to: real-time number of people in each direction, crowd gathering degree, waiting time.

[0015] Meanwhile, the running state original data of each elevator is acquired in real time through the multi-source sensor devices built in the elevator car, such as a load sensor, a photoelectric encoder, a speed sensor and the like, and the running state features corresponding to each elevator are calculated, including the accurate floor position, the running direction, the current load, the running speed, the door state and the like.

[0016] According to the acquisition time stamp, the passenger flow features are fused with the elevator state features to construct a high-dimensional state feature vector which can comprehensively describe the current state of the system.

[0017] It should be noted that the application is applied in an intelligent elevator group control system, as shown in Figure 2 Fig. 1 is a schematic diagram of an overall architecture of the elevator group control system, and the architecture of the preferred embodiment thereof includes a local online execution module (100) and a remote offline optimization module (200).

[0018] The local online execution module (100) is deployed in the building locally and is responsible for all real-time tasks to ensure safety and efficiency, including a multi-modal data acquisition unit, a strategy generation model, a scheduling decision engine, a cooperative execution and interaction unit.

[0019] The multi-modal data acquisition unit (110) is the "senses" of the system and is responsible for collecting data required for decision-making in real time. The core thereof is an infrared thermal imaging sensor deployed on each floor, which is used to anonymously and accurately acquire the real-time waiting number of each elevator hall. The multi-modal data acquisition unit (110) also includes internal sensor data of the elevator itself, such as the load, speed, position and door state.

[0020] The cooperative execution and interaction unit (140) is the "limbs" of the system and is responsible for executing decisions and interacting with users, including an elevator cluster controller for receiving and accurately executing the physical operation instructions output by the planning solver, and a floor LED screen controller for receiving and synchronously displaying dynamic guidance information on the screen, such as "the elevator is full and does not stop at this floor" or "will stop soon and can carry 5 people".

[0021] The strategy generation model (120) is a trained machine learning model deployed on the local edge side and is used as a strategy parameter generator. Instead of directly outputting the control instructions of the elevator, the strategy generation model (120) outputs a set of running weight parameters, including the waiting time weight w_wait, the energy consumption weight w_energy and the equipment wear weight w_travel, according to the current real-time perceived elevator system state (i.e. the high-dimensional state feature vector). The real-time and deterministic decision-making is made locally through the strategy generation model, the learning process of the model (completed offline in the cloud) is separated from the inference process (completed online locally), and the safety and reliability problems caused by directly using the model for real-time control are solved.

[0022] The dispatch decision engine (130) is essentially a mathematical optimization solver, which receives the weight parameters from the strategy generation model output and combines the real-time system state to calculate the current optimal elevator dispatch strategy, which is the key to ensuring that the elevator dispatch behavior is efficient, safe, deterministic, and verifiable.

[0023] After a dispatch instruction is issued to an elevator to go to a target floor, the elevator responds to the corresponding dispatch instruction, and the infrared thermal imaging sensor continuously monitors the number of people waiting at the target floor. If the number of people waiting is zero before the elevator arrives at the target floor, i.e., the preset cancellation condition is met, a new instruction to cancel the elevator's stop at the floor is automatically generated and executed, the stop plan is canceled, and the efficiency is further improved.

[0024] The remote offline optimization module deployed in the cloud is the core of the system's ability to "think" and "evolve". It includes a delay training and optimization unit for working during idle periods such as at night, receiving historical operation data uploaded locally, and building a high-fidelity virtual environment; a reinforcement learning decision engine for model training in the virtual environment, finding a set of target function weight parameters that can optimize the system's overall efficiency through millions of simulations and learning; and a human-machine collaborative review unit that generates an optimization report when the engine finds a better set of parameters, clearly explaining the expected effect of adopting new parameters.

[0025] S102、According to the multi-modal elevator data, the strategy generation model is used to generate the running weight parameters of the target building in the future preset time period.

[0026] The high-dimensional state feature vector is input into the strategy generation model, which is a deep feedforward neural network that performs nonlinear transformation through multiple fully connected layers in the hidden layer to gradually extract high-level abstract features that can represent the system's complex operating mode.

[0027] The high-level abstract features are transmitted to the output layer of the model, which maps the high-level abstract features to a continuous running weight parameter space through a linear transformation layer in the output layer, and outputs the corresponding running weight parameters. For example, output a three-dimensional vector [10.0, 1.5, 1.0] representing waiting time weight, energy consumption weight, and equipment wear weight, indicating that the current strategy places great emphasis on passenger waiting time and relatively less emphasis on energy consumption and equipment wear, quantitatively defining the preference and trade-off of the current dispatch strategy at the moment.

[0028] The high-dimensional state feature vector includes: passenger flow features, including the current up / down waiting passenger numbers of each floor, the longest waiting time of each floor call request, etc.; elevator state features, including the current position, running direction, current load, running speed, door state of each elevator; and space-time context features, including whether the current time is in a peak period, whether the current date is a weekday, etc.

[0029] In S103, a scheduling decision engine is used to construct a scheduling decision objective function based on the operation weight parameters, and to solve the scheduling decision objective function based on the multi-modal elevator data to generate an elevator scheduling instruction.

[0030] The operation weight parameters are input into the scheduling decision engine, which performs weighted summation on a plurality of predefined operation cost items based on the operation weight parameters to construct a scheduling decision objective function.

[0031] Each of the plurality of cost items corresponds to a system optimization objective, and the operation weight parameters define the relative weights of the cost items in the scheduling decision objective function. The scheduling decision objective function is constructed in the form of the product of the waiting time weight coefficient and the total waiting time cost, plus the product of the energy consumption weight coefficient and the total energy consumption cost, plus the product of the equipment wear weight coefficient and the total equipment wear cost.

[0032] Meanwhile, the engine loads the preset constraint conditions corresponding to the scheduling decision objective function, takes the real-time multi-modal elevator data as state variables, and performs numerical optimization calculation in combination with the objective function and the constraint conditions to output an optimal decision variable set. Each optimal decision variable in the set is a binary decision variable (taking the value of 1 or 0) for accurately indicating whether to assign a specific elevator to respond to a specific direction call request of a specific floor.

[0033] The optimal decision variable set is analyzed, and all variables with a value of 1 are traversed to generate corresponding elevator scheduling instructions. Each instruction explicitly includes the assigned elevator number and the target floor number, for example: “assign elevator #B to respond to the 42nd floor down call”.

[0034] In the embodiments of the present application, the scheduling decision engine is a mixed integer linear programming solver. When the current system time enters the preset lunch period, the strategy generation model generates and outputs a set of operation weight parameters optimized for the lunch peak according to the high-dimensional state feature vector of the current elevator operation, for example: wherein, is the waiting time weight, is the energy consumption weight, is the equipment wear weight.

[0035] The dispatch decision engine receives the running weight parameter, and constructs the dispatch decision objective function, wherein the specific form of the objective function is: wherein, the square sum of the down longest waiting time of the floor j is calculated, and the square term is used to preferentially eliminate long waiting; , is the estimated energy consumption of the elevator i moving to the floor j; , is the travel distance of the elevator i moving to the floor j.

[0036] At the same time, the engine loads the constraint condition preset for the lunch peak sub-strategy, including constraint C1 (down capacity locking): for any elevator i not in the special service set (such as only containing e8), if its current position is in the high zone floor set F_high_zone (such as floor >= 30), it is prohibited to assign any new up hall station call for it. The constraint statement is: IF ∈ AND i ∉ , THEN = 0 for all j.

[0037] Constraint C2 (empty elevator active preset): for any idle elevator i (i.e. d_i(t) ==0) not in E_service, if it becomes idle at time t, the system automatically generates a virtual parking call for it to go to the "high zone center floor" (such as 40 floors). This call has the highest priority, which is equivalent to adding a constraint: This constraint realizes the active predictive deployment of the capacity.

[0038] Constraint C3 (task uniqueness): for each floor j with a down request (i.e. C_j_down(t)>0), at least one elevator must be assigned to respond. The constraint statement is: This constraint guarantees the fairness of the service and avoids passengers being omitted.

[0039] Constraint C4 (elevator load limit): for each elevator i, its current load l_i(t) plus the estimated total weight of the newly assigned passengers must not exceed its maximum load Max_Load_i. This constraint is a physical safety red line that must be followed.

[0040] Constraint C5 (Service Elasticity Reservation): For elevators belonging to the E_service set (e.g., e8), constraints C1 and C2 are not applied. This constraint guarantees the robustness of the system and the ability to cope with sudden demand.

[0041] The mixed integer linear programming solver performs numerical optimization calculations to solve a set of optimal decision variables from all possible scheduling schemes that minimize the objective function J(t) .

[0042] S104, based on the elevator scheduling instruction, generate elevator control physical signals, schedule each elevator in the target building, and collect the running state data of each elevator during the scheduling process.

[0043] Based on the elevator scheduling instruction, the interaction unit (140) converts it into specific elevator control physical signals, such as drive motor signals and stop layer signals, and sends them to the controller of the elevator cluster (500) for precise scheduling of each elevator in the target building. At the same time, the unit controls the LED information display screen (400) of each floor, dynamically updating the guidance information (such as "Elevator B will arrive soon, and can carry 5 people").

[0044] During the entire scheduling instruction execution process, the system continuously collects execution result data of each elevator through the multi-modal data acquisition unit (110), including new position, load change, actual energy consumption, and the latest passenger flow change data of each floor, forming new running state data. All these data are recorded in the local log in real time and in detail.

[0045] Through the scheduling decision engine, guidance information is generated at the same time as the elevator scheduling instruction, actively managing passenger expectations and behavior, effectively guiding passengers to choose the optimal elevator scheme, avoiding congestion, reducing invalid waiting and reverse elevator behavior, and optimizing passenger flow distribution. At the same time, the improvement of passenger behavior in turn reduces the total waiting time, energy consumption and equipment wear and tear of the system, and improves the efficiency of physical scheduling. As a result, the system forms a man-machine cooperative closed loop, i.e. scheduling decisions shape passenger behavior, passenger behavior affects system state, and system state drives optimization of scheduling strategy, achieving efficient scheduling with continuous self-adjustment.

[0046] S105, receive the running state data through the remote cloud, train and optimize the strategy generation model based on historical running state data in a historical preset time period, send the optimized strategy generation model to the artificial terminal for review, and update the strategy generation model.

[0047] The local data log and upload unit (150) uploads the massive historical running state data recorded in the historical preset time period to the remote cloud through an encrypted network channel. The offline learning and optimization unit (210) of the cloud offline optimization module (200) obtains the historical running state data. Based on the historical running state data, the unit (210) constructs and continuously calibrates a high-fidelity digital twin environment (211). The environment is a virtual mapping of the physical elevator system and can highly accurately simulate the running characteristics, response time of the elevator and the unique passenger flow mode (such as the tidal passenger flow) of the target building, providing a reliable simulation platform for subsequent model training.

[0048] In the digital twin environment, the strategy generation model is simulated and trained based on the historical running state data. In a preferred embodiment, the training adopts an efficient phased collaborative training process: first, the historical running state data is preprocessed and analyzed, and the corresponding scene features such as time, passenger flow density distribution and main running direction are extracted. Based on the scene features, the historical running state data is automatically divided and classified into several typical running scenes, such as "early morning peak uplink dominant scene", "lunch break downlink peak scene" and "evening sparse passenger flow scene".

[0049] For each typical running scene, a corresponding training data subset is constructed. In the digital twin environment, each training data subset is used to independently train the strategy generation model, obtaining a highly specialized strategy sub-model corresponding to each typical running scene.

[0050] A scheduling model is constructed, the input of the scheduling model is also a high-dimensional state feature vector representing the real-time state of the system, and the output is a set of weight distribution coefficients, each coefficient corresponds to the credibility or importance weight of the output of a strategy sub-model. In the digital twin environment, the meta-scheduling model is trained through historical running state data, and the goal is to learn how to intelligently select, weight and fuse the output of one or several strategy sub-models according to the current real-time system state.

[0051] Finally, the trained strategy sub-models and meta-scheduling model are combined to form a global strategy generation model. The parameters of each expert sub-model can be fixed, and the entire combined model can be further globally fine-tuned in the digital twin environment to optimize the smoothness of the meta-scheduling model in switching between different scenes and the overall collaborative performance, thereby obtaining an optimized strategy generation model. Avoid the difficulty of training a single model to handle all scenarios, and obtain better performance and stronger generalization ability.

[0052] After the training is completed, the cloud-based human-machine collaborative review platform (220) will automatically generate an optimization report. The report details the changes to the strategy generation model, performance on the test set (such as the estimated percentage of average waiting time reduction, energy consumption changes), and optimization performance confidence.

[0053] The optimization report is submitted to the human terminal and presented to the administrator through a graphical interface or integrated natural language interaction interface. The administrator can review the report details and even obtain more information through natural language questioning.

[0054] The system receives and analyzes the approval decision made by the administrator, generates a final approval instruction based on the approval intention (approval or rejection) and the confidence data in the optimization report.

[0055] If an approval instruction is generated, the cloud platform will encrypt and compress the optimized strategy generation model to generate a secure model update package. If a rejection instruction is generated, the process terminates and the model will not be updated. The system can record the rejection reason for subsequent training optimization.

[0056] Finally, the model update package is differentially distributed to the local end through a secure communication link. After receiving the package, the local end verifies the signature and integrity, replaces the old model with the new one, and updates the entire model in the next working period or at a preset time, completing the entire model update closed loop. Differential distribution reduces bandwidth usage, improves update efficiency and security.

[0057] By placing the complex model training and optimization process of artificial intelligence in the cloud offline, and generating and executing the final scheduling instruction locally online, the core contradiction of applying advanced artificial intelligence in the safety-critical field of elevators is fundamentally solved. At the same time, the scheduling uncertainty and reliability risks caused by network delays and interruptions in traditional single cloud intelligent decision-making mode are effectively overcome, and the security risks of unpredictable behavior and difficult to verify and authenticate caused by directly using black box models for real-time control are avoided. The deterministic decision engine deployed locally ensures the absolute stability and predictability of scheduling behavior, meeting the extreme requirements of elevator systems for operational safety, thereby achieving efficient unification of intelligent level and operational reliability.

[0058] Moreover, by continuously learning and adapting to the unique and time-varying passenger flow patterns of specific buildings in the cloud, the most customized optimal scheduling strategy is generated, overcoming the inherent defects of traditional fixed algorithm strategies, such as single strategy and inability to cope with tidal passenger flow. Not only does it significantly shorten passenger waiting time, eliminate anxiety and invalid elevator behavior caused by information opacity, but also reduces elevator invalid stops and empty running through optimized scheduling, thereby reducing overall system energy consumption and equipment wear and tear.

[0059] AsFigure 3 As shown, the embodiment of the present application also proposes a cloud-edge collaborative elevator scheduling device, comprising: at least one processor; and, a memory in communication with the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform a cloud-edge collaborative elevator scheduling method as described in any of the above embodiments.

[0060] The embodiment of the present application also provides a non-volatile computer storage medium, which stores computer executable instructions, and the computer executable instructions are configured to perform a cloud-edge collaborative elevator scheduling method as described in any of the above embodiments.

[0061] Each of the embodiments in the present application is described in a progressive manner, and the same or similar parts of each of the embodiments can be referred to each other. Each of the embodiments mainly describes the differences from other embodiments. In particular, for the device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

[0062] The device and medium provided by the embodiments of the present application are one-to-one corresponding to the method, so the device and medium also have similar beneficial technical effects as the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the device and medium will not be described here.

[0063] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0064] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions described in the flowcharts and / or block diagrams. Figure 1one or more processes and / or blocks Figure 1 an apparatus with the functionality to achieve the specified processes or functions in the block or blocks.

[0065] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 one or more processes and / or blocks Figure 1 an apparatus with the functionality to achieve the specified processes or functions in the block or blocks.

[0066] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the Figure 1 one or more processes and / or blocks Figure 1 an apparatus with the functionality to achieve the specified processes or functions in the block or blocks.

[0067] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0068] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, non-volatile memory, such as read-only memory (ROM), EPROM, and / or flash memory. The memory is an example of computer-readable media.

[0069] Computer-readable media includes permanent and non-permanent, moveable and non- moveable media that can be implemented in any method or technology for storage of information such as 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 technology, compact disc read-only memory (CD-ROM), digital versatile discs (DVDs) or other optical storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to computing devices. According to the definition herein, computer-readable media does not include transitory media, such as modulated data signals and carrier waves.

[0070] It should also be noted that the terms "comprising," "including," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0071] The above description is merely illustrative of the application, and not restrictive. Various modifications and changes can become apparent to those skilled in the art. Incorporating any modification, equivalent substitution, improvement, etc. within the spirit and principle of the application, shall be included in the scope of the claims of the application.

Claims

1. A cloud-edge collaborative elevator dispatching method, characterized by, The method comprises: real-time collection of multi-modal elevator data in a target building; the multi-modal elevator data comprises internal sensor data of each elevator and external passenger flow data of each floor; generating, according to the multi-modal elevator data, an operation weight parameter of the target building in a future preset time period through a strategy generation model; constructing, based on the operation weight parameter, a scheduling decision objective function through a scheduling decision engine, solving the scheduling decision objective function according to the multi-modal elevator data, and generating an elevator scheduling instruction; based on the elevator scheduling instruction, generating an elevator control physical signal, scheduling each elevator in the target building, and collecting operation state data of the each elevator in the scheduling process; receiving the operation state data through a remote cloud, training and optimizing the strategy generation model based on historical operation state data in a historical preset time period, sending the optimized strategy generation model to an artificial terminal for auditing, and updating the strategy generation model.

2. The cloud-edge collaborative elevator dispatching method of claim 1, wherein, The real-time collection of multi-modal elevator data in a target building specifically comprises: real-time acquisition of external passenger flow data corresponding to each floor through infrared sensor devices deployed on each floor of the target building, preprocessing the external passenger flow data, and generating time-series passenger flow density features corresponding to each floor; real-time acquisition of internal sensor data corresponding to each elevator through multi-source sensor devices built-in each elevator in the target building, calculation of operation state features corresponding to each elevator based on the internal sensor data; based on the collection timestamp, fusion of the time-series passenger flow density features and the operation state features, and construction of a high-dimensional state feature vector.

3. The cloud-edge collaborative elevator dispatching method of claim 2, wherein, The generation of an operation weight parameter of the target building in a future preset time period according to the multi-modal elevator data through a strategy generation model specifically comprises: inputting the high-dimensional state feature vector into the strategy generation model, extracting high-level abstract features through multiple fully connected layers in the hidden layer of the strategy generation model, and transmitting the high-level abstract features to the output layer of the strategy generation model; mapping the high-level abstract features to an operation weight parameter space through a linear transformation layer in the output layer, and outputting corresponding operation weight parameters; the operation weight parameters comprise waiting time weight, energy consumption weight, and equipment loss weight.

4. The cloud-edge collaborative elevator dispatching method of claim 1, wherein, The construction of a scheduling decision objective function based on the operation weight parameter through a scheduling decision engine, the solving of the scheduling decision objective function according to the multi-modal elevator data, and the generation of an elevator scheduling instruction specifically comprise: inputting the operation weight parameter into the scheduling decision engine; weighting and summing a plurality of predefined operation cost items based on the operation weight parameter, and constructing a scheduling decision objective function; the operation cost items comprise total waiting time cost, total energy consumption cost, and total equipment loss cost; determining a preset constraint condition corresponding to the scheduling decision objective function, solving the scheduling decision objective function based on the multi-modal elevator data and the preset constraint condition, and obtaining an optimal decision variable set; The optimal decision variable set is parsed to generate corresponding elevator scheduling instructions.

5. The cloud-edge collaborative elevator dispatching method of claim 4, wherein, The scheduling decision objective function is solved based on the multi-modal elevator data and the preset constraint condition to obtain optimal decision variables and generate elevator scheduling instructions, specifically including: The multi-modal elevator data is identified as state variables and the preset constraint condition, and numerical optimization calculation is performed to obtain optimal decision variables; The optimal decision variable is a binary decision variable, taking values of 1 or 0, and is used to indicate whether a specific elevator is assigned to respond to a specific floor call request; The optimal decision variable set is traversed to obtain a variable set with all values of 1, and corresponding elevator scheduling instructions are generated respectively, the elevator scheduling instructions including assigned elevator numbers and target floor numbers.

6. The cloud-edge collaborative elevator dispatching method of claim 1, wherein, The historical running state data in the historical preset time period is obtained, and based on the historical running state data, a digital twin environment of elevator operation of the target building is constructed; The strategy generation model is simulated and trained in the digital twin environment based on the historical running state data; The strategy generation model is trained to converge through iterative simulation training. The strategy generation model is simulated and trained in the digital twin environment based on the historical running state data, specifically including:

7. The cloud-edge collaborative elevator dispatching method of claim 6, wherein, The historical running state data is preprocessed to extract corresponding scene features, and the historical running state is divided into several typical running scenes based on the scene features; For each typical running scene, a corresponding training data subset is constructed, and the strategy generation model is trained in the digital twin environment through the training data subset to obtain a strategy sub-model corresponding to the typical running scene; A strategy scheduling model is constructed to output a weight distribution coefficient corresponding to the strategy sub-model through the strategy scheduling model; In the digital twin environment, the strategy scheduling model is trained through the historical running state data, and the trained strategy sub-models and the strategy scheduling model are combined to obtain an optimized strategy generation model. The optimized strategy generation model is sent to an artificial terminal for review to update the strategy generation model, specifically including:

8. The cloud-edge collaborative elevator dispatching method of claim 7, wherein, Based on the change details and optimization performance confidence of the strategy generation model, an optimization report is generated, the optimization report is submitted to an artificial terminal, and an administrator is presented with a review request through an interactive interface of the artificial terminal; The administrator's approval content is received and parsed, and based on the approval intention and the optimization performance confidence, an approval instruction is generated; the approval instruction includes an approval instruction and a rejection instruction; When the approval instruction is generated, the optimized strategy generation model is encrypted and compressed to generate a model update package; The model update package is differentially issued to the local end through a secure communication link to complete the update at the local end. It includes:

9. A cloud-edge collaborative elevator dispatching device, characterized by, At least one processor; And The memory is connected in communication with the at least one processor; wherein ​ The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the cloud-edge collaborative elevator dispatching method according to any one of claims 1-8.

10. A non-transitory computer storage medium storing computer-executable instructions that, when executed, cause a computer to perform: The computer executable instructions are configured to perform the cloud-edge collaborative elevator dispatching method according to any one of claims 1-8.

Images