System and method for controlling motion of an elevator tank

Abstract

The present disclosure provides a system and method for controlling motion of an elevator bank. The method comprises: accepting a current request served by an elevator group; receiving a partial trajectory of motion of a person moving in an environment served by the elevator group; and obtaining the probability of future elevator requests. The method further includes processing the portion of the trajectory using a neural network trained to estimate a weighted combination of probability density functions for indicating a time-of-arrival distribution of the person; and generating a set of possible future requests that jointly represent the probability of future elevator requests and the time-of-arrival distribution. The method further includes optimizing scheduling of the elevator group to serve a current request and a set of possible future requests; and controlling the elevator group according to the dispatch.

Description

Technical Field

[0001] This disclosure generally relates to an elevator, and more specifically, to a system and method for controlling the movement of an elevator group. Background Technology

[0002] Elevators or elevator groups are installed in buildings with multiple floors. Users use elevator groups to move from one floor to another within the building. Elevator scheduling is a crucial aspect of ensuring the efficient operation of elevator groups, for example, reducing user wait times. Typically, current elevator requests are received from users, and scheduling is determined based on these requests to control the elevator group. However, to determine the optimal scheduling for the elevator group to minimize the average wait time for all users, future elevator requests need to be considered. To consider future elevator requests, the arrival times of potential future passengers need to be predicted. Arrival time refers to the amount of time it might take a user to reach the elevator group.

[0003] Some methods use model-based approaches to predict arrival times. Model-based methods use queuing theory or agent-based modeling to build simulators of user movement. The models used in model-based methods make assumptions about user movement flow, speed, and density. The models used in model-based methods are essentially generative models, so sampling user trajectories from the models is relatively simple. However, trajectory sampling is slow and costly. Furthermore, performance degrades for simulations of longer durations. Additionally, accurate calibration of the physical model can be slow and laborious.

[0004] Therefore, a method and system for scheduling optimization and control of elevator groups is still needed. Summary of the Invention

[0005] Some implementations aim to provide a system and method for predicting the arrival times of future passengers at a specific elevator in an elevator group. Specifically, some implementations aim to predict a probability distribution of future passenger arrival times. Some implementations also aim to obtain the probability of future passenger elevator requests (i.e., future elevator requests) and, based on the obtained probability of future elevator requests and the predicted probability distribution of future passenger arrival times, determine the continuation of the future passenger arrival flow. Furthermore, some implementations aim to determine the scheduling of the elevator group based on the continuation of the future passenger arrival flow, wherein the scheduling optimizes performance metrics by minimizing the average waiting time (AWT) of all passengers.

[0006] Typically, elevator group service is provided upon receiving current elevator requests from passengers, and the scheduling of the elevator group is determined based on these requests. However, to determine the optimal scheduling of the elevator group, it is desirable to consider future elevator requests in addition to current ones. For example, a first passenger on the sixth floor requests elevator group service to go to the first floor. The elevator car assigned to serve the first passenger is currently on the fifteenth floor. It is also considered that a second passenger on the tenth floor will request service to go to the first floor within the next five seconds. In this scenario, the elevator scheduler pre-determines the arrival information associated with the future passenger (i.e., the second passenger) and can determine the scheduling so that the elevator car on the fifteenth floor can wait for five seconds until the future passenger (i.e., the second passenger) requests service to go to the first floor. When the second passenger's request is received, the elevator car can first pick up the second passenger on the tenth floor and then the first passenger on the sixth floor, so that both passengers can arrive at the first floor together. Therefore, it is desirable to consider future elevator requests to further optimize the passenger's AWT (Average Time To Wait).

[0007] To account for future elevator requests, it is necessary to determine the probability of a passenger arriving at the elevator and the probability of the time it may take for the passenger to arrive. The probability of a passenger arriving at the elevator and receiving elevator service can be called the probability of a future elevator request. The probability of the time it may take for the passenger to arrive at the elevator is called the arrival time.

[0008] The problem of determining the probability of a future elevator request is called the future service request problem. The problem of determining the probability of arrival time is called the arrival time prediction problem. Some implementations are based on the understanding that the future service request problem and the arrival time prediction problem can be considered as separate and independent problems. For example, the probability of a passenger arriving at the elevator can be determined independently, and the probability of arrival time can be predicted independently, assuming the passenger will arrive at the elevator. Because the future service request problem and the arrival time prediction problem are considered as separate problems, each problem can be solved using different methods.

[0009] For example, in some implementations, given a passenger's current location and destination, a neural network can be used to predict the passenger's arrival time. However, given the current location and destination, a passenger may choose multiple paths to reach the destination (e.g., a passenger starting from the current location might first go to an intermediate location (e.g., the restroom) and then to the elevator, thus taking longer than going directly to the elevator). To accurately predict arrival times, this multimodal nature of arrival times needs to be considered. Some implementations are based on the understanding that a probability distribution of arrival times can be predicted to account for this multimodal nature.

[0010] To this end, some embodiments of this disclosure disclose a neural travel-time mixture model (NTMM) for predicting the probability distribution of arrival times. The NTMM is trained to estimate a weighted combination of probability density functions, such that this weighted combination indicates the arrival time distribution of passengers arriving at the elevator via one of multiple paths. The weighted combination of probability density functions predicted by the NTMM has the advantage that, in addition to evaluating the probability density, the resulting distribution can also be used to evaluate the conditional cumulative distribution function and generate samples.

[0011] Furthermore, in some implementations, the probability of a future elevator request is determined using an attention-based destination predictor with an encoder-decoder structure. This attention-based destination predictor can be implemented using a neural network based on a transformer architecture. Alternatively, in some implementations, the probability of a future elevator request is determined using a recurrent neural network (RNN). Additionally or alternatively, in some implementations, the future elevator request is directly obtained. For example, in an organization's office space, a user interface can be provided to the user at their seating location (or their cubicle) to submit an elevator service request. The user can submit an elevator service request via the user interface.

[0012] The probability of known future elevator requests and the predicted arrival time probability distribution can be used to determine the optimal scheduling of elevator groups. However, directly combining the probability of known future elevator requests and the predicted arrival time probability distribution to determine the optimal scheduling is a challenge. Some implementations are based on the understanding that a continuation of future passenger arrival flows (also known as a "set of possible future requests") can be generated based on the probability of known future elevator requests and the predicted arrival time probability distribution. Furthermore, the optimal scheduling can be determined based on the continuation of future passenger arrival flows. Additionally, the elevator group is controlled according to the optimized scheduling to serve both current and future elevator requests simultaneously.

[0013] Therefore, one embodiment discloses a control system for controlling the movement of an elevator group, the control system comprising: at least one processor; and a memory storing instructions that cause the at least one processor of the control system to: accept one or more current elevator requests served by the elevator group; accept a partial trajectory of the movement of a person moving in an environment served by the elevator group; obtain the probability of future elevator requests; process the partial trajectory using a neural network trained to estimate a weighted combination of probability density functions, wherein the weighted combination of the probability density functions indicates an arrival time distribution of the person arriving at the elevator group via one of multiple paths in the environment; generate a set of possible future requests jointly representing the probability of the future elevator requests and the arrival time distribution; optimize the scheduling of the elevator group to serve the one or more current elevator requests and the set of possible future requests; and control the elevator group according to the scheduling.

[0014] Therefore, another embodiment discloses a method for controlling the movement of an elevator group, the method comprising: accepting one or more current elevator requests served by the elevator group; accepting a partial trajectory of the movement of a person moving in an environment served by the elevator group; obtaining the probability of future elevator requests; processing the partial trajectory using a neural network trained to estimate a weighted combination of probability density functions, wherein the weighted combination of the probability density functions indicates an arrival time distribution of the person arriving at the elevator group via one of multiple paths in the environment; generating a set of possible future requests jointly representing the probability of the future elevator requests and the arrival time distribution; optimizing the scheduling of the elevator group to serve the one or more current elevator requests and the set of possible future requests; and controlling the elevator group according to the scheduling.

[0015] Therefore, another embodiment discloses a non-transitory computer-readable storage medium having a program implemented thereon, the program being executable by a processor to perform a method for controlling the movement of an elevator group, the method comprising: accepting one or more current elevator requests served by the elevator group; accepting a partial trajectory of the movement of a person moving in an environment served by the elevator group; obtaining a probability of future elevator requests; processing the partial trajectory using a neural network trained to estimate a weighted combination of probability density functions, wherein the weighted combination of the probability density functions indicates an arrival time distribution of the person arriving at the elevator group via one of multiple paths in the environment; generating a set of possible future requests jointly representing the probability of the future elevator requests and the arrival time distribution; optimizing the scheduling of the elevator group to serve the one or more current elevator requests and the set of possible future requests; and controlling the elevator group according to the scheduling.

[0016] The currently disclosed embodiments will be further explained in conjunction with the accompanying drawings. The drawings are not necessarily drawn to scale, but generally focus on illustrating the principles of the currently disclosed embodiments. Attached Figure Description

[0017] Figure 1A A schematic diagram of controlling the movement of an elevator group according to an embodiment of the present disclosure is shown.

[0018] Figure 1B A block diagram of a control system for controlling the movement of an elevator group according to an embodiment of the present disclosure is shown.

[0019] Figure 1C A graph showing the arrival time distribution and the empirical probability density function (PDF) of a sample drawn from the arrival time distribution according to an embodiment of the present disclosure is shown.

[0020] Figure 2 Training of a neural time-of-process model (NTMM) according to some embodiments of the present disclosure is illustrated.

[0021] Figure 3A The floor layout of a building having a trajectory for training NTMM is shown according to some embodiments of the present disclosure.

[0022] Figure 3B The retraining of a trained NTMM according to some embodiments of this disclosure is illustrated.

[0023] Figure 4A A block diagram of a transformer-based neural network according to some embodiments of the present disclosure is shown.

[0024] Figure 4B A schematic diagram of a long short-term memory (LSTM) based neural network according to some embodiments of the present disclosure is shown.

[0025] Figure 5 A schematic diagram of a floor plan of a building according to some embodiments of the present disclosure is shown.

[0026] Figure 6 A flowchart of a method for controlling the movement of an elevator group according to an embodiment of the present disclosure is shown.

[0027] Figure 7 This is a schematic diagram illustrating a computing device used to implement the methods and control systems of this disclosure. Detailed Implementation

[0028] In the following description, numerous specific details are set forth for purposes of explanation in order to provide a thorough understanding of this disclosure. However, it will be apparent to those skilled in the art that this disclosure may be practiced without these specific details. In other instances, apparatus and methods are shown only as block diagrams to avoid obscuring this disclosure.

[0029] As used in this specification and claims, the terms “for example,” “such as,” and “like,” as well as the verbs “comprising,” “having,” “including,” and other verb forms thereof, when used in conjunction with a list of one or more components or other items, are to be interpreted as open-ended, meaning that the list should not be construed as excluding other additional components or items. The term “based on” means at least partially based on. Furthermore, it should be understood that the wording and terminology used herein are for descriptive purposes and should not be considered restrictive. Any headings used in this specification are for convenience only and have no legal or limiting effect.

[0030] Figure 1A A schematic diagram of controlling the movement of elevator group 115 according to an embodiment of the present disclosure is shown. Elevator group 115 may include one or more elevator cars, such as cars 117a and 117b. Elevator group 115 may be installed in an environment, for example, a building containing multiple floors (such as floors 119a, 119b, 119c, 119d, 119e, and 119f). Some embodiments aim to provide a control system for determining the optimal scheduling of elevator group 115, wherein the optimal scheduling optimizes performance metrics by minimizing the average waiting time (AWT) of all passengers. Some embodiments also aim to provide a control system for controlling elevator group 115 based on optimal scheduling. Such a control system is as follows: Figure 1B As stated above.

[0031] Figure 1B A block diagram of a control system 121 for scheduling and controlling elevator group 115 according to an embodiment of the present disclosure is shown. The control system 121 includes a processor 123 and a memory 125. The processor 123 may be a single-core processor, a multi-core processor, a computing cluster, or any other configuration. The memory 125 may include random access memory (RAM), read-only memory (ROM), flash memory, or any other suitable storage system. Furthermore, in some embodiments, the memory 125 may be implemented using a hard disk drive, an optical disk drive, a USB flash drive, a drive array, or any combination thereof.

[0032] Processor 123 receives a partial trajectory 101 of the movement of a person (possibly a potential passenger) moving within the environment served by elevator group 115. The partial trajectory 101 of the person's movement may correspond to a portion of the path the person takes from a starting point to a destination. The partial trajectory 101 may depend on the path taken by the person. For example, the partial trajectory 101 may be a straight line, a curve, or a zigzag shape, depending on the path taken by the person. The environment served by elevator group 115 may be an area of ​​floors among multiple floors (119a to 119f). In an embodiment, processor 123 may use environment-associated sensors (e.g., cameras, etc.) to capture the movement of the person to determine the partial trajectory 101.

[0033] Furthermore, processor 123 receives current elevator requests 111 from passengers for service of elevator group 115 and determines the scheduling of elevator group 115 based on the current elevator requests. However, to determine the optimal scheduling of elevator group 115, it is desirable to consider future elevator requests in addition to the current elevator request 111. For example, the first passenger on the sixth floor of the building requests elevator group service to go to the first floor. Considering that the elevator car assigned to serve the first passenger is currently on the fifteenth floor. Considering that a second passenger on the tenth floor will also request service to go to the first floor within the next five seconds. In this case, the elevator scheduler will pre-determine the arrival information associated with the future passenger (i.e., the second passenger) and can determine the scheduling such that the elevator car on the fifteenth floor can wait for five seconds until the future passenger (i.e., the second passenger) requests service to go to the first floor. When the request from the second passenger is received, the elevator car can first pick up the second passenger on the tenth floor and then pick up the first passenger on the sixth floor, so that both passengers can arrive at the first floor together. Therefore, it is desirable to consider future elevator requests to further optimize the passenger's AWT.

[0034] To account for future elevator requests, it is necessary to determine the probability that a passenger will arrive at elevator group 115 and receive elevator service, as well as the probability of the time it may take for the passenger to arrive at elevator group 115. The probability that a passenger will arrive at elevator group 115 and receive elevator service can be called the probability of a future elevator request. The probability that a passenger may take to arrive at elevator group 115 is called the arrival time.

[0035] The problem of determining the probability of a future elevator request is called the future service request problem. The problem of determining the probability of arrival time is called the arrival time prediction problem. Some implementations are based on the understanding that the future service request problem and the arrival time prediction problem can be considered as separate and independent problems. For example, the probability of a passenger arriving at elevator group 115 can be determined independently, and the probability of arrival time can be predicted independently, assuming the passenger will arrive at elevator group 115. Since the future service request problem and the arrival time prediction problem are considered as separate problems, each problem can be solved using different methods.

[0036] For example, in some implementations, given a passenger's current location and destination, a neural network can be used to predict the passenger's arrival time. However, given the current location and destination, a passenger may choose multiple paths to reach the destination (e.g., a passenger starting from the current location might first go to an intermediate location (e.g., the restroom) and then to the elevator, thus taking longer than going directly to the elevator). To accurately predict arrival times, this multimodal nature of arrival times needs to be considered. Some implementations are based on the understanding that a probability distribution of arrival times can be predicted to account for this multimodal nature.

[0037] To this end, some embodiments of this disclosure disclose a neural network 103 trained to estimate a weighted combination of probability density functions such that the weighted combination of probability density functions indicates the arrival time distribution of passengers arriving at the elevator via one of multiple paths. Processor 123 uses the neural network 103 to process a portion of trajectory 101 to estimate a weighted combination of probability density functions indicating the arrival time distribution 105 of persons arriving at elevator group 115 via one of multiple paths. The weighted combination of probability density functions predicted by the neural network 103 has the advantage that, in addition to estimating the probability density, the resulting distribution can also be used to evaluate the conditional cumulative distribution function and generate samples.

[0038] Furthermore, the processor 123 obtains a probability 107 of a future elevator request. In one embodiment, to obtain the probability 107 of a future elevator request, the processor 123 uses an attention-based destination predictor with an encoder-decoder structure to process a portion of the trajectory 101. This attention-based destination predictor can be implemented using a neural network based on a transformer architecture. Alternatively, in some embodiments, to obtain the probability 107 of a future elevator request, the processor 123 uses a recurrent neural network (RNN) to process a portion of the trajectory 101.

[0039] The probability 107 of future elevator requests and the arrival time distribution 105 can be used to determine the optimal scheduling of elevator group 115. However, directly combining the probability 107 of future elevator requests and the arrival time distribution 105 to determine the optimal scheduling is a challenge. Some implementations are based on the understanding that a possible set of future requests can be generated based on the probability 107 of future elevator requests and the arrival time distribution 105. To this end, processor 123 generates a possible set of future requests 109 for jointly representing the probability 107 of future elevator requests and the arrival time distribution 105. An example of generating this possible set of future requests 109 will be explained below.

[0040] Consider an eight-story building, with each floor (indicated by 1 to 8) comprising one elevator group. The starting time is t0 = 5:00:00 pm. The current elevator request is {(5:00:00 pm, 8, 1)}. The predicted duration is T = 30 seconds. Using a transformer-based neural network and neural network 105, the following future arrival information is extracted: On the 7th floor, passenger 1 from source 2 is expected to arrive at elevator group 115 within ΔT seconds with a probability of 0.8, where ΔT follows the arrival time distribution 105. Sixteen values ​​were sampled from arrival time distribution 105 for ΔT: [19.4, 9.8, 19.9, 12.7, 16.8, 19.7, 11.7, 21.0, 12.2, 21.1, 10.5, 17.3, 21.2, 13.9, 11.1, 12.3]. The continuation of the 20 passenger arrival flows forming the possible future request set 109 can be generated through Monte Carlo simulation, as follows:

[0041] {(5:00:00pm, 8, 1)}

[0042] {(5:00:00pm, 8, 1)}

[0043] {(5:00:00pm, 8, 1)}

[0044] {(5:00:00pm, 8, 1)}

[0045] {(5:00:00pm,8,1),(5:00:19.4pm,7,1)}

[0046] {(5:00:00pm,8,1),(5:00:9.8pm,7,1)}

[0047] {(5:00:00pm,8,1),(5:00:19.9pm,7,1)}

[0048] {(5:00:00pm,8,1),(5:00:12.7pm,7,1)}

[0049] {(5:00:00pm, 8, 1), (5:00:16.8pm, 7, 1)}

[0050] {(5:00:00pm,8,1),(5:00:19.7pm,7,1)}

[0051] {(5:00:00pm,8,1),(5:00:11.7pm,7,1)}

[0052] {(5:00:00pm,8,1),(5:00:21.0pm,7,1)}

[0053] {(5:00:00pm,8,1),(5:00:12.2pm,7,1)}

[0054] {(5:00:00pm,8,1),(5:00:21.1pm,7,1)}

[0055] {(5:00:00pm,8,1),(5:00:10.5pm,7,1)}

[0056] {(5:00:00pm,8,1), (5:00:17.3pm,7,1)}

[0057] {(5:00:00pm,8,1), (5:00:21.2pm,7,1)}

[0058] {(5:00:00pm,8,1),(5:00:13.9pm,7,1)}

[0059] {(5:00:00pm, 8, 1), (5:00:11.1pm, 7, 1)}

[0060] {(5:00:00pm,8,1), (5:00:12.3pm,7,1)}

[0061] Furthermore, processor 123 optimizes the scheduling 113 of the elevator group to serve the current elevator request 111 and the possible set of future requests 109. Processor 123 controls the elevator group 115 based on the optimized scheduling 113.

[0062] Figure 1C The arrival time distribution 105 estimated by the neural network 105 and the empirical probability density function (PDF) of the samples drawn from the arrival time distribution 105 are shown.

[0063] Figure 1CA graph 127 is shown showing an arrival time distribution 105 and an empirical PDF 129 of a sample drawn from the arrival time distribution 105 according to some embodiments of the present disclosure.

[0064] In this implementation, the neural network 105 trained to estimate a weighted combination of probability density functions indicating the arrival time distribution 105 is a neural journey-time mixture model (NTMM). The NTMM is trained on a dataset as follows: Figure 2 As stated above.

[0065] NTMM training

[0066] Figure 2 Training of NTMM 207 according to some embodiments of this disclosure is illustrated. NTMM 207 is trained using a dataset 201 containing coordinates 203 of the origin and destination of different individuals and durations 205 corresponding to the trajectories of different individuals. Additionally or alternatively, dataset 201 contains durations corresponding to the trajectories of the same individuals at different times. The trajectories used to train NTMM 207 are as follows: Figure 3A As stated above.

[0067] Figure 3A The diagram illustrates a layout 300 of a building's floors 301 with trajectories for training NTMM 207, according to some embodiments of this disclosure. In some embodiments, each trajectory (such as trajectory 305 and trajectory 311) can be represented by a sequence of checkerboard layout indices. Each trajectory may have its own start and destination. For example, trajectory 305 includes a start 303 and a destination 307, and trajectory 311 includes a start 309 and a destination 313. The checkerboard layout index can be obtained by discretizing the range of location coordinates, where the range of location coordinates can represent the longitude and latitude associated with the checkerboard layout index corresponding to floor 301. In another embodiment, the trajectory may include trajectories that simultaneously contain location and time information. Furthermore, each trajectory may be represented by a sequence of tuples of checkerboard layout indices, where each tuple may contain the checkerboard layout index obtained by discretizing the range of location coordinates and a prediction period.

[0068] like Figure 3AAs shown, the position coordinates on floor 301 are discretized into a checkerboard layout with indices 1 to 2500, thus simplifying the representation of the trajectories. The checkerboard layout indices 1 to 2500 are uniformly represented by 315. The space of floor 301 is discretized into a 50×50 checkerboard layout, and each trajectory is converted into a sequence of checkerboard layout positions. Such trajectories can be used to simulate a person's movement throughout the floor during a certain time period, where the destinations visited can represent the most frequently visited indoor destinations in real life, such as restrooms, staircases, elevators, and lounges. Mathematically, these trajectories are provided as explained below.

[0069] In the example implementation, given a set of position coordinates {(x l ,y l Let x = 1, 2, ..., L, where x and y represent longitude and latitude, respectively. min =min l∈{1,2,…,L} x l x max =max l∈{1,2,…,L} x l y min =min l∈{1,2,…,L} y l y max =max l∈{1,2,…,L} y l Next, N can be... X ×N Y Any given coordinate tuple (x) in the rectangle l ,y l The chessboard-shaped layout index p) l Determined as in and quantization interval And quantization interval As long as computing resources allow, a positive integer N X and N Y It can be reasonably taken to be large. Furthermore, given a set of trajectories, N can be determined using cross-validation. X and N Y .

[0070] Therefore, the trajectory can be represented as:

[0071] S=(p1,p2,…,p n ),

[0072] Where n represents the trajectory length, and the last entry p n Corresponding to the destination.

[0073] In some implementations, both longitude and latitude indices are used. For example, for the coordinate tuple (x... l ,y l The checkerboard index p of longitude (and corresponding latitude) l1 (correspondingly, p) l2 ) was identified as (correspondingly, Therefore, the same trajectory S is represented as:

[0074] S=((p 11 ,p 12 ),(p 21 ,p 22 ),…,(p n1 ,p n2 )).

[0075] The trajectories in dataset 201 are represented as trajectories S. NTMM 207 uses dataset 201 to learn the conditional probability density p(T|X). This conditional probability density can be parameterized by a model with parameter θ, which can be optimized using maximum likelihood estimation. Furthermore, the probability density function p(·|X) and the cumulative distribution function F(·|X) can be evaluated to enable various applications using PDF.

[0076] The probability density function of the target variable Y can be expressed as a weighted combination of the individual probability densities, as shown below:

[0077]

[0078] Where X is the input variable. k For the weight, φ k The components are Gaussian or normal. In this disclosure, a single Gaussian PDF or normal distribution is used. Therefore, the output of NTMM 207 is a set of K weights w k Mean μ k and standard deviation σ k .

[0079] Since the arrival times are all positive, a hybrid density network is used to model the arrival times in logarithmic space, and then an exponential transformation is used to determine the probability distribution of the arrival times. By transforming the probability distribution expression (1), the probability density of the time variable T can be determined as follows:

[0080]

[0081] Here, θ represents the parameter of NTMM 207. As mentioned above, φ k Indicates the use of mean μ k and standard deviation σ kThe Gaussian PDF is calculated. The mixture density function p(T|X) can be used not only to estimate the probability density but also to estimate the conditional cumulative distribution function (CDF) F(T|X), and further generate sample T. i ~p(T|X).

[0082] CDF can be estimated as:

[0083]

[0084] Where, Φ k It is the CDF of a Gaussian distribution. Samples T can be generated from the mixture density p(T|X) using the latent generating distribution. i :

[0085] Z~Cat(w1,…,w K )

[0086]

[0087] T←exp(Y)

[0088] Among them, Cat(·) and These are categorical distribution and Gaussian distribution, respectively.

[0089] Given a dataset This includes the starting and destination coordinates (X) of trajectory n. n ) and duration (T) n ),Right now and duration The log-likelihood function of NTMM 207 is:

[0090]

[0091] By maximizing the log-likelihood function, and using {π} k} and {σ k The parameters of NTMM 207 are learned by regularizing the prior knowledge of}. The resulting objective function is:

[0092]

[0093] Therefore, the hyperparameters in the objective function include (a) the number of components (K), (b) the regularization penalty term (λ), (c) the number of nodes in the two dense layers, and (d) the learning rate.

[0094] Some implementations are based on the understanding that the trajectories of different people may change over time. In other words, people's movement in the environment changes over time. Therefore, the trained NTMM 207 needs to be retrained on new datasets, for example, on movement patterns where the hyperparameters are no longer optimal.

[0095] Figure 3B The retraining of a trained NTMM 207 according to some embodiments of this disclosure is illustrated. The NTMM 207, already trained using dataset 201, is further retrained using a new dataset 317 to obtain an updated NTMM. The new dataset 317 contains the durations of updated trajectories corresponding to different individuals. Since NTMM 207 is retrained using previously obtained hyperparameters, there is no need to perform expensive hyperparameter searches each time NTMM 207 is retrained. Thus, NTMM 207 is trained in a computationally inexpensive manner.

[0096] NTMM 207 is used to estimate a weighted combination of the probability density functions indicating the arrival time distribution 105. As described below, based on the arrival time distribution and the probability 107 of future elevator requests, processor 123 generates a set of possible future requests 109.

[0097] Generate a set of possible future requests

[0098] Using H = {h1, h2, ..., h J Let} represent the set of passengers arriving within a certain time interval, where passenger h is the set of passengers arriving within that time interval. i You can use tuples (τ) i ,o i ,d i ) represents, where τ i Indicates arrival time, o i Indicates the floor reached, d i Let H(t) represent the destination floor. Let H(t) represent the set of passengers who arrive at time t, have not yet been served, and are still waiting. Let W(H(t)) represent the cumulative waiting time of the passengers in H(t). c (·|·) is a function representing the waiting time of one or more passengers, assuming another group of zero or more passengers are also assigned to the same car, for example, car 117a. The set of passengers assigned to car 117a at time t but not yet served and still waiting is H. c (t), ΔW c (h)≡W c (H c (t)∪h|H c (t)∪h)-W c (H c (t)|H cThe marginal increase in waiting time for passenger h after being assigned to car 117a can be determined as follows:

[0099]

[0100] Among them, W c (h|H c The term (t) represents the time required to use car 117a to serve passenger h, and This indicates an increase in the waiting time for passenger h.

[0101] As defined in this paper, a future passenger is a passenger who has not yet made a service request. The set of possible future requests can be defined as:

[0102]

[0103] The set of potential future requests can also be called the continuation set. For each continuation set, the optimal cumulative wait time (CWT) can be determined, as follows:

[0104]

[0105] in, This represents the continuation set of (passengers) assigned to car 117a. A subset of. The average waiting time (AWT) of this allocation can be determined as:

[0106]

[0107] Furthermore, the assignment is determined by minimizing the AWT on the candidate partitions of the continuation set. For this, either an exact algorithm or an approximate algorithm can be applied.

[0108] In addition, Let ξ be the training trajectory, where Represents trajectory S (ξ) The grid index of the position of the κ-th record, the last entry A grid index (or checkerboard layout index) representing the destination location. Training trajectory set {S (ξ) ξ = 1, ..., N trainThis can be obtained from historical records extracted by sensors (such as cameras installed in the environment). Let the predicted duration T represent the length of the time interval expected to be included in the generation of a set of possible future arrivals. The predicted duration T can be chosen to be large or small depending on the availability of computing resources. On each floor, sensors monitor personnel movement up to the current time t, and processors 123 determine a set of partial trajectories (e.g., partial trajectory 101, etc.) of candidate future arrivals within the time interval [t, t+T].

[0109] set up Let ζ be the ζ-th such partial trajectory, where Representing the trajectory The grid index of the ιth record position,

[0110] In addition, for the partial trajectory of the input The actual output of NTMM 207 is the probability distribution P of the set Ω of all candidate location grid indices. (ζ) (ω). Ω′ represents a subset of Ω, indicating the grid index of the elevator car door corresponding to that floor. Furthermore, the probability set {P} of the subset Ω′ can be extracted. (ζ) (ω); ω∈Ω′} to generate the continuation set. Based on partial trajectories The prediction can then be used with {P} (ζ) The probability information in (ω); ω∈Ω′} is used to form different continuations of the arriving flow through Monte Carlo simulation. Then, the prediction information of all available partial trajectories is considered. Obtain the complete continuation set and multiple further continuation sets.

[0111] Furthermore, in this implementation, to obtain the probability 107 of future elevator requests, the processor 123 uses a transformer-based neural network to process a portion of the trajectory 101. The transformer-based neural network will then... Figure 4A As described in the text.

[0112] Figure 4A A block diagram 400 illustrates components of a transformer-based neural network 435 according to some embodiments of the present disclosure. The transformer-based neural network 435 may include: an encoder 413 comprising a stack of N encoder blocks; and a decoder 431 comprising a stack of N decoder blocks, which may further comprise a stack of self-attention and fully connected feedforward layer components, such that the transformer-based neural network 435 uses positional encoding and parallelizable attention mechanisms.

[0113] Some implementations are based on the understanding that the attention mechanism of the transformer-based neural network 435 can model the dependencies between elements in a sequence without considering the distance between them. These advantages of the transformer-based neural network 435 are not limited to speech-related applications, but are also applicable to extended destination prediction tasks.

[0114] like Figure 4A As shown, firstly, the input trajectory 401 (such as a partial trajectory 101) is processed by embedding and position encoding block 403, where the input token can be converted into a d by the embedding layer. 模型 The input markers can be discrete elements in a dimensional vector. Furthermore, positional encoding can be used to obtain the order information of the elements of the input trajectory 401, which is then combined with the embedding vector to obtain a new vector via a summation operation. Additionally, the new vector can be passed through a multi-head attention block 407. The attention in encoder 413 can be used to reference other markers in the input trajectory 401 when attempting to encode a particular marker.

[0115] The multi-head attention block 407 may contain an array of scaled dot product attention components, where the attention function can be computed based on the data matrix X. The data matrix can be projected using three different matrices learned during training to obtain projection matrices Q, K, and V representing the query, key, and value, respectively.

[0116] In some implementations, attention can be calculated as follows:

[0117]

[0118] Where, d k It is the number of columns in the K matrix.

[0119] In some implementations, multi-head attention mechanisms involve using different learning projections to compute multiple attention functions to obtain improved representation performance.

[0120] Furthermore, the adder and normalization block 409 in encoder 413 may include residual connections and layer normalization (RLNs). The RLNs can connect the inputs and outputs of multi-head attention block 407, add them together, and then normalize them. The resulting element can further pass through feedforward sub-layer 411, and then through another adder and normalization block 415, which includes RLN connections in the process of leaving encoder 413. The adder and normalization block 415 can connect the inputs and outputs of feedforward layer 411, add them together, and then normalize them.

[0121] Furthermore, the target output trajectory 405 can pass through the embedding and position coding block 417 (similar to the embedding and position coding block 403) and then enter the masked multi-head attention block 419. The elements of the masked multi-head attention block 419 can be similar to the multi-head attention block 407 of the encoder 413, except that the target output trajectory 405 is masked, so that the masked multi-head attention block 419 only refers to the position of the previous output trajectory. In addition, the addition and normalization block 421, which includes residual connections and RLNs, can concatenate the input and output of the masked multi-head attention block 419, add them together, and then normalize them.

[0122] After the addition and normalization block 421, there may be another multi-head attention block 423, which is similar to the multi-head attention block 407 of the encoder 413, except that the multi-head attention block 423 uses information from the encoder output as additional input for attention calculation.

[0123] Furthermore, another addition and normalization block 425 concatenates the input and output of the multi-head attention block 423, adds them together, and then normalizes them. The resulting element can then pass through the feedforward sub-layer 427, and then through another addition and normalization block 429 as it exits the decoder 431. The addition and normalization block 429 concatenates the input and output of the feedforward layer 427, adds them together, and then normalizes them.

[0124] Outside the decoder 431, a linear layer and a softmax block 433 may exist, which can receive the output of the decoder 431 and create a logic vector for all possible output elements. Furthermore, the linear layer and softmax block 433 can perform a softmax operation to transform the values ​​in the logic vector, thereby outputting the probability 107 of future elevator requests.

[0125] Alternatively, processor 123 uses an RNN to process a portion of the trajectory 101 to obtain the probability 107 of future elevator requests. Examples of RNNs may include, but are not limited to, neural networks based on Long Short-Term Memory (LSTM), neural networks based on Bidirectional Long Short-Term Memory (BiLSTM), etc.

[0126] Figure 4BAn LSTM-based neural network 437 according to an embodiment of the present disclosure is illustrated. This LSTM-based neural network unit 437 includes an input gate, a forget gate, and an output gate. Each of these gates corresponds to a “standard” neuron in a feedforward (or multi-layer) neural network; that is, each gate computes a weighted sum of activations (using an activation function). Furthermore, it, ot, and ft represent the activations of the input gate, output gate, and forget gate, respectively, at time step t. The exit arrows from memory unit c to gates i, o, and f represent peephole connections. A peephole connection represents the contribution of memory unit c to the activation at time step t-1, i.e., the contribution of ct-1. In other words, gates i, o, and f compute their activations at time step t (i.e., it, ot, and ft, respectively), while taking into account the activation of memory unit c at time step t-1, i.e., ct-1. A single left-to-right arrow exiting memory unit c is not a peephole connection but represents ct, and the “×” symbol indicates element-wise multiplication between inputs. Furthermore, the LSTM-based neural network 437 includes an operation that applies a differentiable function (e.g., a sigmoid function) to a weighted sum.

[0127] In operation, the first LSTM unit receives a portion of the trajectory 439 and passes the hidden state vector ht to subsequent LSTM units, which are then passed to the softmax layer to output the probability 107 of future elevator requests.

[0128] Alternatively, in one implementation, the processor 123 receives elevator service requests from personnel via user equipment and generates a set of possible future requests that jointly represent the elevator service requests and arrival time distributions. This implementation is as follows: Figure 5 As stated above.

[0129] Figure 5 A schematic diagram of a building floor 500 according to some embodiments of the present disclosure is shown. Floor 500 may include elevator groups 115, one or more meeting rooms 501, one or more cubicles 505, and restrooms 507. In embodiments, a user device may be provided in each meeting room and each cubicle. For example, a user device 503 may be provided in meeting room 501. User device 503 may include a user interface for submitting an elevator service request to elevator group 115. Personnel residing in meeting room 501 may submit an elevator service request via the user interface of user device 503. Alternatively, the person may also submit an elevator service request via a user device (such as a smartphone, laptop, etc.).

[0130] Furthermore, processor 123 generates a set of possible future requests that jointly represent elevator service requests received via user equipment 503 and arrival time distribution 105. Processor 123 optimizes the scheduling of elevator group 115 to serve the current elevator request 111 and the generated set of possible future requests. Based on the optimized scheduling, processor 123 controls elevator group 115.

[0131] The following Figure 6 The document describes a general method for controlling the movement of elevator group 115.

[0132] Figure 6 A block diagram of a method 600 for controlling the movement of an elevator group 115 according to an embodiment of the present disclosure is shown.

[0133] At box 601, method 600 includes accepting one or more current elevator requests 111 from one or more passengers. For example, one or more passengers may make a request for an elevator car (such as car 117a or car 117b) by pressing or touching an “up” or “down” button associated with elevator group 115.

[0134] At box 603, method 600 includes receiving a partial trajectory (e.g., partial trajectory 101) of the movement of a person moving in the environment served by elevator group 115. At box 605, method 600 includes obtaining a probability of future elevator requests, e.g., probability 107 of future elevator requests.

[0135] At box 607, method 600 includes processing a portion of the trajectory using a neural network trained to estimate a weighted combination of probability density functions, wherein the weighted combination of probability density functions indicates the arrival time distribution of a person reaching the elevator via one of multiple paths in the environment. In this implementation, the neural network corresponds to... Figure 2 The above-described NTMM 207.

[0136] At box 609, method 600 includes generating a set of possible future requests for jointly representing the probability and arrival time distribution of future elevator requests. At box 611, method 600 includes optimizing the scheduling of elevator group 115 to serve current requests and the generated set of possible future requests.

[0137] At box 613, method 600 includes controlling elevator group 115 according to the optimized schedule.

[0138] Figure 7This is a schematic diagram illustrating a computing device 700 for implementing the method 600 and control system 121 of this disclosure. The computing device 700 includes a power supply 701, a processor 703, a memory 705, and a storage device 707, all of which are connected to a bus 709. Furthermore, a high-speed interface 711, a low-speed interface 713, a high-speed expansion port 715, and a low-speed connection port 717 can be connected to the bus 709. Additionally, a low-speed expansion port 719 is connected to the bus 709. Furthermore, an input interface 721 can be connected to an external receiver 723 and an output interface 725 via the bus 709. A receiver 727 can be connected to an external transmitter 729 and a transmitter 731 via the bus 709. External memory 733, external sensors 735, a machine 737, and an environment 739 can also be connected to the bus 709. Furthermore, one or more external input / output devices 741 can be connected to the bus 709. The network interface controller (NIC) 743 may be adapted to be connected to a network 745 via a bus 709, wherein data or other data may be presented on a third-party display device, a third-party imaging device, and / or a third-party printing device outside the computing device 700.

[0139] Memory 705 may store instructions executable by computing device 700 and any data usable by the methods and systems of this disclosure. Memory 705 may include random access memory (RAM), read-only memory (ROM), flash memory, or any other suitable storage system. Memory 705 may be one or more volatile storage cells and / or one or more non-volatile storage cells. Memory 705 may also be another form of computer-readable medium such as a magnetic disk or optical disk.

[0140] Storage device 707 may be adapted to store supplementary data and / or software modules used by computing device 700. Storage device 707 may include hard disk drives, optical disk drives, USB flash drives, drive arrays, or any combination thereof. Furthermore, storage device 707 may also include computer-readable media such as floppy disk devices, hard disk devices, optical disk devices, magnetic tape devices, flash memory or other similar solid-state storage devices, or device arrays including devices in a storage area network or other configuration. Instructions may be stored in an information carrier. When the instructions are executed by one or more processing devices (e.g., processor 703), one or more methods such as those described above will be performed.

[0141] The computing device 700 can be optionally connected to a display interface or user interface (HMI) 747 via a bus 709. The HMI 747 is adapted to connect the computing device 700 to a display device 749 and a keyboard 751, wherein the display device 749 may include a computer monitor, camera, television, projector, or mobile device, etc. In some implementations, the computing device 700 may include a printer interface for connecting to a printing device, wherein the printing device may include a liquid inkjet printer, solid inkjet printer, large format printer, thermal printer, UV printer, or dye-sublimation printer, etc.

[0142] High-speed interface 711 manages bandwidth-intensive operations of computing device 700, while low-speed interface 713 manages low-bandwidth-intensive operations. This functional allocation is merely illustrative. In some implementations, high-speed interface 711 may be connected to memory 705, user interface (HMI) 747, keyboard 751, and display 749 (e.g., via a graphics processor or accelerator), and to high-speed expansion port 715, which can accept various expansion cards via bus 709. In some implementations, low-speed interface 713 is connected to storage device 707 and low-speed expansion port 717 via bus 709. Low-speed expansion port 717, which may include various communication ports (e.g., USB, Bluetooth, Ethernet, Wireless Ethernet), may be connected to one or more input / output devices 741. Computing device 700 may be connected to server 753 and rack server 755. Computing device 700 may be implemented in a variety of different forms. For example, computing device 700 may be implemented as part of rack server 755.

[0143] This specification provides only exemplary embodiments and is not intended to limit the scope, applicability, or configuration of this disclosure. Rather, the following description of exemplary embodiments will provide those skilled in the art with a feasible description of implementing one or more exemplary embodiments. Various changes to the function and arrangement of the elements are contemplated without departing from the spirit and scope of the disclosed subject matter as set forth in the appended claims.

[0144] Specific details are set forth in the following description to provide a thorough understanding of the embodiments. However, those skilled in the art will understand that embodiments may be practiced without these specific details. For example, systems, processes, and other elements in the disclosed subject matter may be shown as components in block diagram form to avoid obscuring the embodiments with unnecessary details. In other cases, well-known processes, structures, and techniques may be shown without unnecessary details to avoid obscuring the embodiments. Furthermore, the same reference numerals and names in the various figures denote the same elements.

[0145] Furthermore, various implementations can be described as processes drawn as flowcharts, flow diagrams, data flow diagrams, structure diagrams, or block diagrams. Although flowcharts can describe operations as sequential processes, many operations can be performed in parallel or simultaneously. Moreover, the order of operations can be rearranged. A process may terminate upon completion of its operations, but may have additional steps not discussed or included in the diagrams. Furthermore, not all operations in any particular described process may occur in all implementations. A process may correspond to a method, function, program, subroutine, subroutine, etc. When a process corresponds to a function, the termination of the function may correspond to the function returning to the calling function or the main function.

[0146] Furthermore, implementations of the disclosed subject matter can be carried out, at least partially, manually or automatically. They can be performed, or at least assisted by, using machines, hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof, for manual or automatic implementation. When implemented in software, firmware, middleware, or microcode, program code or code segments for performing the necessary tasks can be stored in a machine-readable medium. The processor can then perform the necessary tasks.

[0147] The various methods or processes outlined herein can be encoded as software that can be executed on one or more processors employing any of a variety of operating systems or platforms. Furthermore, such software can be written using a variety of suitable programming languages ​​and / or programming or scripting tools, and can also be compiled into executable machine language code or intermediate code that executes on a framework or virtual machine. Typically, in various implementations, the functionality of program modules can be combined or distributed as needed.

[0148] Implementations of this disclosure can be embodied as a method, examples of which have been provided. Actions performed as part of this method can be ordered in any suitable manner. Therefore, implementations can be constructed in which actions are performed in a different order than those shown, which may include performing some actions simultaneously, even if these actions are shown as sequential actions in the illustrative embodiments.

[0149] Furthermore, the embodiments and functional operations of this disclosure described herein can be implemented in digital electronic circuits, in tangibly implemented computer software or firmware, in computer hardware (including the structures disclosed herein and their equivalents), or in one or more combinations thereof. Additionally, some embodiments of this disclosure can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-transitory program carrier, for execution by a data processing device or for controlling the operation of a data processing device. Furthermore, program instructions can be encoded on artificially generated propagation signals (e.g., machine-generated electrical, optical, or electromagnetic signals) that are generated to encode information for transmission to a suitable receiver device for execution by the data processing device. The computer storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or one or more combinations thereof.

[0150] According to embodiments of this disclosure, the term "data processing apparatus" can encompass all kinds of devices, apparatuses, and machines for processing data, including, for example, programmable processors, computers, or multiple processors or computers. The apparatus may include special-purpose logic circuitry, such as FPGAs (Field-Programmable Gate Arrays) or ASICs (Application-Specific Integrated Circuits). In addition to hardware, the apparatus may also include code that creates an execution environment for associated computer programs, such as code constituting processor firmware, protocol stacks, database management systems, operating systems, or combinations thereof.

[0151] A computer program (which may also be referred to or described as a program, software, software application, module, software module, script, or code) can be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and can be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A computer program may, but does not necessarily, correspond to a file in a file system. A program may be stored as a part of a file that holds other programs or data (e.g., stored in one or more scripts in a markup language document), in a single file dedicated to the program in question, or in multiple coordinating files, such as a file storing one or more modules, subroutines, or portions of code.

[0152] Computer programs can be deployed to execute on one or more computers located at a single site or distributed across multiple sites and interconnected via a communication network. For example, a computer suitable for executing a computer program can be based on a general-purpose or special-purpose microprocessor or both, or any other type of central processing unit (CPU). Typically, the CPU receives instructions and data from read-only memory or random access memory or both. The basic components of a computer are the CPU for executing or running instructions and one or more memory devices for storing instructions and data.

[0153] Typically, a computer will also include one or more mass storage devices (e.g., disks, magneto-optical disks, or optical disks) for storing data, or operatively coupled to receive data from or transfer data to said storage device, or both. However, a computer does not need to have such a device. Furthermore, a computer may be embedded in another device, such as a mobile phone, personal digital assistant (PDA), mobile audio or video player, game console, GPS receiver, or portable storage device such as a Universal Serial Bus (USB) flash drive, to name a few.

[0154] To interact with the user, embodiments of the subject matter described in this specification can be implemented on a computer having a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user, including acoustic, voice, or tactile input, can be received in any form. Furthermore, the computer can interact with the user by sending documents to and receiving documents from the device used by the user; for example, by sending a webpage to a web browser on the user's client device in response to a request received from a web browser.

[0155] The embodiments of the subject matter described in this specification can be implemented in a computing system that includes, for example, a back-end component as a data server, or a middleware component as an application server, or a front-end component as a client computer having, for example, a graphical user interface or a web browser, or any combination of one or more such back-end, middleware, or front-end components, wherein a user can interact with the implementation of the subject matter described in this specification through the graphical user interface or web browser. The components of the system can be interconnected via digital data communication (e.g., a communication network) of any form or medium. Examples of communication networks include local area networks (“LANs”) and wide area networks (“WANs”), such as the Internet.

[0156] A computing system may include clients and servers. Clients and servers are typically geographically separated and usually interact through a communication network. The relationship between clients and servers is established through computer programs running on their respective computers, and they have a client-server relationship with each other.

[0157] Although this disclosure has been described with reference to certain preferred embodiments, it should be understood that various other modifications and variations can be made within the spirit and scope of this disclosure. Therefore, aspects of the appended claims cover all such changes and variations within the true spirit and scope of this disclosure.

Claims

1. A control system for controlling the movement of an elevator group, the control system comprising: At least one processor; and a memory storing instructions that cause the at least one processor of the control system to: Accept one or more current elevator requests served by the elevator group; Accept a portion of the trajectory of movement of persons moving within the environment served by the elevator group; Obtain the probability of future elevator requests; The partial trajectories are processed using a neural network trained to estimate a weighted combination of probability density functions, wherein the weighted combination of probability density functions indicates the arrival time distribution of the person arriving at the elevator group via one of multiple paths in the environment. Generate a set of possible future requests that jointly represent the probability of the future elevator request and the arrival time distribution; Optimize the scheduling of the elevator group to serve the one or more current elevator requests and the set of possible future requests; as well as The elevator group is controlled according to the scheduling.

2. The control system according to claim 1, wherein, The neural network is a neural journey-time mixture model (NTMM).

3. The control system according to claim 2, wherein, The NTMM is trained on a dataset that includes the coordinates of the origin and destination, as well as the duration of the trajectory corresponding to different people.

4. The control system according to claim 3, wherein, The duration of the trajectory corresponds to the same person at different times.

5. The control system according to claim 3, wherein, The processor is also configured to retrain the NTMM based on a new dataset that includes the duration of updated trajectories corresponding to different individuals.

6. The control system according to claim 1, wherein, In order to obtain the probability of the future elevator request, the processor is also configured to use a neural network based on a transformer architecture to process the partial trajectory.

7. The control system according to claim 1, wherein, In order to obtain the probability of the future elevator request, the processor is also configured to use a recurrent neural network (RNN) to process the partial trajectory.

8. The control system according to claim 1, wherein, The processor is also configured to: The person receives an elevator service request via user equipment; and Generate a set of possible future requests that jointly represent the elevator service request and the arrival time distribution.

9. A method for controlling the movement of an elevator group, the method comprising the following steps: Accept one or more current elevator requests served by the elevator group; Accept a portion of the trajectory of movement of persons moving within the environment served by the elevator group; Obtain the probability of future elevator requests; The partial trajectories are processed using a neural network trained to estimate a weighted combination of probability density functions, wherein the weighted combination of probability density functions indicates the arrival time distribution of the person arriving at the elevator group via one of multiple paths in the environment. Generate a set of possible future requests that jointly represent the probability of the future elevator request and the arrival time distribution; Optimize the scheduling of the elevator group to serve the one or more current elevator requests and the set of possible future requests; as well as The elevator group is controlled according to the scheduling.

10. The method according to claim 9, wherein, The neural network is a neural journey-time mixture model (NTMM).

11. The method according to claim 10, wherein, The NTMM is trained on a dataset that includes the coordinates of the origin and destination, as well as the duration of the trajectory corresponding to different people.

12. The method according to claim 11, wherein, The duration of the trajectory corresponds to the same person at different times.

13. The method according to claim 11, wherein, The method further includes retraining the NTMM based on a new dataset that includes the duration of the updated trajectories corresponding to different individuals.

14. The method according to claim 9, wherein, To obtain the probability of the future elevator request, the method further includes using a neural network based on a transformer architecture to process the partial trajectory.

15. The method according to claim 9, wherein, To obtain the probability of the future elevator request, the method further includes using a recurrent neural network (RNN) to process the partial trajectory.

16. The method according to claim 9, wherein, The method further includes the following steps: The person receives an elevator service request via user equipment; and Generate a set of possible future requests that jointly represent the elevator service request and the arrival time distribution.

17. A non-transitory computer-readable storage medium having a program implemented thereon, the program being executable by a processor to perform a method for controlling the movement of an elevator group, the method comprising the steps of: Accept one or more current elevator requests from one or more passengers, which are served by the elevator group; Accept a portion of the trajectory of movement of persons moving within the environment served by the elevator group; Obtain the probability of future elevator requests; The partial trajectories are processed using a neural network trained to estimate a weighted combination of probability density functions, wherein the weighted combination of probability density functions indicates the arrival time distribution of the person arriving at the elevator group via one of multiple paths in the environment. Generate a set of possible future requests that jointly represent the probability of the future elevator request and the arrival time distribution; Optimize the scheduling of the elevator group to serve the one or more current elevator requests and the set of possible future requests; as well as The elevator group is controlled according to the scheduling.

18. The non-transitory computer-readable storage medium according to claim 17, wherein, The neural network is a neural journey-time mixture model (NTMM).

19. The non-transitory computer-readable storage medium according to claim 17, wherein, The NTMM is trained on a dataset that includes the coordinates of the origin and destination, as well as the duration of the trajectory corresponding to different people.

20. The non-transitory computer-readable storage medium according to claim 19, wherein, The method further includes retraining the NTMM based on a new dataset that includes the duration of the updated trajectories corresponding to different individuals.

Images