A method for training a neural network to control a group of elevators, and a system that uses this method.

Abstract

The configurations of the present invention are: a receiver module to receive the state variables (110) of the elevators and the registered calls of a passenger from a group of elevators; a calculation module to calculate the sequencing parameters from the state variables (110) of the elevators and the registered calls of the passengers; a determination module (120) to determine an order of these sequencing parameters according to at least one demand parameter; a sequencing module (130) to sort the state variables (110) of the elevators in the determined order; a neural network module (140) to input the sorted state variables (150) of the elevators into a trained neural network and output the sorted control values ​​for each elevator;It relates to a system for controlling an elevator group (100) which includes a sequencing removal module (160) to remove the sequencing of the sequential control values ​​(170) of each elevator in order to produce control values ​​and a controller to control the elevator group;

Description

1 TARIFF TRAINING A NEURAL NETWORK FOR CONTROLLING A GROUP OF ELEVATORS THE METHOD AND A SYSTEM THAT USES IT Technical Field to Which the Invention Relates The present invention describes a method for training a neural network for controlling a group of elevators, 5 a method and system that uses trained neural networks to control an elevator group This relates to the assignment of an elevator, specifically one dedicated to transporting a new passenger. State of the Art Elevators are vertical transport devices used to move passengers between different levels of a building. They are transport devices. They typically run along a rail or shaft within a building. It consists of a cabin that travels through space. Elevators are commonly found in multi-story buildings. and is a fundamental requirement for efficient vertical movement. Modern multi-story buildings and similar structures utilize elevators to serve many passengers. And therefore it has a group consisting of multiple elevator cabins and shafts. This Multiple elevator cabin and shaft assembly, to increase the operational efficiency of all elevators and 15 generating control decisions for elevators, such as allocating elevators to passengers. For this purpose, a group of control systems is needed. Today, the most advanced elevator control systems are used to control the system and specifically, to assign the most suitable elevator to transport a new passenger. They use networks. However, we are talking about large buildings with many elevators in a group. 20 When it comes to this, the number of possible system states increases exponentially with the number of elevators. This makes the number of required training examples or cycles extremely high. This brings about a need for an efficient elevator group control system. Neural networks use training algorithms to determine connection weights based on a set of training examples. It requires "training" to produce. However, a large 25 with numerous elevators in a group In the case of buildings, the number of possible system states is exponentially greater than the number of elevators. as it increases and the number of required training examples or cycles is extremely high. This situation makes education inefficient. 2 On the other hand, the elevator group is symmetrical according to the changes in the elevator status. It possesses these properties. If this symmetry can be taken into account, the state space is greatly reduced, and Effective education becomes possible. One of the current methods is to create a neural network itself with the same weights in the relevant places. It uses symmetry by making it symmetrical. However, this method limits the expressive power of the network and 5 This limits the quality of control. A general, unrestricted neural network while satisfying symmetry conditions. It is desirable to find a method for training effectively. Brief Description of the Figures The attached figures are for illustrative purposes only of an invention which will be described in detail below. It is given to illustrate its configuration: 10 Figure 1 shows the control of an elevator group where neural network connection symmetry is not considered. It is a schematic drawing of a view of the system. Figure 2 shows the neural network connectivity symmetry in a sample configuration of the present invention. A schematic view of the system for controlling a group of elevators, taking into consideration It is a drawing. 15 The elements shown in the figures are numbered as follows: 100. A system for controlling a group of elevators. 110. State variables of elevators 120. Determination module 130. Sorting module 20 140. Neural network module 150. Ordered state variables of elevators 160. Sort removal module 170. Sequential control values ​​for each elevator. 180. The manufactured control values ​​for each elevator are 25. E1. First elevator E2. Second elevator E3. Third elevator. En. Nth elevator S11, S12, S1m. The status variables of the first elevator are 30. S21, S22, S2m. State variables of the second elevator. S31, S32, S3m. State variables of the third elevator. 3 State variables of the n-th elevator: Sn1, Sn2, Snm. Do1, Do2, Do3, Don. The defined order of the state variables. Oc1. Sequential control values ​​of the first elevator. Oc2. Sequential control values ​​of the second elevator. Oc3. The sequential control values ​​of the third elevator are 5. The sequential control values ​​of the nth elevator. Gc1. Manufactured control values ​​of the first elevator. Gc2. The manufactured control values ​​of the second elevator. Gc3. The manufactured control values ​​of the third elevator. The manufactured control values ​​for the nth elevator are 10. Detailed Description The configurations of the present invention; a state variable for elevators from an elevator group. examples of control value examples for each elevator in the elevator group and one Obtaining state variable samples for at least one recorded call from a passenger; elevators The status variable of at least one recorded call in question is 15, along with examples of status variables. Ranking parameters, which are correlational variables between samples, from examples of elevator state variables and at least one recorded instance of the passenger in question. Calculating the status variable for each elevator from the call's instances; control based on at least one demand parameter that defines the expected conditions from the value samples Determining the order of the sequencing parameters for each elevator; elevator status 20 The specified order of variable samples and control value samples for each elevator The sorting of elevators in an order; an input of sorted state variable instances of elevators. a training set and an output training set of sequenced control value examples for each elevator. a group of elevators that involves the steps of training a neural network by inputting it as input. It relates to a method for training a neural network for control. In other words, 25 an input layer of ordered state variable instances of elevators and elevators A neural network is created by inputting sequenced control value samples as an output layer. It is the training of the network. The method for training a neural network to control a group of elevators is called computer-implemented training. That could be one method. 30 One configuration of the invention is training a neural network for controlling a group of elevators. an elevator trained with any configuration of the method or obtained directly It involves a neural network for controlling the group. 4 The configurations of the present invention also include an elevator assembly comprising the following steps: It relates to the method of checking: the condition of elevators from a group of elevators. the variables (110) and the status variables of at least one registered call of a passenger taking; elevator status variables (110) and at least one recorded call of a passenger Ranking parameters, which are correlational variables between state variables, 5 from the status variables of the elevators (110) and at least one recorded call of a passenger Calculation of state variables for each elevator; a generated control Sorting according to at least one demand parameter that defines the expected conditions for its value. Determining a sequence of parameters; state variables of elevators (110) Arranging the elevators in a specified order; the sorted status variables of the elevators are 10. (150) any method of training a neural network for the control of a group of elevators Entering a neural network that has been trained (in other words, acquired) with a specific configuration; The trained neural network outputs the sequenced control values ​​for each elevator. obtaining; generating control values ​​for each elevator for each elevator Removal of the ordering of the ordered control values ​​(170); each 15 of the elevator group Checking the elevator according to the manufactured control values ​​(180). A method for controlling a group of elevators can be a computer-implemented method. In a configuration of a method for controlling an elevator group, "the elevator group The step is "checking each elevator according to the manufactured control values ​​(180)". At least one elevator must be allocated in response to the passenger's registered request and / or there will be 20 waiting. at least one elevator must be directed to a location in the elevator shaft so that it remains there. includes. The configurations of the present invention also include an elevator assembly comprising the following components: It is related to the system (100) for the control of: the status of elevators from a group of elevators. variables (110) and status variables of at least one registered call of a passenger 25 a receiver module to receive; elevator status variables (110) and at least one passenger The rank is the correlational variable between the state variables of a registered call. parameters of the elevators state variables (110) and at least one passenger In order to calculate the status variables of the recorded call for each elevator The calculation module defines at least 30 conditions expected from a generated control value. a sequence of sorting parameters to determine an order based on a demand parameter Determination module (120); determines the status variables (110) of the elevators in the specified order. a sorting module (130) for sorting in order; sorted status of elevators Training a neural network for controlling a group of elevators with variables (150) a neuron trained (in other words obtained) with any configuration of the method to enter the network and retrieve the sequenced control values ​​for each elevator from the trained neural network. a neural network module to receive as output (140); control values ​​for each elevator To produce the sequence of control values ​​(170) of each elevator, 5 a sequencing lifting module (160) for lifting; lifting group of elevators of each elevator a controller to control according to the generated control values ​​(180). In a configuration of the system (100) for controlling a group of elevators, the controller, by assigning at least one elevator to the passenger's registered call and / or the elevator group At least one elevator must be positioned in the elevator shaft for the purpose of keeping it on standby. by directing each elevator to control according to the generated control values ​​(180). It has been adapted. One of the main aims of the invention is to provide a passenger with the most suitable and optimal boarding call. using a neural network to control a group of elevators for the purpose of allocating elevators It is the implementation of a method and system. To achieve this goal, an elevator 15 an efficient method for training a neural network to control a group It has been accomplished. Neural networks use a training algorithm to weight connections based on a set of training examples. It requires "training" to produce. Buildings with a large number of elevators in a group are an example of this. when it comes to the subject, the relevant state space (state variable 20 which is related to a decision strategy) (offering a viable range of values) its size increases exponentially with the number of elevators. and the necessary training examples (along with the calculated control values ​​of each elevator). The number of state variables (110) or cycles of elevators becomes extremely high income. This situation results in inefficient training for the neural network. Figure 1 shows the control of an elevator group where neural network connection symmetry is not considered. 25 It shows the system (100) for. Referring to Figure 1, the condition of the elevators variables (110), that is, an input layer, directly to the neural network module (140) (that is, a (hidden layer) is entered and the generated control values ​​(180) of each elevator are output. The status variables of the elevators are taken in Figure 1 (110); the status of the first elevator variables (S11, S12, S1m), state variables of the second elevator (S21, S22, S2m), 30 The state variables of the third elevator (S31, S32, S3m) and the state of the n-th elevator. The variables are represented by (Sn1, Sn2, Snm). The generated control values ​​are the outputs; The manufactured control values ​​of the first elevator (Gc1), the manufactured control values ​​of the second elevator 6 values ​​(Gc2), the manufactured control values ​​of the third elevator (Gc3) and the n-th elevator These are represented by the generated control values ​​(Gcn). For a company with a large number of elevators and For elevator groups serving a high number of passengers, an elevator like the one shown in Figure 1 is suitable. The system for controlling the group (100) is not efficient because symmetry is not taken into account. The neural network fails to train. 5 However, elevator groups, state variables of elevators (110) It has symmetry properties depending on how it is modified. If this symmetry is applied to the training of the neural network... If possible, the spatial dimension of the relevant state variables (110) of the elevators can be taken into account. This is greatly reduced, and efficient education becomes possible. A method that uses the symmetry property arranges the neural network itself into 10 corresponding nodes. The goal is to make them symmetrical so that they have the same weights. However, this method distorts the network. It restricts the power of expression and limits the quality of control. In this restrictive method aimed at achieving symmetrical behavior in neural network operation, The process is accomplished by replicating the sets of connection weights. This results in an input-output ratio. In the symmetry condition, it makes the neural network connections symmetrical. For example, {Y1, Y2, ... Yn} 15 Consider a neural network with input vectors {X1, X2, ... Xn} mapped to an output vector. The network will have a link weight matrix at the first layer. The link weight matrix, It consists of vectors {W1, W2, ... Wn} connected to successive layers. The symmetry condition For this to be true, if an input vector {A, B, ... P} is mapped to an output vector {a, b, ... p}, Any permutation of inputs such as the input vector {B, P, ... A} can also be given by {b, p, ... a}. The output vectors must be mapped to the same permutation. This is the link. This will be obtained if the matrices are constrained. For example, the first matrices of the neural network The vectors for the layer are in the form W1 = [ w0, w1, w1, ... w1 ], W2 = [ w1, w0, w1, ... w1], ... Wn = [ w1, w1, ... w0 ] and all other layers must be constrained similarly. However, if such a neural network behaves symmetrically with respect to permuted input-output patterns... 25 a general nerve where the degree of freedom is such that all weights can be adjusted independently. It is much lower than what can be achieved by an unconstrained neural network. There will be mappings. However, in the symmetrical one, it is necessary to obtain the unrestricted state. It's impossible. Therefore, symmetry 30 does not compromise the general, unrestricted concept of the neural network. training a neural network to control a group of elevators while fulfilling the conditions The aim is to find an effective method for this. 7 The application focuses on training a neural network to control a group of elevators. It describes a method. This method explains the general and unrestricted concept of neural networks. It ensures the efficiency of neural network training while maintaining symmetry conditions. The invention incorporates supervised learning and reinforcement learning for the neural network. Different types of training algorithms can be used, including but not limited to these. 5 From a training set of X[i] input instances which are state variable instances, the control value A multilayer neural network learns to map to output examples Y[i] which are examples. To train the network, a supervised learning algorithm, such as backpropagation, is preferred. is used. The supervised learning algorithm uses training examples (state variable examples). (and control value examples) an error in matching will remain until the selected value falls below 10. The neural network adjusts the connection weights sequentially. A neural network for controlling a group of elevators. the method involves minimizing a loss function through a series of control episodes. to train a neural network to control according to a control target, for example Q- Reinforcement learning, such as learning through reinforcement, is preferably used. Reinforcement learning The algorithm observes the development of the controlled method during each control episode. 15 and adjusts the neural network connection weights in such a way that the objective function is successively improved. Figure 2 shows the control of an elevator group considering neural network connection symmetry. The system (100) shows the receiver module in the system, elevators from a group of elevators status variables (110) and the status of at least one registered call of a passenger It takes the variables. In Figure 2; the first elevator (E1), the second elevator (E2), the third elevator 20 (E3) and the nth elevator (En), with their own state variables, i.e. the state of the first elevator. variables (S11, S12, S1m), status variables of the second elevator (S21, S22, S2m), The state variables of the third elevator (S31, S32, S3m) and the state of the n-th elevator. The variables (Sn1, Sn2, Snm) are used to calculate the sorting parameters. Then, the order of the sorting parameters is determined by the order determination module (120). 25 Next, the sorting module (130) sorts the status variables (110) of the elevators. Status The specified order of the variables (Do1, Do2, Do3, Don.), the determination module (120) It is represented in the ordered state variables (150) of the elevators in the neural network module. (140) is entered. Neural network module (140); sequential control values ​​of the first elevator (Oc1), the sequential control values ​​of the second elevator, (Oc2), the sequential 30 of the third elevator Output the control values ​​(Oc3) and the sequential control values ​​(Ocn) of the n-th elevator. It gives the control values ​​of each elevator. The sorting lifting module (160) gives the control values ​​of each elevator. to produce the sequence of the sequential control values ​​(170) of each elevator It performs a reverse permutation of the determined order for lifting. Each lift 8 Control values ​​produced for (180), i.e. the control values ​​produced for the first elevator (Gc1), manufactured control values ​​of the second elevator (Gc2), manufactured control values ​​of the third elevator control values ​​(Gc3) and manufactured control values ​​(Gcn) of the nth elevator, elevator It is carried out to control the group. In the configurations of the invention, examples of state variables of elevators and / or 5 Status variables of elevators (110); current position of each elevator, each the direction of the elevator, the number of passengers in each elevator, the total weight of each elevator, The maximum weight carrying capacity of each elevator, ensuring each elevator fits comfortably. the maximum number of passengers it can carry (the available space for passengers in each elevator) (maximum number), acceleration rate for each lift at various weights, each 10 the speed of the elevator, the speed of each elevator for various weights, the speed of each passenger entering the elevator each Estimated loading time for each elevator upon entry (loading) and each exit (unloading) from the elevator. weight, estimated time taken during loading and unloading, each at various speeds energy consumed by the elevator, weight per unit of distance, and similar factors. from the group (examples of elevator state variables and / or elevator state 15) The observable of each elevator that can be selected from the group of variables (110) can be selected It may include, but is not limited to, the values ​​and calculable values ​​for each elevator. No. Calculated values ​​include the estimated arrival times of the elevator to each floor, and the values ​​at each floor. This may include, but is not limited to, the estimated arrival time of passengers and similar factors. In the invention's configurations, the state variable is 20 for at least one recorded call by a passenger. instances and / or status variables for at least one recorded call by a passenger; a the passenger's location, that is, the location where the call recording took place, the call recording from a group including the timestamp and similar markings of when it was carried out (at least one passenger's instances of the status variable for a registered call and / or at least one registered passenger (from the group from which status variables can be selected for the call) can include selectable values ​​25 but not limited to these. In the invention's configurations; the calculated sequencing parameters; an elevator and a passenger the distance between (in other words, the location of an elevator and a call log) the distance between locations such as the floor it is located on), for passengers in each elevator the number of available seats, the number of currently registered calls, the passenger's expected wait is 30 the duration, the estimated total waiting time for all available passengers when the elevator is selected, the time elapsed since a call was recorded (the time since a call was recorded) (duration), the number of waiting passengers on each floor (and if possible, video capture such as a camera) 9 from the group including (obtained from their vehicles) and similar items [calculated ranking] It can include values ​​that can be selected from the group of parameters that can be selected, but only with these. is not limited. The calculated ranking parameters are those of a fuzzy control algorithm. The output is a partially trained neural network undergoing reinforcement learning for optimal control. The output of a heuristic group controller control algorithm for an elevator group, such as the output of 5 It may include. The calculated sorting parameters also include those mentioned above. It may also include a weighted sum of two or more sorting parameters. The sequencing in the invention's configurations involves instances of state variables for elevators or sequential arrangement of the elevators' state variables (110) in increasing or decreasing order It may include, but is not limited to, a sequence. In a sequential sequence, the elevators are 10 state variables (110) or examples of state variables for elevators; from the lowest They are sorted from highest to lowest, shortest to longest, closest to furthest, and so on. However, it is not limited to these. Sequential ordering is some state variable for elevators. examples or state variables of elevators (110) outside of the order It may have a nonlinear structure that can be held or skipped. 15 "Determining the sequence of sequencing parameters for each elevator" and "the elevators The defined examples of state variables and control value samples for each elevator The purpose of the "ordering in sequence" operation is to create a neural network that satisfies symmetry conditions. The aim is to provide a training example. If an elevator "i" is in the instance of the state variable Xi = A, then... If the elevator "k" is in the instance of the state variable Xk = B and the ordered training instance is […, 20 If the state variable instances are in the form [(Xi=A), …, (Xk=B), …], then the instances of the state variable are (exactly or approximately). When modified (as), the ordered training instance becomes […, (Xk=A), …, (Xi=B), …] It should be done. The assignments – best fit, second best fit, etc. – must be known precisely. In this case, a perfect sequence is achieved. The sequence of each elevator The order of the parameters is determined accordingly. 25 To this end, the sequencing process involves observing each elevator. and / or examples of elevator status variables that may have calculable values ​​and / or according to a sorting function that works on elevator status variables (110) This can be done. The sorting function can be trivial (e.g., the elevator's registered call). Status variable 30 such as distance and / or estimated arrival time of the elevator to a recorded call (Using only one of the instances / state variables), multiple instances The variable can be a weighted sum of the instance / state variable (0.5 * distance + 0.5 * (such as arrival time), or elevators, a "fitness" set for call service (that is, arranging according to a sequence of orderings that are considered appropriate for the purpose) It can be a detailed function. Therefore, suitable sorting functions are optimal. These are "good" approaches to assignments. "The distance of the elevator to a registered call (at most Sorting based on situational variables such as "from closest to furthest" is appropriate. That's why there are rankings. 5 In the design of the invention; demand parameters, minimum waiting time for the passenger (waiting time for the passenger, the elevator allocated to the passenger based on the passenger's registered call time) (meaning the time interval between the moment it reaches its destination), lowest energy consumption, service the time period from a passenger entering (loading) the elevator to the same passenger exiting the elevator. The minimum service time, which refers to the time elapsed between (emptying) and a specific time, is 10. Maximum efficiency, which refers to the maximum number of passengers carried in a segment, the most from the group that includes low congestion and similar conditions [from the group from which demand parameters can be selected] It may include, but is not limited to, selectable values. Clarifying the method of training a neural network for controlling a group of elevators 15 A simplified example is provided below for this purpose: • In the first step, that is, "state variable for elevators from a group of elevators examples of control value examples for each elevator in the elevator group; Obtaining state variable samples for at least one recorded call from a passenger" In this stage, input sample data and output are obtained. This data, in this example, is a vector 20 It can be represented by a vector. • "Examples of elevator status variables received and at least one recorded call from a passenger. Examples of state variable for "vectors X such that X = [X1, X2 ... Xn]" It can be represented by the vector X1 = [X11, X12, ... X1m]. X1 and Xn The interval represents each elevator and a passenger's recorded call. The interval between X11 and X1m is 25. or instances of state variables for elevators from the elevator group or one It represents instances of the state variable for at least one recorded call from the passenger. • "Examples of control values ​​taken for each elevator in the elevator group", Y = [Y1, Y can be represented by a vector of vectors Y such that Y2, ... Yn], where Y1 = The lines are [Y11, Y12, ... Y1m]. Each elevator is represented between Y1 and Yn. Y11 30 Between Y1m and Y1m, there are examples of control values ​​for each elevator in the elevator group. It represents. 11 • For this example, the sequencing parameters for each elevator are the vector S = [s1, s2, ... sn]. It can be calculated as follows. • The order of the sequencing parameters for each elevator is s(r1) <= s(r2) <= ... <= s(rn) R can be defined as R = [r1, r2, ... rn] such that... • Examples of ordered state variables for elevators X' = [X(r1), X(r2) ... X(rn)] 5 will be. • The sequential control value of each elevator is Y' = [Y(r1), Y(r2), ... Y(rn)] will be. • The neural network will be trained with X' as the input layer and Y' as the output layer. Similarly, clarifying the method and system (100) for the control of a group of elevators 10 A simplified example is provided below for this purpose: • "Received elevator status variables (110) and at least one recorded call from a passenger "state variables", a vector of vectors X such that X = [X1, X2 ... Xn] It can be represented as follows: Here, X1 = [X11, X12, ... X1m]. Each interval between X1 and Xn... It represents an elevator and a passenger's recorded call. The distance between X11 and X1m is 15. status variables of elevators from elevator group (110) or a passenger's most It represents the state variables of a small, registered call. • For this example, the sequencing parameters for each elevator are the vector S = [s1, s2, ... sn]. It can be calculated as follows. • The order of the sequencing parameters for each elevator is s(r1) <= s(r2) <= ... <= s(rn) 20 R can be defined as R = [r1, r2, ... rn] such that... • Ordered state variables of elevators (150) X' = [X(r1), X(r2) ... X(rn)] will be. • By inputting the data 'X' into the neural network, the sequential control values ​​of each elevator are obtained. (170) gives the output; these values ​​will be in the form Y' = [Y(r1), Y(r2) ... Y(rn)]. 25 • De-ordering of Y' values, ensuring each elevator manufactured It creates the control values ​​(180). Sorted inputs for training [sorted state variable examples for elevators or The basic idea behind the use of state variables of elevators (110)] is training 12 It is necessary to present all permutations of possible outcomes [all inputs] during the process. without delay, it can indicate the most suitable elevator to be allocated to a passenger at each decision point. The goal is to obtain a neural network. The outputs of the neural network are, in most cases, the first (sorted) output [each [Sequential control values ​​for an elevator] are trained to show the optimal one. However, if this always yielded the optimal result, a neural network would not be needed; 5 The sequencing parameters could be used directly to control a group of elevators. The neural network output is the first-line control for determining the most suitable elevator to be assigned to a passenger. Not elevators with these values, but second or even higher-order control values. There are situations where it is stated that there is an elevator available. For example, as a demand parameter. In a scenario where "minimum waiting time for passengers" is taken into account: In many cases, 10 The elevator that offers the shortest wait time (i.e., the one in first place) is actually allocated. will be done. However, in some special cases, allocating this elevator to those already waiting may be necessary. This could extend the waiting time for another passenger. In such cases, this applies to all passengers. Suitable for assigning an alternative elevator that minimizes waiting time optimally. Maybe. 15

Claims

13 REQUESTS 1. It is a method for training a neural network to control a group of elevators, The method includes the following: − Examples of state variables for elevators from a group of elevators, for a passenger instances of the status variable for at least one recorded call and elevator 5 in question. Taking control value samples for each elevator in the group - Examples of elevator state variables and at least one recorded instance of the passenger in question. the correlational variables between the state variable instances of the call sorting parameters, state variable examples of elevators and words The subject is each of the status variable samples from at least one recorded call of the passenger. calculation for an elevator - at least one request that defines the expected conditions from the control value samples According to the parameter, each elevator has one of the aforementioned ranking parameters. determining the order − Examples of elevator state variables and the control value of each elevator are 15 the examples arranged in the specified order − An input training for the aforementioned ordered state variable examples of elevators as a set and examples of the sequential control values ​​of each elevator Training a neural network by inputting an output as a training set.

2. A non-volatile 20-bit neural network model trained according to the method in Claim 1. a memory device.

3. A method for checking a group of elevators, the method is... includes the following - the status variables of elevators (110) from a group of elevators and a passenger Retrieving the state variables of at least one recorded call 25 - the status variables of the elevators (110) and at least one registered passenger in question The ordering is the correlational variable between the state variables of the call. parameters of the elevators from the status variables (110) and the said for each elevator from the status variables of at least one recorded passenger call calculation 30 14 - at least one requirement that defines the conditions expected from a generated control value. According to the parameter, one of the sorting parameters in question is a sequence determination − the status variables of the elevators (110) in the specified order 5th place − The ordered state variables of the elevators (150), according to the method in claim 1 entering a trained neural network - Output of sorted control values ​​for each elevator from the trained neural network to be taken as - In order to generate the control values ​​for each elevator, 10 for each elevator De-ordering of the ordered control values ​​(170) - the elevator group, each elevator's manufactured control values Checked according to (180).

4. This is a method for checking a group of elevators according to claim 3, The feature is that each elevator group 15 is according to the manufactured control values ​​(180) of each elevator. The verification step is to ensure that at least one elevator is assigned to the passenger's registered call. and / or an elevator for the purpose of keeping at least one elevator in standby mode. It involves directing it to a location in the well.

5. A system for controlling a group of elevators (100), and the system includes the following 20 - the status variables of elevators from a group of elevators (110) and of a passenger configured to retrieve the state variables of at least one registered call receiver module - the status variables of the elevators (110) and at least one registered passenger in question The correlation variables between the state variables of the call are the ranking 25 parameters from the elevators' status variables (110) and the said for each elevator from the status variables of at least one recorded passenger call a calculation module structured for calculation - at least one requirement that defines the conditions expected from a generated control value. To determine the order of the sorting parameters according to the parameter. a determination module structured for (120) - the status variables of the elevators (110) in the specified order a sorting module structured for sorting (130) 5 - the ordered state variables of the elevators (150), according to the method in claim 1 to enter a trained neural network and from the trained neural network for each elevator configured to output the ordered control values ​​(170) neural network module (140) - In order to generate the control values ​​for each elevator, 10 for each elevator configured to deorder the sorted control values ​​(170) a sort removal module (160) and - the elevator group, with each elevator having its own manufactured control values. a controller configured to control according to (180).

6. A system for controlling a group of elevators according to claim 5 (100) 15 its characteristic is that the controller in question responds to at least one elevator's registered passenger call. an elevator for the purpose of allocation and / or keeping at least one elevator on standby by directing the elevator group to a location in the shaft, each elevator individually It is adapted to control according to the produced control values ​​(180).

Images