JP2962174B2 - Elevator group control device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an elevator group management control system for allocating calls using a neural network, and more particularly to a system in which the specifications of each elevator differ or the situation temporarily changes. The present invention relates to an elevator group management control device that can always perform highly accurate assignment.

[0002]

2. Description of the Related Art Conventionally, elevator group management control is mainly based on a call assignment method using an evaluation function, but recently, call assignment is performed using a neural network modeled on a biological neural circuit. A new scheme has been proposed.

A neural network is a network imitating the human brain. A plurality of neural cell models (neurons) are connected in a complex manner, and the operation of each neuron and the form of connection between the neurons are properly determined to perform pattern recognition. The function and the knowledge processing function can be embedded. For example, "Nikkei Electronics" 1987
P115-P124 and 1 of August 10, issue (No. 427)
It is disclosed in the book "PDP model" published by Sangyo Tosho Co., Ltd. in February 989, etc. In particular, those in which neurons are arranged in a hierarchical structure can use an autonomous learning algorithm called "back propagation". There are features.

When this neural network is used, it is not necessary for a human to consider an assignment algorithm at all, and a decision system for deciding an optimum assigned car in response to various traffic conditions can be automatically generated. For example, Japanese Patent Application Laid-Open No. 1-275381, "Elevator Group Management Control Device" and Japanese Patent Application Laid-Open No. 2-52875, "Elevator Automatic Learning Management Device," Japanese Utility Model Laid-Open No. 3-59971, "Elevator group management control device", and the like.

As shown in FIG. 4, a neural network N for call assignment includes an input layer corresponding to an input pattern I, an output layer corresponding to an output pattern O, and an intermediate layer located between the input layer and the output layer. It consists of layers of neurons.

[0006] The input pattern I is composed of various data representing the state of the elevator system (floor and direction of landing call, position and operation direction of each car, car call, load state, etc.), and parameters required for call assignment. Is converted into a form that can be input to a neural network. For example, if a plurality of parameters m are used in n elevators, the input pattern I is represented by the input parameters Pxy (x = 1 to
n, y = 1 to m).

I = {P11, P12,... P1m}, {P21, P22,.
m ｛,..., Pn1, Pn2,... Pnm｝ In the input layer, there are a number of neurons corresponding to the total number of parameters of the input pattern I (the number of parameters m × the number of elevators n). When data is given, a signal is sequentially transmitted to the output layer, and as a result, a certain value is output from each neuron in the output layer.

[0008] In the output layer, there are neurons for the number of elevators, and for various input patterns, the neuron corresponding to the car which is optimal for allocation is "1".
And the other neurons are trained to output “0”. Accordingly, among the values (O1 to On) of the neurons of the output pattern O, the largest value of the neuron indicates the optimal unit, and the unit corresponding to this neuron is selected as the assigned unit.

The intermediate layer (one layer in the embodiment,
The number of neurons may be appropriately determined depending on the number of elevators, the properties of the building, and the like.

Although not shown, a connection weight (synapse weight) indicating the strength of connection of the neurons is set between the neurons. The connection weight is initially set to an appropriate value, but can be modified so that a more accurate call assignment can be performed using a learning algorithm called “back propagation”.

In this back propagation, the output pattern of the neural network is compared with the output target of a previously created teacher signal for the same input pattern, and the connection weight is corrected so as to minimize the error. First, all weights are initialized (for example, set to random values), and an input pattern of a teacher signal is given to each neuron of the input layer for learning. Then, the output pattern at this time is compared with the output target of the teacher signal, and using the difference (error), the values of the respective connection weights are sequentially corrected from the output layer side so as to reduce the difference.

If this is repeated using a large number of teacher signals until the error converges, the call assignment function at the same level as the teacher signal is automatically embedded in the neural network, and the input pattern for learning is input. In addition to the unknown input pattern, the same level of call assignment as the teacher signal can be performed.

[0013]

The learning of the neural network for assignment is performed based on data obtained in advance by simulation or the like or actual measurement data at an actual site. In general, elevators are symmetrical in each car. Therefore, the learning efficiency can be improved by utilizing the symmetry.

That is, as shown in FIG. 5, at the time of learning, a net of one size (the number of neurons in the input layer is m equal to the number of parameters, and the number of neurons in the output layer is 1)
If the nets of the same size are learned and the nets of the total number are configured based on the learned nets, the net to be learned can be small, and the time required for the learning can be shortened.

When the whole is learned by one neural network, only one output pattern for one input pattern (information of all units) can be learned in one situation (one assignment). When learning on the same number of nets, it is possible to learn the number of patterns in one situation. Therefore, if learning is performed using one net, convergence is quickened, and learning can be performed efficiently.

As described above, if it is assumed that each elevator has symmetry, learning of the neural network can be performed efficiently. Absent. For example, the service floor may be different depending on the number, or the specifications such as speed may be different. It is also conceivable that a certain car temporarily goes out of group control due to a failure or maintenance, and the number of elevators that can service the hall call changes.

In such a case, a neural network that has learned on the premise that each car has symmetry cannot make an appropriate call assignment. It may be assigned to a unit that is out of control.

To avoid this, a method of masking the output pattern with the service number pattern of the floor and selecting the maximum value from the number of units that can service the floor can be considered. However, at the time of learning, only the most suitable car is assigned as the assigned car, and the other car is not assigned (that is, either "1" or "0"), which is enemy to car assignment after the second car. I have not learned the order.

Therefore, in this method, if the masked car is the optimum car, it is impossible to determine which of the remaining cars is the next appropriate car. In addition, each value of the output pattern corresponding to each unit is not only 1 and 0,
In practice, various intermediate values may be output. However, since the values are not necessarily proportional to the allocation suitability of each car, appropriate allocation cannot be performed by this method.

Therefore, when learning the neural network,
It is conceivable to learn in advance all possible cases, such as when the service floor of each unit is different, or when a unit that is out of order for failure or maintenance is out of group management, but in order to do so, There is a problem that an enormous number of learning data is required, and a long time is required for learning.

[0021] The present invention has been made in view of the above-described problems, and is intended to perform optimal assignment according to the situation even when the specifications of each unit are different or the situation temporarily changes. It is an object of the present invention to provide a group management system that does not deteriorate learning efficiency.

[0022]

In order to achieve the above-mentioned object, the present invention comprises a determination means for determining whether or not each car is to be allocated to a call when a hall call is generated. The determination result is added as one of the parameters of the input pattern to the allocation neural network.

Further, if the number of elevators to be allocated is different, the traffic condition is different for the group management system, and if the information is not correctly given to the neural network, the optimum allocation may not be possible. Therefore, when the number of elevators is used for the calculation of the parameter, the calculation is performed only for the car currently assigned.

[0024]

In the elevator group management apparatus according to the present invention, when a hall call occurs, various data representing the state of the elevator system at that time are digitized and input to the neural network as an input pattern. At this time, it is determined at the same time whether or not each car is to be assigned, and the result is one of the parameters of the input pattern.

If the floor is set as an unstoppable floor, or if there is a car that cannot be serviced on that floor because it has been removed from group management for some reason, the above parameter of that car will be "0". , The above parameters of the other units are “1”. In the neural network, since learning is performed in advance so as not to assign to a car whose parameter is 0, assignment is performed only for a car whose parameter is 1.

[0026]

FIG. 1 is a block diagram showing the entire configuration of the present invention. Here, for convenience of explanation, an elevator to be controlled is A
The following describes a case where the number of elevators is three, but the present invention can be applied irrespective of the number of elevators.

In FIG. 1, reference numeral 1 denotes a hall call button provided on each floor (only one floor is shown and other floors are omitted), 2 is a hall call signal, and 3A to 3C are An operation control device for managing the operation of each of the units A to C, and the state of each of the units (car position, driving direction, traveling or stop, door open / closed state, car call, load, service floor, abnormality, etc.) It is a car information signal to represent.

Reference numeral 5 denotes a microcomputer which functions as a group management device. A means for registering a hall call of each floor based on data of the hall call signal 2 and the car information signal 4 read through the input / output interface 6. 7 and a function as determination means 8 for determining whether or not each car is to be assigned, and various data representing the state of the elevator system are converted into numerical values and converted into a form that can be input to a neural network. A function as the input pattern calculating means 9, a function as the neural network calculating means 10 for realizing the neural network by software, and a car which seems to be optimum from the output patterns of the neural network are selected and assigned to the hall call. Assignment means 1
1 and outputs the assignment result as an assignment signal 12.

Each of the operation control devices 3A to 3C controls the operation of the car so as to sequentially respond to the hall call assigned by the assignment signal 12 and the car call registered in the own car.

FIG. 2 shows an example of each parameter of the input pattern. As shown in FIG. 2, each parameter of the input pattern is calculated by the input pattern calculation means 9 based on various data representing the state of the elevator system, such as information on a hall call and a car information signal of each car.

In the figure, PA1 is a parameter indicating whether or not the car A is to be assigned. For example, when the car A is out of the group management control due to a failure or the like, or when the landing call generation floor is If it is set as a non-stop floor,
Since it is impossible for Car A to service the hall call, PA1 = 0, and conversely, PA1 = 1 when it is to be assigned.

Similarly, PB1 and PC1 are parameters indicating whether or not each of the B-unit and the C-unit is to be assigned, and is also “1” when they are to be assigned, and “0” when they are not to be assigned. ".

PA2 to PC2 determine the floor difference between the floor / direction at which the hall call occurs and the car position / driving direction of each car.
PA3 to PC3 are parameters representing the load state of each unit, PA4 to PC4 are parameters representing the distribution of the position of each unit, and PA5 to PA5. PC5 is a parameter indicating the number of calls each of which is assigned to each car. In addition, an input pattern is created by appropriately using other parameters deemed necessary for call assignment.

Each of these parameters is normalized to the range of 0 to 1. For example, in the case of PA2, PA2 = floor difference / 2 (total floor number-1). Then, it can be obtained by dividing the current value by the maximum value that can be taken, such as PA3 = current load / rated load, but this normalization processing is not always necessary.

If the number of elevators to be allocated is different, the traffic condition is different for the group management system. If this information is not correctly given to the neural network, there is a possibility that the optimum allocation cannot be performed. For this reason, when the number of elevators is used in the calculation of the above parameters, the calculation is performed using the number of vehicles currently assigned.

For example, when calculating the parameters of the distribution status of elevators, the distance between each car (substantial distance,
Alternatively, the variance value is obtained from the time length represented by the traveling time or the like), but at this time, the calculation is performed only for the car to which the current assignment is to be performed using the above-mentioned parameters PA1 to PC1. Similarly, when the number of elevators itself is one of the input parameters, the number of elevators to be allocated is the input parameter itself. In this way, by constantly monitoring the number of vehicles to be allocated at the present time by using the parameters PA1 to PC1, the number of vehicles can be used for calculating input parameters.

FIG. 3 is a flowchart showing a procedure of the assignment according to the present invention. In FIG. 3, first, it is determined in step S1 whether or not there is a newly generated hall call, and if there is, the process proceeds to step S2. In step S2,
Information such as the status (position, driving direction, etc.) of each unit and the floor / direction of the hall call is read.

In step S3, it is determined from the hall call and the data on the status of each car whether or not each car can be serviced for that hall call, that is, whether there is a car which is not to be allocated. Among the parameters, that is, PA1 to PC1, the parameter of the corresponding car is set to “0”, and the parameter of the other car is set to “1”. If there is no corresponding car, the parameters PA1 to PC1 are all "1".

In step S6, other parameters are calculated to create an input pattern. The weight value representing the degree of connection between each unit of the neural network is
Since it is set in advance by learning, when the input pattern is determined, the output of each layer can be sequentially obtained by calculation, and this processing is performed in step S7. Then, in step S8, an optimum car is selected from the output pattern value, and this is assigned to a hall call.

In the learning, when PA1 to PC1 in the input pattern, that is, the parameter indicating whether or not to be assigned is "0", the value of the output pattern is determined regardless of the values of the other parameters. A teacher signal is always created with "0", and when this parameter is "1", a teacher signal is created so that the assignment is determined by the values of other parameters as in the past, and learning is performed in advance. .

In this way, the parameters PA1 to PC1
Among them, the output of the car whose number is "0", that is, the car whose assignment is impossible for any reason is always "0", so that the call is not erroneously assigned to such car. Moreover, since the number of parameters is increased by one, efficient learning using the symmetry of the elevator can be performed using only the neural network corresponding to one vehicle, as in the related art.

[0042]

According to the present invention, a call can be serviced for various reasons, such as when a certain floor is set as a non-stop floor, or when a certain unit is temporarily out of group management due to failure or maintenance. Since all impossible cases are collected into one parameter, the parameter is set to "0"
It is possible to prevent a call from being assigned to a car that cannot be serviced for various reasons simply by adding simple learning to prevent a call from being assigned to a car corresponding to.

Furthermore, since only one parameter is added and the content of the learning is simple, the conventional learning efficiency is hardly impaired.

[0044]

[Brief description of the drawings]

FIG. 1 is a block diagram showing the overall configuration of the present invention.

FIG. 2 is a diagram showing an example of each parameter of an input pattern according to the present invention.

FIG. 3 is a flowchart illustrating an example of an assignment procedure according to the present invention.

FIG. 4 is a diagram showing an example of a neural network for call assignment.

FIG. 5 is a diagram illustrating an example of a configuration of a neural network during learning.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 hall call button 3A-3C A-C operation control device 4 car information signal 5 microcomputer 6 input / output interface 7 hall call registration means 8 determination means 9 input pattern calculation means 10 neural network calculation means 11 assignment means

Claims (2)

(57) [Claims] When a plurality of elevators are put into service on a plurality of floors and a hall call occurs, various data representing the state of the elevator system at that time are converted into a form that can be used as an input pattern of a neural network. In the elevator group management control device, which selects the most suitable car from the output pattern and assigns it to the hall call, the elevator group management control device, for each elevator unit, generates the call when the hall call occurs. And a determination means for determining whether or not to make an allocation target, and the determination result for each car is added as one of the parameters of the input pattern. 2. The method according to claim 1, wherein, of the various parameters of the input pattern, the parameters calculated using the number of elevators are determined by using the number of elevators to be assigned by the determination unit. Item 3. An elevator group management control device according to Item 1.