JP2964900B2 - Learning method of neural network for call assignment - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for learning a neural network in an elevator group management control device for allocating calls using a neural network.

[0002]

2. Description of the Related Art Conventionally, elevator group management control has mainly been a call allocation method using an evaluation function or a call allocation control by an expert system using fuzzy logic. A new method of assigning calls using a neural network 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, there is no need 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. 3-31173 "Elevator group management control device" Japanese Patent Application No. 5-2438
No. 17, "Method of Learning Neural Network for Elevator Call Assignment" and the like.

FIG. 4 shows an example of a neural network for call assignment. As shown in FIG. 4, the neural network NN for call assignment includes an input layer NR1 corresponding to an input pattern (elevator system state data), an output layer NR3 corresponding to an output pattern (assignment suitability),
It is composed of neurons of an intermediate layer NR2 located between the input layer and the output layer.

The input pattern uses various data representing the state of the elevator system (floor and direction of landing call, position and driving direction of each car, car call, load state, etc.) as parameters necessary for call assignment. , Into a form that can be input to a neural network, and the number of neurons in the input layer corresponds to the total number of parameters.

When input data is given to each neuron in the input layer, 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. In the output layer, there are neurons for the number of elevators, and for various input patterns, the neuron corresponding to the car that is optimal for allocation outputs “1” (maximum value), and the other neurons output “0”. Is learned in advance.

Accordingly, among the values of each neuron in the output pattern, a neuron that outputs a value closest to “1” (maximum value) indicates that it is optimal for allocation.
The unit corresponding to this neuron is selected as the assigned unit. Note that the number of neurons in the intermediate layer (one layer in the embodiment, but may be plural) is appropriately determined according to the number of elevators, the properties of the building, and the like.

Although not shown, a connection weight (synapse weight) indicating the strength of the connection between 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”.

Since this back propagation is well known and will not be described in detail, a learning sample (an input pattern and a desired output pattern corresponding to the input pattern, that is, a teacher signal are paired). An algorithm that compares the output pattern for the same input pattern with the teacher signal and corrects the connection weights to minimize the error. Initially, all weights are initialized (for example, random The input pattern of the learning sample is given to each neuron of the input layer. The output pattern at this time and the output pattern of the learning sample (teacher signal)
Then, using the difference (error), the values of the connection weights are sequentially corrected from the output layer side so that the difference becomes smaller.

If this is repeated using a large number of learning samples 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 learning input The same level of call assignment as the teacher signal can be performed not only for patterns but also for unknown input patterns.

[0012]

The learning of the neural network for assignment is performed using a large number of learning samples created in advance by simulation or the like, or learning samples created during operation of the elevator at an actual site. However, there are learning samples that are easy and difficult to learn depending on their contents, and there are some samples in which the output of the neural network does not match the teacher signal even after a certain amount of learning is completed. If such learning difficult samples are included in a large number of learning sample groups, it is necessary to repeat learning using the same learning sample group many times until the error converges. Had the problem of getting worse.

The present invention has been made in view of such problems, and has as its object to improve the learning efficiency of a neural network.

[0014]

In order to achieve the above object, according to the present invention, after learning a neural network with a learning sample group including a large number of learning samples, the learning sample is added to the neural network. Apply sequentially,
As a result, the ratio of the number of learning samples in which the output of the neural network matches the teacher signal to the total number of learning samples (adaptation rate) is calculated. A new learning sample group is created by duplicating and adding the learning sample to the learning sample group,
The learning of the neural network is performed again using this new learning sample group.

Further, as a result of learning again, when the above-mentioned matching rate rises from the previous value, the above procedure is repeated until the matching rate becomes larger than the predetermined value or falls below the previous value.

[0016]

According to the present invention, after learning using the learning sample group, only a learning sample in which the output of the neural network does not match the teacher signal is duplicated and added to create a new learning sample group. At the time of learning, learning samples whose outputs do not match are subjected to overlapping learning.

[0017]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a flowchart showing the procedure of the learning method according to the present invention. First, in step S1, N learning samples are created. This learning sample is a pair of an input pattern obtained by converting elevator system state data at the time of occurrence of a hall call into a form that can be input to a neural network, and a number assigned to the input pattern (or an actual response number), that is, a teacher signal. Although it can be created by simulating the operation of the elevator or during the operation of the elevator at an actual site, it is not directly related to the content of the present invention, and thus the detailed description is omitted.

Next, in step S2, a variable i and a relevance ratio Ri described later are initialized, and learning of the neural network is performed using the learning sample group created in step S1 (step S3).

After the learning is completed, in step S4, as shown in FIG. 2, N learning samples are sequentially applied to the neural network, and for each learning sample, the assigned number indicated by the output pattern and the assigned number indicated by the teacher signal. Then, it is determined whether or not the number matches the above-mentioned number.
The number n of learning samples that do not match among the learning samples is counted.

Then, in step S6, the value of the variable i is incremented, and in step S7, the relevance ratio Ri at this time is calculated. The matching rate is a ratio of the number of learning samples (N−n) in which the output pattern matches the teacher signal in step S4 to the total number N of learning samples, that is, Ri = (N −n) / N, and the closer this value is to 1, the higher the learning result.

In step S8, it is determined whether or not the value of the precision is equal to or greater than a predetermined value. If the value is equal to or less than the predetermined value, further learning is required, and the process proceeds to step S9. Determine whether it has been improved. If the precision is improved from the previous value, the process proceeds from step S9 to step S10, where the replica of the n learning samples whose output pattern and the teacher signal do not match with each other in the learning sample group is used as the original. A new learning sample group is added to the learning sample group, and step S3 and subsequent steps are repeated again. FIG. 3 shows how the new learning sample group is created.

As long as the precision is improved as compared with the previous time, the procedure of steps S3 to S10 is repeated assuming that the learning sample group has the effect of learning. If the precision is lower than the previous value, Since the effect of the learning sample group can no longer be expected, step S1
The process proceeds to step 1, and the learning sample group is discarded, and a new learning sample group is created again in step S1.

In this way, the above procedure is repeated, and when the relevance rate exceeds a predetermined value, the learning is terminated, and the neural network can be used for actual call assignment thereafter.

[0024]

According to the present invention, only learning samples whose neural network output does not match the teacher signal are duplicated and added. As a result, it is possible to prevent learning samples that are easy to learn from repeating learning more than necessary, and as a result, learning can be performed very efficiently.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating an example of a learning procedure according to the present invention.

FIG. 2 is a diagram for explaining calculation of a precision in the present invention.

FIG. 3 is a diagram for explaining how a new learning sample group is created in the present invention.

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

[Explanation of symbols]

 A1 to An 1st to nth operation control devices 1 hall call button 2 hall call signal 3 car information signal 4 assignment signal 10 group management device 11 input / output interface 12 learning sample creation means 13 traffic situation determination means 14 learning sample Storage means 15 neural network 16 assignment determination means 17 learning means

Continued on the front page (72) Inventor Mami Nakagawa 1-28-10 Sho, Ibaraki-shi, Osaka Prefecture Inside Fujitec R & D Co., Ltd. (72) Inventor Takeshi Midoriya 1-28-10 Sho, Ibaraki-shi, Osaka Stock Company Company Fujitec intra-technology research Institute (72) inventor Tanabe TomoAkira Ibaraki, Osaka Zhuang 1-chome 28 No. 10 No. stock Company Fujitec technology research Institute within the examiner Yuji Shimizu (58) investigated the field (Int.Cl. ^6, DB name ) B66B 1/18-1/20

Claims (2)

(57) [Claims] When a hall call occurs, various data representing the state of the elevator system at that time are converted into a form usable as an input pattern of a neural network and input, and an optimum car is selected from the output pattern. In the method for learning a neural network for call assignment of a group management elevator, the learning sample group comprising a large number of learning samples is used for learning the neural network, and then the learning sample is added to the neural network. Are sequentially applied, and as a result, the ratio (accuracy) of the number of learning samples in which the output of the neural network matches the teacher signal to the total number of learning samples (calculation rate) is calculated. Create a new learning sample group by duplicating and adding the learning sample that did not match to the learning sample group A learning method of a neural network for call assignment, wherein learning of the neural network is performed again using the new learning sample group. 2. When a hall call occurs, various data representing the state of the elevator system at that time are converted into a form usable as an input pattern of a neural network and input, and an optimum car is selected from the output pattern. In the method for learning a neural network for call assignment of a group management elevator, the learning sample group comprising a large number of learning samples is used for learning the neural network, and then the learning sample is added to the neural network. Are sequentially applied, and as a result, the ratio (the precision) of the number of learning samples in which the output of the neural network matches the teacher signal to the total number of training samples is calculated, and when the precision is equal to or less than a predetermined value, Create a new learning sample group by duplicating and adding the learning sample that did not match to the learning sample group Then, the learning of the neural network is performed again using the new learning sample group, and as a result, when the precision is higher than the previous value, the precision is equal to or more than a predetermined value or lower than the previous value. A learning method of a neural network for call assignment, characterized by repeating the above procedure up to this point.