JP3060879B2 - Construction of a Neural Network for Call Assignment and Its Learning Method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a neural network for allocating hall calls in a group control elevator, and more particularly to a configuration and a learning method for improving the learning efficiency and allocation performance.

[0002]

2. Description of the Related Art Conventionally, group management control of elevators has mainly been a call allocation method using an evaluation function or a call allocation control using 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 complicated manner, and the operation of each neuron and the connection form between the neurons are determined in a well-defined manner. 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, and 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. Further, examples used for estimating the waiting time include JP-A-2-286581 “Elevator group management control device” and JP-A-5-246633 “Elevator group management control device construction method”.

FIG. 3 shows an example of a neural network for call assignment. As shown in FIG. 3, 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 the 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, ie, 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.

The neural network for estimating the waiting time has the same configuration as described above. Various data representing the state of the elevator system at the time of the occurrence of a hall call are used as input patterns, and the actual waiting time for the pattern is used as a teacher signal. When learning is performed in the same manner as described above using the learned learning sample, the same level of waiting time as the teacher signal can be predicted for an unknown input pattern.

[0013]

The assignment of a hall call is closely related to the prediction of waiting time. For example, if the state of the elevator system at the time when the hall call occurs is regarded as "problem", the optimum assignment at that time is considered. It can be considered that the car has an "answer" and the predicted waiting time (response time) of each car for the hall call has a relationship like a "hint".

[0014] By the way, the neural network imitates the structure of the human brain, and replacing the mechanism of the operation of the neural network with the function of the human brain seems to be significant.

For example, when learning a neural network for call assignment, the method of learning with a learning sample in which the state of the elevator system and the assigned car (assignment suitability) are paired as in the past is a problem for humans. It is similar to presenting only the answer and learning it. Therefore, in this method, it is difficult to understand the process leading to the answer, and not only takes a lot of time for learning, but also there is a risk of learning how to give the answer with a misleading interpretation.

On the other hand, Japanese Patent Laid-Open Publication No. Hei 5-246633 mentioned above discloses a method of efficiently constructing a neural network for call assignment, which is to obtain an accurate teacher signal such as a waiting time for a hall call in the first stage. It is described that information that can be learned is learned, and assignment of a hall call is learned in the second and subsequent stages.

However, when this method is replaced with the function of the human brain in the same manner as described above, learning the waiting time first, and then learning the assigned machine, is based on the method of creating hints for the problem (which leads to the answer). This is similar to having the process learn first and then learn how to derive the answer. Therefore, in this method, there is a risk that the process is emphasized more than giving a correct answer. Further, since learning is performed in two stages, there is a problem that a certain amount of learning time is required.

[0018]

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and it is clearly more efficient for the human brain to simultaneously learn both a hint and an answer to a problem as compared to the above two methods. Since it is considered to be a learning method, it can be expected that learning the waiting time and the assigned car at the same time leads to an improvement in the learning efficiency and the assignment performance in the neural network.

For this reason, the call assignment neural network according to the present invention increases the number of neurons in the output layer by only increasing the number of neurons in the output layer. And the learning of the neural network is performed by simultaneously providing the teacher signal of the assigned unit and the teacher signal of the estimated waiting time to the same input pattern. Features.

[0020]

According to the present invention, the neural network for call assignment is provided with two teacher signals, that is, a waiting time and an assigned number, at the same time, and learning is performed. Knowledge that reflects both the "process" will be automatically embedded.

[0021]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing a connection configuration of a neural network according to the present invention. As is clear from FIG. 1, in the present invention, an output layer corresponding to a predicted waiting time of a hall call is provided as an output layer, in addition to an output layer corresponding to an assigned car (assignment suitability). That is, the number of neurons in the output layer is increased from the configuration of the conventional neural network, and learning can be performed by simultaneously providing the assigned input device and the waiting time teacher signal to the same input pattern.

The learning procedure of the present invention in the neural network having such a configuration will be described with reference to the flowchart of FIG.

Before starting the learning, it is necessary to prepare a large number of learning samples in advance. Since the method of creating the learning sample is not directly related to the contents of the present invention, the detailed description is omitted. However, the learning sample used in the present invention is the learning sample used in learning the conventional call assignment neural network. It is the same as the sample or the learning sample used for learning the waiting time prediction neural network, and can be created in the same manner.

That is, a learning sample is created by a simulation operation of an elevator or during operation of an elevator at an actual site, while a learning sample for call assignment inputs an elevator system state at the time of a landing call to a neural network. The input pattern is converted into an input pattern, and the assigned pattern (or actual response number) for the input pattern is used as a teacher signal. The input pattern and the teacher signal are paired. Similarly, the learning sample for waiting time prediction converts the elevator system state at the time of the occurrence of a hall call into a form that can be input to the neural network to obtain an input pattern, and waits for the input pattern (predicted waiting time or each unit). The time required for the actual response) is used as a teacher signal, and this input pattern and the teacher signal are paired.

In the present invention, since the teacher signal of the assigned unit and the teacher signal of the waiting time are given at the same time during learning, one set of a learning sample for the call assignment and a learning sample for the waiting time for the same input pattern. As a learning sample or from the beginning,
It is preferable to create a learning sample in which the assigned unit and the waiting time are used as teacher signals for the same input pattern.

When the number of learning samples thus created reaches a predetermined number or more, first, in step S1, a variable N representing the number of learning samples is initialized, and in step S2, the value of N is incremented.

In step S3, for the first learning sample, an input pattern and an output target (teacher signal of assigned car and teacher signal of waiting time) are set.
In step S4, the output of the neural network for the input pattern is calculated. In step S5, an error from the output pattern is calculated from the output pattern,
In step 6, the weight between the intermediate layer and the output layer is sequentially corrected from the error value.

Similarly, the error of the hidden layer is calculated in step S7, and the weight between the input layer and the hidden layer is sequentially corrected in step S8. Then, the above procedure is repeated until the processing is completed for all the samples via step S9, and further, the above procedure is repeated until the convergence of the error is completed via step S10, thereby completing the learning.

[0029]

According to the present invention, learning is performed by simultaneously giving a wait time and an assignment teacher signal, so that it is the same as learning both a hint and an answer to a problem at the same time. It is possible to create a good neural network for assignment. Also, by adding an output layer for waiting time prediction, one neural network can be used for both assignment and waiting time prediction. It is also effective in shortening and saving memory used.

[Brief description of the drawings]

FIG. 1 is a diagram showing a connection configuration of a neural network according to the present invention.

FIG. 2 is a flowchart illustrating a learning procedure according to the present invention.

FIG. 3 is a diagram showing an example of a neural network for call assignment.

[Explanation of symbols]

 NN Neural network for call assignment NR1 Input layer neurons NR2 Hidden layer neurons NR3 Output layer neurons

──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Takeshi Midoriya 1-28-10 Sho, Ibaraki-shi, Osaka Prefecture Inside Fujitec R & D Co., Ltd. (72) Inventor Tomoaki Tanabe 1-28 Shojo, Ibaraki-shi, Osaka No. 10 Inside Fujitec R & D Co., Ltd. (72) Inventor Naoki Ota 1-28-10 Sho, Ibaraki-shi, Osaka Prefecture Inspector in Fujitec R & D Co., Ltd. Yuji Shimizu (56) References JP-A-2-286581 ( JP, A) (58) Field surveyed (Int. Cl. ⁷ , DB name) B66B 1/18-1/20

Claims (2)

(57) [Claims] [Claim 1] as input patterns various data representing the state of the elevator system at the time of hall call occurs, allocates No. machine (assignment aptitude) and the output pattern, the neural network for allocation call group management elevator, the call Output layer of the neural network for assignment
By increasing only the number of Ron, as its output layer,
A configuration of a call assignment neural net, characterized in that an output layer corresponding to a predicted waiting time of a hall call is provided separately from an output layer corresponding to an assigned car. Wherein an input pattern various data representing the state of the elevator system at the time of hall call occurs, allocates No. machine (assignment aptitude) and the output pattern, the group management elevator call assignment for neural network output layer
By increasing only the number of neurons, its output layer and
In addition to the output layer corresponding to the assigned
Assignment with an output layer corresponding to the estimated waiting time
Oite learning process of the neural network for Te, for the same input pattern, the call assignment neural and teacher signal assignment Unit teacher signal and latency
A learning method of a neural network for call assignment, characterized in that learning is performed by simultaneously providing the learning to an output layer of the net.