JPH06329352A - Elevator operation demand anticipating device - Google Patents

Abstract

(57) [Summary] [Purpose] An elevator operation demand forecasting device that can predict traffic flow in response to daily minute fluctuations and predict the occurrence of hall calls and car calls that are highly correlated with specific times and events. Offer. A demand forecasting determination device for an elevator, comprising a demand forecasting determination device 3 for forecasting demand for elevator operation after a short time using stored past traffic information and present traffic information 3 is a traffic information collection unit 9 that collects traffic information such as hall call information and car call information, a long-term information storage unit 12 and a temporary information storage unit 11 that classify and store the collected traffic information, and a long-term information storage unit 11. It includes a learning unit 13 that performs required learning based on the storage content of the information storage unit 12, and a demand prediction unit 14 that changes the learning result of the learning unit 13 according to the storage content of the temporary information storage unit 11.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an elevator operation demand prediction device for predicting elevator operation demand, and more particularly, to a traffic flow after a short time in order to sufficiently cope with daily minute traffic fluctuations. The present invention relates to an elevator operation demand prediction device that predicts and generates an operation prediction output.

[0002]

2. Description of the Related Art Conventionally, as an elevator operation demand forecasting device, some devices have been already developed, and among them, the following devices are included.

First, the device disclosed in Japanese Patent Laid-Open No. 59-153770 detects and records the occurrence times of various daily traffic patterns, and based on the records, present or future various traffic patterns. The time of occurrence of is predicted.

Next, the device disclosed in Japanese Patent Laid-Open No. 59-48369 measures the number of passengers in each floor in the vertical direction, learns the pattern obtained at the time of measurement as a traffic flow mode, and determines the past number. The required traffic flow mode is determined from the number of passengers in the vertical direction during a minute, and different operation control is performed according to the mode determined here,
When a mode that does not match the default mode occurs, it is stored as a new traffic flow mode, and at the same time, a new program that matches this new traffic flow mode is automatically created so that any traffic flow mode changes It is designed so that an operation system that can be adapted by itself can be created.

Subsequently, the device disclosed in Japanese Patent Laid-Open No. 57-62179 stores the car calls and the hall calls that have already been serviced in a table, respectively, and learns the call occurrence probability based on the storage result. Is used to predict the occurrence of subsequent calls.

The device disclosed in Japanese Unexamined Patent Publication (Kokai) No. 60-106774 continues to store data such as the demand for car calls generated in response to each hall call, and based on this stored data, By predicting a derived car call that is created by deriving from a hall call, serviceability is improved.

Next, the device disclosed in Japanese Patent Laid-Open No. 60-232378 tests the significance of the derived car call between the expected value by learning and the demand actually generated, and only truly effective data is obtained. By using it, the performance is improved.

The device disclosed in Japanese Unexamined Patent Publication No. 61-130188 continues to calculate the ratio of the average number of car calls generated in the past predetermined time zone to the current number of car calls generated in the same time zone as the past average. The frequency of hall calls is predicted by multiplying the number of hall calls generated.

[0009]

However, among the known operation demand forecasting devices for each elevator, the device disclosed in Japanese Patent Laid-Open No. 153770/1984 is a device for generating various daily traffic patterns (traffic flow). Adjust the time for each building,
Although it is possible to accurately predict the time of occurrence of the various traffic patterns, there are different traffic patterns for each building, and by just setting some traffic patterns in advance,
There is a problem in that it cannot be used for various purposes.

Further, the device disclosed in the above-mentioned Japanese Patent Laid-Open No. 59-48369 is intended to solve the above problems, and when a traffic flow (traffic pattern) that has not been set is observed, The traffic flow is learned as a new traffic flow mode, and a control program suitable for this traffic flow mode is sought. Thereafter, this traffic flow is recognized as a new traffic flow mode, and control suitable for this mode is performed. It is possible to do. The device is excellent in that it captures the characteristics of the traffic flow within a certain period of time, but when there is a slight change in the start time or a change in the traffic volume, the change occurs. There is a new problem that cannot be dealt with correctly. That is, although the device can be used for various purposes by the function of creating a new traffic flow mode when the range of the fluctuation is extremely small or the fluctuation is almost negligible, it can be used for various fine daily operations. To deal with all the fluctuations, a huge number of traffic flows must be created in advance, which is practically impossible.

Further, the above-mentioned JP-A-57-62179,
JP-A-60-106774, JP-A-60-23237
No. 8, JP-A-60-232378, JP-A-61-113.
Although each of the devices disclosed in No. 0188 predicts the generation of a derived car call based on the generation of a call, a car call, etc. is generated before a hall call is generated or a hall call on a specific floor is generated. Is not considered at all, and the prediction is still insufficient.

The present invention solves the above-mentioned problems and inconveniences, and its purpose is to predict traffic flow in response to daily minute changes and to identify specific events occurring at specific times. An object is to provide an elevator operation demand prediction device capable of predicting the occurrence of hall calls and car calls from the correlation.

[0013]

In order to achieve the above object, the present invention provides a demand for predicting demand for elevator operation after a short time using stored past traffic information and present traffic information. In an elevator demand prediction apparatus including a prediction determination device, the demand prediction determination device classifies the traffic information collection unit that collects traffic information such as hall call information and car call information, and the collected traffic information. And a long-term information storage unit and a temporary information storage unit that respectively store,
And a demand predicting unit that changes the learning result obtained by the learning unit according to the stored contents of the temporary information storage unit, the learning unit performing required learning based on the stored contents of the long-term information storage unit. 1 means are provided.

In order to achieve the above-mentioned object, the present invention provides a demand forecast judging device for forecasting demand for elevator operation after a short time by using stored past traffic information and present traffic information. In the elevator demand forecasting device, the demand forecasting determination device collects traffic information such as hall call information and car call information, and temporary information in the collected traffic information. The neural network includes a temporary information storage unit for storing and a neural network, and the neural network stores long-term information in the collected traffic information for a long period of time, and performs necessary learning based on the stored long-term information. And a second means for changing the learning result according to the contents stored in the temporary information storage unit.

[0015]

According to the first means, the traffic information collecting unit is
Collects hall call information, car call information, and traffic information such as the number of people in the car and the number of people getting on and off, and at the same time, the general information collection unit collects information related to other traffic flows as necessary. The collected information is classified, then sent to the temporary information storage unit and the long-term information storage unit, and stored and stored therein. The learning unit learns the correlation and characteristics of each event based on the information stored in the long-term information storage unit. The demand prediction unit predicts an event that occurs after a short time based on the current storage information in the temporary information storage unit and the correlation or feature of each event obtained by the learning result of the learning unit. In this prediction, not only is the prediction of whether the stored information in the temporary information storage unit conforms to a certain pattern taken into consideration, but also the occurrence time and occurrence interval of each event, the size of each event, etc. are taken into consideration. Prediction is made.

Therefore, according to the first means, it is possible to predict the most suitable and accurate traffic flow at each point of time, taking into consideration minute daily fluctuations. It is possible to predict hall calls and car calls from the correlation in.

According to the second means, the traffic information collecting unit collects hall call information, car call information, and traffic information such as the number of people in the car and the number of people getting on and off, and at the same time, if necessary. The general information collection unit collects other information related to traffic flow, and the collected information is classified into long-term information and temporary information and then sent to the neural network and the temporary information storage unit. Are stored and stored therein. The neural network learns the correlation and characteristics of each event based on the stored long-term information, and the correlation between each event obtained by the learning and the currently stored information in the temporary information storage unit. The phenomenon occurring after a short time is predicted based on the characteristics and the characteristics, and the mode of prediction is almost the same as the mode of prediction described in the first means.

As described above, according to the second means, in addition to the same operation as that obtained by the first means, the configuration of the demand forecast determination device can be simplified.

[0019]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing the outline of the construction of an elevator operation demand forecasting apparatus according to the present invention.

In FIG. 1, 1 is a hall call collecting unit, 2
Basket call information collection unit, 3 is a demand determination device, 4 is an external information source, 5 is a group management control device, 6 is an elevator control device,
7 is an elevator car, 7 ₁ is a call button inside the car, 8 is a hall call button, 8 ₁ is a hall call button on the first floor, 8 _n is a hall call button on the nth floor, 9 is a traffic information collection unit, and 10 is general information collection. Section, 11 is a temporary information storage section, 12 is a long-term information storage section, 1
3 is a learning unit, and 14 is a demand prediction unit.

[0022] Then, the hall call collecting unit 1 is coupled to a hall call button 8 _n of the first floor of the hall call buttons ₈₁ through n floors to each other, car call information collecting unit 2 is coupled to the elevator control device 6. The elevator controller 6 is coupled to an in-car call button 7 ₁ located in the elevator car 7. The demand determination device 3 includes a traffic information collection unit 9 and a general information collection unit 10
It includes a temporary information storage unit 11, a long-term information storage unit 12, a learning unit 13, and a demand prediction unit 14, and these units 9 to 13 are coupled to each other as shown in the figure. The external information source 4 is connected to the general information collecting unit 10, and the group management control device 5 is connected to the demand forecasting unit 14. The hall call collecting unit 1 is also connected to the traffic information collecting unit 9, and the car call information collecting unit 2 is also connected to the traffic information collecting unit 9.

The elevator operation demand forecasting device having the above-described configuration operates roughly as follows.

[0024] First, hall call collecting unit 1, a hall call button 8, i.e., collect button information obtained in the first floor hall call button ₈₁ to n floor hall call button 8 _n, hall calls were collected here The information is supplied to the traffic information collecting unit 9. Car call information collecting unit 2, the position and car call buttons 7 ₁ obtained in button information such as information of the car 7 (hereinafter, referred to as the information cage call information) to collect, car calls or collected here Information Is also supplied to the traffic information collecting unit 9. On the other hand, external information source 4
Collects general information related to the elevator and supplies the collected general information to the general information collecting unit 10. next,
The various information supplied to the traffic information collection unit 9 and the general information collection unit 10 is temporary storage information suitable for temporary storage or long-term storage information suitable for long-term storage. After being classified according to the information content such as or the like, they are separately sent to the temporary information storage unit 11 and the long-term information storage unit 12 and stored and stored respectively. Then, the learning unit 13
Learning is performed based on the long-term storage information stored and stored in the long-term information storage unit 12, and the learning result is supplied to the demand forecasting unit 5. Details of the learning in the learning unit 13 will be described below. The demand prediction unit 5 selects the most suitable traffic flow from the learning results based on the temporary storage information stored and stored in the temporary information storage unit 11, and selects the selected traffic flow as an operation prediction output. It is supplied to the management control device 5. The management control device 5 controls the service elevator for hall calls and the like based on the operation prediction.

Next, FIG. 2 is a block diagram showing a first embodiment of an elevator operation demand forecasting apparatus according to the present invention, in which the learning unit 13 shown in FIG. And a correlation table.

In FIG. 2, reference numeral 15 is a correlation calculation unit, 16 is a correlation table, 17 is a clock unit, and other components shown in FIG.
The same components as the components shown in FIG.

The input of the correlation calculation unit 15 is connected to the temporary information storage unit 11 and the long-term information storage unit 12, and the output is connected to the correlation table 16. The output of the correlation table 16 is coupled to the input of the demand prediction unit 14, and the clock unit 17 is connected to the temporary information storage unit 11 and the long-term information storage unit 12.
Combined with and.

By the way, in this embodiment, the correlation and characteristics of each event are obtained, and the operation thereof will be described in detail below.

First, the traffic information collecting unit 9 receives the hall call information collected by the hall call collecting unit 1 and the car call information collected by the car call information collecting unit 2, and receives the information according to the contents. After the classification, they are supplied to the temporary information storage unit 11 and the long-term information storage unit 12, respectively, and stored and stored therein. In this case, the temporary storage unit 11 supplies the supplied information and the time information from the clock unit 17 indicating the generation time of the information to the correlation calculation unit 15. Long-term information storage unit 1
2 stores and stores the supplied information together with the time information indicating the generation time of the information from the clock unit 17.

Next, the correlation calculation unit 15 calculates the long-term correlation of events from the stored contents of the long-term information storage unit 15. What is meant by the correlation of long-term events here? For example, if there is a large number of passengers from the lobby floor around 8:30 on Tuesday, the number of passengers from a specific floor will increase around 10:30 on the same day. The relationship is written and stored in the correlation table 16 as a correlation coefficient or an occurrence probability between each event. At the same time, the correlation calculation unit 15 calculates a temporary correlation between events from the stored contents of the temporary information storage unit 11, and also writes and stores it in the correlation table 16. Here, the correlation between temporary events means, for example, predicting the traffic volume 10 minutes before, 5 minutes before, and 5 minutes after the current traffic volume.

Subsequently, the demand predicting unit 14 finds the one suitable for the current traffic situation received from the temporary information storage unit 11 from the stored contents written in the correlation table 16, and the correlation obtained here. The operation demand is predicted based on the contents of the relationship table 16, and the operation demand prediction output is supplied to the group management control device 5.

As described above, according to the present embodiment, the information contents stored and stored in the long-term information storage unit 12 are periodically changed,
It is possible to make a prediction based on a global correlation and to make a prediction based on a temporary correlation of events by the information in the temporary information storage unit 11.
By correcting the periodic prediction according to the current situation, it becomes possible to make a detailed and accurate prediction.

Further, in general, calculation of the correlation requires enormous amount of calculation by a large-scale computer, but by calculating the correlation, what kind of event It is possible to numerically grasp the deep relationship between them, and it is easy to apply the correlation learned in one building to the correlation in another building, and it is easy to point out wrong learning.

Continuing with FIG. 3, the embodiment shown in FIG.
It is explanatory drawing which shows the case applied to the prediction of traffic volume.

In FIG. 3, (a) shows a case where the traffic volume situation at time t2 is predicted from the traffic volume situation up to time t1, and (b) is stored in the long-term information storage unit 12. It is an example of various types of stored traffic data. The horizontal axis of each figure is time t, and the vertical axis is traffic volume. In addition, the same components as those shown in FIG. 1 are denoted by the same reference numerals.

Referring to FIG. 3, in order to predict the traffic volume e at the time t2 when the traffic has not yet arrived, the current time t is used.
The traffic volume d of 1 and the past traffic volumes a, b, c are used, and the past traffic volume transition data having the transitions corresponding to the traffic volume transitions a, b, c, d have already been The traffic volume transition data stored in the long-term information storage unit 12 is selected, and the situation of the traffic volume e at the time t2 is predicted based on the selected traffic volume transition data. For this prediction, the measured value of the traffic volume at time t2 is f
And when it is different from the predicted value e,
The traffic volume transition data in which the traffic volume at time t2 is e is used as correction data, and the stored contents of the long-term information storage unit 12 are corrected. In addition, the correlation based on the data stored in the temporary information storage unit 11 is based on the past and present traffic volumes of the transitions a, b, c, and d in order to correct the daily deviation. The forecast of the traffic change data stored in 12 is corrected.

In the embodiment shown in FIG. 2 and each embodiment described below, the external information source 4 and the general information collecting unit 10 are omitted, and the external information source 4 and the general information collecting unit are omitted. 10 does not necessarily need to be used, but in the general information collection unit 10, for example, the cycle of traffic lights in front of the entrance of the building, the operation (arrival) time of trains and / or buses, conference events, etc. By storing and storing information related to the elevator, such as the start and end times of the lunch break, it becomes possible to more accurately predict the operation demand.

Next, FIG. 4 is a block diagram showing a second embodiment of the elevator operation demand prediction apparatus according to the present invention, in which the learning unit 13 shown in FIG. 1 is replaced with a multiple regression analysis unit. It is composed of a regression equation table.

In FIG. 4, 18 is a multiple regression analysis unit, and 19 is
Is a regression equation table, and the same components as those shown in FIG. 2 are denoted by the same reference numerals.

The input of the multiple regression analysis unit 15 is connected to the temporary information storage unit 11 and the long-term information storage unit 12, and the output is connected to the regression equation table 19. The output of the regression equation table 19 is coupled to the input of the demand prediction unit 14.

Here, in this embodiment, multiple regression analysis is performed from the traffic volume transition data as shown in FIG. 3B to obtain the regression equation, and the traffic volume is predicted by the obtained regression equation. The operation will be described below.

The traffic information collecting unit 9 classifies the hall call information and the car call information according to the contents and stores them in the temporary information storage unit 11 and the long-term information storage unit 12, respectively. Also in this case, the temporary information storage unit 11 stores the supplied information,
The multiple regression analysis unit 1 uses the time information indicating the generation time of the information.
8, and the long-term information storage unit 12 stores and stores the supplied information together with the time information indicating the generation time of the information.

The multiple regression analysis unit 18 includes a long-term information storage unit 15.
Performs a multiple regression analysis of long-term events from the stored memory contents of
A regression equation is obtained from the analysis result, and is written and stored in the regression equation table 19.

The demand forecasting unit 14 finds a regression equation suitable for the current traffic situation received from the temporary information storage unit 11 from the stored contents written in the regression equation table 19, and obtains the regression equation obtained here. Predict operating demand based on In this case, if the regression equation is provided with a correction term that corrects the daily deviation based on the current and past several points of data, more accurate prediction can be performed. Further, the operation demand prediction output is supplied to the group management control device 5 as in the previous example.

In this embodiment, the prediction is performed by the regression equation by the multiple regression analysis, but the Kalman filter may be used instead of the regression equation.

As described above, also in the present embodiment, the regression equation is corrected by comparing the prediction result by the multiple analysis / analysis unit 18 with the actual measurement result from the temporary information storage unit 11, and the daily deviation due to the temporary data. If you try to fix
It is possible to accurately predict the operation demand that is suitable for each building.

Next, FIG. 5 is a block diagram showing the third embodiment of the elevator operation demand forecasting apparatus according to the present invention. The traffic information collecting unit 9 and the temporary information storage shown in FIG. Each part except the part 11 is configured by a neural network.

In FIG. 5, reference numeral 20 denotes a neural network, and other components that are the same as those shown in FIG. 2 are designated by the same reference numerals.

Then, the neural network 20 is
The input is connected to the traffic information collection unit 9, the clock unit 17, and the temporary information storage unit 11, and the output is connected to the group management control device 5. In this case, since the neural network 20 has the learning function and the information storage function by itself, the configuration thereof is extremely simplified as compared with the first and second embodiments. have.

The present embodiment having the above-mentioned configuration operates as follows.

First, the traffic information collecting unit 9 collects the hall call information and the car call information as in the first and second embodiments, and stores the traffic information collected here for a long time. Information is classified into information and temporary storage information. Then, the long-term storage information is supplied to the neural network 20, and the temporary storage information is supplied to the temporary information storage unit 11. The clock unit 17 supplies the current time information to the neural network 20 and the temporary information storage unit 11. At this time, in the neural network 20, a short time t _n
Assuming that the subsequent traffic volume is predicted, the neural network 20 uses the current traffic volume, some past traffic volumes stored in the temporary information storage unit 11, and the current time information. A predicted value of traffic volume after the short time t _n is output, and the predicted traffic volume is supplied to the group management control device 5. Then, when the time elapses and the time t _n at which the prediction is performed is reached, the neural network 20 causes the number of traffic volumes used for the prediction stored in the temporary information storage unit 11 to be input, and Time t _n
Learning is performed by using the actually measured traffic volume obtained as a teacher signal, and the learning result is stored.

As described above, according to this embodiment, the first
Not only can the functions similar to those of the first embodiment and the second embodiment be achieved, but the use of the neural network 20 can simplify the configuration of the riding prediction determination device 3.

By the way, the neural network 20 has the property that the predictive predictive value is not good until the learning state converges. As in the third embodiment, only one neural network 20 is used. Therefore, in the case where learning and prediction are alternately executed, it may not be possible to obtain a sufficient correct answer rate.

Next, FIG. 6 is a block diagram showing a fourth embodiment of the elevator operation demand forecasting apparatus according to the present invention, in which two sets of neural networks are used for forecasting and learning, and forecasting and learning are performed. Is executed by a separate neural network to improve the correct answer rate of prediction.

In FIG. 6, 21 is a first neural network, 22 is a second neural network, and 2 is a second neural network.
Reference numeral 3 denotes a determination switching unit, and other components that are the same as those shown in FIG. 2 are given the same reference numerals.

Then, the first neural network 2
The inputs of the first and second neural networks 22 are connected to the traffic information collection unit 9, the clock unit 17, and the temporary information storage unit 11, respectively, and the outputs are connected to the determination switching unit 23.
The determination switching unit 23 is coupled to the temporary information storage unit 11 and the group management control device 5.

The operation of this embodiment will be described. First, it is assumed that the first neural network 21 is for prediction and the second neural network 22 is for learning. Long-term memory information is for prediction (first)
The point that the temporary memory information is supplied to the neural network 21 and the learning (second) neural network 22 and the temporary storage information is supplied to the temporary information storage unit 11 is exactly the same as the fourth embodiment. At this time, the prediction neural network 21 uses the long-term storage information from the traffic information collection unit 9,
Based on the time information from the clock unit 17 and the temporary storage information from the temporary information storage unit 11, the traffic volume after a short time t _n is predicted and the predicted output is supplied to the group management control device 5. On the other hand, the learning neural network 22 similarly predicts the traffic volume after a short time t _n based on the long-term storage information, the time information, and the temporary storage information.
Then, when the measured value of the traffic volume is obtained at time t _n when the time has passed and the prediction is performed, the determination switching unit 23 causes the prediction calculation result of the prediction neural network 21 and the prediction calculation result of the learning neural network 22. When the prediction calculation result of the learning neural network 22 is closer to the actual measurement value than the prediction calculation result of the prediction neural network 21, the learning neural network 22 is used for determination (second time) in the next prediction. 2) and the determination neural network 21 as the learning (first) neural network 21 so that learning and prediction are performed. On the other hand, when the prediction calculation result of the judgment (first) neural network 21 is closer to the actual measurement value than the prediction calculation result of the learning (second) neural network 22, the judgment (first) neural network 21 Inputs some traffic volumes used for the prediction stored in the temporary information storage unit 11, learns by using the actually measured traffic volume obtained at the time t _n as a teacher signal, and outputs the learning result. Try to remember it.

When the functions for judgment and learning in the first and second neural networks 21 and 22 are switched, when the switching is performed only by the judgment of one prediction calculation result, the switching is performed. Although it may be performed frequently and impair the operation stability of the demand prediction determination device 3, in such a case, the learning neural network 22 changes the measured value from the determination neural network 21 continuously several times. Switching may be performed for the first time when a similar prediction calculation result is output.

If the neural network is constructed by a program on a digital computer, only two sets of weighting coefficient tables for each neuron connecting section are prepared and the same program is used for the determination. And a learning neural network can be configured.

Further, in the present embodiment, the case where two sets of neural networks 21 and 22 are used has been described, but a plurality of neural networks may be provided as a learning neural network, for learning or prediction. Instead of using a neural network, a predictive means such as a multiple regression analysis, a correlation coefficient, a Kalman filter, etc. may be adopted, and the means that has obtained the closest measured value may be used for the prediction.

As described above, according to the present embodiment, at least the first and second neural networks 21 and 22 are used, and they are divided into the learning network and the determination network to perform the prediction. Until the learning state of the application neural network converges, stable prediction can be performed using the determination neural network. Then, when the learning state of the learning neural network has converged and more accurate prediction can be made, if switching to the prediction neural network, the demand for elevators corresponding to the characteristics of each building and aging It is possible to execute the prediction in a stable state.

In each of the above-mentioned embodiments, the case of predicting the traffic volume after a short time has been described as an example, but the present invention is not limited to the case of predicting the traffic volume. The traffic flow including the traffic volume and destinations of each floor can be predicted in the same manner as the traffic volume.

Next, FIG. 7 is an explanatory diagram in the case of predicting the call generation, and FIG. 7A shows an example of the call state of each floor corresponding to each time in the residential elevator,
(B) shows an example of the calling state of each floor corresponding to each time in an elevator such as a tenant building.

In FIGS. 7A and 7B, the horizontal axis represents the time and the vertical axis represents the call occurrence probability, and the hall call occurrence probability in each floor at each time.

As shown in FIG. 7A, in a residential elevator installed in a high-rise house, hall calls are concentrated in almost constant time especially in the morning work hours. The reason for this is that in the above-mentioned residential elevators, the users on each floor are limited to the residents on that floor in the early morning hours, and most of the users in the above hours are the time to start work and start work. This is because the same person is often used at the same time because they are commuting attendants who have fixed times, and the car call is often on the first floor, which is the exit.
Therefore, in the residential elevator, the generation of calls on each floor will occur at a substantially constant time depending on the time of day, which makes it easy to predict the generation of calls on each floor, and furthermore, using the prediction. It becomes possible to perform group management control such as prior arrangement of cars to the floor where call is scheduled to occur and efficiency of allocation.

Further, as shown in FIG. 7B, in an elevator installed in a tenant building where one building is used by several companies, when a hall call on a specific floor occurs, A few minutes later, a situation occurs in which hall calls for other specific floors occur, or car calls derived from hall calls for one specific floor concentrate on another specific floor. The reason for this is that usually one company is concentrated on a certain floor, and users move between the same companies much more often than they do between different companies. This is because it is used to move from one floor to another floor, finish work on the other floor, and then return to the original specific floor. Therefore, in the elevator such as the tenant building, since there is a correlation between a hall call and a car call between a certain specific floor and another specific floor, if the correlation is used, it is possible to predict the call occurrence. As in the previous case, the prediction can be used to perform group management control such as pre-arrangement of cars to the planned floor where call is to occur and efficient allocation of cars.

Here, the example shown in FIG.
In the example in which the same company resides on the 3rd and 4th floors, when calls a and c occur on the 3rd floor at times t ₁ and t ₃ , ₄ minutes at the times t ₂ and t 4 Calls b and d occur on the floor.
In addition, the derived car call by the hall call on the 3rd floor is almost on the 4th floor, and the derived car call by the hall call on the 3rd floor is on the 1st floor of the lobby floor. Such a correlation between specific floors remarkably appears in a tenant building or the like, but even when only one company is in one building, it may appear due to the strength of the connection between the departments of the company. Therefore, by learning the correlation between each hall call and the car call, it is possible to predict the occurrence of a call with a configuration exactly the same as in the case of demand prediction.

By the way, in the case of a residential elevator or an elevator such as a tenant building, the usage pattern may change when the resident of the building or the resident company changes.
In such a case, it is necessary to newly start learning.

Therefore, FIG. 8 is a block diagram showing the fifth embodiment of the elevator operation demand forecasting apparatus according to the present invention, in which the learning result can be initialized for each floor so that It is capable of responding to changes.

In FIG. 8, reference numeral 24 is a floor-by-floor initialization device, and other components that are the same as those shown in FIG. 2 are designated by the same reference numerals.

The floor-level initialization device 24 is connected to the long-term information storage unit 12 and the correlation table 16.

Since the operation of this embodiment is basically almost the same as that of the first embodiment shown in FIG. 2, a detailed explanation of the operation of this embodiment will be omitted, but particularly In this embodiment, demand forecast is performed for each floor so that only the learning result of the floor where the resident or the moving company has moved can be initialized. In this case, the floor-by-floor initialization device 24 is provided outside the demand prediction determination device 3, and the long-term information storage unit 12 and the corresponding floor of the correlation table 16 are stored by the output of the floor-by-floor initialization device 24. Only the contents are initialized.

As described above, according to the present embodiment, the learning is re-executed only for the floors on which the resident or the moving company has moved, so that while predicting the other floors, new learning is performed. You can do various learning.

In the present embodiment, the learning calculation unit 15 uses the correlation calculation unit 15 and the correlation table 16, but the learning unit 13 uses the multiple regression analysis unit 18 and the regression equation table 19. Of course, the same can be applied to the case.

In the case where a neural network is used for the demand forecasting / determining device 3, the neural network is generally one in which neurons are related to each other to obtain an output. It is not possible to initialize only the stored contents of the department.

FIG. 9 is a block diagram showing the sixth embodiment of the elevator operation demand prediction apparatus according to the present invention. Even when a neural network is used, the learning result can be initialized for each floor. It was done like this.

In FIG. 9, reference numeral 25 is a neural network, 26 is an input layer neuron, 27 and 28 are a plurality of intermediate layer neurons, 29 and 30 are a plurality of output layer neurons, and the others are shown in FIGS. 5 and 8. The same components as those described above are designated by the same reference numerals.

Then, the neural network 25 is
One input layer neuron 26, a plurality of intermediate layer neurons 27 and 28, and a plurality of output layer neurons 29 and 30 are provided, and the intermediate layer neurons 27 and 28 and the output layer neurons 29 and 30 are respectively placed on the floor. Prepared as many as.

The operation of this embodiment is basically almost the same as the operation of the third embodiment shown in FIG. 5, so a detailed description of the operation of this embodiment will be omitted. In the present embodiment, the intermediate layer neurons 27, 2 provided for each floor
For the floor where the resident or resident company moved after learning in 8, the middle layer neurons 27, 2 of that floor
Only 8 is initialized to reduce the influence on the prediction outputs of other floors. In this case, the neural network 25 stores information about each floor obtained by learning in the intermediate layer neurons 27 and 28 and the output layer neurons 29 and 30 corresponding to each floor, and the specific floor There is the movement in
If this changes the usage of the elevator,
By controlling the floor-level initialization device 8, it is possible to initialize only the intermediate layer neurons and the output layer neurons of the specific floor, and start new learning without affecting the prediction performance of other floors. It is possible.

As described above, according to the present embodiment, it is possible to predict demand for other floors while learning for a particular floor, and to achieve that function using the neural network 25. Therefore, the configuration of the demand forecast determination device 3 can be simplified.

[0081]

As described above, according to the present invention,
Based on the long-term information in the traffic information related to the operation of the elevator, we learn the correlation and characteristics of each event, and the learning result is changed and corrected with the temporary information in the traffic information. It is possible to predict the most suitable and accurate traffic flow at each point of time, taking into account minute changes in the daily usage of elevators, and hall calls and car calls based on the correlation of specific events that occur at specific times. There is an effect that the prediction of can be performed.

That is, according to the present invention, the traffic volume and other various information related to the elevator can be accurately and accurately predicted with a relatively simple structure, so that the traffic flow and the traffic volume generated after a short time can be matched. The control can be prepared in advance, and the traffic flow after a short time can be promptly dealt with. For this reason, deterioration of service functions due to non-conformance of control corresponding to the traffic flow and traffic volume,
There is an effect that it is possible to prevent the service function from being deteriorated due to an increase in waiting time or the like due to the change in traffic flow or traffic volume. Further, by accurately predicting the occurrence of a call, there is also an effect that the waiting time in a quiet time can be significantly shortened.

In addition to learning the correlation and characteristics of each event and changing and correcting the learning result by using a neural network, the configuration can be further simplified. There is an effect.

[Brief description of drawings]

FIG. 1 is a block configuration diagram showing an outline of a configuration of an elevator operation demand prediction device according to the present invention.

FIG. 2 is a block configuration diagram showing a first embodiment of an elevator operation demand prediction device according to the present invention.

FIG. 3 is an explanatory diagram showing a case where the embodiment shown in FIG. 2 is applied to traffic volume prediction.

FIG. 4 is a block diagram showing a second embodiment of an elevator operation demand prediction apparatus according to the present invention.

FIG. 5 is a block configuration diagram showing a third embodiment of an elevator operation demand prediction apparatus according to the present invention.

FIG. 6 is a block diagram showing a fourth embodiment of the operation demand forecasting apparatus for an elevator according to the present invention.

FIG. 7 is an explanatory diagram in the case of predicting call generation.

FIG. 8 is a block configuration diagram showing a fifth embodiment of an elevator operation demand prediction apparatus according to the present invention.

FIG. 9 is a block diagram showing a sixth embodiment of an elevator operation demand prediction apparatus according to the present invention.

[Explanation of symbols]

1 Hall call collection unit 2 Car call information collection unit 3 Demand determination device 4 External information source 5 Group management control device 6 Elevator control device 7 Elevator car 7 _{1 In-} car call button 8 Hall call button 8 ₁ 1st floor hall call button 8 _n nth floor hall call button 9 traffic information collection unit 10 general information collection unit 11 temporary information storage unit 12 long-term information storage unit 13 learning unit 14 demand forecasting unit 15 correlation calculation unit 16 correlation table 17 clock unit 18 multiple regression analysis unit 19 Regression formula table 20, 25 Neural network 21 First neural network 22 Second neural network 23 Judgment switching unit 24 Floor-specific initialization device 26 Input layer neurons 27, 28 Multiple intermediate layer neurons 29, 30 Multiple output layers Neuron

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Kenji Yoneda 1070 Ige, Katsuta City, Ibaraki Prefecture Hitachi Ltd. Mito Plant

Claims (9)

[Claims] 1. An elevator demand prediction apparatus comprising a demand prediction determination apparatus for predicting demand for elevator operation after a short time by using stored past traffic information and present traffic information. The demand prediction determination device is a traffic information collection unit that collects traffic information such as hall call information and car call information, a long-term information storage unit and a temporary information storage unit that classify and store the collected traffic information,
It includes a learning unit that performs required learning based on the stored contents of the long-term information storage unit, and a demand prediction unit that changes the learning result obtained by the learning unit according to the stored contents of the temporary information storage unit. An elevator operation demand forecasting device. 2. The operation demand forecasting apparatus for an elevator according to claim 1, wherein the required learning in the learning unit is to learn a correlation between each piece of traffic information. 3. The elevator operation demand prediction apparatus according to claim 1, wherein the required learning in the learning unit is to perform multiple regression analysis of each traffic information. 4. The long-term information storage unit and the learning unit,
The floor-by-floor initialization device for initializing the learning content of the specific floor in response to changes in the usage status of the specific floor in the number of elevator service floors. 4. The elevator operation demand prediction device according to any one of 3 to 3. 5. An elevator demand forecasting apparatus comprising a demand forecasting determination apparatus for forecasting demand for elevator operation after a short time by using stored past traffic information and present traffic information. The demand prediction determination device includes a traffic information collection unit that collects traffic information such as hall call information and car call information, a temporary information storage unit that stores temporary information in the collected traffic information, and a neural network. The neural network includes long-term storage of long-term information in the collected traffic information, performs required learning based on the stored long-term information, and stores the learning result in the temporary information storage unit. An elevator operation demand prediction device characterized by being changed according to stored contents. 6. The neural network comprises at least a prediction neural network and a learning neural network connected in parallel, and the output sides of the prediction neural network and the learning neural network are connected by a decision switching unit. The elevator operation demand prediction device according to claim 5, wherein the operation demand prediction device is an elevator. 7. The neural network is coupled to a floor-specific initializing device for initializing learning contents of a specific floor in response to changes in the usage status of the specific floor in the number of service floors of the elevator. The elevator operation demand prediction device according to any one of claims 5 to 6, characterized in that: 8. The demand forecast determination device further includes a general information collection unit that collects other information related to traffic flow,
6. The other information collected is classified into long-term information and temporary information, and the classified information is stored in a storage unit corresponding to the information content thereof. Elevator demand forecasting device. 9. The operation demand of an elevator according to claim 1, wherein the demand forecast determination device generates a demand forecast output for forecasting a hall call and a car call after a short time. Prediction device.