JPS63106281A - Group control method of elevator - Google Patents

Abstract

(57) [Abstract] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Objective of the Invention] (Industrial Field of Application) The present invention relates to group management control of elevators, and in particular, by applying knowledge engineering, it is possible to achieve each target value in various group management controls. The present invention relates to a group management control method for elevators that enables good tracking.

(Prior art) In recent years, when multiple elevators are installed in parallel, in order to improve the operating efficiency of the elevators and improve the service to elevator users, the response number for hall calls on each floor has been changed to a small device such as a microcomputer. Computers are used to allocate information rationally and quickly.

In other words, this is a control method called group management control.
When a hall call occurs, the system selects the most suitable elevator to handle the hall call, quickly assigns the elevator to respond to the hall call, and prevents other elevators from responding to the hall call. Alternatively, the system can switch to a preset driving mode according to traffic changes unique to the building, such as when traffic demand increases, such as when leaving work in the morning, when leaving work, or during lunch, or even when traffic decreases, such as at night. It controls each elevator for efficient operation.

By the way, due to the remarkable development of small computers in recent years and the cost reduction of small combinations, small computers have come to be used not only for group control devices but also for elevator control devices that control individual elevators. Ta.

Furthermore, serial transmission is becoming the mainstream method for transmitting information to these combinations. These group management control devices and individual elevator control devices are controlled by each transmission device and a certain procedure using software.
Group management data can be exchanged freely by simply connecting wires (or optical cables).

In addition, even in large buildings where a large number of elevators are managed and controlled in groups, building management computers and OA (office automation) computers are used to manage the entire building and exchange information between each floor. ing. This information includes many items related to building traffic. Further, the group management control combination information also includes hall data (for example, hall passenger detection) necessary for building management. However, the exchange of these building management data and group management control data is
is almost never done. For this reason, there are cases in which sensors and notification devices are installed on each device to obtain information.

However, in general, group management control of elevators focuses on control of allocation to hall calls and special operation control for high traffic demand.

Conventionally, allocation control for hall calls has been performed by evaluation using various calculation data such as arrival time. For this reason, an allocation failure may occur due to a prediction failure.

In order to eliminate this problem, there are systems that learn daily traffic flows, and systems that require reliability of predicted data.

(Problem to be solved by the invention) However, in an elevator system, no matter how much learning is performed as described above, hall calls occur stochastically, and it is impossible to completely predict derived car calls. Well, the reliability of that data is low. However, even if the reliability of the data is low, a value of a certain order of magnitude can be obtained by predictive calculation and must be incorporated into evaluation calculation.

This means that evaluation decisions are fraught with many possibilities of failure. This also shows that there is a discrepancy between the setting of conditions using predicted values and the conditions considered by a true designer. By aiming for the desired conditions, the movement tends to be conservative and slow, rather than the nimble movement that the designer had intended.

Furthermore, when incorporating the designer's ideas into the evaluation calculation of allocation, there is a tendency that the changes in function values, weights, etc. become indirect, resulting in sluggish movement.

Furthermore, this fact indicates that trends are difficult to emerge even when verifying effects.

Even in special operations for handling high demands such as elevator group management control, due to the ambiguity of forecast data), as mentioned earlier, fine-grained control tends to become sluggish and conservative control.

Therefore, the purpose of this invention is to directly utilize the knowledge of a designer, that is, a group management control expert, for control.
To achieve fine-grained control, and at the same time to take into account the ambiguity of the conditions that form the basis of that control, to make it possible to reduce failures in allocation control, and to perform control that leaves the possibility of good service intact. An object of the present invention is to provide a group management control method for elevators that enables the following.

[Configuration of the Invention (Means for Solving Problems) In order to achieve the above object, the present invention is carried out as follows. In other words, in elevator group management control that centrally controls multiple elevators, when obtaining predictive data to be used for control from the current situation and determining various control instructions based on the conditions, it is necessary to determine the intended control purpose. In order to achieve this, we set the contents of control instructions according to the above various conditions based on empirical rules, and at the same time, we use expert knowledge to create the attribution function that determines each condition and its confidence based on the attribution value of the predicted data. The above predicted data used for control is artificially set based on the subjective quantity based on the above, and the confidence value is set by adding a confidence level to the ambiguity. The effectiveness of the assumption that the control instruction is given using the condition as the degree of attribution to the control instruction set above is determined by weighting by this degree of attribution, and the result is calculated by inference. It is characterized by determining control instructions according to.

(Function) In other words, in group management control, when obtaining predictive data to be used for control from the current situation and setting various control instructions based on the conditions based on this, the ambiguity of the predictive data used for control is The value confidence value is calculated by taking into account the corresponding confidence level, and using this confidence value, the above 8172
The conditions and their degree of belonging are obtained from the degree function, and based on this, the applicable ones are extracted from the content of control instructions based on various conditions based on empirical rules set to achieve the intended control purpose, and the degree of belonging mentioned above is determined by The effectiveness in the case where it is assumed that the extracted control instruction is issued using the Rf degree as the Rf degree to the extracted control instruction is determined by weighting based on the degree of attribution and by inference. Then, control instructions are determined according to the results.

In this way, the present invention obtains a control instruction based on the experience and subjectivity of an expert as a comprehensive evaluation value that takes into account the ambiguity of the materials used to judge the control instruction. This is to select the most suitable control command under the current situation. Since the content of this judgment reflects the experience and intuition of experts, it is possible to perform optimal control by avoiding inappropriate control commands due to changes in the situation that cannot be predicted using only actual data. It becomes like this.

Since the knowledge of the designer, that is, the group management control expert, can be directly used for control, the average non-response time can be reduced in order to perform control that always leaves the possibility of improving the service, and at the same time, the control Since it is possible to make selections that take into account the ambiguity of the underlying conditions, it becomes possible to provide an elevator group management control method that makes it possible to reduce failures in assignment control.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a group management control system according to an embodiment of the present invention.

In the figure, 1 is a group management control device that manages the entire group management control, 2 is an elevator single control device that controls the operation of a single elevator, and 3 is a transmission controller that exchanges data with the outside. 4 is a monitoring monitor that monitors the operating status of the elevator. The group management control device 1 includes an elevator control device 2, a transmission controller 3. It is connected to the elevator monitoring monitor 4 through a serial transmission line using a transmission-only LSI. Further, I/I (input/output) coupling between the hall dart, Langsenna, display, etc. and the hall input/output device 5 is performed by a serial transmission line. Tore is L for transmission only
Based on SI and general-purpose transmission software. Further, an in-car controller 6 provided in the car and a corresponding elevator control device 2 are also connected by a serial transmission line. In addition, the data of the building management duo - Yu 11,
The data of the OA computer 12, the data of the time recorder 13, and the notification data input terminal device 14 and the system 110
For example, force notification data, entrance counter data (such as usage and termination of a conference room, etc.) given via the controller 15 are
The number of people entering and leaving) is combined by the interface of the transmission controller 3 and transmitted to the serial system path on the elevator system side. As a result, in addition to being able to provide the above-mentioned various data to the group management control device 1, data related to group management can also be provided to the computers 11 and 12.
It is configured so that it can be given to

This system is close to the largest scale example. Therefore, the present invention can be applied even to a system in which a portion is deleted.

Next, the software configuration of the group management control device 1 will be explained. As shown in FIG. 2, the configuration of the software for group management control consists of various functional tasks, their salaried routines, and a task management program for managing these tasks. When the small combination computer (generally a microcomputer is used, henceforth referred to as a microcomputer) in the group management control device 1 is started, this microcomputer first starts and then executes the task management program 2o. determines which task, that is, which functionally separated software module to start. Tasks are software modules with different functions, and are activated according to conditions.

Here, each task will be briefly explained. 32 is an initialization task], which is a task that initializes RAM (random access memory) and CPU (processor) registers, and LSI (large scale integrated circuit), and initializes the initial state and operation. is activated when the mode changes.

21 is an external input task, which is CCT, KCT
This is an input task that sets external inputs such as HCT on RAM (random access memory). This task has a high priority, and if started, it will need to be restarted every 100mm or so.

Additionally, trigger checks are performed during the original task.

ζThe above-mentioned HCT includes a hall condition table, into which the data registered with the hall call in the elevator control device is input.

OCT is Kirkcondige, Ntable, and KCT is KCT, and currently, it is assumed that there are four elevators, A-D, making up the group, and the service floors are 1 to 8 floors. Then the above HCT.

CCT and KCT have bit configurations as shown in FIGS. 3, 4, and 5, respectively.

That is, the HCT representing the hole state shown in FIG.
, 0-130 hole sub-index (H8)
For K, 8th floor descent (8D) to 7th floor rise (7U) t
8-bit information is stored in each. The hall conditions for each floor will be explained in detail.

For example, if the up call hall button is pressed in the elevator hall on the 5th floor, the 7th bit of Ha's J J (5U) is 1, and the service elevator corresponding to this hall call is pressed using the method described below. When it is determined that it is machine A, the 0th bit and the 6th bit in 11 of H8 become 1.

Then, when the A car arrives at the 5th floor, the 0, 6 and 7 bits in 11 of Ha are all reset to OK.

That is, bits 0 to 3 indicate the number set of each elevator, pit 6 indicates whether an elevator is assigned to a hall call, and bit 7 indicates whether there is a hall call.

In the OCT representing the car status in FIG. 4, information on 16 pits of elevators A to D is stored for indexes 0 to 3. That is, 0
- The load condition of the car is shown in binary notation in the 3rd pit. The meaning of these bits 0 to 3 is @0001",'"
0010”.

@0011’, @0100”, @010”.

@0110’, @0111”, @1000”.

@1001”, @1010”, @1011”.

0 to 10% for ``1100'', respectively.

11-20 inches, 21-30 inches, ~31-40%.

41-50%, 51-60eII, 61-70%.

71-8 (1,81-90%, 91-100%.

101 to 110 inches, 111 inches or more.

The 5th bit indicates the running status of the car, ``1'' indicates running, @0 indicates decelerating. The 7th bit indicates the open/closed status of the door, ``l'' indicates open, and 0 indicates Indicates that it is closed.8-1
The third bit indicates the heel position in binary notation. 1
4. The 15th bit indicates the direction of movement of the car, "10" indicates rising, @Of" indicates descending, and @00' indicates no direction.
In other words, it indicates that it is stopped.

FIG. 5 shows a KCT representing a call state. In this KCT, like the HCT shown in FIG.
The pit indicates whether there is a car call for elevators A-D.

Through the above steps, the status of the elevator and hall call is input.

Reference numeral 22 denotes an allocation task, which performs corner allocation. This task checks new hall calls every 100m5.C or so, and if any new hall calls occur, evaluates them based on predicted unanswered time, fullness, etc., and determines the hall with the best evaluation.

26 is an allocation review task, and this allocation review task 26 is a low-level task that is activated approximately once every second. Then, the allocation will be changed.

28 is a task for communicating with each individual elevator, and in addition to cyclic data transmission, this task also outputs control output (assignment, assignment cancellation, etc.) as necessary. Or the number of passengers,
Requests are made for data such as the number of people getting off the train and whether there is a new outbreak. These operations are performed using a buffer, and data as shown in FIG. 6 is transferred as shown in FIG. 7.

29 is the annual timer and various timers, which is 10m5
This is a routine for various interval/quarter timers such as ec, 100mm5e, and 1 second, and an annual timer combined with them. Additionally, these data are corrected by an external timer. The yearly timer has information such as the month, date, day of the week, holidays, Rokuyo, and other events, the second I/L task for floppy disks, and the first I/L task for CRT. Oki task information will be updated.

30 is the first Ilo (input/output) task for CRT (Character Display Terminal) transmission, and this task is used to transmit information with external terminals, other combinations, etc. Ru. This task is started in a time-sliced manner at a low interrupt level so as not to harm other group management masks. Also, 31 is the second I1 for 70-tubi (flexible) disk control.
This task is activated when learning data etc. are stored in an external 70 disk. Like the first I10 task 30 for CRT transmission, it is activated at a low interrupt level.

22 is a learning data processing task, and this learning data processing task sets the current state in a data table based on external input or data from a single elevator, and whenever the state changes to the next state, the learning data processing task sets the current state in the data table. A task that replaces data, and is activated when data changes or the state changes. Additionally, this task is a low interrupt level task and is activated so as not to harm the high interrupt level group management task. However, if there is a special 7 lag or if the priority order is changed, it will change. The learning data is classified into several equivalent transportation modes based on factors such as month, day of the week, six days of the week, holiday, and time band as shown in Figure 8, and each mode has the following data. .

Examples thereof are shown in FIGS. 9 and 10.

Therefore, HCT$RAT is the average number of hall calls in 15 minutes,
KCT$RAT is the average number of car calls, IN$RAT
indicates the average number of passengers, OUT$RAT indicates the average number of passengers getting off the train, and KCTlSET indicates the call occurrence rate for each floor.

HCT$RAT, ~OUT$RAT is indicated by index H8 (hall sub-index) of the oriented floor.

For example, KCTl$RAT is indicated by a matrix of A and B to store the number of calls from floor A to floor B and the call occurrence rate.

Also, during times of high demand, these changes are learned at small intervals. This is for each H8 and tAV$M
Denoted as ENSP (HS, t).

However, t is time.

These learning data do not need to be used directly in the present invention, but since they contribute to improving accuracy when performing various predictive calculations, it is better to use them to obtain better results. It goes without saying that even if there is no learning data, control can be performed with reasonable accuracy.

Other tasks shown in FIG. 2 include the building computer communication and data collection task 34, which is started every second and performs data input and output data communication with external computers, as well as data collection. Traffic demand forecasting task 33 (100 mA) uses data to anticipate demand, predict traffic demand, and determine driving models.
There are also driving tasks 35 as driving model tasks that are activated by this.

In the system of the present invention, the group management control device 1 is further provided with a hall call assignment control function and a traffic model adaptive control function using the inference described below.

Inference here refers to the data conditions used to perform such control.
Based on the knowledge of elevator control experts, the degree of contribution σ to the instruction content given by the conditions including these and the judgment based on the instruction content is determined based on the knowledge of elevator control experts, and based on those comprehensive judgments. It refers to the method of determining the final table instructions.

Traffic model adaptive control refers to data (including learning data) on the demand situation that changes depending on, for example, dispatch time, lunch time, weekdays, holidays, nighttime, etc., or temporary demand changes such as meetings. Data such as special demand for meeting rooms on floors due to events such as holding conferences (data on suitability can be obtained by making reservations to the building management, for example, 2nd floor 1-1)
Kyoshi forecasts demand, such as the concentration of users on a specific floor of a building at each time of the day, and the distribution of users from a specific floor to various floors, and then develops a preemptive system to cover the predicted demand. This refers to performing control that is compatible with the demand forecast model that has been created.

Next, the group management control method based on the present invention will be explained using assignment control for hall calls as an example.

FIG. 11 shows a flowchart of allocation control.

Furthermore, for convenience of explanation, it is assumed that the following three elevators (man to C) are subject to group management control. For convenience, the floors covered by the service are from the 1st floor (IF) to the 12th floor (1
2F).

The allocation control method for newly generated hall calls differs depending on the operation mode of the elevator. Therefore, as an example, allocation control in the operation mode when the traffic flow on each floor is in a balanced state will be explained.

In the allocation control routine during balance no lean, the target values are (1) reduce long-term calls (reduce long-waiting calls of 60 seconds or more to 0), and (2) increase good calls (total number of calls answered within 30 seconds). (3) Maintain the sergis on high demand floors in good condition.

(4) Eliminate full penetration (when load is 80 inches or more) is set. Here, the above (1) will be explained in detail below, but the above (2) to (4) are also carried out in the same way.

The control strategy for the above-mentioned goals is based on the experience and knowledge possessed by experts, and is the control strategy of experts in allocation control.

Therefore, artificially defined control rules expressed by conditions and instructions are prepared corresponding to each control strategy.

For example, the conditions and control instructions corresponding to the goal of ``reducing the number of long-lasting calls'' are ``do not allocate if it will result in long-lasting calls''; If so, we will allocate it."

For other control strategies, control rules of the form "If it is a person, then B" are similarly prepared.

In the present invention, the degree to which the conditions indicated in the above-mentioned control rules hold true and the weighting of control instructions under those conditions are determined for each control rule by an inference function. The degree to which the above conditions are met is "if it lasts a long time"
If so, it means "the degree to which it lasts a long time."

A high degree of this means that you will have to wait a very long time, while a low degree of this means that you will have to wait a little longer. In the above-mentioned reasoning, the degree of long waiting time caused by provisional assignment to a certain electric machine is predicted, and the control instructions are weighted according to the degree of waiting.

That is, when performing tentative allocation, if it is likely that there will be a long wait, the control instructions are weighted in the direction of not performing the allocation. The weighting of the instructions determined for each control rule is comprehensively evaluated, and in the case of assignment control, the evaluation value as the strength of the control command "assign" is determined.

This is done for each elevator number, and the hall call is assigned to the elevator number showing the strongest value.

When determining the allocation, if there are a large number of goals, if you try to allocate to an elevator that completely satisfies each goal, you will not be able to make the allocation. 1 In the present invention, allocation is made to the elevator number that satisfies all the targets to the highest degree overall.

For this purpose, a control rule is prepared based on the above-mentioned control strategy, and the degree to which the condition is met and the strength of the instruction are determined to determine the strength of the control command.

This part will be explained in detail.

The major control elements of group management control include ■ Assignment control for hall calls ■ Special operations during high demand. In addition, in (2), there are cases where demand is concentrated, cases where demand diverges, or cases where these are combined, and appropriate allocation control can be considered for each.

Therefore, in the present invention, as shown in FIG. 11, the Fl routine determines the operation mode corresponding to ■, and the routine F2-1 to F2-1 to perform allocation control corresponding to the respective operation modes OPI to OPn. F! - In the control of , control routines based on inferences of the direct representation of expert knowledge are likewise used.

In the following, the control method will be explained focusing on the allocation control F2-1 at the time of second-run.

In particular, in the present invention, the inference routine is the central part and is commonly used in Fl, F2-1 to F2-m in FIG. 11, so this part will be explained first.

In the inference routine, if a control instruction is given to a certain machine, the error amount/(deviation) with respect to the target value and the error increment Δ caused by that control instruction are calculated.
- In other words, the output Δe, which indicates whether or not to issue the instruction, is determined based on how much the possibility is to be increased and the rate of increase. In addition, for . and Δ., there are several predicted values depending on the subjective quantities of experts and the confidence level based on probability data.

The control to reduce the number of long-waiting calls as a target value in the balance/4-turn allocation control routine will be described in detail below. First of all, with respect to the target value of (1) (reducing long-lasting calls), it is necessary to determine in advance whether or not to reduce the value according to the error amount, error increment Δ, as shown in FIG. The condition-instruction table for the long waiting part is determined based on artificial judgment criteria obtained from knowledge and experience.

In FIG. 13, PB, PM, and ZZ are at the levels e and Δ·, PR is "positive and large," PM is "positive, and the output ΔU indicates the content of judgment, that is, the instruction, and po is "
J and PS to be assigned are "may be assigned".

20 is ``normal J, NS is ``no need to be assigned.''

NE means "not allocated."

Further, FIG. 12 shows a flowchart of the inference part. However, this is a general flowchart, and these steps are performed for each target rule.

That is, if a hall call occurs on a certain floor, the group management control device 1 responds to the hall call by
In 3T12-11C, a certain control instruction is temporarily set. For example, suppose that a hall call is assigned to car A. Next, in the routine of step BT12-2, the number of long-waiting hall calls that will occur in machine A due to the assignment is determined by the error amount 6 and the error increment Δe for each hall call. Prediction is made based on the degree of attribution of prediction data to the condition.

Here, each error amount and error increment Δ will be explained. In this case, the degree of belonging is determined based on the predicted value (rp2) of a long waiting time for a hall call on each floor (that is, the certainty of a long wait) and the system importance (Hsp) of those floors.

The importance of this system is that on floors where high demand is predicted, if the linearity of the system is disrupted due to long waiting times on that floor and frequent car calls, predictions become impossible. However, considering the influence, the degree of importance has been replaced with a numerical value. This is a learned HC
It is effective to make predictions using data such as T$RAT and KCT$RAT. If learning data is not used, the designer's subjective quantities and experience values are used.

FIG. 15 shows a membership function as an example of long-lasting possibility data for hall call assignment. Letting the membership function shown in this example (FIG. 15(a)) be S, it is expressed as follows. However, a, b, and c are appropriately added according to the data. Furthermore, the X-axis uses the number of long waits as a percentage, and the y-axis data represents the degree of belonging.

Here, the function changes depending on the level when the certainty factor (certainty) of TP2 is small (see FIG. 15, 6). 1st
In Figure 5 (b), the confidence level is 2 to 5 or 5〉K2)K
3〉K4, the corresponding curves are 15-5゜15-2.15-3.15- in Fig. 15, respectively.
It becomes 4. As the certainty increases, the membership function S shown above is changed.

For example, if "certainty A" is double "certainty B", the first
In Figure 5 (b), the “certainty B” curve is 15-2
On the other hand, the curve of r certainty A' is K as shown in 15-5.
Become.

Based on the relationship between the number of long-waiting hall calls for that machine due to hall call provisional allocation and the confidence level for the predicted value, the increase in the number is calculated in the form of error minute and error increment Δe. First, the degree of attribution of the long waiting probability data (TPJ) as the predicted value to be used is determined.

Think of this as follows. That is, as shown in FIG. 15(c), for confidence degrees of 801s, 50%, 20chi...゜15-12...
Use. This means that there is one new hole call for each (x = 1), and if it is assigned to this, the degree of membership y will be approximately 0.9 (P4) and approximately 0.5 (P4), respectively, depending on the confidence level. P2) and approximately 0.0 (P6). Incidentally, Pn (n=1.2.3...) indicates 4 India. Also, if there are 3 hall calls (x = 3)
The degree of membership y is approximately 0.98 (P4) and approximately 0.8, respectively.
5 (P5), approximately 0.12 (P 6 '').

If there are calls with multiple confidence degrees for one machine, the sum of those attribution degrees is used. However, if the total is 1 or more, it is counted as 1.

Basically, the degree of attribution of predicted data is determined by using the degree of attribution function according to the degree of certainty for the hall call, and if there are multiple hall calls with different degrees of certainty, the degree of attribution corresponding to each degree of certainty is used. The total value of the degree of membership determined by the degree function is K. However, in the present invention, we change this way of thinking and calculate the degree of certainty when there are multiple hall calls.
First, a composite value F is obtained by summing the reliability of each hall call, this is averaged by the number of hall calls, and this value is used to form the overall reliability.

Overall confidence = data certainty = average confidence = −=
K... (A) However, the confidence level is the probability that the car will arrive with 60 or more TPj, n
is the total number of hall calls including provisional assignments.

Then, using the membership function as shown in FIG. 21 for this overall confidence level K, the error amount e and error increment Δ· are determined.

The degree of membership function shown in FIG. 21 is independent of the number of calls, and with respect to the composite value F and the error increments ΔF and K of the composite value,
For each value of K (with a certain range), K is "assigned", "may be assigned", and "normal".

Indices such as ``no need to allocate'' and ``no allocation'' are artificially shown in relation to the error increment. Therefore, here we choose the υ function for K and F
, ΔF is a function that allows the above index to be selected. Therefore, once K and F are determined, the degree of belonging can be determined as the error amount e and the error increment Δ· for each index, regardless of the number of hall calls.

Therefore, when the composite value F and the overall confidence K for machine A are first determined using equation (4) and (A9) above, the membership function (Fig. 21) corresponding to K is selected, and The value of composite value F before and after provisional assignment to F@1 +
Find the increment ΔF1 of Fa2 with respect to FB2 and Fl.

Then, calculate Fa2, ΔF1, Scherrer's component, and error increase Δe, and use the corresponding conditions and instructions based on artificially set empirical rules as shown in Figure 13 (,). seek.

Assuming that the membership degree function used at that time is as shown in FIG. If ΔF1 is as shown in the figure, this 212. ΔF, error increment Δ・is condition z
The degree of membership for z is 0.5. Then, the degree of attribution for condition PM is 0.5. Furthermore, the degree of attribution for condition zz is 0.1 for the error. Then, the degree of attribution is 0 for condition PM.
．． It will be 9.

In other words, when considering whether a certain object is an element of set A, rather than strictly dividing it, we use a membership degree function, which is a function that indicates the degree of membership, in order to consider the degree to which it is an element of set A. Use. This is as shown in Fig. 21(a), where the horizontal axis shows the above-mentioned composite value F and the composite value increment ΔF, and the vertical axis shows the work 2-minute and the error increment Δ. be.

In addition, in the figure, the sets zz, PM, and P
Each membership degree function (-example) of B is shown. The set zz is the set that is "approximately zero," the set PM is the set that is "positive and medium," and the set PR is the set that is "approximately zero."
indicates a set that is positive and large.

Each membership degree function is defined as a set of values ZZ, P for all composite values F or composite value increments ΔF.
Give the degree of inclusion in M and PB. This degree is
This indicates the degree of belonging to the above set, also called degree of membership, and is expressed as a number between ro, OJ and rl, OJ.

For example, if the degree of membership is 1.0, it indicates that the object is a complete element of a set called A, and if the degree of membership is 0.
If it is 0, it indicates that the target is not an element of a certain set called A.

Here, let us consider the degree of belonging when the composite value F is f. As can be seen from FIG. 21 (&), the membership degree of the set PR for one composite value f is 0.7, and
The membership degree of the set PM is 0.3.

In other words, the composite value f is "positive and large" with a membership degree of 0.7.
It belongs to the set called "Positive and Moderate" with a degree of membership of 0.3. Note that the membership function to be used also changes depending on the overall confidence level K. That is, when the data is accurate, the sections are clearly separated as shown in FIG. 21(b), and when the data is inaccurate, the sections appear gently as shown in FIG. 21(c), and the sections are not clearly distinguishable.

After obtaining the corresponding conditions and their degree of belonging in this way, the above-mentioned error amount, error increment Δ, and degree of control command are determined. In other words, A of the elevator to be controlled
Since the provisional assignment of the newly generated hall call is given to the No. 1 car as the control command ΔU, here we will specifically specify how strongly the hall call that was provisionally allocated to the No. A car will be made into the official assignment. Find it based on.

The condition-command table is as shown in FIG. 13 (&), which shows the contents of the control command ΔU corresponding to the error event and error increment Δe determined based on the experience and knowledge of experts. .

For example, if the error e is "approximately zero" (set zz) and the error increment Δe is "positive and large" (set PB), the control command ΔU is PO, PS, ZO, N
There are 5 types of S and NEO, po means "assigned", ps means "may be assigned", zO means "normal", N8 means "no need to assign", and NE means "not assigned". . Then, the closer the content is to the direction of "assign", the more positive the value is, and the further the content (meaning) is from the direction of "assign", the more negative the value is. The distribution of is set artificially. Also, the number of sets as engineering 9e is set PB
, set PM, and set NE, and the error increment Δ・
There are three similar types in the case of . There are nine types of control commands ΔU depending on the combination of error 〇 and error increment Δe. Therefore, we will consider these nine types of rules here.

The rules of this Qal class have been determined based on experience, and the details are shown in FIG. 13(b). In the diagram, for example, "Rule 1" is "Engine 2-
is "Positive and large. J(PB)J" and "Error increment Δ is [Positive and large. J(PB)j", then control command ΔU is "not assigned" (NE). It means to do.

Already, the composite value F, 2 and the increment Δ for the tentative allocation of machine A
F The degree of membership for each set is determined, and the composite value Fa
2 has a membership degree of set PM of 0.9 and a membership degree of set 22 of 0.
．． It is 1. Also, the degree of membership of the increment ΔF for that set PM is 0.5, and the degree of membership for the set 22 is 0.
．． It is 5.

Therefore, the composite value Fa2 belongs to the set PM and the set 22, and the increment ΔF of the composite value also belongs to the set PM and the set ZZI.
Belongs to C. Therefore, in order to determine the control command ΔU based on the error e and the error increment Δe, “Rule 5
”, “Rule 6”, “Rule 8” and “Rule 9” are used.

If these rules are illustrated in Figure 14, "Rule 5"
Regarding error 0, the set PM is filled to a degree of 0.9, and for an error increment Δ·, the set PM is filled to a degree of 0.5.

As the degree to which "Rule 5" is satisfied, the smaller value of the degrees to which the two sets are satisfied is adopted. Therefore, the degree of strength of the control command ΔU is limited to 0.5 in this case.

Thereafter, a set indicating the control command ΔU including the degree of strength is obtained for "Rule 6", "Rule 8", and "Rule 9" in the same manner.

Next, the logical sum of the sets of control commands ΔU obtained using these four rules is calculated, and this is weighted and averaged according to the degree of belonging to the set, to obtain the final strength U of the control commands. Since the control command ΔU has a value as a distribution of a set, as a result of the above-mentioned weighted average, the strength of control as a direction of comprehensive control is obtained according to the weight of each of the four control commands. In this case, as shown in FIG. 14, a general command to "assign" is issued to the hall call, and the value of the strength U of the control command is -0.69.

Since the value of the control command is negative, it means that it is "difficult to assign." The greater the value of the strength U, the easier it is to assign, and the smaller the value, the more difficult it is to assign. It means that.

With the above, the ability to allocate a newly generated hall call to machine A- is required.

Hereinafter, the strength U to be assigned to other machines will be determined using the same method.

This concludes the explanation of the reasoning part.

In this way, K obtains instructions for each condition using the empirical rules of condition-indication based on expert knowledge, and obtains instructions with a degree that reflects the obtained instructions at a rate corresponding to the ambiguity of the condition. From the disjunction of each instruction, a total instruction based on experience and ambiguity can be determined. This routine replaces the reference table in step 5T12-3, so that p 1 e F j -J ~F in FIG.
Can be used for 2-m.

Next, allocation control using the above reasoning will be explained.

This routine will be explained by taking as an example the predicted arrival time calculation in hall call allocation control. The algorithm of this routine can be used as is for predictive load calculation, etc.

FIG. 16 shows a flowchart of the hall call allocation control routine. The inference part explained earlier is step 5T16-
Corresponds to 4.

A hall call has occurred, and the group management control device that detected it
enters the hall call allocation control routine. First, step 5T16-1 is executed, in which it is determined which car is not in a state such as in a group or at rest, that is, which car cannot be controlled to allocate hall calls. Then, it is set so that impossible machine numbers are not subject to calculation. Next, the program moves to Stella ff3T1g-2, where each seed measurement calculation is performed. Here, predictive calculations are performed on data necessary for the inference calculation in the subsequent step 5T16-4, such as the predicted arrival time of each car, maximum arrival time, and certainty of the arrival time, and are used as inference data.

In step ST1g-4, this inference data, its confidence level, and membership function are used to establish the establishment of the condition based on each control rule that expresses the expert's control strategy by conditions and instructions as described above. The strength of the two control commands for each control rule of the machine to which the hall call has been tentatively assigned is determined by weighting the instructions and the strength of the two control instructions for each control rule of the machine to which the hall call has been tentatively assigned. The strength of control for that unit is determined as a numerical index.

Then, after performing each calculation in steps 5T16-2 to 5T16-4 described above for each machine, step 5T1
At 6-5, the strength of the control command for each car is compared, and a command is output to finally allocate the newly generated hall call to the car for which the command is the strongest (the car with the thickest value).

This completes the allocation control of newly generated hall calls.

In addition, when performing inference, each seed measurement calculation (STle-x) is performed to obtain data such as the predicted arrival time, maximum arrival time, and certainty of arrival time for each car in hall call assignment control. Among them, the flowchart of the predicted arrival time routine) is shown in the IE20 diagram.

In FIG. 20, step 5T20-1 is actually changed,
Here, a derivative call such as a derivative call or call is generated for the hall call etc. (If there is learning data, 1. Tori through hole tiH
Committee to explain momme.

FIG. 22 shows the hole sub-index H8.

The hall sub-index H8 takes into consideration the floor on which the corner is currently located and the driving direction of the car.

To explain an elevator on the 12th floor, if there is a car on the 12th floor (12F) and a descending operation is performed, the hall sub-index H8 of this car is 10'. 1
As you descend from the second floor, the value of the hole sub-index H8 increases as shown in FIG. 22. Further, when the car is on the first floor and the car is in an upward operation, the hole sub-index H8 of the car is -11', and increases as the car ascends from the first floor. For example, when car No. A descends on the 11th floor, the hall sub-index Ha of this car is -1'. Further, the fact that the hall sub-index Ha of the car is 45# means that the car is on the 7th floor and the driving direction is downward. The value of this hall sub-index Ha indicates the starting point of the calculation of the predicted arrival time to each floor, which will be performed in the following steps.

Next, step 5T20-2 is performed. Steg 8T2
In step 0-2, a derived car call for the already registered hall call is generated based on learning data and the like. This derived call is a simulated call for a hall call, and may differ from the actual call.

FIG. 23 shows the hall call and its derivative calls.

For example, if the hall calls are on the 9th and 4th floors,
These derivatives are called 7 for the hall call on the 9th floor.
This is the 2nd floor for the 4th floor hall call. As described above, after the generation of the derived car call is completed, Stella f8T20
-3 is performed.

In Stella fsT20-3, the predicted arrival time Tx of each floor is calculated. For convenience of explanation, we will focus only on hall calls and perform calculations assuming that there are no car calls.

FIG. 19 shows the predicted arrival time Tx for each floor. Here, let us assume that the time required to go up or down between floors on the first floor is 1 second, and when the car arrives at a floor with a hall call or corner call, the time lost due to passengers getting on and off at that floor is 10 seconds. Here is a simple example. In reality, highly accurate calculations are performed.

Consider the case where the car is on the 11th floor and descends.

1 from the 11th floor (Hall sub-index H8 is 11')
It takes 1 second to travel from the 0th floor (hall sub-index Ha is "2'). Also, it takes 2 seconds to travel from the 11th floor to the 9th floor (hall sub-index H8 "3'). On the 9th floor, there is a hall call, so 10 seconds are lost as time. Therefore, from the 11th floor to the 8th floor (
The hall sub-index H8 is "4").Similarly, the hall sub-index H8 from the 11th floor to the 4th floor is "8". ) takes 27 seconds. VC: A basket on the 11th floor descends to the 1st floor, then ascends from the 1st floor to the 12th floor, and returns to the 11th floor again.
It takes 62 seconds. This time is applicable when there are hall calls on the 9th and 4th floors, and as the number of hall calls increases, the predicted arrival time to each floor increases accordingly.

Estimated arrival time RESPT (H
a) can be expressed by the following formula.

RANT (Stal, Endl) is the travel time required to travel from one stop floor to the next stop floor.

Furthermore, LO8T (Kndl) is the lost time at the scheduled stop floor. KEIKAT (H8) indicates the elapsed time since the allocation (hall call allocation) was set for the hall sub-index H8 where the hall call occurred. KEIKAT (J(S) can be considered to be 0', so it is ignored. t indicates the number of calls to the floor for which the predicted arrival time is to be calculated. ) In the case of FIG. 23, for example, the predicted arrival time RESPT (3) for the 9th floor where there is a hall call is expressed by the following equation.

RESPT (3) = RANT (S ta 1.
End, ) #2 (seconds) Since there is no call between the 11th floor and the 9th floor, the car does not stop at the 10th floor, so the lost time is "0". Similarly, the predicted arrival time RESPT (8) for the fourth floor is expressed by the following equation.

RESPT(8)=RAMT (8ta 1.End
1) + RANT (Sta 2, End 2) +
RANT (5ta3.End + LO8T (Endl) + LO8T (End
2) RANT (Sta2.End2) is the running time from the 9th floor to the 7th floor. RANT (Sta3.End
3) is the travel time from the 7th floor to the 4th floor. In the case of
RANT (Sta2.End2) is 2 seconds, RA
NT (8ta5.End5) is 3 seconds.

LO8T (Endl) is the loss time on the 9th floor, L
O8T (End 2) is the loss time on the 7th floor. Therefore, RESPT(8) is 27 seconds. With the above steps, the predicted arrival time for each floor shown in FIG. 19 is determined.

Next step 5T20-4 and Stella 7'5T20-5
will be held. Here, only for floors with hall calls, the predicted minimum arrival time Tm1n and the predicted maximum arrival time Tma
Determine the probability distribution mode of x and arrival time.

The predicted arrival time Tx is the arrival time at each floor when a derivative call is generated in response to an actual hall call, and the actual hall call, its derivative call, and car call are considered. For this predicted arrival time Tx, the predicted minimum arrival time Tm1n is the arrival time at each floor when only existing hall calls and car calls are considered. Furthermore, the predicted maximum arrival time Tmax is the arrival time at each floor when hall calls occur on all floors and when derived car calls prolong the arrival time. Predicted minimum arrival time Tm1n and predicted maximum arrival time Tmax
is determined by the above-mentioned equation (1).

In order to obtain the predicted maximum arrival time Tmax in the example of FIG. 19, as an example, we predicted the occurrence of demand from the 5th floor (5F) conference room to the 1st floor (IF) and the 12th floor (12F). (Usually, only the floors with hall calls are found.) Next, the probability distribution of arrival times is found. In order to find the probability that a car with a hall call will arrive within a predetermined time, a probability distribution mode is set for each floor with a hall call.

FIG. 18 shows two types of probability distribution modes of the arrival time at a floor with a hall call (, ) and (b). This distribution mode is distributed around the predicted arrival time Tx. Moreover, this distribution mode is the predicted minimum arrival time Tm1
n, the predicted arrival time Tx, and the predicted maximum arrival time Tmax have different distribution states. However, this distribution mode always requires a predicted minimum arrival time Tm1n and a predicted maximum arrival time T
It is set so that the areas of the 81st part and the 82nd part are equal and the sum of the areas is 1.

The selection of the two probability distribution modes shown in Figures 18(a) and (b) is based on the fact that the derived car calls frequently generated using learning data for hall calls on a certain floor match the actual car calls. It is determined by the size of the possibility.

In other words, if there is a low possibility that the derived car call that has been generated many times, such as learning data, will match the car call that actually occurs, the probability of the pattern in which TL1 + TL2 is long as shown in Figure 18 (i). A distribution mode is selected.

FIG. 17a shows a case where a derivative car call occurs on the 12th floor in response to an ascending hall call on the 5th floor. In this case, the probability distribution mode of the arrival time of someone on the third floor to the eighth and seventh floors becomes the mode shown in FIG. 18 (, ). However, Tm
1n is within the range of Tmax.

On the other hand, if there is a high possibility that the derived car call that has been generated many times, such as learning data, will match the car call that actually occurs, a probability distribution with short TLl and TL2 as shown in Figure 18 (b) will be used. mode is selected.

TLl + TL2 changes depending on its certainty. Then, the longest value of certainty K is determined. However, Tmfn.

Limited by Tmax.

That is, in the case shown in FIG. 17, there is a high possibility that a car call will actually occur on the first floor, which is the reference floor, in response to a ball call, so there is a high possibility that the car will go to the first floor.

Therefore, the probability distribution mode of the arrival time on any third floor even though the user is on the 10th floor becomes the distribution mode shown in FIG. 18(b).

Based on the probability distribution of arrival times determined as described above, the certainty of arrival times is determined. This confidence level indicates the possibility that Kaka will arrive within a predetermined time. For example, consider the probability distribution mode of arrival times as shown in FIG. 18(c). Here, the horizontal axis shows the predicted arrival time, and the vertical axis shows the probability value. The probability that the car will arrive within 30 seconds, that is, the certainty factor TPO, can be obtained by calculating the area of part A. Similarly, the confidence level TPI that the car will arrive within 31 seconds or more but less than 60 seconds can be obtained from the area of part B.
The confidence level TP2 in the case of 60 seconds or more can be determined from the area of the C portion. Since the probability distribution mode is normalized so that the area is IK, the reliability of each arrival time can be determined by determining the area. This confidence is expressed as a probability.

Above, steps 5T20-3 to 5T20 shown in FIG.
-5 gives the predicted arrival time and its confidence.

By repeating steps 5T20-3 to 8T20-7 for each layer, the reliability of the predicted arrival time for each floor and the predicted arrival time for the floor with the hall call is determined.

This completes the calculation of the predicted arrival time by Stella fsT16-2 in FIG. 16.

The membership function shown in Fig. 21 used in the inference calculation of allocation control and the error 〇 and its increment Δ shown in Fig. 13
The relationship between .DELTA.U and the instruction .DELTA.U is artificially determined as described above.

That is, the membership degree function is determined using an expert's empirical rule. Moreover, the error shown in FIG. 13(b) and its increment Δ
Which instruction to use for e is also determined using the empirical rules of experts. Therefore, in the allocation control, it is possible to make inferences based on direct expression of the empirical rules of experts, so that accurate allocation can be performed. Hall calls and the like occur probabilistically, and it is extremely difficult to accurately perform allocation calculations using mathematical formulas that take into account the probabilities. and
Accurate allocation control can be performed by using a direct expression of Iruma's empirical rule.

In addition, since the "confidence" of the predicted arrival time is considered in the allocation control, even if the predicted arrival time is the same, the value of
Since the call can be assigned to a machine with a high degree of certainty, the occurrence of long-waiting calls can be reduced. Finally, control can be exercised that leaves open the possibility of good service.

Furthermore, since the direct algorithm expressions of experts are used without using complicated evaluation formulas when determining control commands, forecast accuracy can be easily improved.

Since rules representing the algorithm can be easily added and changed, it can be easily and quickly adapted to various buildings with different traffic demands. The following objectives can be considered in elevator group management control.

(1) Reduce long-term calls. (Reduce calls that last more than 60 seconds.) (2) Increase good calls. (Increase the number of calls that can be answered within 30 seconds.) (3) Maintain good service on high demand floors. (Arrival time to high-demand floors should be within 30 seconds.) (4) Reduce the number of full passes. (Eliminating cases where the load is 80 inches or more.) (5) Assign a nearby car. (Allocate the car that arrives quickly and accurately.) etc. These are connected in a list format, and only the necessary parts are used by the operation model. The list format has a pointer to the next data, which makes it easy to connect, add, and change data.

In this way, the above-mentioned inference calculation routine is expressed in list form for each goal in the above-mentioned group management control assignment control. Therefore, each routine can be easily added or changed.

In elevator group management control, transportation capacity is increased by determining the operation model in response to traffic demand.

This driving model includes a divergent model, a convergent model, a composite model, etc., and the inference calculation according to the present invention can also be used in determining switching of these driving models. Predetermined allocation control is performed for each of the above-mentioned driving models. Inference calculations are also used in this allocation control, but the goals of each allocation control are (1) to (5) above.
) are selected from the goals of

In addition, we modeled the micro and macro traffic flow for periodic high demand that occurs during Akpu peak and lunch times, as well as temporary high demand for floors with conference rooms, etc. The present invention can also be applied to determining an operation mode that can correspond to the following.

[Effects of the Invention] As detailed above, according to the present invention, the knowledge of a designer, that is, a group management control expert, is directly utilized for control,
Since it is possible to achieve fine-grained control and at the same time to make selections that take into account the ambiguity of the conditions that form the basis of that control, it is possible to reduce failures in allocation control and always leave open the possibility of improving the service. Since the control is performed, it is possible to provide a group management control method for elevators, which has characteristics such as being able to shorten the average non-response time.

[Brief explanation of the drawing]

Figure 1 is a configuration diagram of the system used in the present invention, Figure 2 is a diagram showing an example of the configuration of the program module, and Figures 3 to 6 and 8 to 10 are data used by this system. A diagram showing an example of a table, FIG. 7 is a diagram for explaining the data communication buffer, FIG. 11 is a diagram for explaining the software module of the group management control element used in the present invention, and FIG. 12 is a diagram for explaining the present invention. FIG. 13 is a diagram for explaining the relationship between conditions, commands, and rules used in the present invention. FIG. 14 is a diagram for explaining how to obtain control commands in the present invention.
The figure is a diagram for explaining the method of determining the reliability of data used for inference in the present invention, Figure 16 is a flowchart of allocation control in the present invention, and Figures 17 and 18 are diagrams showing the probability distribution of predicted arrival times. Diagram for explanation, No. 19
The figure shows the predicted arrival time and minimum calculated in the present invention. A diagram showing an example of the maximum arrival time, FIG. 20 is a flowchart of the prediction calculation used for allocation control in the present invention, FIG. 21 is a diagram showing an example of the membership degree function used in the present invention, and FIG. FIG. 23 is a diagram showing the hall sub-index of the car used in the prediction calculation, and is a diagram for explaining the state of the derived car call with respect to the hall call. 1...Group management control device, 2...Elevator control device. 4...Elevator monitoring monitor, 6...Patent attorney Takehiko Suzue, applicant's representative for in-car control. x, y) Fig. 12 Fig. 13 (a) Fig. 13 (b) Fig. 15 (a)・ ■ Fig. 15 (b) Fig. 15 (C) Fig. 16 Fig. 17 (a) (b ) Probability value ρ (C) Fig. 18 Fig. 19 Fig. 2.0 □ΦH1) - (b) (c) Fig. 21 Fig. 22/ Fig. 23

Claims (1)

[Claims] In elevator group management control where multiple elevators are managed and controlled, the intended control purpose is achieved when obtaining predictive data to be used for control from the current situation and determining various control instructions based on the conditions based on this data. Therefore, we set control instruction contents according to the above various conditions based on empirical rules, and subjectively based on expert knowledge, we set the attribution function that determines each condition and its confidence based on the attribution value of the predicted data. The above prediction data that is artificially set based on the amount of data and used for control is given a confidence value according to its ambiguity, and this confidence value is used to calculate the condition and its attribution from the above attribution function. The effectiveness of the assumption that the control instruction was given using that condition as the degree of attribution to the control instruction set above is weighted by this degree of attribution, and is determined by inference based on the result. A group management control method for elevators, the method comprising determining control instructions.