JPH0790998B2 - Group management control method for elevators - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Object of the Invention (Industrial field of use) The present invention relates to elevator group management control, and particularly to allocation control for hall calls, which provides good allocation in response to changes in traffic demand. The present invention relates to an elevator group management control method to be performed.

(Prior Art) Elevator group management control is centered on control of allocation for hall calls and special operation control for high demand. In the allocation control for hall calls, conventionally, allocation was performed by evaluation using various predictive calculation data such as arrival time. Therefore, the failure of the prediction sometimes caused the failure of the allocation. In order to eliminate such a situation, there have been studies such as learning the daily traffic flow and demanding the reliability of the prediction data.

In the group control control of elevators, the allocation control for hall calls is conventionally performed by the method, which evaluates each machine according to the evaluation formula obtained by the function that uses the predicted arrival time, and thereby determines the allocation machine for hall calls. Was there. Although various methods are used for this evaluation formula, the following three formulas are often used.

The predicted arrival time is calculated, and the maximum predicted arrival time among the predicted arrival times for all hall calls is used as the evaluation value.

The predicted arrival time is calculated, and the predicted arrival time is used to obtain an evaluation value by using a quadratic function or a cubic function so as to match human damage.

 The average and standard deviation of the predicted arrival times are used as evaluation values.

Various damages were added to these evaluation values and the assigned number was decided as a comprehensive evaluation.

Further, with respect to changes in traffic demand, the coefficient of the above-described evaluation calculation was changed.

For this reason, in particular, when the balance demand was high, the control output sometimes became unsatisfactory due to saturation.

On the contrary, there were cases in which unnecessary operation was performed when demand was low.

(Problems to be Solved by the Invention) In the conventional method, the following problems occur due to the following two points.

When traffic demand changes, it is difficult to allocate an appropriate amount based on the change in the coefficient.

Since the evaluation formula does not show a direct goal, it becomes impossible to determine which evaluation method should be used for traffic demand. In other words, there are many interferences, and many targets change due to the change of one coefficient. The coefficient that is changed for one goal is not unique.

For this reason, there has been a problem that due to a change in the coefficient or the like, it is difficult to cope with the change in the traffic demand, and wasteful driving at the time of low demand or unsatisfactory hall call allocation at the time of high demand occurs. In other words, it is not possible to respond to a drastic change in traffic demand. It is an object of the present invention to provide an elevator group management control method which performs hall call allocation control adapted to changes in traffic demand and always provides satisfactory services.

[Structure of Invention]

(Means for Solving the Problems) In order to achieve the above object, the present invention is as follows. That is, in the hall call allocation control of the group management control that activates a plurality of elevators for a plurality of service floors and controls the operation of the elevators, the function that predicts the traffic demand is controlled by the function that predicts the traffic demand. Use a set of control rules that set goals and conditions and show expert control strategies to achieve these control goals,
The reasoning function is used to determine the final control command by using the degree to which the condition of the control rule is satisfied and the weight of the instruction.

(Operation) The expert control strategy is a practical knowledge of highly efficient group management control empirically obtained by an expert who is familiar with elevator group management (allocation control).

Next, the control rule expressed by the condition and the instruction is a rule expressed in a general form of "if A, then B". In the control by hall call allocation, a control strategy of an expert is expressed as one of the control rules expressed by conditions and instructions, such as "if long waiting, do not allocate." In this way, a plurality of rules expressing the expert's control strategy in the form of "if A then B" are prepared.

In the inference function, the degree to which the condition shown in the control rule is satisfied when a control command based on the control rule is given to the controlled object is predicted, and the instruction shown in the control rule is weighted from this degree. For example, suppose that the control rule is "if long wait, do not allocate." At that time, if "allocate a hall call." Is given as a control command to the elevator No. 1 which is one of the controlled objects, it predicts the degree to which the condition "long wait", which is the condition shown in the control rules, will be satisfied. To do. That is, the degree of long waiting due to the hall call assignment is predicted. Based on this predicted degree, weighting is determined for the instruction to “assign”.

The inference function described above determines the degree to which the condition of the control rule is satisfied and the weighting of the instruction for each of the plurality of control rules, and finally determines the control command from the weighted instruction. Here, the target value of each control is set to a point that is considered to be optimal with respect to traffic demand.

(Embodiment) An elevator group management control method according to an embodiment of the present invention will be described below with reference to FIG.

In FIG. 1, reference numeral 1 denotes a group management control device, which includes a group management control unit 1A, a knowledge engineering application unit 1B, and an auxiliary storage unit 1C.
The elevator controller 2, the transmission controller 3, and the elevator monitor 4 are connected to each other via a system bus formed of a transmission-dedicated LSI by serial transmission. The hall gate, the lamp, the sensor, and the display are connected to the I / O controller 5 by serial transmission using a transmission-dedicated LSI and general-purpose transmission software. The in-car controller 6 and the elevator control device 2 are also connected by serial transmission. I / O of building management computer 7 data, OA computer 8 data, time recorder 9A data input device 7, notification data and entrance counter 10A data
The controller 10 is coupled by the interface of the transmission controller 3 and is transmitted to the serial system bus.

The present system is an example close to the maximum specifications, and therefore, the present invention can be used even for a system without a part (performs what is input). In addition, 7A, 8A, 4A
It is an operation display system for CRT terminals and key inputs.

Next, the software configuration will be described with reference to FIG.

In FIG. 2, after the group management control device 1 (FIG. 1) is started (S), the task management program 20 determines which task is to be activated. A task is a software module for each function and is activated by a condition.

Here, each task will be briefly described.

Reference numeral 32 denotes an initialization task for initializing RAM and CPU registers and each LSI.
It is activated when the initial state or operation mode is switched.

Reference numeral 21 is an external input task for setting external inputs such as CCT (car condition table), KCT (car condition table), and HCT (hole condition table) on the RAM. This external input task 21 has high priority and 100mse
Rebooting is performed every about c. Here, HCT is an abbreviation for the hall condition table, and hall call registration is performed by the elevator control device, and the data is input.

Assuming that there are 4 units A to D in the group, and 1 to 8 floors, the above HCT, CCT, and KCT are shown in FIG. 3, respectively.
The bit configuration is as shown in FIGS. That is, in the HCT showing the hole state shown in FIG. 3, 8 for the hole sub-index (HS) of 0 to 13
8 bits of information are stored from the floor down (8D) to the floor 7 up (7U). The state of the hall on each floor will be specifically described. For example, if the up switch is pressed in the elevator hall on the 5th floor, the 7th bit of HS11 (5U) becomes 1, and if the service elevator corresponding to this hall call determines to be Unit A by the method described later, it will be 0bit and 6 of HS11. The bit becomes 1. When the above-mentioned Unit A arrives at the 5th floor, all bits 0, 6, 7 of HS11 are reset to 0. That is, 0 to 3 bits indicate the number set of each elevator, 6 bits indicate whether elevators are assigned to hall calls, and 7 bits indicate the presence of hall calls.

In the CCT showing the car state in Fig. 4, each of elevators A to D is indexed from 0 to 3
16-bit information is stored. That is, the load state of the car is shown in the binary system in bits 0 to 3. The meaning of these 0 to 3 bits is “0001” “0010” “0011” “010
0 "" 0101 "" 0110 "" 0111 "" 1000 "" 1001 "" 1010 "" 101
1 "to 1100" 0-10%, 11-20%, 21-3
0%, 31-40%, 41-50%, 51-60%, 61-70%, 71-8
0%, 81-90%, 91-100%, 101-110%, 111% or more.

5 bits indicate the running state of the car, "1" is running, "0"
Indicates that the vehicle is decelerating. 7-bit indicates the open / closed state of the door, "1"
Indicates open and “0” indicates closed. Bits 8 to 13 represent the car position in binary notation. Bits 14 and 15 indicate the moving direction of the car. “10” indicates rising, “01” indicates falling, and “00” indicates no direction, that is, stopped.

In the KCT showing the car call state in Fig. 5, H in Fig. 3
Similar to CT, bits 0 to 3 indicate the presence or absence of a car call for elevators A to D.

As a result, the elevator and hall call states have been entered.

In FIG. 2, reference numeral 22 is an allocation task for allocation.
This allocation task 22 checks the newly-generated hall call every 100 msec, and if there is any occurrence, it predicts the unanswered response time calculation subroutine 24, fills up, etc.
25 and the evaluation subroutine 23 perform composition for damage such as predicted non-response time and fullness, and determine the machine with the best composition quality.

26 is an assignment review task, and this assignment review task 26
Is a low-level task that is activated about once a second,
The assignment is changed for hall calls that are long waited, full, or predicted. 28 is each single elevator communication task, and in addition to cyclic data transmission, allocation of control output and data request as necessary, allocation cancellation, number of passengers, number of people getting off, new call And so on. These are performed using the buffer, and the data having the contents shown in FIG. 6 are transmitted in the format shown in FIG.

Reference numeral 29 denotes an annual timer and various timers, which are routines of various interval timers after 10 ms, 100 ms, one second, and an annual timer combined with them. Also, these data are corrected by an external timer.

The yearly timer has information about the month, date, day of the week, holidays, six days, and other events, and the information is updated by the floppy disk of the second I / O task 31 and the CRT of the first I / O task 30. To be done.

CRT transmission input output of the first I / O task 30,
The task for the character display terminal is used to transfer information with an external terminal or another computer. This task 30 is time-sliced and activated at a low level so as not to damage other group management rusks.

The second (flexible) floppy disk control of the I / O task 31 is started when the learning data and the like are stored in an external floppy disk. First
Like the I / O task 30, it is started at a low level. The learning data processing task 27 is a task that sets data in the state at that point in time according to external input or data from a single unit, and replaces that data when changing to the next state. It is activated when there is a change or state change. It is also a low-level task that is activated so as not to hurt high-level group management tasks. However, it changes when a special flag or priority order is changed. Here, the learning data is shown in FIG. 8 (a) (b) (c).
(D) As shown in (e), it is classified into several equivalent traffic modes according to factors such as month, date, day of the week, six days of the week, holidays, and time zones (time bands). It has the data shown in Figure 10.

Examples of these are shown in FIGS. 9 and 10. The symbols in FIGS. 9 and 10 are as follows.

HCT $ RAT: Average number of hall calls generated in 15 minutes.

HCT $ RAT: Average number of car calls generated.

IN $ RAT: Average number of passengers.

OUT $ RAT: Average number of people getting off.

KCT $ SET: Car call generation rate for each floor.

HCT $ RAT to OUT $ RAT is indicated by the index HS (hall sub-index) of the directional floor. KCT $ RAT is shown by the matrix of A and B from the A floor to the B floor.

Also, when the demand is high, those changes are learned at a detailed interval. This is AV $ MEN $ P (HS ・ t) and each HS
And t. However, t is time.

As other tasks, in FIG. 2, task 34 is started every 1 second and performs data input / output data communication with an external building management computer and data collection by it, and there is a demand using the data. There is a communication demand forecasting task 33 that predicts the traffic demand and determines the driving model.
Start each time. Further, there are various driving tasks 35 as tasks of the driving model activated by these.

Next, allocation control for hall calls in the group management control method according to the present invention will be described. The allocation control is performed by the group management actual control unit 1A and the knowledge engineering application unit 1B in the group management control device 1 shown in FIG. 11th
The figure shows a flowchart of allocation control. Further, for convenience of explanation, four elevators (Unit A, Unit B, Unit C and D
(Unit No.) shall be group controlled.

In the above description of the configuration, the case of the 8th floor has been described, but in the following description, the case of the 12th floor will be described for convenience.

Allocation control for hall calls in the elevator group management control will be described. The allocation control method differs depending on the operation mode of the elevator.

As an example, allocation control in the operation mode when the traffic flows on each floor are in a balanced state will be described.

In the allocation control, it is assumed that the following control targets are given as an example.

(1) Reduce long waiting calls.

(2) Increase good calls.

(3) Make the maximum length wait good.

(4) Maintain good service on high demand floors.

(5) Reduce the number of people who are full.

Target values for these targets are set at optimum points according to traffic demand. This method will be described later.

The above-mentioned goals are based on the knowledge possessed by experts and are expert control strategies in allocation control. Control rules represented by conditions and instructions are prepared corresponding to the respective control strategies described above. For example, the control rule corresponding to “reduce long waiting time” is “do not allocate if long waiting”. For "increase good calls", "if good calls are assigned", etc., and for other control strategies, control rules of the form "if A then B" are prepared. However, these control rules are combined by a narrower series of control rules.

In the present invention, the degree of satisfaction of the conditions shown in the above control rules and the weighting of instructions are determined for each control rule by the inference function. The degree to which the above-mentioned condition is satisfied is the degree to which a long wait occurs if the condition is “if long wait”. A high degree of this means a very long wait, and a low degree of this means a long wait. In the above reasoning, the degree of long waiting is predicted by temporarily assigning a certain elevator machine, and the instruction is weighted based on the degree. That is, if a long wait is likely to occur when the temporary allocation is performed, the weighting of the instruction is performed in the direction in which the allocation is not performed. Comprehensively evaluate the weighting of the instructions determined for each control rule, and "allocate" in allocation control
Determines the strength of the control command.

The above-described determination of the control command output strength is performed for each elevator car by using a series of control rules for each target, and allocation is performed for the car that satisfies the target to the highest degree.

In deciding the allocation, if you try to allocate to an elevator that completely meets all the above goals,
No allocation can be made. Therefore, the elevators that meet the target to the highest degree are assigned.

Therefore, a control rule is prepared based on the control strategy described above, and the strength of the control command is determined by obtaining the degree to which the condition is satisfied and the strength of the instruction.

Next, allocation control will be described. Determination of allocation in allocation control will be described using the flowchart shown in FIG. In step 1, the elevators that cannot perform allocation control due to failure, out-of-group operation, etc. are excluded from allocation control. In the next step 2, the prediction calculation of the data necessary for the inference calculation in the subsequent step 4 is performed. In step 3, preprocessing is performed to convert the various data obtained in step 2 into a form suitable for performing the inference operation in step 4. In step 4, an inference operation is performed. This inference operation weights the degree to which the condition of each control rule, which represents the expert's control strategy by conditions and instructions, and the instructions are weighted. The strength of the control command is determined from the weighted instruction for each control rule. This control command is to "assign to hall call." Each operation from step 2 to step 4 described above is performed for each machine. Further, in step 4, a series of control rules for many goals are performed respectively. Hereinafter, for convenience of description, the degree of instruction weighting is referred to as instruction strength. In step 5, the control command strengths of the respective units are compared with each other, and a command for finally allocating the newly generated hall call to the unit satisfying the target at the highest degree is output. With the above, it is decided which machine to allocate to the newly generated hall call, and the allocation control is completed.

In step 4 described above, the degree to which the condition is satisfied and the strength of the instruction are determined for each control rule, but one of the control rules will be described below. The determination of the degree of satisfaction of the condition and the strength of the instruction will be described. Therefore, in accordance with the above-mentioned control rule, in step 2, the predicted non-response time of each unit,
The minimum unresponsiveness time, the maximum unresponsiveness time, and the certainty factor of the unresponsiveness time are calculated. The certainty factor indicates the possibility that the car will arrive within a predetermined time.

The above prediction calculation will be described with reference to the flowchart shown in FIG.

In step 2a, the hole sub-index HS where the car is currently set is set. Hole sub-index in Fig. 13
Indicates HS. The hall sub-index HS takes into consideration the floor where the car is currently located and the driving direction of the car. Explaining the elevator on the 12th floor, if you have a car on the 12th floor and you want to perform descending operation, the hall sub index HS of this car will be '0'. As you descend from the 12th floor
As shown in Fig. 13, the value of the hole sub index HS becomes large. In addition, if there is a car on the first floor and the car is going up, the hall sub index HS for that car is '11'.
It gets bigger as you go up from the floor. For example, if the car of Unit A descends on the 11th floor, the hall sub index HS of this car will be '1'. In addition, the fact that the hall sub index HS of the car is '5' means that the car is on the 7th floor and the driving direction is descending. The value of the hole sub-index HS indicates the starting point of the calculation of the predicted non-response time for each floor, which is performed in the following steps.

Then step 2b is performed. In step 2b, a derived car call for the already registered hall call is generated based on the learning data or the like. This derived car call is a call generated imitatively to a hall call and may be different from an actual call. Figure 14 shows the hall call and its derivative calls. For example, the hall call is 9
Assuming they are on the 4th and 4th floors, their derived basket calls are
It is the 7th floor for hall calls on the 9th floor and the 2nd floor for hall calls on the 4th floor. After the generation of the derived car call is completed as described above, step 2c is performed.

In step 2c, the predicted non-response time Tx of each floor is calculated. For convenience of explanation, the calculation is performed assuming that the call is focused only on the hall call.

Figure 15 shows the predicted non-response time Tx of each floor. Here, the time required to rise or fall between floors is 1 second,
When a car arrives at a floor with a hall call or a car call, the loss time is 10 seconds due to passengers getting on and off the floor.

Consider the case where the car descends on the 11th floor. It takes 1 second from the 11th floor (Hall sub index HS is '1') to the 10th floor (Hall sub index HS is '2'). Also, it takes 2 seconds from the 11th floor to the 9th floor (Hall Sub Index HS is '3'). On the 9th floor, there is a hall call, so 10 seconds are spent as lost time. Therefore 11
Floor to 8th floor (hall sub index HS is '4')
It will take 13 seconds. Similarly, the 11th to 4th floor hall sub-index HS is '8'. ) Takes 27 seconds. The car on the 11th floor descends to the 1st floor, then 1
It took 62 seconds to climb from the first floor to the 12th floor and return to the 11th floor again. This time is when there are hall calls on the 9th and 4th floors, and as the number of hall calls increases, the predicted non-response time to each floor also increases accordingly.

Predicted unanswered time RESPT (HS) for floors with hall car calls
Can be expressed by the following equation.

RANT (Sta _i , End _i ) is the travel time required to travel from one stop floor to the next stop floor.

LOST (End _i ) is the time lost at the scheduled stop floor.
KEIKAT (HS) indicates the elapsed time since the allocation was set for the hall sub index HS where the hall call occurred. KEIKAT (HS) can be considered as '0', so it is ignored. l indicates the number of calls to the floor for which the predicted unanswered time is obtained. (Including the call of the floor for which the predicted unanswered time is to be obtained.) In the case of FIG. 14, for example, the predicted unanswered time RESPT (3) of the 9th floor where there is a hall call is expressed by the following equation.

RESPT (3) = RANT (Sta ₁ , End ₁ ) ≈ 2 (sec) There is no call between the 11th floor and the 9th floor, so the car does not stop on the 10th floor, so the loss time is '0'. Similarly, the predicted non-response time RESPT (8) on the fourth floor is expressed by the following equation.

RESPT (8) = RANT (Sta ₁ , End ₁ ) + RANT (Sta ₂ , En
_{d 2) + RANT (Sta 3} , End 3) + LOST (End 1) + LOST (End 2) RANT (Sta 2, End 2) is the travel time to 7 levels 9. RANT (Sta ₃ , End ₃ ) is the traveling time from the 7th floor to the 4th floor. In this case, RANT (Sta ₂ , End ₂ ) is 2 seconds,
RANT (Sta ₃ , End ₃ ) is 3 seconds. LOST (End ₁ ) is the lost time on the 9th floor, and LOST (End ₂ ) is the lost time on the 7th floor. Therefore, RESPT (8) is 27 seconds. This is the first
The predicted non-response time for each floor shown in Fig. 5 is obtained.

Next, step 2d and step 2e are performed. here,
Minimum unanswered time Tmin only for floors with hall calls,
The maximum non-response time Tmax and the probability distribution mode of the non-response time are determined.

The predicted unanswered time Tx is the arrival time of each floor when a derivative call is generated for an actual hall call and the actual hall call, its derivative call, and the car call are considered. The minimum predicted unanswered time Tmin for this predicted unanswered time Tx is the arrival time of each floor when only actual hall calls and car calls are considered. Also, the maximum unanswered time Tmax is the arrival time of each floor when hall calls of all floors occur. The minimum non-response time Tmin and the maximum non-response time Tmax are obtained by the above equation (1).

Next, the probability distribution mode of the non-response time is obtained. In order to determine the possibility that a car with a hall call will arrive within a predetermined time, the probability distribution mode of each floor with a hall call is set. Fig. 16 shows two types of probability distribution modes of unanswered time for floors with hall calls Fig. 16 (a) and (b)
Indicates. This distribution mode is distributed around the predicted non-response time Tx. In addition, this distribution mode has a minimum unanswered time Tmin,
The distribution state varies depending on the values of the predicted non-response time Tx and the maximum non-response time Tmax. However, this distribution mode always exists between the minimum dead time Tmin and the maximum dead time Tmax,
The areas of S ₁ and S ₂ are equal, and the sum of the areas of both is 1 so that TL1 and TL2 are equal. The choice of the two probability distribution modes shown in Fig. 16 is determined by the possibility of matching the derived car call generated by learning data etc. for the hall call on a certain floor with the actual car call. . That is, when the derived car call generated by learning data or the like is unlikely to match the car call actually generated, the probability distribution mode of the pattern shown in FIG. 16 (a) is selected. Fig. 17 (a) shows the case where a derived car call is generated on the 12th floor in response to an ascending hall call on the 5th floor. In this case, the probability distribution mode of the non-response time to the 8th floor and the 7th floor of the car on the 3rd floor is the mode shown in FIG. 16 (a).

On the other hand, if the derived car call generated by learning data etc. is likely to match the car call actually generated,
The probability distribution mode of the pattern shown in FIG. 16 (b) is selected.

That is, in the case shown in FIG. 17 (b), there is a high possibility that a car call will actually occur on the first floor of the reference floor with respect to the hall call, so the possibility that the car will go to the first floor is high. Therefore, the probability distribution mode of the non-response time on the 3rd floor of the 10th floor car is the distribution mode shown in FIG. 16 (b).

The certainty factor of the non-response time is obtained from the probability distribution of the non-response time obtained as described above. The certainty factor indicates the possibility that the car will arrive within a predetermined time. In FIG. 18, the horizontal axis represents the non-response time, and the vertical axis represents the probability value. Basket
The probability of arrival within 30 seconds, that is, the certainty factor TPO is obtained by determining the area of the A portion. Similarly, the confidence TP1 that a car will arrive in 31 seconds or more and less than 60 seconds can be obtained from the area of the B part, and the confidence TP2 in the case of 60 seconds or more TP2
Can be obtained from the area of the C portion. Since the probability distribution mode is normalized so that the area becomes 1, the certainty factor of each non-response time can be obtained by finding the area. This certainty factor is expressed as a probability.

As described above, the predicted non-response time and the certainty factor thereof are obtained by step 2c, step 2d and step 2e shown in FIG. By repeating steps 2c to 2g for each floor, the confidence level of the predicted non-response time of each floor and the predicted non-response time of the floor with the hall call is obtained. This completes the calculation of the predicted non-response time in step 2 of FIG.

The inference operation of allocation control in step 4 shown in FIG. 11 will be described.

In the inference calculation of step 4, calculation of allocation control is performed. In this allocation control, with the goal of "reducing the long waiting time", that is, "0" for a long waiting call of 60 seconds or more, when a newly generated hall call is tentatively allocated, It is difficult to allocate to a machine that is more likely to become, that is, the deviation from the target of "reducing the long waiting time" by making a temporary allocation (hereinafter,
Say an error. ) E and the error increment ΔE are used to numerically represent the strength of the instruction “assign to hall call”. Determine the strength of this instruction for each elevator.

The inference calculation of allocation control will be described with reference to the flowchart shown in FIG.

In step 4a, a control command for "temporarily assigning a newly generated hall call" is given to elevator No. A, which is a control target.

Then step 4b is performed. The error E from the target and the error increment ΔE are calculated from the following equations.

Confidence; Confidence that a car will arrive in 60 seconds or more TP ₂ n; Number of all hall calls including temporary allocation From the above formula (2), before temporary allocation to Unit A and after temporary allocation. increment values and error E _a2 error E _a2 delta
Find E _a . If the value of the error E before the temporary allocation is E _a1 and the error E after the temporary allocation is E _a2 , the increment ΔE _{a of the} error E is calculated by the following equation.

ΔE _a = E _a2 −E _a1 (3) Next, using the membership function, the error E _a2 and the error increment ΔE
to evaluate the value of _a. Here, the membership function will be described.

In general, when considering whether or not a certain object is an element of the set A, a membership function is used to consider the degree of being an element of the set A, rather than strictly dividing it. In the membership degree function shown in FIG. 20, the horizontal axis represents the error E and the error increment ΔE, and the vertical axis represents the membership degree. FIG. 20 shows the respective membership functions of the set ZO, the set PM, and the set PB as a set. The set ZO is a set in which the error E and the error increment ΔE are “approximately zero”, the set PM is a set in which the error E and the error increment ΔE are “positive and medium”, and the set PB is the error E and the error increment ΔE. Indicates that the set is “positive and large”. Each membership function has all errors E
Alternatively, the degree of inclusion of each value in the set ZO, the set PM and the set PB is given to the error increment ΔE. This degree indicates the degree of belonging to a set, and is called the degree of membership, which is indicated by a number between 0.0 and 1.0. A degree of membership of 1.0 indicates that the target is a complete element of set A, and a degree of membership of 0.0 indicates that the target is not a complete element of set A.

For example, when the error E is e, consider the degree of belonging. As can be seen from FIG. 20, the degree of membership of the set PB with respect to the error e is 0.7, and the degree of membership of the set PM is 0.3. That is, the error e belongs to the set PB “positive and large” with the degree of belonging of 0.7, and belongs to the set PM “positive and medium” with the degree of belonging of 0.3.

Next, the degree to which the condition of the control rule is satisfied is obtained. That is, since this condition is "if long wait", the degree of long wait is calculated. This degree is represented by the degree of belonging of the error E and the error increment ΔE. Therefore, the degree to which the condition is satisfied is obtained by obtaining the degree of belonging of the error E and the increment ΔE of the error.

The degree of membership of the error E _a2 after provisional allocation and the increment ΔE _{a at} that time is obtained from the membership function shown in FIG. From FIG. 20, the degree of membership of the error E _a2 with respect to the set PM is 0.9, and the degree of membership with respect to the set ZO is 0.1. Also, the set P of error increments ΔE _a
The degree of belonging to M is 0.5, and the degree of belonging to the set ZO is
It is 0.5.

As described above, to which of the sets ZO, PM, PB the values of the error E and the increment ΔE of the error belong was determined in consideration of the degree of belonging. This membership function gives an evaluation of the value of the error E and the error increment ΔE as to whether the value is large or small. That is, the fact that the value of the error E belongs to the set PB means that the value is large, and that the larger the degree of belonging to the set PB is, the larger the value of the error E is. The same applies to the value of the error increment ΔE. Since the degree of long waiting is represented by the set to which the error E and the error increment ΔE belong and the degree of belonging thereof, a large error E and an error increment ΔE mean a large degree of long waiting.

As shown in FIG. 20, the error E _a belongs to the set PM with a degree of belonging of 0.9, and the error increment ΔE _a has _a degree of belonging to the set PM and the set ZO.
Since it belongs to 0.5, it means that it will be a long wait.

The strength of the instruction shown in the above-mentioned control rule is determined from the evaluation result of the error E and the error increment ΔE described above. This evaluation result is the set of error E and error increment ΔE ZO, PM, PB
That is, what degree of membership belongs to which set.

The condition-instruction table of step 4d in FIG. 19 is shown in FIG. FIG. 21 shows the instruction ΔU corresponding to the error E and the error increment ΔE. For example, the error E is “approximately zero” (set ZO), and the error increment ΔE is “positive and large”.
In the case of (set PB), “no need to allocate” is obtained as the instruction ΔU. There are five types of instructions ΔU, PO is “allocated”, PS is “allocated”, ZO is “normal”, NS is “not required to be allocated”, and NE is “not allocated”. The number of sets as error E is set PB, set
There are three types, PM and set NE, and also three types in the case of the error increment ΔE, and there are nine types of rules depending on the combination of the error E and the error increment ΔE. Therefore, nine kinds of rules are considered, and the rules are shown in FIG. Rule 1 means that when the error E is “positive and large” and the error increment ΔE is “positive and large”, the instruction ΔU is “not assigned”, and the same applies to rule 2 and thereafter.

As shown in FIG. 21, the condition-instruction table indicates the condition Δ based on the set to which the error E and the increment ΔE of the error belong.
It determines U. Therefore, the condition-instruction table shown in FIG. 21 is a matrix of the production rules represented by "if A, then B" type. Further, the instruction for the condition is artificially determined, and is based on the control strategy based on the expert knowledge.

Next, from FIG. 21, the instruction ΔU is obtained by evaluating the error E _a2 and the error increment ΔE _a . Error E _a2 is set PM and set Z
Belongs to O and the error increment ΔE _a belongs to the sets PM and ZO. Therefore, when the instruction ΔU is obtained as the condition and the set to which the error E _a2 and the increment ΔE _a belong, there are the following four ways.

(B) The error E _a2 belongs to the set PM and the increment ΔE _a is set P
If it belongs to M, the instruction ΔU becomes NE (not assigned).

(B) Error E _a2 belongs to set PM and increment ΔE _a is set ZO
If it belongs to, the instruction ΔU becomes ZO (normally assigned).

(C) Error E _a2 belongs to set ZO and increment ΔE _a is set PM
If the instruction belongs to, the instruction ΔU is NS (it is not necessary to allocate).
Becomes

(D) Error E _a2 belongs to set ZO and increment ΔE _a is set Z
If it belongs to O, the instruction ΔU becomes PO (assign).

As described above, four rules are extracted from the above-mentioned nine types of rules by the combinations (b) to (d) of the set to which the error E _a2 belongs and the set to which the increment ΔE _a belongs. The extracted rules are rule 5, rule 6, rule 8 and rule 9 shown in FIG. 4 shown in four rules for error E _a2 and increment ΔE _a
We got four instructions ΔU, but we can't give these four instructions ΔU to the elevator with the same strength. That is, among the four rules, there are rules that can be applied strongly and rules that can only be applied weakly. Therefore, the instructions given by each rule are compared according to the degree to which the conditions of the rule are satisfied. That is, weighting the instructions of each rule,
The weighted average of the weighted instructions is used to determine the strength U of the instruction.

The weighting of the instructions of each rule and the strength of the instructions obtained by weighted averaging will be described with reference to FIG. 23rd
In the figure, the horizontal axis of the graph of error E and increment ΔE for each rule is the value of error E or increment ΔE, and the vertical axis is the degree of membership. Further, the positive direction on the horizontal axis of the graph showing the instruction ΔU indicates the assigning direction, the negative direction indicates the non-assigning direction, and the vertical axis indicates the belonging degree.

With respect to Rule 5 shown in FIG. 23, for error E the set PM is filled to a degree of 0.9 and for error increment ΔE the set PM is filled to a degree of 0.5. The degree to which Rule 5 is satisfied is the degree to which the two sets are satisfied.
It will be a small value. Therefore, Rule 5 will be satisfied with a degree of 0.5. The set indicating the instruction ΔU is limited by the degree of 0.5. Similarly, rule 6, rule 8, rule 9
A set indicating the instruction ΔU is obtained. Step by step
4c is finished.

Next, step 4e is performed. In step 4e, the logical sum of the set of instructions for each rule obtained in step 4c is taken, and the weighted average is calculated by the degree of belonging to the set to finally obtain the strength U of the instruction. Here, the strength U of the instruction is -0.69 as shown in FIG. The above-mentioned instruction strength U indicates the degree of weighting with respect to the instruction "not assign" indicated in the control rule "do not assign if long wait".

As described above, the determination of the degree of satisfaction of the condition and the weighting of the instruction has been described based on the control rule that “there is no allocation if waiting for a long time”. The degree to which the condition is satisfied and the weight of the instruction are determined. The weight of the instruction obtained for each control rule, that is, the strength of the control command is determined from the strength U of the instruction. The control command referred to here is "assigned to hall call". This control command is determined for each unit. With the above, step 4e is completed and step 4 shown in FIG. 11 is completed.

Next, in step 5, it is determined which vehicle is finally assigned based on the strength of the control command obtained for each vehicle in step 4. After deciding which car number to allocate to the newly generated hall call, a control command "allocate" is output to that car number.

By ending step 5, the allocation control for hall calls is completed.

The relationship between the membership function shown in FIG. 20 and the error E shown in FIG. 21, the increment ΔE, and the instruction ΔU used in the inference operation of the allocation control is artificially determined.

That is, the degree-of-attribute function is determined by using an expert's rule of thumb. Further, which instruction to use for the error E and its increment ΔE shown in FIG. 21 is also determined by using the empirical rule of the expert. Therefore, in assignment control, inference can be performed by a direct expression of an expert's empirical rule, and accurate assignment can be performed. Hall calls etc. occur probabilistically, and it is very difficult to perform the assignment operation accurately with a mathematical formula in consideration of the probability, but as shown in the above inference operation, various data are weighted. Accurate allocation control can be performed by using a direct expression of human experience.

Also, in allocation control, the "confidence" of the predicted non-response time
Therefore, the number of long-waiting calls can be reduced because it can be assigned to a machine with a high “confidence” of the same predicted unanswered time.

Since the expert's direct algorithmic expression is used when determining the control command without using a complicated evaluation expression, it is possible to easily improve the forecasting accuracy, and it is easy to add or change the rule that is the expression of the algorithm. Therefore, it can be easily and quickly adapted to various buildings with different traffic demands. The following goals can be considered in group control of elevators.

(1) Reduce long waiting calls.

(2) Increase good calls.

(3) Reduce the longest waiting call.

(4) Maintain good service on high demand floors.

(5) Reduce the number of people who are full.

(6) Reduce first-come-first-served basis.

(7) Increase early calls.

These are just examples, and there are other automatically generated goals and a series of rules.

The inference operation routine described above is represented in a list format for each target in the allocation control of the group management control. Therefore, it is possible to easily add or change each routine.

Here, the algorithm for the above inference routine will be described again by using mathematical expressions and the like.

It is a general software package that can be used to determine the operation amount (output by control instruction) in the control of various group management controls.

The processing of the pan-software package will be described below again.

The terms used in the following description are general terms.

From the current state of the elevator group system and the possible manipulated variable, predict the future state assuming that the manipulated variable is output, and calculate the predicted control deviation and deviation rate for multiple target values. Thus, the evaluation index of the manipulated variable is configured to determine the optimum manipulated variable.

The set of changes in the quantized manipulated variable is DU.

DU = {Δu1, Δu2, ... Δun} E is the control deviation for a certain target value and the change rate of the control deviation is
With DE, the control law is expressed as follows.

Rij: if (DE is Δui → E is Aj and DE is Bj) then W is
Cj means that, if the control deviation E satisfies the property of Aj and the change rate DE of the control deviation satisfies the property of Bj when Δui is selected as the change in the manipulated variable, then the change in the manipulated variable Δui. On the other hand, Cj is taken as the evaluation index W. "

(Note) In this example, Aj, Bj,
An example in which Cj is common is shown.

If ΔUi is different, it is necessary to set Aij, Bij, and Cij.

Here, the membership function representing Aj, Bj is μAj (e),
Let μbj (Δe) and the predicted values of e and Δe be e ^ and Δe ^.
Further, the membership function representing Cj is μCj (w).

At this time, the membership function μPj (e ^, Δ
e ^ | Δui) is μPj (e ^, Δe ^ | Δui) = μAj (e ^) ^ μbj (Δe ^) (^
= Min) where, in DE is Δui (E is eO and DE is Δe
O) is observed, the inference of the value W regarding the evaluation index W with respect to Δui by the control law Rij is μCj (eO, ΔeO, W | Δui) = [μAj (e ^) ^ μbj (Δe
^)] ΜCj (w) = Pj μCj (w) where Pj indicates how much the observed fact (eO, ΔeO) satisfies the conditional part of the control law.

With M control rules, Becomes

Here, W * is determined as follows: W * = ∫WμR (eO, ΔeO, W | Δui) d (W) / ∫μR
(EO, ΔeO, W | Δui) d (W) That is, the center of gravity of μR (eO, ΔeO, W | Δui) is calculated to determine W *.

The above is performed for a plurality of target values r to determine the manipulated variable Δui which is the optimum evaluation index.

This control system is shown in FIG.

Here, the description of FIGS. 25 (a) to 25 (d) will be given.

(A) e ^ (K) = r (K) −Y ^ (K) (b) Δe ^ (K) = e ^ (K) −e ^ (K-1) (c) W * = ∫WμR ( e ^ (k), Δe ^ (k), W | Δu
i) d (W) / ∫μR (e ^ (k), Δe ^ (k), W | Δui)
d (W) Based on a plurality of targets r, Δui that is the optimum W * is determined as Δu.

(D) u (K) = u (K-1) + Δu * (K) (e) is shown in FIG.

Here, another example of the present embodiment will be shown, and the above description will be described in more detail. However, the method is exactly the same as the previous example.

In these control rules, e, Δe, W, and Δu are quantized and handled.

e and Δe are the three values shown below. (Aj, Bj) PB: Very big.

PM: Somewhat big.

ZO: It's near zero.

W has 7 values. (Cj) PB: Large on the plus side (appropriate) PN: On the plus side.

PS: Small on the plus side.

Z: It is near zero.

NS: Small on the minus side.

NM: It's on the minus side.

NB: Large on the minus side. (Inappropriate) Further, Δu is a control instruction such as allocation.

Here, as an example, the condition-instruction for three target values is De
The cision table format is shown in Fig. 27 (a) to (c). These are common to Δu. The goal of (a) in FIG. 27 is “eliminate long waiting”, the goal of (b) is “to increase the number of calls that can be answered quickly”, and the goal of (c) is “maximum waiting is good”. I will. "

When the above is linked to the hall call assignment control in the elevator group management control, the change in the operation amount Δui indicates the hall call assignment control, and from the target value when the hall call is provisionally assigned to each unit. The allocation control that minimizes the error is performed. In addition, since the condition part can have a vague element, the vagueness of data prediction can be absorbed.

Next, the target and the output function will be described.

Figure 28 shows multiple goals and the output function that displays them.
These output functions are functions using the predicted non-response time distribution.

However, when K is the current time, the error is ei = yi ^ (K + 1) and the error increment is Δei = yi ^ (K + 1) −yi ^ (K). (I: Target number) Tup60sec (I): Predicted non-response time for hall call I
Possibility of 60 seconds or more Tmin (I): Minimum predicted unanswered time for hall call I Tmax (I): Maximum predicted unanswered time for hall call I. Here, the target values b1,2,3 change with traffic demand. This part will be explained later.

Other target items include (d)-(g) shown below.

(D) Increase good calls. (Increase the possibility within 30 seconds.) (E) Keep the service on a specific floor in good condition. (Increase the possibility of reaching a specific floor within 30 seconds) (Note) Setting the top floor and the bottom floor will prevent dumpling operation.

(F) Reduce the number of full cancellations.

(G) Decrease first-come-first-served basis.

Next, the membership function will be shown as an example in which the membership function is slightly modified.

μAij and μBij (i: target number, j: quantization index) will be described with reference to FIG. Figure 28 shows the membership functions for error values and error increments. In this case, the quantization index of j is ternary, and 1: ZO,
2: PM, 3: PB. Where the membership function is the minimum value a,
It was designed to change depending on the maximum value b and the central ratio c. This figure is an example of a membership function when the X axis is normalized, and is an S function and a Z function. Details are shown below.

(S function) (Z function) z (u | a, b, c) = 1-s (u | a, b, c) (i) Membership function for ZO (μAi1, μBi1) (Ii) Membership function for PM (μAi2, μBi2) (when u ≦ c) (When u ≧ c) (Iii) Membership function for PB (μAi3, μBi3) Next, the constants a, b, and c of the meccaship function for the target i are shown below.

Here, a is the minimum value, b is the maximum value, and c is the center ratio, which can be changed.

The membership function constants are those shown in FIG. 30, where μA is the membership function for e, and μB
Is a membership function for Δe.

Membership function μCj (j: Quantization index) Figure 31 shows the output synthesis membership function. Here, the quantization index of j is 7 and is 1: PB ↑ (appropriate) 2: PM 3: PS 4: Z (normal) 5: NS 6: NM 7: NB ↓ (inappropriate). Denotes a degree of good or bad for the instruction. In addition, the shape uses a triangle with a base of ± 1 (length 2).

The Decision table, which is the rule for each goal, uses 1-7 above.

Next, the composition of rules between goals will be explained.

The output of each degree of conformity to the target ri (i = 1-3) is set to wi, and the overall composite value W is weighted by αi And

The allocation is output to the best machine of W.

The detailed explanation of the inference routine is given above.

Next, the forecast of traffic demand and the routine for changing the target value by it are shown. This routine operates in the 1B knowledge engineering application section of FIG. Traffic demand forecast is performed in step 1 of FIG. This is for the top floor, the middle floor, and the reference floor, which are combined from the smallest predicted maximum unanswered time of each unit (group number / 2). However, this is the number shown as an example. If the number of groups is four, then two units are combined, if the number of groups is five, then three units are combined, and if the number of groups is six, then three units are combined. In other words, if at least half of the number of units in those floors are in good service, then the system will be in good condition and the demand for the entire system will be low. (The reverse becomes worse and the demand becomes higher.) Floors with these calculations set in advance (top floor, intermediate floor, standard floor)
The maximum value is used as the traffic demand for the elevator group system. Also, make some corrections (± 10%) based on the learning data.

Next, in step 2, the importance of the rule matching the traffic demand value is obtained. It has a curve like that shown in Fig. 33 in memory beforehand and takes its importance according to the traffic demand. This curve is based on expert know-how. a is a curve of the target "eliminate long waiting", b is a curve of the target "enhance calls that respond early", and c is a curve of "make the maximum long waiting good".

In this example, the emphasis is placed on the "improve calls that respond quickly" when the demand is low, and the focus is shifted to the "elimination of long wait" as the demand increases.

As described above, the target can be changed according to the change in traffic demand. Corresponding to the change (0 to 100%) bx1 to 3 of the target, the target values b1 to 3 described above are changed.

Becomes However, i is a target number.

As described above, it is possible to realize the elevator group management control in which the target priority point can be changed according to the traffic demand and the service that is always satisfactory can be performed.

(Other Embodiments) In the above embodiment, the weight α in the case of combining the rules for each target in changing the priority points
Similar effects are obtained when i is changed.

Further, even if this inference routine is used as it is for the traffic demand determination routine, it can be similarly performed, which is more effective.

〔The invention's effect〕

We were able to directly connect the expert's control strategy to the control, and also change the priority target according to changes in traffic demand, and realized hall call allocation control that was always convincing from low demand to high demand. .

[Brief description of drawings]

FIG. 1 is a system configuration diagram of an elevator group supervisory control device according to an embodiment of the present invention, FIG. 2 is a software configuration diagram for realizing the same embodiment, and FIGS. 3 to 10 are the same embodiment. FIG. 11 is a flow chart of the allocation control of the same embodiment, FIG. 12 is a flow chart of the predictive calculation used for the allocation control, and FIG. 13 is a hall of the car used in the predictive calculation. Fig. 14 shows the sub-index, Fig. 14 shows the state of the derived car call with respect to the hall call, and Fig. 15 shows the calculation result of the predicted arrival time.
Figure 16 shows the probability distribution mode of predicted arrival time, and Figure 17 shows the state of derived car calls for hall calls.
FIG. 18 is a diagram for obtaining a certainty factor of predicted arrival time, FIG. 19 is a flowchart of inference operation, FIG. 20 is a diagram showing a membership function, and FIGS. 21 and 22 are conditions-instruction diagrams, 23rd
Fig. 24 is a diagram for obtaining a control command, Fig. 25 is a configuration diagram of a control system, Fig. 26 is a diagram showing a control law and a membership function, Fig. 27 is a diagram showing condition-instruction, Figure 28 shows the goal and output function, Figure 29 shows the membership function,
FIG. 30 is a diagram showing the constants of the membership function, FIG. 31 is a diagram showing the membership function for output composition, and FIG. 32 is a flow chart diagram for determining the importance of the rule matching the traffic demand value. The figure shows the relationship between traffic demand value and importance. 1 ... Group management control device, 2 ... Elevator control device, 3 ... Transmission controller, 4 ... Elevator monitoring monitor, 5 ... Hall gate, lamp, sensor, display I / O controller, 6 ... In-car controller.

Claims (1)

[Claims] 1. A group management control method for elevators, which allocates an optimum elevator from a plurality of elevators to a generated hall call, and responds to the predicted demand of the traffic demand by a function of predicting the traffic demand. Set the control target, using a plurality of control rules represented by the conditions and instructions prepared corresponding to this control target, the hall call is provisionally assigned to each of the plurality of elevators, for each elevator The degree of satisfaction of the conditions in the plurality of control rules and the strength of the instruction are calculated by the inference function, and the result of synthesizing the strength of the instructions calculated for each control rule is set as the strength of the control command of each elevator. An elevator group management control method characterized in that an optimal elevator is determined from the strength of a control command and assigned to the hall call.