JPH0790995B2 - Optimal allocation method for group control elevators - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a call assignment for a group control elevator, and more particularly to a method for searching for an optimal assignment solution for a known call generation.

[0002]

2. Description of the Related Art In the current group management elevators, a call allocation control system is mainly used in which a hall call is given only to a selected specific car. As a call allocation control system, a call allocation control by an evaluation function has been conventionally used. In addition, recently
For example, call assignment control using a fuzzy theory as in JP-A-1-197287 "Elevator group management control method", or neural network as in JP-A-1-275381 "Elevator group management device". Various methods such as the call allocation control used have been proposed.

On the other hand, when the optimum solution of the allocation for the known call generation is obtained, that is, when the call situation of the passengers (occurrence time of all hall calls, generation floor, destination floor, etc.) is known, a plurality of units are set. It has been practiced to find an optimum call allocation solution for each car (optimal operation of each car) and utilize this for call allocation control using the fuzzy theory or neural network as described above.

Of course, in the actual group control elevator, future call generation is uncertain when call allocation is performed. Therefore, the optimum solution for the known call generation as described above is directly applied to the call allocation control algorithm. It cannot be used, but it can be used, for example, to extract call assignment rules using fuzzy theory, or can be used as a teacher signal when learning a control law effective for group management in call assignment using neural networks. In addition to this, various data can be obtained because it is possible to obtain data that can be greatly referred to when considering the control algorithm.

By the way, when the generation states of all calls are known and a plurality of cars are assigned to those calls,
The problem of finding an optimal solution from all combinations of allocations can be regarded as a combinatorial optimization problem that is well known in the traveling salesman problem. There are various solutions to this combinatorial optimization problem, but one well-known as an approximate solution method is simulated.
There is an annealing method (Simulated Annealing method, hereinafter abbreviated as SA method), and a method for searching for an optimal solution (optimum operation) of call assignment in a group control elevator using this SA method is disclosed in, for example, Japanese Patent Laid-Open No. 62-205972. Elevator group optimal operation analysis system ”.

[0006] Here, the SA method is, for example, "Operations Research Vol. 31, No. 1, 4" issued in January 1986.
Although it is a well-known method as described in "Pages 3 to 48", the procedure will be briefly described below. Generate an initial solution X. The initial temperature T is set. Perform solution transformations (also called perturbations). The change amount Δ of the solution value due to the solution conversion is calculated. If Δ> 0, that is, if the value of the solution is improved, the solution conversion is accepted. When Δ <0, that is, when the solution value is not improved, the solution conversion is accepted with a probability of P (Δ) = exp (−Δ / T). The above steps 1 to 3 are repeated until the distribution of solution values reaches a steady state. Reset the temperature T to a slightly lower temperature. The above steps (1) to (5) are repeated until the variation in the value of the solution becomes sufficiently small. In this way, if the temperature T is controlled well, the optimum solution or the best solution close to the optimum solution can be obtained.

[0007]

By the way, in carrying out this SA method, as the above procedure, that is, the Annealing schedule, initial temperature, temperature series, steady-state determination conditions at each temperature, stop determination conditions, etc. However, there is a problem in that it is very difficult to determine this schedule, and there is a problem in that the calculation of probabilities is always involved, and thus the calculation takes a very long time.

For this reason, as a recent improvement of the SA method, for example, "JOURNAL OF COMPUTATION" issued in 1990.
AL PHYSICS 90, P161-175 ”has proposed a method called Threshold Accepting method (hereinafter referred to as TA method), and the procedure is as follows. Generate an initial solution X. Set the threshold value T. Transform the solution. The amount of change Δ in the value due to the conversion of the solution is calculated. Accept the transformation of the solution when Δ> -T, otherwise do not. That is, if the value of the solution is improved, or if the change is smaller than the threshold value T even in the case of deterioration, the conversion of the solution is accepted. The above steps 1 to 3 are repeated until the distribution of solution values reaches a steady state. Reset the threshold T to a slightly smaller value. The above steps (1) to (5) are repeated until the variation in the value of the solution becomes sufficiently small.

As described above, the TA method accepts the conversion of the solution with a certain probability in the case where the value of the solution is deteriorated, whereas the TA method accepts all the values within the range smaller than the threshold value. This is different from the SA method. As a result, it has been confirmed that the TA method is considerably simpler than the SA method in that the determination of the schedule is unnecessary and the calculation of the probability is not necessary, and that the SA method can obtain a better result.

However, in the case of an elevator, the car is m
If the number of vehicles and calls is ⁿ , the number of combinations of assignments, that is, the number of solutions is m ^n. There is a problem that it becomes a combinatorial optimization problem on a large scale and it takes too much time even if the TA method is used.

The present invention has been made in view of the above points,
Considering call assignment in a group-controlled elevator as a combinatorial optimization problem and applying a TA method to obtain a higher-quality optimal solution, the number of operations is reduced by utilizing the property peculiar to the elevator, and a search method with fast convergence Is provided.

[0012]

SUMMARY OF THE INVENTION In order to solve the above problems, the present invention seeks an optimal solution when a plurality of cars are assigned to a plurality of known hall calls, and the number of cars is substantially equal. The above-mentioned hall calls are arbitrarily assigned so as to be an initial solution, and the solution is converted by exchanging any two hall calls assigned to different cars among the hall calls. Thre as a step
The feature is that the optimal solution is searched by applying the hold accepting method.

[0013]

When the number of cars is m and the number of hall calls is n,
Although the number of solutions of the allocation is m ⁿ , according to the present invention, first, the hall calls are allocated to each car so that the numbers are even, and the initial solution is generated. Since the two hall calls assigned to different cars are exchanged, the number of solutions to be searched is from m ⁿ to about n! / ((N / m)!) It is reduced to ^m ways, and convergence is very fast. Moreover, among the solutions that have an equal number of calls assigned to each car, there is no solution with an extremely bad solution value due to the nature of the elevator, and there is a high possibility that an optimal solution is included. Not only that,
The solution transformation does not cause the solution value to be extremely deteriorated or greatly fluctuated by the solution transformation because the calls are interchanged, and the search space becomes a very smooth one suitable for the TA method.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 shows, as an example, the occurrence status of hall calls (arrival status of passengers) in a 7-floor building with time as the horizontal axis, and the assignment of these known calls is considered as a combinatorial optimization problem.

In FIG. 1, H ₁ is an ascending hall call generated by a passenger arriving at the first floor hall, H ₂ is a descending hall call generated by a passenger arriving at the fourth floor hall, and H ₃ to H ₉ shows a hall call generated at floor position shown similarly. In addition, B _{1 to} B ₃ respectively indicate passengers arriving late to the hall on the first floor after the hall call H ₁ is generated, and other ◯ marks similarly indicate passengers arriving at each hall at the illustrated time points. .

In the example of FIG. 1, the hall call H on the first floor, for example.
Passengers one basket assigned to ₁ (not shown) B ₁ .about.B
₃ also shows boarding at the same time and departing, and then a hall call H ₅ is generated by passengers arriving after that, but if the initial position of each car is different or another car is assigned to the hall call H ₁ Response time is different,
For example, if passengers B ₁ and B ₂ depart and passengers B ₃ arrive later to generate a hall call on the first floor, even if the passengers have the same time of occurrence, the halls may be allocated depending on the allocation and the initial position of each car. The call generation situation may be different from that in FIG. However, since the point of occurrence of each passenger, the destination floor and the initial position of each car are all known and the problem is treated as unchanged, the call assignment of each car is decided sequentially in the order of occurrence, and the operation of each car is automatic. Therefore, the simulation described later can be easily executed.

FIG. 2 is a flow chart for explaining the optimum solution search method according to the present invention. The case where the contents of each step are applied to the example of FIG. 1 will be described. First,
In step S1, allocation is initialized as follows. A _i = ((i−1) mod m) +1 (i = 1 ... n) where A _i represents the car number assigned to the hall call H _i , n is the number of calls and m is the car It is the number of vehicles. Note that ((i-1) mod m) indicates the remainder when (i-1) is divided by m. That is, this formula shows that n hall calls are sequentially assigned to m cars.

In the example of FIG. 1, if n = 9 and the number of cars m = 3, the initial solution of the car assigned to each hall call H _i is shown in FIG. like,
To 3 Unit 1 Unit will be assigned repeatedly in sequence on each hall call H _i. As a result, if the calls assigned to Units ₁ to ₃ are shown as C _{1 to} C ₃ for each unit, it means that each unit has 3 hall calls assigned, as shown in Fig. 3 (b). , Let this be the initial solution.

As described above, when the initial car position and the assigned car are different, the calling situation may be different. Therefore, H _{1 to} H ₉ in FIG. 1 and FIG. H ₁
~ H ₉ does not always match, here H
_{1 to} H ₉ represent the order in which the hall calls occur, and FIG. 3 (b) shows that the calls assigned to C _1, that is, the No. 1 machine, are the hall calls that occur at the 1st, 4th, and 7th places. There is.

Next, in step S2, the initial value of the threshold value T that plays an important role in the TA method is set to T _o for initialization.

Next, in step S3, a simulation is executed for the initial solution to calculate the value of the optimization objective function. Although various optimization objective functions can be considered here, here, the average waiting time t _a for each general hall call is considered as the optimization objective function.

The waiting time of each hall call by simulation is calculated as follows. First, as shown in FIG. 3 (b), the first hall call H ₁ generated is assigned to Unit 1. In FIG. 1, the hall call H _{1 in} the ascending direction on the first floor corresponds to this, and its occurrence. The waiting time for the hall call H ₁ can be easily obtained from the time point and the initial position of the car of the first car. In addition, since the response time of Unit 1 is fixed by this,
The passengers arriving from the time of occurrence of each passenger on the 1st floor to the time of the response of Unit 1 boarded the car, and the time of occurrence of the next hall call on the 1st floor generated by the passengers arriving immediately after that Determine.

This time is compared with the time when the 4th floor landing call H _{2 in} FIG. 1 is generated. If the 4th floor landing call H ₂ is earlier, this is H ₂ shown in FIG. The waiting time for the hall H ₂ is obtained from the hall call H ₂ and the car position of the _second vehicle, and the response time of the second vehicle is also determined.

In this way, according to FIG. 3 (b), when the generated hall calls are sequentially assigned to Nos. 1 to 3, since the generation time of each passenger is known, the operation of each car and the generation of the call are confirmed. Then, the waiting time for each hall call can be obtained, and in FIG. 1, as a result of the allocation,
The number of hall calls may be 10 or more, but similarly, the 10th and subsequent calls may be sequentially assigned to the respective units.

[0025] Thus the waiting time for calls each landing in the initial solution can be obtained, the average value calculated t _a, stores it as the optimal value t ^* at the present time at step S4. With this, the initialization is completed, and the search for the optimum solution is started in step S5 and subsequent steps.

In step S5, the solution is converted. Here, the solution conversion is not performed on all the solutions, but is performed by replacing only two arbitrary hall calls assigned to different cars with respect to the immediately preceding solution. That is, the hall call to be replaced is selected according to the following conditions, and this is replaced with a new solution. P (j = i) = 1 / n (i = 1 ... n) P (k = i) = 1 / n (i = 1 ... n) where j ≠ k, A _j ≠ A _k

Here, P (j = i) or P (k = i)
Represents the probability that H _j or H _k (j, k = 1 ... N) will be selected as a replacement target in the hall call H _i . Now, let us say that the cars assigned to H _j and H _k before the replacement are represented by A _j and A _k , respectively, and the cars assigned to H _j and H _k after the replacement are represented by A ′ _j and A ′ _k , respectively. , H _j and H _k are exchanged, A ′ _j = A
_k , A ′ _k = A _j .

For example, suppose that j and k are j =
If 8 and j = 9 are selected, that is, if the assignments of the hall calls H ₈ and H ₉ are exchanged with respect to the assignment of FIG. 3 (b), then A ′ ₉ = A ₈ and A ′ ₈ = A ₉ , and H _{Since 9} is assigned to Unit 2 and H ₈ is assigned to Unit 3, new solutions after conversion are as shown in FIGS. 4 (a) and 4 (b).

In step S6, a simulation is executed for this new solution in the same manner as described above to calculate the average waiting time t _a for each hall call. And step S
In 7, the difference between the average value t _a of the waiting time in the new solution and the optimum value t ^* at the present time is divided by t ^* and normalized to the range of 0 to 1, and it is determined whether or not this is smaller than the threshold value T. to decide. That is, in step S7, by converting the solution, the worse amount as the value for improved or (mean value t _a waiting time the optimum value t ^* than or becomes smaller the current) is or worse solutions are given Judge whether it is less than the value.

[0030] As a result, the average value t _a waiting time in the solution proceeds to step S8 if the corruption within or predetermined value the value of the solution is improved as the optimum value t ^*, whereas, the solution If the value has become worse than a predetermined value, the process proceeds to step S9 to undo the conversion of the solution.

In step S10, it is determined whether or not the number of searches reaches a predetermined value. If not, step S11.
Then, the value of the threshold value T is slightly reduced by a constant amount δT and reset, and then the solution conversion is repeated again. The search is repeated in this way, and when the number of times reaches the predetermined number, the search is terminated, and the value of t ^* at that time point is the optimum value of the average value of the waiting time, and the solution for the optimum value t ^* is the optimum solution for call allocation.

Needless to say, in the above description, the number of floors, the number of cars, or the number of calls (passengers) generated is not limited to the above-mentioned embodiment. Not limited to the average value of time, for example, the maximum waiting time, energy consumption and maximum transportation amount,
Alternatively, various functions such as their combination (logical sum, weighted sum) can be used.

In the above description of step S5, the probability P of selecting the hall and H _j or H _k is 1 / n.
However, at the initial stage of the search, for example, P (j =
1)> P (j = n), that is, the selection probability of the hall call with the earlier generation order is increased, and P (j = 1) <P in the latter half stage.
(J = n), that is, it is possible to make them unequal so that the probability of selection of hall calls whose generation order is late becomes large. Since a landing call with a faster generation order has a greater effect on the generation of subsequent calls by switching, a landing call with a faster generation order is selected in the earlier stage of the search in this way to improve convergence. be able to.

In the above embodiment, the hall calls are evenly assigned to each car, and the solution is converted only by replacing any two calls. Therefore, the number of calls assigned to each car is always equal. ing. This is because elevator operation and passenger generation for a sufficiently long time can be approximated by Poisson process with average arrival rate μ and average incidence rate λ, respectively.
The utilization rate ρ _i of each car at that time is ρ _i = λ _i / μ (i = 1 ... m, Σλ _i = λ), where λ _i is the passenger generation rate shared by each car. The average waiting time t _t of the whole is _{t t = Σ (1 / mμ} ) · (ρ i / (1-ρ i)), the minimum value _{ρ 1 = ρ 2 = ···· =} ρ m That is, the call allocation of each car is even (this can be easily proved).

Further, even if not considering the above formula, it is easy to predict that the result of the optimum allocation to the calls during that time is that the allocation to each car will be even if the time is considered to be long enough. Therefore, the search is performed only when the number of calls is always equal. But,
The optimal solution does not always exist when the calls are perfectly evenly assigned, and even if the assignments are slightly uneven, it is likely that a solution equivalent to the optimal solution will be included. It may include equality.

For that purpose, for example, with a certain probability, _A'j =
This can be achieved such that A _k , A ′ _k = A _k , ie by allowing the transformation of the solution to reassign only one of the hall calls H _j to another car.

Further, the optimum solution obtained by the procedure of the above embodiment is set as one solution, and the same procedure is used to obtain the optimum solution for each of the different random numbers. It is also possible to select the best one as the final optimal solution.

[00038]

According to the present invention, since the property peculiar to the group control elevator is utilized and only the uniform call allocation or its vicinity is searched, the search space is a search space for all solutions. This is much smaller than that of, and the optimum call assignment can be searched very efficiently.

Further, the quality of the solution in the search space is relatively good, and the solution is converted only by exchanging two arbitrary hall calls or by changing the allocation of one hall call. , There is little possibility that the solution value will be extremely deteriorated or greatly changed due to the solution conversion, and the smooth search space will increase the possibility of overcoming the local solution and increase the probability of reaching the optimal solution. .

That is, since the TA method uses a threshold value to determine acceptance / rejection of solution conversion, it is impossible to overcome a local solution in which the amount of change in value due to solution conversion is larger than the threshold value. However, it was difficult to apply it to an optimization problem in which the search space is not smooth, but according to the present invention, the TA method can be applied, and a more optimal solution can be efficiently obtained.

As described above, according to the present invention, it is possible to efficiently obtain a high-quality optimal allocation solution. Therefore, in the call allocation rule extraction using the fuzzy theory or in the call allocation using the neural network, It can be effectively used as a teacher signal when learning a control law effective for group management.

[Brief description of drawings]

FIG. 1 is a diagram showing an example of a hall call generation situation as a combination optimization problem.

FIG. 2 is a flow chart for explaining an optimum solution search method according to the present invention.

FIG. 3 is a diagram showing an example of an initial solution when the present invention is applied to the hall call of FIG.

FIG. 4 is a diagram showing an example of solution conversion to the initial solution of FIG.

[Explanation of symbols]

H ₁ to H ₉ hall call B ₁ .about.B ₃ Passenger A _i hall call average waiting time for car call car number T threshold t _a respective landing assigned to H _i t ^* optimum value of the average value t _a

Claims (2)

[Claims] 1. In an optimum allocation method of a group management elevator for searching and extracting an optimum solution when a plurality of cars are allocated to a plurality of known hall calls, the number is substantially equal to each car. A hall call is arbitrarily assigned and used as an initial solution, and the solution is converted by exchanging any two hall calls assigned to different cars among the hall calls, and this is used as a basic step of the search.
Optimal allocation method for group-controlled elevators, characterized by searching for an optimal solution by applying the Accepting method. 2. The optimum allocation method for a group controlled elevator according to claim 1, wherein the allocation of one of the arbitrary hall calls is changed to another car with a predetermined probability as a solution conversion.