JPH05204891A - Method and device for schedule planning - Google Patents

Abstract

PURPOSE:To obtain optimal solution within allowable realistic finite time by a universal computer by generating an objective function from an environmental variable and a state variable, and finding the minimum value of the objective function. CONSTITUTION:This device is equipped with a means 1 which inputs the environmental variable representing environment for which an instructed plan is targeted, and the state variable representing a plan that can be taken in supplied environment, objective function generating means 2-5 which generate the objective function from supplied environmental variable and state variable, a minimum value retrieval means 2-5 which find the minimum value of the objective function F(x), and an output means 6 which outputs a retrieval result at the minimum value retrieval means 2-5. When an objective function value by a present plan candidate before change is compared with the objective function value by a next plan candidate after change and no satisfactory value can be obtained(no reduction of the objective function value is attained), i.e. when it is judged that a process is a retrieval process arriving at the optimal solution even when no improvement is attained, such plan after change can be employed as the optimal solution retrieval route.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a planning method, and more particularly to a method and apparatus for solving a problem in which the number of combinations or permutations is enormous and there is always an optimal planning method at an extremely high speed.

[0002]

2. Description of the Related Art In various fields, methods have been proposed for obtaining an optimum solution that maximizes or minimizes various parameters. For example, "(1) Flexible production scheduling using fuzzy inference" (Journal of Fuzzy Society of Japan, Vol.3, No.2 pp339-346, 1991), (2) "Development of knowledge-based scheduling system" (Hitachi Microcomputer Technique) , Vol.4, No.2 pp56-61, 1990) etc., many methods of making good plans in a very short time using empirical information have been proposed and put to practical use. It has been.

However, such knowledge engineering, fuzzy or OR (operations research) methods are not always satisfactory in the following three points.

First of all, the optimum solution cannot be obtained. Second, there is a lack of flexibility in changing the target problem. Thirdly, the structure of the planning device is complicated and enormous manufacturing man-hours and costs are required. The following is a brief description of each.

Regarding the first point, it was thought that there is only one type of method for obtaining the only optimum solution from among a huge number of combinations.

This is a method called "enumeration method", in which all combinations are examined and evaluated to detect the optimum solution.

However, according to this method, it can be said that it is impossible to solve a real problem within an allowable realistic finite time.

For example, in order to visit only 24 points once and find the shortest route, 24! It is necessary to consider the combination of streets, but 24! Is much larger than 10 ²⁰ , so even if it is a “high-speed computer” that requires only 10 to ⁶ seconds to find one route, it is 10 ¹⁴
Seconds, that is, 10 ⁶ years or more will be required.

Further, it can be said that the demand for optimization targeting about 24 is actually the smallest case. For the above reasons, it is impossible to find an optimal solution within an allowable finite time, and so research has begun in what is called OR (Operations Research).
Since it was researched mainly on knowledge engineering and fuzzy theory, it can be said that the optimum solution cannot be obtained. The second point, the lack of flexibility, is emphasized as the speed and optimality are pursued due to the nature peculiar to knowledge engineering. That is, in the conventional method, a good planning system is nothing but knowing in detail the specific phenomenon of the target problem. Therefore, if any change in the target occurs, the system will fail at all. In recent years, in fields such as the manufacturing industry where high-mix low-volume production has become a major issue, frequent equipment changes and expansions have become the most important issues in order to address this, but as mentioned above, , It is considered that introduction of expert planning system applying knowledge engineering, fuzzy, etc. should be avoided in such fields. With respect to the third point, the above description of the second point applies as it is. In other words, in order to get a good system, learning and research on the problem
It is necessary enough, and the expression requires a lot of man-hours and cost.

[0010]

As described above, when considering, for example, the solution of the scheduling problem, the method using the conventional technique does not provide a satisfactory solution, and all the methods are incomplete. Was there. In addition, the calculation was stopped based on the operator's judgment and the calculation allowable time, etc. without any idea of when to stop the calculation. However, there is a problem that the obtained solution lacks generality.

Further, there has been expected an invention of a solution for various problems, which is not affected by changes in the target.

An object of the present invention is to solve the above-mentioned problems, to solve various problems, and to solve an optimal solution within a realistic finite time allowed by a general-purpose computer without human intervention. It is to provide a planning method and apparatus.

[0013]

[Means for Solving the Problems] In order to solve the above problems, the following means can be considered.

In order to create an objective function for setting an optimum plan, an environment variable representing an environment of a given planning target and a state variable representing a plan that can be taken in the given environment are input. Means, an objective function creating means for creating an objective function from given environment variables and state variables, a minimum value searching means for finding the minimum value of the objective function F (x), where x is the state variable, and a minimum value searching means. It is a planning apparatus having an output means for outputting the search result in (1).

Further, in order to create an objective function for setting an optimum plan, an environment variable representing a given environment of a plan object and a state variable representing a plan that can be taken in the given environment are input. Input means, an objective function creating means for creating an objective function from given environment variables and state variables, a minimum value searching means for obtaining the minimum value of the objective function F (x), where x is the state variable, and a minimum value. A planning apparatus having output means for outputting the search result by the search means and minimum value search termination means for terminating the search for the minimum value at a predetermined time which is predetermined according to the number of state variables is also conceivable. Also,
The output means may be a display device.

Further, the display device may have a function of displaying a variable that changes depending on the number of times of planning.

Further, the minimum value searching means approximates the state probability distribution of the Markov chain to the optimal state probability distribution by simulating the Markov chain using the Gibbs matrix as the state transition probability matrix, and determines the minimum value of the objective function. It can also be considered as a means for obtaining the given state variable.

In this case, the minimum value searching means sets the temperature T, which is a search parameter, to Δ / L with respect to the number of searches i.
og (i + a) (Δ is a sufficiently large positive real number, Log
Is a common logarithm, and a is a positive real number).

Further, the minimum value searching means may be provided with a means for decreasing the temperature T, which is a search parameter, by Δ / i with respect to the number of searches i.

In addition, the minimum value search means sets the temperature T, which is a search parameter, to Δ / Lo with respect to the number of searches i.
g (i + a) (Δ is a sufficiently large positive real number, Log is
The common logarithm, a is a positive real number, and a means for decreasing the temperature T, which is a search parameter, by Δ / i with respect to the number of searches i, and the two means can be selected. A configuration is also possible.

The following is conceivable as a concrete plan making apparatus using the above means. First, there can be considered a planning apparatus that determines the objective function, which is a state variable, in the order of loading and unloading work by using the objective function as a distance between loading and unloading work shelves in loading and unloading work of a three-dimensional warehouse.

Further, a planning apparatus may be considered in which the objective function is the moving time of a crane used in the loading and unloading work of a three-dimensional warehouse, and the loading and unloading work order, which is a state variable, is determined. Further, a planning apparatus may be considered in which the objective function is set as a time required for movement between the points in the work of visiting each of the plurality of points once, and the order of visit, which is a state variable, is determined.

Further, a planning apparatus may be considered in which the objective function is the energy required to move between the points in the work of visiting each of the points once, and the order of visit, which is a state variable, is determined.

Further, a planning apparatus may be considered in which the objective function is a service degree determined by the expected arrival time with respect to the expected time in the work of visiting each of a plurality of points once, and the order of visit, which is a state variable, is determined.

Further, in the job shop plan in which a plurality of works are performed in a predetermined facility, the objective function is the total work required time from the start of work to the end of work, and the facility used for each work, which is a state variable, A planning device that determines the work start time of the equipment is also conceivable.

Further, in the job shop plan in which a plurality of works are performed in a predetermined facility, the objective function is defined as a ratio of operating time and non-operating time of each facility, and the equipment used for each work which is a state variable A planning device that determines the work start time of the equipment is also conceivable.

Further, the objective function is defined as a total value of a difference between a completion deadline time defined for each work and a completion time of the work in a job shop plan in which a plurality of works are performed in a defined facility. A planning device that determines the equipment used for each work, which is a state variable, and the work start time of the equipment is also conceivable.

Further, a planning apparatus may be considered in which the objective function is the total value of the moving time of the mounting machine in the work of mounting a plurality of components on the electronic board, and the mounting sequence which is a state variable is determined.

In addition, the objective function is the sum of the differences between the control target value and the predicted value in the process control using a plurality of control means, and the start / stop sequence of the plurality of control means which is a state variable. And a planning device for a control plan for determining the controlled variable is also conceivable.

Next, as a method for solving the above-mentioned problems,
The following means are possible. First, in order to set the objective function for setting the optimal plan, the environment variables that represent the environment of the given planning target and the plan candidates that can be taken in the given environment are represented. A state variable is input, a search for recombining state variables is performed, a temperature T that determines the energy of a random system is set so as to decrease with each search, and a first uniform random number is generated. , Rearranging the state variables based on the first random number to create a next plan candidate from the current plan candidates, and according to a predetermined objective function, the objective function value fold of the current plan candidates, and
The objective function value fnew of the next plan candidate is calculated and
d, fnew, the temperature T, and the newly generated second uniform random number α are α <exp (-(fnew-fol
d) / T) (exp represents the power of the natural logarithm)
When the inequality is satisfied, a planning method may be considered in which the process of making the next plan candidate the optimum plan candidate is performed a predetermined number of times.

Further, in order to set an objective function for setting an optimum plan, an environment variable representing a given environment to be planned and a state variable representing a plan candidate that can be taken in the given environment are set. A search is performed by inputting and recombining the state variables, and the temperature T that determines the energy of the random system is set so as to decrease with each search, and a first uniform random number is generated. The state variables are rearranged based on the first random number to create a next plan candidate from the current plan candidates, and according to a predetermined objective function, the objective function value fold of the current plan candidate and the purpose of the next plan candidate. The function value fnew is calculated, and the fold,
fnew, the temperature T, and the newly generated second uniform random number α are α <exp (-(fnew-fold) /
T) is satisfied (exp represents the exponentiation of natural logarithm), the number of searches is greater than n ³ (n is the number of components of the state variable) A planning method may be conceived in which the slope of the change of the objective function value before the predetermined number of search times from the current objective function value becomes larger than a certain negative real number.

Further, in order to set an objective function for setting the optimum plan, an environment variable representing a given planning target environment and a state variable representing a plan candidate that can be taken in the given environment are set. A search is performed by inputting and recombining state variables, and the temperature T that determines the energy of the random system is Δ / Log (i + a), where i is the number of times of the search.
(Δ is a sufficiently large positive real number, Log is a common logarithm, a
Is a positive real number), or is set to decrease by Δ / i, further, a first uniform random number is generated, and the state variables are rearranged based on the first random number. The next plan candidate is created, the objective function value fold of the current plan candidate and the objective function value fnew of the next plan candidate are calculated according to a predetermined objective function, and the f
old, fnew, the temperature T, and the newly generated second uniform random number α are α <exp (-(fnew-fo
It is also possible to consider a planning method in which the process of making the next plan candidate the optimum plan candidate is performed a predetermined number of times when the inequality (ld) / T) (exp represents the power of natural logarithm) is satisfied.

The means for specifically determining the objective function used in the above method and solving various problems will be described below.

First, a planning method can be considered in which the objective function is the distance between the loading and unloading work shelves in the loading and unloading work of a three-dimensional warehouse, and the loading and unloading work order, which is a state variable, is determined.

A planning method may be considered in which the objective function is the moving time of the crane used in the loading and unloading work of the three-dimensional warehouse, and the loading and unloading work order, which is a state variable, is determined. In addition, a planning method may be considered in which the objective function is set as a time required for movement between the points in the work of visiting each of the plurality of points once, and the order of visit, which is a state variable, is determined.

A planning method may also be considered in which the objective function is the energy required for movement between the points in the work of visiting each of the points once, and the order of visits, which is a state variable, is determined.

Also, a planning method may be considered in which the objective function is the service degree determined by the expected arrival time with respect to the expected time in the work of visiting each of the plurality of points once, and the visit order, which is a state variable, is determined.

Further, the objective function is a total work required time from the start of work to the end of work in a job shop plan in which a plurality of works are carried out at predetermined equipment, and equipment used for each work which is a state variable A planning method for determining the work start time of the equipment is also conceivable.

Further, in the job shop plan for performing a plurality of works in a predetermined facility, the objective function is defined as a ratio of operating time and non-operating time of each facility, and the equipment used for each work, which is a state variable, A planning method for determining the work start time of the equipment is also conceivable.

Further, the objective function is a total value of the differences between the completion deadline time defined for each work and the completion time of the work in a job shop plan in which a plurality of works are carried out in a defined facility. A planning method for determining the equipment used for each work, which is a state variable, and the work start time of the equipment is also conceivable.

Further, a planning method may be considered in which the objective function is set to the total value of the moving time of the mounting machine in the work of mounting a plurality of components on the electronic board, and the mounting sequence, which is a state variable, is determined.

In the process control using a plurality of control means, the objective function is a total value of the difference between the control target value and the predicted value, and the start and stop sequence of the plurality of control means, which is a state variable, and A planning method of a control plan for determining the controlled variable is also conceivable. Furthermore, the followings are conceivable as a solution means in the production line management system that performs high-mix low-volume production.

That is, in a production line management system for producing a wide variety of products in small quantities, equipment allocation information including at least product manufacturing specification information, equipment specification information, and an objective function representing an evaluation item to be minimized in a production plan. An input unit for inputting raw information, a first storage means for storing various inputted various pieces of information, and a lot of lots for minimizing the objective function from various pieces of various information stored in the first storage means. An arithmetic unit for calculating an optimum loading sequence, a second storage unit for storing a random loading sequence, a third storage unit for storing the calculated optimal loading sequence for the lot, and a third storage unit. An output unit for outputting the stored optimal loading order of the lot is provided, and the computing unit minimizes in the production plan, starting from an arbitrary state probability distribution r when the loading order x of the lot is a state variable. The objective function F representing the evaluation item you want to aim
By using the Gibbs matrix Gt determined by (x) as a state transition probability matrix and simulating the Markov chain, the state probability distribution of the Markov chain is made as close as possible to the optimal state probability distribution, and the objective function F (x) with a high probability is obtained. It has a minimum value or a quasi-minimum value and a stochastic minimum value search means for calculating the value of x that gives it, and further, the random lot loading order stored in the second storage means, and the first lot The various specification information stored in the storage means is input to the stochastic minimum value search means, and the minimum value of the objective function and the optimum lot input order that gives it are calculated within a finite time. And a means for outputting the result to the output section to minimize the value of the target evaluation item or to output an optimum lot input order in accordance with it. It is a planning apparatus.

The stochastic minimum value searching means may be configured to handle an objective function to which a forbidden condition is added.

Further, the stochastic minimum value searching means is m
Objective function expressed as the sum of products of the individual evaluation functions Hj (x) and the coefficient Kj, F (x) = Σ Kj · Hj (x)
(Σ represents the sum total of j = 1 to m) may be used.

In addition, the stochastic minimum value searching means uses the values of the plurality of evaluation functions Hj (x) determined by the state x as fuzzy variables to perform qualitative fuzzy inference according to a priori-determined fuzzy control rule. The input means may be configured to infer a comprehensive evaluation value and treat the value of the inference result as the value of the objective function F (x). Further, the input means interrupts or interrupts the search process of the minimum value search means. And a function of inputting an instruction to stop the search processing, and further, the minimum value searching means interrupts the search processing in accordance with the command input from the input means, and processes from the interruption. It is also possible to consider a planning device that restarts the search and terminates the search process.

Further, as the temperature lowering means, the following constitutions can be considered. The minimum value searching means reduces the temperature T according to a cooling curve of the temperature T determined based on a plurality of search times given through the input means and a temperature data set corresponding to the search times. It is a planning device equipped with. Furthermore, the following means are conceivable as a configuration in which the temperature lowering means can be selected. The minimum value searching means sets the temperature T, which is a search parameter, to Δ / Log (i + a) (Δ is a sufficiently large positive real number, Log is a common logarithm, and a is a positive real number with respect to the number of searches i. ), A means for decreasing the temperature T, which is a search parameter, by Δ / i with respect to the number of searches i, a plurality of times of search given via the input means, and a temperature for the number of searches. It is a planning apparatus that includes means for lowering the temperature according to the cooling curve of the temperature T determined based on the data set of 1.

Further, the following means can be considered as a configuration for setting the initial temperature. That is, the minimum value searching means calculates the initial temperature Ts based on the final temperature Te at the final search count N input in advance from the input means, and sets Ts = T.
e · log (N + a) / log (1 + a) or T
This is a planning apparatus provided with means for giving s = Te · N (where a is a positive real number).

[0049]

In order to solve the above-mentioned problems, the present invention performs the processing using the method described below in the combination evaluation problem.

For convenience of explanation, all processes are divided into first and second processes.

First, in the first processing, the order or the combination is changed little by little, and a value intended by the plan (hereinafter referred to as an objective function value) is calculated each time, and the objective function of the plan before the change is calculated. This is the process of comparing with the value.

To change the order or the combination, a random number is generated and, for example, all the elements included in a certain set of the elements constituting the combination are rearranged according to the random number. , Create a new plan candidate.

Next, in the second processing, when the objective function value by the current plan candidate before the change is compared with the objective function value by the next plan candidate after the change, it is not good (reduction of the objective function value). Is not performed), that is, even if it is not improved, if it is determined that the search process reaches the optimum solution, the changed plan is adopted as the optimum solution search route. ..

By repeating the above-mentioned first and second processes and observing the transition of the objective function value, the planning is finished when the optimum solution is reached.

Conventional OR (Operations Research)
In the method, AI (artificial intelligence) method, etc., since the evaluation does not temporarily go down in the planning process (the direction in which the objective function value becomes high) as described in the second processing, the true It was impossible to reach the optimal solution,
In the present invention, since it is accurately (or stochastically) determined that the search process reaches the optimum solution, the optimum solution, which has been impossible in the past, is detected.

Further, by the above processing, the optimum solution can be searched at a high speed, and when the optimum solution is reached,
By observing the route that has been reached, it can be judged that the obtained solution is optimal.

The above-described processing is created based on the results extracted by conducting a number of simulation experiments for the properties common to various target problems, and is an invention applicable to a wide variety of fields. You can say that.

The end of the planning is, for example, until the number of planned searches i becomes larger than n ³ (n is the number of constituent elements of the state variable) or from the current objective function value before the predetermined number of searches. It may be performed by performing until the slope of the change of the objective function value of becomes larger than a certain negative real number.

When searching for the optimum solution, the combination is changed (hereinafter also referred to as "perturbation") using the temperature as a parameter, but the number of searches depends on how the temperature is lowered, that is, the cooling is performed. i is the temperature Δ / Log
(I + a) (Δ is a sufficiently large positive real number, Log is a common logarithm, and a is a positive real number), or a method of setting the temperature to decrease by Δ / i and cooling the temperature may be used.

As a concrete example of the objective function to be optimized,
Examples include, but are not limited to:

First, the objective function is defined as the distance between the loading / unloading work shelves in the loading / unloading work of the three-dimensional warehouse, and the loading / unloading work order is determined. Secondly, the objective function is set in the three-dimensional warehouse. The moving time of the crane used in loading and unloading work is used to determine the loading and unloading work order. Thirdly, in the work of visiting each of the plurality of points once, the objective function is required to move between the points. Fourth, determining the order of visits, and fourth, determining the order of visits by using the objective function as energy required to move between the points in the work of visiting each of the plurality of points once. Fifthly, in the work of visiting each of a plurality of points once, the objective function is defined as a service degree determined by an expected arrival time with respect to an expected time, and the order of visit is determined. Sixthly, the objective function is determined. From In a job shop plan in which a plurality of works are performed at different facilities, the total work required time from the start of work to the end of work is determined, and the facility to be used for each work and the work start time of the facility are determined. The objective function is the ratio of the operating time and the non-operating time of each equipment in a job shop plan in which a plurality of operations are performed at a specified equipment, and the equipment used for each work and the work start time of the equipment are determined. Eighth, the objective function is defined as a total value of differences between a completion deadline time defined for each work and a completion time of the work in a job shop plan in which a plurality of works are performed in a defined facility. Determining the equipment used for each work and the work start time of the equipment,
Ninth, the objective function is the total value of the moving time of the mounting machine in the work of mounting a plurality of components on the electronic board,
Deciding the mounting order, tenthly, the objective function is
In process control performed using a plurality of control means, a control plan for determining a start / stop sequence and a control amount of the plurality of control means is set as a total value of a difference between a control target value and a predicted value. In the production line management system for high-mix low-volume production,
It is possible to set the evaluation order (minimum delivery date, facility utilization rate, completion time of all processes, etc.) that should be minimized, and to plan the lot loading sequence.

It goes without saying that various problems can be dealt with by adopting other parameters as the objective function.

The above method can be realized with the following hardware configuration.

In order to set an objective function, input means for inputting environment variables and state variables, objective function creating means for producing an objective function from given environment variables and state variables, and x for the state variable, the minimum Minimum value searching means for finding the minimum value of the objective function F (x) representing the evaluation item to be converted,
It is a plan making device comprising output means for outputting a search result by the minimum value search means.

Further, in order to provide the minimum value search ending means for ending the search for the minimum value at a predetermined time to terminate the calculation at a predetermined time without the intervention of an operator, the following device can be considered.

In order to set an objective function, input means for inputting environment variables and state variables, objective function creating means for producing an objective function from given environment variables and state variables, and x for the state variable, the minimum Minimum value searching means for finding the minimum value of the objective function F (x) representing the evaluation item to be converted,
The plan making device includes an output unit that outputs a search result of the minimum value search unit and a minimum value search end unit that ends the search for the minimum value at a predetermined time.

The output means is a display means, for example, a variable (eg, temperature,
By displaying the objective function value, etc.), it becomes possible to provide a user-friendly system.

Further, the minimum value searching means sets the temperature T, which is a search parameter, to Δ / Log with respect to the number of searches i.
(I + a) (Δ is a sufficiently large positive real number, Log is a natural logarithm, and a is a positive real number).
The number of searches i is reduced by Δ / i. Further, Δ / Log (i
+ A) (Δ is a sufficiently large positive real number, Log is a natural logarithm, and a is a positive real number), and means for decreasing the number of searches i by Δ / i are provided. By making the means selectable, the search for the optimum solution can be performed more accurately and quickly.

Further, it becomes possible to provide a device for making the above-mentioned various eigenfunctions handled by the above-mentioned device so as to find an optimum solution for various problems.

As the objective function handled by the minimum value searching means, it is possible to handle an objective function to which a forbidden condition is added. The prohibition condition is, for example, a condition for complying with a delivery date of a product in a scheduling problem of a manufacturing line management system that performs high-mix low-volume production.

If the objective function is the sum of the products of each evaluation function and the weighting coefficient, the value of each evaluation item is weighted by each coefficient, so that comprehensive evaluation can be performed. The means is an objective function F (x) represented as a sum of products of m evaluation functions Hj (x) and coefficients Kj.
= ΣKj · Hj (x) (Σ represents the total sum of j = 1 to m).

Further, it is also possible to make a comprehensive inference by giving ambiguity to the value of each evaluation item.

In this case, the minimum value searching means determines the state variable x.
By using the values of the plurality of evaluation functions Hj (x) determined by the fuzzy variables as a fuzzy variables, qualitative fuzzy inference is performed according to a priori-determined fuzzy control rule, and a comprehensive evaluation value is inferred. It will be treated as the value of the objective function F (x).

Hereinafter, the operation of the main part of the present invention will be described more specifically with reference to examples. For example, assume that a planning problem is given. This can be grasped as a problem of determining the order of n pieces of work.

Of course, this may be considered as the problem of finding the shortest route connecting n points or the problem of finding the longest route. That is, it is only necessary to find out how to arrange the n tasks so that the value of the target function becomes the minimum or the maximum. Here, each work requires a different setup time depending on the immediately preceding work, but the time required from the start to the end of a certain work is set. The time required for each work is Ti (i = 1 to n), the setup time is tij (i, j = 1 to n, j is a work number performed before the work i), and the objective function F (x) is ,
Assuming that the objective function for a certain state x is F (x) = Σ (t
ij + Ti). The state x at this time is regarded as a vector and described as follows.

X = (S ₁ , S ₂ , S ₃ , ………………, S _n ) (where S _i represents each work) This plan, that is, the state x is perturbed to a new state. get y.

Objective function F (y) for this new state y
Is calculated similarly.

Thus obtained F (x) and F (y)
Then, it is determined whether or not y is a new combination according to the new state acceptance probability distribution calculated from the temperature that is controlled so as to decrease as the number of examinations increases and the temperature.

Here, the new state acceptance probability distribution is expressed by the following equation. exp (-(F (y) -F (x)) / T). (However, exp represents the power of the natural logarithm) By repeating the above processing, the minimum value of the energy of the system can be obtained in an extremely short time. It should be noted that the solution obtained is not a realistic suboptimal solution but a true optimal solution that gives the minimum value. It goes without saying that the energy of this system is, as described above, the optimum order that minimizes the total work time of the n target works. As described above, practically, a method for obtaining an optimum solution in a sufficiently short time has been found. However, further speedup can be realized by the method described below.

Simulation experiments targeting various problems have revealed that the method further has the following features.

The first is that "assuming the number of target works is n, an optimum solution can be obtained by searching n ³ times or less".

Secondly, the objective function value does not change significantly after the degree of change of the objective function value with respect to the increase in the number of searches has become sufficiently small in a fixed number of search times frame. The first point indicates that a sufficiently satisfactory plan can be obtained even if the number of searches is automatically terminated by n ³ , and as a result, the time required for planning can be guaranteed. This is because, for example, when n = 24 as described above, even if a high-speed computer is used, the calculation that requires at least 10 ⁶ years is only 24 ³ × 10 to ⁶ (seconds) <10 to ¹
(Second), that is, about 0.1 second, the completion is shown. In this case, it means that much higher speed is achieved as compared with 10 ¹⁵ times. Furthermore, the second point is n
The larger the value, the more important it becomes. The optimal solutions in the fields that actually require optimal planning, such as various industries, vary greatly depending on the object in question, but as described in the second point above, the objective function value after the inclination becomes sufficiently small It is a fact that the amount of change mostly converges. In other words, in the second point, the plan at the time described can be the optimal solution. This is fundamentally different from the quasi-optimal solution obtained by the conventional techniques such as knowledge engineering, fuzzy and OR, and is an optimal solution in practical use. In other words, the error is smaller than the minimum accuracy error that can be practically executed, and the effect is no different from the optimum solution. The timing for aborting the calculation described in the second point is much earlier than that for n ³ ,
n ² or less. Considering the case of n = 24, it is 1/24 as compared with n ^3. Therefore, in the above example, about 5 × 10 ⁴ (s
The optimum solution is obtained when ec) = 0.5 (msec).

Conversely, using the same high speed computer, 1
The number of n that can be calculated per minute is about 7,800. Current,
The value of n, which is usually required, is 1000 or less at most, and a planning method and apparatus capable of solving the present requirement of the planning problem within a finite time are provided.

Even if the method is to be realized by software on a sequential execution type computer (also referred to as Neumann type), the main logic can be realized in only tens of steps. Moreover, not only the structure becomes simpler by using the parallel processing type processing means, but also the higher speed can be realized, and the same operation can be performed by the mutual connection type neural computer composed of a plurality of layers. Become.

Further, in the present invention, the optimal solution is stochastically obtained by using random numbers that follow uniform distribution. However, by using this as a chaotic search applying the chaotic phenomenon, it is possible to further increase the speed.

[0086]

Embodiments of the present invention will now be described in detail with reference to the drawings. FIG. 1 is a block diagram showing the functional configuration of an embodiment of the present invention.

The present embodiment is an embodiment in which the present invention is applied to an apparatus for determining an optimum work sequence in a manufacturing line.

Therefore, the planning device according to the present invention is used as the optimum work sequence determining device.

The optimum work sequence determining apparatus of this embodiment is provided with the input section 1 for inputting product manufacturing specification information, apparatus specification information, and equipment allocation specification information, and performing operations such as instructions. A memory 2 for storing various kinds of information, a memory 3 for storing a random lot loading order, and the above memory 2
And an arithmetic unit 4 for executing a process for obtaining an optimum lot input sequence based on the information stored in Nos. 3 and 3, a memory 5 for storing the arithmetic result, and an output unit 6 for outputting the contents stored in the memory 5. Have.

The above functions of this embodiment can be realized by using a computer system. For example, the input unit 1 is configured to include a keyboard 1a for staff to perform an input operation and a display device 1b having a display such as a CRT for displaying input contents.

The arithmetic unit 4 and the memories 2, 3 and 5 are
It can be composed of a central processing unit (CPU) and a main memory. The output unit 6 includes a display device 6a having a display such as a CRT and a printer 6b. Note that either one of the display devices 1b and 6a can be used as the other.

From the input section 1, the staff inputs product manufacturing specification information, apparatus specification information and equipment allocation specification information,
It is stored in the memory 2.

Here, the product manufacturing specification information is, for example,
The information includes the number of lots of various manufactured parts to be delivered, delivery date, number of processes, usable devices for each process, setup change time for each process, process time, transfer time between devices, process in-process information, and the like.
The device specification information is, for example, device failure information, device maintenance information, or the like. Further, the equipment allocation specification information includes equipment allocation constraint conditions, equipment allocation rules, an objective function which is an evaluation item to be minimized in a production plan, and the like.

The product manufacturing specification information, the apparatus specification information, and the equipment allocation specification information will be described by taking the manufacturing line shown in FIG. 2 as an example. FIG. 3 shows a production line for the machine parts shown in FIG. 2. Figure 2
Each of the press machine 1, the press machine 2, the assembling machine 1, the assembling machine 2 and the coating machine 1 shown in FIG. 1 can handle a wide variety of parts.

The product type A consists of two steps, and step 1 is a pressing step, and the pressing machine 1 or the pressing machine 2 can be used.
Step 2 is an assembling step, which is the assembling machine 1 or the assembling machine 2.
Can be used. The product type B comprises three steps, and step 1 is a pressing step, and the press machine 1 or the press machine 2 can be used. Step 2 is an assembling step, and the assembling machine 1 or the assembling machine 2 can be used. Step 3 is a painting step, and the painting machine 1 can be used. The product type C comprises two steps, step 1 is an assembling step, and the assembling machine 1 or the assembling machine 2 can be used. Step 2 is a painting step, and the painting machine 1 can be used.

First, the product manufacturing specification information will be described with reference to FIGS. 3 to 9. FIG. 3 is an explanatory diagram showing an example of correspondence between the product type number, the name, and the number of steps of each product type. Figure 4
FIG. 4 is an explanatory diagram showing a correspondence example between a process name for each process of each type, an available device number, and a processing condition number when the device is used. The usable device numbers in FIG. 4 are, for example, as shown in FIG. 2, press machine 1 = 1, press machine 2 = 2,
Assembling machine 1 = 3, assembling machine 2 = 4, and coating machine 1 = 5. FIG. 5 is an explanatory diagram showing an example of the correspondence between the setup change time and the processing time for each processing condition number. The setup change time is the time required to change the setting of the processing condition required when the processing condition number of the previous work of the used device and the processing condition number of the work to be processed this time are different. The processing time is the time required to actually perform the work to be processed.

FIG. 6 is an explanatory diagram showing an example of the transfer time between the apparatuses. For example, the device No. 2 to the device NO. The time required for transporting to 4 is “3”. FIG. 7 is an explanatory diagram showing an example of correspondence between lot numbers and product numbers. It shows which lot number corresponds to which product type. FIG. 8 is an explanatory diagram showing an example of correspondence between the product type number, the input lot number, and the delivery date. The number of input lots and the delivery date of each product are shown.

FIG. 9 is an explanatory diagram showing an example of lots currently placed on the manufacturing line at the current time 52, planned equipment numbers to be used, and use time zones. The symbols in FIG. 9 represent the lot number and the process number assigned to the device m at time t. In the process allotted time zone in FIG. 9, when the process condition number of the process is different from the previous process condition number of the device, the setup change time required to change the setting of the process condition and the device of the process are required. It includes the processing time required to work in.

Next, the device specification information will be described with reference to FIGS. 10 and 11. FIG. 10 is an explanatory diagram showing an example of a failure time zone for each device. In the example of FIG. 1 indicates that there is a failure during the period from the current time 52 to the time 54. FIG. 11 is an explanatory diagram showing an example of a scheduled maintenance time period for each device. In the example of FIG. No. 3 indicates that maintenance is scheduled during the period from time 57 to time 62. During a time when the device is out of order or is being maintained, the device cannot be used for work.

Next, the equipment allocation specification information will be described. FIG. 12 is an explanatory diagram showing an example of equipment allocation constraint conditions. FIG. 13 is an explanatory diagram showing an example of equipment allocation rules. FIG. 14 is an explanatory diagram showing an example of the objective function. In this case, the objective function F (X) of FIG. 14 indicates that the average delivery delay of all lots is minimized.

FIG. 15 is an explanatory diagram showing an example of a random lot input order. This is a random rearrangement of the input lots shown in FIG.

Next, the operation of the arithmetic unit 4 will be described.

First, in-process information shown in FIG.
The equipment allocation initial state shown in FIG. 16 is created from the equipment failure information shown in FIG. 10 and the equipment maintenance information shown in FIG. At the present time 52, the device used in the current process of the lot in the manufacturing line and the device scheduled to be used in the next process or later are assigned in advance. In addition, the device that has failed at the current time 52 cannot be operated during the failure time zone, so it is occupied in advance as a failed state. Further, since the scheduled maintenance time of the device cannot be operated, it is occupied in advance as a state under maintenance.

Next, the initial lot loading sequence shown in FIG. 17 is created from the memory 3 storing the random lot loading sequence shown in FIG. The number n of input lots is n = 8 from the production planning information shown in FIG.

Since the following operation is complicated, it will be described in detail with reference to a flowchart. 18 to 25 are explanatory diagrams showing the processing procedure of the calculation unit 4 using a flowchart. In addition, the program for executing the operation procedure shown in these flow charts is the above-mentioned memory 2,
It is stored in the main storage devices constituting 3 and 5.
First, referring to the initial lot loading sequence shown in FIG.
In steps 100 to 102 of 8, the lot loading sequence shown in FIG. 26 is created. Next, step 103 shown in FIG.
The optimum lot introduction sequence shown in FIG. 27 is created from the lot introduction sequence shown in FIG. Next, from the initial state of equipment allocation shown in FIG. 16, in step 104 of FIG.
Create the equipment allocation table shown in. Next, the objective function shown in FIG. 14 is embodied according to an actual processing procedure.
Give the lot input order as an argument.

The objective function allocates all the processes of all lots to the equipment allocation table according to the lot input order according to the equipment allocation constraint condition shown in FIG. 12 and the equipment allocation rule shown in FIG. FIG. 29 shows an example of the allocation result.
Here, as an example of how to obtain the objective function F (x), a subroutine starting with ENTRY in FIG. 21 (FIGS. 21 to 25)
The flowchart will be described. First, the function F (x)
Is set to 0 (step 400). Next, the value of the input lot number counter j is set to zero. Then, a loop for changing the value of j from 0 to "the total number of input lots-1" is entered (step 401). In this loop, j corresponds to "lot input sequence-1". The argument x of the objective function indicates the order of loading lots. xj represents the j + 1th lot number to be input. Next, the input lot information is searched by xj to find the corresponding product type number. Further, the current time is set to t, and the process number pr is set to 0 (steps 402 to 404).

Next, a loop for changing pr from 0 to "the number of processes corresponding to the product type-1" is entered (step 4).
05 to 608). In this loop, pr corresponds to the process number.

Then, the earliest ending device number is set to -1. This means that the device number that finishes the process for the earliest among a plurality of usable devices is undecided. Next, as a preparation for obtaining the ending device number,
The earliest end time is set to the latest time, that is, the final simulation time. Next, the value of the counter k for the number of usable devices, which is determined by the product type number and the process number, is set to zero. Next, a loop for changing k from 0 to "the number of usable devices-1" is entered. In this loop, k means the (k + 1) th available device when there are a plurality of available devices. Next, by referring to the product type-specific process information based on the product type number, the process number, and the value of k + 1, the corresponding usable device number and processing condition number are obtained.

Next, steps 500 to 5 shown in FIG.
The processing of 04 is performed. First, the carrying time information is searched for the carrying time based on the obtained available device number and the assigned device number of the process one step before. However, when pr = 0,
Since there is no step immediately before, the transfer time is set to 0. Next, the processing condition information is searched by the processing condition number thus obtained, and the setup change time and the processing time are obtained. When the previously used processing condition number of the obtained usable device number and the processing condition number obtained this time do not match, the setup change time is added to the obtained processing time.

Next, steps 505 to 5 shown in FIG.
08 processing is performed. Here, t represents the start time at which the device can be allocated, and the time obtained by adding the transport time to this becomes the start time at which the device can be actually allocated, so let this be l. Next, it is determined whether or not it is possible to allocate the processing time from the start time 1 at which the apparatus can be allocated, that is, whether the lot is not allocated from the time 1 to the time "l + processing time-1" of the usable apparatus number. However, when the allocation is impossible, the start time l when the device can be allocated is updated by 1 unit time, and l
When the final simulation time is reached, it is determined that the usable device number cannot be assigned and the process proceeds to the connector G. On the other hand, when l has not reached the final simulation time, it is repeatedly determined from the new time l whether the processing time can be allocated. During the repetition, when an allocatable time zone is found, that is, when there is no lot allocation from the time 1 to the time "l + processing time-1", the process proceeds to the connector F and the device allocation end time at that time, that is, When the time "l + processing time -1" is earlier than the earliest end time, the earliest end time is updated to the time "l + processing time -1", and the earliest end device number is updated to the available device number obtained, and the connector G and Join together.

From the connector G, the usable device counter k is updated by 1, the process information for each product type is searched by the product number and the process number, and the corresponding number of usable devices is obtained. k
If the number of available devices does not reach the calculated available device number, it is necessary to obtain the end time when the next available device number can be assigned and compare it with the earliest end time to update the earliest end time. , Connector H. On the other hand, when k reaches the calculated number of usable devices, it is determined whether the earliest ending device number is found, that is, whether the earliest ending device number is positive or negative. If not found, that is, if no allocatable time zone from the time t to the final simulation time is found, the process proceeds to the connector I and outputs a message of insufficient simulation time. Then, the processing of the calculation unit 4 is completed. On the other hand, when the earliest ending device number is found, "processing time-1" is started from the earliest ending time of the earliest ending device number.
The lot number xj is occupied until the time of returning to the previous time.

Next, the current time t is set to "early end time + 1"
To update. Furthermore, the process number pr is updated by 1, the product type information is searched from the product type number, and when the corresponding process number is not reached by pr, it is necessary to allocate the device for the next process. Go to K. On the other hand, pr
When the number of processes reached is reached, the earliest end time indicates the end time of the final process.Therefore, the production plan information is searched by the product type number to find the corresponding delivery date, and "early end time-obtained delivery date Is calculated to obtain the delivery delay time, and the calculated delivery delay time is divided by the total number of input lots to obtain the function F.
Add to the value of (x). Here, F (x) represents the lot average delivery delay time.

Next, the input lot number counter j is updated by 1, and when j has not reached the total input lot number, it is necessary to obtain the lot average delivery delay time of the next lot and add it to F (x). , Connector L. On the other hand, when j reaches the total number of input lots, F (x) has the result of calculating the average lot delivery time of all input lots as a numerical value, so the processing of the function F (x) ends. This subroutine ends.

The above is the description of the subroutine relating to the function F (x). Now, when the value of the objective function is calculated from the allocation result of FIG. 29, it becomes -3.0.

This value is used as the variable buffers fold and opt.
To enter. Next, the minimum value search count is first set to i = 0,
Enter the operation loop up to i = N-1. In the loop,
First, the temperature T is calculated. Here, Δ that is a factor that determines the temperature T is a sufficiently large positive number.

Next, the equipment allocation table shown in FIG. 28 is created from the equipment allocation initial state shown in FIG. Then 1
~ Two uniform random numbers of different integers are obtained, and this is rs
And re. Swap rs and re so that rs <re. Next, the new lot introduction sequence shown in FIG. 30 is created from the lot introduction sequence shown in FIG.

Now, assuming that rs = 5 and re = 2 are obtained, rs and re are exchanged, and rs = 2 and re = 5. Next, from the rsth position in the new lot input sequence in FIG.
Rearrange the contents of the th order in reverse order. FIG. 31 shows a new lot input order after rearrangement.

Next, the objective function is called with the new lot input order shown in FIG. 31 as an argument. The objective function allocates all the processes of all lots to the equipment allocation table according to the new lot input order according to the equipment allocation constraint condition shown in FIG. 12 and the equipment allocation rule shown in FIG. FIG. 32 shows an example of the allocation result. When the value of the objective function is calculated from the allocation result of FIG. 32, it becomes −3.75. This value is stored in the variable buffer fnew. Next, a uniform random number of 0 to 1 is generated, and this value is stored in the variable buffer a.

When the value of exp (− (fnew-fold) / T) = exp (− (− 3.75 − (− 3.0)) / T) = exp (0.75 / T) is calculated, T Is a positive number, exp (0.
75 / T)> 1.

Therefore, a <exp (0.75 / T) is established, and it is judged that the new lot introduction order should be accepted.

Here, if a ≧ exp (-(fnew-
In the case of (fold) / T), the new lot introduction order is rejected, the previous lot introduction order is maintained, and the next state perturbation (referred to as a process for optimizing the objective function) is performed. Since the new lot input order has been accepted, the lot input order in FIG. 33 is created from the new lot input order in FIG. Next, the objective function value fne according to the new lot input order
The objective function value fold is created from w. Next, the minimum value of the objective function is compared with the objective function value according to the new lot submission order. If the objective function value according to the new lot submission order is smaller, the minimum value of the objective function is updated and the new lot submission order is updated. Create a more optimal lot entry order. Now, the minimum value of the objective function = -3.0, the objective function value by the new lot input order =-
Since it is 3.75, the minimum value of the objective function is updated to -3.75, and the optimum lot input order is updated as shown in FIG. Next, the minimum value search count is updated, and when the predetermined count N is reached, the search is terminated, and the optimum lot input order at the end point is stored in the memory 5. When the minimum value search count is less than N, the next search by state perturbation is repeated.

It can be said that the above processing procedure faithfully simulates the Markov chain of the above-mentioned annealing method. Therefore, the minimum value can be searched with the limited number of times N of searches. For example, with the experimental value of the present invention, when n = 24, the minimum value could be searched with the number of searches of N = 4 · 24 ² = 2304 times. This number is the number of searches N2 =
24! On the other hand, it is shown that even a small computer such as a personal computer or a workstation, which has a relatively slow processing speed, can derive a solution within an actually allowable time.

As described above, all processes of all lots are assigned to the equipment allocation table based on the lot introduction sequence x, the product manufacturing specification information, the equipment specification information, and the equipment allocation specification information.
The value of the objective function F (x) is calculated from the allocation result.

The information in the memory 5, which stores the optimum order of loading lots, which is the output result, is output to the output unit 6 and finally to the CRT display, printer, etc. Is an easy-to-read material, and by introducing lots into the manufacturing line in accordance with the optimum lot introduction order, which is the output result, it is possible to realize the production method with the smallest delivery delay, in other words, the largest delivery time margin. That is, maximization of production efficiency can be realized.

The objective function shown in FIG. 14 can be freely determined. For example, by describing the lead time, the facility idle rate, the average in-process time, the average waiting time, etc., it is possible to derive the lot loading sequence that minimizes the target evaluation item.

The equipment allocation rule shown in FIG. 13 is as shown in FIG.
It is desirable to set a rule that contributes to minimize the objective function shown in. Therefore, when changing the objective function, at the same time, the equipment allocation rule needs to be changed to a rule that contributes to minimizing the objective function.

According to the present invention, an optimal solution or a solution corresponding thereto can be automatically derived without an operator, so that a scheduling problem can be automatically solved, and true factory automation can be realized. Moreover, compared with the heuristic method, it can be applied to any production line and has high versatility.

As another embodiment of the present invention, it is conceivable to add a forbidden condition to the objective function to determine the optimum work order. The prohibition condition may be, for example, a condition that the delivery date of the product is strictly adhered to. In this case, eliminate the solution that does not adhere to the product delivery date in the simulation process, or stop the calculation when the solution that meets the prohibition condition is found in the simulation process, and set the objective function value at that time to infinity. It is conceivable to take measures such as programming so that As described above, according to the present embodiment, the solution corresponding to the prohibition condition can be eliminated, and the calculation is aborted when it is found that the prohibition condition is satisfied during the calculation of the objective function value, so that the calculation speed is improved. To do.

Further, as another embodiment of the present invention, it is conceivable to determine the optimum work order by using the objective function as the sum of products of the evaluation function and the weighting coefficient. That is, the objective function is regarded as a composite function of various conditions. In this case, the weighting coefficient is set according to the importance of each condition. That is, if the evaluation function representing the condition is important, a method of increasing the weighting coefficient can be considered. At the time of simulation, an objective function expressed by the weighting coefficient and the evaluation function, for example, F (x)
= ΣKj · Hj (x) (where Kj is a weighting coefficient, H
j (x) means an evaluation function, and Σ means that the total sum is equal to the number of evaluation functions).

As described above, according to the present embodiment, by changing the weighting coefficient of each evaluation function, the value of each evaluation item can be weighted by each coefficient for comprehensive evaluation.

Further, as another embodiment of the present invention, the objective function may be determined as follows, and the ambiguity possessed by human beings may be given so that evaluation close to human beings can be performed.
That is, the state variable, which is the order of loading lots, is x,
The values of a plurality of evaluation functions Hj (x) determined by the state x are used as fuzzy variables, and fuzzy inference is performed according to the a priori-determined fuzzy control rule to infer a comprehensive evaluation value and the inference result In this embodiment, the value is the objective function F (x).

As described above, according to the present embodiment, it is possible to make an evaluation close to that of a human by comprehensively inferring the value of each evaluation item with ambiguity.

By the way, the system described above can be implemented by the stochastic feedback neural network shown in FIG. Therefore, when the Neumann computer is used, it can be realized by the procedure shown in the above flow, and when it is realized by the neural computer, it can be realized by the configuration of FIG. 36, for example. The use of a neural computer has a simpler configuration and a faster solution than the case of using a Neumann-type computer. The operation logic is the same as the above description.

Next, another embodiment will be described in order to explain the present invention in detail.

As an example, the delivery plan shown in FIG. 45 is set aside in order to facilitate understanding.

FIG. 45 (b) shows a problem of visiting 16 points 1 to 16 only once and determining the shortest route among the routes returning to the starting point.

In this example, "1" is the starting point and the black circle is "●".
, The route that visits 16 places is 16! = 16 × 15 × ... × 1> 20 × 10 ¹² , so the enumeration method in which all routes are searched by a high-speed computer requires a long time and is not practical.
As shown in FIG. 45 (c), when the number of visited sites is 32, the number of possible route combinations is more than 10 ³⁰ , and the enumeration method cannot actually solve the problem. I will end up. Therefore, it will be described below that an optimum solution can be obtained in a short time using the present invention.

FIG. 46 shows how the objective function approaches the optimum solution in the present invention. The vertical axis represents the value of the objective function (in this example, the sum of distances of visited routes). Value) and the horizontal axis shows the number of searches.

In process A of FIG. 38, a plan as shown on the left side of FIG. 45 (b) is randomly created, and this is designated as plan x. Next, in process B, two types of random numbers i1 and i2, which are integers in the range of 1 to 16, are generated in order to generate a plan y in which the plan x is slightly changed. Also, process F
In order to use it for determining whether or not it is the optimum route, 0.
It also generates a uniform random number a having a value from 0 to 1.0. Next, in process C, an amount called temperature, which is a parameter used for determining whether or not the route is the optimum route in process F, is determined. In the present embodiment, this is set to T and is decreased each time the search number i increases.

For example, the temperature may be lowered by the equation T = Δ / log (i + 2) or the equation T = Δ / i (Δ
Is a sufficiently large positive real number).

Next, in process D, a plan y is created by slightly changing x. For example, in the case of i1 = 2 and i2 = 5, the visit orders of the second to fifth visits of the plan x are all exchanged.

In FIG. 45B, x = (1, 4, 16, 5, 14, 15, 6, 12, ...
, ..., 11), y = (1,14,5,16,4,15,6,12, ...
…, 11).

For y, four of the constituent elements of x have been changed. Considering the route, only point 4 and point 14 have been exchanged, and the major elements should not be changed. There is.

This is devised so that, no matter what values i1 and i2 are, the visit order is changed at most two places or less.

In process E, the objective function value F (y) of the newly generated plan y, that is, the total distance is calculated. Now,
Regarding the combination x that is approaching the optimal solution at the present time and the combination y in which x is slightly changed, y is also approaching the optimal solution, and whether y is closer to the optimal solution or not Need to judge.

The whole of this relationship is the graph shown in FIG.

Let j be the current number of searches and F (x) _j shown in FIG. 46A be the objective function value of the combination x. Here, F (x) _j = 200. Assuming that the objective function value of the combination examined last time is F (x) _j-1 , since F (x) _j-1 = 210, the changed combination is improved and replaced. However, since the value of the objective function value F (y) = F (x) _{j + 1} of the combination y slightly changed from the current x is 220, empirically, x is an optimal solution rather than y. Should be close to.

Therefore, y is not on the route to the optimum solution, and it is easy to judge that the next time, it is sufficient to continue the examination based on x. However, if this judgment is made, the optimal solution will not be reached. This is an experimentally clear matter, and FIG.
This is because, as shown in (a), once the combination x is generated in the process of obtaining the optimum solution (here, the point M that gives the minimum distance), no other combination can be a candidate. Because no matter how x is changed,
This is because the objective function value cannot be reduced. Such a point is referred to as a pole point (also referred to as a minimum point in this embodiment). That is, it deviates from the route to the optimum solution. Therefore, in the process F, if the comparison of the objective functions is simply performed, in most cases,
It will not be possible to approach the optimal solution.

As a result of various examinations, it was found that the optimum solution was reached by making the judgment as shown in FIG.

In other words, conceptually, even if the extreme point is encountered as described above, a new process for exiting the extreme point may be performed.

The transition of the objective function value becomes as shown in FIGS. 46 (b) and 46 (c) by changing the plan little by little, and reaches the optimum solution. That is, assuming that the objective function value at i = j is MX1, for example, the next maximum point C is passed. This process is repeated several times until the optimum solution region M
The plan will be changed so that

At this time, if the plan is extremely large and the plan is changed, an optimum solution will be erroneously obtained, so the property of the transition of the objective function value is used, and a value that becomes smaller as the number of searches increases is defined. Then, the criterion is changed.

This is the temperature T determined in the process C described above. As shown in FIG. 46 (a), after approaching the optimum value considerably, and further observing the details, the objective function value shifts as shown in FIG. 46 (b). That is, the closer to the optimum solution, the smaller the absolute value of the change amount of the objective function value, but the objective function value increases or decreases. Figure 46
In (c), the transition of the objective function value is further observed microscopically, but finally the minimum value, that is, the optimum solution M is reached. Since FIG. 46 is a conceptual diagram, the objective function value fluctuates even after the optimum solution search, but the actual fluctuation is small.

The determination of the process F, which is the core part of the above description, is performed as follows based on the above description. That is, when a <exp ((-F (y) -F (x)) / T) (Equation 1) (where exp is a power of natural logarithm) is satisfied, a combination y in which x is slightly changed is satisfied. Is newly adopted as a plan on the route leading to the optimum value.

Equation 1 is the difference of the objective functions (F (y) -F
The value obtained by dividing (x)) by the temperature T that decreases with the number of searches is taken as a power of the natural logarithm and compared with a uniform random number a having a value of 0.0 to 1.0. If it is larger, it is a judgment formula for performing the process of adopting the new plan y as a plan on the route leading to the optimum solution.

For example, in FIG. 46 (a), if a plan of j = 1 is x, a plan of j = 2 is y, and i = 1 at present, F (x) ₁ = 200, F (x) ₂ = F (y) = 220, Δ = 4.0 T = Δ / (Log (1 + 2)) ≈3.64 (where Δ
Is a sufficiently large positive real number, and Log is a natural logarithm) When a = 0.040, exp ((-F (y) -F (x)) / T) = exp (-
(220-200) /3.64) = exp (-20 / 3.64) ≒ 0.004
Since it is 11, the formula (1) is established.

Therefore, the plan y is replaced with x ← y in the process G in FIG. 38 as a candidate on the route leading to the optimum solution.

By thus providing a temperature that gradually decreases with respect to the number of searches, even if the temperature is trapped at the minimum point, it is possible to escape from it, and the minimum value, that is, the optimum solution can be certainly reached. ..

In this way, it was found that the optimum solution can be obtained by repeating the processes B to I, but when considering a practical application device, it cannot be used unless the number of searches is as small as possible.

However, as a result of various experiments, it was found that the number of times to reach the optimal solution exists as a realistic number depending on the number of planned objects (for example, n of the number of points to visit) and the temperature T. It was

FIG. 46 shows the optimum solution search situation for 16 points. The optimum solution search is completed in about 1610 times. In this embodiment, since the visiting points are defined in a circle, if a circle is formed, that is obviously the optimum solution. Utilizing this fact, the number of visiting points is arranged in a circle, the number of visiting points is changed, and how many times the optimum solution is reached is examined. The results shown in FIG. 47 are obtained. Figure 4
In the graph of 7, the horizontal axis defines the number of searches that reach the optimum solution, and the vertical axis defines the frequency of the experiment. FIG. 47
In the 6 types of graphs shown in the upper part of the figure, the planned points are 8, 16, 32, 64, 100, and 50, respectively.
It is assumed that the number is 0, and the number of times of reaching the optimum solution is shown in each of the number of experiments around 100 times. Even if a plan is made with the same configuration, the number of times of arrival changes because the initial state is different. This is the optimum solution M in FIG.
However, it depends on how far away the planning starts from.

For example, the initial value (initial position) is the symbol F.
(See FIG. 46), the optimal solution M is reached earlier, but if it is the symbol B (see FIG. 46), the symbol F
More searches are needed than in.

When the six types of diagrams in FIG. 47 are examined in detail, it is understood that the number of planned points is n and the optimum solution is reached within the number of searches within the cube of n. Ie n
= 8 is at most 400 times, so that n ³ = 8 ³ = 512> 400 holds. The same can be said when n = 16, 32, 64, 100, 500. In the lower diagram of FIG. 47, n = 8,
Each of the cases of 16, 32, 64, 100 is the total. From this, the number of times to reach the optimum solution is n at most.
^It can be seen that it is less than ³ , and most of it is less than c · n ² (c ≦ 4). That is, the following can be said when planning is made by the above method.

(1) The optimum solution can be reached by searching n ³ at most.

(2) The optimum solution is almost reached within the number of searches of 4 · n ² .

Furthermore, whether or not the optimum solution has been reached can be determined by observing the transition of the objective function value, and FIG. 39 explains this. In FIG. 39, as in FIG. 46, the horizontal axis defines the number of searches and the vertical axis defines the objective function value.
When the optimum solution is reached, the objective function value does not change because it is the optimum value. Therefore, from the property of (1) above, the slope of the objective function value transition with respect to the number of past n ² / L (L is a real number of 1.0 or more) searches counted by the current number of searches is a negative real number. If it is larger than b, it can be said that the optimum solution has been almost reached. The real numbers L and b are determined according to the problem to be dealt with. The present embodiment describes the case where the minimum value is the optimum solution, but when the maximum value is the optimum solution, a completely similar method can be used, considering the reciprocal of the objective function value.

Since the above properties have been clarified by various simulation experiments, the optimum solution can be obtained at high speed by the above method. That is, in the process K, when the following conditions are satisfied, the planning is finished and the process I
The plan held in can be the optimal solution.

That is, the conditions for ending the plan are as follows.

When the number of searches i is greater than n ³ , or the slope of the transition of the objective function value from the number of searches i to the number of past n ² / L searches is greater than a certain negative real number b This is the case.

The condition indicates that, for example, in the case of n = 16, the process may be unconditionally ended when the number of searches i becomes larger than n ³ = 16 ³ = 4096. For example, as shown in FIG. 48, if n = 16, the current number of searches i = 500, L = 1.0, and b = -0.00391, the conditions are as follows. That is, n ² /
From L = 16 ² /1.0=256, from i = 500 to 2
Observe the slope of the transition of the objective function value up to 56 times before.

From i = 500, i = 2 before 256 times
Since it is 45, the objective function value at this time is F (x) ₂₄₅ =
200, F (x) ₂₄₆ to F (x) ₅₀₀ are all 19
9, the slope dF (x) _i / di is dF (x) _i /di=(199-200)/256≈−0.00390, and the slope is larger than b = −0.00391. It can be seen that the optimum solution has been reached. In other words, the value is sufficiently stable and the amount of displacement before that is sufficiently small.

As is apparent from the above description, the present invention is
It has the following important meanings.

This is to enable optimal planning, optimal design, and optimal control, which were completely impossible with the conventional method. Moreover, since this method perfectly matches with statistical physics such as thermodynamics, it does not include any uncertainties such as empirical knowledge, and is a very stable and reliable method. That is,
A certain method of the "reduction problem" that minimizes a target value is
It means that it is established. As will be apparent from various embodiments described below, many problems that could not be solved by the related art can be solved reliably, at high speed, and with an apparatus having an extremely simple structure.

As described above, the present invention can be applied to most planning problems in which there is always one type of optimal solution, and the object is not limited to various plants, processes, fields, purposes, etc. Here, after describing a device configuration and a functional configuration example having an extremely primitive general-purpose logic structure showing basic logic, some specific examples will be described.

FIG. 49 shows a structural example of the apparatus according to the present invention.

This example of the apparatus is configured to have an input means, a display means, an objective function creating means, and a minimum value searching means.

The input means is means for inputting state variables such as the lot loading sequence and various environmental variables such as equipment maintenance time, and is composed of, for example, a keyboard, a mouse, a light pen and the like.

The objective function creating means is means for creating an objective function composed of parameters to be minimized, and has, for example, a CPU, a ROM, a RAM, a hard disk and the like.

The minimum value searching means is means for minimizing the created objective function, and is, for example, CPU, ROM, RA.
It is configured by including M, a hard disk, and the like.

The display means is means for displaying the search result obtained by the minimum value search means, and is composed of, for example, a CRT, a liquid crystal display, an EL display or the like.

Further, although not shown, a structure having a means for changing the temperature and a means for stopping the minimum value search may be considered.

The means for changing the temperature and the means for terminating the minimum value search are composed of, for example, a CPU, a ROM, a RAM, a hard disk and the like.

FIG. 37 shows an example of a functional block diagram for explaining the function of the planning apparatus 1 according to the present invention.

The processing of this apparatus is started by the planning request from the operator. This apparatus 100 stores a storage unit 1100 for storing information necessary for planning such as facility information and production instruction information, and an initial plan creation for reading out the necessary information from the storage unit 1100 to create a random initial plan. Means 200, an integer uniform random number used when generating a new plan, a random number generating means 300 for generating a real uniform random number used to determine whether or not an optimal solution, and an increase in the number of searches. A new plan generating means 500 for generating a new plan by the temperature setting means 400 that determines the temperature so as to decrease with it and the integer random number generated by the random number generating means.
And an objective function calculating means 700 for calculating an objective function that is preset or selected for each target plan,
Evaluation for comparing a difference between a real random number value generated by the random number generating means 300 and the objective function values of the old and new two types of plans and a value determined by the temperature to determine a plan on the route leading to the optimum solution. The inclination of the objective function value corresponding to the plan selected by the means 510 and the evaluation means 500 with respect to the number of searches is observed, and when it is determined that the optimum solution has been reached, the planning result is displayed on the display device, and In the case of arrival, the display means 600 for displaying the progress of planning on the display device, and in the case of non-arrival, the optimum solution estimation means 800 for repeating similar optimum solution search processing from the random number generation means 300, and the target problem. The objective function selecting unit 1200 selects or changes the objective function according to the above.

FIG. 38 is a processing flow chart for explaining the operation of each means described above. For ease of understanding, an explanation will be given taking as an example the problem of planning the shortest entry / exit operation time in a three-dimensional automated warehouse shown in FIGS. 40 and 41.

FIG. 41 shows an outline of the process. In order to store a large amount of various kinds of articles, in recent years, a three-dimensional warehouse 22 having several thousand to several hundred thousand shelves has been constructed.
Dozens to thousands of loading and unloading operations are carried out daily by stacker cranes.

In FIG. 40, one row, one row and one row, and one stacker crane 21 that constitutes a three-dimensional warehouse.
Shows the operation of. It is assumed that the loading and unloading work from and to is shown in FIG. In the management system, the shelf position for the stored items is fixed, and there is a "fixed address system" in which a table called a "pallet" is always present except during loading and unloading work. There is an address system ", but in the present embodiment, the former is adopted. 40. In FIG. 40, the stacker crane 21 moves [3] to the shelf after the operation of unloading the article from the shelf [1] and again loading the pallet and the remaining article [2] after the work is completed.

Since [1] and [2] are necessary works in any order, in order to minimize the work time shown in FIG. The total value of moving time D (i → i + 1) F (x) = ΣD (i → i + 1) must be minimized.

(However, Σ represents the total sum of i from 1 to 5.) The moving time D (i → i + 1) depends on the acceleration / deceleration characteristics and the moving vector of the stacker crane 21, as shown in the lower diagram of FIG. For example, in this embodiment, the step direction y
The required travel time ty ₁ for the moving direction is compared with the required travel time tx ₁ for the continuous direction x, and the longer required time ty ₁ is the actual moving time.

In initialization processing A shown in FIG. 38, as shown in FIG.
The randomly determined work order combination vector shown in x = (,,,,,) is determined, and i
= 1 (the number of searches), Fmin = F (x) (optimal objective function value), and M = x (optimal combination) are set. Next, the random number generation means 3 performs a random number generation process B, and 1 to n,
In the present embodiment, the uniform random numbers i ₁ and i ₂ (i ₁ <i ₂ ) which are integers from 1 to 6 and any real numbers between 0.0 and 1.0, Generate a uniform random number a. next,
A temperature control process C is performed in the temperature setting means 4, and the temperature is determined by an equation, for example, T = Δ / log (i + 2) (Δ represents a sufficiently large positive real number, log represents a common logarithm). Be done. T is temperature, and as the number of searches i increases,
T decreases. Next, in the new candidate plan generation process D, a new candidate plan proposal y = (,,,,) shown in FIG. 41 is generated. The vector y is obtained by exchanging the work order between the i ₁ -th work and the i ₂ -th work of the vector x. In the present embodiment, i ₁ = 2 and i ₂ = 3, and are interchanged. Next, the objective function value calculation means 7 performs the objective function calculation process E, and as shown in FIG. 41, F (y) = ΣDi = 321 (where Σ is from 1 to 5).
Represents the total sum up to) is calculated. Next, the evaluation means 51
Then, the evaluation determination process F is performed as follows, for example. That is, if a <exp (-(f (y) -f (x)) / T) is satisfied, "replacement" is performed, and if not satisfied, "replacement unnecessary" is determined.

In the above equation, a is a uniform real-valued random number of 0.0 to 1.0 generated in advance in the random number generation process B.

In the case of "replacement", the process "x ← y" is performed in the plan candidate replacement process G. In process H, the minimum objective function value Fmin up to that point is compared with F (y), and if Fmin> F (y) holds, then Fmin ← F (y) (optimal objective function value ), M ← y (optimum combination) is performed in process L.

FIG. 39 shows the processing result of the display processing J of the display means 6. F (1), that is, the objective function value at the time of the first examination is set at the midpoint of the objective function value axis on the vertical axis, the number of searches is set on the horizontal axis, and the objective function value is set at a position corresponding to the corresponding number of searches. Plot and display as a connected graph. At this time, the maximum value on the search frequency axis is n ³ . Further, the optimum objective function value Fmin (m) (m is the number of searches) is displayed on the same scale. In FIG. 39, it is indicated by a broken line. With such a display, the planner can accurately grasp the drafting situation, and the planner with sufficient experience determines the practical optimal solution acquisition timing himself and finishes the plan. You can also let it. By setting the objective function value F (1) at the time of the first examination to the midpoint of the vertical axis, there is no movement of the starting point and smoothing even if the optimization direction of the objective function of the plan is minimum or maximum. Can be displayed. The scale of the vertical axis may be changed according to the degree of change in the objective function value.

Being able to visually grasp the planning situation not only enhances the sense of trust, but also makes it easier to intuitively understand the essence of the problem and to judge whether or not a sufficient solution is realistically obtained. For operators, it is possible to construct a system that is easy to use.

Next, in process K, the planning process up to this point is observed and, for example, the planning process is completed as follows.

As described in the outline of the present invention, for example, the amount of change in the objective function value with respect to the number of past n ² / L (L is a real number of 1.0 or more) searches from the current number of searches. If the slope of n is larger than a certain negative real number b, the planning is finished. The real number b is determined by the minimum resolution which is a realistic value according to the problem to be solved. In the case of the maximization problem, it goes without saying that b is a positive real number.

With the above operation, the optimization of the loading / unloading work of the automatic warehouse is completed in a short time. Using a computer that calculates the total travel time for each operation in 10 to ⁶ (seconds), the optimal plan for work with about 100 shelves is only 1
It will be completed in 0 to ² (seconds). In the conventional enumeration method, what required 10 ⁹⁴ (seconds) or more was shortened to 1/10 ⁹⁶ .

In the loading and unloading work of the automatic warehouse, in addition to the present embodiment, the need to minimize the difference from the expected time for each work in the next process after warehousing, the minimization of attendance time for each product type, That is, although there are various purposes such as leaving the stock in the order of the earliest entry, in the present invention, if these are simply set in advance in the objective function value calculation means 7, the optimum solution can be obtained.

FIG. 42 shows an example of a delivery plan which is another embodiment of the present invention.

This plan is a plan to return to the starting point at the end while delivering a package by visiting a plurality of delivery places once. In the present embodiment, the optimum route is obtained by setting the start place as the starting point, visiting the delivery place from, and finally returning to the starting place. For example, the combination C = (,,,,,) is an example in which the required time is optimized. However, this plan does not consider deviations from the amount of cargo (cargo volume) and the "expected time" (time when the cargo is expected to arrive), that is, various services at all, so the plan is unsatisfactory. May be In this example, "load"
Indicates the amount of parcels loaded for delivery at a certain delivery place. For example, at the start point A, the total amount of parcels is 2600 (Kg). Also, "energy" determined by the amount of gasoline consumed
Is the product of the traveling distance and the load amount in the section defined by two points. The expected time is indicated by a relative allowable time (unit: minutes) from the start time. For example,
The delivery to the delivery location may be done after 6 hours, but the delivery to the delivery location is expected to be completed by 75 minutes later. In the above-mentioned example, the total value function F1 of the required time is the objective function, but in the case of minimizing the energy, F2 in FIG. 42 may be the objective function. In addition, if you want to deliver to the expected time, that is, to deliver as soon as possible than the expected time,
The objective function may be F3, which is the total deviation between the expected time and the estimated arrival time.

In some cases, it is desired that the delivery be as close as possible to the expected time. In other words, minimize the delay,
In addition, it is a case where delivery is done not too early. For such a request, the objective function may be F4, which is the sum of the squares of the deviations between the expected time and the expected arrival time.

When the above objective functions are strictly expressed by mathematical expressions, F1 (i) = Σf1P (required time function) F2 (i) = Σ (ΣWm × f1p) (energy function) ) F3 (i) = Σ (THp-f1p) (Expected time margin function) F4 (i) = Σ (THp-f1p) ² (Expected time deviation function) (However, f1p is required from location P to P + 1 Time, Wm
Is the load on location m, THp is the expected time at location P, Σ
Indicates that the sum of all p = 1 to 6 is taken.) As described above, it is extremely easy to optimize the delivery plan by defining the objective function according to various needs. You can

FIG. 43 is an explanatory diagram for formulating a production plan in the manufacturing industry or the like. The delivery planning problem is easy to understand because it is easy to grasp visually, but in this problem, since there are many aspects in which the target must be grasped temporally and spatially, it is very difficult for humans to make plans. Have difficulty.

The solution according to the present invention will be described below.

[0205] In a general production process, a plurality of general-purpose manufacturing facilities are installed to improve production efficiency,
Produce different types and quantities of products.

In this embodiment, it is considered to make a plan to execute from work in a certain manufacturing facility A. Of course, the same can be considered when there are a plurality of manufacturing facilities. At the start of each work, it is necessary to perform “setup time”, which is the total time of cleaning up the previous work and preparing for this work. When the production of products of the same type is continuous, the setup time may be short because there are few changes in jigs, parts, etc., cleaning, etc. However, in the case of production of different products, the setup time is long. An example of this setup time is shown in the table of FIG. 43 with the previous work number on the vertical axis and the current work number on the horizontal axis. For example, when the previous work is done and the work this time is, the setup time of 18 (minutes) is required. In the manufacturing plan, there are various other factors such as time for transportation between equipments, breakdown, periodic repair, interruption of express goods, etc., but in the present invention, the production plan can be easily drafted in consideration of them. it can. This is as described in the first embodiment.

The plan for executing the work sequence combination i = (,,,,,) according to this setup time table requires 420 (minutes) from the start, as shown in the figure.
For work order combination, j = (,,,,,)
It only takes 382 (minutes).

Since the objective function value in this case is the time from the start of work to the completion of work, F (i) = Σ (Tp + Dp-1 → p). (However, Tp
Represents the work time of the work P, Dp-1 → p represents the setup time from the work P-1 to P, and Σ represents the sum of p = 1 to 6) In the present embodiment, Considering that Tp is not related to the order, F (i) = Σ (Dp-1 → p) is sufficient.

(However, Σ represents that the total sum of p = 1 to 6 is taken.) As described above, according to the present embodiment, the plan that maximizes the production efficiency and maximizes the equipment operation rate is set. Can plan.

Further, in order to realize multi-product variable quantity production, it is important to handle sudden orders from customers and strictly adhere to delivery dates. To achieve this, the "work end punctuality time" that is set for each work,
That is, the total value of the differences between the delivery date and the scheduled end time may be maximized. If THp is the delivery date of the work P and TEp is the scheduled end time, then F (i) = Σ (THp-TEp), (where Σ represents the sum of p = 1 to 6) The objective function may be set so that the determined F (i) is maximized. Of course, there are various needs other than the above-mentioned embodiment, but all of them can make an optimum plan by using the same method as described above, and the present invention makes it possible to optimize the plan in any production plan. Will be achieved.

Needless to say, there are a plurality of functions, and when considering them comprehensively (for example, when the optimization target is the sum of a plurality of functions), the problem of planning an optimal plan is solved. It is also possible to correspond.

For example, the objective function is defined as follows. F (i) = W ₁ × (ΣDp-1 → p) + W ₂ × (ΣTHp−TEp) (where Σ means to take the total sum of p = 1 to 6) It is efficient and has a sufficient margin for delivery, and the former and the latter are considered in the ratio determined by W ₁ and W _2. ”

However, considering the user's position in reality, by defining a plurality of objectives, conflicting elements are mixed between the objectives, and the planned plan is what the planning staff really wanted. It may be difficult to judge whether or not there is, and the result may not always be satisfactory.

Therefore, in a planning problem in which there is always one kind of optimal solution, the purpose should be clarified by narrowing down the objective function to one kind, and the above-mentioned multi-objective evaluation plan should be carefully expanded. Need to go to.

FIG. 44 shows another embodiment, which deals with the problem of planning a sequence for mounting various electronic components and the like on an electronic substrate. In recent years, this is often done automatically by a robot or the like, but like the delivery plan, for example, the objective function is set so as to minimize the taking-out time of the mounted parts, the setup time, the movement according to the mounting order, and the positioning time. .. That is, F (i) = ΣD ₁ (p) + Σ (ΣD (p, m)).

(However, D ₁ (p) is the time for taking out the mounting parts, D (p, m) is the mounting time in the p-th work, and m is
Number of mounted parts in p-th operation, first and 2 from the left
The third Σ represents that the sum total of p = 1 to 6 is taken, and the third Σ represents that the sum total of m = 1 to 6 is taken. Not only can the production plan and the above installation plan be applied,
For example, the present invention can be applied to pattern optimization, signal path minimization, impedance optimization, and the like in board layout design.

Although the application examples have been described in detail with reference to the drawings, the application of the present invention is not limited to the above examples, and can be applied to various planning problems shown below, for example. There is an outline of an application example of the present invention for a typical planning problem.

Again, these are merely illustrative of the problems solved by the present invention.

First, the problem of timetable formation and reorganization will be described.

The formation of the departure and arrival timetables that determine the departure and arrival times at multiple stations through which the train passes, and the formation of the takeoff and landing timetables of airplanes are considered to be extremely complicated planning. The system is built. The purpose in this case is summarized as how to keep the arrival and departure time and reduce the transportation cost. Detailed information for planning is essential, but empirical knowledge is often unnecessary. To improve the service, for example, a function that minimizes the deviation between the departure and arrival times and the scheduled time may be used as the objective function, and if priority is given to safety, for example, the allowance time of cross flights can be set. The function that maximizes may be the objective function. Similarly, for other purposes, an objective function may be created according to the purpose.

Secondly, the network optimization problem will be described.

The planning problem for reaching a destination from a certain source through a complicated communication network is an important problem in the fields of telephone lines, wide area computer networks (WAN) and the like. Also, minimizing energy loss in the power supply network or minimizing power failure recovery time in the event of a failure is an important social issue. The same applies to the optimal planning problem for stable water supply and flood prevention.

These can apply the same method as the delivery plan described above, and the complexity of the problem is smaller than that of the delivery plan. In other words, in the above delivery plan, all the delivery locations must be visited, so there are n! Although there were individual, in the network or tree problem, it is not necessary to visit all nodes (points),
This is because a route from the source node to the destination node may normally exist regardless of any route.

It is practically impossible to have all the nodes in the route, and the number of possible solutions is much smaller than that of a delivery plan having the same number of nodes. From the above, an optimal plan can be obtained by defining the objective function in the same way as the delivery plan and examining only possible solutions. Next, thirdly, the design optimization problem will be described. It is needless to say that the present invention can be applied to the automation of the design because the design is considered to be one aspect of the plan as described in the application example of the work planning in the electronic substrate production.

For example, when taking out a product having a desired pattern from a regular material, it is possible to minimize the surplus material. Conversely, it is possible to define a fixed shape, a minimum outer shape of a material required for a certain number of parts, and the like. As a result, a significant cost reduction can be achieved.

Finally, fourthly, the optimal control problem will be described. In particular, it can be said that the control that determines the controlled variable while predicting the future state is a kind of optimization planning problem. For example, in the control of a process having a plurality of actuators, it is necessary to determine in what order and by how much the actuators should be displaced.
Simulate a huge number of combinations in an allowable time,
It is necessary to establish the most effective operation plan. On the other hand, by setting the objective function to the deviation between the control target value and the predicted value according to each operation plan, the present invention can be applied as it is. The planning according to the invention is very fast and can be applied to real-time control.

Further, by using a screen as one of the display and input means as shown in FIG. 50, it is possible to arbitrarily create a schedule for cooling the temperature in the annealing. Input of necessary data to the screen can be realized by a configuration using, for example, an electronic light pen. The cooling schedule is specified by inputting the required number of searches and the temperature in the optional cooling schedule table shown in the figure. The cooling schedule of the specified result is displayed in the cooling schedule graph in the figure.
It is easy for the operator to understand and improves operability. The actual cooling curve may be obtained by, for example, linear interpolation, polynomial approximation, or the like between the points determined by the number of searches i and the input temperature. In this way, in the initial stage of the search, the temperature is relatively high and escapes from the risk of reaching the minimum point. In the middle stage of the search, the temperature is gradually lowered, and further in the final stage of the search. , It is possible to bring the temperature closer to 0 and to approach the optimum solution endlessly. As described above, regardless of the initial state of the search, it is possible to set the cooling schedule capable of searching the suboptimal solution at high speed within the designated predetermined number of searches.

Further, by using a screen which is an example of display and input means as shown in FIG. 51, it is possible to set and input the initial value of the temperature in the annealing and the maximum number of searches for stopping the search on the way.

Further, the temperature cooling schedule function is set to normal cooling (T = Δ / log (i + a)), high-speed cooling (T =
Δ / i), and optionally specified cooling (a method of connecting a specified number of times of search and temperature and obtaining a cooling curve of temperature by a linear interpolation method or the like) can be selected from three types. As described above, if the initial value of the temperature, the maximum number of searches, and the cooling schedule can be adjusted, it is possible to obtain the quasi-optimal solution that is closest to the optimal solution with the minimum number of searches. Also, using the screen shown in FIG.
The maximum average equipment utilization rate up to the present is displayed, and at the same time, for example, a time series transition with an increase in the number of searches for the average equipment availability and a time series transition with an increase in the number of temperature searches are displayed in a graph. Therefore, there is an advantage that the state of the search can be closely observed. Also, when the operator can estimate from the screen that the objective function value is close to the optimal solution, from the screen via the mouse, keyboard, etc.,
For example, by instructing "interrupt", "restart", "discontinue", etc., "discontinue" the search for the optimal solution, "resume" the search from the point of interruption, or "discontinue" the search.
be able to.

With such a configuration, the search can be terminated by the operator's judgment before the maximum number of searches is reached. Therefore, useless search can be eliminated and an optimum solution can be obtained early. . Further, as shown in FIG. 52, if the value of the temperature Te in the final search number N is determined, the initial value Ts of the temperature can be calculated by the back calculation by the cooling schedule function. This will be described below. In the final search N, when fnew> fold, the new state has a larger objective function value than the old state. now,
Assuming that the minimum objective function value is obtained, the probability of accepting a new state is exp (-(fnew-fold) / Te).

(However, exp represents the power of the natural logarithm).

To make this acceptance probability 0.001 or less,-(fnew-fold) / Te <ln0.001 (ln is
Natural logarithm Here, from In 0.001 = −6.9, Te <(fnew
-Fold) /6.9. This is 0.001
With a probability of, it means that a certain state transits to a state in which the objective function value becomes large such as fnew> fold.
At this time, in order to suppress (fnew-fold) to 0.01, that is, (fnew-fold) <0.01, Te <0.01 / 6.9 = 0.0015. If Δ is obtained from this Te and Ts is calculated, Te = Δ /
log (N + a), Δ = 0.0015 * log (N +
From a), Ts = 0.0015 * log (N + a) / log (1+
In the case of a), the initial value of the temperature is obtained.

That is, in the final search round N, it is a condition for converging to an optimum solution that the probability that the objective function value slightly increases is slightly suppressed. Therefore, the initial value of the temperature is calculated backward from the condition. Determining means that a quasi-optimal solution is always obtained within the number N of searches when the temperature is lowered according to the cooling schedule, starting from the initial value of the obtained temperature. By automatically calculating the initial value of the temperature for always reaching the sub-optimal solution from the number of searches N and the cooling schedule, it is not necessary for a human to set the initial value of the temperature. , Operability is greatly improved.

Further, as shown in FIG. 53, the initial state of the search is changed, the search is tried a plurality of times, and the number of searches per time is reduced to reduce the total number of searches to obtain the optimum solution or quasi-optimal solution. The optimal solution can be searched, which will be described below. Now, when the search is started from in the figure, in order to reach the minimum value while avoiding falling into the minimum value, it is necessary to increase the initial value of the temperature and increase the number of searches. Is clear from. However, it is also clear that when the search is started from in the figure, the initial value of the temperature is lowered and the minimum value is reached even with a small number of searches.

Therefore, if the initial value of the temperature is lowered and the initial state is changed with a small number of searches and the search is tried a plurality of times, it can be seen that the minimum value can be searched while reducing the total number of searches. That is, as shown in the figure ...
From each of the four types of initial states shown in (4), if the initial value of the temperature is set low and the value of N is set small, the minimum value can be searched in 4N times of searches. .. In this case, when compared with the number of searches M for searching the minimum value starting from the initial state of, 4N <M can be satisfied.

Thus, by changing the initial state and performing a plurality of searches with a small number of searches, it is possible to perform a faster optimum solution search.

Further, as shown in FIG. 54, when the number of states satisfying the constraint condition is significantly smaller than the number of all states, when the search is started from a random initial state,
A large number of searches are required to reach a state that satisfies the constraint condition. Further, depending on the initial state, there are cases where the state that satisfies the constraint condition cannot be reached forever. In such a case, if one of the solutions satisfying the constraint condition is searched as the initial state, the optimal solution always exists in the adjacent states, and therefore the optimal solution can be searched with a small number of searches. For example, in a production line, if the delivery date for each lot is divided into several lots each day, if the lots are rearranged in the order of earliest delivery date, the initial delivery order will be the initial state. It is clear that all the lots meet the constraints because they meet the delivery date of each lot.

In this way, even for a planning problem with severe constraints, the optimal solution can be searched with a small number of searches, so the search speed is improved.

As described above, according to the present invention, a probabilistic search method is used to solve a planning problem in which the number of combinations is enormous and an optimal plan always exists. .. Further, the optimal solution can be obtained by the number of searches within n ³ , where n is the number of elements of the target plan. The practical effect is the same as the optimal solution within n ² number of searches. Utilizing the properties obtained, the planning device itself grasps the status of optimization and completes the planning.

Therefore, it is possible to obtain an optimal solution that cannot be obtained by the conventional OR, knowledge engineering, fuzzy, and other planning systems.

Also, the optimum solution can be obtained extremely quickly. If the number of elements of the target planning problem is n,
The enumeration method is n! Although it has been necessary to search the number of times, in the present invention, a sufficiently satisfactory solution can be obtained with the number of searches of about n ² .
This means that if n is 100, the speed can be increased to 1/10 ⁹⁶ . Conventionally, only a plan with 20 to 30 elements could be studied at most, but in the present invention, a planning problem having hundreds to thousands of elements can be solved in seconds to minutes.

Further, the present invention is highly adaptable to the change even if the information on the target problem is changed. That is, since the main logic of the present invention is not based on empirical information or limited information, it is not affected by, for example, changes in equipment layout, occurrence of failures, changes in the purpose of planning, and the like.

Furthermore, the present invention does not have a complicated structure, and, for example, software for realizing the main logic of the method according to the present invention on a Neumann computer is:
A program with only a few dozen steps is sufficient. Further, in a device configuration having parallel processing type hardware, it can be realized by only a few logic devices, and further speedup can be expected.

Further, by using the mutual coupling type neural computer, further simplification and speedup can be obtained.
Use of a chaos computer can further increase the speed. It was also explained that one of the major features of the present invention is that design and control problems can be solved. That is, various objectives can be expressed as a function, and the design or control problem in which the value of the objective function has one kind of minimum value or maximum value can be solved. In particular, since the processing time for various problems according to the present invention is extremely short and the device configuration is simple, the device configuration using a microcomputer or sequencer can be used for a control problem for which a real-time solution is strongly desired. It will be available.

[0245]

EFFECTS OF THE INVENTION It has become possible to find an optimum solution for a wide variety of planning problems in an extremely short time without using a large computer.

[0246] Further, it becomes possible to deal with changes in various kinds of information, and a knowledge database or the like becomes unnecessary. Furthermore, since the optimum solution search is automatically terminated, the judgment based on the experience of the operator becomes unnecessary.

[Brief description of drawings]

FIG. 1 is an explanatory diagram showing a configuration of an embodiment of the present invention.

FIG. 2 is an explanatory diagram of a conventional manufacturing line example.

FIG. 3 is an explanatory diagram of a correspondence example of a product type number, a name, and the number of processes of each product type.

FIG. 4 is an explanatory diagram of an example of correspondence between a process name for each process of each product type, an available device number, and a processing condition number when the device is used.

FIG. 5 is an explanatory diagram of an example of correspondence between setup change time and processing time for each processing condition number.

FIG. 6 is an explanatory diagram showing an example of transfer time information between devices.

FIG. 7 is an explanatory diagram of an example of correspondence between a lot number and a product type number.

FIG. 8 is an explanatory diagram of an example of correspondence between a product type number, the number of input lots, and a delivery date.

FIG. 9 is an explanatory diagram of an example of a lot in process on the manufacturing line at the current time, a planned device number to be used, and a use time zone.

FIG. 10 is an explanatory diagram of an example of device failure information.

FIG. 11 is an explanatory diagram of an example of device maintenance information.

FIG. 12 is an explanatory diagram of an example of equipment allocation constraint conditions.

FIG. 13 is an explanatory diagram of an example of equipment allocation rule.

FIG. 14 is an explanatory diagram of an example of an objective function.

FIG. 15 is an explanatory diagram of an example of a random lot input sequence.

FIG. 16 is an explanatory diagram of an example of an initial state of equipment allocation.

FIG. 17 is an explanatory diagram of an example of an initial lot input sequence.

FIG. 18 is an explanatory diagram of a flowchart of an example of a processing procedure of a calculation unit.

FIG. 19 is an explanatory diagram of a flowchart of an example of a processing procedure of a calculation unit.

FIG. 20 is an explanatory diagram of a flowchart of a processing procedure example of a calculation unit.

FIG. 21 is an explanatory diagram by a flowchart of a facility allocation subroutine.

FIG. 22 is an explanatory diagram by a flowchart of a facility allocation subroutine.

FIG. 23 is an explanatory diagram of a flowchart of a facility allocation subroutine.

FIG. 24 is an explanatory diagram by a flowchart of a facility allocation subroutine.

FIG. 25 is an explanatory diagram by a flowchart of an equipment allocation subroutine.

FIG. 26 is an explanatory diagram of an example of a lot loading order.

FIG. 27 is an explanatory diagram of an example of an optimum lot input sequence.

FIG. 28 is an explanatory diagram of an example of a facility allocation table.

FIG. 29 is an explanatory diagram of an example of equipment allocation result.

FIG. 30 is an explanatory diagram of an example of a new lot input sequence.

FIG. 31 is an explanatory diagram of an example of a new lot input order after rearrangement.

FIG. 32 is an explanatory diagram of an example of equipment allocation result.

FIG. 33 is an explanatory diagram of an example of a lot loading order.

FIG. 34 is an explanatory diagram of an example of an optimum lot input sequence.

FIG. 35 is an explanatory diagram of an example of an annealing algorithm.

FIG. 36 is an explanatory diagram of an example of a probabilistic feedback type neural network.

FIG. 37 is an example of the functional configuration of the planning apparatus according to the present invention.

FIG. 38 is an explanatory diagram showing a flow of processing of the present invention.

FIG. 39 is a display example of the transition of the objective function value.

FIG. 40 is an explanatory diagram of planning of loading and unloading work of a three-dimensional automated warehouse.

FIG. 41 is an example of planning a loading / unloading operation of a three-dimensional automated warehouse.

FIG. 42 is an explanatory diagram of a delivery plan planning example.

FIG. 43 is an explanatory diagram of production planning in the production line.

FIG. 44 is an explanatory diagram of a situation where components are mounted on an electronic board.

FIG. 45 is an explanatory diagram of an example of the present invention.

FIG. 46 is an explanatory diagram of an optimum solution search process.

FIG. 47 is an explanatory diagram of a delivery plan drafting example.

FIG. 48 is an explanatory diagram of an example of operation end logic.

FIG. 49 is an explanatory diagram of an example of a planning apparatus according to the present invention.

FIG. 50 is an explanatory diagram of a display example according to the present invention.

FIG. 51 is an explanatory diagram of a display example according to the present invention.

FIG. 52 is an explanatory diagram of the present invention.

FIG. 53 is an explanatory diagram of the present invention.

FIG. 54 is an explanatory diagram of the present invention.

[Explanation of symbols]

1 ... Input unit, 1a ... Keyboard, 1b ... Display device, 2 ...
Memory for storing various specification information, 3 ... Memory for storing random lot loading order, 4 ... Calculation section, 5 ... Memory for storing optimum lot loading order, 6 ... Output section, 6a ...
Display device, 6b ... Printer, 21 ... Stacker crane, 100 ... This device, 200 ... Initial plan creating means, 30
0 ... Random number generation means, 400 ... Temperature setting means, 500 ... New plan generation means, 510 ... Evaluation means, 600 ... Display means,
700 ... Objective function computing means, 800 ... Optimal solution estimating means,
1100 ... Storage means 1200 ... Objective function selection means

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Kenichi Nakamura 3-2-1, Sachimachi, Hitachi City, Ibaraki Prefecture Hitachi Engineering Co., Ltd.

Claims (39)

[Claims] 1. In order to create an objective function for setting an optimal plan, an environment variable representing a given environment of a planning target and a state variable representing a plan that can be taken in the given environment are input. Input means, an objective function creating means for creating an objective function from given environment variables and state variables, a minimum value searching means for obtaining the minimum value of the objective function F (x), where x is the state variable, and a minimum value. A planning apparatus having output means for outputting a search result of the search means. 2. In order to create an objective function for setting an optimal plan, an environment variable representing a given environment of a plan object and a state variable representing a plan that can be taken in the given environment are input. Input means, an objective function creating means for creating an objective function from given environment variables and state variables, a minimum value searching means for obtaining the minimum value of the objective function F (x), where x is the state variable, and a minimum value. A planning apparatus having output means for outputting a search result by the search means and minimum value search termination means for terminating the search for the minimum value at a predetermined time which is predetermined according to the number of state variables. 3. The planning apparatus according to claim 1, wherein the output means is a display device. 4. The display device according to claim 3,
A planning apparatus having a function of displaying a variable that changes depending on the number of times of planning. 5. The minimum value search means according to claim 1 or 2, wherein the Gibbs matrix is a state transition probability matrix, and the Markov chain is simulated so that the state probability distribution of the Markov chain becomes an optimal state probability distribution. A planning device characterized in that it is means for obtaining a state variable that gives a minimum value of an objective function. 6. The minimum value searching means according to claim 5, wherein the temperature T which is a search parameter is Δ / Log (i + a) (Δ is a sufficiently large positive real number, Log Is a common logarithm, and a is a positive real number). 7. The plan making apparatus according to claim 5, wherein the minimum value searching means includes means for decreasing the temperature T, which is a search parameter, by 1 / i with respect to the number of searches i. .. 8. The minimum value search means according to claim 5, wherein the temperature T which is a search parameter is Δ / Log (i + a) (Δ is a sufficiently large positive real number, Log with respect to the number of searches i. Is a common logarithm, a is a positive real number), and means for decreasing the temperature T, which is a search parameter, by Δ / i with respect to the number of searches i, and the two means can be selected. A planning device characterized in that 9. The planning apparatus according to claim 1 or 2, wherein the objective function is a distance between loading and unloading work shelves in loading and unloading work of a three-dimensional warehouse, and determines a loading and unloading work order which is a state variable. 10. The planning apparatus according to claim 1 or 2, wherein the objective function is a moving time of a crane used in loading and unloading work of a three-dimensional warehouse, and determines a loading and unloading work order which is a state variable. 11. The objective function according to claim 1 or 2, wherein the objective function is a time required for movement between respective points in the work of visiting each of the plurality of points once, and determines the order of visit, which is a state variable. Planning device. 12. The objective function according to claim 1 or 2, wherein the objective function is energy required for movement between the points in the work of visiting each of the points once, and a visit order as a state variable is determined. A planning device. 13. The objective function according to claim 1, wherein the objective function is a service degree determined by an expected arrival time with respect to an expected time in a work of visiting each of a plurality of points once, and a visit order as a state variable is determined. A planning device. 14. The job function according to claim 1 or 2, wherein the objective function is a total work required time from the start of work to the end of work in a job shop plan in which a plurality of works are performed in a predetermined facility, and is a state variable. A planning device that determines the equipment used for each work and the work start time of the equipment. 15. The job function according to claim 1 or 2, wherein the objective function is a ratio of an operating time and a non-operating time of each facility in a job shop plan in which a plurality of works are performed in a predetermined facility, and is a state variable. A planning device that determines the equipment used for each work and the work start time of the equipment. 16. The job function according to claim 1, wherein the objective function is a job shop plan for performing a plurality of works in a predetermined facility, and the completion deadline time and the completion of the work set for each work. A planning device that determines the equipment used for each work, which is a state variable, and the work start time of the equipment, which is the total value of the difference from the time. 17. The objective function according to claim 1, wherein the objective function is a total value of movement times of the mounting machine in a work of mounting a plurality of components on an electronic board, and determines a mounting order which is a state variable. Planning device. 18. The objective function according to claim 1, wherein the objective function is a total value of a difference between a control target value and a predicted value in process control performed using a plurality of control means.
A planning device for a control plan, which determines a start / stop sequence and a control amount of the plurality of control means which are state variables. 19. In order to set an objective function for setting an optimum plan, an environment variable representing a given environment of a planning target and a state variable representing a plan candidate that can be taken in the given environment are set. A search is performed by inputting and recombining the state variables, and the temperature T for determining the energy of the random system is set so as to decrease every time the search is performed. Further, the first uniform random number is generated, and the first uniform random number is generated. The state variables are rearranged based on one random number to create a next plan candidate from the current plan candidates, and according to a predetermined objective function, the objective function value fold of the current plan candidate and the objective function value of the next plan candidate. fnew is calculated, and the above-mentioned fold, fnew,
The temperature T and the newly generated second uniform random number α are α <exp (-(fnew-fold) / T) (e
xp represents the exponentiation of the natural logarithm) When satisfying the inequalities, a planning method for performing a process of making the next plan candidate the optimum plan candidate a predetermined number of times. 20. In order to set an objective function for setting an optimum plan, an environment variable representing a given environment to be planned and a state variable representing a plan candidate that can be taken in the given environment are set. A search is performed by inputting and recombining the state variables, and the temperature T that determines the energy of the random system is set so as to decrease with each search, and a first uniform random number is generated. The state variables are rearranged based on the first random number to create a next plan candidate from the current plan candidates, and according to a predetermined objective function, the objective function value fold of the current plan candidate and the purpose of the next plan candidate. The function value fnew is calculated, and the fold and fnew are calculated.
And the temperature T and the newly generated second uniform random number α
Becomes α <exp (-(fnew-fold) / T) (e
xp represents the exponentiation of the natural logarithm) When the inequality is satisfied, the process of selecting the next plan candidate as the optimum plan candidate is performed until the number of searches exceeds n ³ (n is the number of components of the state variable), or , A planning method that is performed until the slope of the change of the objective function value from the current objective function value a predetermined number of times before the search number becomes larger than a certain negative real number. 21. In order to set an objective function for setting an optimum plan, an environment variable representing a given environment of a planning target and a state variable representing a plan candidate that can be taken in the given environment are set. A search is performed by inputting and recombining state variables, and the number of searches is i, and the temperature T that determines the energy of the random system is Δ / Log (i + a) (Δ is
A sufficiently large positive real number, Log is a common logarithm, a is a positive real number), or is set to decrease by Δ / i, and further, a first uniform random number is generated and First, the state variables are rearranged to create a next plan candidate from the current plan candidates, and the objective function value fold of the current plan candidate and the objective function value fnew of the next plan candidate are calculated according to a predetermined objective function. And the fold,
fnew, the temperature T, and the newly generated second uniform random number α are α <exp (-(fnew-fold) / T) (e
xp represents the exponentiation of the natural logarithm) When satisfying the inequalities, a planning method for performing a process of making the next plan candidate the optimum plan candidate a predetermined number of times. 22. The object function according to claim 19, 20 or 21, wherein:
A planning method that determines the order of loading and unloading work, which is a state variable, by using the distance between the loading and unloading work shelves. 23. The planning method according to claim 19, 20 or 21, wherein the objective function is a moving time of a crane used in loading and unloading work of a three-dimensional warehouse, and a loading and unloading work order which is a state variable is determined. 24. The objective function according to claim 19, 20 or 21, wherein the objective function is a time required to move between points in a work of visiting each of a plurality of points once, and the order of visit, which is a state variable, is set. How to make a decision. 25. The objective function according to claim 19, 20 or 21, wherein the objective function is energy required to move between points in a work of visiting each of a plurality of points once, and the order of visit as a state variable is set. How to make a decision. 26. The objective function according to claim 19, 20 or 21, wherein the objective function is a degree of service determined by an expected arrival time with respect to an expected time in a work of visiting each of a plurality of points once, and is a state variable. A planning method that determines the order of visits. 27. The job function according to claim 19, 20 or 21, wherein the objective function is a total work required time from the start of work to the end of work in a job shop plan in which a plurality of works are performed in defined equipment. A planning method for determining equipment used for each work, which is a variable, and a work start time of the equipment. 28. The objective function according to claim 19, 20 or 21, wherein the objective function is a ratio of an operating time to a non-operating time of each equipment in a job shop plan for performing a plurality of works on a predetermined equipment, and A planning method for determining equipment used for each work, which is a variable, and a work start time of the equipment. 29. In the job shop plan according to claim 19, 20 or 21, said objective function is a job shop plan in which a plurality of works are performed in a predetermined facility, and the completion deadline time and the work specified for each work. The planning method for determining the equipment to be used for each work, which is a state variable, and the work start time of the equipment, which is the sum of the difference from the completion time. 30. The objective function according to claim 19, 20 or 21, wherein, in the work of mounting a plurality of components on an electronic board, the objective function is a total value of moving times of the mounting machine, and a mounting sequence which is a state variable is set. How to make a decision. 31. The objective function according to claim 19, 20 or 21, which is a state variable, which is a total value of a difference between a control target value and a predicted value in a process control using a plurality of control means. A method of planning a control plan for determining the start-up / shutdown order and control amount of the plurality of control means. 32. In a production line management system for high-mix low-volume production, equipment allocation information including at least product manufacturing specification information, device specification information, and an objective function representing an evaluation item to be minimized in a production plan. An input unit for inputting raw information, a first storage means for storing various inputted various pieces of information, and a lot of lots for minimizing the objective function from various pieces of various information stored in the first storage means. An arithmetic unit for calculating an optimum loading sequence, a second storage unit for storing a random loading sequence, a third storage unit for storing the calculated optimal loading sequence for the lot, and a third storage unit. An output unit for outputting the stored optimum lot input sequence is provided, and the calculation unit starts from an arbitrary state probability distribution r and minimizes the production plan when the lot input sequence x is a state variable. Planned Objective function F that represents the evaluation item
By using the Gibbs matrix Gt determined by (x) as a state transition probability matrix and simulating the Markov chain, the state probability distribution of the Markov chain is made as close as possible to the optimal state probability distribution, and the objective function F (x) with a high probability is obtained. It has a minimum value or a quasi-minimum value and a stochastic minimum value search means for calculating the value of x that gives it, and further, the random lot loading order stored in the second storage means, and the first lot The various specification information stored in the storage means is input to the stochastic minimum value search means, and the minimum value of the objective function and the optimum lot input order that gives it are calculated within a finite time. And a means for outputting the result to the output section to minimize the value of the target evaluation item or to output an optimum lot input order in accordance with it. Planning apparatus. 33. The planning apparatus according to claim 32, wherein the stochastic minimum value searching means handles an objective function to which a prohibition condition is added. 34. In claim 32, the stochastic minimum value searching means is an objective function expressed as a sum of products of m evaluation functions Hj (x) and coefficients Kj, F (x) = Σ Kj · A planning device in a manufacturing line, which handles Hj (x) (Σ represents the sum total of j = 1 to m). 35. The probabilistic minimum value search means according to claim 32, wherein a plurality of evaluation functions Hj (x) determined from a state x.
Is used as a fuzzy variable to perform qualitative fuzzy inference according to a priori-determined fuzzy control rule to infer a comprehensive evaluation value, and the inference result value is used as the objective function F (x).
A planning device in a production line, which is treated as the value of. 36. The function according to claim 1 or 2, wherein the input means inputs a command for instructing interruption of the search processing of the minimum value search means, resumption of processing from interruption, and termination of the search processing. Further, the minimum value searching means further comprises: suspending the search processing, resuming the processing from the interruption, and stopping the search processing in accordance with an instruction input from the input means. 37. The method according to claim 5, wherein the minimum value searching means is a plurality of search times given through the input means.
And a plan making device comprising means for lowering the temperature according to a cooling curve of the temperature T determined based on a set of temperature data with respect to the number of searches. 38. The temperature according to claim 8, wherein the minimum value search means is determined based on a plurality of search times given through the input means and a temperature data set corresponding to the search times. A means for lowering the temperature T, which is a search parameter, according to the cooling curve of T is provided, and the means and the temperature T, which is a search parameter, are set to Δ / Log (i + a) (Δ is a sufficiently large positive , Log is a common logarithm, a is a positive real number), and the temperature T, which is a search parameter, is set to the search count i.
The plan making device is characterized in that the three means for decreasing by Δ / i can be selected. 39. The minimum value search means according to claim 5, based on a final temperature Te at the final search count N input in advance from the input means, and an initial temperature Ts is expressed as Ts = Te · log. (N + a) / log (1 + a) or means for giving Ts = Te · N (where a is a positive real number).