JP7156666B2 - Parameter search device, parameter search method, and program - Google Patents

Description

The present disclosure relates to a parameter search device, parameter search method, and program.

With reference to FIG. 1, chronological changes in the number of products in stock at a distribution center or the like will be described. FIG. 1 is a diagram showing an example of time-series data of the number of products in stock at a distribution center or the like.
In FIG. 1, the horizontal axis is time, and the vertical axis is the number of products in stock.
In addition, on the vertical axis of FIG. 1, the maximum inventory quantity is the maximum required inventory quantity of the product. The average number of inventories is the average value of the number of inventories of products during a specific period. The number of items to be ordered is the number of items in stock that serves as a reference for the timing of ordering products. The number of safety stocks is the number of stocks required to prevent out-of-stock or shortage of products.

In addition, on the horizontal axis of FIG. 1, the maximum inventory days (=holding days) is the number of days until the number of products corresponding to the maximum inventory decreases to zero. The number of days of order lead time is the number of days required to procure the product. The number of safety stock days is the number of days until the number of safety stock items decreases to 0. The safety stock days is obtained by dividing the number of safety stocks by the average number of shipments.

In the example of FIG. 1, the product is ordered at the timing when the inventory quantity of the product has decreased to the order quantity. The number of items to be ordered is calculated based on the maximum inventory days and safety inventory days. Therefore, in order to order products at appropriate timing, it is necessary to appropriately set the maximum number of days in stock and the number of days in safety stock as parameters for determining the number of items to be ordered.

Conventionally, the maximum number of days in stock and the number of days in safety stock were manually set by the user based on past experience. However, since the maximum days in stock and the safety days in stock are determined depending on multiple indicators such as the average days in stock and the number of orders, even an experienced user cannot determine the maximum days in stock and the safety days in stock. It was difficult to set properly.

Therefore, recently, a technique of searching for the optimum solution of parameters using AI (Artificial Intelligence) technique has been proposed. For example, Patent Literatures 1 and 2 disclose techniques for searching for optimal solutions for parameters using the gradient descent method.

Japanese Patent No. 4349271 Japanese Patent Publication No. 2005-527034

However, the general gradient descent method disclosed in Patent Documents 1 and 2 uses differentiation to search for the optimum solution of parameters, so the search requires complicated calculations, and the search takes a long time. There is a problem that it hangs.

Therefore, an object of the present disclosure is to solve the above-described problems and to provide a parameter search device, parameter search method, and program capable of searching for the optimum solution of parameters without performing complicated calculations.

A parameter search device according to one aspect includes:
a dividing unit that divides a predetermined N-dimensional space with N (N is a natural number equal to or greater than 2) parameters as axes;
a searching unit for searching, from among the squares in the predetermined N-dimensional space, a square having a minimum value of an objective function obtained using the respective values of N parameters corresponding to the square;
a determination unit that determines the values of the N parameters corresponding to the mass at which the value of the objective function is the minimum value as an optimal solution;
Prepare.

A parameter search method according to one aspect comprises:
A parameter search method by a parameter search device,
a step of dividing a predetermined N-dimensional space with N (N is a natural number equal to or greater than 2) parameters as axes;
a step of searching from among the squares in the predetermined N-dimensional space for a square having the minimum value of an objective function obtained using the respective values of N parameters corresponding to the square;
a step of determining the values of each of the N parameters corresponding to the mass at which the value of the objective function is the minimum value as an optimal solution;
including.

A program according to one aspect comprises:
the computer,
dividing means for dividing a predetermined N-dimensional space with N (N is a natural number equal to or greater than 2) parameters as axes;
search means for searching, from among the squares in the predetermined N-dimensional space, a square having a minimum value of an objective function obtained using respective values of N parameters corresponding to the square;
Determination means for determining the values of each of the N parameters corresponding to the mass at which the value of the objective function is the minimum value as an optimal solution;
function as

According to the above aspect, it is possible to provide a parameter search device, a parameter search method, and a program capable of searching for parameters without performing complicated calculations.

FIG. 4 is a diagram showing an example of time-series data of the number of products in stock in a distribution center or the like; 1 is a block diagram showing a configuration example of a parameter search device according to an embodiment; FIG. FIG. 7 is a diagram showing an image example of a method of searching for a square by a searching unit according to the embodiment; It is a figure which shows the example of the search method of the square by the search part which concerns on embodiment. It is a figure which shows the example of the search method of the square by the search part which concerns on embodiment. 6 is a diagram showing an example of a search path when the searching unit according to the embodiment searches for squares by the method of FIG. 5; FIG. FIG. 4 is a flow chart showing an example of a parameter searching method by the parameter searching device according to the embodiment; FIG. 8 is a flowchart showing an example of processing in step S102 of FIG. 7; FIG. 9 is a diagram showing an example of a search path when the searching unit according to the embodiment searches for squares by the method of FIG. 8; It is a figure which shows the example of the search method of the square by the search part which concerns on embodiment. 1 is a block diagram showing a configuration example of a parameter search device conceptually showing a parameter search device according to an embodiment; FIG.

Embodiments of the present disclosure will be described below with reference to the drawings. Note that the following descriptions and drawings are appropriately omitted and simplified for clarity of explanation. Further, in each drawing below, the same elements are denoted by the same reference numerals, and redundant description is omitted as necessary.

<Embodiment>
In this embodiment, it is assumed that an optimum solution is searched for two parameters, ie, the maximum number of days in stock and the number of days in safe stock of products in a distribution center or the like. It is also assumed that the four indices for determining the maximum number of days in stock and the number of safe days in stock of a product are the average number of days in stock, the average amount of inventory, the number of orders placed, and the out-of-stock rate.

Here, the average number of days in stock is the number of days until the average number of stocked items decreases to 0. The average inventory value is an amount obtained by dividing the accumulated inventory value of the product during the specified period by the number of days during the specified period. The number of orders is the number of times a product is ordered during a specific period. The out-of-stock rate is the ratio of products that could not be supplied to orders for products during a specific period. The out-of-stock rate is obtained by dividing the number of out-of-stock items (number of orders received - number of shipments) by the number of orders received.

First, the configuration of the parameter search device 10 according to the present embodiment will be described with reference to FIG. FIG. 2 is a block diagram showing a configuration example of the parameter search device 10 according to this embodiment.
As shown in FIG. 2, the parameter search device 10 according to the present embodiment includes a division section 11, a search section 12, a determination section 13, and an input section .

The dividing unit 11 divides a predetermined two-dimensional space with the maximum inventory days and the safety inventory days as axes into squares.

ここで、ｋ_ｉ（ｉ＝１～４）は、各指標の桁合わせをするための除数であり、指標毎に計算される。また、ω_ｉは、各指標の重要度に応じて各指標の重み付けをするための重み係数であり、指標毎に設定される。 The search unit 12 has a function of calculating the value of the objective function f for a predetermined two-dimensional space. The objective function f is represented, for example, by the following formula (1).

Here, k _i (i=1 to 4) is a divisor for adjusting the digits of each index and is calculated for each index. ω _i is a weighting factor for weighting each index according to the importance of each index, and is set for each index.

ここで、ｘ_ｉは、指標である。また、ｆｌｏｏｒ（ｔ）は、ｔ以下の整数の中で最大の整数を求める床関数である。
これにより、後述の探索開始点となるマスでは、数式（１）におけるｘ_ｉ／ｋ_ｉの値は、０．５以上１．０未満になる。 The divisor k _i is calculated, for example, by the searching unit 12 by the following formula (2) based on the value of the index in the cell serving as the search starting point, which will be described later.

where x _i is the index. Also, floor(t) is a floor function that obtains the largest integer among integers less than or equal to t.
As a result, the value of x _i /k _i in the equation (1) is 0.5 or more and less than 1.0 in the squares that serve as search starting points, which will be described later.

For example, all weighting factors ω _i are set to 1.0 when all indices are considered equivalent. Also, with respect to ω _i , for example, if the importance of the out-of-stock rate is higher than the importance of other indices, only the out-of-stock rate ω _i is set to 2.0, and the other indices ω _i are set to 1 .0 for example.
The weighting factor ω _i is input by the user through the input unit 14, for example.

When the search unit 12 calculates the value of the objective function f for a predetermined two-dimensional space, the calculation is performed as follows.
First, the search unit 12 sets the average number of days in stock, the average number of days in stock, the number of orders, and the out-of-stock rate so that the values of the maximum number of days in stock and the number of days in safety stock correspond to the corresponding square ( (by simulation, etc.).
Next, the search unit 12 substitutes the values of the average number of days in stock, the average amount of inventory, the number of orders, and the out-of-stock rate calculated above into the objective function f.
Thereby, the value of the objective function f of the corresponding mass is calculated.

The search unit 12 searches for a square having the minimum value of the objective function f from among the squares in a predetermined two-dimensional space.
The determining unit 13 determines the values of the maximum number of days in stock and the number of safety days in stock corresponding to the mass having the minimum value of the objective function f as the optimum solution.
The input unit 14 is a part where various inputs are performed by the user, and for example, the weighting coefficient ω _i set for each index x _i is input as described above.

Here, with reference to FIG. 3, an image of a method of searching for a square by the searching unit 12 according to the present embodiment will be described. FIG. 3 is a diagram showing an image example of a square search method by the search unit 12 according to the present embodiment. In FIG. 3, the two axes of a predetermined two-dimensional space are the maximum inventory days and the safe inventory days, and the axis perpendicular to this predetermined two-dimensional space is the objective function f.

As shown in FIG. 3, the search unit 12 uses one of the squares in a predetermined two-dimensional space as a search starting point, and uses the direction in which the value of the objective function f decreases the most (that is, the direction in which the slope is steepest). ), searching for the square with the minimum value of the objective function f.

Here, with reference to FIGS. 4 and 5, a method of searching for a square by the search unit 12 according to the present embodiment will be described in detail. 4 and 5 are diagrams showing an example of a method of searching for a square by the searching unit 12 according to the present embodiment.

In the example of FIG. 4, the searching unit 12 first sets one of the squares in a predetermined two-dimensional space as a search starting point. The square of the search starting point is set, for example, to a square that is estimated to have the maximum value of the objective function f among squares in a predetermined two-dimensional space.

Subsequently, the search unit 12 calculates the values of the objective function f for the square at the search start point (assumed to be the current square) and the squares in eight directions around it (that is, all around). Move to the square with the smallest value of the function f, and set the destination square as the new current square.

After that, the search unit 12 calculates the values of the objective function f of the current square and the squares in the eight directions around it, and selects the squares with smaller objective function f values than the current square from among the squares in the eight directions around the current square. recursively search for . However, squares outside the search range (such as squares where the number of days is less than 0) are excluded from the search.

When the searching unit 12 fails to find the surrounding squares to be searched in the search processing, the current square is the square having the minimum value of the objective function f among the squares in the predetermined two-dimensional space. , and ends the search for the square.

However, in the example of FIG. 4, the searching unit 12 calculates all the values of the objective function f of the squares in the eight directions around the current square every time the square is moved, so the search time for the square increases. is concerned.

The example in FIG. 5 is intended to shorten the search time for cells compared to the example in FIG. In the example of FIG. 5, the search processing performed after moving from the search start point square includes two search modes, an oblique search mode and an up/down/left/right search mode.

In the diagonal search mode, the values of the objective function f are calculated for the current square and its surrounding squares in four diagonal directions. This is the process of searching for small squares.

In the up, down, left, and right search mode, the values of the objective function f of the current square and the surrounding squares in four directions, up, down, left, and right, are calculated. This is a process of searching for a square with a small value of .

The example of FIG. 5 is an example in which the direction of movement from the square of the search start point is one of four oblique directions. In this case, among the squares around the square of the search start point, the squares in the oblique direction have the smallest value of the objective function f. Therefore, it is highly probable that the steepest direction in the vicinity of the search start point is the diagonal direction from the search start point square. Therefore, the search unit 12 executes the diagonal search mode.

When the search unit 12 finds the surrounding squares that were the search target in the diagonal search mode, the search unit 12 moves to the surrounding squares having the smallest value of the objective function f among the surrounding squares found in the diagonal search mode, The diagonal search mode is recursively executed with the destination square as the new current square. In this way, when the surrounding squares to be searched in the oblique search mode are found, calculation of squares in four directions, up, down, left, and right, can be omitted, so that the square search time can be shortened.

On the other hand, if the search unit 12 cannot find the surrounding squares that were the search target in the oblique search mode, it switches the oblique search mode to the up, down, left, and right search mode, and executes the up, down, left, and right search mode. When the search unit 12 finds the surrounding squares that were the search target in the up/down/left/right search mode, the search unit 12 moves to the surrounding squares having the smallest value of the objective function f among the surrounding squares found in the up/down/left/right search mode. Then, the up/down/left/right search mode is recursively executed with the destination square as the new current square. Conversely, if the search unit 12 cannot find the surrounding squares that were the search target even in the up, down, left, and right search mode, the search unit 12 determines that the current square is the smallest value of the objective function f among the squares in the predetermined two-dimensional space. It is determined that the square is a value, and the search for the square ends.

In this manner, in the example of FIG. 5, the search unit 12 performs the search process while switching between the diagonal search mode and the up/down/left/right search mode. Then, if the search unit 12 fails to find the surrounding squares that were the search target in both the diagonal search mode and the up/down/left/right search mode, the search unit 12 determines that the current square is a square in a predetermined two-dimensional space with the objective function f value is the minimum value.

FIG. 6 is a diagram showing an example of a search path when the searching unit 12 according to this embodiment searches for a square by the method of FIG.
In the example of FIG. 6, the upper left square is set as the search starting point, and the diagonal search mode is first executed, followed by switching to the up/down/left/right search mode, and then switching to the diagonal search mode again. , the search ends in the lower right square. That is, in the example of FIG. 6, among the squares in the predetermined two-dimensional space, the lower right square is determined to be the square having the minimum value of the objective function f.

A parameter searching method by the parameter searching device 10 according to the present embodiment will be described below with reference to FIG. FIG. 7 is a flowchart showing an example of a parameter searching method by the parameter searching device 10 according to this embodiment.

As shown in FIG. 7, the dividing unit 11 divides a predetermined two-dimensional space around the maximum number of days in stock and the number of days in safety stock into squares (step S101). For example, division is performed in units of one day.
The searching unit 12 searches for a square having the minimum value of the objective function f from among the squares in a predetermined two-dimensional space (step S102).
The determination unit 13 determines the values of the maximum number of days in stock and the number of days in safety corresponding to the mass having the minimum value of the objective function f as the optimum solution (step S103).

Next, referring to FIG. 8, the process of step S102 in FIG. 7 will be described in detail. FIG. 8 is a flow chart showing an example of the processing in step S102 of FIG. Note that FIG. 8 shows the processing when searching for a square by the method of FIG.

As shown in FIG. 8, the search unit 12 first determines one of the squares in a predetermined two-dimensional space as a search start point (step S201). Subsequently, the search unit 12 calculates the value of the objective function f for the square at the search start point and the squares in the eight directions around it (step S202). Move to a square and set the destination square as a new current square (step S203).

Subsequently, the search unit 12 determines the direction of movement (one of the four directions up, down, left, and right, or one of the four diagonal directions) from the square of the search start point (step S204).

If the direction of movement from the square of the search start point is one of the four directions of up, down, left, and right, the search unit 12 executes the up, down, left, and right search mode. That is, the search unit 12 calculates the values of the objective function f of the current square and the squares in four directions, up, down, left, and right (step S205). A square with a smaller value of the objective function f than is searched for (step S206).

When the search unit 12 finds the surrounding squares that were the search target in the up/down/left/right search mode (Yes in step S206), the search unit 12 finds the smallest value of the objective function f among the surrounding squares found in the up/down/left/right search mode. Move to a surrounding square (step S207), and set the new square as the new current square. Then, the search unit 12 proceeds to step S205 and recursively executes the up/down/left/right search mode.

If the search unit 12 cannot find the surrounding squares that were the search target in the up/down/left/right search mode (No in step S206), the search unit 12 then executes the diagonal search mode. That is, the search unit 12 calculates the values of the objective function f of the current square and its surrounding squares in four diagonal directions (step S208), A square with a small value of the objective function f is searched (step S209).

When the search unit 12 finds the surrounding squares that were the search target in the diagonal search mode (Yes in step S209), the searching unit 12 searches the surrounding squares having the smallest value of the objective function f among the surrounding squares found in the diagonal search mode. (step S210), and the destination square is set as a new current square. Then, the search unit 12 proceeds to step S211 and recursively executes the diagonal search mode.

If the search unit 12 cannot find the surrounding squares that were the search target even in the diagonal search mode (No in step S209), the search unit 12 determines that the current square is the value of the objective function f among the squares in the predetermined two-dimensional space. is determined to be the square with the minimum value, and the search for the square ends.

On the other hand, if the movement direction from the search start point square is one of the four diagonal directions, the search unit 12 executes the diagonal search mode similar to steps S209 and S210 (steps S211 and S212). .

If the search unit 12 finds the surrounding squares that were the search target in the diagonal search mode (Yes in step S212), the searching unit 12 searches the surrounding squares that have the smallest value of the objective function f among the surrounding squares found in the diagonal search mode. (step S213), and the destination square is set as a new current square. Then, the search unit 12 proceeds to step S211 and recursively executes the diagonal search mode.

If the search unit 12 cannot find the surrounding squares to be searched in the oblique search mode (No in step S212), then the search unit 12 executes the up/down/left/right search mode similar to steps S206 and S207 described above. (Steps S214, S215).

If the search unit 12 finds the surrounding squares that were the search target in the up/down/left/right search mode (Yes in step S215), the search unit 12 finds the smallest value of the objective function f among the surrounding squares found in the up/down/left/right search mode. Move to a surrounding square (step S216), and set the destination square as a new current square. Then, the search unit 12 proceeds to step S205 and recursively executes the up/down/left/right search mode.

If the search unit 12 cannot find the surrounding squares that were the search target even in the up/down/left/right search mode (No in step S215), the search unit 12 determines that the current square is the target function f among the squares in the predetermined two-dimensional space. It is determined that the square has the minimum value, and the search for the square ends.

FIG. 9 is a diagram showing an example of a search path when the searching unit 12 according to the present embodiment searches for squares by the method of FIG.
In the example of FIG. 9, the search unit 12 has determined the lower right square with the maximum inventory days of 20 days and the safety inventory days of 10 days as the search start point.
In addition, in the search starting point, the cell with the smallest value of the objective function f among the eight surrounding cells is the cell in the left direction.

Therefore, the search unit 12 moves from the search start point to the leftward square, and thereafter recursively executes the up, down, left, and right search mode until the maximum inventory days is 16 days and the safety inventory days is 10 days. moving.

However, in the current cell with the maximum inventory days of 16 days and the safety inventory days of 10 days, none of the cells in the vertical and horizontal directions surrounding it has a smaller value of the objective function f than the current cell. On the other hand, in the current cell, among the cells in the oblique direction around it, there is a cell with a smaller objective function f value than the current cell. is the square in the upper left diagonal direction.

Therefore, the search unit 12 moves from the current square to the upper left diagonal square, recursively executes the diagonal search mode, and moves to the square with the maximum inventory days of 9 days and the safety inventory days of 3 days. .

However, for the current cell with the maximum inventory days of 9 days and the safety inventory days of 3 days, there is no cell with a smaller value of the objective function f than the current cell among the cells in the oblique direction around it. Further, in the current cell, there is no cell having a smaller value of the objective function f than the current cell in the cells in the vertical and horizontal directions around it.
Therefore, the search unit 12 determines that the current mass is the mass with the minimum value of the objective function f, and the determination unit 13 determines the value of the maximum inventory days (9 days) corresponding to the current mass and the safety stock The value of days (3 days) is determined to be the optimal solution for maximum days of supply and safety days of supply, respectively.

As described above, according to the present embodiment, the parameter search device 10 divides a predetermined two-dimensional space with the maximum inventory days and the safety inventory days as axes into squares, and divides the predetermined two-dimensional space into squares. Among them, the mass with the minimum value of the objective function f is searched, and the respective values of the maximum number of days in stock and the number of days in safety stock corresponding to the found mass are determined as optimum solutions. Therefore, it is possible to search for the optimum solution for each of the two parameters, the maximum number of days in stock and the number of safety days in stock, without performing complicated calculations.

Therefore, even an inexperienced user can set the maximum number of days in stock and the number of days in safety stock appropriately. You can place an order for the product in time. This can contribute to preventing the occurrence of out-of-stock or shortage of products due to shortage of the number of products in stock.

<Modified example of the embodiment>
In the above embodiment, the divisor k _i for digit alignment for each index x _i of the objective function f is calculated based on the value of the index x _i in the square at the search start point, and thereafter, even if the square moves, , the divisor k _i was fixed. However, since this divisor k _i is calculated based on the value of the index x _i in the search starting point square, it may not become an appropriate value as the square moves.
Therefore, in this disclosure, the divisor k _i may be calculated each time a square is moved. In this case, the divisor k _i for each index x _i of the objective function f used for searching the surroundings in the destination cell should be the value of the index x _i in the destination cell.

In addition, in the above embodiment, an example in which the indicators for determining the maximum number of days in stock and the number of days in safety stock are the average number of days in stock, the average amount of inventory, the number of orders, and the out-of-stock rate has been described, but the present disclosure is not limited to this. In the present disclosure, the indices for determining the maximum number of days in stock and the number of safety days in stock may be a plurality of indices including at least one of these four indices.

Further, in the above-described embodiment, an example has been described in which two parameters for searching for the optimum solution are the maximum inventory days and the safety inventory days of the product, but the present disclosure is not limited to this. In the present disclosure, the parameters for searching for the optimum solution may be parameters in fields other than the field of inventory management in distribution centers and the like. In this case, the indices for determining the parameters of other fields may be appropriately determined according to the relevant fields. In addition, the present disclosure is applicable to searching for optimal solutions for N (N is a natural number equal to or greater than 2) parameters, and N is not limited to 2. In this case, a predetermined N-dimensional space with N parameters as an axis is divided into squares, and a square having the minimum value of the objective function is searched from among the squares in the predetermined N-dimensional space, The value of each of the N parameters corresponding to the mass with the minimum value of the objective function is determined as the optimum solution.

<Case where N is other than 2>
A specific case where N is other than 2 will be described below.
This example is an application of the present disclosure to the field of breakfast nutrition management. In this case, for example, when the breakfast in the dormitory is a system in which a choice is made from several menus, nutritional management is performed so as to reduce salt and sugar content and reduce food costs.

In this example, the parameter is the number of days to select each menu. Specifically, the following four parameters are used (that is, N is 4).
Rice with natto …… N ₁ day/month Salmon grilled with salt …… N ₂ days/month Ham and eggs …… N ₃ days/month Cereal …… N ₄ days/month Here, the constraint conditions for N ₁ to N ₄ are as follows. .
_N1 + _N2 +N3+ _N4 = ₃₀ (or 31)
In addition, four indices for determining the number of days for each menu are food expenses, salt intake, sugar intake, and cholesterol intake.

In the case of this example, a square having the minimum value of the objective function f is searched from squares in a predetermined four-dimensional space.
The objective function f in this example conforms to the above formula (1), and is a function using the four indices of food expenses, salt intake, sugar intake, and cholesterol intake.

In this example, when calculating the value of the objective function f for a predetermined square in a four-dimensional space, first, food expenses, salt content Values for intake, sugar intake, and cholesterol intake are calculated. Next, the values of food expenses, salt intake, sugar intake, and cholesterol intake calculated above are substituted into the objective function f. As a result, the value of the objective function f of the corresponding mass is calculated.

<Processing to search for squares in N-dimensional space>
In the following, a detailed description will be given of the process of searching for a square in a predetermined N-dimensional space.
(A) For two-dimensional space (i.e., when N is 2)
The above embodiment is an example of a two-dimensional space.
In the case of a two-dimensional space as in the above embodiment, in the search processing performed after moving from the search start point square, as described with reference to FIG. Two search modes with different directions are switched and executed. These two search modes are, in other words, as follows.
First search mode (up/down/left/right search mode):
One axis is extracted from the two axes, and a square located at a position ±1 from the current square in the axial direction of the extracted one axis is searched (other axes are 0). The search directions are ₂ C ₁ ×2 ¹ =4 directions.
Second search mode (diagonal search mode):
2 axes are extracted from the 2 axes, and a square located at a position ±1 from the current square in each axial direction of the extracted 2 axes is searched. The search directions are ₂ C ₂ ×2 ² =4 directions.
In the case of a two-dimensional space, the total circumference of the current square is 3 ² -1=8 directions.

(B) For 3-dimensional space (i.e., when N is 3)
In the case of a three-dimensional space, in the search processing performed after moving from the search start point square, as shown in FIG. 10, three search modes with mutually different search directions are switched and executed. The three search modes are as follows.
First search mode:
One axis is taken out of the three axes, and a mass located at a position ±1 from the current mass in the axial direction of the taken out one axis is searched (other axes are 0). The search directions are ₃ C ₁ ×2 ¹ =6 directions.
Second search mode:
2 axes are extracted from the 3 axes, and a square located at a position ±1 from the current square in each axial direction of the extracted 2 axes is searched (other axes are 0). The search directions are ₃ C ₂ ×2 ² =12 directions.
Third search mode:
3 axes are taken out from the 3 axes, and a square located at a position ±1 from the current square in each axial direction of the taken out 3 axes is searched. The search directions are ₃ C ₃ ×2 ³ =8 directions.
In the case of a three-dimensional space, there are 3 ³ −1=26 directions around the current square.

(C) Case of N-Dimensional Space Search processing for two-dimensional space and three-dimensional space is as described above.
Based on this, in the case of the N-dimensional space, in the search processing performed after moving from the search start point square, N search modes with different search directions are switched and executed. The N search modes are as follows.
First search mode:
One axis is taken out from the N axis, and a mass located at a position ±1 from the current mass in the axial direction of the taken out one axis is searched (other axes are 0). The search direction is the _{NC 1} ×2 ¹ direction.
Second search mode:
Two axes are extracted from the N axis, and a mass located at a position ±1 from the current mass is searched for in each axial direction of the extracted two axes (other axes are 0). The search direction is _{NC 2} ×2 ² directions.
Nth search mode:
The N-axis is extracted from the N-axis, and a mass located at a position ±1 from the current mass is searched for in each axial direction of the extracted N-axis. The ^{search direction is the NCN×2N} _{direction .}
In the case of an N-dimensional space, the directions around the current square are represented by the following formula (3).

<Concept of Embodiment>
Next, a configuration conceptually showing the parameter search device 10 according to the above embodiment will be described. FIG. 11 is a block diagram showing a configuration example of a parameter searching device 100 conceptually showing the parameter searching device 10 according to the above embodiment.

As shown in FIG. 11, the parameter searching device 100 comprises a dividing section 101, a searching section 102, and a determining section 103. FIG. The dividing section 101 corresponds to the dividing section 11 , the searching section 102 corresponds to the searching section 12 , and the determining section 103 corresponds to the determining section 13 .

A dividing unit 101 divides a predetermined N-dimensional space into grids.
The searching unit 102 searches for a square having the minimum value of the objective function from squares in a predetermined N-dimensional space.
The determination unit 103 determines the values of the N parameters corresponding to the mass with the minimum value of the objective function as the optimum solution.

The search unit 102 will be described in detail below.
First, the search unit 102 sets one of the squares on a predetermined N-dimensional space as a search start point. The search starting point is, for example, the square in which the value of the objective function is assumed to be the maximum among the squares in the predetermined N-dimensional space.

Next, the search unit 102 calculates the objective function values of the search start point square and the squares around it, moves to the squares around it that have the smallest objective function value, and moves to the surrounding squares. is the new current cell. Then, the search unit 102 calculates the objective function values of the current square and its surrounding squares, and performs a search process of searching for a square having a smaller objective function value than the current square from among the squares surrounding the current square. Run recursively.

If the searching unit 102 fails to find the surrounding squares to be searched in the search processing, the search unit 102 determines that the current square is the square having the minimum value of the objective function among the squares in the predetermined N-dimensional space. Make a decision and end the search.

Here, the search processing includes N search modes in which the square search directions are different from each other, and the search unit 102 performs the search processing while switching between the N search modes. The search process when N is 2 includes two search modes, the up/down/left/right search mode and the oblique search mode, as described above.

Also, the objective function is obtained by dividing each value of a plurality of indices that determine the values of each of the N parameters by a divisor calculated for each index and multiplying by a weighting factor set for each index. Above is the added function. The search unit 102 calculates the values of each of the plurality of indices so as to be the values of the N parameters corresponding to the square, and substitutes the calculated values of the plurality of indices into the objective function. , to calculate the value of the objective function of the corresponding cell.

Note that the search unit 102 may calculate the divisor for each index of the objective function based on the value of the index in the search start point cell. Alternatively, the search unit 102 may calculate the divisor for each index of the objective function each time the square is moved. When calculating each time a square is moved, the search unit 102 may set the divisor for each index of the objective function used to search the surroundings of the destination square as the value of the index in the destination square. .

Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above embodiments. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure.

For example, in the above embodiment, the parameter search device of the present disclosure has been described as a hardware configuration, but the present disclosure is not limited to this. According to the present disclosure, the parameter search device incorporates a processor and memory, and the processor reads and executes a computer program stored in the memory, thereby realizing arbitrary processing of the parameter search device.

In the above examples, the programs can be stored and delivered to the computer using various types of non-transitory computer readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (eg, flexible discs, magnetic tapes, hard disk drives), magneto-optical recording media (eg, magneto-optical discs), CD-ROMs (Compact Disc-Read Only Memory), CDs - R (CD-Recordable), CD-R/W (CD-ReWritable), semiconductor memory (e.g. mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (Random Access Memory)) include. The program may also be delivered to the computer by various types of transitory computer readable medium. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. Transitory computer-readable media can deliver the program to the computer via wired channels, such as wires and optical fibers, or wireless channels.

Some or all of the above embodiments can also be described in the following additional remarks, but are not limited to the following.
(Appendix 1)
a dividing unit that divides a predetermined N-dimensional space with N (N is a natural number equal to or greater than 2) parameters as axes;
a searching unit for searching, from among the squares in the predetermined N-dimensional space, a square having a minimum value of an objective function obtained using the respective values of N parameters corresponding to the square;
a determination unit that determines the values of the N parameters corresponding to the mass at which the value of the objective function is the minimum value as an optimal solution;
A parameter search device comprising:
(Appendix 2)
The search unit is
With one of the squares on the predetermined N-dimensional space as a starting point,
Calculate the value of the objective function for the current square serving as the starting point and surrounding squares, move to the surrounding square with the smallest objective function value among the squares around the starting square, and recursively execute search processing with the square as the new current square,
If the search process fails to find the surrounding squares to be searched, it is determined that the current square is the square having the minimum value of the objective function among the squares in the predetermined N-dimensional space. ,
A parameter search device according to appendix 1.
(Appendix 3)
The square of the starting point is, among the squares in the predetermined N-dimensional space, the square that is estimated to have the maximum value of the objective function.
The parameter search device according to appendix 2.
(Appendix 4)
The search process includes
including N search modes in which the search directions of the squares are different from each other;
The search unit is
performing the search process while switching between N search modes;
A parameter search device according to appendix 2 or 3.
(Appendix 5)
N is 2;
The search process includes
calculating the value of the objective function of the current square and its surrounding squares in the first search direction; a first search mode for searching for
calculating the value of the objective function of the current square and its surrounding squares in the second search direction; a second search mode that searches for
The search unit is
If the direction of the destination square to which the current square has been moved from the starting point is the first search direction, the first search mode is executed, and the surrounding squares that were searched in the first search mode are searched. If it is found, it moves to the surrounding square having the smallest value of the objective function among the surrounding squares found in the first search mode, sets the destination square as a new current square, and resumes the first search mode. run recursively,
If the direction of the destination square to which the current square as the starting point has been moved is the second search direction, the second search mode is executed, and the surrounding squares that were searched in the second search mode are searched. If it is found, it moves to the surrounding square having the smallest value of the objective function among the surrounding squares found in the second search mode, sets the destination square as a new current square, and resumes the second search mode. run recursively,
The parameter search device according to appendix 4.
(Appendix 6)
The search unit is
When the search target surrounding squares cannot be found in the first search mode, the second search mode is subsequently executed, and the search target surrounding squares cannot be found even in the second search mode. In this case, the current square is determined to be the square having the minimum value of the objective function among the squares in the predetermined two-dimensional space, and the surrounding squares that were searched in the second search mode are searched. If it is found, it moves to the surrounding square having the smallest value of the objective function among the surrounding squares found in the second search mode, sets the destination square as a new current square, and resumes the second search mode. run recursively,
If the search target surrounding squares cannot be found in the second search mode, then the first search mode is executed, and the search target surrounding squares cannot be found even in the first search mode. In this case, the current square is determined to be the square having the minimum value of the objective function among the squares in the predetermined two-dimensional space, and the surrounding squares that were searched in the first search mode are searched. If it is found, it moves to the surrounding square having the smallest value of the objective function among the surrounding squares found in the first search mode, sets the destination square as a new current square, and resumes the first search mode. run recursively,
A parameter search device according to appendix 5.
(Appendix 7)
The objective function divides each value of a plurality of indices that determine the values of each of the N parameters by a divisor calculated for each of the indices, and multiplies by a weighting factor set for each of the indices. and then add the function,
The search unit is
When calculating the value of the objective function of the mass on the predetermined N-dimensional space, calculating the values of each of the plurality of indices so as to be the respective values of N parameters corresponding to the mass, and calculating substituting the value of each of the plurality of indices obtained into the objective function;
5. The parameter search device according to any one of appendices 2 to 4.
(Appendix 8)
The search unit is
calculating the divisor for each index based on the value of the index in the starting point square;
A parameter search device according to appendix 7.
(Appendix 9)
The search unit is
calculating the divisor for each index each time a square is moved;
Let the divisor for each index of the objective function used to search the surroundings in the destination square be the value of the index in the destination square;
A parameter search device according to appendix 7.
(Appendix 10)
N is 2;
The parameters are the maximum inventory days and safety inventory days of the product,
The indicator includes at least one of the average inventory days, average inventory value, number of orders, and out-of-stock rate of the product,
The parameter search device according to any one of appendices 7 to 9.
(Appendix 11)
A parameter search method by a parameter search device,
a step of dividing a predetermined N-dimensional space with N (N is a natural number equal to or greater than 2) parameters as axes;
a step of searching from among the squares in the predetermined N-dimensional space for a square having the minimum value of an objective function obtained using the respective values of N parameters corresponding to the square;
a step of determining the values of each of the N parameters corresponding to the mass at which the value of the objective function is the minimum value as an optimal solution;
Parameter search method, including
(Appendix 12)
the computer,
dividing means for dividing a predetermined N-dimensional space with N (N is a natural number equal to or greater than 2) parameters as axes;
search means for searching, from among the squares in the predetermined N-dimensional space, a square having a minimum value of an objective function obtained using respective values of N parameters corresponding to the square;
Determination means for determining the values of each of the N parameters corresponding to the mass at which the value of the objective function is the minimum value as an optimal solution;
A program to function as

REFERENCE SIGNS LIST 10 parameter search device 11 division unit 12 search unit 13 determination unit 14 input unit 100 parameter search device 101 division unit 102 search unit 103 determination unit

Claims (6)

a dividing unit that divides a predetermined N-dimensional space with N (N is a natural number equal to or greater than 2) parameters as axes;
a searching unit for searching, from among the squares in the predetermined N-dimensional space, a square having a minimum value of an objective function obtained using the respective values of N parameters corresponding to the square;
a determination unit that determines the values of the N parameters corresponding to the mass at which the value of the objective function is the minimum value as an optimal solution;
with
The search unit is
With one of the squares on the predetermined N-dimensional space as a starting point,
Calculate the value of the objective function for the current square serving as the starting point and surrounding squares, move to the surrounding square with the smallest objective function value among the squares around the starting square, and recursively execute search processing with the square as the new current square,
If the search processing fails to find the surrounding squares to be searched, it is determined that the current square is the square having the minimum value of the objective function among the squares in the predetermined N-dimensional space. ,
The objective function divides each value of a plurality of indices that determine the values of each of the N parameters by a divisor calculated for each of the indices, and multiplies by a weighting factor set for each of the indices. and then add the function,
The search unit is
When calculating the value of the objective function of the mass on the predetermined N-dimensional space, calculating the values of each of the plurality of indices so as to be the respective values of N parameters corresponding to the mass, and calculating substituting the value of each of the plurality of indices obtained into the objective function;
Parameter searcher. The search process includes
including N search modes in which the search directions of the squares are different from each other;
The search unit is
performing the search process while switching between N search modes;
The parameter searching device according to claim 1 .
The search unit is
calculating the divisor for each index based on the value of the index in the starting point square;
The parameter searching device according to claim 1 . The search unit is
calculating the divisor for each index each time a square is moved;
Let the divisor for each index of the objective function used to search the surroundings in the destination square be the value of the index in the destination square;
The parameter searching device according to claim 1 . A parameter search method by a parameter search device,
a step of dividing a predetermined N-dimensional space with N (N is a natural number equal to or greater than 2) parameters as axes;
a searching step of searching, from among the squares in the predetermined N-dimensional space, the square having the minimum value of the objective function obtained using the respective values of the N parameters corresponding to the square;
a step of determining the values of each of the N parameters corresponding to the mass at which the value of the objective function is the minimum value as an optimal solution;
including
In the search step,
With one of the squares on the predetermined N-dimensional space as a starting point,
Calculate the value of the objective function for the current square serving as the starting point and surrounding squares, move to the surrounding square with the smallest objective function value among the squares around the starting square, and recursively execute search processing with the square as the new current square,
If the search processing fails to find the surrounding squares to be searched, it is determined that the current square is the square having the minimum value of the objective function among the squares in the predetermined N-dimensional space. ,
The objective function divides each value of a plurality of indices that determine the values of each of the N parameters by a divisor calculated for each of the indices, and multiplies by a weighting factor set for each of the indices. and then add the function,
In the search step,
When calculating the value of the objective function of the mass on the predetermined N-dimensional space, calculating the values of each of the plurality of indices so as to be the respective values of N parameters corresponding to the mass, and calculating substituting the value of each of the plurality of indices obtained into the objective function;
Parameter search method. the computer,
dividing means for dividing a predetermined N-dimensional space with N (N is a natural number equal to or greater than 2) parameters as axes;
search means for searching, from among the squares in the predetermined N-dimensional space, a square having a minimum value of an objective function obtained using respective values of N parameters corresponding to the square;
Determination means for determining the values of each of the N parameters corresponding to the mass at which the value of the objective function is the minimum value as an optimal solution;
A program for functioning as
The search means is
With one of the squares on the predetermined N-dimensional space as a starting point,
Calculate the value of the objective function for the current square serving as the starting point and surrounding squares, move to the surrounding square with the smallest objective function value among the squares around the starting square, and recursively execute search processing with the square as the new current square,
If the search processing fails to find the surrounding squares to be searched, it is determined that the current square is the square having the minimum value of the objective function among the squares in the predetermined N-dimensional space. ,
The objective function divides each value of a plurality of indices that determine the values of each of the N parameters by a divisor calculated for each of the indices, and multiplies by a weighting factor set for each of the indices. and then add the function,
The search means is
When calculating the value of the objective function of the mass on the predetermined N-dimensional space, calculating the values of each of the plurality of indices so as to be the respective values of N parameters corresponding to the mass, and calculating substituting the value of each of the plurality of indices obtained into the objective function;
program .

Images