JP2019197374A - Number-of-purchases determination assisting system for foods - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a food purchase quantity determination support system capable of combining the purchase quantity of each item so as to maximize the gross profit while minimizing the discount, disposal and opportunity loss. A food purchase number determination support system includes a learning unit that creates a learning model of an optimum combination of food purchase numbers, and an evaluation value for each purchase number candidate of each item based on the learning model. And a calculation unit for calculating an evaluation value for each allocation ratio candidate when the number of purchases of an item is allocated to a plurality, the learning model is at least a purchase number candidate for each item, a state description parameter, and a discount rate, And a first evaluation function, an allocation ratio candidate when the purchase number of one item is allocated to a plurality of items, a state description parameter, and a second evaluation function. [Selection diagram] Figure 1

Description

  The present disclosure relates generally to a food purchase number determination support system, and more particularly, to a food purchase number determination support system for maximizing gross profit considering opportunity loss, disposal loss, and discount loss.

  In the food procurement business, the number of items to be purchased is determined for each item from the number of stocks, expiration date, sales record, and the like. Items that are close to the expiration date are sold at a discount, and if the items whose expiration date has expired remain unsold, they are discarded. Discounts are accompanied by discounts, disposal is lost, and out of stock is accompanied by opportunity losses.

  At present, the decision-making part in the food procurement business is done by humans and is treated as a highly specialized business. The number of items (products) is, for example, about 20,000, and it is difficult for a person with specialized knowledge to grasp the whole. For this reason, human resources are always in short supply.

  In addition, since the number of items (products) is tens of thousands and there are several to tens of options for the number of purchases for each item, a combined explosion will occur. It does not function effectively in determining the number of food items purchased. Furthermore, since there are many items having a correlation with respect to the number of purchases, for example, it is difficult to apply the existing AI method.

JP 2003-182854 A

R. Sutton and A. Barto: Reinforcement Learning, MIT Press (1999). Y. Hirashima: “A reinforcement learning method for train marshaling based on movements of locomotive”, IAENG Int. J. Comput. Sci., Vol. 38, No. 3, 242/248 (2011).

  In this disclosure, the number of purchases of each item can be combined so as to maximize the gross profit while minimizing discounts / disposals / opportunity loss. The object is to present a food purchase number determination support system that is a decision support system that can show superiority or inferiority.

The present disclosure is intended to achieve the above object. The food purchase number determination support system according to the present disclosure is a food purchase number determination support system according to the optimal combination learning and derivation method of food purchase numbers,
A learning unit that creates a learning model for an optimal combination of food purchases, and
Based on the learning model, an evaluation value for each purchase number candidate of each item, and a calculation unit that calculates an evaluation value for each allocation ratio candidate when the number of purchases of one item is assigned to a plurality,
The learning model includes at least a purchase candidate for each item, a state description parameter, a discount rate, a first evaluation function, and an allocation ratio candidate when the number of purchases of one item is assigned to a plurality of items, A state description parameter, and a second evaluation function,
The learning unit
For each item, a discount rate is determined based on the loss amount, and for each purchase candidate, a first evaluation function is determined based on the state description parameter and the determined gross profit,
Furthermore, when the number of purchases of one item is allocated to a plurality, a second evaluation function is determined for each allocation ratio candidate based on the state description parameter and the determined gross profit,
The computing unit is
Deriving an evaluation value for each purchase number candidate for each item from the first evaluation function for each purchase number candidate for each item in the learning model,
The evaluation value for each allocation ratio candidate when the number of purchases of one item is allocated to a plurality is derived from the second evaluation function for each allocation ratio candidate in the learning model.

  The loss of various losses can be reduced by carrying out the highly specialized work currently performed by humans by the food purchase number determination support system according to the present disclosure and reviewing the purchase of the entire handling item. Furthermore, the food purchase quantity determination support system according to the present disclosure can provide an educational effect related to purchase operations.

FIG. 1 is a diagram schematically illustrating a decision-making flow in a food purchase number determination support system according to the present disclosure. FIG. 2 is a block diagram of the food purchase number determination support system according to the first embodiment. FIG. 3 is a flowchart illustrating the operation of the food purchase number determination support device in the food purchase number determination support system according to the first embodiment. FIG. 4 is a flowchart showing the detailed operation of the preliminary learning process in the food purchase quantity determination support device according to the first embodiment. FIG. 5 is an example of a screen displayed by the display unit of the food purchase number determination support apparatus according to the first embodiment, which displays the number of purchases for the item A and its evaluation.

  Hereinafter, embodiments will be described in detail with reference to the drawings as appropriate. However, more detailed description than necessary may be omitted. For example, detailed descriptions of already well-known matters and repeated descriptions for substantially the same configuration may be omitted. This is to avoid the following description from becoming unnecessarily redundant and to facilitate understanding by those skilled in the art.

  In addition, the inventors provide the accompanying drawings and the following description in order for those skilled in the art to fully understand the present disclosure, and these are intended to limit the subject matter described in the claims. is not.

[Background to this disclosure]
The food purchasing operation involves determining and ordering the number of items to be purchased every day for tens of thousands of items, and is an important operation that directly affects the profits of stores such as supermarkets and grocery stores.

  Each item has a expiration date and is discarded after the expiration date. The cost of disposal is called a disposal loss. There are cases where sales promotion measures are taken with discounts for products whose expiry date is approaching in order to reduce waste loss. In this case, a discount loss with respect to the suggested retail price occurs. In general, inventory is prepared in order to prevent loss of sales opportunities due to out of stock, but a sales loss caused by out of stock is called an opportunity loss.

  Furthermore, fresh foods often have shorter expiry dates than processed products, and the period of storage as stock is not always the same for each item. And there is one in which the purchase of one item correlates with a plurality of sales items, such as diverting fish that can be eaten raw to grilled foods, side dishes, and lunch boxes. In other words, in the food procurement business, when individually determining the number of items to be purchased, the number of stocks, sales, expiry date, purchase / sales status of correlated products, disposal / discount / opportunity loss status, gross profit It is necessary to consider.

  With the goal of maximizing gross profit while suppressing losses, the number of combinations to be examined to determine the number of purchases based on the correlation between the number of items and factors to be considered is enormous. Therefore, the current situation is that a small number of employees with high specialties are in charge of food procurement operations. These employees are not substitutable, and considerable on-site experience is required to develop and secure human resources with equivalent skills. Furthermore, no education method or training method has been established for human resources with such skills. For these reasons, there is a chronic shortage of human resources who can properly perform food procurement operations.

  On the other hand, it is possible to model a combination optimization problem with a loss amount or the like as an evaluation target for a combination of the number of purchases. However, since the combination of the number of purchases increases exponentially with respect to the number of items having a correlation, when the number of items exceeds 100, the solution by the conventional mathematical programming becomes difficult. It is well known that the sales of a plurality of related items increase or decrease depending on weather, seasons, events, current affairs information, etc. at the site of purchasing operations. Actually, the conventional method has not solved the problem of practical scale.

  In order to solve the above problems, the inventors of the present application support the determination of the number of food purchases according to the present disclosure related to the optimal combination learning / derivation method of purchase numbers using loss as an evaluation criterion. The system was devised. In the food purchase number determination support system according to the present disclosure, the number of purchases is sequentially determined, and the determined number of purchases is sequentially added as a state description parameter to cope with the correlation between items. Then, for a specific combination of individual purchase numbers, evaluation is calculated from the determined gross profit and loss amount, and the evaluation value is propagated to each purchase number. Propagation is based on optimal equations to ensure that the optimal number of purchases for optimal gross margin and loss is obtained at the decision-making stage for each item.

[1. Basic features of the present disclosure]
The basic features of the food purchase quantity determination support system according to the present disclosure will be described below.

[1.1. modeling]
[1.1.1 Purchase number determination work]
Assume that the number of items is k, the number of options for the number of purchases for the i-th item is m _i, and the number of purchases o _i is determined in the same order as the on-site purchase decision work. Although it depends on the operation mode at the site, at this time, as the external input (for example, from an external management server), the stock quantity of the item to be determined is set to s _i (i = 1,. _{Represents i} and gives the initial value. Also, let l _{i be} the expiry date, M _{i be} the remaining budget, and p _i = [p _i1 , _... , P _il ] be the number of purchases of other related products, and the state description parameter x _i is expressed by the following equation:

However, if p _ij (1 ≦ j ≦ l) is not yet determined, p _ij = 0. After the number of purchases is determined for all items, the number of sales on the day is estimated by a predetermined probability model, and the loss amount and gross profit are determined. For a given budget, the state transition combination [x ₁ , x ₂ , _... , X _k ] that maximizes the gross profit is the optimal solution.

  In the above model, it is difficult to realize a method that needs to hold all solution evaluations on a practical scale of tens of thousands of solution components.

から選択する。この際、状態パラメータx_iに対し、仕入れ数候補との組み合わせに対する評価値を

としてQ1_iの値に基づいて選択を行う。 [1.1.2 Learning model]
(1) Purchase value evaluation value and update rule For the model described in [1.1.1 Purchase number determination work], the number of purchases for the i-th item is determined for each candidate.

Select from. At this time, the evaluation value for the combination with the number of purchase candidates for the state parameter x _i

To select based on the value of Q1 _i .

ここで、αは学習率（係数）、γは割引率である。 When the number of purchases is determined, the selection result is represented by o _i and the corresponding evaluation value Q1 is updated by the following formula:

Here, α is a learning rate (coefficient), and γ is a discount rate.

つまり、Rは目標達成率であり、粗利確定までは値が０である。 R is a reward, and after the sale on the day, the target gross profit amount is set as _Go and the gross profit amount is determined as G:

That is, R is the target achievement rate, and the value is 0 until the gross profit is determined.

から選択する。この際、状態パラメータx_iに対し、割当比率との組み合わせに対する評価値を

としてQ2_iの値に基づいて関連品目への割当比率を選択する。 (2) When there are multiple sales forms for one item When the purchase of one item correlates with multiple sales items, such as diverting raw fish that can be eaten to pottery, side dishes, lunch boxes, etc., the corresponding sales The allocation ratio for the item affects the loss amount. Therefore, local evaluation is performed using the allocation ratio as an option. In other words, the allocation ratio candidate

Select from. At this time, the evaluation value for the combination with the allocation ratio for the state parameter x _i

Select the allocation ratio to the related item based on the value of Q2 _i .

の中から選択した結果w_iに対応するQ2_iを次式で更新する：

And

Update Q2 _i corresponding to the result w _i selected from with:

を選択するために、必要に応じてp_iの構成要素の一部を他品目の仕入れ数から割当比率に変更する。 Properly

In order to select, a part of the components of p _i is changed from the number of other items purchased to the allocation ratio as necessary.

ここで、D_iは第i品目で生じたロス額、T_pはk品目によって得られた当日の総売上額、δ は割引率の範囲を調整するパラメータである。上式（５）はロス額が小さい仕入れ数の選択肢に対する割引率を小さく設定する。また、粗利が大きくなる仕入れ数の組合せに対しては相対的に割引率が小さくなる。 [1.1.3 Calculation of discount rate]
The number of purchases for each item affects the gross margin and loss after sales.
In this case, in order to reflect the gross profit and the loss amount in the evaluation of candidates for each purchase number, o _i of the i-th item in the following equation, the discount rate for _{w i γ i (i = 1} , · · ·, k) Calculate:

Here, D _i is the loss amount generated in the i-th item, T _p is the total sales amount for the day obtained by the k item, and δ is a parameter for adjusting the range of the discount rate. In the above equation (5), the discount rate for the option of the number of purchases with a small loss amount is set small. In addition, the discount rate is relatively small for a combination of purchases in which the gross profit increases.

に収束する。Ｒは粗利確定後にのみに与えるので、仕入れ数の決定作業の進捗に伴って評価値が単調増加する。上式（２）〜（５）は、最良の選択肢に対する評価値を用いて伝播を行うので、各状態における評価値が目標状態に至る最良の仕入れ数の組合せを反映することになる。 The above formulas (2) and (4) are calculated after a sufficient number of updates

Converge to. Since R is given only after the gross profit is determined, the evaluation value increases monotonically with the progress of the work for determining the number of purchases. Since the above formulas (2) to (5) are propagated by using the evaluation value for the best option, the evaluation value in each state reflects the best purchase number combination that reaches the target state.

[1.1.4 Operation selection]
The target arrangement, the movement target, the wagon destination, and the number of moving wagons are selected stochastically by the Soft-Max method (see Non-Patent Document 1). In this update rule, the value of Q _i (i = 1,2) decreases as the number of discounts increases. In general, in a large-scale problem, as the number of items increases, the number of purchases increases, so that the evaluation value particularly around the initial arrangement becomes excessively small. In this case, the search range in the derivation of the first half of learning is excessively widened, and a sufficient learning effect cannot be obtained.

を用い、

に対する動作の選択確率Pn(x_i,u)を次式で決定する（非特許文献２参照）。

ただし、ξは温度定数であり、

の各値に対して動作選択確率の違いを拡大/縮小することによって、探索範囲を調整する役割を果たす（非特許文献２参照）。 Q _i (i = 1,2) was normalized as an improvement method

Use

The selection probability Pn (x _i , u) of the action for is determined by the following equation (see Non-Patent Document 2).

Where ξ is the temperature constant,

It plays the role of adjusting the search range by enlarging / reducing the difference in the motion selection probability for each value (see Non-Patent Document 2).

[1.2. Decision support]
After performing sufficient learning, the number of purchases for each item is associated with an evaluation value that reflects at least the gross margin and the loss amount derived within the searched area, so that all candidate values are extracted. By comparing the magnitude relations, it is possible to rank the number of purchases.

  Since the evaluation value corresponds to each option for each item, the evaluation value can be sequentially referred to according to the state transition accompanying the purchase number determination. A combination of selecting the highest-ranked option for all items is the best solution, and information on the magnitude of deviation from the best solution can be obtained using the evaluation value of each candidate.

[1.3. System operation]
Preliminary learning is performed to support decision making, and an evaluation function based on Q _i is acquired. Q _i starts with an initial value of 0, and R received when the gross profit is determined propagates with a discount. By repeating preliminary learning, the evaluation value gradually spreads from the state in which the total number of purchases is determined to the initial state, but at this time, propagation from the state having the maximum evaluation value is performed regardless of the number of purchases determined. carry out. In other words, if the evaluation value reflects the loss rate against gross profit, the evaluation of the option that minimizes the loss rate is propagated, and the evaluation function based on Q _i evaluates the best combination of purchases from any state to the function value. Will be taken as.

  When sales or store visit prediction is configured using a predetermined probability model, the evaluation of the number of purchases in each state learns the expected value of the loss rate according to the set probability distribution.

[1.4. Example of decision flow]
FIG. 1 is a diagram schematically illustrating a decision-making flow in a food purchase number determination support system according to the present disclosure. For example, for item A, it is shown that there are options for the number of purchases, such as “Number of purchases 1”, “Number of purchases 2”,... “Number of purchases n”. Further, for example, it is shown that the item A can be assigned to “for raw food”, “for sushi”, and “for lunch”. Furthermore, the evaluation based on the gross profit and the loss rate is shown to be sequentially propagated to each purchase number candidate.

[1.5. Originality and superiority]
The purchase number determination model described in [1.1.1 Purchase number determination work], the learning model of [1.1.2 Learning model], and the discount calculation method of [1.1.3 Calculation of discount rate] are as follows. Each is a new method based on the basic theory necessary for decision support, optimality, gross margin and loss rate, and there is no alternative method for existing technology.

  A large amount of teacher data has not been accumulated for the subject problem. For example, although the order record data of professional staff remains (exists), no records are taken on judgment criteria and decision rules. In general, in order to extract decision-making rules from on-site data, it is necessary to conduct detailed interviews with specialized staff, so it is not possible to complete the analysis necessary for imitation of this disclosure in a short period of time. It is difficult, and it is considered that the speed of extension / extraction of candidates and extension / development of the store visit prediction model by the research according to the present disclosure can exceed that of imitation.

  Research in Japan regarding methods for extracting decision rules from data was actively conducted around 2005 mainly on the operation decision rules related to operations that the reactor operators of chemical plants had. There is a background. In a chemical plant, numerical data related to the operation of the reactor is accumulated, but the intentions and judgment criteria regarding the operation amount and operation timing by the operator were not recorded. As soon as research to extract decision rules is needed. A lot of research has been done in the field, but no effective solution has been shown.

[First Embodiment]
Next, embodiments will be described in detail with reference to the drawings as appropriate. However, more detailed description than necessary may be omitted. For example, detailed descriptions of already well-known matters and repeated descriptions for substantially the same configuration may be omitted. This is to avoid the following description from becoming unnecessarily redundant and to facilitate understanding by those skilled in the art.

  The inventor provides the accompanying drawings and the following description in order for those skilled in the art to fully understand the present disclosure, and is not intended to limit the subject matter described in the claims. Absent.

[2.1. Configuration of support system for determining the number of food purchases]
FIG. 2 is a block diagram of the food purchase number determination support system 1 according to the first embodiment. A food purchase number determination support system 1 shown in FIG. 1 includes a food purchase number determination support device 2 configured by a PDA or the like and connected to an external network 16, and a management server 18 connected to the external network 16. The external network 16 is, for example, a LAN, a WAN, or the Internet.

  The food purchase number determination support device 2 shown in FIG. 1 includes a control unit 4, an input unit 8, a display unit 6, a storage unit 10, and a communication interface unit 12. The control unit 4 performs arithmetic processing and control processing for the entire apparatus. The input unit 8 is a part that generates or receives input data for the apparatus 2 and is generally configured by a keyboard, a mouse, a touch panel, and the like. The input unit 8 may be configured by a touch sensor provided so as to overlap the display surface of the display unit 6. At this time, the input unit 8 accepts a touch (contact) operation on the image displayed by the display unit 6. Thus, the input part 8 and the display part 6 may be comprised integrally, and may be comprised separately.

  The display unit 6 is a part that displays a processing result or the like by the control unit 4 as an image, and includes a liquid crystal display, an organic EL display, a plasma display, or the like. The display unit 6 may include a hearing device such as a speaker as an output unit. The storage unit 10 stores a program that operates in the control unit 4, parameter data necessary for food purchase number determination support, and the like. The storage unit 10 is configured by a flash memory, for example. The communication interface unit 12 is a part that operates as an interface with the external network 16. The control unit 4, the input unit 8, the display unit 6, the storage unit 10, and the communication interface unit 12 are connected to each other by an appropriate bus 14.

  The food purchase number determination support device 2 is configured by an information processing device such as a PDA (Portable Digital Assistant), a tablet terminal, a notebook personal computer, a desktop personal computer, or a workstation.

The control part 4 is comprised by CPU and MPU, for example, and controls operation | movement of the food purchase number determination assistance apparatus 2 whole. Arithmetic processing in the control unit 4 and control processing of the entire apparatus are performed in cooperation with a processor, which is a hardware resource, and a program, which is software, recorded in the storage unit 10 or imported from the outside via the external network 16. Realized by work. The control unit 4 may be realized by a hardware circuit (ASIC, FPGA, etc.) such as an electronic circuit designed exclusively or a reconfigurable electronic circuit. The characteristic operation of the food purchase number determination support device 2 according to the first embodiment is realized by arithmetic processing and control processing in the control unit 4.

  The management server 18 constitutes a computer system that supports and manages the entire business relating to a supermarket or a grocery store that performs food procurement operations. The management server 18 stores data such as the number of stocks, sales forecasts, expiration date, remaining budget, sales, events, number of visits, etc. The food purchase quantity determination support device 2 includes the external network 16 and the communication interface unit 12. Thus, the data is appropriately captured and recorded in the storage unit 10 or used for arithmetic processing in the control unit 4.

  Data such as the number of inventory, sales forecast, expiration date, remaining budget, sales, event, number of visits, etc. are data of state description parameters necessary for the food purchase quantity determination support as described above.

  Note that all or part of the processing in the food purchase quantity determination support device 2 may be performed by the management server 18. Moreover, the computer program which performs the process in the above-mentioned food purchase number determination support apparatus 2 may be stored in the memory | storage part (not shown) of the management server 18, and when needed, the food purchase number determination support apparatus 2 It may be configured to download to its own storage unit 10.

[2.2. Operation of the food purchase number decision support system]
FIG.3 and FIG.4 is a flowchart which shows operation | movement of the food purchase number determination support apparatus 2 of the food purchase number determination support system 1 which concerns on 1st Embodiment. Operation | movement of the food purchase number determination assistance apparatus 2 which concerns on 1st Embodiment is demonstrated using FIG.3 and FIG.4.

  When the work starts (step S02), the number of business days is first determined (step S04). The number of business days is input from the input unit 8, for example. Next, the number of purchased items to be determined is initialized (step S06).

  Next, preliminary learning is executed (step S08). The preliminary learning process is executed for each business day. For example, the inventory number and the sales forecast data are input from the management server 18 to calculate the evaluation of each purchase quantity. Details of the preliminary learning process will be described later with reference to FIG.

After the preliminary learning is performed, it is determined whether there is an item for which the number of purchases has not been determined (step S10). If there is an item for which the number of purchases has not been determined (step S10: YES), the number of purchases for the undecided item and its evaluation are displayed on the display unit 6, and the number of purchases is selected based on this ( Step S12). The evaluation of the number of purchase candidates is calculated by preliminary learning. The selected number of purchases constitutes a purchase table p _i (step S14).

  FIG. 5 is an example of a screen on the display unit 6 of the food purchase quantity determination support device 2 that displays the purchase quantity candidates for the item A and the evaluation thereof. The operator of the food purchase quantity determination support device 2 selects one of the purchase quantity candidates displayed on the screen.

Next, it is determined whether there is an item for which the allocation ratio has not been determined (step S16). If there is an item for which the allocation ratio is undetermined (step S16, YES), the allocation ratio candidates and evaluations for the undetermined item are displayed on the display unit 6, and the allocation ratio is selected based on this ( Step S18). The evaluation of the allocation ratio candidate is calculated by preliminary learning. The selected allocation ratio constitutes an allocation table p _i (step S20). Further, if there is an item for which the allocation ratio has not been determined (step S16 / YES), step S18 and step S20 are executed for that item. If there is no item for which the allocation ratio has not been determined (NO in step S16), the process returns to step S10.

  If there are more items for which the number of purchases is not yet determined in step S10 (step S10, YES), steps S12 to S16 are executed (repeated). If it is determined in step S10 that there is no item whose purchase number has not yet been determined (NO in step S10), a reward R is calculated and accepted, and the stock quantity is updated (step S22).

  Further, it is confirmed whether or not the work is to be ended (step S24). If all the work is completed for the number of business days determined in step S04 (step S24, YES), the entire work is ended (step S26). If not finished (NO in step S24), the process returns to step S06. Now switch to work on the next business day. The number of items to be determined is initialized, and the number of orders for each item returns to 0. However, the total number of stocks and the number of stocks on the expiration date are updated by taking over the sales results of the previous day. Further, steps S08- are executed.

  FIG. 4 is a flowchart showing a detailed operation of the preliminary learning process in the food purchase quantity determination support device 2 according to the first embodiment. The detailed operation of the preliminary learning process will be described with reference to FIG.

  Learning is started (step S0802), and the number of purchased items to be determined is initialized (step S0804).

を格納する（ステップＳ０８１０）。 Next, it is determined whether there is an item for which the number of purchases has not been determined (step S0806). If there is an item for which the number of purchases is undetermined (step S0806 YES), the number of purchases for the undecided item is selected, and based on this, the discount rate γ _i is calculated, and Q1 _i and Q2 _i are It is updated (step S0808). In updating Q1 _i and Q2 _i, the update target is determined depending on whether the latest selection result is the number of purchases or allocation (ratio). Subsequently, the purchase table p _i is constructed,

Is stored (step S0810).

を格納する（ステップＳ０８１６）。更に割当比率が未決定の品目が存在するのであれば（ステップＳ０８１２・ＹＥＳ）、その品目についてステップＳ０８１４、ステップＳ０８１６が実行される（繰り返される）。なお、割当比率が未決定の品目が存在しないのであれば（ステップＳ０８１２・ＮＯ）、ステップＳ０８０６に戻る。 Next, it is determined whether there is an item for which the allocation ratio has not been determined (step S0812). If there is an item for which the allocation ratio is undetermined (step S0812, YES), the allocation ratio for the undetermined item is selected, and based on this, the discount rate γ _i is calculated, and Q2 _i and Q1 _i are It is updated (step S0814). In the update of Q2 _i and Q1 _i, the update target is determined depending on whether the latest selection result is the number of purchases or the allocation (ratio). Subsequently, the purchase table p _i is constructed,

Is stored (step S0816). Further, if there is an item for which the allocation ratio has not been determined (step S0812, YES), step S0814 and step S0816 are executed (repeated) for that item. If there is no item for which the allocation ratio has not been determined (step S0812, NO), the process returns to step S0806.

  If there are more items for which the number of purchases is not yet determined in step S0806 (YES in step S0806), steps S0808 to S0812 are executed (repeated). If it is determined in step S0806 that there is no item for which the number of purchases has not yet been determined (NO in step S0806), a reward R is calculated and accepted, and the stock quantity is updated (step S0818).

  Further, it is confirmed whether or not the work is to be finished (step S0820). If the work is to be finished (step S0820, YES), the whole work is finished (step S0822). If not finished (NO in step S0820), the process returns to step S0804, and the preliminary learning is restarted from the initialization of the number of items to be determined.

[Other Embodiments]
As described above, the first embodiment has been described as an example of the technique disclosed in the present application. However, the technology in the present disclosure is not limited to this, and can also be applied to embodiments in which changes, replacements, additions, omissions, and the like have been made as appropriate.

  In addition, the accompanying drawings and detailed description are provided to explain the embodiments. Accordingly, among the components described in the accompanying drawings and the detailed description, not only the components essential for solving the problem, but also the components not essential for solving the problem in order to illustrate the above technique. May also be included. Therefore, it should not be immediately recognized that these non-essential components are essential as those non-essential components are described in the accompanying drawings and detailed description.

  Moreover, since the above-mentioned embodiment is for demonstrating the technique in this indication, a various change, replacement, addition, abbreviation, etc. can be performed in a claim or its equivalent range.

DESCRIPTION OF SYMBOLS 1 ... Food purchase number determination support system, 2 ... Food purchase number determination support apparatus, 4 ... Control part, 6 ... Display part, 8 ... Input part, 10 ... Memory | storage part, 12 ... Communication interface unit, 14 ... Bus, 16 ... External network, 18 ... Management server.

Claims (1)

In the food purchase quantity decision support system related to the optimal combination learning and derivation method of food purchase quantity,
A learning unit that creates a learning model for an optimal combination of food purchases, and
Based on the learning model, an evaluation value for each purchase number candidate of each item, and a calculation unit that calculates an evaluation value for each allocation ratio candidate when the number of purchases of one item is assigned to a plurality,
The learning model includes at least a purchase candidate for each item, a state description parameter, a discount rate, a first evaluation function, and an allocation ratio candidate when the number of purchases of one item is assigned to a plurality of items, A state description parameter, and a second evaluation function,
The learning unit
For each item, a discount rate is determined based on the loss amount, and for each purchase candidate, a first evaluation function is determined based on the state description parameter and the determined gross profit,
Furthermore, when the number of purchases of one item is allocated to a plurality, a second evaluation function is determined for each allocation ratio candidate based on the state description parameter and the determined gross profit,
The computing unit is
Deriving an evaluation value for each purchase number candidate for each item from the first evaluation function for each purchase number candidate for each item in the learning model,
An evaluation value for each allocation ratio candidate when the number of purchases of one item is allocated to a plurality is derived from a second evaluation function for each allocation ratio candidate in the learning model.
Food purchase number decision support system.

Images