DE102023132434A1 - Procedure for controlling the availability of a product - Google Patents

Abstract

The invention relates to a method for controlling the availability of goods, wherein at least one of the following input variables is determined: availability value of the goods (12), date (14), weather data (16), weekdays (18), holidays, vacation days, bed occupancy data and/or event data; wherein, based on the input variables (12, 14, 16, 18), a probability distribution (10) of the quantity of required goods in a defined availability interval is forecast based on a trained model (22), and an output value (34) is determined based on the probability distribution (10); and wherein the availability of goods is controlled (36) based on the output value (34).

Description

The invention relates to a method for controlling the availability of goods according to patent claim 1. Furthermore, the invention relates to a computer program according to patent claim 10.

The method for controlling product availability described in this application and the associated computer program generally relate to goods with an expiration date or best-before date. The remaining shelf life until the expiration date can be hours, days, or even weeks. In particular, the method relates to perishable goods such as food, fresh produce, and/or cut flowers. The product can be, in particular, a manufactured food, such as a bakery product (rolls, bread, cakes), or generally fresh or perishable goods that are sold, for example, in supermarkets, hardware stores, or drugstores. The method is particularly applicable in the food trade, the food industry, the catering industry, and the out-of-home catering sector.

Especially with the aforementioned goods, managing product availability influences many aspects. For example, if too much product is produced and offered than is currently demanded (oversupply), the surplus product can spoil. This results in many resources, especially material resources, being wasted unnecessarily. On the other hand, too little product is also detrimental. Both oversupply and undersupply lead to underutilization of production and storage capacity, and staff utilization is suboptimal.

The following example primarily uses baked goods as a commodity to explain the process. For example, if too many rolls are produced on one day or at a certain time and they are not sold, the resulting rolls cannot be sold the next day, or only poorly. Surplus rolls are usually discarded the following day, resulting in the loss of valuable raw materials produced with effort and energy. This example can also be applied to other goods, such as dairy products, fruit and vegetables, ready-to-eat meals, cut flowers, meat products, etc.

The underlying object of the present invention is therefore to provide a method for controlling the availability of goods and a computer program by means of which the use of energy and resources, such as personnel, time, money and in particular raw materials for the goods, can be reduced.

The problem is solved with the features of the independent patent claims; further practical embodiments and advantages are described in connection with the dependent claims.

The invention relates to a method for controlling the availability of a product. As already described above, the product can, in particular, be a product with an expiration date, especially a perishable product, which should therefore be available in the right quantity at the right time. The product can, in particular, be food, fresh produce, and/or cut flowers.

• - Verfügbarkeitswert der Ware
• - Datum, Wetterdaten, Wochentage, Feiertage, Ferientage, Bettenbelegungsdaten und/oder Veranstaltungsdaten.

According to the procedure, at least one of the following input variables is first determined:

• - Availability value of the goods
• - Date, weather data, weekdays, holidays, vacation days, bed occupancy data and/or event data.

The availability value can - as described below - be read from, among other things, an inventory management system and/or cash register system and/or determined based on recorded measured values. For example, the inventory levels still in stock are recorded, e.g., from the difference between the goods produced the previous day and the goods sold. The availability value can also be measured. Such a measured value can, in particular, be a measurement from a sensor, e.g., a scale, an optical sensor, a fill level sensor, and/or a pressure sensor, from which the currently available supply of the goods can be read.

In addition to the data relating purely to the goods, other input variables can be determined, including one or more dates. Based on the date, for example, the season can be deduced. As described below, a correlation to a past date can also be established.

Weather data can be used as an additional input variable. In particular, current weather data or weather data in a defined future time interval are used. Weather data up to 14 days in advance can be used. In particular, weather data is used on an hourly basis.

Alternatively or additionally, weekdays, public holidays, vacation days, bed occupancy data, and/or event data serve as input variables. These data, among other things, also affect the demand for certain goods and, accordingly, influence the corresponding expected supply of the goods.

Returning to the example of bakery products, it is conceivable that more rolls should be available on Saturdays, Sundays and public holidays than on weekdays.

Using one or more of the previously mentioned input variables, a probability distribution of the quantity of required goods in a defined availability interval is then predicted based on a trained model, and an initial value is determined based on the probability distribution. The probability distribution indicates the quantity of the specific goods that will likely be purchased in the certain, previously defined availability interval. The one or more input variables are analyzed, in particular using AI, to determine the probability distribution. The availability interval can be freely selected for the respective application and can, for example, be one or more days or just one or more hours. For example, the availability interval can cover the period of a sales shift. The selected initial value is, in particular, the value, i.e., the quantity of goods, with the greatest probability.

In the bakery product example, a baseline value can be predicted based on the availability value (what is in stock), the next day's date, and the forecast weather for the next day. In this example, the baseline value is the quantity of rolls most likely to be sold the next day.

After determining the baseline, product availability is controlled based on the baseline. In particular, an ordering process is controlled based on the baseline. For example, a bakery branch orders the quantity of rolls from the wholesale bakery that are likely to be sold based on the baseline.

In particular, a production process is controlled based on the initial value. If possible, only as many goods are produced or completed as are likely to be needed. If the availability interval is chosen relatively short, for example, covering only a few hours, the exact quantity of goods expected to be needed for this number of hours can be produced.

In the example of rolls, for example, a baking process can be started throughout the day whenever a larger quantity of rolls is expected to be in demand. The procedure then specifically includes starting such a baking process.

Overall, the method according to the invention makes it possible to predict the availability of goods in such a way that as little surplus as possible is produced or ordered, and as little as possible is offered. The input data used and the trained model enable an accurate data-based prediction. This can save energy and resources.

In particular, the individual input variables can be downloaded automatically from one or more sources.

In particular, the output value is transmitted to an interface, which serves as a link to a device for controlling product availability. In particular, the interface communicates with a merchandise management system, cash register system, and/or a production facility. In particular, it is intended that the output value be retrieved from the interface.

In particular, the probability distribution is predicted using a learning algorithm that uses historical availability values as training data. Historical availability values are values from the past. Historical availability values can be determined from a specific point in time in the past, starting from a specific day or time. In particular, historical availability values are determined on an hourly or daily basis, i.e., there is a value per hour or per day. However, other granularities are also conceivable.

In particular, the historical availability values are ordered goods, sold goods and/or the difference between ordered and sold goods. The historical availability values are determined in particular from an inventory management system and/or cash register system, which contains order and/or sales figures. Alternatively or additionally, measured values are used, e.g. values from a scale, an optical sensor, a camera, a pressure sensor and/or a level sensor, from from which the goods available and withdrawn or the difference between them can be derived.

In particular, the historical availability values for a specific time interval and, in particular, corresponding data regarding dates, weather, days of the week, holidays, special days, events, and/or bed occupancy data are used. From this, an AI-based model is developed, which forms the basis for the current forecast of the probability distribution.

To determine the most accurate initial value possible, when forecasting the probability distribution, availability values from a shorter time period within a defined time interval are given a higher weighting than availability values from before this defined time interval. Such an autoregressive model can thus quickly adapt to short-term changes in sales and is particularly accurate.

In a practical embodiment of the method, the mean µ of the probability distribution is used as the initial value. For example, only as many rolls are ordered as the mean of the probability distribution. The output of such an initial value is designated "None," so that no surplus is most likely produced.

It is also conceivable to choose a surplus value as the output value instead of the mean µ. The surplus value is greater than the mean µ. The surplus value can, in particular, be µ + σ. The output of such an initial value is referred to as the "target." In this case, a quantity of goods is made available; in the case of µ + σ, a surplus of goods remains with a defined probability. The surplus value can be freely chosen and can also be µ + 1/2σ or µ + 2σ, depending on how much surplus is to be accepted.

If the output value is higher than the mean, it could be bread rolls, for example. In this case, a surplus can be accepted, as rolls can be reused, e.g., for breadcrumbs, dough, or animal feed.

The initial value, and in particular whether it is the average value or a surplus value, is determined based on a characteristic of the goods. In particular, this characteristic is the perishability of the goods or the possibility of reusing the goods for other purposes.

In particular, the process is carried out at specific times. These times are specifically tailored to the production and sales processes of the respective goods. In particular, the process can be carried out daily, at intervals of several days, or even several times a day.

In a practical embodiment of the method, the model on which the calculation of the probability distribution is based is recalculated if an actual deviation between the most recently determined output values of a first time interval I and the previous output values of a second time interval II, where time interval II lies further in the past than time interval I, exceeds a target deviation. The values in time interval I are then used in particular to retrain the model. This is intended to react to possible sudden changes in demand for a product and to adapt the forecast to new trends as quickly as possible.

In a further practical embodiment, the availability value and/or the historical availability values are measured values recorded by a sensor. In particular, the availability value and/or the historical availability values can be recorded by a scale, an optical sensor, a camera, a pressure sensor and/or a fill level sensor. For example, a scale can be used to determine the amount of goods still available in a display. Alternatively or additionally, a camera and subsequent image analysis can be used to determine how many or what volume of goods are currently still available. The same applies to the historical availability values; these can also be measured values previously recorded by sensors.

As already described above, in step c), the availability of goods can be controlled by starting a production process. By determining the output value for a specific point in time, the production process can be started at that specific point in time. This ensures that fresh goods are produced as close as possible to their sale. In particular, it can also be used to optimize production processes, for example, by scheduling production processes to maximize synergy effects. For example, several baking processes can be started one after the other, each utilizing the heat from the previous processes.

One production step can be the better timed baking of baked goods In particular, this allows a baking process to be started precisely, for example, when goods are likely to be needed. In step c), an oven is then controlled so that it heats up and the desired baking program is set.

In particular, a message can be sent to a person to start the production of additional goods, possibly requiring human intervention or action. For example, a person might be asked to place rolls into the preheated oven or generally load a production line. Alternatively, loading the oven or a production line can also be automated using handling robots.

The invention also relates to a computer program containing machine-readable instructions which, when executed on one or more computers, cause the computer or computers to carry out a method as described above.

• 1 ein Blockschaltbild für ein Verfahren zur Steuerung einer Warenverfügbarkeit,
• 2 ein erstes Beispiel für eine Wahrscheinlichkeitsverteilung und
• 3 ein zweites Beispiel für eine Wahrscheinlichkeitsverteilung.

Further practical embodiments and advantages are described below in conjunction with the figures. They show:

• 1 a block diagram of a method for controlling the availability of goods,
• 2 a first example of a probability distribution and
• 3 a second example of a probability distribution.

In 1 A block diagram is shown, which is intended to visualize a method for controlling the availability of goods. To determine a probability distribution 10 of goods required in the future, several input variables 12, 14, 16, 18 are fed to a calculation unit 20 on the left side.

The input variables 12, 14, 16, and 18 shown here are an availability value 12, the date 14, weather data 16, and weekdays 18. It is conceivable that other or even further input variables, such as holidays or bed occupancy data, could be fed to the calculation unit 20. In particular, the availability values 12 can be measured values recorded by a scale, an optical sensor, a pressure sensor, and/or a fill level sensor.

The input variables 12, 14, 16, 18, 20 are fed to the calculation unit 20. The calculation unit 20 is also based on a model 22. The model 22 was trained with historical availability data 24 and corresponding data such as data 26, weather data 28, and weekdays 30.

The calculation unit 20 then predicts a probability distribution 10 from the input variables 12, 14, 16, 18 on the basis of the model 22 (see also 2 ). The probability distribution 10 indicates the probability P with which a certain quantity N of a product will be purchased for the forecasted availability interval.

It may also be provided that a target for a surplus value of 32 is included.

Based on the probability distribution 10, an initial value N (symbolized by box 34) is determined, which is then used to control the availability of goods (symbolized by box 36).

In 2 and 3 Two examples of determined probability distributions 10 are shown. These are, for example, probability distributions 10 that were forecast for the same product but for different locations.

When looking at the 2nd 3 it can be seen that the probability distribution 10 in 2 is narrower than the probability distribution 10, which in 3 is shown.

If a “no value” is chosen as the initial value, which corresponds to the mean value µ, this would correspond in the example shown to a number of 10 pieces for both probability distributions 10. With the “no value” it is likely that no surplus is to be produced.

However, if a certain surplus can be tolerated, a surplus value can also be chosen as the starting value. For example, a "can-value" (µ + σ) can be selected as the starting value. As can be seen, this would result in a smaller number of goods in the first distribution than in the broader distribution. The same applies to the other surplus values, the "target" and "super-target" (µ + 2σ).

List of reference symbols

10

Probability distribution

12

Input variable (availability value)

14

Input variable (date)

16

Input variable (weather data)

18

Input variable (weekdays)

20

Calculation unit

22

Model

24

Training data (historical availability values)

26

Training data (date)

28

Training data (weather data)

30

Training data (weekdays)

32

Default surplus value

34

Determination of initial value

36

Control availability

Claims (10)

Method for controlling the availability of goods, a) wherein at least one of the following input variables is determined: - availability value of the goods (12) - date (14), weather data (16), days of the week (18), holidays, vacation days, bed occupancy data, and/or event data; b) wherein, based on the input variables (12, 14, 16, 18), a probability distribution (10) of the quantity of required goods in a defined availability interval is predicted based on a trained model (22), and an initial value (34) is determined based on the probability distribution (10); c) and wherein the availability of goods is controlled (36) based on the initial value (34). Method according to the preceding claim, characterized in that the forecast of the probability distribution (10) is carried out using a self-learning algorithm which uses historical availability values (24) as training data. Method according to the preceding claim, characterized in that historical availability values from a shorter period of time in a defined time interval are given a higher weighting than historical availability values before this defined time interval. Method according to one of the preceding claims, characterized in that the mean value (µ) of the probability distribution (10) or a defined excess value (32) is selected as the starting value (34). Method according to one of the preceding claims, characterized in that the initial value (34) is selected on the basis of a property of the goods. Method according to one of the preceding claims, characterized in that the method is carried out at specified times. Method according to one of the preceding claims, characterized in that the model (22) on which the calculation of the probability distribution (10) is based is redetermined if an actual deviation between the last determined output values of a first time interval I and the previously determined output values of a second time interval II exceeds a target deviation. Method according to one of the preceding claims, characterized in that the availability value (12) and/or the historical availability values (24) are a measured value which was recorded by a sensor. Method according to one of the preceding claims, characterized in that in step c) the availability of goods is controlled by starting a production process. A computer program containing machine-readable instructions which, when executed on one or more computers, cause the computer(s) to perform a method according to any one of the Claims 1 until 9 to execute.