TWI879256B - Processing system and method for order forecasting - Google Patents

Abstract

A processing system and method for order forecasting is provided. The processing system comprises a server and a client. The server comprises a first communication unit, a first storage unit, and a first processing unit. The client comprises a second communication unit, a second storage unit, and a second processing unit. The first storage unit stores a selling parameter and a selling history information. The processing unit performs the multiple regression model and selecting a selected parameter form the selling parameter. The processing unit generates a forecasted quantity according to the selected parameter. The server sends the forecasted quantity to the client. The second storage unit stores a on-shelf quantity information and an inventory quantity information. The second processing unit calculates a forecasted inventory result according to the forecasted quantity, the on-shelf quantity information and the inventory quantity information.

Description

Order estimation processing system and method

The invention relates to a system and method for processing the quantity of ordered goods, and more particularly to a system and method for processing order estimation applied to store orders.

For convenience stores with different customer groups, the product demands of each convenience store will also be different. Because different customer groups have different product demands, supermarket owners can only make purchase decisions based on past sales history in addition to the recommendations of the head office. In addition to considering the types of goods to be purchased, supermarket owners also need to consider the quantity of goods to be purchased. This is a severe challenge for inexperienced store owners.

If the quantity of goods purchased is too large, different problems will arise for goods with different shelf life. For long-life goods, the store will have a situation of hoarding. For short-life goods, the store may have a large number of scrapped goods.

In view of this, in one embodiment, the order estimation processing system includes a server and an ordering end. The server has a first communication unit, a first storage unit and a first processing unit. The first processing unit is connected to the first communication unit and the first storage unit. The first storage unit stores a plurality of sales parameters and sales history information. Each sales parameter matches a corresponding commodity category. The first processing unit performs a multivariate regression model on the sales parameters according to the sales history information and the commodity category to select at least one sales parameter. The selected sales parameter is the selected parameter. The multivariate regression model generates a quantity estimation value of the commodity category according to the selected parameter. The first processing unit generates a quantity estimation value of the commodity category according to the selected parameter. The unit sends a cargo volume estimation through a first communication unit; the ordering end has a second communication unit, a second storage unit and a second processing unit, the second processing unit is connected to the second communication unit and the second storage unit, the second communication unit is network-connected to the first communication unit, the second communication unit receives the cargo volume estimation, the second storage unit stores the virtual inventory quantity information and the inventory quantity information corresponding to at least one commodity category, and the second processing unit calculates the purchase estimation result of the commodity category according to the cargo volume estimation, the virtual inventory quantity information and the inventory quantity information.

The order estimation processing system can provide corresponding quantity estimation values to the ordering end according to different commodity categories. The ordering end only needs to adjust the purchase estimation result according to the quantity of commodities on the shelf and the inventory quantity.

In one embodiment, the order estimation processing method includes a server side performing a multivariate regression model on a plurality of sales parameters to select at least one sales parameter, wherein the selected sales parameter is a selected parameter; the server side generates a quantity estimation value according to the selected parameter; the server side sends the quantity estimation value to the ordering side; the ordering side generates a purchase estimation result according to the quantity estimation value, the simulated inventory quantity information, and the inventory quantity information.

The order estimation processing system and processing method can be applied to the order quantity estimation suggestions for each commodity category. The order estimation processing system adopts a multivariate regression model to select the corresponding parameters of each store from a variety of sales parameters. Therefore, the processing system can provide the corresponding quantity estimation value of each store. The ordering end only needs to adjust the virtual inventory quantity information and the inventory quantity information to obtain the purchase estimation result of the commodity category. In this way, the order estimation processing system can reduce the evaluation cost of the store for the commodity category and its quantity when placing an order.

Please refer to FIG. 1 and FIG. 2, which are respectively a schematic diagram of the order estimation processing system architecture and a schematic diagram of the system components of an embodiment. The order estimation processing system (hereinafter referred to as the processing system 10) includes a server 100 and at least one order terminal 200. The order terminal 200 corresponds to the computing equipment of each store, such as a POS (Point of Sale), a personal computer, a notebook or a tablet computer. Each order terminal 200 is connected to the server 100 via a network. The server 100 can be a cloud server or an independent server installed within the enterprise.

The server 100 has a first communication unit 110, a first storage unit 120 and a first processing unit 130. The first processing unit 130 is connected to the first communication unit 110 and the first storage unit 120. The first storage unit 120 stores a plurality of commodity categories 320, each of which has a plurality of sales parameters 121, sales history information 122 and a multivariate regression model 123. The sales parameter 121 includes but is not limited to month, week, season, weather information, period sales, holiday sales, weekday sales, weekday holiday sales difference, store location, order frequency or a combination thereof. Each sales parameter 121 has a plurality of parameter data. The sales history information 122 is the sales volume information of each product category 320 of the store. If the ordering end 200 has multiple order records, each sales parameter 121 includes multiple parameter data. In other words, the server 100 will update the parameter data after each order. The parameter data can be the sales data of the product, or it can be a hyperparameter or node weight learned by machine. If the ordering end 200 places an order for the first time, the parameter data of the sales parameter 121 is an empty set.

Each sales parameter 121 matches a corresponding product category 320. In other words, different product categories 320 have their own sales parameters 121. The types of product categories 320 can be any of pasta, department stores, room temperature beverages, food and seasonings, beauty and health care, candy, bodybuilding functions, paper, sanitary products, beer, ice cream, frozen food, snacks, spirits, convenience products, international boutiques, cigarettes, customer service, toys, commission collection and sales, proprietary communication products, electronic payment recharge, etc. The first processing unit 130 performs a multivariate regression model 123 calculation on each product category 320 to generate selected parameters and cargo volume estimation 410, and its operation is described in detail later.

The ordering terminal 200 has a second communication unit 210, a second storage unit 220 and a second processing unit 230. The second processing unit 230 is connected to the second communication unit 210 and the second storage unit 220. The second communication unit 210 can be connected to the first communication unit 110 via an Ethernet or a telecommunications network. The second communication unit 210 is used to transmit sales history information 122 or receive a quantity estimation 410. The second storage unit 220 stores a front-end ordering program 221, a simulated inventory quantity information 222 and an in-stock quantity information 223.

The simulated inventory quantity information 222 is the quantity of goods that the specified commodity category 320 is expected to have in the store. In some embodiments, the simulated inventory quantity information 222 is the sum of the quantity of goods that can be displayed on the shelf and the safety inventory. The quantity of goods that can be displayed on the shelf is the quantity of goods that can be displayed on the store shelf 310 of the commodity category 320. The height, width and depth of the store shelf 310 will affect the quantity that can be displayed of the commodity category 320, and the number of store shelves 310 in each store is different, and the actual total quantity that can be displayed will also be different. In addition, the store shelves 310 will also be adjusted according to decoration or other factors, so that the quantity of goods that can be displayed on the shelves during each activity period will change accordingly. Furthermore, the quantity of goods that can be displayed on the shelf can be adjusted according to the display method or the convenience of putting on the shelf and taking out. Therefore, each order terminal 200 needs to record the quantity of goods that can be displayed on the shelf. The safety stock quantity is a multiple value of the sales quantity of a single store per day (per store per day), where the multiple value is greater than or equal to "1".

The inventory quantity information 223 is the sum of the actual inventory quantity of the product category 320 in the store and the in-transit quantity. The in-transit quantity is the quantity of products that have been ordered by the ordering end 200 but have not yet been put into the warehouse.

To clearly illustrate the estimation of order quantity by the processing system 10, please refer to FIG. 3. The processing method of order estimation includes the following steps: Step S310: The server end 100 performs a multivariate regression model calculation on the multiple sales parameters 121 of each product category 320 to select at least one sales parameter 121, and the selected at least one sales parameter 121 is a selected parameter; Step S320: The server end 100 calculates and generates a quantity estimation value 410 of each product category 320 according to the selected parameter; Step S330: The server end 100 sends the quantity estimation value 410 to the order end 200; and Step S340: The order end 200 calculates the purchase estimation result according to the quantity estimation value 410, the simulated inventory quantity information and the inventory quantity information.

The server 100 receives the sales history information 122 from each ordering terminal 200. The ordering terminal 200 can periodically send the sales history information 122 corresponding to the product category 320 in a specified time period to the server 100, or send the sales history information 122 to the server 100 after the sales history information 122 of the ordering terminal 200 accumulates to a specified amount. The server 100 appends the new sales history information 122 to the existing sales history information 122.

After receiving the sales history information 122 from the ordering end 200, the server end 100 can execute the multivariate regression model 123 calculation on all the sales parameters 121 for the plurality of product categories 320 according to the sales history information 122. Alternatively, the server end 100 executes the multivariate regression model 123 calculation at a specified time, for example, the server executes the multivariate regression model 123 calculation at 00:00 AM every Monday. The first processing unit 130 executes the multivariate regression model 123 calculation on the sales parameters 121 (corresponding to step S310). The multivariate regression model 123 is any one of stepwise regression, multiple linear regression, polynomial regression, principal component regression (PCR), or time series regression.

The multivariate regression model 123 selects a sales parameter 121 in an iterative manner. After the sales parameter 121 is selected, the relevant calculation of the determination coefficient is performed to determine whether the sales parameter 121 is a qualified parameter. If it is a qualified sales parameter 121, it is called a selected parameter. Since not all the selected sales parameters 121 are selected parameters, the selected sales parameters 121 are called temporary parameters (no labels). For example, the set of sales parameters 121 is (X ₁ ,X ₂ ,X ₃ ...,X _n ), and the selected parameter is (X` _i ), Y is the prediction target, R _i ² is the coefficient of determination, where i is the number of sales parameters 121 and i∈(1,2,3...,n).

In the initial stage of the multivariate regression model 123, the first processing unit 130 determines at least one prediction target based on the default cycle and the corresponding sales history information 122. In addition, the multivariate regression model 123 can also select a default value as a prediction target. Generally speaking, the default cycle is a set of order cycles of the store outlets (i.e., the ordering end 200). For example: if the default cycle is one year and the order cycle is one week, the default cycle includes a set of 52 weeks in that year. Or during a special sale, the default cycle can be defined by the ordering end 200 or the server end 100.

The first processing unit 130 calculates the judgment coefficient of each temporary parameter for the prediction target until the first processing unit 130 obtains the judgment coefficients of all temporary parameters. For the judgment coefficient, the value of the judgment coefficient is between [0,1]. The closer the value of the judgment coefficient is to "1", the better the fitting result of the multivariate regression model 123. Then, the first processing unit 130 selects the temporary parameter with the maximum judgment coefficient and temporarily stores this temporary parameter, which is referred to as the first parameter (that is, the selected parameter). The multivariate regression model 123 completes this round of calculation, and for the convenience of explanation, the prediction target Y of this round of calculation and the corresponding selected parameter are recorded as "Y-> _X`a ", a∈(1,2,3...,n)).

In some embodiments, the first processing unit 130 may use a stepwise regression analysis method. In the training phase of the stepwise regression analysis method, the most suitable factor set and regression coefficient are iteratively refitted for each stepwise regression analysis model. The first processing unit 130 first considers the explanatory power of a single factor for the dependent variable, and gradually adds other factors.

The first processing unit 130 selects any one of the multiple sales parameters 121 and calculates the judgment coefficient of the selected sales parameter 121. The first processing unit 130 saves the sales parameter 121 with the largest judgment coefficient. Then, the first processing unit 130 selects a second sales parameter 121 from the remaining sales parameters 121. The first processing unit 130 selects the sales parameter 121 that can maximize the judgment coefficient from the remaining sales parameters 121. This selection process forms a loop and uses an information criterion such as Akaike Information Criterion, Bayesian information criterion, or Mallows's CP as a termination condition. When the fit is no longer significantly improved or the predetermined threshold condition is not met, the loop ends and the final set of factors in the model is determined to obtain the best fit result.

Taking the Akagi information criterion as an example, the smaller the calculation result of the Akagi information criterion, the better the fitting result of the multivariate regression model 123. The threshold condition can be a specified value or the difference between the fitting results of different calculation rounds. By arranging the difference between the fitting results of two adjacent rounds, it can be judged that the difference between the fitting results gradually decreases, which means that the fitting results begin to converge.

The first processing unit 130 selects any one of the remaining sales parameters 121, and records the corresponding relationship between the newly selected sales parameter 121 (i.e., the newly selected temporary parameter) and the forecast target Y as "Y->_X'a + _X'b ". The first processing unit 130 calculates the corresponding fitting result for the forecast target based on the newly selected temporary parameter and the first parameter. The first processing unit 130 repeatedly selects sales parameters 121 until all sales parameters 121 are traversed and the determination coefficient corresponding to each sales parameter 121 and the first parameter is obtained.

Next, the first processing unit 130 selects the one with the largest judgment coefficient from all judgment coefficients in this calculation round as the fitting result. The two temporary parameters selected in this calculation round are called second parameters (i.e., selected parameters). The first processing unit 130 then uses the information criterion to determine whether the fitting result of the judgment coefficient converges to the threshold condition.

If the convergence to the threshold condition is not met, the first processing unit 130 repeatedly executes the selection of sales parameters 121 and the determination of the fitting result until the fitting result converges to the threshold condition. The first processing unit 130 selects a set of selected parameters that meet the threshold condition. The selected selected parameters and the prediction target Y can be expressed as follows: Y = a ₁ *X` <sub>a</sub> + a <sub>2</sub> *X` _b + ..... + a _m *X` _m , m ≤ n, a is the regression coefficient Formula 1

Formula 1 is a cargo volume estimation F formula output by the selected parameters through the multivariate regression model 123. The first processing unit 130 calculates the cargo volume estimation value 410 according to the selected parameters and the cargo volume estimation F formula (corresponding to step S320).

For example, the selected parameters are [month, week, season, temperature], and the calculation results of the judgment coefficient and Akagi information criteria are as follows: Selected parameters Determination coefficient (R ² ) Akagi Information Guidelines month 0.6 100 Month, Week 0.7 90 Month, Week, Season 0.8 80 Month, Week, Season, Temperature 0.85 85 Table 1. Calculation results of information criteria and judgment coefficients Sales volume = a ₁ * month + a ₂ * week + a ₃ * season + a 4 * temperature Formula 2

As can be seen from Table 1, after the selected parameters are added with "week", the degree of fit of the multivariate regression model 123 begins to improve (i.e., the aforementioned convergence). "Month" is the first parameter, and "month, week" is the second parameter. For other calculation rounds, the third parameter or more can be inferred.

After the selected parameters "week" and "season" are added, the degree of fit of the multivariate regression model 123 is also improved. Subsequently, when "temperature" is introduced into the multivariate regression model 123, although the judgment coefficient increases, its increase is smaller than the increase of the first two parameters. This shows that the explanatory power of "temperature" for changes in sales volume is relatively weak. Nevertheless, there is a certain correlation between "temperature" and sales volume. Here, the selected parameter "temperature" is also regarded as meeting the threshold condition. The first processing unit 130 obtains the cargo volume estimation F formula (i.e., Formula 2) based on the aforementioned selected parameters and the forecast target.

Then, the first processing unit 130 can substitute the sales parameter 121 and related information to be predicted into Formula 2. For the convenience of explanation, it is assumed that the regression coefficients are [a1=0.4, a2=3, a3=2, a4=0.1], and the prediction target of "Monday, the third week of August 2023, temperature: 30 degrees" is predicted, but the actual regression coefficients need to be obtained by minimizing the Ordinary Least Squares (OLS) of all sales parameters 121. The following formula is the sum of the contributions of each selected parameter to obtain the predicted target Y: Prediction target = 0.4*8+3*1+2*3+0.1*30 = 15.2≒16                Formula 3

The first processing unit 130 substitutes the selected parameters into Formula 2, where "month" is "8", "week" is "1", and "quarter" is "3", thereby obtaining the calculation result of Formula 3. This calculation result is the estimated quantity 410. The first processing unit 130 sends the estimated quantity 410 to the ordering terminal 200 (corresponding to step S330).

Compared to the implementation mode in which the selected parameters are directly introduced, in some embodiments, before the first processing unit 130 performs the processing of the prediction target, the first processing unit 130 can perform one-hot encoding processing (One-Hot Encoding) on the selected parameters belonging to the discrete variable type, so that the selected parameters belonging to the discrete variable type are assigned corresponding one-hot codes and regression coefficients.

Continuing with the above example, if the selected parameter "week" is a discrete variable, the first processing unit 130 can perform one-hot encoding on the selected parameter "week". The selected parameter "week" can be represented by 7 bits, for example, the one-hot encoding of the selected parameter "week" is (0,0,0,0,0,0,0). After the first processing unit 130 obtains the selected parameter "week" and the predicted sales parameter 121, the first processing unit 130 can select the one-hot encoding corresponding to the selected parameter "week" according to the predicted sales parameter 121, please refer to the following table. Selected parameters One-hot encoding Regression coefficient Monday (1,0,0,0,0,0,0) 0.5 Tuesday (0,1,0,0,0,0,0) 0.6 Wednesday (0,0,1,0,0,0,0) 0.7 Thursday (0,0,0,1,0,0,0) 0.8 Friday (0,0,0,0,1,0,0) 0.9 Saturday (0,0,0,0,0,1,0) 0.8 Sunday (0,0,0,0,0,0,1) 0.7 Table 2. One-hot encoding correspondence of selected parameters

When the first processing unit 130 selects the predicted sales parameter 121 of "Monday", the first processing unit 130 selects the one-hot code (1,0,0,0,0,0,0) and the regression coefficient. Then, the first processing unit 130 multiplies the selected one-hot code and the regression coefficient to obtain the contribution of the selected parameter. Still taking "Monday" as an example, the first processing unit 130 selects the one-hot code (1,0,0,0,0,0,0) and the regression coefficient "0.5", and multiplies them by 1*0.5 to obtain the contribution of the selected parameter "0.5". The first processing unit 130 adds the contribution of the selected parameter (corresponding to the “week” in this example) with the contributions of other selected parameters to obtain the predicted targets of all selected parameters.

In addition, the first processing unit 130 can add an intercept term to the sum of the contributions of each selected parameter to adjust the unexplained portion generated by each selected parameter in the multivariate regression model 123. For example, if the intercept term is "3.5", the first processing unit 130 will add the intercept term to Formula 1, as shown in Formula 4 below: Prediction target = a ₁ *X` <sub>a</sub> +a <sub>2</sub> *X` _b +.....+ _am *X` _m + 3.5 Formula 4

After the ordering terminal 200 receives the estimated quantity 410, the second processing unit 230 executes the front-end ordering program 221. The front-end ordering program 221 calculates the purchase estimation result of the commodity category 320 according to the estimated quantity 410, the virtual inventory quantity information 222, and the inventory quantity information 223, please refer to FIG. 4. In some embodiments, the second processing unit 230 adds the estimated quantity 410 to the virtual inventory quantity information 222, and then subtracts the inventory quantity information 223 to obtain the purchase estimation result I (corresponding to step S340). After the ordering terminal 200 obtains the purchase estimation result I, the ordering terminal 200 can submit the purchase estimation result I to the server 100 or the related backend. In the screen of the front-end ordering program 221 in FIG. 4, "Stock quantity estimation" is the stock quantity estimation value 410, and its value is "F" as an example, "Pseudo inventory quantity" is the pseudo inventory quantity information 222, and its value is "M" as an example, and "In-stock quantity" is the in-stock quantity information 223, and its value is "O" as an example. Therefore, the calculation result of the purchase estimation result I is "F+M-O=I".

In some embodiments, for a newly established storefront in the server 100, before obtaining the sales parameter 121 of the first ordering end 200, the first storage unit 120 has not recorded any sales parameter 121 of the ordering end 200. Therefore, when the first processing unit 130 executes the multivariate regression model 123, it will determine whether the sales parameter 121 exists in the first storage unit 120. When the sales parameter 121 is an empty set, the first processing unit 130 directly regards the preset quantity as the purchase estimation result I, and omits the calculation of steps S320 to S340. The preset quantity can be provided by the ordering end 200, or it can be determined based on the sales parameter 121 of other ordering ends 200.

In some embodiments, after a preset period, the ordering end 200 sends an update request to the server 100 to obtain the estimated quantity 410 of the preset period from the server 100. For example, the ordering end 200 may send an update request to the server 100 every Monday to obtain the estimated quantity 410 of the next week.

The order estimation processing system 10 and processing method can be applied to the recommendation of purchase quantity estimation for each commodity category 320. The order estimation processing system 10 adopts a multivariate regression model 123 to select the best corresponding parameters for each store from a variety of sales parameters 121, rather than using the same parameters uniformly. In this way, it is possible to avoid ignoring the best sales parameters 121 that may be applicable to each store, or adding sales parameters 121 that are not applicable to the store, thereby affecting the final calculation result. Therefore, the processing system 10 can provide each store with a corresponding quantity estimation value 410 according to different store. The order end 200 only needs to adjust the virtual inventory quantity information 222 and the inventory quantity information 223 to obtain the purchase estimation result I of the commodity category 320. In this way, the order estimation processing system 10 can reduce the evaluation cost of the merchandise category 320 and its quantity when the store places an order.

10: Processing system 100: Server 110: First communication unit 120: First storage unit 121: Sales parameters 122: Sales history information 123: Multivariate regression model 130: First processing unit 200: Ordering end 210: Second communication unit 220: Second storage unit 221: Front-end ordering program 222: Pseudo inventory quantity information 223: In-stock quantity information 230: Second processing unit 310: Store shelves 320: Product category 410: Quantity estimation F: Quantity estimation M: Pseudo inventory quantity information O: In-stock quantity I: Purchase estimation result S310, S320, S330, S340: Steps

Figure 1 is a schematic diagram of the processing system architecture of an order estimation in an embodiment. Figure 2 is a schematic diagram of the system components in an embodiment. Figure 3 is a schematic diagram of the processing flow of order estimation in an embodiment. Figure 4 is a schematic diagram of the interface of the front-end ordering program in an embodiment.

10: Processing system

100: Server side

110: First communication unit

120: First storage unit

121: Sales parameters

122: Sales history information

123:Multivariate regression model

130: First processing unit

200: Order side

210: Second communication unit

220: Second storage unit

221: Front-end ordering program

222: Specified inventory quantity information

223: Inventory quantity information

230: Second processing unit

410: Valuation based on volume

Claims (13)

An order estimation processing system, comprising: A server end has a first communication unit, a first storage unit and a first processing unit. The first processing unit is connected to the first communication unit and the first storage unit. The first storage unit stores a plurality of commodity categories. The commodity categories respectively have a plurality of sales parameters and a sales history information. The first processing unit performs a multivariate regression model calculation on the sales parameters of the commodity categories according to the sales history information to select at least one of the sales parameters. The selected at least one of the sales parameters is a selected parameter. The first processing unit calculates a cargo volume estimation value for the commodity categories according to the selected parameter using the multivariate regression model. The first processing unit sends the cargo volume estimation value through the first communication unit; and An ordering end has a second communication unit, a second storage unit and a second processing unit. The second processing unit is connected to the second communication unit and the second storage unit. The second communication unit is network-connected to the first communication unit. The second communication unit receives the estimated quantity value. The second storage unit stores a simulated inventory quantity information and an in-stock quantity information corresponding to the commodity categories. The second processing unit calculates a purchase estimation result for the commodity categories according to the estimated quantity value, the simulated inventory quantity information and the in-stock quantity information: Among them, the second processing unit subtracts the in-stock quantity information from the sum of the estimated quantity value and the simulated inventory quantity information to obtain the purchase estimation result. The order estimation processing system as described in claim 1, wherein the virtual inventory quantity information is the sum of a shelf displayable quantity and a safety stock quantity. The order estimation processing system as described in claim 1, wherein the multivariate regression model is a stepwise regression process, multiple linear regression, polynomial regression, principal component regression or time series regression. The order estimation processing system as described in claim 1, wherein the sales parameters include month, week, season, weather information, periodic sales, holiday sales, weekday sales, weekday-holiday sales difference, store location, order frequency or a combination thereof. In the order estimation processing system as described in claim 1, if the sales parameter is an empty set, the first processing unit obtains a default quantity as the purchase estimation result. A processing system for order estimation as described in claim 1, wherein the multivariate regression model obtains a judgment coefficient based on at least one of the selected parameters, and the multivariate regression model determines a fitting result of the judgment coefficient based on an information criterion. If the fitting result does not converge to a threshold condition, the multivariate regression model selects other selected parameters and the current selected parameters to generate a new judgment coefficient and the fitting result until the fitting result converges to the threshold condition. A processing system for order estimation as described in claim 6, wherein the information criterion further includes Akagi information criterion, Bayesian information criterion or Kwai Fong information criterion. The order estimation processing system as described in claim 1, wherein the order end sends an update request to the server end every preset period, and the order end obtains the estimated value of the cargo quantity for the next preset period from the server end. A method for processing order estimation includes: A server performs a multivariate regression model calculation on a plurality of sales parameters of each of a plurality of commodity categories to select at least one of the sales parameters, and the selected at least one of the sales parameters is a selected parameter; The multivariate regression model is used to calculate and generate a quantity estimation value of each of the commodity categories according to the selected parameter; The server sends the quantity estimation value to an ordering end; and The ordering end calculates a purchase estimation result by subtracting the quantity in stock information from the sum of the quantity estimation value and a simulated inventory quantity information. The processing method for order estimation as described in claim 9, wherein the step of "the server performs the multivariate regression model calculation on the multiple sales parameters of each of the commodity categories to select at least one sales parameter, and the selected at least one sales parameter is the selected parameter" includes: The server performs the multivariate regression model calculation on the sales parameters according to sales history information. The order estimation processing method as described in claim 9, wherein after the step of "the server performs the multivariate regression model calculation on the multiple sales parameters of each of the commodity categories to select at least one sales parameter, and the selected at least one sales parameter is the selected parameter", further includes: If the sales parameter is an empty set, the server uses a preset quantity as the purchase estimation result. The processing method for order estimation as described in claim 9, wherein after the step of "the server performs the multivariate regression model calculation on the multiple sales parameters of each of the commodity categories to select at least one sales parameter, and the selected at least one sales parameter is the selected parameter", the method includes: The server determines a fitting result of the multivariate regression model according to an information criterion; If the fitting result converges with the information criterion, the server regards the sales parameters corresponding to the fitting result as the selected parameters; and If the fitting result does not converge with the information criterion, the server reselects other sales parameters and executes the multivariate regression model. The order estimation processing method as described in claim 9, wherein after the step of "the ordering end generates the purchase estimation result based on the estimated quantity, the simulated inventory quantity information and the in-stock quantity information", further includes: The ordering end sends an update request to the server end every preset period to obtain the estimated quantity of the corresponding preset period.