Cigarette goods source putting method based on linear regression model

Abstract

The invention relates to the technical field of goods source putting algorithm models, in particular to a cigarette goods source putting method based on a linear regression model, which comprises the following steps: establishing a standard terminal sales rate prediction model; two linear goods source putting strategies based on the product and regulation terminal sales rate are constructed, the strategies comprise a capability priority strategy and an overall planning and sharing strategy, the capability priority strategy takes the product and regulation terminal sales rate as a goods source putting weight coefficient, and the higher the sales rate is, the larger the putting amount is; according to an overall planning and sharing strategy, the sales rates of the product and regulation terminals are ranked on a coordinate axis, and a plurality of numerical intervals are divided according to a certain threshold value, so that balanced distribution of goods sources is realized. The product and regulation terminal sales rate is innovatively constructed, the tradition that research on cigarette marketing of a tobacco commercial system basically focuses on an order terminal of a tobacco commercial company is changed, and the method adapts to the consumption trend of dissimilation, interestingness, branding and cost performance.

Description

Technical Field

[0001] The present application belongs to the technical field of supply placement, and specifically relates to a cigarette supply placement method based on a linear regression model. Background Art

[0002] Cigarette supply strategy is an important means for the tobacco business system to adjust the market status. Therefore, how to make supply delivery more consistent with the actual needs of retail customers under the constraints of macro factors such as regional economy and population, as well as actual conditions such as sales plans and supply combinations is a very important topic.

[0003] Although existing delivery strategies such as grade-level delivery, point-of-sale delivery and label delivery have their own advantages, they also have obvious shortcomings: grade-level delivery refers to dividing cigarette products into thirty different grades according to the traditional customer comprehensive index scores, and formulating different delivery strategies for customers of different grades, but this method does not take cigarette product specification information into consideration. For customers of the same grade, the delivery strategy for all product specifications is the same, and the flexibility is low; point-of-sale delivery refers to selecting retail outlets with high sales capabilities for key delivery through market research and data analysis, but it is highly subjective; label delivery refers to the use of big data and customer relationship management systems (CRM) to achieve a one-to-one correspondence between product specifications and customers, and accurately deliver products that meet the needs of different customer groups, but it requires continuous system updates and maintenance, which is costly and complex to operate.

[0004] In actual research, it was found that retail customers of the same level have different sales capabilities for different cigarette specifications. Researching the delivery strategy based on product sales capabilities to adapt to the consumption trends of differentiation, interest, branding, and cost-effectiveness has become a technical problem that needs to be solved urgently in this field. Summary of the invention

[0005] The present application provides a cigarette supply delivery method based on a linear regression model to solve or partially solve the problems raised in the above background technology.

[0006] The present application provides a method for placing cigarette supplies based on a linear regression model, comprising the following steps:

[0007] Establish a product terminal sales rate prediction model;

[0008] Construct two linear supply delivery strategies based on the terminal sales rate of product specifications, including the capability priority strategy and the overall equal distribution strategy, where:

[0009] The capability priority strategy uses the terminal sales rate of the product specification as the weight coefficient for supply release. The higher the sales rate, the greater the supply release.

[0010] The overall average strategy realizes the balanced distribution of the supply by sorting the terminal sales rates of product specifications on the coordinate axis and dividing them into multiple numerical intervals according to a certain threshold.

[0011] Preferably, the construction of the product specification terminal sales rate prediction model is based on fitting the sample data using any one of the linear regression model LR, extreme gradient boosting model XGBoost, SVM support vector machine, and neural network NN.

[0012] Preferably, the specific method for establishing a product specification terminal sales rate prediction model based on the linear regression model is as follows:

[0013] S1: Obtain the sample data and preprocess the sample data;

[0014] S2: Divide the sample data into a training set and a test set;

[0015] S3: After passing the multicollinearity test, use the stepwise regression method combined with the AIC criterion for variable selection to establish a preliminary model, specifically as follows:

[0016] Product specification terminal sales rate = β0 + β1 * region - β2 * business type + β3 * cigarette display area + β4 * theme promotion activity - β5 * cooperation in publicity and promotion + β6 * product specification order volume, where β0 is a constant term, and β1 to β5 are regression coefficients.

[0017] Preferably, the method for establishing a product specification terminal sales rate prediction model based on the linear regression model further includes:

[0018] S4: Use the preliminary model to predict the product sales rate of the prediction target set, and use the "kσ" rule to screen out outliers, where k is an adjustable parameter;

[0019] S5: For the outliers and non-outliers screened out, establish sample one and sample two respectively. For sample one and sample two, based on the same steps of the preliminary model, establish steps to establish model one and model two respectively to predict them.

[0020] Preferably, in step S1, the method for predicting the sample data includes:

[0021] S101: Quantify the qualitative data in the customer indicators;

[0022] According to the degree of correlation between the static attributes and business capabilities in the customer indicators, assign values to the attributes from small to large. The larger the value, the stronger the business capabilities, and it has a positive correlation with the terminal sales rate;

[0023] S102: Screen and complete the sample;

[0024] Delete the samples with missing values in the product sales rate, i.e., the dependent variable, because the prediction accuracy cannot be measured for them, and assign the missing values in other attributes, i.e., the independent variables, to 0;

[0025] S103: For the dynamic attributes in the customer metrics, perform standardization on them through "normal normalization" to eliminate the influence of dimensions.

[0026] Preferably, in the step S2, the test set and the training set are set in a ratio of 2:8.

[0027] Preferably, the specific methods for constructing the ability - priority strategy and the overall - balance and equal - distribution strategy are as follows:

[0028] Construct the source - supply distribution task:

[0029] Let the total policy distribution period be T, which is divided into n periods in total, the total number of on - sale product specifications be m, the available distribution quantity of a certain product specification in a certain period be Q (usually in the unit of strips), the total number of retail customers available for distribution be u, and the distribution quantity of each retail customer for this product specification in this period be a j , k = 1, 2…u, aj≥0, and the sales rate of each retail customer predicted by the model is denoted as r j , j = 1, 2,..., u, r j >0. Let the cumulative sum of the sales rates of the distributed retail customers be R, then according to the definition:

[0030]

[0031] Sort r j in descending order from large to small, the maximum value is denoted as r max , the minimum value is denoted as r min , and the average value is denoted as r ave . Divide the retail customers into two parts according to their sales capabilities. If r min ≤r j ≤r ave , then x j ∈E (weak sales ability). If r ave <r j <r max , then x j ∈F (strong sales ability). Let the number of elements in E be s, then the number of elements in F is u - s;

[0032] Construct the ability - priority strategy and the overall - balance and equal - distribution strategy based on the source - supply distribution task.

[0033] Preferably, the specific method for constructing the ability - priority source - supply distribution strategy is as follows:

[0034] Let Q = Rw + b, 0≤b<R, w∈N + , then

[0035] If Q < R, then w = 0 and b = R, indicating that the current supply of this product specification is relatively scarce. Arrange the elements in F in descending order of r, and distribute 1 piece to each household until the supply reaches 0. If there is still remaining supply after traversing the elements in F, then perform recursion until the supply reaches 0 to end the distribution; j Reduce the order, and distribute 1 piece to each household until the supply is 0. If there is still remaining supply after traversing the elements in F, then perform recursion until the supply is 0 to end the distribution;

[0036] If Q ≥ R, let the distribution quantity in F be The distribution quantity in E is where the symbol represents rounding up, and the symbol represents rounding down. In this way, the total distribution quantity Q' ≈ Q, and fine-tuning is required;

[0037] If Q' > Q, then arrange the elements in F in ascending order of r j and decrease the distribution quantity by 1 in turn until Q' = Q;

[0038] If Q' < Q, let the elements in E be arranged in descending order of r j and increase the distribution quantity by 1 in turn until Q' = Q; If Q' = Q, then no additional fine-tuning is required.

[0039] Preferably, the specific method for constructing the overall coordinated and evenly distributed supply distribution strategy is as follows:

[0040] Let the average distribution quantity Then Q = ud + g, where 0 ≤ g < u;

[0041] If d = 0, at this time Q < u, indicating that the current supply of this product specification is relatively scarce. Arrange the elements in F in descending order of rj, and distribute 1 piece to each household until the supply reaches 0. If there is still remaining supply after traversing the elements in F, then perform recursion until the supply reaches 0 to end the distribution;

[0042] If d ≥ 1, let Divide the set F into v subsets F1, F2,..., F v , where the element attribute in F1 is and so on. Let the number of elements in F1 be g1, the number of elements in F2 be g2, and so on. The number of elements in F v is g v , then there is being the number of elements in F. Let the initial distribution quantity of the elements in F be d, and each household in F1 gets 1 more piece, and so on. Each household in F v gets v more pieces. Denote the total additional quantity in F as P = g1 + 2g2 + … + vg v ;

[0043] Let Divide the set \(E\) into \(z\) subsets \(E_{1}\), \(E_{2}\), \(\cdots\), \(E_{z}\), where the element attributes in \(E_{i}\) are z , \(E_{2}\), z-1 , \(\cdots\), \(E_{1}\), where the z element attributes in \(E_{i}\) are and so on. Let the number of elements in \(E_{i}\) be \(h_{i}\). z The number of elements in \(E\) is \(h\). z , \(E_{2}\), z-1 The number of elements in \(E_{2}\) is \(h_{2}\). z-1 And so on. The number of elements in \(E_{1}\) is \(h_{1}\). Then we have Let \(h\) be the number of elements in \(E\). Let the initial delivery volume of each element in \(E\) be \(d\). Then, make multiple deductions on the delivery volume of elements in \(E\) (skip if it is 0). For the first time, deduct 1 item from each customer in \(E_{1}\). For the second time, deduct 1 item from each customer in \(E_{1}\) and \(E_{2}\) in sequence. For the \(k\)th time, deduct 1 item from each customer in \(E_{1}\) to \(E_{k}\) in sequence until the cumulative deduction quantity reaches \(P\). k until the cumulative deduction quantity reaches \(P\).

[0044] Compared with the prior art, the beneficial effects of the present application are:

[0045] (1) The present application innovatively constructs the terminal sales rate of product specifications, changes the tradition that the research on cigarette marketing in the tobacco business system mainly focuses on the order side of tobacco commercial companies. The capacity - priority source delivery strategy is applicable to scenarios such as new product launch, brand cultivation, high market inventory of product specifications, and many seasonal customers. The overall - average source delivery strategy is applicable to scenarios such as stabilizing the basic sales volume, product - specification complementarity, stabilizing the profit level of retail customers, large - customer management, and balancing the cigarette market, adapting to the consumption trends of diversification, interest - orientation, branding, and cost - performance.

[0046] (2) The present application generates a prediction model for the terminal sales rate of product specifications based on fitting and iteration of a linear regression model, screens and assigns weights to each index affecting the sales rate, and the model has a high prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present application is further described below with reference to the drawings and embodiments.

[0048] Figure 1 is the preliminary model prediction effect diagram of the present application.

[0049] Figure 2 is the schematic diagram of the construction process of the product - specification terminal sales rate prediction model of the present application.

[0050] Figure 3 is the schematic diagram of the implementation of the delivery strategy of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] As used in the specification and claims, certain terms are used to refer to specific components. Those skilled in the art should understand that hardware manufacturers may use different terms to refer to the same component. The specification and claims do not distinguish components by the difference in names, but by the difference in functions of the components. As used throughout the specification and claims, the term "comprising" is an open-ended term and should be interpreted as "comprising but not limited to". "Substantially" means within an acceptable error range, and those skilled in the art can solve the technical problem within a certain error range and basically achieve the technical effect.

[0052] In the description of the present application, it should be understood that the orientation or positional relationship indicated by the terms "upper", "lower", "front", "rear", "left", "right", "horizontal", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present application.

[0053] In the present application, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed", etc. should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0054] Embodiment 1

[0055] As Figures 1 to 3 shown, the present application provides a cigarette supply placement method based on a linear regression model, and the specific steps are as follows:

[0056] Establish a prediction model for the terminal sales rate of product specifications;

[0057] Construct two linear supply placement strategies based on the terminal sales rate of product specifications, including the capacity priority strategy and the overall balance strategy, where:

[0058] The capacity priority strategy uses the terminal sales rate of product specifications as the supply placement weight coefficient. The higher the sales rate, the larger the placement quantity;

[0059] The overall balance strategy realizes the balanced distribution of supply by sorting the terminal sales rate of product specifications on the coordinate axis and dividing multiple numerical intervals according to a certain threshold.

[0060] Specifically, the definition of the terminal sales rate of product specifications is as follows:

[0061] Let \(r = q / t\), where \(r\) represents the terminal sales rate of a certain product specification, \(q\) represents the sales volume (in cartons) of this product specification, and \(t\) represents the number of days with inventory for this product specification. Product specification refers to brand and specification. For example, the product specification of Seven Wolves - Gutian Red Army Gray, where Seven Wolves is the brand and Gutian Red Army Gray is the specification.

[0062] In the actual terminal sales market, it is difficult to count the sales volume of each retail customer. Taking the tobacco system in a certain city in Fujian as an example, there are 33,328 retail customers who place regular orders, and only about 900 retail customers are e - Futong users. According to the definition of the terminal sales rate of product specifications, the \(r\) of e - Futong retail customers for a certain product specification can be directly calculated. However, the \(r\) of other retail customers cannot be directly calculated. Therefore, customer clustering is required, and regression analysis is used to estimate the \(r\) of other retail customers. E - Futong is a terminal sales information system deployed by the Fujian tobacco commercial system for some retail customers, which can obtain the transaction data of cigarette sales.

[0063] The construction of the product - specification terminal - sales - rate prediction model can be based on the linear regression model LR, XGBoost (eXtreme Gradient Boosting), SVM (Support Vector Machine), and NN (Neural Network) to fit the sample data and establish the product - specification terminal - sales - rate prediction model. By measuring the performance of the four models on the training set and test set through the mean squared error MSE, it is found that the linear regression model performs best on the test set. Therefore, this application uses the linear regression model as the preferred solution.

[0064] Specifically, the specific method for establishing the product - specification terminal - sales - rate prediction model based on the linear regression model is as follows:

[0065] S1: Obtain the sample data and pre - process the sample data;

[0066] S2: Divide the sample data into a training set and a test set;

[0067] S3: After performing a multicollinearity test, use the stepwise regression method combined with the AIC criterion for variable selection to establish a preliminary model, specifically as follows:

[0068] Product - specification terminal sales rate=\(\beta_0+\beta_1\times Region-\beta_2\times Business format type+\beta_3\times Cigarette display area+\beta_4\times Theme promotion activity-\beta_5\times Cooperation in promotional activities+\beta_6\times Product - specification order quantity\), where \(\beta_0\) is a constant term, and \(\beta_1\) to \(\beta_5\) are regression coefficients.

[0069] The main factors affecting customer sales rate include regional economic factors, past order information, social factors and seasonal factors. These factors interact with each other and have a common impact on local cigarette market sales. Through system query, marketing personnel evaluation, e-Futong and external collaboration, relevant data were collected, cleaned and processed, as shown in Table 1, which are 28 customer indicators that affect sales rate.

[0070] Table 1

[0071]

[0072]

[0073] In step S1, the method for predicting sample data includes:

[0074] S101: Quantify qualitative data from customer metrics;

[0075] According to the correlation between the static attributes in the customer indicators and the operating ability, the attributes are assigned values from small to large. The larger the value, the stronger the operating ability, which is positively correlated with the terminal sales rate;

[0076] S102: Screen and complete samples;

[0077] The samples with missing values in the dependent variable, i.e., product sales rate, are deleted because they cannot be used to measure the prediction accuracy. The missing values in other attributes, i.e., independent variables, are assigned a value of 0.

[0078] S103: The dynamic attributes in the customer indicators are standardized through “normalization” to eliminate the impact of the dimension.

[0079] In step S101, the static attributes in the customer indicators refer to qualitative data such as classification, sequence or character type. For example, among the 11 tags of business segmentation, the data analysis results show that the average terminal sales rate of "leisure and entertainment store" is the highest, so the operating ability is the strongest, and the value is assigned to 11, while the supermarket is second and the value is assigned to 10, and the rest of the tags are similar.

[0080] In step S103, the dynamic attributes in the customer index refer to various numerical independent variables, such as "order volume" and "business area".

[0081] In step S2, the ratio of the test set to the training set is 2:8.

[0082] In step S3, taking the product specification of "Furongwang (hard)" as an example, its linear regression model is as follows:

[0083] Product sales rate = 0.2275 + 0.0084 * Region - 0.0119 * Business type + 0.0114 * Cigarette display area + 0.0189 * Theme promotion activity - 0.0211 * Coordinated publicity and promotion + 0.1965 * Order volume of Furongwang (Hard).

[0084] As Figure 1 shown, after establishing the preliminary model, the model was fitted and visualized for the test set and the training set. The results showed that the error MSE on the training set was 0.01160543, and the error MSE on the test set was 0.01378244. Observing the fitting images of the training set and the test set, it was found that the preliminary established model had good prediction results on most samples with low values and concentrated distributions, but the prediction results were average on samples with large values and discrete distributions.

[0085] Considering screening outlier points from the data set, and at the same time considering that these outlier points are data samples existing in the real situation and contain certain valid information and cannot be easily deleted. To improve the prediction efficiency of the model, for the screened outlier samples (in this embodiment, 1305 outlier points were screened out from a total of 32442 prediction samples), according to the above process of establishing the preliminary model, multiple regression methods were used to construct Model 1, and for the remaining samples after screening out the outlier points, Model 2 was established according to the same process. The specific steps are as follows:

[0086] S4: Use the preliminary model to predict the product sales rate of the prediction target set, and use the "kσ" rule to screen outlier points, where k is an adjustable parameter;

[0087] S5: For the screened outlier points and non-outlier points, Sample 1 and Sample 2 are established respectively. For Sample 1 and Sample 2, based on the same steps of the preliminary model, Model 1 and Model 2 are established respectively to predict them.

[0088] After testing, the MSE (error) of Model 1 on the training set was 0.003520614, and on the test set was 0.003617384. While the error MSE of the preliminary model on the training set was 0.01160543, and on the test set was 0.01378244. It can be seen that Model 1 has a greater improvement in prediction accuracy compared to the preliminary model.

[0089] Specifically, the specific methods for constructing the capacity priority strategy and the overall balance strategy based on the terminal sales rate of product specifications are as follows:

[0090] S100: Construct the task of source allocation;

[0091] Let the total strategy delivery period be T, which is divided into n periods in total. The total number of in-sale product specifications is m. The available delivery quantity of a certain product specification in a certain period is set as Q (usually in the unit of cartons). The total number of retail customers that can be delivered to is u. Let the delivery quantity of each retail customer for this product specification in this period be a j , k = 1, 2…u, aj≥0. The sales rate of each retail customer predicted by the model is denoted as r j , j = 1, 2,..., u, r j >0. Let the cumulative sum of the sales rates of the delivered retail customers be R. Then, according to the definition, there is

[0092]

[0093] Sort r j in descending order from large to small. The maximum value is denoted as r max , and the minimum value is denoted as r min . The average value is denoted as r ave . Divide the retail customers into two parts according to their sales capabilities. If r min ≤r j ≤r ave , then x j ∈E (weak sales ability). If r ave <r j <r max , then x j ∈F (strong sales ability). Let the number of elements in E be s, then the number of elements in F is u - s.

[0094] S200: Construct a source delivery strategy with priority given to ability;

[0095] Let Q = Rw + b, 0≤b<R, w∈N + , then

[0096] If Q < R, then w = 0, b = R, indicating that the current supply of this product specification is relatively scarce. Sort the elements in F in descending order according to r j , and deliver 1 carton to each household until the supply is 0. If there is still remaining supply after delivering to all elements in F, then perform recursion until the supply is 0 and the distribution ends;

[0097] If Q ≥ R, let the delivery quantity in F be and the delivery quantity in E be where the symbol represents rounding up, and the symbol represents rounding down. In this way, the total delivery quantity Q'≈Q and fine-tuning is required;

[0098] If Q'>Q, then sort the elements in F in ascending order according to r j , and reduce the delivery quantity by 1 in turn until Q' = Q;

[0099] If Q' < Q, arrange the elements in E in descending order according to r j and increase the delivery quantity by 1 in sequence until Q' = Q;

[0100] If Q' = Q, no additional fine-tuning is required.

[0101] S300: Construct an overall coordinated and evenly distributed supply source delivery strategy;

[0102] Let the average delivery quantity Then Q = ud + g, where 0 ≤ g < u;

[0103] If d = 0, at this time Q < u, indicating that the current supply of this product specification is relatively scarce. Arrange the elements in F in descending order according to rj, and deliver 1 piece to each household until the supply source is 0. If there is still remaining supply source after traversing the elements in F, then perform recursion until the supply source is 0 to end the distribution;

[0104] If d ≥ 1, let Divide the set F into v subsets F1, F2,..., F v , where the element attributes in F1 are And so on. Let the number of elements in F1 be g1, the number of elements in F2 be g2, and so on. The number of elements in F v is g v , then there is is the number of elements in F. Let the initial delivery quantity of the elements in F be d, and each household in F1 is delivered 1 more piece, and so on. Each household in F v is delivered v more pieces. Denote the additional delivery quantity in F as P = g1 + 2g2 + … vg v ;

[0105] Let Divide the set E into z subsets E z , E z-1 ,..., E1, where the element attributes in E z are And so on. Let the number of elements in E z be h z , the number of elements in E z-1 be h z-1 , and so on. The number of elements in E1 is h1. Then there is is the number of elements in E. Let the initial delivery quantity of the elements in E be d, and then deduct the delivery quantity of the elements in E multiple times (skip if it is 0). For the first time, deduct 1 piece from each household in E1. For the second time, deduct 1 piece from each household in E1 and E2 in sequence. For the kth time, deduct 1 piece from each household in E1 to E k in sequence until the cumulative deduction quantity is P.

[0106] The basic idea of the supply allocation strategy based on ability priority is "robbing the poor to assist the rich". A horizontal axis is established, and the "product specification sales rate" is sorted on the numerical axis, and multiple numerical intervals are divided according to a certain threshold. In the interval where r ave is located, the average value d (total supply volume / total number of customers) is allocated. In the intervals on the right, the supply quantity is increased regularly, and in the intervals on the left, the supply quantity is deducted regularly, so as to complete the supply allocation for all customers. This kind of thinking is relatively radical, and the difference in the supply quantity among customers is relatively large. The applicable scenarios of the supply allocation strategy based on ability priority are new product launch, brand cultivation, high market inventory of product specifications, and many seasonal customers, etc.

[0107] The overall planning and equal distribution supply allocation strategy uses the "product specification sales rate" calculated by the model as the supply allocation weight coefficient for customers. For a specific product specification, customers with strong sales ability are allocated more, and customers with weak sales ability are allocated less or not at all. This kind of thinking is relatively balanced, and the difference in the supply quantity among customers is relatively small. The applicable scenarios of the overall planning and equal distribution supply allocation strategy are to stabilize the basic sales volume, product specification complementarity, stabilize the profitability level of retail customers, large customer management, balance the cigarette market, etc.

[0108] The allocation objects of the two supply allocation strategies are dynamically adjusted according to the cycle. The principle of balanced application is the combination of specialized sales. Comprehensive reference is made to data on the outflow and inflow of genuine cigarettes, large customer management, the profitability of retail customers, the state of the cigarette market, etc. The key is to conform to the consumption trends of differentiation, interest, branding, and cost performance, so that the supply allocation is more in line with the actual needs of retail customers.

[0109] Taking a certain product specification as an example, the supply allocation strategy for the 21st week is formulated according to the sales situation of some allocated customers of a certain city's Tobacco Monopoly Bureau for this product specification in the 20th week of 2024.

[0110] First, the sales situation and basic information of 18,411 allocated customers are input into the sales rate prediction model to obtain the predicted sales rate values of each customer. Among them, the maximum sales rate is 2.392561, the minimum value is 0.000634, the average value is 0.210409, the total sales rate is 3873.850156, the number of elements in E with weak sales ability is 9507, the number of elements in F with strong sales ability is 8904, the total supply volume for the 21st week is 25,321 pieces, and the average supply volume d = 1. Then, the specific results of different supply allocation strategies are calculated, as shown in Figure 3 , and the +N allocation strategy in the figure is the original allocation strategy.

[0111] To verify the effectiveness of this strategy, experiments and observations were conducted on 171 product specifications that were all in circulation in a certain city. During the experiment period, operations were carried out strictly in accordance with the established placement strategy, and key indicators such as the single-product scanned sales volume, inventory quantity, and days with inventory were tracked and recorded in detail. The statistical results showed that the inventory turnover rates of 85% of customers were significantly improved, the customer gross profit margin increased by 1.7% year-on-year, the outflow of genuine cigarettes decreased by 2.2% year-on-year, and the market condition was optimized.

[0112] The embodiments of the present application have been described in detail above in conjunction with the accompanying drawings. However, the present application is not limited to the above embodiments, and various changes can be made without departing from the gist of the present application within the scope of knowledge possessed by those of ordinary skill in the art.

Claims

1. A cigarette supply delivery method based on a linear regression model, characterized in that It includes the following steps: Establish a prediction model for the terminal sales rate of product specifications; Construct two linear supply source allocation strategies based on the terminal sales rate of product specifications, including the capacity priority strategy and the overall planning and equal distribution strategy, where: For the capacity priority strategy, the terminal sales rate of product specifications is used as the supply source allocation weight coefficient. The higher the sales rate, the larger the allocation volume; For the overall planning and equal distribution strategy, by sorting the terminal sales rate of product specifications on the coordinate axis and dividing it into multiple numerical intervals according to a certain threshold, the balanced allocation of supply sources is achieved.

2. The cigarette supply source allocation method based on the linear regression model according to claim 1, wherein: The construction of the prediction model for the terminal sales rate of product specifications is based on fitting the sample data using any one of the linear regression model LR, extreme gradient boosting model XGBoost, SVM support vector machine, and neural network NN.

3. The cigarette supply source allocation method based on the linear regression model according to claim 2, wherein: The specific method for establishing a prediction model for the terminal sales rate of product specifications based on the linear regression model is as follows: S1: Obtain sample data and preprocess the sample data; S2: Divide the sample data into a training set and a test set; S3: After passing the multicollinearity test, use the stepwise regression method combined with the AIC criterion for variable selection to establish a preliminary model, specifically as follows: Terminal sales rate of product specifications = β0 + β1 * region - β2 * business format type + β3 * cigarette display area + β4 * theme promotion activity - β5 * cooperation in promotion + β6 * product specification order volume, where β0 is a constant term, and β1 to β6 are regression coefficients.

4. The cigarette supply source allocation method based on the linear regression model according to claim 3, wherein: The method for establishing a prediction model for the terminal sales rate of product specifications based on the linear regression model further includes: S4: Use the preliminary model to predict the product sales rate of the prediction target set, and use the "kσ" rule to screen out outliers, where k is an adjustable parameter; S5: For the outliers and non-outliers screened out, establish sample one and sample two respectively. For sample one and sample two, based on the same steps of the preliminary model, establish model one and model two respectively to predict them.

5. The cigarette supply source allocation method based on the linear regression model according to any one of claims 3 or 4, wherein: In the step S1, the method for predicting the sample data includes: S101: Quantify the qualitative data in the customer indicators; According to the correlation degree between the static attributes and business capabilities in the customer indicators, assign values to the attributes from small to large. The larger the value, the stronger the business capabilities, and it has a positive correlation with the terminal sales rate; S102: Screen and complete the sample; Delete the samples with missing values in the product sales rate, i.e., the dependent variable, because the prediction accuracy cannot be measured. Assign the missing values in other attributes, i.e., the independent variables, as 0; S103: For the dynamic attributes in the customer indicators, perform standardization processing on them through "normal normalization" to eliminate the influence of dimensions.

6. The cigarette supply source distribution method based on a linear regression model according to any one of claims 3 or 4, characterized in that: In step S2, the test set and the training set are set in a ratio of 2:

8.

7. The cigarette supply source distribution method based on a linear regression model according to claim 1, characterized in that: The specific methods for constructing the ability priority strategy and the overall planning and equal distribution strategy are as follows: Construct the supply source distribution task: Let the total strategy delivery period be T, which is divided into n periods in total. The total number of in-sale product specifications is m. The deliverable quantity of a certain product specification in a certain period is set as Q (usually in the unit of cartons). The total number of retail customers that can be delivered to is u. Let the delivery quantity of each retail customer for this product specification in this period be a j , j = 1, 2…u, a j ≥0. The sales rate of each retail customer predicted by the model is denoted as r j , j = 1, 2,..., u, r j >0. Let the cumulative sum of the sales rates of the delivered retail customers be R. Then, according to the definition, we have: Sort r j in descending order from largest to smallest, and denote the maximum value as r max , and the minimum value as r min , and the average value as r ave . Divide retail customers into two parts according to their sales ability. If r min ≤r j ≤r ave , then x j ∈E (weak sales ability). If r ave <r j <r max , then x j ∈F (strong sales ability). Let the number of elements in E be s, then the number of elements in F is u - s; Based on the supply source distribution task, construct the ability priority strategy and the overall planning and equal distribution strategy.

8. The cigarette supply source distribution method based on a linear regression model according to claim 7, characterized in that: The specific method for constructing the ability priority supply source distribution strategy is as follows: Let \(Q = Rw + b\), where \(0\leq b < R\) and \(w\in N\). + , then If Q < R, then w = 0 and b = R, indicating that the current supply of this product specification is relatively scarce. Arrange the elements in F in descending order according to r j and distribute 1 piece to each household until the supply runs out. If there is still remaining supply after traversing the elements in F, then perform recursion until the supply runs out and the distribution ends; If Q ≥ R, let the input amount in F be the input amount in E be where the symbol represents rounding up, and the symbol represents rounding down. In this way, the total input amount Q' ≈ Q and fine-tuning is required; If Q'>Q, then arrange the elements in F in ascending order according to r j and decrease the delivery quantity by 1 in turn until Q' = Q; If Q' < Q, arrange the elements in E in descending order according to r j and increase the delivery volume by 1 in sequence until Q' = Q; if Q' = Q, no additional fine-tuning is required.

9. The cigarette supply source distribution method based on a linear regression model according to claim 7, characterized in that: The specific method for constructing the overall planning and equal distribution supply source distribution strategy is as follows: Set the average delivery volume Then Q = ud + g, where 0 ≤ g < u; If d = 0, at this time Q < u, indicating that the current supply of this product specification is relatively scarce. Arrange the elements in F in descending order of r j and distribute 1 piece to each household until the supply is 0. If there is still remaining supply after traversing the elements in F for distribution, then perform recursion until the supply is 0 to end the distribution; If d ≥ 1, let Partition the set F into v subsets F1, F2, ..., F v , where the element attribute in F1 is And so on. Let the number of elements in F1 be g1, the number of elements in F2 be g2, and so on. The number of elements in F v is g v , then there is being the number of elements in F. Let the initial delivery quantity of each element in F be d, and each customer in F1 gets 1 more item, and so on. Each customer in F v gets v more items. Denote the additional quantity in F as P = g1 + 2g2 + … vg v ; Let Partition the set E into z subsets E z , E z-1 ,..., E1, where the element attributes in E z are And so on. Let the number of elements in E z be h z , the number of elements in E z-1 be h z-1 , and so on. The number of elements in E1 is h1. Then there is being the number of elements in E. Let the initial delivery quantity of each element in E be d. Then, perform multiple deductions on the delivery quantity of the elements in E (skip if it is 0). For the first time, deduct 1 item from each customer in E1. For the second time, deduct 1 item from each customer in E1 and E2 in sequence. For the kth time, deduct 1 item from each customer in E1 to E k in sequence until the cumulative deduction quantity is P.

Images