Product sales prediction method and device, electronic equipment and medium

Abstract

The invention relates to the technical field of data processing, and provides a product sales prediction method and device, electronic equipment and a medium, and the method comprises the steps: obtaining historical sales information of a plurality of marketing categories of a product and a historical sales proportion information set of each marketing category; predicting the total sales volume of each marketing category in a future set time period according to each piece of historical sales volume information; according to each historical sales volume ratio information set, predicting a sales volume ratio of each product model in a future set time period; and according to the total sales volume of each marketing category and the sales volume proportion of each product model in each marketing category, predicting the sales volume of the product model in a future set time period. According to the method, the set of the plurality of product models, namely the overall total sales volume of each marketing category in the future time period, is predicted firstly, and then the sales volume proportion of the future time period is predicted according to the sales volume proportion change of the single product model in the marketing category to which the single product model belongs, so that the product sales volume prediction is realized, and the prediction accuracy and efficiency are improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a product sales forecasting method, device, electronic device and medium. Background Art

[0002] With the popularity of online sales models such as shelf e-commerce and interest e-commerce, the main problem of the supply chain in the traditional home appliance manufacturing industry is the serious bullwhip effect, that is, some fluctuations in upstream material procurement and other links are transmitted layer by layer, and will be amplified step by step to the downstream. The greater the change in the demand side at the end of the supply chain, the greater the change, which often leads to a backlog of goods or a long-term shortage of goods, thus having a negative impact on the ever-changing online sales. Therefore, in order to ensure the stability and accuracy of the entire product supply chain, it is necessary to further reduce costs and increase efficiency, and predict product sales in advance, which can minimize the impact of demand fluctuations on the quality of upstream material procurement, production planning and other decision-making. Therefore, how to achieve product sales forecasting has become an urgent problem to be solved. Summary of the invention

[0003] The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application proposes a product sales forecasting method to achieve product sales forecasting and improve the accuracy and efficiency of the forecast.

[0004] The application also proposes a product sales forecasting device, an electronic device and a storage medium.

[0005] The product sales forecasting method according to the first aspect of the present application includes: Obtaining historical sales information of multiple marketing categories of the product, and a historical sales ratio information set of each marketing category; the historical sales ratio information set of any marketing category includes historical sales ratio information of various product models in the marketing category; Based on the historical sales information, predict the total sales of each marketing category within a future set time period; Based on each of the historical sales volume share information sets, predict the sales volume share of each of the product models within a set time period in the future; Based on the total sales of each of the marketing categories in the future set time period, and the sales proportion of each of the product models in each of the marketing categories in the future set time period, the sales of the product models in the future set time period are predicted.

[0006] According to an embodiment of the present application, predicting the total sales volume of each marketing category within a future set time period based on each of the historical sales volume information includes: According to each of the historical sales information, obtain the year-on-year sales data of each of the marketing categories in the previous year; the year-on-year sales data includes the historical total sales volume and the year-on-year change ratio of the sales volume in the same period; Get the year-on-year change rate of total product market sales in the most recent consecutive historical weeks in the future set time period; Determine the predicted value of the year-on-year change in total sales volume for each marketing category within a future set time period based on the preset year-on-year calculation weight parameters, the year-on-year sales volume change ratio and the year-on-year change rate; Based on the historical total sales volume for the same period and the predicted value of the year-on-year change in the total sales volume, the total sales volume of each of the marketing categories in a set time period in the future is predicted.

[0007] According to an embodiment of the present application, the predicted value of the year-on-year change in total sales volume of each marketing category in a future set time period is determined based on the preset year-on-year calculated weight parameter, the year-on-year change ratio of sales volume and the year-on-year change rate, including: Determine the year-on-year change value and the first year-on-year change rate forecast value of the target historical week according to the year-on-year change rate; Determine a second year-on-year change rate forecast value of the total market sales volume of each marketing category within a future set time period according to a preset exponential smoothing weight parameter, the year-on-year change value and the first year-on-year change rate forecast value; Based on the preset year-on-year calculation weight parameters, the year-on-year sales volume change ratio and the second year-on-year change rate forecast value, the forecast value of the year-on-year change in total sales volume of each marketing category within a future set time period is determined.

[0008] According to an embodiment of the present application, predicting the sales share of each product model in a future set time period based on each of the historical sales share information sets includes: According to the historical sales volume share information, obtain the mid- to long-term sales volume share characteristics and the short-term sales volume share characteristics of each product model; Inputting the mid- to long-term sales volume share characteristics and the short-term sales volume share characteristics into a first prediction model, obtaining the mid- to long-term sales volume share and the short-term sales volume share of each product model output by the first prediction model; the first prediction model includes a plurality of linear layers, and the linear layers are used to predict the sales volume share of each product model; The static attribute characteristics of each product model, the mid- to long-term sales ratio and the short-term sales ratio are integrated to obtain integrated data; the static attribute characteristics at least include door body type, product positioning and product life cycle; The fused data is input into a second prediction model to obtain the sales share of each product model in a future set time period output by the second prediction model; the second prediction model includes a self-attention mechanism layer, and the self-attention mechanism layer is used to determine the relevant information between the sales share change at each time node and the prediction target.

[0009] According to an embodiment of the present application, the linear layer includes a first linear layer and a second linear layer; the first linear layer is used to predict the medium- and long-term sales share based on the medium- and long-term sales share characteristics; the second linear layer is used to predict the short-term sales share based on the short-term sales share characteristics.

[0010] According to one embodiment of the present application, the marketing category is determined based on the following method: According to the product characteristic parameters of the product, the product models are divided to obtain a division result; According to the classification result, a plurality of the marketing categories are determined; any one of the marketing categories includes a plurality of product models with the same product characteristic parameters.

[0011] According to an embodiment of the present application, predicting the sales volume of the product model in the future set time period according to the total sales volume of each marketing category in the future set time period and the sales volume proportion of each product model in each marketing category in the future set time period includes: Calculate the product of the total sales volume of each marketing category in a future set time period and the sales volume proportion of each product model in each marketing category in a future set time period; The product is used as the sales volume of the product model in a future set time period.

[0012] A product sales forecasting device according to an embodiment of the second aspect of the present application includes: An acquisition module, used to acquire historical sales information of multiple marketing categories of a product, and a historical sales ratio information set of each marketing category; the historical sales ratio information set of any marketing category includes historical sales ratio information of various product models in the marketing category; A first prediction module, used to predict the total sales volume of each marketing category within a future set time period based on each of the historical sales volume information; A second prediction module is used to predict the sales share of each product model in a future set time period based on each of the historical sales share information sets; The third prediction module is used to predict the sales volume of the product model in a future set time period based on the total sales volume of each marketing category in the future set time period and the sales volume proportion of each product model in each marketing category in the future set time period.

[0013] According to an electronic device of an embodiment of the third aspect of the present application, the electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the product sales forecasting method described in any one of the above methods is implemented.

[0014] According to the non-transitory computer-readable storage medium of the fourth aspect embodiment of the present application, a computer program is stored thereon, and when the computer program is executed by a processor, the product sales forecasting method as described in any one of the above is implemented.

[0015] The above one or more technical solutions in the embodiments of the present application have at least one of the following technical effects: By first predicting a collection of several product models, that is, the overall total sales volume of each marketing category in the future time period, and then predicting the sales share of the future time period based on the changes in the sales share of individual product models in the marketing category to which they belong, we can achieve product sales forecasting and improve the accuracy and efficiency of the forecast.

[0016] Furthermore, production plans, procurement plans and inventory management can be planned better and more efficiently, and costs can be better controlled, inventory backlogs and waste can be reduced, and the efficiency and flexibility of the supply chain can be improved.

[0017] Furthermore, intelligent sales forecasting can also help companies better understand market demand, thereby better formulating sales strategies and promotion plans, and increasing market share and sales.

[0018] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 This is one of the flow charts of the product sales forecasting method provided in the embodiment of the present application.

[0021] Figure 2This is the second flow chart of the product sales forecasting method provided in the embodiment of the present application.

[0022] Figure 3 It is a schematic diagram of the sales forecasting model framework provided in an embodiment of the present application.

[0023] Figure 4 It is a module diagram of a product sales forecasting device provided in an embodiment of the present application.

[0024] Figure 5 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following is a further detailed description of the implementation of the present application in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present application but cannot be used to limit the scope of the present application.

[0026] In the description of the embodiments of the present application, it should be noted that the terms "first", "second" and "third" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0027] In the embodiments of the present application, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0028] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiments of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0029] In the face of fluctuations in demand on online e-commerce platforms, sales forecasting or demand forecasting is the first line of defense for supply chain management. Online sales forecasting requires the use of prediction algorithms to characterize historical sales patterns and relatively accurately predict demand-side fluctuations in the next month, thereby helping upstream procurement, production, and warehousing departments to prepare appropriate resources in advance, thereby improving resource utilization efficiency and reducing costs. In this application, in the supply chain scenario of household appliances represented by refrigerator products, due to the limitations of production and operation scale and the requirements of market competition, the number of product models on the market and the life cycle of product models are limited, so the product model sample data that can be used for prediction algorithm training is also very limited, and the length of the time series data sequence of each product model is also very limited, which requires the prediction model to find the sales trend pattern of each household appliance model in a small amount of sample data, and for some new products that have been on the market for a short time, the prediction model needs to predict its future sales pattern based on very short sales sequence data.

[0030] Figure 1 This is one of the flowcharts of the product sales forecasting method provided in the embodiment of the present application. Figure 1 The present application embodiment provides a product sales forecasting method, including: Step 100, obtaining historical sales information of multiple marketing categories of the product, and a historical sales share information set for each of the marketing categories; the historical sales share information set for any of the marketing categories includes historical sales share information for various product models in the marketing category.

[0031] In the embodiments of the present application, the products include various household appliances and electronic devices, such as washing machines, refrigerators, air conditioners, televisions, microwave ovens, induction cookers, etc.

[0032] According to the product characteristic parameters of the product, the product models are divided to obtain the division results; according to the division results, multiple marketing categories are determined. Any marketing category includes multiple product models with the same product characteristic parameters, and any marketing category includes the characteristics that the sales trend is relatively more regular and predictable. Among them, the product characteristic parameters include door type, size, capacity, energy efficiency level, functions and characteristics, price segment, etc. For example, according to the door types of refrigerators and freezers, such as single door, double door, cross four doors, multi-door, direct cooling two doors, air cooling two doors, direct cooling three doors, air cooling three doors, French four doors and household cabinets, multiple product models with the same door type are divided into one marketing category. Or, according to the energy efficiency level of the product, such as A+, A++, etc., multiple product models with the same energy efficiency level are divided into one marketing category. Or, according to the price segment of the product, such as high-end products, mid-range products, low-end products, etc., multiple product models with the same price segment are divided into one marketing category. Alternatively, based on the product's volume or size, such as small appliances, medium appliances, large appliances, etc., multiple product models of the same volume or size can be divided into one marketing category.

[0033] Obtain historical sales information of multiple marketing categories of a product, wherein the historical sales information includes sales quantity, sales time, sales location, sales channel, sales trend, etc. For example, obtain historical sales information of each marketing category from a sales database. Alternatively, connect to an e-commerce platform and extract product sales data to obtain historical sales information of each marketing category.

[0034] Furthermore, a historical sales share information set of each marketing category is obtained. The historical sales share information set of any marketing category includes the historical sales share information of various product models in the marketing category, wherein the historical sales share information includes the sales share of each product model in the marketing category to which it belongs. For example, assuming that a marketing category includes refrigerator model A, refrigerator model B, and refrigerator model C, and the total annual sales of the marketing category last year was 6,000 units, of which the annual sales of refrigerator model A last year was 1,000 units, the annual sales of refrigerator model B last year was 2,000 units, and the annual sales of refrigerator model C last year was 3,000 units, then the historical sales share information of each refrigerator model is: Sales volume of refrigerator model A = (1000 / 6000) × 100% = 16.67%; Sales volume of refrigerator model B = (2000 / 6000) × 100% = 33.33%; The sales share of refrigerator model C = (3000 / 6000) × 100% = 50.00%.

[0035] Based on this, assuming that the marketing category is refrigerator, the historical sales share information set of this marketing category includes the sales share of refrigerator model A of 16.67%, the sales share of refrigerator model B of 33.33%, and the sales share of refrigerator model C of 50.00%.

[0036] Step 200: predict the total sales volume of each marketing category within a future set time period based on the historical sales volume information.

[0037] A statistical model is built in advance, and the historical sales information of each marketing category is input into the statistical model to obtain the total sales of each marketing category in a set time period in the future (such as the next month) output by the statistical model.

[0038] Step 300: predict the sales share of each product model in a future set time period based on each of the historical sales share information sets.

[0039] A first prediction model and a second prediction model are pre-constructed, wherein the first prediction model is used to predict the mid- to long-term and short-term sales share of each product model; the second prediction model is used to predict the sales share of each product model in a set time period in the future based on the prediction results of the first prediction model and the static attribute characteristics of each product model.

[0040] On the basis of the first prediction model and the second prediction model, the sales share of each product model in a set time period in the future is predicted according to the historical sales share information of each product model in each historical sales share information set.

[0041] Step 400, predicting the sales volume of the product model in a future set time period based on the total sales volume of each marketing category in a future set time period and the sales volume proportion of each product model in each marketing category in a future set time period.

[0042] Calculate the product of the total sales of each marketing category and the sales share of each product model in each marketing category, and then use the product as the sales of the product model in the future set time period. For example, assume that the total sales forecast for a marketing category in the next month is 800 units, and that the sales share of model A in the marketing category in the next month is 40%, the sales share of model B is 30%, and the sales share of model C is 30%.

[0043] Calculate the sales volume of each product model in the next month: Model A’s predicted sales = 40% × 800 = 320 units; Model B’s forecasted sales = 30% × 800 = 240 units; Model C’s forecasted sales = 30% × 800 = 240 units; From this, we can see that in the next month, Model A may sell about 320 units, while Model B and Model C may sell about 240 units each.

[0044] It is understandable that the historical sales trend of a single product model is characterized by short time series and high volatility, and the number of product models is very limited. The product model samples that can be used for algorithm learning are far fewer than other prediction scenarios. The above problems pose great challenges to predicting the sales volume of a single product model in the next month.

[0045] Taking refrigerators and freezers as an example, this application divides refrigerators and freezers into several marketing categories according to the door type (each marketing category is a collection of several product models). By observing the data, it is found that the overall online sales trend of each marketing category has a strong similarity compared with the sales trend of the same period last year, so that the sales trend of a single product model that is difficult to predict can be turned into a relatively easy-to-predict overall sales trend of the marketing category. Then, by calculating the sales share of each product model in its marketing category each month, the sales fluctuations that vary greatly between individual product models can be compressed within a unified dimension range. For algorithm prediction, predicting the sales share of product models with unified dimensions has higher predictability. On the other hand, some refrigerator models have certain regularities in the long-term sales share changes of 6 months, and some refrigerator models have traces of sales share changes in the past 60 days. Therefore, the sales share change characteristics of refrigerator models in the medium and long term (6 months) and short term (nearly 60 days) need to be input into the prediction algorithm for learning, so that a better prediction effect can be achieved overall. Based on the above analysis, this application successfully transforms the problem of predicting the sales volume of a product model in the next month into: first predict a collection of several product models, that is, the overall total sales volume of each marketing category in the next month, and then predict the sales volume share in the next month based on the changes in the sales volume share of individual product models in their respective marketing categories.

[0046] The product sales forecasting method provided in the embodiment of the present application obtains the historical sales information of multiple marketing categories of the product, and the historical sales proportion information of various product models in each marketing category; according to the historical sales information of each marketing category, the total sales of each marketing category in the future set time period is predicted; according to the historical sales proportion information of each product model, the sales proportion of each product model in the future set time period is predicted; according to the total sales and sales proportion, the sales of each product model in the future set time period is predicted. The present application first predicts a collection of several product models, that is, the overall total sales of each marketing category in the future time period, and then predicts the sales proportion of the future time period according to the changes in the sales proportion of a single product model in the marketing category to which it belongs, so as to realize product sales forecasting. At the same time, production plans, procurement plans and inventory management can be planned better and more efficiently, and costs can be better controlled, inventory backlogs and waste can be reduced, and the efficiency and flexibility of the supply chain can be improved. In addition, intelligent sales forecasting can also help companies better understand market demand, so as to better formulate sales strategies and promotion plans, and increase market share and sales.

[0047] Based on the above embodiment, predicting the total sales volume of each marketing category within a future set time period according to each of the historical sales volume information includes: Step 211, based on each of the historical sales information, obtain the year-on-year sales data of each of the marketing categories in the previous year; the year-on-year sales data includes the historical total sales volume and the year-on-year change ratio of the sales volume in the same period; Step 212, obtaining the year-on-year change rate of the total market sales volume of the product in the most recent consecutive historical weeks in a future set time period; Step 213, determining a predicted value of the year-on-year change in total sales volume of each marketing category within a future set time period according to the preset year-on-year calculation weight parameter, the year-on-year change ratio of sales volume and the year-on-year change rate; Step 214, predicting the total sales of each marketing category within a set time period in the future based on the historical total sales for the same period and the predicted value of the year-on-year change in the total sales.

[0048] Based on historical sales information, calculate the total historical sales of each marketing category in the same period of the previous year ( marketing categories) and year-on-year sales change ( Marketing categories), where the same period can be understood as the same time period as the current time point, such as the same quarter or the same month of the previous year. The year-on-year sales volume change ratio includes the percentage of year-on-year sales volume growth or decrease, which is calculated as follows: Calculate the increase or decrease in sales volume during the same period of the previous year: current sales volume - sales volume during the same period of the previous year; Calculate the percentage increase or decrease in year-on-year sales: ((current sales - sales in the same period last year) / sales in the same period last year) × 100%.

[0049] For example, assuming that the total historical sales volume of refrigerators in the same period of the previous year was 5,000 units, and the total sales volume in the current period is 6,000 units, therefore, the year-on-year sales growth rate of refrigerators = ((6,000-5,000) / 5,000) × 100% = 20%.

[0050] Get the year-on-year change rate of the total market sales of the product in the most recent consecutive weeks in the future set time period. For example, if the sales volume is predicted for month M+1, calculate the year-on-year change rate of the total market sales volume of the online channel marketing category for each of the 8 consecutive weeks before the 15th of month M. = 1-8 weeks before the 15th of the month. Among them, the year-on-year change rate is an indicator used to compare the change of the same indicator between two different time points, and is used to analyze the change of an indicator (such as sales, profit, market share, etc.) in different time periods within the same cycle. It can be understood that the year-on-year change rate calculates the percentage change of the indicator between the current period and the same period last year (or the same period of the last complete cycle).

[0051] Determine the predicted value of the year-on-year change in total sales volume for each marketing category within a set time period in the future based on the year-on-year change ratio, the year-on-year change rate, and the preset weight parameters for year-on-year calculation. In one embodiment, determine the year-on-year change value and the first year-on-year change rate prediction value of the target historical week based on the year-on-year change rate; determine the second year-on-year change rate prediction value of the total market sales volume for each marketing category within a set time period in the future based on the year-on-year change ratio, the second year-on-year change rate prediction value, and the preset weight parameters for exponential smoothing; determine the predicted value of the year-on-year change in total sales volume for each marketing category within a set time period in the future based on the year-on-year change ratio, the second year-on-year change rate prediction value, and the preset weight parameters for year-on-year calculation. For example,

[0052] ; ; in, It is The forecast value of the year-on-year change in total sales volume of marketing categories M+1, is the weight parameter for year-on-year calculation, It is the forecast of the year-on-year change rate of total sales volume of the marketing category market in the last 8 weeks. The predicted value of the year-on-year change rate of the marketing category M+1 month, i.e. the second predicted value of the year-on-year change rate; It is the year-on-year change value in the week before the 15th of May, that is, the year-on-year change rate; It is the predicted value of the year-on-year change rate in the week before the 15th of May, that is, the first predicted value of the year-on-year change rate; is the weight parameter of exponential smoothing, is the year-on-year change ratio. It is calculated based on the following method: = ; = ; = ; = ; = ; = ; = ; = ; in, , , , , , , They are the year-on-year change rates from 2 weeks to 8 weeks before March 15th, , , , , , , They are the predicted values ​​of the year-on-year change rate from 2 weeks to 8 weeks before May 15th.

[0053] Finally, based on the historical total sales volume and the forecasted value of the year-on-year change in total sales volume during the same period, the total sales volume of each marketing category in the future set time period is predicted. Get the predicted total sales of each marketing category for month M+1 .

[0054] The embodiment of the present application improves the accuracy and efficiency of product sales forecasting by predicting the total sales of regular marketing categories.

[0055] Based on the above embodiment, predicting the sales share of each product model in a future set time period according to each of the historical sales share information sets includes: Step 311, obtaining the mid- to long-term sales volume ratio characteristics and the short-term sales volume ratio characteristics of each product model according to the historical sales volume ratio information; Step 312: input the mid- to long-term sales volume share feature and the short-term sales volume share feature into a first prediction model to obtain the mid- to long-term sales volume share and the short-term sales volume share of each product model output by the first prediction model; the first prediction model includes a plurality of linear layers, and the linear layers are used to predict the sales volume share of each product model; Step 313, the static attribute characteristics of each product model, the mid- and long-term sales ratio, and the short-term sales ratio are integrated to obtain integrated data; the static attribute characteristics at least include door body type, product positioning, and product life cycle; Step 314, input the fused data into a second prediction model to obtain the sales share of each product model in a future set time period output by the second prediction model; the second prediction model includes a self-attention mechanism layer, and the self-attention mechanism layer is used to determine the relevant information between the sales share change at each time node and the prediction target.

[0056] In the embodiment of the present application, the medium- to long-term is greater than or equal to 6 months, and the short-term is less than or equal to 2 months, wherein the medium- to long-term and the short-term can be determined based on the predicted demand, and are not limited here. The medium- to long-term sales volume share features include product model codes, marketing category sales volume from M-11 (last year's same period last year), year-on-year changes in the marketing category market, sales volume share of product models in the marketing category from M-6, average sales volume share of product models in the marketing category in the first 6 months, and sales volume share series features of product models in the marketing category with daily granularity. The short-term sales volume share features include product model codes, year-on-year changes in the marketing category market, sales volume share of product models in the marketing category from M-1, sales volume share of product models in the marketing category from M-2, and sales volume share series features of product models in the marketing category with daily granularity.

[0057] Then, the mid- to long-term sales volume share characteristics and the short-term sales volume share characteristics are input into the first prediction model, such as the MLP model, to obtain the mid- to long-term sales volume share and the short-term sales volume share of each product model output by the first prediction model. The MLP model includes multiple linear layers, which are used to predict the sales volume share of each product model. The calculation formula is: (two prediction tasks), are model parameters, is the characteristic data of the input model, is the residual part, is the model output, including and ,in, It is the sales volume share in the medium and long term. It is the short-term sales share.

[0058] The mid- to long-term sales share, short-term sales share and static attribute features of each product model are fused to obtain fused data, which is then input into the second prediction model to obtain the sales share of each product model in the future set time period output by the second prediction model. The second prediction model is a deep learning model, which includes a self-attention layer, a Linear layer and a Softmax layer. The self-attention layer is used to determine the relevant information between the sales share change at each time node and the prediction target.

[0059] For example, the output of the MLP model in the previous step is combined with the static attribute features and input into the self-attention model structure in the deep learning model to examine the correlation between the sales share change at each time node and the prediction target, and then pass through a Linear layer and a Softmax layer to obtain the final sales share prediction result. .

[0060] The embodiment of the present application improves the accuracy and efficiency of product sales forecasting by predicting the sales share of product models in their corresponding marketing categories.

[0061] In order to further analyze and explain the product sales forecasting method proposed in this application, refer to the following examples.

[0062] The embodiment of the present application specifically proposes a product sales forecasting method based on small sample learning, which mainly includes: first, according to the characteristics of each product (such as the door type of refrigerators and freezers), a number of SKUs (Stock Keeping Units) in each product are divided into several marketing categories, wherein each product corresponds to a SKU number, i.e., a product model. At the same time, a statistical learning method is used to predict the total sales of each marketing category. Secondly, based on the sales data of product models that have been sold online in the past, the sales share change trend of each SKU in the marketing category to which it belongs is obtained, so as to predict the sales proportion of the marketing category to which it belongs in the next month based on the sales share change trend, and finally, the total sales of the marketing category and the SKU sales share are combined to obtain the predicted sales of each SKU in the next month.

[0063] In one embodiment, the prediction of refrigerator and freezer products is taken as an example for analysis and explanation. Figure 2-Figure 3 , the specific implementation steps are as follows: Step 1: Integrate the sales data of refrigerators and freezers that have been sold online, static attribute data (including attributes such as door type, product positioning, and product life cycle), and online sales market change data of refrigerators and freezers purchased from third-party platforms: 1.1) According to the door type of refrigerators and freezers, SKUs are divided into several marketing categories: single door, double door, cross four door, multi-door, direct cooling two door, air cooling two door, direct cooling three door, air cooling three door, French four door, household cabinet, etc., and a data table as shown in Table 1 is generated: Table 1

[0064] 1.2) The total sales data of the marketing category is processed to generate a data table as shown in Table 2. The year-on-year sales data of the marketing category is mainly used to predict the total sales of the marketing category in the next month: Table 2

[0065] 1.3) Based on the weekly change data of the refrigerator online market purchased from the third-party platform, the data of the last two years is processed to generate the data table shown in Table 3. This data is used to determine the overall changes in the future sales market of each marketing category: Table 3

[0066] 1.4) Perform feature processing on the historical sales data of refrigerator and freezer product models, and integrate them into a feature data wide table based on the data generated in the first three steps, calculate the correlation coefficient between each feature and the target variable, and the importance index in the machine learning model, and select the feature table involved in the machine learning model, as shown in Table 4: Table 4

[0067] The above feature data includes the changing pattern of the sales share of SKUs in each month in the past six months, which is used to characterize the medium- and long-term sales share changing trend of SKUs; secondly, the above feature data also includes the short-term (such as the past 60 days) daily sales share changing trend of SKUs. In order to minimize the negative impact of the drastic ups and downs of daily granularity sales data on the algorithm model, the embodiment of the present application adopts a sliding average processing method, using 35 days as a fixed window sliding average to calculate the SKU sales share of the past 60 days, and finally obtains 26 consecutive average daily sales values.

[0068] The feature processing method of the embodiment of the present application is inspired by the idea of ​​self-attention mechanism. Similar to the self-attention mechanism, it has the characteristics of both long-term memory and short-term memory. The year-on-year data of the marketing category and the sales volume share change trend data of medium and long-term SKUs are introduced into the feature data to characterize the long-term memory data. At the same time, the sales share changes of short-term SKUs, that is, short-term memory data, are also taken into account. These two types of feature data are equally important in the prediction model.

[0069] Step 2: Use statistical learning methods to build a forecasting algorithm for the total sales of marketing categories in month M+1 on the 15th of month M. In view of the small sample size and large fluctuations in sales trends in this sales forecasting scenario, several SKUs are first combined into one marketing category, and then the total sales of several marketing categories with relatively more regular and predictable sales trends are determined, and statistical learning methods are used to obtain robust forecasting results with a small sample size.

[0070] 2.1) Count the total sales of each marketing category in the same period last year ( marketing categories) and the percentage of sales growth or decrease in each marketing category compared with last year, that is, the year-on-year change ratio of sales ( marketing categories).

[0071] 2.2) Calculate the year-on-year change rate of total sales volume of online channel marketing categories in each of the 8 consecutive weeks before the 15th of May = 1-8 weeks before the 15th.

[0072] 2.3) In the related art, as time goes by, the weight value of the exponential smoothing algorithm becomes smaller, so the farther away the data is, the smaller the impact on the result is. However, in the embodiment of the present application, the year-on-year total sales volume of the marketing category last year is crucial for predicting the total sales volume in the next month. Therefore, the year-on-year change ratio of the marketing category last year and the year-on-year change rate of the total sales volume of the marketing category market in the last 8 weeks (such as the last 60 days) contribute the same to the model: ; ; in, It is The forecast value of the year-on-year change in total sales volume of marketing categories M+1, is the weight parameter for year-on-year calculation, It is the forecast of the year-on-year change rate of total sales volume of the marketing category market in the last 8 weeks. The predicted value of the year-on-year change rate of marketing categories in M+1 months, is the year-on-year change value in the week before March 15th. is the predicted value of the year-on-year change rate in the week before March 15th. is the weight parameter for exponential smoothing.

[0073] 2.4) Finally, according to Get the predicted total sales of each marketing category for month M+1 .

[0074] Step 3: Use machine learning methods to build a prediction algorithm to predict the online sales of each product model in month M+1 on the 15th of month M.

[0075] 3.1) Construct an MLP model structure with two single Linear layers. The two Linear layers are used to fit the mid- to long-term sales volume share change trend of SKU and the short-term sales volume share change trend of SKU respectively. The calculation formula is: (two prediction tasks), are model parameters, It is the characteristic data of the input model, which comes from Table 4, including the characteristics of SKU's medium- and long-term sales share and SKU's short-term sales share; is the residual part, is the model output, including and ,in, It is the sales volume share of SKU in the medium and long term. It is the short-term sales ratio of the SKU.

[0076] 3.2) The output of the MLP model in the previous step is combined with the SKU static attribute features and input into the self-attention model structure in the deep learning model to examine the correlation between the sales share change at each time node and the prediction target. After passing through a Linear layer and a Softmax layer, the final sales share prediction result is obtained. .

[0077] 3.3) Finally, according to Calculate the final prediction result, that is, the predicted online sales volume of each product model in M+1 month (the next month). Due to the characteristics of the short product time series and small sample size mentioned above, the embodiment of the present application uses a Linear layer model structure that does not require high sample data volume and time series length when fitting the medium- and long-term and short-term sales volume share trends of SKUs. This structure can use the least computing resources and obtain better prediction results, and can avoid the overfitting problem of deep model structures in the case of small samples. In addition, when predicting sales, the embodiment of the present application jumps out of the framework of calculating in sequence according to the time series order, and draws on the idea of ​​self-attention in the sales forecasting scenario. At the same time, it examines the year-on-year change trend last year, the medium- and long-term change trend in the past 6 months, and the short-term change trend in the past 60 days, and ultimately improves the accuracy of sales forecasting.

[0078] In one embodiment, the historical sales data and static attribute data of refrigerator and freezer models, as well as the online market data of refrigerators and freezers are updated in the HIVE database every day. Data is obtained from the HIVE database and feature processed, and the processed data is stored in the Python cluster. Figure 2 The algorithm model shown in the figure uses the historical sales volume of each month in the previous 6 months as the model training target, and the data of the 24 months before the training target month as the model training set data source. Then the training set data is input into the model, and the historical sales volume of each month in the previous 6 months is used as the model prediction target to obtain the model parameters with the best prediction performance in the previous 6 months. The trained model is further used to predict the sales volume of each refrigerator and freezer product model in the online sales channel in the next month (such as month M+1), and the data shown in Table 5 is output: Table 5

[0079] Based on the data generated by intelligent prediction, business personnel formulate corresponding production and procurement plans for each product model, update the actual sales data in the next month, and evaluate the accuracy of the intelligent prediction results of the previous month.

[0080] The embodiment of the present application is aimed at the product sales forecasting scenario, and converts the sales forecasting task into the forecast of the total sales of a regular marketing category and the forecast of the sales share of the SKU in the marketing category to which it belongs. On the other hand, the idea of ​​self-attention in machine learning is borrowed in the forecasting method. Taking the online sales forecast of refrigerators and freezers as an example, the forecasting algorithm simultaneously examines the year-on-year sales situation last year and the medium- and long-term sales share in the past 6 months and the short-term sales share change trend in the past 60 days. The above technical solution can surpass the manual forecasting method in actual business use and meet the business use requirements. It can effectively help the business predict the sales trend in the next month, so as to better and more efficiently plan production plans, procurement plans and inventory management. At the same time, it can help enterprises better control costs, reduce inventory backlogs and waste, and improve the efficiency and flexibility of the supply chain. In addition, intelligent sales forecasting can also help enterprises better understand market demand, so as to better formulate sales strategies and promotion plans, and increase market share and sales.

[0081] Based on the above embodiments, reference Figure 4 , an embodiment of the present application proposes a product sales forecasting device, including an acquisition module 401, a first forecasting module 402, a second forecasting module 403 and a third forecasting module 404.

[0082] The acquisition module 401 is used to acquire historical sales information of multiple marketing categories of the product, and a historical sales ratio information set of each marketing category; the historical sales ratio information set of any marketing category includes historical sales ratio information of various product models in the marketing category; The first prediction module 402 is used to predict the total sales volume of each marketing category within a future set time period based on the historical sales volume information; The second prediction module 403 is used to predict the sales volume share of each product model in a future set time period according to each of the historical sales volume share information sets; The third prediction module 404 is used to predict the sales volume of the product model in a future set time period based on the total sales volume of each marketing category in a future set time period and the sales volume proportion of each product model in each marketing category in the future set time period.

[0083] The product sales forecasting device provided in the embodiment of the present application obtains the historical sales information of multiple marketing categories of the product, and the historical sales proportion information set of each marketing category; the historical sales proportion information set of any marketing category includes the historical sales proportion information of various product models in the marketing category; according to each historical sales information, the total sales of each marketing category in the future set time period is predicted; according to each historical sales proportion information set, the sales proportion of each product model in the future set time period is predicted; according to the total sales of each marketing category and the sales proportion of each product model in each marketing category, the sales of the product model in the future set time period is predicted. The present application first predicts a set of several product models, that is, the overall total sales of each marketing category in the future time period, and then predicts the sales proportion of the future time period according to the sales proportion change of a single product model in the marketing category to which it belongs, so as to realize product sales forecasting. At the same time, it can better and more efficiently plan production plans, procurement plans and inventory management, and can better control costs, reduce inventory backlogs and waste, and improve the efficiency and flexibility of the supply chain. In addition, intelligent sales forecasting can also help companies better understand market demand, so as to better formulate sales strategies and promotion plans, and increase market share and sales.

[0084] In one embodiment, the first prediction module 402 is specifically configured to: According to each of the historical sales information, obtain the year-on-year sales data of each of the marketing categories in the previous year; the year-on-year sales data includes the historical total sales volume and the year-on-year change ratio of the sales volume in the same period; Get the year-on-year change rate of total product market sales in the most recent consecutive historical weeks in the future set time period; Determine the predicted value of the year-on-year change in total sales volume of each marketing category within a future set time period based on the preset year-on-year calculation weight parameter, the year-on-year sales volume change ratio and the year-on-year change rate; Based on the historical total sales volume for the same period and the predicted value of the year-on-year change in the total sales volume, the total sales volume of each of the marketing categories in a set time period in the future is predicted.

[0085] In one embodiment, the first prediction module 402 is specifically configured to: Determine the year-on-year change value and the first year-on-year change rate forecast value of the target historical week according to the year-on-year change rate; Determine a second year-on-year change rate forecast value of the total market sales volume of each marketing category within a future set time period according to a preset exponential smoothing weight parameter, the year-on-year change value and the first year-on-year change rate forecast value; Based on the preset year-on-year calculation weight parameters, the year-on-year sales volume change ratio and the second year-on-year change rate forecast value, the forecast value of the year-on-year change in total sales volume of each marketing category within a future set time period is determined.

[0086] In one embodiment, the second prediction module 403 is specifically configured to: According to the historical sales volume share information, obtain the mid- to long-term sales volume share characteristics and the short-term sales volume share characteristics of each product model; Inputting the mid- to long-term sales volume share characteristics and the short-term sales volume share characteristics into a first prediction model, obtaining the mid- to long-term sales volume share and the short-term sales volume share of each product model output by the first prediction model; the first prediction model includes a plurality of linear layers, and the linear layers are used to predict the sales volume share of each product model; The static attribute characteristics of each product model, the mid- to long-term sales ratio and the short-term sales ratio are integrated to obtain integrated data; the static attribute characteristics at least include door body type, product positioning and product life cycle; The fused data is input into a second prediction model to obtain the sales share of each product model in a future set time period output by the second prediction model; the second prediction model includes a self-attention mechanism layer, and the self-attention mechanism layer is used to determine the relevant information between the sales share change at each time node and the prediction target.

[0087] In one embodiment, the linear layer includes a first linear layer and a second linear layer; the first linear layer is used to predict the medium- and long-term sales share according to the medium- and long-term sales share characteristics; the second linear layer is used to predict the short-term sales share according to the short-term sales share characteristics.

[0088] In one embodiment, the marketing category is determined based on: According to the product characteristic parameters of the product, the product models are divided to obtain a division result; According to the classification result, a plurality of the marketing categories are determined; any one of the marketing categories includes a plurality of product models with the same product characteristic parameters.

[0089] In one embodiment, the third prediction module 404 is specifically configured to: Calculate the product of the total sales volume of each marketing category in a future set time period and the sales volume proportion of each product model in each marketing category in a future set time period; The product is used as the sales volume of the product model in a future set time period.

[0090] Figure 5An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the following method: obtaining historical sales information of multiple marketing categories of the product, and a historical sales ratio information set of each marketing category; the historical sales ratio information set of any marketing category includes historical sales ratio information of various product models in the marketing category; predicting the total sales of each marketing category in a future set time period based on each of the historical sales information; predicting the sales ratio of each of the product models in a future set time period based on each of the historical sales ratio information sets; predicting the sales of the product model in a future set time period based on the total sales of each marketing category in a future set time period and the sales ratio of each of the product models in each of the marketing categories in a future set time period.

[0091] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0092] On the other hand, an embodiment of the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the product sales prediction method provided by the above-mentioned embodiments, for example, including: obtaining historical sales information of multiple marketing categories of products, and a historical sales share information set for each of the marketing categories; the historical sales share information set for any of the marketing categories includes historical sales share information of various product models in the marketing category; based on each of the historical sales information, predicting the total sales of each of the marketing categories in a future set time period; based on each of the historical sales share information sets, predicting the sales share of each of the product models in a future set time period; based on the total sales of each of the marketing categories in the future set time period, and the sales share of each of the product models in each of the marketing categories in the future set time period, predicting the sales of the product model in a future set time period.

[0093] On the other hand, an embodiment of the present application discloses a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the product sales prediction method provided by the above-mentioned method embodiments, for example, including: obtaining historical sales information of multiple marketing categories of products, and a historical sales share information set for each of the marketing categories; the historical sales share information set of any of the marketing categories includes historical sales share information of various product models in the marketing category; based on each of the historical sales information, predicting the total sales of each of the marketing categories in a future set time period; based on each of the historical sales share information sets, predicting the sales share of each of the product models in the future set time period; based on the total sales of each of the marketing categories in the future set time period, and the sales share of each of the product models in each of the marketing categories in the future set time period, predicting the sales of the product model in the future set time period.

[0094] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0095] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

[0097] The above implementation modes are only used to illustrate the present application, but not to limit the present application. Although the present application is described in detail with reference to the embodiments, a person skilled in the art should understand that various combinations, modifications or equivalent replacements of the technical solutions of the present application do not depart from the spirit and scope of the technical solutions of the present application, and should be included in the scope of the claims of the present application.

Claims

1. A product sales forecasting method, characterized in that: include: Obtain historical sales information of multiple marketing categories of the product, and a historical sales percentage information set of each marketing category; The historical sales volume share information set of any of the marketing categories includes the historical sales volume share information of various product models in the marketing category; Based on the historical sales information, predict the total sales of each marketing category within a future set time period; Based on each of the historical sales volume share information sets, predict the sales volume share of each of the product models within a set time period in the future; Based on the total sales of each of the marketing categories in the future set time period, and the sales proportion of each of the product models in each of the marketing categories in the future set time period, the sales of the product models in the future set time period are predicted.

2. The product sales forecasting method according to claim 1, characterized in that: The predicting of the total sales volume of each marketing category within a future set time period based on each of the historical sales volume information includes: According to each of the historical sales information, obtain the year-on-year sales data of each of the marketing categories in the previous year; the year-on-year sales data includes the historical total sales volume and the year-on-year change ratio of the sales volume in the same period; Get the year-on-year change rate of total product market sales in the most recent consecutive historical weeks in the future set time period; Determine the predicted value of the year-on-year change in total sales volume for each marketing category within a future set time period based on the preset year-on-year calculation weight parameters, the year-on-year sales volume change ratio and the year-on-year change rate; Based on the historical total sales volume for the same period and the predicted value of the year-on-year change in the total sales volume, the total sales volume of each of the marketing categories in a set time period in the future is predicted.

3. The product sales forecasting method according to claim 2, characterized in that: Determining the predicted value of the year-on-year change in total sales volume of each marketing category within a future set time period based on the preset year-on-year calculated weight parameter, the year-on-year change ratio of sales volume and the year-on-year change rate includes: Determine the year-on-year change value and the first year-on-year change rate forecast value of the target historical week according to the year-on-year change rate; Determine a second year-on-year change rate forecast value of the total market sales volume of each marketing category within a future set time period according to a preset exponential smoothing weight parameter, the year-on-year change value and the first year-on-year change rate forecast value; Based on the preset year-on-year calculation weight parameters, the year-on-year sales volume change ratio and the second year-on-year change rate forecast value, the forecast value of the year-on-year change in total sales volume of each marketing category within a future set time period is determined.

4. The product sales forecasting method according to claim 1, characterized in that: The predicting the sales volume share of each product model in a future set time period based on each of the historical sales volume share information sets includes: According to the historical sales volume share information, obtain the mid- to long-term sales volume share characteristics and the short-term sales volume share characteristics of each product model; Inputting the mid- to long-term sales volume share characteristics and the short-term sales volume share characteristics into a first prediction model, obtaining the mid- to long-term sales volume share and the short-term sales volume share of each product model output by the first prediction model; the first prediction model includes a plurality of linear layers, and the linear layers are used to predict the sales volume share of each product model; The static attribute characteristics of each product model, the mid- and long-term sales ratio, and the short-term sales ratio are integrated to obtain integrated data; the static attribute characteristics at least include door body type, product positioning, and product life cycle; The fused data is input into a second prediction model to obtain the sales share of each product model in a future set time period output by the second prediction model; the second prediction model includes a self-attention mechanism layer, and the self-attention mechanism layer is used to determine the relevant information between the sales share change at each time node and the prediction target.

5. The product sales forecasting method according to claim 4, characterized in that: The linear layer includes a first linear layer and a second linear layer; the first linear layer is used to predict the medium- and long-term sales share according to the medium- and long-term sales share characteristics; the second linear layer is used to predict the short-term sales share according to the short-term sales share characteristics.

6. The product sales forecasting method according to claim 1, characterized in that: The marketing categories are determined based on: According to the product characteristic parameters of the product, the product models are divided to obtain a division result; According to the classification result, a plurality of the marketing categories are determined; any one of the marketing categories includes a plurality of product models with the same product characteristic parameters.

7. The product sales forecasting method according to claim 1, characterized in that: The predicting of the sales volume of the product model in the future set time period according to the total sales volume of each marketing category in the future set time period and the sales volume proportion of each product model in each marketing category in the future set time period includes: Calculate the product of the total sales volume of each marketing category in a future set time period and the sales volume proportion of each product model in each marketing category in a future set time period; The product is used as the sales volume of the product model in a future set time period.

8. A product sales forecasting device, characterized in that: include: An acquisition module, used to acquire historical sales information of multiple marketing categories of a product, and a historical sales ratio information set of each marketing category; The historical sales volume share information set of any of the marketing categories includes the historical sales volume share information of various product models in the marketing category; A first prediction module, used to predict the total sales volume of each marketing category within a future set time period based on each of the historical sales volume information; A second prediction module is used to predict the sales share of each product model in a future set time period based on each of the historical sales share information sets; The third prediction module is used to predict the sales volume of the product model in a future set time period based on the total sales volume of each marketing category in the future set time period and the sales volume proportion of each product model in each marketing category in the future set time period.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the product sales forecasting method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the product sales forecasting method according to any one of claims 1 to 7 is implemented.

Images