Data processing method and device, equipment and storage medium

By combining historical sales data and economic data, using feature engineering and machine learning models to predict sales in the target period, the problem of poor prediction accuracy of sparse sales data in the existing technology is solved, and more accurate sales forecasts are achieved.

CN119027166AInactive Publication Date: 2024-11-26BEIJING JINGDONG QIANSHITECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310594986.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-24
Publication Date
2024-11-26
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art has poor accuracy when predicting the sales volume of products with sparse sales historical data, mainly because it only mines the historical sales data itself and does not fully utilize relevant economic data.

Method used

By obtaining the historical sales data and historical economic data of the prediction object, using feature engineering and classification models, generating feature engineering features, and using stack autoencoder and density clustering methods for dimensionality reduction and clustering, training classification and prediction models, and predicting targets Whether sales and specific sales volumes occur during the period.

Benefits of technology

The accuracy of sales forecasting of sparse sales historical data products has been improved, especially in intermittent demand scenarios, which significantly improves the accuracy and reliability of forecast results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027166A_ABST
    Figure CN119027166A_ABST
Patent Text Reader

Abstract

The invention provides a data processing method and device, equipment and a storage medium, and relates to the technical field of computers. The method comprises the following steps: acquiring historical sales volume data of a prediction object, wherein the historical sales volume data of the prediction object comprises sales volume of the prediction object at a plurality of time points in a preset time period before a target period; acquiring historical economic data of a plurality of time points in a preset time period before the target period; obtaining feature engineering features of the predicted object through feature engineering according to the historical sales data of the predicted object and the historical economic data of the corresponding time; predicting whether the sales of the prediction object occurs in the target period according to the feature engineering features of the prediction object through a classification model, and obtaining a prediction result; and if the prediction result is that the prediction object is sold in the target period, predicting the sales volume of the prediction object in the target period through a prediction model according to the prediction result and the feature engineering features of the prediction object. According to the method, the product sales volume prediction accuracy is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of computers, and in particular, to a data processing method, device, electronic device, and readable storage medium. Background Art

[0002] In some industries, product sales often show intermittent demand. The demand for products is zero in many periods, and the demand quantity is unstable during the period when the products are sold. Therefore, the historical sales data of such products are relatively sparse. Due to the lag of the supply chain, it is usually necessary to predict product sales in the future target period in order to formulate a sales plan as accurate as possible, and then formulate production and procurement plans based on the sales plan. When predicting the sales of products in the target period, the relevant technology only mines the historical sales data of the products themselves, resulting in poor accuracy in sales prediction for products with sparse historical sales data.

[0003] As mentioned above, how to improve the accuracy of sales forecasting for products with sparse historical sales data has become an urgent problem to be solved.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0005] The purpose of the present disclosure is to provide a data processing method, device, electronic device and readable storage medium, which can at least to some extent overcome the problem that the sales prediction accuracy of products with sparse sales history data is poor due to the fact that the relevant technology only mines the product sales history data itself.

[0006] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0007] According to one aspect of the present disclosure, a data processing method is provided, including: obtaining historical sales data of a prediction object, the historical sales data of the prediction object including the sales of the prediction object at multiple time points within a preset time period before a target period; obtaining historical economic data at multiple time points within a preset time period before the target period; obtaining feature engineering features of the prediction object based on the historical sales data of the prediction object and the historical economic data of the corresponding time through feature engineering; predicting whether sales of the prediction object will occur in the target period based on the feature engineering features of the prediction object through a classification model to obtain a prediction result; if the prediction result is that sales of the prediction object will occur in the target period, predicting the sales of the prediction object in the target period through a prediction model based on the prediction result and the feature engineering features of the prediction object.

[0008] According to an embodiment of the present disclosure, feature engineering features of the prediction object are obtained based on the historical sales data of the prediction object and the historical economic data of the corresponding time through feature engineering, including: using a sliding window to obtain target sales index values ​​at multiple preset time points within the preset time period based on the historical sales data of the prediction object, to obtain sales features; using the sliding window to obtain target economic data index values ​​at multiple preset time points based on the historical economic data of the corresponding time, to obtain economic features; splicing the sales features and the economic features according to the multiple preset time points to obtain the feature engineering features of the prediction object.

[0009] According to an embodiment of the present disclosure, feature engineering features of the prediction object are obtained based on the historical sales data of the prediction object and the historical economic data of the corresponding time through feature engineering, and also include: marking the multiple preset time points with holidays to obtain date features; splicing the sales features and the economic features according to the multiple preset time points to obtain the feature engineering features of the prediction object, including: splicing the sales features, the economic features and the date features according to the multiple preset time points to obtain the feature engineering features of the prediction object.

[0010] According to an embodiment of the present disclosure, the method further includes: obtaining historical sales data of multiple training objects, the historical sales data of each training object among the multiple training objects including the sales of the training object at multiple time points within a preset time period before a preset period; for each training object, using a stacked autoencoder to obtain a neural network output of the training object based on the historical sales data of the training object; using a density clustering method to cluster the multiple training objects based on the neural network output of each training object to obtain a clustering result; and training an initial classification model and an initial prediction model based on the clustering results to obtain the classification model and the prediction model.

[0011] According to an embodiment of the present disclosure, the method also includes: obtaining historical economic data of multiple time points within a preset time period before the preset period; for each of the training objects, obtaining the feature engineering features of the training object according to the historical sales data of the training object and the historical economic data of the corresponding time through the feature engineering; for each of the training objects, using a stacked autoencoder to obtain the dimensionality reduction features of the training object according to the historical sales data of the training object; based on the clustering results, respectively training the initial classification model and the initial prediction model to obtain the classification model and the prediction model, including: taking the feature engineering features and dimensionality reduction features of each training object as input features, taking the clustering results as labels, training the initial classification model, and obtaining the classification model.

[0012] According to an embodiment of the present disclosure, the initial classification model and the initial prediction model are trained respectively based on the clustering results to obtain the classification model and the prediction model, and also include: for each of the training objects, predicting the probability of sales of the training object in the preset period according to the feature engineering features and dimensionality reduction features of the training object by the classification model to obtain the probability features of each training object; determining multiple target training objects from the multiple training objects according to the probability features of the each training object; taking the feature engineering features, dimensionality reduction features and probability features of each target training object among the multiple target training objects as input features, taking the sales volume of each target training object in the preset period as a label, training the initial prediction model to obtain the prediction model.

[0013] According to an embodiment of the present disclosure, the method further includes: obtaining the feature importance of feature engineering features, dimensionality reduction features and probability features respectively based on the prediction model.

[0014] According to an embodiment of the present disclosure, the method also includes: using a stacked autoencoder to obtain the dimensionality reduction features of the prediction object based on the historical sales data of the prediction object; the prediction result includes the probability features of the prediction object; predicting whether the sales of the prediction object will occur in the target period based on the feature engineering features of the prediction object through a classification model to obtain the prediction result, including: predicting the probability of the sales of the prediction object occurring in the target period based on the feature engineering features and dimensionality reduction features of the prediction object through the classification model to obtain the probability features of the prediction object; if the prediction result is that the sales of the prediction object will occur in the target period, predicting the sales of the prediction object in the target period based on the prediction result and the feature engineering features of the prediction object through the prediction model, including: if the probability feature of the prediction object is greater than the prediction probability threshold, predicting the sales of the prediction object in the target period based on the probability features, feature engineering features and dimensionality reduction features of the prediction object through the prediction model.

[0015] According to an embodiment of the present disclosure, the historical economic data includes historical cost data, historical price data and historical capital data.

[0016] According to another aspect of the present disclosure, there is provided a data processing device, comprising: an acquisition module, for acquiring historical sales data of a prediction object, the historical sales data of the prediction object including the sales of the prediction object at multiple time points within a preset time period before a target period; the acquisition module, further for acquiring historical economic data at multiple time points within a preset time period before the target period; a feature engineering module, for obtaining feature engineering features of the prediction object according to the historical sales data of the prediction object and the historical economic data of the corresponding time through feature engineering; a classification module, for predicting whether sales of the prediction object will occur in the target period according to the feature engineering features of the prediction object through a classification model to obtain a prediction result; and a prediction module, for predicting the sales of the prediction object in the target period through a prediction model according to the prediction result and the feature engineering features of the prediction object if the prediction result is that the sales of the prediction object will occur in the target period.

[0017] According to another aspect of the present disclosure, an electronic device is provided, comprising: a memory, a processor, and executable instructions stored in the memory and executable in the processor, wherein the processor implements any of the above methods when executing the executable instructions.

[0018] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which computer-executable instructions are stored, and when the executable instructions are executed by a processor, any of the above methods is implemented.

[0019] The data processing method provided by the embodiments of the present disclosure obtains feature engineering features of the prediction object based on historical sales data including sales of the prediction object at multiple time points within a preset time period before the target period and historical economic data of the corresponding time through feature engineering, and then uses a classification model to predict whether the sales of the prediction object will occur in the target period based on the feature engineering features of the prediction object. If the prediction result is that the sales of the prediction object will occur in the target period, the prediction model is used to predict the sales of the prediction object in the target period based on the prediction result and the feature engineering features of the prediction object, thereby improving the accuracy of sales prediction for products with sparse historical sales data.

[0020] It is to be understood that the foregoing general description and the following detailed description are exemplary only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above and other objects, features and advantages of the present disclosure will become more apparent by describing in detail example embodiments thereof with reference to the attached drawings.

[0022] Figure 1 An exemplary diagram showing a comparison of sales statistics between intermittent demand and regular demand is shown.

[0023] Figure 2 A schematic diagram showing a system structure in an embodiment of the present disclosure.

[0024] Figure 3 A flow chart of a data processing method in an embodiment of the present disclosure is shown.

[0025] Figure 4 Shows Figure 3 FIG. 5 is a schematic diagram of the processing process of step S306 in one embodiment.

[0026] Figure 5 A schematic diagram of a sliding window operation is exemplarily shown.

[0027] Figure 6 Shows Figure 3 FIG. 5 is a schematic diagram of the processing process of step S306 in another embodiment.

[0028] Figure 7 is based on Figure 3 A flowchart of a model training method is shown.

[0029] Figure 8 is based on Figure 7 A flowchart of a classification model training method is shown.

[0030] Fig. 9 is based on Figure 7 and Figure 8A flowchart of a prediction model training method is shown.

[0031] Fig.10 The figure is a diagram showing the result of feature importance according to an exemplary embodiment.

[0032] Fig.11 is based on Figures 3 to 9 A flowchart of another data processing method is shown.

[0033] Fig.12 A sales volume forecasting business logic schematic diagram is shown according to an exemplary embodiment.

[0034] Fig.13 is based on Figures 3 to 11 A flowchart of product sales forecasting is shown.

[0035] Fig.14 A block diagram of a data processing device in an embodiment of the present disclosure is shown.

[0036] Fig.15 A block diagram showing another data processing device in an embodiment of the present disclosure.

[0037] Fig.16 A schematic structural diagram of an electronic device in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0038] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The accompanying drawings are only schematic illustrations of the disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted.

[0039] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, devices, steps, etc. may be adopted. In other cases, well-known structures, methods, devices, implementations or operations are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0040] In addition, the terms "first", "second", etc. are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present disclosure, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. The symbol " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0041] In this disclosure, unless otherwise clearly specified and limited, terms such as "connection" should be understood in a broad sense, for example, it can be an electrical connection or can communicate with each other; it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in this disclosure can be understood according to specific circumstances.

[0042] The terms involved in the embodiments of the present disclosure are explained below.

[0043] SMAPE: Accuracy index, the full name is Symmetric Mean Absolute Percentage Error, the formula is as follows:

[0044]

[0045] FFORMA: The second prize method of the 2020 M4 competition. The full name of the method is Feature-based forecast model averaging. It integrates multiple time series models to output the final forecast results. It is a method with high prediction accuracy and generalization.

[0046] PPI: full name is Producer Price Index, producer price index.

[0047] CPI: full name is Consumer Price Index, consumer price index.

[0048] ROC: The full name is Receiver Operating characteristic Curve, receiver operating characteristic curve.

[0049] AUC: The full name is Area Under Curve, which is defined as the area under the ROC curve and the coordinate axis. The value is generally between (0.5, 1). The larger the AUC, the better the quality of the ROC curve.

[0050] AAC (Accuracy): also known as accuracy, is a commonly used indicator to measure classifier performance. It indicates the proportion of correct predictions made by a model on a test data set.

[0051] As described above, the sales volume of products in some industry fields usually presents an intermittent demand state, such as slow-moving retail goods (sales volume is high only during promotional activities), auto spare parts in the automotive industry, etc. The embodiments of the present disclosure are described using the product (object) being auto spare parts as an example, but are not limited to this.

[0052] In recent years, with the advancement of the digitalization of the supply chain in the automotive spare parts industry, demand forecasting is particularly important as the first step in supply chain planning. There are many methods for sales forecasting based on time series data in related technologies. The most basic ones include simple moving averages, and more complex ones include some time series models, such as the Autoregressive Integrated Moving Average model (ARIMA), etc. There are also some machine learning models, deep learning models, etc. Many types of time series models have shown different effects in different application scenarios. These models usually focus more on mining the historical sales data itself. For example, in order to predict sales in January 2023, the above-mentioned related models will fully mine sales before December 2022. Due to the characteristics of the automotive spare parts industry, sales are usually sparse. There are also some methods that can predict intermittent periods and dynamic sales of time series data, such as the croston series.

[0053] The effect of using relevant technologies to predict the sales of auto spare parts needs to be improved. The sales of auto spare parts usually show an intermittent demand state. For example, due to the particularity of the industry, its customers usually purchase in a centralized manner. Therefore, 80% of the spare parts of the auto company usually only have less than 5 sales in a year. This intermittent demand usually presents small batch / block demand and intermittent demand state. Figure 1 The following is an example of a sales statistics comparison diagram of intermittent demand and regular demand. Figure 1 As shown, unstable demand 102 and stable demand 104 are regular, continuous demand states, and small batch / block demand 106 and intermittent demand 108 are intermittent demand states. Figure 1 As shown in the figure, from the vertical perspective, the CV square (Squared Coefficient of Variation, the ratio of the standard deviation to the mean) gradually increases from bottom to top, with the median value being 0.49; from the horizontal perspective, the ADI (Average inter-Demand Interval, the average demand time between two demands) gradually increases from left to right, with the median value being 1.32.

[0054] Related technologies have difficulty achieving good results in intermittent prediction scenarios, partly because these methods are only for historical sales data. Auto parts sales data is sparse and is zero in most periods, so it does not contain enough information. Another reason is that the prediction methods for sparse sales scenarios in related technologies focus more on predicting average sales over a period of time. When sales are mostly zero, using conventional accuracy indicators such as 1-smape usually results in negative values.

[0055] Therefore, the present disclosure provides a data processing method, which combines the historical sales data of the forecast object and related public economic data, adopts a two-stage model of classification and forecasting, predicts whether the forecast object will be sold in the target period, and predicts the specific sales volume more accurately.

[0056] Figure 2 An exemplary system architecture 20 to which the data processing method or data processing apparatus of the present disclosure can be applied is shown.

[0057] like Figure 2 As shown, the system architecture 20 may include a terminal device 202, a network 204, and a server 206. The terminal device 202 may be a variety of electronic devices having a display screen and supporting input and output, including but not limited to a smart phone, a tablet computer, a laptop portable computer, a desktop computer, a wearable device, a virtual reality device, a smart home, etc. The network 204 is used to provide a medium for a communication link between the terminal device 202 and the server 206. The network 204 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc. The server 206 may be a server or server cluster that provides various services, such as a processing server, a network server, a database server, etc.

[0058] The user can use the terminal device 202 to interact with the server 206 through the network 204 to receive or send data, etc. For example, the user can operate on the terminal device 202 to upload the historical sales data of the predicted object and the historical economic data of the corresponding time to the server 206 through the network 204 for feature engineering processing. For another example, the user can obtain the feature engineering features of the predicted object according to the historical sales data of the predicted object and the historical economic data of the corresponding time through feature engineering through the software on the terminal device 202. For another example, after the server 206 predicts whether the sales of the predicted object will occur in the target period according to the feature engineering features of the predicted object through the classification model, it obtains the prediction result, and predicts the sales of the predicted object in the target period through the prediction model according to the prediction result and the feature engineering features of the predicted object, and then sends the predicted sales to the terminal device 202 through the network 204 for the user to view.

[0059] It should be understood that Figure 2 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0060] Figure 3 FIG. 1 is a flow chart of a data processing method according to an exemplary embodiment. Figure 3 The method shown can be applied to the server side of the above system, for example, and can also be applied to the terminal device of the above system.

[0061] refer to Figure 3 , the method 30 provided in the embodiment of the present disclosure may include the following steps S302 to S312.

[0062] In step S302, historical sales data of the forecast object is obtained, where the historical sales data of the forecast object includes sales of the forecast object at multiple time points within a preset time period before the target period.

[0063] In some embodiments, the prediction object may be a product or commodity whose sales volume is to be predicted, such as a spare part for an automobile. The prediction object may be represented by a minimum stock keeping unit (SKU).

[0064] In some embodiments, the preset time period may be one year, two years, three years, etc. For example, if the preset time period is one year, and the target period is April 2023, the preset time period is from April 2022 to March 2023. The multiple time points may be every half month, every month, every two months, etc. For example, if the multiple time points are every month, the target period is April 2023, and the preset time period is from April 2022 to March 2023, the sales volume of the predicted object at multiple time points in the preset time period before the target period is the sales volume of the predicted object in April, May, June, ... December 2022, January, February, and March 2023.

[0065] In some embodiments, the features may also use more time series features, such as sales duration, frequency, trend, etc. The sales data disclosed in the present invention is not limited to the sales volume itself.

[0066] In step S304, historical economic data at multiple time points within a preset time period before the target period is obtained.

[0067] In some embodiments, historical economic data may include historical cost data, historical price data, and historical capital data.

[0068] Taking automobile spare parts as an example, based on existing academic research and business experience, the macroeconomic data related to the purchasing behavior of automobile spare parts dealers are as follows in Table 1:

[0069] Table 1

[0070]

[0071]

[0072] These data have official public data sources and are also summarized in common public financial databases. The demand side of auto spare parts is spare parts dealers. The data categories in Table 1 are defined from the perspective of dealers. Spare parts sales are affected by the inventory strategy of spare parts buyers. The specific business experience logic can be referred to Fig.12 .

[0073] The disclosed embodiment only selects part of the data based on business experience for cost considerations, but is not limited thereto. Consideration can be given to introducing a large amount of public economic data to automatically evaluate the importance of external data without considering performance.

[0074] In step S306, feature engineering features of the prediction object are obtained based on the historical sales data of the prediction object and the historical economic data of the corresponding time through feature engineering.

[0075] In some embodiments, a sliding window can be used to extract sales characteristics and economic characteristics from historical sales data and historical economic data, respectively. For specific implementation methods, please refer to Figure 4 and Figure 5 .

[0076] In some embodiments, holidays can also be marked to obtain date features within a preset time period, and the date features are spliced ​​with sales features and economic features according to the date to obtain feature engineering features. For specific implementation methods, please refer to Figure 6 .

[0077] In step S308, whether sales of the predicted object will occur in the target period is predicted according to the feature engineering features of the predicted object through the classification model to obtain a prediction result.

[0078] In some embodiments, the initial classification model can be trained using the training object to obtain a trained classification model. For specific implementations, see Figure 7 and Figure 8 .

[0079] In some embodiments, the Figure 7 and Figure 8 The feature processing method used during training creates the features of the prediction object and inputs them into the trained classification model for prediction. The specific implementation method can be found in Fig.11 .

[0080] In step S310, if the prediction result is that the sales of the predicted object occur in the target period, the sales volume of the predicted object in the target period is predicted by the prediction model based on the prediction result and the feature engineering features of the predicted object.

[0081] In some embodiments, the initial prediction model can be trained using the training object to obtain a trained prediction model. For specific implementations, please refer to Fig. 9 .

[0082] In some embodiments, in some embodiments, the Fig. 9 The feature processing method used during training creates the features of the prediction object and inputs them into the trained prediction model for sales prediction. For specific implementation methods, please refer to Fig.11 .

[0083] In step S312, if the forecast result is that the sales of the forecast object will not occur in the target period, the process ends. Only the forecast result that the sales of the forecast object will not occur in the target period may be output.

[0084] According to the data processing method provided by the embodiment of the present disclosure, feature engineering is used to obtain the feature engineering features of the prediction object based on the historical sales data including the sales of the prediction object at multiple time points within a preset time period before the target period and the historical economic data of the corresponding time. Then, a classification model is used to predict whether the sales of the prediction object will occur in the target period based on the feature engineering features of the prediction object. If the prediction result is that the sales of the prediction object will occur in the target period, the prediction model is used to predict the sales of the prediction object in the target period based on the prediction result and the feature engineering features of the prediction object. By combining the historical sales data of the prediction object and the relevant public economic data, a two-stage model of classification and prediction is used to predict whether the prediction object will be sold in the target period and predict the specific sales volume, thereby improving the accuracy of sales prediction for products with sparse historical sales data.

[0085] Figure 4 Shows Figure 3 The processing process of step S306 in one embodiment is shown in FIG. Figure 4 As shown, in the embodiment of the present disclosure, the above step S306 may further include the following steps.

[0086] Step S402, using a sliding window to obtain target sales index values ​​at multiple preset time points within a preset time period based on historical sales data of the prediction object, and obtaining sales characteristics.

[0087] In some embodiments, the multiple preset time points within the preset time period can be 1 week, 2 weeks, 1 month, 3 months, half a year, or a year before the target period, or can be evenly distributed as monthly, every two months, etc., and can be adjusted according to actual conditions, and this disclosure does not limit it.

[0088] In some embodiments, the target sales index value can be a weekly average, a 5-day average, a week and a half average, etc., or a weekly median, a 5-day median, a week and a half median, etc. It can be adjusted according to actual conditions and is not limited in this disclosure.

[0089] Step S404, using a sliding window to obtain target economic data index values ​​at multiple preset time points according to historical economic data at corresponding times, and obtaining economic characteristics.

[0090] In some embodiments, the release time intervals of economic data are different, and the data date granularity is different. Futures and stock-related data include the closing, opening, highest, lowest, and price fluctuations of each trading day. During the data processing stage, the two fields of closing prices on Mondays can be filtered. Most of the macro data are released monthly. When performing feature engineering of economic data, the weekly granularity can also be used as the basic granularity, and the missing values ​​can be filled with the most recent historical data. A sliding window can also be used to take multiple preset time points corresponding to the historical sales data, such as the weekly average of 1 week, 2 weeks, 1 month, 3 months, half a year, and one year before the target period.

[0091] Step S406, concatenating the sales volume characteristics and the economic characteristics according to a plurality of preset time points to obtain feature engineering characteristics of the prediction object.

[0092] According to the method provided in the embodiments of the present disclosure, the sales features and economic features produced by the sliding window can be used to extract the weekly averages of one week, two weeks, one month, three months, half a year, and one year before the target period. This extraction method can summarize the main information such as the mean variance of a sequence to a certain extent, and pay more attention to recent information. The features obtained in this way are used in classification models and prediction models to improve the accuracy of sales forecast results.

[0093] Figure 5 A schematic diagram of a sliding window operation is shown as an example. Figure 5 As shown, the horizontal axis represents time in days, with 250 days as the current period, and the vertical axis represents the indicator value. Curve 502 represents real historical data, curve 504 represents predicted future data, and 506 is the weekly average of each preset time point obtained by the sliding window operation.

[0094] Figure 6 Shows Figure 3 FIG. 5 is a schematic diagram of the processing process of step S306 in another embodiment. Figure 6 and Figure 4 The difference is that the date feature is added in the feature engineering process. Figure 6 As shown, in the embodiment of the present disclosure, the above step S306 may further include the following steps S602 to S608.

[0095] Step S602, using a sliding window to obtain target sales index values ​​at multiple preset time points within a preset time period based on historical sales data of the prediction object, and obtaining sales characteristics.

[0096] In some embodiments, the specific implementation of step S602 may refer to step S402.

[0097] Step S604, using a sliding window to obtain target economic data index values ​​at multiple preset time points based on historical economic data at corresponding times, and obtaining economic characteristics.

[0098] In some embodiments, the specific implementation of step S604 may refer to step S404.

[0099] Step S606: Mark multiple preset time points with holidays to obtain date features.

[0100] In some embodiments, the date marking may include month, quarter, Spring Festival and other holiday markings.

[0101] Step S608, concatenating the sales volume features, economic features and date features according to multiple preset time points to obtain feature engineering features of the prediction object.

[0102] According to the method provided in the embodiment of the present disclosure, sales features and economic features are produced using a sliding window, and dates are marked to produce date features. The date features are then concatenated with the sales features and economic features according to the dates to complete feature engineering, thereby obtaining feature engineering features for classification models and prediction models, which can improve the accuracy of sales forecast results.

[0103] Figure 7 is based on Figure 3 A flowchart of a model training method is shown. Figure 7 , the method 70 provided in the embodiment of the present disclosure may include the following steps S702 to S708.

[0104] In step S702, historical sales data of multiple training objects are obtained, where the historical sales data of each training object in the multiple training objects includes the sales of the training object at multiple time points within a preset time period before a preset period.

[0105] In some embodiments, the time series data of historical sales can be adjusted into data that can be used for supervised learning training. The historical sales data of one SKU can include multiple periods. One period can be used as a training sample, and the sales in that period is the prediction target y of the training set.

[0106] In step S704, for each training object, a stacked autoencoder is used to obtain the neural network output of the training object according to the historical sales data of the training object.

[0107] In some embodiments, a stacked autoencoder can be used to build a network according to the period of SKU sales data, and the number of neurons in the middle layer of the autoencoder is set to 1 / 16 of the input, and the output of the middle layer is extracted as a dimensionality reduction feature. The number of layers of the neural network can be, for example, 5 layers (i.e., the number of neurons in each layer is one-fourth of the previous layer), the activation function can be selected as tanh, the number of training epochs can be determined based on the error after decoding, and the total output of the 5-layer network is the output of the neural network.

[0108] In step S706, a density clustering method is used to cluster the multiple training objects according to the neural network output of each training object to obtain a clustering result.

[0109] In some embodiments, a density clustering method can be used for preliminary clustering, and the silhouette coefficient can be used as an optimization indicator to search for the minimum radius and the minimum number of samples within the radius in the parameter space to obtain model categories and a small number of outliers.

[0110] In some embodiments, clustering categories can be obtained in the dimensionality reduction space, and similar ones will be clustered into one category, for example, sales with relatively high continuous fluctuations are grouped into one category, and sales with relatively low volatility are grouped into another category.

[0111] In step S708, the initial classification model and the initial prediction model are trained respectively based on the clustering results to obtain a classification model and a prediction model.

[0112] In some embodiments, the feature engineering features and dimensionality reduction features of each training object can be used as input features, and the clustering results can be used as labels to train the initial classification model. For specific implementation methods, please refer to Figure 8 .

[0113] In some embodiments, training objects predicted to have dynamic sales can be selected to train the initial prediction model. For specific implementations, please refer to Fig. 9 .

[0114] According to the method provided in the embodiment of the present disclosure, the training objects are preliminarily classified through an unsupervised learning method, namely, a density clustering method, as category labels to train the initial classification model, and similar SKUs are clustered into one category, so as to fully explore the relationship between sales volume and historical sales volume and economic data of the same period, thereby improving the accuracy of the classification model and further improving the accuracy of sales forecasts.

[0115] Figure 8 is based on Figure 7 A flowchart of a classification model training method is shown. Figure 8 and Figure 7 The difference is that Figure 8 Describes the training of an initial classification model. Figure 8 , the method 80 provided in the embodiment of the present disclosure may include the following steps S802 to S814.

[0116] In step S802, historical sales data of multiple training objects are obtained, where the historical sales data of each training object in the multiple training objects includes the sales of the training object at multiple time points within a preset time period before a preset period.

[0117] In some embodiments, the specific implementation of step S802 may refer to step S702.

[0118] In step S804, for each training object, a stacked autoencoder is used to obtain the neural network output of the training object according to the historical sales data of the training object.

[0119] In some embodiments, the specific implementation of step S804 may refer to step S704.

[0120] In step S806, a density clustering method is used to cluster the multiple training objects according to the neural network output of each training object to obtain a clustering result.

[0121] In some embodiments, the specific implementation of step S806 may refer to step S706.

[0122] In step S808, historical economic data of multiple time points within a preset time period before a preset period is obtained.

[0123] In some embodiments, the preset period is similar to the target period, and the specific implementation of step S808 can refer to step S304.

[0124] In step S810, for each training object, feature engineering features of the training object are obtained based on the historical sales data of the training object and the historical economic data of the corresponding time through feature engineering.

[0125] In some embodiments, the processing of the training object is similar to the prediction object. The specific implementation of step S810 can refer to Figure 4 and Figure 5 .

[0126] In step S812, for each training object, a stacked autoencoder is used to obtain the dimension reduction features of the training object according to the historical sales data of the training object.

[0127] In some embodiments, the specific implementation of step S812 may refer to step S704.

[0128] In step S814, the feature engineering features and dimensionality reduction features of each training object are used as input features, and the clustering results are used as labels to train an initial classification model to obtain a classification model.

[0129] In some embodiments, the initial classification model may be an XGboost (xgb) binary classification model. The input training features may include sales volume features, economic features, date features, and dimension reduction features, and the clustering results are labels.

[0130] According to the method provided in the embodiment of the present disclosure, by inputting the feature engineering features and autoencoding dimensionality reduction features of the training object and using the density clustering results as category labels to train the initial classification model, the accuracy of the classification model can be improved, thereby improving the accuracy of sales forecasts.

[0131] Fig. 9 is based on Figure 7 and Figure 8 A flowchart of a prediction model training method is shown. Fig. 9 and Figure 8 The connection is that in Fig. 9 Adopted in Figure 8 The input results of the trained classification model are used to train the initial prediction model. Fig. 9 , the method 90 provided in the embodiment of the present disclosure may include the following steps S902 to S906.

[0132] In step S902, for each training object, the probability of sales of the training object occurring in a preset period is predicted by a classification model according to the feature engineering features and dimensionality reduction features of the training object, thereby obtaining the probability features of each training object.

[0133] In some embodiments, the probability feature of the training object can be input into the feature engineering feature and the dimensionality reduction feature. Figure 8 After the classification model is trained, the output is the probability of sales of the training object occurring in the preset period.

[0134] In step S904, a plurality of target training objects are determined from the plurality of training objects according to the probability characteristics of the respective training objects.

[0135] In some embodiments, a probability threshold may be preset, such as 0.4, 0.5, or 0.6, etc. The probability threshold may be used to determine whether a SKU will be actively sold in the next period, and only training samples (training objects) predicted to be actively sold are used to train the initial prediction model.

[0136] In step S906, feature engineering features, dimensionality reduction features and probability features of each target training object among multiple target training objects are used as input features, and the sales volume of each target training object in a preset period is used as a label to train an initial prediction model and obtain a prediction model.

[0137] In some embodiments, the initial prediction model may be an XGBoost regression model.

[0138] In some embodiments, the feature importance of feature engineering features, dimensionality reduction features, and probability features can be obtained based on the prediction model. Taking the initial prediction model as the XGBoost regression model as an example, the feature to be selected as the split point can be calculated based on the gain of the structure score, and the importance of a feature is the sum of the number of times it appears in all trees. In other words, the more an attribute is used to build decision trees in the model, the higher its importance is. An exemplary feature importance output result can be referred to Fig.10 .

[0139] According to the method provided in the embodiment of the present disclosure, the initial prediction model is trained by using samples with active sales predicted by the trained classification model, and the sales characteristics, economic characteristics, date characteristics, dimensionality reduction characteristics and probability characteristics of these samples are input into the initial prediction model for training, which can improve the accuracy of the prediction model and thus improve the accuracy of sales forecasts.

[0140] Fig.10 is a feature importance result display diagram according to an exemplary embodiment, using the information gain evaluation criterion to evaluate feature importance. Fig.10 As shown, the most important features 1002 first include the predicted sales average of the SKU in 3 months, half a year, and one year; followed by capital cost data (such as stock index, shibor interest rate, etc.), raw material futures related, etc. The output of feature importance makes the embodiment of the disclosure interpretable, and business personnel and related customers have a higher acceptance of the model of the embodiment of the disclosure.

[0141] Fig.11 is based on Figures 3 to 9 A flowchart of another data processing method is shown. Fig.11 Combined with Figures 4 to 9 method, Figure 3 Describe. Fig.11 , the method 110 provided in the embodiment of the present disclosure may include the following steps S1102 to S1116.

[0142] In step S1102, historical sales data of the prediction object is obtained.

[0143] In some embodiments, the specific implementation of step S1102 may refer to step S302.

[0144] In step S1104, a stacked autoencoder is used to obtain the dimensionality reduction features of the prediction object according to the historical sales data of the prediction object.

[0145] In some embodiments, the processing of the predicted object is similar to the training object, and the specific implementation of step S1104 can refer to step S812.

[0146] In step S1106, historical economic data at multiple time points within a preset time period before the target period is obtained.

[0147] In some embodiments, the specific implementation of step S1106 may refer to step S304.

[0148] In step S1108, feature engineering features of the prediction object are obtained based on the historical sales data of the prediction object and the historical economic data of the corresponding time through feature engineering.

[0149] In some embodiments, the specific implementation of step S1108 may refer to step S306.

[0150] In step S1110, the probability of sales of the predicted object occurring in the target period is predicted according to the feature engineering features and dimensionality reduction features of the predicted object through the classification model to obtain the probability features of the predicted object.

[0151] In some embodiments, the processing of the predicted object is similar to the training object, and the specific implementation of step S1110 can refer to step S902.

[0152] In step S1112, it is determined whether the probability feature of the prediction object is greater than the prediction probability threshold.

[0153] In some embodiments, the prediction probability threshold may be 0.4, or 0.5, or 0.6, and so on.

[0154] In step S1114, if the probability feature of the predicted object is greater than the predicted probability threshold, the sales volume of the predicted object in the target period is predicted by the prediction model according to the probability feature, feature engineering feature and dimensionality reduction feature of the predicted object.

[0155] In step S1116, if the probability feature of the predicted object is not greater than the predicted probability threshold, the process ends.

[0156] Fig.12 According to an exemplary embodiment, a sales forecasting business logic diagram is shown. Fig.12 As shown, taking automobile spare parts as an example, the sales volume 1202 of automobile spare parts depends on the purchase and inventory decisions of dealers, and the inventory plan 1204 of dealers depends on the return on investment 1208 of purchasing spare parts and sales. The return on investment 1208 will be affected by upstream costs 1210, consumer-side prices 1214, and capital costs 1218. In the static profit 1212 of the purchase price 1214 of spare parts minus the cost 1210, if the capital cost 1218 is lower (that is, the return on investment of the purchase behavior is higher), the purchaser is more likely to purchase. Therefore, the capital data 1216 and price data 1214 in the economic data can be used to decide whether the dealer will purchase in the next period, and how much to purchase, that is, the operating data 1206.

[0157] Fig.13 is based on Figures 3 to 11 A flowchart of product sales forecasting is shown. Fig.13 The training and actual prediction processes are described comprehensively. Fig.13 The process shown may include the following steps S1302 to S1310.

[0158] In step S1302, if it is a training process, a stacked autoencoder is used to obtain the dimensionality reduction features of the training object based on the historical sales data of the training object. The specific implementation method can refer to step S812; if it is a prediction process, a stacked autoencoder is used to obtain the dimensionality reduction features 13004 of the prediction object based on the historical sales data of the prediction object. The specific implementation method can refer to step S1104.

[0159] In step S1304, if it is a training process, a density clustering method is used to cluster multiple training objects according to the neural network output of each training object to obtain a clustering result 13002. The specific implementation method can refer to step S706; if it is a prediction process, this step is skipped.

[0160] In step S1306, the clustering result 13002 is used as a label, and the dimension reduction features 13004 and feature engineering features of each training object are input into the xgb binary classification model for training, wherein the feature engineering features may include sales features 13006, time features 13008, and economic features 13010. For a specific implementation, please refer to Fig. 9 .

[0161] In step S1308, the dimension reduction features 13004 and feature engineering features of the prediction object are input into the xgb binary classification model to predict the probability of sales of the prediction object occurring in the target period to obtain probability features 13012. The specific implementation method can refer to step S1110.

[0162] In step S1310, the probability feature 13012, the dimension reduction feature 13004 and the feature engineering feature of the prediction object are input into the (trained) xgb regression model, and the sales prediction result 13014 of the prediction object is output. For specific implementation methods, please refer to Fig.11 The process of training the xgb regression model can be found in Fig. 9 .

[0163] The following describes the experimental results of comparing the method provided by the embodiment of the present disclosure with the FFORMA method used in the related art.

[0164] Taking the spare parts data of a certain auto spare parts company as an example, the prediction object set has a total of 15,000 SKUs, and the time range is from January 2020 to December 2022. The data from January 2018 to June 2022 is used as the training object set, and the data from June to December 2022 is used as the test set.

[0165] 1) The method provided by the embodiment of the present disclosure can accurately predict whether dynamic sales will occur during the target period

[0166] The dynamic sales ratio in the monthly dimension of the training object set is about 0.4. The AUC performance on the test set is 0.83 when the method provided by the embodiment of the present disclosure is used for training. The probability judgment threshold is set to 0.5, and the ACC is 0.82.

[0167] 2) The method provided by the embodiment of the present disclosure can more accurately predict the sales volume during the sales promotion period

[0168] The integrated learning time series prediction method FFORMA is used as the control group. In business scenarios, the prediction capabilities of two types of SKUs are usually compared, namely the full SKU and the high-flow SKU with relatively high sales frequency (the average number of sales months in the past year is greater than 10 times). The specific results are shown in Table 2 below. The effect of the full SKU is significantly improved. This is because there are more zeros in the full SKU. The method provided by the embodiment of the present disclosure has the ability to predict zeros, so the prediction accuracy of the full amount is significantly better. Secondly, the performance on high-flow SKUs is also significantly better than FFORMA. In addition to the accurate prediction of zeros, economic data also helps to predict specific sales quantities more accurately.

[0169] Table 2

[0170]

[0171] Fig.14FIG. 1 is a block diagram of a data processing device according to an exemplary embodiment. Fig.14 The device shown can be applied to the server side of the above system, for example, and can also be applied to the terminal device of the above system.

[0172] refer to Fig.14 The device 140 provided in the embodiment of the present disclosure may include an acquisition module 1402 , a feature engineering module 1404 , a classification module 1406 and a prediction module 1408 .

[0173] The acquisition module 1402 may be used to acquire historical sales data of the forecast object, where the historical sales data of the forecast object includes the sales of the forecast object at multiple time points within a preset time period before a target period.

[0174] The feature engineering module 1404 may be used to obtain feature engineering features of the prediction object based on the historical sales data of the prediction object and the historical economic data of the corresponding time through feature engineering.

[0175] The classification module 1406 may be used to predict whether sales of the prediction object will occur in the target period according to the feature engineering features of the prediction object through a classification model to obtain a prediction result.

[0176] The prediction module 1408 may be used to predict the sales volume of the prediction object in the target period through a prediction model based on the prediction result and feature engineering features of the prediction object if the prediction result indicates that the prediction object will be sold in the target period.

[0177] Fig.15 FIG. 1 is a block diagram of another data processing device according to an exemplary embodiment. Fig.15 The device shown can be applied to the server side of the above system, for example, and can also be applied to the terminal device of the above system.

[0178] refer to Fig.15 The device 150 provided in the embodiment of the present disclosure may include an acquisition module 1502, a dimensionality reduction module 15032, a clustering module 15034, a feature engineering module 1504, a training module 1505, a classification module 1506 and a prediction module 1508.

[0179] The acquisition module 1502 may be used to acquire historical sales data of the forecast object, where the historical sales data of the forecast object includes the sales of the forecast object at multiple time points within a preset time period before the target period.

[0180] The acquisition module 1502 may also be used to acquire historical sales data of multiple training objects, where the historical sales data of each training object in the multiple training objects includes the sales of the training object at multiple time points within a preset time period before a preset period.

[0181] The acquisition module 1502 may also be used to acquire historical economic data at multiple time points within a preset time period before a preset period.

[0182] Historical economic data may include historical cost data, historical price data, and historical capital data.

[0183] The dimension reduction module 15032 can be used to obtain the neural network output of each training object based on the historical sales data of the training object using a stacked autoencoder.

[0184] The dimension reduction module 15032 can also be used to obtain the dimension reduction features of each training object based on the historical sales data of the training object using a stacked autoencoder.

[0185] The dimension reduction module 15032 can also be used to use a stacked autoencoder to obtain the dimension reduction features of the prediction object based on the historical sales data of the prediction object.

[0186] The clustering module 15034 may be used to cluster multiple training objects according to the neural network output of each training object using a density clustering method to obtain a clustering result.

[0187] The feature engineering module 1504 may be used to obtain feature engineering features of the prediction object based on the historical sales data of the prediction object and the historical economic data of the corresponding time through feature engineering.

[0188] The feature engineering module 1504 can also be used to use a sliding window to obtain target sales index values ​​at multiple preset time points within a preset time period based on the historical sales data of the prediction object, and obtain sales features; use a sliding window to obtain target economic data index values ​​at multiple preset time points based on the historical economic data of the corresponding time, and obtain economic features; and splice the sales features and economic features according to multiple preset time points to obtain feature engineering features of the prediction object.

[0189] The feature engineering module 1504 can also be used to obtain feature engineering features of each training object based on the historical sales data of the training object and the historical economic data of the corresponding time through feature engineering.

[0190] The training module 1505 may be used to train the initial classification model and the initial prediction model based on the clustering results to obtain a classification model and a prediction model.

[0191] The training module 1505 can also be used to use the feature engineering features and dimensionality reduction features of each training object as input features and the clustering results as labels to train an initial classification model and obtain a classification model.

[0192] The feature engineering module 1504 can also be used to mark holidays at multiple preset time points to obtain date features; and to concatenate sales features, economic features, and date features according to multiple preset time points to obtain feature engineering features of the prediction object.

[0193] The classification module 1506 may be used to predict whether sales of the prediction object will occur in the target period according to the feature engineering features of the prediction object through a classification model to obtain a prediction result.

[0194] The prediction results may include probability characteristics of the predicted object.

[0195] The classification module 1506 can also be used to predict the probability of sales of the training object occurring in a preset period for each training object through a classification model according to the feature engineering features and dimensionality reduction features of the training object, thereby obtaining the probability features of each training object.

[0196] The classification module 1506 can also be used to predict the probability of sales of the predicted object occurring in the target period according to the feature engineering features and dimensionality reduction features of the predicted object through a classification model to obtain the probability features of the predicted object.

[0197] The prediction module 1508 may be used to predict the sales volume of the prediction object in the target period through a prediction model based on the prediction result and feature engineering features of the prediction object if the prediction result indicates that the prediction object will be sold in the target period.

[0198] The prediction module 1508 can also be used to determine multiple target training objects from multiple training objects based on the probability characteristics of each training object; use the feature engineering features, dimensionality reduction features and probability features of each target training object in the multiple target training objects as input features, and use the sales volume of each target training object in a preset period as a label to train an initial prediction model and obtain a prediction model.

[0199] The prediction module 1508 may also be used to obtain the feature importance of feature engineering features, dimensionality reduction features, and probability features based on the prediction model.

[0200] The prediction module 1508 may also be used to predict the sales volume of the prediction object in the target period through the prediction model according to the probability characteristics, feature engineering characteristics and dimensionality reduction characteristics of the prediction object if the probability characteristics of the prediction object are greater than the prediction probability threshold.

[0201] The specific implementation of each module in the device provided by the embodiment of the present disclosure can refer to the content of the above method, which will not be repeated here.

[0202] Fig.16 The following is a schematic diagram showing the structure of an electronic device in an embodiment of the present disclosure. It should be noted that: Fig.16The device shown is only an example of a computer system, which should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0203] like Fig.16 As shown, the device 1600 includes a central processing unit (CPU) 1601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1602 or a program loaded from a storage part 1608 into a random access memory (RAM) 1603. In the RAM 1603, various programs and data required for the operation of the device 1600 are also stored. The CPU 1601, the ROM 1602, and the RAM 1603 are connected to each other via a bus 1604. An input / output (I / O) interface 1605 is also connected to the bus 1604.

[0204] The following components are connected to the I / O interface 1605: an input section 1606 including a keyboard, a mouse, etc.; an output section 1607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1608 including a hard disk, etc.; and a communication section 1609 including a network interface card such as a LAN card, a modem, etc. The communication section 1609 performs communication processing via a network such as the Internet. A drive 1610 is also connected to the I / O interface 1605 as needed. A removable medium 1611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1610 as needed, so that a computer program read therefrom is installed into the storage section 1608 as needed.

[0205] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part 1609, and / or installed from a removable medium 1611. When the computer program is executed by a central processing unit (CPU) 1601, the above-mentioned functions defined in the system of the present disclosure are executed.

[0206] It should be noted that the computer-readable medium shown in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0207] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0208] The modules involved in the embodiments described in the present disclosure may be implemented by software or by hardware. The modules described may also be set in a processor, for example, it may be described as: a processor includes an acquisition module, a feature engineering module, a classification module, and a prediction module. The names of these modules do not constitute a limitation on the modules themselves in some cases, for example, the acquisition module may also be described as a "module for acquiring raw data".

[0209] As another aspect, the present disclosure further provides a computer-readable medium, which may be included in the device described in the above embodiment; or may exist independently without being assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by a device, the device includes:

[0210] Obtain historical sales data of the prediction object, the historical sales data of the prediction object includes the sales of the prediction object at multiple time points within a preset time period before the target period; obtain historical economic data at multiple time points within a preset time period before the target period; obtain feature engineering features of the prediction object based on the historical sales data of the prediction object and the historical economic data of the corresponding time through feature engineering; predict whether sales of the prediction object will occur in the target period based on the feature engineering features of the prediction object through a classification model to obtain a prediction result; if the prediction result is that sales of the prediction object will occur in the target period, predict the sales of the prediction object in the target period through the prediction model based on the prediction result and the feature engineering features of the prediction object.

[0211] The exemplary embodiments of the present disclosure are specifically shown and described above. It should be understood that the present disclosure is not limited to the detailed structures, configurations or implementations described herein; on the contrary, the present disclosure is intended to cover various modifications and equivalent configurations included in the spirit and scope of the appended claims.

Claims

1. A data processing method, characterized in that: include: Acquire historical sales data of the forecast object, wherein the historical sales data of the forecast object includes sales of the forecast object at multiple time points within a preset time period before a target period; Obtaining historical economic data at multiple time points within a preset time period before the target period; Obtaining feature engineering features of the prediction object based on historical sales data of the prediction object and historical economic data of the corresponding time through feature engineering; Predicting whether sales of the forecast object will occur in the target period according to the feature engineering features of the forecast object by using a classification model to obtain a forecast result; If the prediction result is that the predicted object is sold during the target period, the sales volume of the predicted object during the target period is predicted by a prediction model based on the prediction result and the feature engineering features of the predicted object.

2. The method according to claim 1, characterized in that The feature engineering features of the prediction object are obtained according to the historical sales data of the prediction object and the historical economic data of the corresponding time through feature engineering, including: Using a sliding window to obtain target sales index values ​​at multiple preset time points within the preset time period according to the historical sales data of the predicted object, and obtaining sales characteristics; Using the sliding window to obtain the target economic data index values ​​of the plurality of preset time points according to the historical economic data of the corresponding time, and obtaining the economic characteristics; The sales volume characteristics and the economic characteristics are spliced ​​according to the multiple preset time points to obtain the feature engineering characteristics of the prediction object.

3. The method according to claim 2, characterized in that The feature engineering features of the prediction object are obtained according to the historical sales data of the prediction object and the historical economic data of the corresponding time through feature engineering, and also include: Marking the multiple preset time points with holidays to obtain date features; The sales volume feature and the economic feature are spliced ​​according to the multiple preset time points to obtain the feature engineering features of the prediction object, including: The sales volume characteristics, the economic characteristics and the date characteristics are spliced ​​according to the multiple preset time points to obtain the feature engineering characteristics of the prediction object.

4. The method according to claim 1, characterized in that Also includes: Acquire historical sales data of a plurality of training objects, wherein the historical sales data of each training object in the plurality of training objects includes sales of the training object at a plurality of time points within a preset time period before a preset period; For each of the training objects, a stacked autoencoder is used to obtain a neural network output of the training object according to the historical sales data of the training object; Clustering the plurality of training objects according to the neural network output of each training object using a density clustering method to obtain a clustering result; Based on the clustering results, the initial classification model and the initial prediction model are trained respectively to obtain the classification model and the prediction model.

5. The method according to claim 4, characterized in that Also includes: Obtaining historical economic data at multiple time points within a preset time period before the preset period; For each training object, obtaining the feature engineering features of the training object according to the historical sales data of the training object and the historical economic data of the corresponding time through the feature engineering; For each of the training objects, a stacked autoencoder is used to obtain a dimension reduction feature of the training object according to the historical sales data of the training object; Based on the clustering results, the initial classification model and the initial prediction model are trained respectively to obtain the classification model and the prediction model, including: The feature engineering features and dimensionality reduction features of each training object are used as input features, the clustering results are used as labels, and the initial classification model is trained to obtain the classification model.

6. The method according to claim 5, characterized in that Based on the clustering results, the initial classification model and the initial prediction model are trained respectively to obtain the classification model and the prediction model, and further comprising: For each training object, the probability of sales of the training object occurring in the preset period is predicted by the classification model according to the feature engineering features and the dimensionality reduction features of the training object, so as to obtain the probability features of each training object; Determining a plurality of target training objects from the plurality of training objects according to the probability characteristics of the respective training objects; The feature engineering features, dimensionality reduction features and probability features of each target training object among the multiple target training objects are used as input features, the sales volume of each target training object in the preset period is used as a label, the initial prediction model is trained, and the prediction model is obtained.

7. The method according to claim 6, characterized in that Also includes: The feature importance of feature engineering features, dimensionality reduction features and probability features are respectively obtained based on the prediction model.

8. The method according to claim 1, characterized in that Also includes: Using a stacked autoencoder to obtain the dimension reduction features of the prediction object according to the historical sales data of the prediction object; The prediction result includes the probability characteristics of the prediction object; Predicting whether the sales of the forecast object will occur in the target period according to the feature engineering features of the forecast object through the classification model to obtain a forecast result, including: Predicting the probability of sales of the predicted object occurring in the target period according to the feature engineering features and dimensionality reduction features of the predicted object by the classification model, thereby obtaining the probability features of the predicted object; If the prediction result is that the predicted object is sold during the target period, the sales volume of the predicted object during the target period is predicted by a prediction model according to the prediction result and the feature engineering features of the predicted object, including: If the probability feature of the predicted object is greater than the predicted probability threshold, the sales volume of the predicted object in the target period is predicted by the prediction model according to the probability feature, feature engineering feature and dimensionality reduction feature of the predicted object.

9. The method according to claim 1, characterized in that: The historical economic data includes historical cost data, historical price data and historical capital data.

10. A data processing device, characterized in that: include: An acquisition module, used to acquire historical sales data of a forecast object, wherein the historical sales data of the forecast object includes sales of the forecast object at multiple time points within a preset time period before a target period; The acquisition module is further used to acquire historical economic data at multiple time points within a preset time period before the target period; A feature engineering module, used for obtaining feature engineering features of the prediction object according to the historical sales data of the prediction object and the historical economic data of the corresponding time through feature engineering; A classification module, used to predict whether the sales of the forecast object will occur in the target period according to the feature engineering features of the forecast object through a classification model to obtain a forecast result; A prediction module is used to predict the sales volume of the prediction object in the target period through a prediction model based on the prediction result and the feature engineering features of the prediction object if the prediction result is that the prediction object will be sold in the target period.

11. An electronic device, comprising: A memory, a processor, and executable instructions stored in the memory and executable in the processor, wherein the processor implements the method according to any one of claims 1 to 9 when executing the executable instructions.

12. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: When the executable instructions are executed by a processor, the method according to any one of claims 1 to 9 is implemented.