Return data prediction model training method, device and equipment, and return data prediction method, device and equipment

Abstract

The invention provides a return data prediction model training method, device and equipment, and a return data prediction method, device and equipment, and the return data prediction model training method comprises the steps: determining sample demand data and sample return data; determining a first covariable corresponding to the sample demand data, the first covariable being used for representing a variable condition related to the sample demand data; according to the first covariable, determining a sample demand feature corresponding to the sample demand data, and determining a sample return feature corresponding to the sample return data; processing the sample demand features and the sample return features to obtain sample prediction features; and training an initial return data prediction model according to the sample demand features, the sample return data and the sample prediction features to obtain a target return data prediction model, and solving the technical problem of low accuracy of return data prediction in the prior art.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a return data prediction model training, return data prediction method, device and equipment. Background Art

[0002] In e-commerce scenarios, positive demand data (purchases) can be predicted to guide sales plans and replenishment. Reverse return data (returns) can also be predicted to support the subsequent second-hand goods warehousing and resale process (or return to suppliers). The accurate prediction of return data at future points in time can effectively improve the planning of subsequent processes, facilitate the optimization of the second-hand warehousing process capacity, and guide the sales of second-hand goods. Since there is a certain delay between the time when the return data is generated and the positive demand data, it is impossible to directly calculate the estimated return data proportionally based on the predicted value of the positive demand data. However, there is a strong correlation between the positive demand data and the reverse return data. Furthermore, due to the characteristics of the e-commerce scenario, during the promotion period, the positive demand data usually has a higher peak, while the return data appears relatively smoothly after the peak of the demand data. In daily sales, return data may appear intermittently. As a result, the distribution of return data in the daily sales and after the promotion period is obviously different.

[0003] In the related art, the prediction of return data is a general time series prediction method, which does not effectively use the demand data in the time series dimension, resulting in low accuracy of return data prediction. Summary of the invention

[0004] The present application aims to solve one of the technical problems in the related art at least to some extent.

[0005] To this end, the present application proposes a return data prediction model training method, a return data prediction method, a return data prediction model training device, a return data prediction device, an electronic device, a computer-readable storage medium and a computer program product to effectively improve the accuracy of return data prediction.

[0006] To achieve the above-mentioned purpose, the first aspect embodiment of the present application proposes a return data prediction model training method, including: determining sample demand data and sample return data; determining a first covariate corresponding to the sample demand data, wherein the first covariate is used to represent the variable situation related to the sample demand data; according to the first covariate, determining the sample demand characteristics corresponding to the sample demand data, and determining the sample return characteristics corresponding to the sample return data; processing the sample demand characteristics and the sample return characteristics to obtain the sample prediction characteristics; and training the initial return data prediction model according to the sample demand characteristics, the sample return data and the sample prediction characteristics to obtain the target return data prediction model.

[0007] To achieve the above-mentioned purpose, the second aspect embodiment of the present application proposes a return data prediction method, including: determining reference demand data and reference return data; inputting the reference demand data and reference return data into the target return data prediction model trained by the return data prediction model training method proposed in the first aspect embodiment of the present application, so as to obtain the predicted return data output by the target return data prediction model.

[0008] To achieve the above-mentioned purpose, the third aspect embodiment of the present application proposes a return data prediction model training device, including: a first determination module, used to determine sample demand data and sample return data; a second determination module, used to determine a first covariate corresponding to the sample demand data, wherein the first covariate is used to represent the variable situation related to the sample demand data; a third determination module, used to determine the sample demand characteristics corresponding to the sample demand data according to the first covariate, and determine the sample return characteristics corresponding to the sample return data; a processing module, used to process the sample demand characteristics and the sample return characteristics to obtain the sample prediction characteristics; and a training module, used to train the initial return data prediction model according to the sample demand characteristics, the sample return data and the sample prediction characteristics to obtain the target return data prediction model.

[0009] To achieve the above-mentioned purpose, the fourth aspect embodiment of the present application proposes a return data prediction device, including: a fourth determination module, used to determine reference demand data and reference return data; a prediction module, used to input the reference demand data and reference return data into the target return data prediction model trained by the return data prediction model training device proposed in the third aspect embodiment of the present application, so as to obtain the predicted return data output by the target return data prediction model.

[0010] To achieve the above-mentioned purpose, the fifth aspect embodiment of the present application proposes an electronic device, comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory to implement the return data prediction model training method proposed in the first aspect embodiment of the present application, or to implement the return data prediction method proposed in the second aspect embodiment of the present application.

[0011] To achieve the above-mentioned objectives, the sixth aspect embodiment of the present application proposes a computer-readable storage medium, in which computer execution instructions are stored. When the computer execution instructions are executed by the processor, they are used to implement the return data prediction model training method proposed in the first aspect embodiment of the present application, or to implement the return data prediction method proposed in the second aspect embodiment of the present application.

[0012] To achieve the above-mentioned purpose, the seventh aspect embodiment of the present application proposes a computer program product, including a computer program, which, when executed by a processor, implements the return data prediction model training method proposed in the first aspect embodiment of the present application, or implements the return data prediction method proposed in the second aspect embodiment of the present application.

[0013] The return data prediction model training method, return data prediction method, return data prediction model training device, return data prediction device, electronic device, computer-readable storage medium and computer program product provided by the present application determine sample demand data and sample return data, and determine a first covariate corresponding to the sample demand data, wherein the first covariate is used to represent the variable situation related to the sample demand data, determine the sample demand characteristics corresponding to the sample demand data according to the first covariate, and determine the sample return characteristics corresponding to the sample return data, process the sample demand characteristics and the sample return characteristics to obtain sample prediction characteristics, and train an initial return data prediction model according to the sample demand characteristics, the sample return data and the sample prediction characteristics to obtain a target return data prediction model. Since in the process of training the return data prediction model, not only the sample return characteristics of the sample return data, but also the sample demand characteristics of the sample demand data and the first covariate affecting the sample demand data are considered, the modeling accuracy of the return data prediction model can be effectively improved. When predicting future return data based on the trained target return data prediction model, the accuracy of the return data prediction can be effectively improved.

[0014] 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

[0015] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0016] Figure 1 It is a flowchart of a return data prediction model training method proposed in an embodiment of the present application;

[0017] Figure 2 It is a flowchart of a return data prediction model training method proposed in another embodiment of the present application;

[0018] Figure 3 It is a flowchart of a return data prediction model training method proposed in another embodiment of the present application;

[0019] Figure 4 It is a structural diagram of a return data prediction model in an embodiment of the present application;

[0020] Figure 5 is a schematic diagram of the return volume prediction interaction of the time series distribution migration in the embodiment of the present application;

[0021] Figure 6 It is a flowchart of a return data prediction method proposed in an embodiment of the present application;

[0022] Figure 7 It is a structural schematic diagram of a return data prediction model training device proposed in one embodiment of the present application;

[0023] Figure 8 It is a structural diagram of a return data prediction device proposed in an embodiment of the present application;

[0024] Fig. 9 A block diagram of an exemplary electronic device suitable for implementing the embodiments of the present application is shown. DETAILED DESCRIPTION

[0025] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0026] The following describes the return data prediction model training method, return data prediction method, return data prediction model training device, return data prediction device, electronic device, computer-readable storage medium and computer program product of the embodiments of the present application with reference to the accompanying drawings.

[0027] Figure 1 It is a flowchart of the return data prediction model training method proposed in one embodiment of the present application.

[0028] In this embodiment, the return data prediction model training method is configured as a return data prediction model training device. In this embodiment, the return data prediction model training method can be configured in the return data prediction model training device. The return data prediction model training device can be set in a server, or can also be set in an electronic device, without limitation.

[0029] This embodiment takes the case where the return data prediction model training method is configured in an electronic device as an example, wherein the electronic device is a hardware device such as a smart phone, a tablet computer, a personal digital assistant, an e-book, etc. having various operating systems.

[0030] It should be noted that the execution entity of the embodiment of the present application may be, for example, a central processing unit (CPU) in a server or electronic device in terms of hardware, or may be, for example, a related background service in a server or electronic device in terms of software, without limitation.

[0031] like Figure 1 As shown, the return data prediction model training method includes:

[0032] S101: Determine sample demand data and sample return data.

[0033] In some embodiments, the demand data used to train the return data prediction model may be referred to as sample demand data. Demand data refers to positive demand data (e.g., purchase data). The return data used to train the return data prediction model may be referred to as sample return data. The return data prediction model may be an artificial intelligence model for predicting future return data, such as a neural network model, a machine learning model, and the like.

[0034] In some embodiments, the sample demand data may be sample demand data within a period of history, and / or recent sample demand data, and / or future sample demand data. The sample return data may be, for example, sample return data within a period of history, and / or known sample return data. The return data predicted based on the return data model may be future return data.

[0035] In some embodiments, the sample demand data and sample return data can be obtained from any possible data source; or the sample demand data and sample return data can be parsed from the model training script; of course, any other possible method can also be used to determine the sample demand data and sample return data, without limitation.

[0036] In some embodiments, an input time series may be determined, wherein the input time series includes: multiple input times, and initial demand data and initial return data are obtained, the initial demand data is instance-normalized according to the multiple input times to obtain sample demand data, and the initial return data is instance-normalized according to the multiple input times to obtain sample return data. Since the input initial demand data is instance-normalized and the input initial return data is instance-normalized, the processed sample demand data can effectively retain the individual characteristics of the initial demand data, and the processed sample return data can effectively retain the individual characteristics of the initial return data. The non-stationary information of the initial demand data and the initial return data can also be effectively removed.

[0037] Among them, instance normalization is a normalization technology, which is mainly used in neural network models in deep learning. It performs normalization on each independent sample instance to improve the learning effect and generalization ability of the model. Instance normalization is to perform normalization on each independent sample instance.

[0038] In some embodiments, in the process of performing instance normalization processing on the initial demand data according to multiple input times to obtain sample demand data, the initial demand data may be mapped to each input time to obtain input demand data corresponding to each input time, and the length of the input time series and the number of first covariates of each input demand data are determined, the mean and standard deviation of the multiple input demand data are determined according to the length and the number of first covariates, and each input demand data is standardized according to the mean and standard deviation, and the multiple input demand data obtained by the standardized processing are used as sample demand data. Thereby, the return data prediction model to be trained can effectively capture the data change characteristics between different input demand data, and by introducing the first covariate of the input demand data into the instance normalization processing, the variables affecting the input demand data are taken into consideration, and the distribution characteristics between different input demand data can be effectively learned.

[0039] In some embodiments, the input time series may be a time series that includes multiple input times. The initial demand data may be divided based on multiple input times, indicating that a corresponding input demand data is input at different input times. The first covariate may be used to represent the variable situation related to the sample demand data. The first covariate may be, for example, the date, activity information (such as activity promotion volume, promotion discount, etc.), price, and other information related to the input demand data. The length of the input time series may refer to the number of multiple input times. In the embodiments of the present application, the mean and standard deviation of multiple input demand data may be determined with reference to the length and the number of first covariates. The obtained mean and standard deviation may be used to standardize each input demand data, and the multiple input demand data obtained by the standardized processing may be used as sample demand data.

[0040] For example, when inputting demand data, the data is subjected to instance normalization, which can effectively remove the non-stationary information of the input demand data. Then, the processed sample demand data is input into the demand quantity-feature extractor (a neural network structure suitable for time series guarantee such as a recurrent neural network (RNN) and / or a convolutional neural network (CNN) can be used). The steps of instance normalization are as follows: first, the mean and standard deviation of multiple input demand data are calculated according to the instance:

[0041]

[0042]

[0043] in, is the input demand data, i represents the i-th input demand data, K is the number of the first covariate corresponding to the input demand data, T x is the length of the input time series. Represents the mean of multiple input demand data, Represents the standard deviation of multiple input demand data.

[0044] Then standardize the input demand data:

[0045]

[0046] in, represents the standardized input demand data, and ε represents a constant.

[0047] In some embodiments, in the process of performing instance normalization processing on the initial return data according to multiple input times to obtain sample return data, the initial return data may be mapped to each input time to obtain input return data corresponding to each input time, and the length of the input time series and the number of second covariates of each input return data are determined, the mean and standard deviation of the multiple input return data are determined according to the length and the number of second covariates, and each input return data is standardized according to the mean and standard deviation, and the multiple input return data obtained by the standardized processing are used as sample return data. In this way, the return data prediction model to be trained can effectively capture the data change characteristics between different input return data, and by introducing the second covariate of the input return data into the instance normalization processing, the variable situation affecting the input return data is taken into consideration, and the distribution characteristics between different input return data can be effectively learned.

[0048] In some embodiments, the input time series may be a time series that includes multiple input times. The initial return data may be divided based on multiple input times, indicating that a corresponding input return data is input at different input times. The second covariate may be used to represent the variable situation related to the sample return data, and the second covariate may be, for example, information such as date, activity, price, etc. related to the input return data. The length of the input time series may refer to the number of multiple input times. In the embodiments of the present application, the mean and standard deviation of multiple input return data may be determined with reference to the length and the number of second covariates, and the obtained mean and standard deviation may be used to standardize each input return data, and the multiple input return data obtained by the standardized processing may be used as sample return data.

[0049] In some embodiments, the above formulas (1), (2), and (3) may be referred to to perform instance normalization processing on the initial return data according to multiple input times to obtain sample return data.

[0050] S102: Determine a first covariate corresponding to the sample demand data, wherein the first covariate is used to represent a variable situation related to the sample demand data.

[0051] After determining the sample demand data and the sample return data, the embodiment of the present application can determine the first covariate corresponding to the sample demand data, wherein the first covariate is used to represent the variable situation related to the sample demand data, and the first covariate is, for example, information such as date, activity, price, etc. related to the input demand data. Then, feature analysis is performed on the sample demand data and the first covariate affecting the sample demand data to guide the modeling of the return data prediction model.

[0052] S103: Determine, based on the first covariate, the sample demand feature corresponding to the sample demand data, and determine the sample return feature corresponding to the sample return data.

[0053] After determining the sample demand data and the sample return data, as well as the first covariate corresponding to the sample demand data, the first covariate can be referenced to determine the sample demand features corresponding to the sample demand data, and determine the sample return features corresponding to the sample return data.

[0054] Among them, the sample demand feature is used to represent the feature distribution of the sample demand data. The sample return feature is used to represent the feature distribution of the sample return data. In some embodiments, the sample demand feature corresponding to the sample demand data can be determined in combination with the first covariate. For example, the first covariate can be integrated into the initial demand feature of the sample demand data, and the initial demand feature integrated with the first covariate is used as the sample demand feature. Alternatively, the first covariate and the sample demand data can be input into the feature learning model together to obtain the sample demand feature corresponding to the sample demand data. Of course, any other possible method can also be used to refer to the first covariate to determine the sample demand feature corresponding to the sample demand data, and there is no limitation to this. In some embodiments, after determining the sample return feature corresponding to the sample return data, feature analysis can be performed on the sample return data to determine its sample return feature.

[0055] S104: Process the sample demand features and the sample return features to obtain sample prediction features.

[0056] In some embodiments, after determining the sample demand features corresponding to the sample demand data and the sample return features corresponding to the sample return data, the sample demand features and the sample return features may be fused and the features obtained by the fusion process may be used as sample prediction features; or the sample demand features and the sample return features may be input into a feature fusion model to obtain sample prediction features; of course, any other possible method may be used to process the sample demand features and the sample return features to obtain sample prediction features, such as modeling methods, mathematical algorithm methods, and so on.

[0057] S105: Train an initial return data prediction model according to sample demand characteristics, sample return data, and sample prediction characteristics to obtain a target return data prediction model.

[0058] In some embodiments, after determining the sample prediction features, an initial return data prediction model may be trained with reference to the sample demand features, sample return data, and sample prediction features to obtain a target return data prediction model.

[0059] In some embodiments, the return data prediction model to be trained may be referred to as an initial return data prediction model. The return data prediction model may be an artificial intelligence model for predicting future return data, such as a neural network model, a machine learning model, etc. The return data prediction model obtained by training may be referred to as a target return data prediction model.

[0060] In some embodiments, the initial return data prediction model can be iteratively trained at least once based on sample demand characteristics, sample return data, and sample prediction characteristics until the trained return data prediction model meets the convergence conditions, and the trained return data prediction model is used as the target return data prediction model.

[0061] In this embodiment, by determining sample demand data and sample return data, and determining a first covariate corresponding to the sample demand data, wherein the first covariate is used to represent the variable status related to the sample demand data, according to the first covariate, determining the sample demand features corresponding to the sample demand data, and determining the sample return features corresponding to the sample return data, processing the sample demand features and the sample return features to obtain sample prediction features, and training an initial return data prediction model according to the sample demand features, the sample return data and the sample prediction features to obtain a target return data prediction model. In the process of training the return data prediction model, not only the sample return features of the sample return data, but also the sample demand features of the sample demand data and the first covariate affecting the sample demand data are considered. Therefore, the modeling accuracy of the return data prediction model can be effectively improved. When predicting future return data based on the trained target return data prediction model, the accuracy of the return data prediction can be effectively improved.

[0062] Figure 2 It is a flowchart of a return data prediction model training method proposed in another embodiment of the present application.

[0063] like Figure 2 As shown, the return data prediction model training method includes:

[0064] S201: Determine sample demand data and sample return data.

[0065] S202: Determine a first covariate corresponding to the sample demand data, wherein the first covariate is used to represent a variable situation related to the sample demand data.

[0066] For the description of S201 - S202 , please refer to the above embodiment, which will not be repeated here.

[0067] S203: Determine initial demand characteristics corresponding to the sample demand data.

[0068] In some embodiments, features obtained by performing feature analysis on sample demand data may be referred to as initial demand features.

[0069] S204: Determine a first covariate feature corresponding to the first covariate.

[0070] In some embodiments, the feature obtained by performing feature analysis on the first covariate may be referred to as the first covariate feature.

[0071] S205: Concatenate the initial demand feature and the first covariate feature, and use the concatenated feature as the sample demand feature.

[0072] In some embodiments, after determining the initial demand feature corresponding to the sample demand data and the first covariate feature corresponding to the first covariate, the initial demand feature and the first covariate feature can be spliced, and the spliced ​​feature can be used as the sample demand feature. This can improve the convenience of feature fusion, effectively integrate the features of the first covariate into the features of the sample demand data, and avoid introducing too many computing resources, thereby improving the efficiency of feature fusion.

[0073] For example, the standardized input requirement data can be Put it into the feature processor to extract the features of the basic demand data (the features of the extracted basic demand data can be an optional example of the initial demand features), and output the feature representation vector Vs (which can be a vectorized initial demand feature). Then, when the future covariate (an optional example of the first covariate) is input, the first covariate is embedded (the embedding operation is to convert the discrete input into a continuous vector representation so that the network can better process the input data), mainly representing the future time information to capture the corresponding peak value, and the corresponding output is the future covariate vector Vv (an optional example of the first covariate feature). Then Vs and Vv are spliced, and after being processed by a multilayer perceptron (Multilayer Perceptron, MLP) (represented as MLP1), the final positive demand representation Vd (an optional example of the sample demand feature) is output.

[0074] S206: Determine a second covariate corresponding to the sample return data, wherein the second covariate is used to represent a variable condition related to the sample return data.

[0075] In the embodiment of the present application, a second covariate corresponding to the sample return data may also be determined, wherein the second covariate is used to represent the variable situation related to the sample return data, and the second covariate is, for example, information such as the time series of commodity ratings related to the sample return data. Then, feature analysis is performed on the sample return data and the second covariate affecting the sample return data to guide the modeling of the return data prediction model.

[0076] S207: Determine initial return features corresponding to the sample return data.

[0077] In some embodiments, features obtained by performing feature analysis on sample return data may be referred to as initial return features.

[0078] S208: Determine a second covariate feature corresponding to the second covariate.

[0079] In some embodiments, the feature obtained by performing feature analysis on the second covariate may be referred to as the second covariate feature.

[0080] S209: Concatenate the initial return feature and the second covariate feature, and use the concatenated feature as the sample return feature.

[0081] In some embodiments, after determining the initial return feature corresponding to the sample return data and the second covariate feature corresponding to the second covariate, the initial return feature and the second covariate feature can be concatenated, and the concatenated feature is used as the sample return feature. This can improve the convenience of feature fusion, effectively integrate the features of the second covariate into the features of the sample return data, avoid introducing too many computing resources, and improve the efficiency of feature fusion.

[0082] For example, the sample return features of the sample return data can be obtained based on the analysis of the return quantity characterization module. Since the sample return data has a certain trend and is related to the characteristics of the product, the historical situation of the return data and the return-related covariates (an optional example of the second covariate), such as the product rating time series, can be referred to to predict future return data. The process of instance normalization of the sample return data can refer to the above-mentioned instance normalization process of the sample demand data to remove the non-stationary information of the sample return data. Then, the sample return data after instance normalization is input into the return quantity-feature extractor (RNN and / or CNN) and another MLP (denoted as MLP2) to extract global information. The extracted global information can be an optional example of the sample return feature.

[0083] S210: Concatenate the sample demand features and the sample return features, and use the concatenated features as sample prediction features.

[0084] In some embodiments, after obtaining the sample demand feature and the sample return feature, the two features can be fused to obtain the sample prediction feature. In the embodiment of the present application, the sample demand feature and the sample return feature can be directly spliced, and the spliced ​​feature can be used as the sample prediction feature, thereby effectively improving the efficiency and accuracy of feature fusion.

[0085] For example, the forward demand representation (an optional example of the sample demand feature) and the return quantity representation (an optional example of the sample return feature) can be combined to predict the return data for each day in the future. The sample demand feature and the sample return feature can be first concatenated to obtain the representation required for reverse prediction (an optional example of the sample prediction feature).

[0086] S211: Train an initial return data prediction model based on sample demand characteristics, sample return data, and sample prediction characteristics to obtain a target return data prediction model.

[0087] For the description of S211, please refer to the above embodiment, which will not be repeated here.

[0088] In this embodiment, since in the process of training the return data prediction model, not only the sample return characteristics of the sample return data, but also the sample demand characteristics of the sample demand data, and the first covariate affecting the sample demand data are considered, the modeling accuracy of the return data prediction model can be effectively improved. When predicting future return data based on the trained target return data prediction model, the accuracy of return data prediction can be effectively improved. The convenience of feature fusion can be improved. While effectively integrating the features of the first covariate into the features of the sample demand data, it can also avoid introducing too many computing resources and improve the efficiency of feature fusion. The convenience of feature fusion can be improved. While effectively integrating the features of the second covariate into the features of the sample return data, it can also avoid introducing too many computing resources and improve the efficiency of feature fusion.

[0089] Figure 3 It is a flowchart of a return data prediction model training method proposed in another embodiment of the present application.

[0090] like Figure 3 As shown, the return data prediction model training method includes:

[0091] S301: Determine sample demand data and sample return data.

[0092] S302: Determine a first covariate corresponding to the sample demand data, wherein the first covariate is used to represent a variable situation related to the sample demand data.

[0093] S303: Determine, based on the first covariate, the sample demand feature corresponding to the sample demand data, and determine the sample return feature corresponding to the sample return data.

[0094] S304: Process the sample demand features and the sample return features to obtain sample prediction features.

[0095] For the description of S301 - S304 , please refer to the above embodiment for details, which will not be repeated here.

[0096] S305: Input the sample demand characteristics into the cumulative demand prediction submodule of the initial return data prediction model to obtain the predicted cumulative demand data output by the cumulative demand prediction submodule.

[0097] The initial return data prediction model in the embodiment of the present application may include a cumulative demand prediction submodule and a return quantity distribution prediction submodule, and the cumulative demand prediction submodule can be used to predict the cumulative demand data.

[0098] In some embodiments, the demand data accumulated at multiple time points obtained by prediction can be referred to as predicted cumulative demand data. That is to say, the initial return data prediction model in the embodiment of the present application can not only predict the return data, but also output the predicted cumulative demand data according to the sample demand characteristics, and the predicted cumulative demand data can be used to guide the prediction of the return data. Therefore, the demand data is effectively used at the time series granularity to estimate the return data, which effectively improves the accuracy and interpretability of the predicted return data.

[0099] In the embodiment of the present application, the initial return data prediction model may include a cumulative demand prediction submodule. Since the main task of the initial return data prediction model is to predict the return volume, the demand prediction task performed by the cumulative demand prediction submodule may be to predict the demand data (sales volume) for the next n days. The cumulative demand prediction submodule has two goals. One is to restore the recent total sales volume to ensure that the recent characterization can be used as a reference for returns. On the other hand, the total sales volume for a period of time in the future is predicted at the same time to guide longer-term returns. Based on the output features of MLP1 as a whole, two numerical results are produced, which are used for supervised learning in subsequent modules.

[0100] S306: Input the sample prediction features into the return volume distribution prediction submodule of the initial return data prediction model to obtain the predicted return volume fluctuation data output by the return volume distribution prediction submodule.

[0101] In some embodiments, after processing sample demand features and sample return features to obtain sample prediction features, the sample prediction features can also be input into the return quantity distribution prediction submodule of the initial return data prediction model to obtain the predicted return quantity fluctuation data output by the return quantity distribution prediction submodule.

[0102] Among them, the return volume distribution prediction submodule can be used to predict the distribution of return volume, and the return data can also be called return volume. The data describing the fluctuation of return volume predicted based on the return volume distribution prediction submodel can be called predicted return volume fluctuation data. The return volume distribution prediction submodule can be used to output predicted return volume fluctuation data.

[0103] S307: Perform at least one round of iterative training on the initial return data prediction model based on the predicted cumulative demand data, sample return data, and predicted return volume fluctuation data until the initial return data prediction model converges, and use the trained initial return data prediction model as the target return data prediction model.

[0104] After determining the predicted cumulative demand data and the predicted return volume fluctuation data, the initial return data prediction model can be iteratively trained for at least one round using the predicted cumulative demand data, sample return data, and predicted return volume fluctuation data until the initial return data prediction model converges, and the initial return data prediction model obtained through training is used as the target return data prediction model.

[0105] In some embodiments, the sample return data includes: multiple input return data obtained through standardization, each input return data corresponds to an input time in the input time series; the predicted return volume fluctuation data includes: predicted return data corresponding to each future time in the future time series. In other words, the sample return data has been mapped to multiple input times of the input time series, and the sample return data is also a data series, including multiple input return data. The predicted return volume fluctuation data is the predicted return volume fluctuation data, which can also be a data series. The predicted return volume fluctuation data may include predicted return data corresponding to each future time in the future time series, and the future time series refers to a series including several future times.

[0106] In some embodiments, the initial return data prediction model can be trained for at least one round of iterations based on multiple input return data, multiple predicted return data, and predicted cumulative demand data until the initial return data prediction model converges. In this way, in the process of training the initial return data prediction model, the distribution of positive demand data is modeled and learned to guide the prediction of return data, thereby supporting the improvement of the accuracy of return data prediction and making it more in line with real e-commerce scenarios.

[0107] In some embodiments, in the process of performing at least one round of iterative training on the initial return data prediction model according to the predicted cumulative demand data, sample return data, and predicted return volume fluctuation data until the initial return data prediction model converges, the mean and standard deviation of multiple input return data and multiple predicted return data may be determined, and the initial return data prediction model may be iteratively trained according to the predicted cumulative demand data and the mean and standard deviation until the initial return data prediction model converges. Since the mean can reflect the central tendency of multiple input return data and multiple predicted return data, and the standard deviation can reflect the degree of dispersion of multiple input return data and multiple predicted return data, both can accurately characterize the distribution characteristics of return data. When the initial return data prediction model is iteratively trained based on the mean and standard deviation of multiple input return data and multiple predicted return data, the effectiveness of model training can be effectively improved, and the prediction accuracy of the return data prediction model can be effectively supported.

[0108] For example, the future mean of the future return volume can be output after processing with the new RevIN (i) and standard deviation std (i) The new RevIN processing can refer to learning the mean and standard deviation of future prediction data through the network using covariates, and conducting comprehensive learning through multi-objective tasks. Relevant and clear covariates can be effectively used to perform differential processing on different samples to obtain distribution conditions. On the one hand, the prediction accuracy is improved, and on the other hand, the data results are disassembled to improve the interpretability.

[0109] In some embodiments, the initial return data prediction model can be iteratively trained according to the predicted cumulative demand data and the mean and standard deviation until the initial return data prediction model converges. For example, a loss value can be calculated based on the predicted cumulative demand data and the mean and standard deviation. If the loss value meets the set conditions, it can be determined that the return data prediction model in the training process converges.

[0110] In some embodiments, multiple predicted return data can also be destandardized according to the mean and standard deviation to obtain multiple reference return data, and the model loss value can be determined based on the predicted cumulative demand data, multiple reference return data, the mean and standard deviation. If the model loss value is less than or equal to the loss threshold, it is determined that the initial return data prediction model has converged. If the model loss value is greater than the loss threshold, a new round of iterative training is performed on the initial return data prediction model based on the predicted cumulative demand data and the mean and standard deviation.

[0111] like Figure 4 As shown, Figure 4: is a structural diagram of the return data prediction model in an embodiment of the present application, wherein the demand quantity can be an optional example of sample demand data, the return quantity can be an optional example of sample return data, the future information can be an optional example of the first covariate, and the future information can include: date, activity, price, etc. The demand quantity-feature extractor and the global MLP1 submodule are used together to determine the sample demand feature, the return quantity-feature extractor and the global MLP2 submodule are used together to determine the sample return feature, the covariate embedding represents an optional example of the first covariate feature, and the local MLP submodule can be used to process the sample demand feature And sample return features to obtain sample prediction features. Cumulative demand prediction can be performed based on the cumulative demand prediction submodule, and return volume fluctuation prediction can be performed based on the return volume distribution prediction submodule. Std represents the standard deviation of multiple input return data and multiple predicted return data. Mean represents the mean of multiple input return data and multiple predicted return data. Different loss functions can be modeled according to different data contents, such as loss function 1, loss function 2, loss function Std, and loss function Mean. Then, the final loss function Final Loss Function is modeled according to loss function 1, loss function 2, loss function Std, and loss function Mean.

[0112] According to the De-Normalization method, the specific formula is as follows:

[0113]

[0114] in, is the network output result of the i-th time series (forecast return data), The final forecast output of the i-th time series (the forecast return data after de-normalization) is restored based on this process, which not only makes up for the loss of non-stationary information caused by Instance Normalization, but also uses the mean and standard deviation of the output of the neural network and the covariates of historical returns and future demand to effectively improve the utilization rate of the input information. At the same time, when facing data with weak regularity, the corresponding mean and standard deviation can be output according to the covariate. Considering the standard deviation and mean in the Normalization step, as well as the form of the affine coefficient, different affine coefficients can be obtained for different time series inputs.

[0115] An example of overall loss calculation is as follows:

[0116] In the embodiment of the present application, the model loss value loss consists of 3 parts, all of which can use the loss of regression tasks, such as Mean Square Error (MSE) and the like.

[0117] The first part, Loss1, includes the recent cumulative data fitting and future cumulative data fitting of the forward demand forecast. They can be given the same weight. That is, Loss1 = 0.5*Loss1 pass +0.5Loss1 future .

[0118] In the second part, Loss2 calculates the fluctuation of return volume. The corresponding label is the actual future data after normalization. The loss function can be designed according to the conventional future multi-point prediction, and the mean of the prediction deviation at each time point t in the future can be calculated. That is:

[0119]

[0120] The third part measures the change of future distribution. The corresponding label can be obtained by calculating the corresponding mean and variance based on the overall return volume in the historical period. That is, Loss3 = 0.5*Loss1 mean +0.5Loss1 std The overall final Loss = a*Loss1+b*Loss2+c*Loss3, where a, b, and c are hyperparameters of the corresponding weights. Since all three are of the same dimension, the weights of different parts can be adjusted appropriately to ensure the prediction effect of the final model.

[0121] In this embodiment, since in the process of training the return data prediction model, not only the sample return characteristics of the sample return data are considered, but also the sample demand characteristics of the sample demand data are considered, and the first covariate affecting the sample demand data is also considered, therefore, the modeling accuracy of the return data prediction model can be effectively improved, and when the future return data is predicted based on the trained target return data prediction model, the accuracy of the return data prediction can be effectively improved. The demand data is effectively used at the time series granularity to estimate the return data, effectively improving the accuracy and interpretability of the predicted return data. The cumulative demand prediction submodule has two goals, one is to restore the recent sales volume to ensure that the recent characterization can be used as a return reference. On the other hand, the total sales volume for a period of time in the future is predicted at the same time to guide the longer-term return situation. Based on the output characteristics of MLP1 as a whole, two numerical results are output, and supervised learning is performed in the subsequent modules. The overall prediction scheme based on positive demand and reverse return can be combined end-to-end, such as overall model learning, to better introduce positive information for direct learning. At the same time, the model is more interpretable, and the future return situation can be quantitatively explained in combination with positive demand information.

[0122] Optional examples, such as Figure 5 As shown, Figure 5It is an interactive schematic diagram of the return volume prediction of the time series distribution migration in the embodiment of the present application. Including the interactive system and the return volume prediction interactive system of the time series distribution migration, the business personnel can manually input the sample data for training the model, and the sample data includes sample demand data and sample return data, and the sample data can be stored in the data storage module. The return volume prediction interactive system of the time series distribution migration can read the sample demand data and sample return data from the data storage module, and perform data preprocessing on the sample demand data and sample return data to train the model, and predict based on the trained model to obtain the model output, and provide the model output to the visualization module for visual display. The model prediction module can use the latest historical data and future data in the model prediction, and use the model output by the above-mentioned model training module to perform model prediction. The forward demand part can output the future cumulative prediction, and the reverse return part can output the return volume point prediction of the corresponding segment in the future, as well as the mean and standard deviation of the predicted overall segment. The model output module can restore the model data to the original dimension and output it in the interactive system. Since the model considers both positive and negative situations, the overall output of the model has a certain degree of interpretability, including multiple outputs of positive near-term, positive future, and negative return future. At the same time, it can also output the future return rate to assist in commodity management, thus improving the entire process of supply chain management.

[0123] Figure 6 It is a flowchart of a return data prediction method proposed in an embodiment of the present application.

[0124] This embodiment takes the return data prediction method being configured as a return data prediction device as an example. In this embodiment, the return data prediction method can be configured in the return data prediction device, and the return data prediction device can be set in a server, or can also be set in an electronic device, without limitation.

[0125] This embodiment takes the case where the return data prediction method is configured in an electronic device as an example, wherein the electronic device is a hardware device such as a smart phone, a tablet computer, a personal digital assistant, an e-book, etc. having various operating systems.

[0126] like Figure 6 As shown, the return data prediction method includes:

[0127] S601: Determine reference demand data and reference return data.

[0128] S602: Input the reference demand data and the reference return data into the target return data prediction model trained based on the return data prediction model training method to obtain the predicted return data output by the target return data prediction model.

[0129] In this embodiment, by determining reference demand data and reference return data, and inputting the reference demand data and reference return data into a target return data prediction model trained based on a return data prediction model training method, to obtain predicted return data output by the target return data prediction model, the prediction accuracy of the predicted return data can be effectively improved.

[0130] Figure 7 It is a structural diagram of a return data prediction model training device proposed in one embodiment of the present application.

[0131] like Figure 7 As shown, the return data prediction model training device 70 includes:

[0132] The first determination module 701 is used to determine sample demand data and sample return data.

[0133] The second determination module 702 is used to determine a first covariate corresponding to the sample demand data, wherein the first covariate is used to represent a variable condition related to the sample demand data.

[0134] The third determination module 703 is used to determine the sample demand characteristics corresponding to the sample demand data and determine the sample return characteristics corresponding to the sample return data based on the first covariate.

[0135] The processing module 704 is used to process the sample demand features and the sample return features to obtain the sample prediction features.

[0136] The training module 705 is used to train the initial return data prediction model according to the sample demand characteristics, sample return data and sample prediction characteristics to obtain the target return data prediction model.

[0137] It should be noted that the aforementioned explanation of the return data prediction model training method is also applicable to the return data prediction model training device of this embodiment and will not be repeated here.

[0138] In this embodiment, by determining sample demand data and sample return data, and determining a first covariate corresponding to the sample demand data, wherein the first covariate is used to represent the variable status related to the sample demand data, according to the first covariate, determining the sample demand features corresponding to the sample demand data, and determining the sample return features corresponding to the sample return data, processing the sample demand features and the sample return features to obtain sample prediction features, and training an initial return data prediction model according to the sample demand features, the sample return data and the sample prediction features to obtain a target return data prediction model. In the process of training the return data prediction model, not only the sample return features of the sample return data, but also the sample demand features of the sample demand data and the first covariate affecting the sample demand data are considered. Therefore, the modeling accuracy of the return data prediction model can be effectively improved. When predicting future return data based on the trained target return data prediction model, the accuracy of the return data prediction can be effectively improved.

[0139] Figure 8 It is a structural diagram of a return data prediction device proposed in one embodiment of the present application.

[0140] like Figure 8 As shown, the return data prediction device 80 includes:

[0141] The fourth determining module 801 is used to determine reference demand data and reference return data.

[0142] The prediction module 802 is used to input the reference demand data and the reference return data into the target return data prediction model trained by the return data prediction model training device as mentioned above, so as to obtain the predicted return data output by the target return data prediction model.

[0143] It should be noted that the aforementioned explanation of the return data prediction method is also applicable to the return data prediction device of this embodiment and will not be repeated here.

[0144] In this embodiment, by determining reference demand data and reference return data, and inputting the reference demand data and reference return data into a target return data prediction model trained based on a return data prediction model training method, to obtain predicted return data output by the target return data prediction model, the prediction accuracy of the predicted return data can be effectively improved.

[0145] Fig. 9 A block diagram of an exemplary electronic device suitable for implementing the embodiments of the present application is shown. Fig. 9 The electronic device 12 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0146] like Fig. 9As shown, the electronic device 12 is in the form of a general purpose computing device. The components of the electronic device 12 may include, but are not limited to: one or more processors or processing units 16, a memory 28, and a bus 18 that connects various system components (including the memory 28 and the processing unit 16).

[0147] The bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus and Peripheral Component Interconnection (PCI) bus.

[0148] The electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 12, including volatile and non-volatile media, removable and non-removable media.

[0149] The memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache 32. The electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Fig. 9 Not shown, often called a "hard drive").

[0150] although Fig. 9Not shown, a disk drive for reading and writing a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing a removable non-volatile optical disk (e.g., a compact disc read only memory (CD-ROM), a digital versatile disc read only memory (DVD-ROM), or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present application.

[0151] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in the memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0152] The electronic device 12 may also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), may also communicate with one or more devices that enable the human body to interact with the electronic device 12, and / or communicate with any device that enables the electronic device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). Such communication may be performed through an input / output (I / O) interface 22. In addition, the electronic device 12 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 20. As shown, the network adapter 20 communicates with other modules of the electronic device 12 through a bus 18. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0153] The processing unit 16 executes various functional applications and data processing by running the programs stored in the memory 28, such as implementing the sorting method mentioned in the above embodiment.

[0154] In order to implement the above embodiments, the present application also proposes a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the methods provided by the above embodiments.

[0155] In order to implement the above embodiments, the present application also proposes a computer program product, including a computer program, which implements the methods provided by the above embodiments when executed by a processor.

[0156] The collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in this application are in compliance with relevant laws and regulations and do not violate public order and good morals.

[0157] It should be noted that personal information from users should be collected for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. In addition, such collection / sharing should be carried out after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign the agreement / authorization including authorization of relevant user information before the user uses the function. In addition, any necessary steps should be taken to protect and safeguard access to such personal information data and ensure that others who have access to personal information data comply with its privacy policy and procedures.

[0158] This application is expected to provide an implementation scheme for users to selectively block the use or access of personal information data. That is, this application is expected to provide hardware and / or software to prevent or block access to such personal information data. Once the personal information data is no longer needed, risks can be minimized by limiting data collection and deleting data. In addition, when applicable, personal identifiers are removed from such personal information to protect the privacy of users.

[0159] In the description of the aforementioned embodiments, 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 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.

[0160] In addition, the terms "first" and "second" 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" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0161] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0162] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0163] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0164] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0165] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0166] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A return data prediction model training method, characterized in that: The following steps are involved: Determine sample demand data and sample return data; Determine a first covariate corresponding to the sample demand data, wherein the first covariate is used to represent a variable condition related to the sample demand data; Determining, based on the first covariate, a sample demand feature corresponding to the sample demand data, and determining a sample return feature corresponding to the sample return data; Processing the sample demand features and the sample return features to obtain sample prediction features; and An initial return data prediction model is trained according to the sample demand characteristics, the sample return data and the sample prediction characteristics to obtain a target return data prediction model.

2. The method according to claim 1, characterized in that The determining of sample demand data and sample return data includes: Determine an input time sequence, wherein the input time sequence includes: a plurality of input times; Obtain initial demand data and initial return data; Performing instance normalization processing on the initial demand data according to the multiple input times to obtain the sample demand data; The initial return data is instance normalized according to the multiple input times to obtain the sample return data.

3. The method according to claim 2, characterized in that The performing instance normalization processing on the initial demand data according to the multiple input times to obtain the sample demand data includes: Mapping the initial demand data to each of the input times to obtain input demand data corresponding to each of the input times; Determine the length of the input time series and the number of first covariates of each of the input demand data; Determine the mean and standard deviation of the plurality of input demand data according to the length and the number of the first covariates; Each of the input demand data is standardized according to the mean and the standard deviation, and the multiple input demand data obtained by the standardized processing are used as the sample demand data.

4. The method according to claim 2, characterized in that: The performing instance normalization processing on the initial return data according to the multiple input times to obtain the sample return data includes: Mapping the initial return data to each of the input times to obtain input return data corresponding to each of the input times; Determine the length of the input time series and the number of second covariates of each input return data; Determine the mean and standard deviation of the plurality of input return data according to the length and the number of the second covariates; Each of the input return data is standardized according to the mean and the standard deviation, and the multiple input return data obtained by the standardized processing are used as the sample return data.

5. The method according to claim 1, characterized in that The determining, according to the first covariate, a sample demand feature corresponding to the sample demand data includes: Determining initial demand characteristics corresponding to the sample demand data; determining a first covariate feature corresponding to the first covariate; The initial demand feature and the first covariate feature are concatenated, and the concatenated feature is used as the sample demand feature.

6. The method according to claim 1, characterized in that The determining of the sample return feature corresponding to the sample return data includes: Determine a second covariate corresponding to the sample return data, wherein the second covariate is used to represent a variable condition related to the sample return data; Determining initial return features corresponding to the sample return data; determining a second covariate characteristic corresponding to the second covariate; The initial return feature and the second covariate feature are concatenated, and the concatenated feature is used as the sample return feature.

7. The method according to claim 1, characterized in that The processing of the sample demand feature and the sample return feature to obtain a sample prediction feature includes: The sample demand feature and the sample return feature are concatenated, and the concatenated feature is used as the sample prediction feature.

8. The method according to claim 1, characterized in that The initial return data prediction model includes: a cumulative demand prediction submodule and a return quantity distribution prediction submodule; The training of the initial return data prediction model according to the sample demand characteristics, the sample return data and the sample prediction characteristics to obtain the target return data prediction model includes: Inputting the sample demand characteristics into the cumulative demand prediction submodule to obtain predicted cumulative demand data output by the cumulative demand prediction submodule; Inputting the sample prediction features into the return volume distribution prediction submodule to obtain the predicted return volume fluctuation data output by the return volume distribution prediction submodule; The initial return data prediction model is trained for at least one round of iteration according to the predicted cumulative demand data, the sample return data and the predicted return volume fluctuation data until the initial return data prediction model converges, and the initial return data prediction model obtained through training is used as the target return data prediction model.

9. The method according to claim 8, characterized in that The sample return data includes: a plurality of input return data obtained through standardization, each of the input return data corresponds to an input time in the input time series; the predicted return volume fluctuation data includes: predicted return data corresponding to each future time in the future time series; The step of performing at least one round of iterative training on the initial return data prediction model according to the predicted cumulative demand data, the sample return data and the predicted return volume fluctuation data until the initial return data prediction model converges includes: Determining means and standard deviations of the plurality of input return data and the plurality of predicted return data; The initial return data prediction model is iteratively trained according to the predicted cumulative demand data and the mean and standard deviation until the initial return data prediction model converges.

10. The method according to claim 9, characterized in that The iterative training of the initial return data prediction model according to the predicted cumulative demand data and the mean and standard deviation until the initial return data prediction model converges includes: Destandardizing the plurality of predicted return data according to the mean and the standard deviation to obtain a plurality of reference return data; Determine a model loss value according to the predicted cumulative demand data, the plurality of reference return data, the mean value and the standard deviation; If the model loss value is less than or equal to the loss threshold, it is determined that the initial return data prediction model has converged; If the model loss value is greater than the loss threshold, a new round of iterative training is performed on the initial return data prediction model based on the predicted cumulative demand data and the mean and standard deviation.

11. A method for predicting return data, characterized in that: The following steps are involved: Determine reference demand data and reference return data; The reference demand data and the reference return data are input into the target return data prediction model trained by the return data prediction model training method as described in any one of claims 1 to 10 above, so as to obtain the predicted return data output by the target return data prediction model.

12. A return data prediction model training device, characterized in that: include: A first determination module is used to determine sample demand data and sample return data; A second determination module, used to determine a first covariate corresponding to the sample demand data, wherein the first covariate is used to represent a variable situation related to the sample demand data; a third determination module, configured to determine, based on the first covariate, a sample demand feature corresponding to the sample demand data, and determine a sample return feature corresponding to the sample return data; a processing module, configured to process the sample demand feature and the sample return feature to obtain a sample prediction feature; and A training module is used to train an initial return data prediction model according to the sample demand characteristics, the sample return data and the sample prediction characteristics to obtain a target return data prediction model.

13. A return data prediction device, characterized in that: include: A fourth determination module, used to determine reference demand data and reference return data; A prediction module is used to input the reference demand data and the reference return data into the target return data prediction model trained by the return data prediction model training device as described in claim 12 above, so as to obtain the predicted return data output by the target return data prediction model.

14. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 11.

15. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 11 when executed by a processor.

16. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 11 when being executed by a processor.

Images