Clothing sales forecasting method, system and device based on artificial intelligence

Abstract

The present invention provides an artificial intelligence-based clothing sales forecasting method, which includes the following steps: extracting clothing sales data and performing data preparation; identifying distorted sales data in the clothing sales data, and performing data cleaning and data repair; establishing an artificial intelligence-based forecasting algorithm model based on the cleaned clothing sales data, and forecasting the sales trend of the next sales cycle; performing temperature correction; and performing holiday and promotion activity correction.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to an artificial intelligence-based clothing sales forecasting method, system and device. Background Art

[0002] Accurately predicting future sales in the apparel industry, identifying the real demand for different styles of clothing, and achieving precise matching of supply and demand can not only minimize the loss of sales opportunities, but also greatly reduce clothing inventory, which is of great significance to improving the profits of apparel companies.

[0003] Since clothing sales behavior is relatively random, it is not only affected by clothing itself, but also by store factors, holiday factors, promotional activities factors, weather factors, etc. The multi-dimensional influencing factors make it a major challenge to accurately predict clothing sales. The current prediction of clothing sales in the industry is mainly based on its historical sales data for the same period, relying on a relatively single information dimension, with poor prediction accuracy, and is not suitable for new clothing. With the accumulation of historical sales, inventory and other data in the clothing industry, it has become possible to use artificial intelligence methods based on big data to solve the problem of clothing sales prediction. Artificial intelligence methods can process high-dimensional data information and can greatly improve the accuracy of clothing sales prediction.

[0004] There are very few existing technologies for clothing sales forecasting based on artificial intelligence methods such as machine learning. For example, the patent CN110322263A established a clothing sales forecasting model based on sales data and online social media data. On the one hand, the model only used one dimension of sales data on the clothing side, and did not consider the impact of inventory and store information, so the data dimension was relatively small; on the other hand, only macro-level popular elements can be learned from social media data, which has limited significance for predicting the future sales of a specific store or a certain style of clothing. Summary of the invention

[0005] One of the purposes of the present invention is to provide an artificial intelligence-based clothing sales forecasting method, system and device, which can obtain and process store sales, inventory and other data, and correct the sales decline caused by insufficient inventory based on the inventory data, so as to obtain more realistic clothing sales data.

[0006] One of the purposes of the present invention is to provide a clothing sales prediction method, system and equipment based on artificial intelligence, which can take the sales of specific stores and clothing styles as the modeling and forecasting objects, introduce the temperature information of the store location, and extract the multi-dimensional label data (color, style number, length, weight, etc.) of the corresponding style, and use the historical data related to the historical sales of the clothing style itself and similar styles to predict the future sales of the clothing style, while considering holiday information and promotional activity information, and using as much dimensional information as possible to ensure the prediction accuracy.

[0007] One of the purposes of the present invention is to provide a clothing sales forecasting method, system and device based on artificial intelligence, which can effectively guide clothing stores to sell and distribute goods.

[0008] One of the purposes of the present invention is to provide a clothing sales forecasting method, system and device based on artificial intelligence, which can establish an intelligent forecasting model through big data and artificial intelligence methods to solve the problem of clothing sales forecasting for specific models in a single store or a regional store group.

[0009] In order to achieve at least one of the above invention purposes, the present invention provides a clothing sales forecasting method based on artificial intelligence, and the clothing sales forecasting method based on artificial intelligence comprises the following steps:

[0010] Extract apparel sales data and perform data preparation;

[0011] Identify distorted sales data in clothing sales data and perform data cleaning and data repair;

[0012] Based on the sales data of cleaned clothing, an AI-based forecasting algorithm model is established to forecast the sales trend of the next sales cycle;

[0013] Apply temperature corrections; and

[0014] Implement holiday and promotional adjustments.

[0015] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0016] Extract historical sales data of the clothing style to be predicted;

[0017] Extracting forecast reference features; and

[0018] Extract the overall sales characteristics of the product.

[0019] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0020] Execute a judgment on whether the clothing model to be predicted has historical sales data, and when the clothing model to be predicted has historical sales data, extract the historical sales data of the clothing model to be predicted; and

[0021] Execute the judgment on whether the clothing model to be predicted is a new clothing launched in the current year. When the clothing model to be predicted is a new clothing launched in the current year, extract the historical sales data of the model most similar to the clothing model to be predicted as the historical sales data of the clothing model to be predicted.

[0022] In some embodiments, the extracted forecast reference features are sales data of 5-10 clothing models that are similar to the clothing model to be predicted in terms of color, style, thickness, and length.

[0023] In some embodiments, the extracted overall sales features of the goods are the overall sales data of the sub-category and the major category to which the clothing style number to be predicted belongs.

[0024] In some embodiments, the distorted sales data includes data missing, over-limit anomalies, data jumps, or sales data distortion caused by insufficient inventory.

[0025] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0026] Identify and remove data that exceeds the limit;

[0027] Identify and remove jump data; and

[0028] Identify and eliminate anomalous apparel inventory data.

[0029] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0030] Set upper and lower limits for clothing sales data, identify and eliminate data that exceeds the limit.

[0031] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0032] The data at each moment is subtracted from the previous and next data. When the absolute values ​​of the two differences are greater than the set threshold, the value at that moment is identified as jump data and is removed.

[0033] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0034] Obtain clothing inventory data and theoretical sales volume data at the corresponding time, perform data comparison, and when the clothing inventory data is 0, or the inventory data is less than the theoretical sales volume at the corresponding time, eliminate the sales data at that time.

[0035] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0036] Generate sales time series curve;

[0037] Divide sales into multiple stages based on the sales timeline curve; and

[0038] Based on the data of each stage, correction of sales data is performed for each abnormal point.

[0039] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0040] Obtain non-over-limit sales data within the normal range and generate a sales time series curve.

[0041] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0042] According to the sales time series curve, sales are divided into four stages: introduction, growth, maturity and decline;

[0043] Based on the data of the four stages respectively, the sales growth slope data values ​​k1, k2, k3 and k4 of each stage are counted and obtained;

[0044] For each abnormal point, the average sales data value s of the week before each abnormal point is calculated; and

[0045] The slope data value ki of the life cycle of each abnormal point is multiplied by the average sales data value s, and the product is used as the correction value of the sales data of each abnormal point.

[0046] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0047] Obtain the training set and test set of the forecast algorithm model;

[0048] Execute the training of forecast algorithm model;

[0049] Perform validation on forecast algorithm models; and

[0050] Acquire feature data, input it into the forecasting algorithm model, and obtain initial forecasting results.

[0051] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0052] Get year data n, split the historical data of n years into training set and test set of the forecast algorithm model, use the data of the first n-1 years as the test set, and the data of the most recent year as the validation set.

[0053] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0054] The historical sales data of the clothing model to be predicted, the historical sales data of similar models, and the historical sales data of the subcategories and major categories are input into the forecasting algorithm model as features, and the sales results at the corresponding time of the next year's data are output as the target to execute the training of the forecasting algorithm model.

[0055] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0056] The prediction algorithm model is verified using the validation set data, and the root mean square error between the prediction result and the actual result is calculated. When the error meets the requirements, the prediction algorithm model is output. When the error does not meet the requirements, the prediction algorithm model and parameter adjustment are returned and retrained.

[0057] In some embodiments, the artificial intelligence-based clothing sales forecasting method further includes the following steps: obtaining feature data for the most recent year, inputting the data into a forecasting algorithm model, and obtaining initial forecasting results.

[0058] In some embodiments, the machine learning model is selected from an extreme gradient boosting tree model, a random forest model, a support vector machine model, and a neural network model.

[0059] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0060] Obtain historical temperature data and perform average temperature forecasts for the coming weeks;

[0061] Establish a temperature difference correction regression model for the impact of temperature on sales; and

[0062] The established temperature difference correction regression model and the predicted temperature are obtained to correct the initial sales results for the next week.

[0063] In some embodiments, the artificial intelligence-based clothing sales forecasting method also includes the following steps: obtaining temperature data for one or more months before the current moment, as well as temperature data for the same period of 5-10 years in the past year, and forecasting the weekly average temperature for the past month, wherein the algorithm time series model is selected from a long short-term memory model, a gated recurrent unit, and a differential integrated moving average autoregressive model.

[0064] In some embodiments, the artificial intelligence-based clothing sales forecasting method also includes the following steps: obtaining historical temperature data and sales data, and establishing a temperature difference correction regression model for the impact of temperature on sales, wherein the algorithm model is selected from a support vector machine model, a Ridge regression model, and a Lasso regression model.

[0065] In some embodiments, the clothing sales forecasting method based on artificial intelligence further includes the following steps:

[0066] Establish sales gain models caused by holidays and promotional activities;

[0067] Calculate sales gains from holidays and promotions; and

[0068] Capture sales gains and revise initial forecast results.

[0069] In some embodiments, the artificial intelligence-based clothing sales forecasting method further includes the following steps: obtaining the activity content and sales data of historical holidays and promotional activities, and establishing a sales gain model caused by holidays and promotional activity scenarios.

[0070] In some embodiments, the artificial intelligence-based clothing sales forecasting method further includes the following steps: obtaining holiday information and promotion strategy information within one or more days in the future and calculating the sales gain brought by holidays and promotional activities based on a sales gain model.

[0071] According to another aspect of the present invention, there is also provided an artificial intelligence-based clothing sales forecasting device, which includes: a software application, a memory for storing the software application, and a processor for executing the software application, wherein each program of the software application correspondingly executes the steps in the artificial intelligence-based clothing sales forecasting method.

[0072] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the clothing sales forecasting method based on artificial intelligence are performed.

[0073] According to another aspect of the present invention, there is also provided a clothing sales forecasting system based on artificial intelligence, the clothing sales forecasting system based on artificial intelligence comprising:

[0074] A clothing sales data extraction subsystem configured to: extract clothing sales data and perform data preparation;

[0075] A data cleaning and data repair subsystem, which is configured to: identify distorted sales data in clothing sales data, and perform data cleaning and data repair;

[0076] A forecasting algorithm model subsystem, which is configured to: establish a forecasting algorithm model based on artificial intelligence based on the sales data of the washed clothing, and forecast the sales trend of the next sales cycle;

[0077] An air temperature correction subsystem configured to: perform air temperature correction; and

[0078] The holiday and promotion activity correction subsystem is configured to perform holiday and promotion activity correction.

[0079] In some embodiments, the clothing sales data extraction subsystem also includes a historical sales data extraction unit for clothing models to be predicted, a forecast reference feature extraction unit and a sales feature extraction unit, wherein the historical sales data extraction unit for clothing models to be predicted is configured to: extract historical sales data of clothing models to be predicted; the forecast reference feature extraction unit is configured to extract forecast reference features; and the sales feature extraction unit is configured to extract overall sales features of the product.

[0080] In some embodiments, the historical sales data extraction unit for the clothing model to be predicted is further configured to: determine whether the clothing model to be predicted has historical sales data, and when the clothing model to be predicted has historical sales data, extract the historical sales data of the clothing model to be predicted; and determine whether the clothing model to be predicted is a new product of the year, and when the clothing model to be predicted is a new product launched that year, extract the historical sales data of the model most similar to the clothing model to be predicted as the historical sales data of the clothing model to be predicted.

[0081] In some embodiments, the extracted forecast reference features are sales data of 5-10 clothing models that are similar to the clothing model to be predicted in terms of color, style, thickness, and length.

[0082] In some embodiments, the extracted overall sales features of the goods are the overall sales data of the sub-category and the major category to which the clothing style number to be predicted belongs.

[0083] In some embodiments, the distorted sales data includes data missing, over-limit anomalies, data jumps, or sales data distortion caused by insufficient inventory.

[0084] In some embodiments, the data cleaning and data repair subsystem includes an out-of-limit data identification and elimination unit, a jump data identification and elimination unit, and an abnormal inventory data identification and elimination unit, wherein the out-of-limit data identification and elimination unit is configured to: identify and eliminate out-of-limit data; the jump data identification and elimination unit is configured to identify and eliminate jump data; the abnormal inventory data identification and elimination unit is configured to identify and eliminate abnormal clothing inventory data.

[0085] In some embodiments, the over-limit data identification and elimination unit is further configured to: set upper and lower limits for clothing sales data, identify over-limit data, and eliminate the over-limit data.

[0086] In some embodiments, the jump data identification and elimination unit is further configured to: subtract the data at each moment from the previous and next data respectively, and when the absolute values ​​of the two differences are both greater than a set threshold, identify the value at that moment as jump data and eliminate it.

[0087] In some embodiments, the abnormal inventory data identification and elimination unit is further configured to: obtain clothing inventory data and theoretical sales volume data at a corresponding moment, perform data comparison, and when the clothing inventory data is 0, or the inventory data is less than the theoretical sales volume at a corresponding moment, eliminate the sales data at that moment.

[0088] In some embodiments, the data cleaning and data repair subsystem also includes a sales timing curve generation unit, a sales stage division unit and an abnormal point sales data correction unit, wherein the sales timing curve generation unit is configured to: generate a sales timing curve; the sales stage division unit is configured to: divide the sales into multiple stages according to the sales timing curve; the abnormal point sales data correction unit is configured to: perform sales data correction for each abnormal point based on the data of each stage.

[0089] In some embodiments, the sales timing curve generating unit is further configured to: obtain non-overlimit sales data within a normal range and generate a sales timing curve.

[0090] In some embodiments, the sales stage division unit is further configured to: divide sales into four stages: introduction, growth, maturity, and decline according to the sales timing curve; count and obtain the sales growth slope data values ​​k1, k2, k3, and k4 of each stage based on the respective data of the four stages; for each abnormal point, count the average sales data value s of each abnormal point in the week before; and multiply the slope data value ki of the life cycle of each abnormal point by the average sales data value s, and use the product as the correction value of the sales data of each abnormal point.

[0091] In some embodiments, the forecasting algorithm model subsystem includes a forecasting algorithm model training unit, a forecasting algorithm model verification unit and an initial forecasting result unit, wherein the forecasting algorithm model training unit is configured to: obtain a training set and a test set of the forecasting algorithm model and perform training of the forecasting algorithm model; the forecasting algorithm model verification unit is configured to: perform verification on the forecasting algorithm model; and the initial forecasting result unit is configured to: obtain feature data, input it into the forecasting algorithm model, and obtain an initial forecasting result.

[0092] In some embodiments, the forecast algorithm model training unit is further configured to: obtain year data n, split the historical data of n years into a training set and a test set of the forecast algorithm model, use the data of the previous n-1 years as the test set, and use the data of the most recent year as the verification set.

[0093] In some embodiments, the forecasting algorithm model training unit is further configured as follows: the historical sales data of the clothing style to be predicted, the historical sales data of similar styles, and the historical sales data of the corresponding subcategories and major categories are input into the forecasting algorithm model as features, and the sales results at the corresponding moment of the next year's data are output as targets to execute the training of the forecasting algorithm model.

[0094] In some embodiments, the forecasting algorithm model verification unit is further configured to: use the verification set data to verify the forecasting algorithm model, calculate the root mean square error between the forecast result and the actual result, and when the error meets the requirements, output the forecasting algorithm model; when the error does not meet the requirements, return to execute the forecasting algorithm model and adjust parameters, and retrain.

[0095] In some embodiments, the initial forecast result unit is further configured to: obtain characteristic data of the most recent year, input the data into the forecast algorithm model, and obtain initial forecast results.

[0096] In some embodiments, the machine learning model is selected from an extreme gradient boosting tree model, a random forest model, a support vector machine model, and a neural network model.

[0097] In some embodiments, the temperature correction subsystem includes a temperature difference correction regression model establishment unit and an initial sales result correction unit, wherein the temperature difference correction regression model establishment unit is configured to: obtain historical temperature data and perform average temperature forecasts for future weeks; and establish a temperature difference correction regression model for the impact of temperature on sales; the initial sales result correction unit is configured to: obtain the established temperature difference correction regression model and the forecasted temperature, and perform corrections on the initial sales results for the future week.

[0098] In some embodiments, the temperature difference correction regression model building unit is further configured to: obtain temperature data for one or more months before the current moment, and temperature data for the same period of 5-10 years in the past year, and perform a forecast for the weekly average temperature for the next month, wherein the algorithmic time series model is selected from a long short-term memory model, a gated recurrent unit, and a differential integrated moving average autoregressive model.

[0099] In some embodiments, the temperature difference correction regression model building unit is further configured to: obtain historical temperature data and sales data, and build a temperature difference correction regression model of the impact of temperature on sales, wherein the algorithm model is selected from a support vector machine model, a Ridge regression model, and a Lasso regression model.

[0100] In some embodiments, the holiday and promotional activity correction subsystem includes a sales gain model establishment unit, a sales gain calculation unit and an initial forecast result correction unit, wherein the sales gain model establishment unit is configured to: establish a sales gain model caused by holiday and promotional activity scenarios; the sales gain calculation unit is configured to: calculate the sales gain brought by holidays and promotional activities; the initial forecast result correction unit is configured to: obtain sales gain and correct the initial forecast result.

[0101] In some embodiments, the sales gain model building unit is further configured to: obtain activity content and sales data of historical holidays and promotional activities, and build a sales gain model caused by holiday and promotional activity scenarios.

[0102] In some embodiments, the sales gain calculation unit is further configured to obtain holiday information and promotion strategy information within one or more days in the future and calculate the sales gain brought by holidays and promotion activities based on the sales gain model.

[0103] According to another aspect of the present invention, there is also provided an artificial intelligence-based clothing sales forecasting device, which includes a clothing sales data extraction device, a data cleaning and data repair device, a forecasting algorithm model device, a temperature correction device, and a holiday and promotional activity correction device, wherein the clothing sales data extraction device is used to extract clothing sales data and perform data preparation; the data cleaning and data repair device is used to identify distorted sales data in clothing sales data, and perform data cleaning and data repair; the forecasting algorithm model device is used to establish an artificial intelligence-based forecasting algorithm model based on the cleaned clothing sales data, and forecast the sales trend of the next sales cycle; the temperature correction device is used to perform temperature correction; the holiday and promotional activity correction device is used to perform holiday and promotional activity correction.

[0104] In some embodiments, the clothing sales data extraction device also includes a historical sales data extraction module for the clothing style to be predicted, a forecast reference feature extraction module and a sales feature extraction module, wherein the historical sales data extraction module for the clothing style to be predicted is used to extract the historical sales data of the clothing style to be predicted; the forecast reference feature extraction module is used to extract forecast reference features; and the sales feature extraction module is used to extract the overall sales features of the product.

[0105] In some embodiments, the historical sales data extraction module for the clothing model to be predicted is also used to: determine whether the clothing model to be predicted has historical sales data, and when the clothing model to be predicted has historical sales data, extract the historical sales data of the clothing model to be predicted; and determine whether the clothing model to be predicted is a new product of the year, and when the clothing model to be predicted is a new product launched that year, extract the historical sales data of the model most similar to the clothing model to be predicted as the historical sales data of the clothing model to be predicted.

[0106] In some embodiments, the extracted forecast reference features are sales data of 5-10 clothing models that are similar to the clothing model to be predicted in terms of color, style, thickness, and length.

[0107] In some embodiments, the extracted overall sales features of the goods are the overall sales data of the sub-category and the major category to which the clothing style number to be predicted belongs.

[0108] In some embodiments, the distorted sales data includes data missing, over-limit anomalies, data jumps, or sales data distortion caused by insufficient inventory.

[0109] In some embodiments, the data cleaning and data repair device includes an over-limit data identification and elimination module, a jump data identification and elimination module, and an abnormal inventory data identification and elimination module. The over-limit data identification and elimination module is used to: identify and eliminate over-limit data; the jump data identification and elimination module is used to identify and eliminate jump data; the abnormal inventory data identification and elimination module is used to identify and eliminate abnormal clothing inventory data.

[0110] In some embodiments, the over-limit data identification and elimination module is further used to: set upper and lower limits for clothing sales data, identify over-limit data, and eliminate the over-limit data.

[0111] In some embodiments, the jump data identification and elimination module is also used to: subtract the data at each moment from the previous and next data respectively, and when the absolute values ​​of the two differences are greater than a set threshold, the value at that moment is identified as jump data and eliminated.

[0112] In some embodiments, the abnormal inventory data identification and elimination module is also used to: obtain clothing inventory data and theoretical sales volume data at the corresponding time, perform data comparison, and when the clothing inventory data is 0, or the inventory data is less than the theoretical sales volume at the corresponding time, eliminate the sales data at that time.

[0113] In some embodiments, the data cleaning and data repair device also includes a sales timing curve generation module, a sales stage division module and an abnormal point sales data correction module, wherein the sales timing curve generation module is used to: generate a sales timing curve; the sales stage division module is used to: divide the sales into multiple stages according to the sales timing curve; the abnormal point sales data correction module is used to: perform sales data correction for each abnormal point based on the data of each stage.

[0114] In some embodiments, the sales timing curve generation module is further used to obtain non-overlimit sales data within a normal range and generate a sales timing curve.

[0115] In some embodiments, the sales stage division module is also used to: divide sales into four stages: introduction, growth, maturity, and decline according to the sales timing curve; count and obtain the sales growth slope data values ​​k1, k2, k3, and k4 of each stage based on the respective data of the four stages; for each abnormal point, count the average sales data value s of each abnormal point in the week before; and multiply the slope data value ki of the life cycle of each abnormal point by the average sales data value s, and use the product as the correction value of the sales data of each abnormal point.

[0116] In some embodiments, the forecasting algorithm model device includes a forecasting algorithm model training module, a forecasting algorithm model verification module and an initial forecasting result module, wherein the forecasting algorithm model training module is used to obtain a training set and a test set of the forecasting algorithm model and to perform training of the forecasting algorithm model; the forecasting algorithm model verification module is used to perform verification on the forecasting algorithm model; and the initial forecasting result module is used to obtain feature data, input it into the forecasting algorithm model, and obtain an initial forecasting result.

[0117] In some embodiments, the forecast algorithm model training module is also used to: obtain year data n, split the historical data of n years into a training set and a test set of the forecast algorithm model, use the data of the previous n-1 years as the test set, and use the data of the most recent year as the verification set.

[0118] In some embodiments, the forecasting algorithm model training module is also used for: the historical sales data of the clothing style to be predicted, the historical sales data of similar styles, and the historical sales data of the corresponding subcategories and major categories are input into the forecasting algorithm model as features, and the sales results at the corresponding time of the next year's data are output as targets to execute the training of the forecasting algorithm model.

[0119] In some embodiments, the forecasting algorithm model verification module is also used to: use the verification set data to verify the forecasting algorithm model, calculate the root mean square error between the forecast result and the actual result, and when the error meets the requirements, output the forecasting algorithm model; when the error does not meet the requirements, return to execute the forecasting algorithm model and adjust parameters, and retrain.

[0120] In some embodiments, the initial forecast result module is further used to obtain characteristic data of the most recent year, input the data into the forecast algorithm model, and obtain initial forecast results.

[0121] In some embodiments, the machine learning model is selected from an extreme gradient boosting tree model, a random forest model, a support vector machine model, and a neural network model.

[0122] In some embodiments, the temperature correction device includes a temperature difference correction regression model establishment module and an initial sales result correction module, the temperature difference correction regression model establishment module is used to: obtain historical temperature data and perform average temperature forecasts for future weeks; and establish a temperature difference correction regression model for the impact of temperature on sales; the initial sales result correction module is used to: obtain the established temperature difference correction regression model and the forecasted temperature, and perform corrections on the initial sales results for the next week.

[0123] In some embodiments, the temperature difference correction regression model building module is also used to obtain temperature data for one or more months before the current moment, as well as temperature data for the same period of 5-10 years in the past year, and to forecast the average weekly temperature for the past month, wherein the algorithm time series model is selected from a long short-term memory model, a gated loop module, and a differential integrated moving average autoregressive model.

[0124] In some embodiments, the temperature difference correction regression model establishment module is also used to: obtain historical temperature data and sales data, and establish a temperature difference correction regression model of the impact of temperature on sales, wherein the algorithm model is selected from a support vector machine model, a Ridge regression model, and a Lasso regression model.

[0125] In some embodiments, the holiday and promotional activity correction device includes a sales gain model establishment module, a sales gain calculation module and an initial forecast result correction module, wherein the sales gain model establishment module is used to: establish a sales gain model caused by holiday and promotional activity scenarios; the sales gain calculation module is used to: calculate the sales gain brought by holidays and promotional activities; the initial forecast result correction module is used to: obtain sales gain and correct the initial forecast result.

[0126] In some embodiments, the sales gain model building module is further used to: obtain activity content and sales data of historical holidays and promotional activities, and build a sales gain model caused by holiday and promotional activity scenarios.

[0127] In some embodiments, the sales gain calculation module is further used to obtain holiday information and promotion strategy information within one or more days in the future and calculate the sales gain brought by holidays and promotion activities based on the sales gain model. BRIEF DESCRIPTION OF THE DRAWINGS

[0128] Figure 1 The present invention is a flowchart of the steps of a clothing sales forecasting method based on artificial intelligence according to an embodiment of the present invention.

[0129] Figure 2 It is a curve diagram of a clothing sales forecasting device based on artificial intelligence according to an embodiment of the present invention.

[0130] Figure 3 It is a curve diagram of the clothing sales prediction device based on artificial intelligence according to the above embodiment of the present invention.

[0131] Figure 4 It is a curve diagram of the clothing sales prediction device based on artificial intelligence according to the above embodiment of the present invention.

[0132] Figure 5 It is a curve diagram of the clothing sales prediction device based on artificial intelligence according to the above embodiment of the present invention.

[0133] Figure 6 It is a curve diagram of the clothing sales prediction device based on artificial intelligence according to the above embodiment of the present invention.

[0134] Figure 7 It is a curve diagram of the clothing sales prediction device based on artificial intelligence according to the above embodiment of the present invention.

[0135] Figure 8 It is a curve diagram of the clothing sales prediction device based on artificial intelligence according to the above embodiment of the present invention.

[0136] Fig. 9 It is a curve diagram of the clothing sales prediction device based on artificial intelligence according to the above embodiment of the present invention. DETAILED DESCRIPTION

[0137] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0138] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the element may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0139] The present invention is an invention related to a computer program. Figure 1 The figure shows a flowchart of a clothing sales forecasting method based on artificial intelligence according to the present invention, which explains a solution for solving the problem proposed by the present invention, based on the computer program processing flow, by executing a computer program compiled according to the above flow to control or process the external or internal objects of the computer. Through the clothing sales forecasting method based on artificial intelligence of the present invention, it is possible to utilize a computer system, integrate manual experience and machine learning results, and improve analysis efficiency, accuracy and universality. It can be understood that the "computer" referred to in the present invention not only refers to desktop computers, laptops, tablets and other devices, but also includes other intelligent electronic devices that can run according to the program and process data.

[0140] The artificial intelligence-based clothing sales forecasting method of the present invention can establish an intelligent forecasting model through big data and artificial intelligence methods to solve the problem of forecasting clothing sales of specific models in a single store or a regional store group.

[0141] Those skilled in the art will appreciate that the "clothing sales data" referred to in the present invention does not only include clothing sales data in a narrow sense, but refers to data related to clothing sales, such as clothing sales data, clothing inventory data, etc.

[0142] Specifically, in a preferred embodiment of the present invention, Figure 1 As shown, the clothing sales forecasting method based on artificial intelligence includes the following steps:

[0143] S100: Extract clothing sales data, including sales volume, inventory data, etc.

[0144] S200: Identify distorted sales data in clothing sales data, and perform data cleaning and data repair.

[0145] S300: Based on the sales data of cleaned clothing, an artificial intelligence-based forecasting algorithm model is established to forecast the sales trend of the next sales cycle.

[0146] S400: Air temperature correction.

[0147] S500: Holiday and promotional event corrections.

[0148] Furthermore, in the step S100, the extracted clothing sales data is clothing sales-related data, including clothing sales data and clothing inventory data, etc., which is a data preparation step.

[0149] Specifically, the step S100 further includes the following steps:

[0150] S110: Execute a judgment on whether the clothing model to be predicted has historical sales data, and when the clothing model to be predicted has historical sales data, extract the historical sales data of the clothing model to be predicted;

[0151] S120: Execute the judgment on whether the clothing model to be predicted is a new clothing model launched in the current year. When the clothing model to be predicted is a new clothing model launched in the current year, extract the historical sales data of the model most similar to the clothing model to be predicted;

[0152] S130: extracting sales data of 5-10 clothing models that are similar to the clothing model to be predicted in terms of color, style, thickness, and length as reference features for prediction; and

[0153] S140: Extract the overall sales data of the sub-category and the major category of the clothing model number to be predicted as the overall sales characteristics of this type of goods. Among them, the sub-category includes but is not limited to white short-sleeved shirts, long dresses, thin down jackets, etc., and the major category includes but is not limited to short-sleeved shirts, dresses, down jackets, etc.

[0154] Furthermore, in step S200, distorted sales data in the data is identified, and data cleaning and data repair are performed. The distorted sales data includes but is not limited to some data missing, over-limit abnormalities, data jumps, or sales data distortion caused by insufficient inventory. Specifically, step S200 also includes the following steps:

[0155] S211: setting upper and lower limits for clothing sales data, identifying over-limit data, and eliminating the over-limit data;

[0156] S212: Subtract the data at each moment from the previous and next data respectively. When the absolute values ​​of the two differences are both greater than a set threshold, the value at that moment is identified as jump data and is removed; and

[0157] S213: Acquire clothing inventory data and theoretical sales data at corresponding moments, and perform data comparison. When the clothing inventory data is 0, or the inventory data is less than the theoretical sales volume at the corresponding moment (such as the weekly average sales volume), the sales data at that moment will be eliminated. This is because the sales data of the day cannot reflect the actual sales demand of the product. Therefore, the algorithm inside the computer program will eliminate the sales data at that moment, thereby improving the accuracy.

[0158] After the above abnormal data is removed, in order to reflect the real sales situation, the missing data needs to be repaired. Specifically, the step S200 also includes the following steps:

[0159] S221: Acquire non-over-limit sales data within a normal range and generate a sales time series curve;

[0160] S222: According to the sales time series curve, sales are divided into four stages: introduction stage, growth stage, maturity stage and decline stage;

[0161] S223: Based on the data of the four stages respectively, count and obtain the sales growth slope data values ​​k1, k2, k3 and k4 of each stage;

[0162] S224: for each abnormal point, the average sales data value s of the abnormal point in the week before is counted; and

[0163] S225: Multiply the slope data value ki of the life cycle where each abnormal point is located by the average sales data value s, and use the product as the correction value of the sales data of each abnormal point.

[0164] It is worth mentioning that sales behavior has obvious periodicity on a long time scale. Generally, the sales trend of clothing will go through four stages in its sales life cycle: introduction, growth, maturity, and decline. Therefore, in the step of performing abnormal data repair in the clothing sales prediction method based on artificial intelligence of the present invention, sales are divided into four stages: introduction, growth, maturity, and decline according to the sales timing curve.

[0165] It is worth mentioning that since the sales data between days are random, the repair of sales data cannot be done by simple linear interpolation. In the step of performing abnormal data repair in the clothing sales forecasting method based on artificial intelligence of the present invention, it does not adopt simple interpolation, but combines the sales business scenario and considers the data change slope of the entire cycle in which the data point is located, which can improve the final forecast accuracy in combination with the sales business scenario.

[0166] Furthermore, in the step S300, the following steps are also included:

[0167] S310: Obtain year data n, split the historical data of n years into a training set and a test set of the forecast algorithm model, the data of the previous n-1 years is used as the test set, and the data of the most recent year is used as the verification set;

[0168] S320: The historical sales data of the clothing model to be predicted, the historical sales data of similar models, and the historical sales data of the sub-category and the major category are input as features into the forecasting algorithm model, and the sales results at the corresponding time of the next year's data are output as the target, and the training of the forecasting algorithm model is performed, wherein the optional machine learning models include but are not limited to the extreme gradient boosting tree model (XGB), the random forest model (RF), the support vector machine model (SVM), the neural network model, etc.;

[0169] S330: Use the validation set data to validate the forecast algorithm model, calculate the root mean square error between the forecast result and the actual result, and when the error meets the requirement, output the forecast algorithm model; when the error does not meet the requirement, return to step S320 to execute the forecast algorithm model and parameter adjustment, and retrain;

[0170] S340: Obtain the characteristic data of the most recent year and input it into the forecasting algorithm model to forecast the sales results of the next year and obtain the initial forecast results.

[0171] It is worth mentioning that the forecasting algorithm model in the artificial intelligence-based clothing sales forecasting method of the present invention not only considers the sales data of the target style number as a feature, but also introduces the sales data of similar style number products and the sales data of the corresponding subcategories and major categories, enriching the dimension of the features and making the forecasting results more reliable.

[0172] Furthermore, in the step S400, the following steps are also included:

[0173] S410: Obtain temperature data for one or more months before the current moment, as well as temperature data for the same period of 5-10 years in the future month, and perform a forecast for the average weekly temperature in the future month, wherein the optional algorithm models include but are not limited to time series models such as long short-term memory model (LSTM), gated recurrent unit (GRU), and differential integrated moving average autoregressive model (ARIMA);

[0174] S420: Obtain historical temperature data and sales data, and establish a temperature difference correction regression model for the impact of temperature on sales, where the models that can be used include but are not limited to support vector machine (SVM), Ridge regression, Lasso regression, etc.;

[0175] S430: Obtain the established temperature difference correction regression model and the predicted temperature, and perform correction on the initial sales results for the next week.

[0176] It is worth mentioning that since the sales trend of many garments is obviously affected by the temperature, and the fluctuation characteristics of the temperature from year to year are also different, the clothing sales forecasting method based on artificial intelligence of the present invention makes temperature corrections to the initial forecast results. In actual use, the sales forecast results can be corrected in real time, and the forecast results for the next week can be periodically corrected during the sales cycle, for example, corrections can be performed every week.

[0177] It is worth mentioning that the artificial intelligence-based clothing sales forecasting method of the present invention takes into account the impact of temperature differences between years by executing a temperature correction step, which is helpful to correct the deviation of using historical data from previous years to forecast current year sales.

[0178] Furthermore, in the step S500, the following steps are also included:

[0179] S510: Obtain the activity content and sales data of historical holidays and promotional activities, and establish a sales gain model caused by holiday and promotional activity scenarios, wherein the promotional activities include but are not limited to discounts, full discounts, etc.;

[0180] S520: Obtain holiday information and promotion strategy information for one or more days in the future and calculate sales gains brought by holidays and promotion activities based on a sales gain model;

[0181] S530: Obtain sales gain and revise the initial forecast result.

[0182] It is worth mentioning that since holidays and promotional activities have a significant impact on the trend of clothing sales, for these scenarios, the artificial intelligence-based clothing sales forecasting method of the present invention corrects the original sales forecast results based on holiday characteristics or the strength of promotional activities.

[0183] It is worth mentioning that the artificial intelligence-based clothing sales prediction method of the present invention takes into account the impact of holidays and different promotional activity scenarios on sales results, so that this patent can be applied to sales forecasts in holidays and promotional activity scenarios.

[0184] It is worth mentioning that the artificial intelligence-based clothing sales forecasting method of the present invention first predicts the overall trend and then makes local corrections. Specifically, the smoothed sales curve is first predicted, and then local corrections are made for short-term temperature effects, holidays and promotional activities, which can ensure the final forecast accuracy.

[0185] According to another aspect of the present invention, there is also provided a clothing sales forecasting device based on artificial intelligence, the clothing sales forecasting device based on artificial intelligence comprising: a software application, a memory for storing the software application, and a processor for executing the software application. Each program of the software application can correspondingly execute the steps in the clothing sales forecasting method based on artificial intelligence of the present invention.

[0186] It will be understood by those skilled in the art that the artificial intelligence-based clothing sales forecasting device can be embodied as a desktop computer, a notebook, a mobile smart device, etc., but the foregoing is only for example and also includes other intelligent analysis devices equipped with the software application of the present invention.

[0187] The present invention may be embedded in a computer program product, which includes all the features that enable the method described herein to be implemented. The computer program product is contained in one or more computer-readable storage media, and the computer-readable storage medium has a computer-readable program code contained therein. According to another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the artificial intelligence-based clothing sales forecasting method of the present invention are executed. Computer storage media is a medium in a computer memory for storing a certain discontinuous physical quantity. Computer storage media include, but are not limited to, semiconductors, disk storage, magnetic cores, magnetic drums, magnetic tapes, laser disks, etc. It can be understood by those skilled in the art that computer storage media are not limited to the aforementioned examples, and the aforementioned examples are merely examples and are not limited to the present invention.

[0188] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided in the form of methods, systems or computer program products. Therefore, the present invention may take the form of an all-hardware embodiment, an all-software embodiment, or a combination of software and hardware embodiments. A typical combination of hardware and software may be a general-purpose computer system with a computer program, which, when loaded and executed, controls the computer system so that the artificial intelligence-based clothing sales forecasting method disclosed in the present invention may be executed.

[0189] It will be appreciated by those skilled in the art that the artificial intelligence-based clothing sales forecasting method of the present invention can be implemented by hardware, software, or a combination of hardware and software. The present invention can be implemented in a centralized manner in at least one computer system, or in a decentralized manner by different parts distributed in several interconnected computer systems. Any computer system or other device that can implement the method is applicable. The combination of common hardware and software can be a general-purpose computer system with a computer program installed, which controls the computer system by installing and executing the program to make it run according to the method.

[0190] Corresponding to the embodiment of the method of the present invention, according to another aspect of the present invention, an artificial intelligence-based clothing sales forecasting system is also provided, and the artificial intelligence-based clothing sales forecasting system is the application of the artificial intelligence-based clothing sales forecasting method of the present invention on the improvement of computer programs.

[0191] Specifically, the artificial intelligence-based clothing sales forecasting system includes a clothing sales data extraction subsystem, a data cleaning and data repair subsystem, a forecasting algorithm model subsystem, a temperature correction subsystem, and a holiday and promotion activity correction subsystem.

[0192] Specifically, the clothing sales data extraction subsystem is configured to: extract clothing sales data and perform data preparation. More specifically, the clothing sales data extraction subsystem also includes a historical sales data extraction unit for clothing models to be predicted, a forecast reference feature extraction unit, and a sales feature extraction unit. The historical sales data extraction unit for clothing models to be predicted is configured to perform a judgment on whether the clothing models to be predicted have historical sales data. When the clothing models to be predicted have historical sales data, the historical sales data of the clothing models to be predicted are extracted, and the judgment on whether the clothing models to be predicted are new clothing products of the year is performed. When the clothing models to be predicted are new clothing products launched that year, the historical sales data of the models most similar to the clothing models to be predicted are extracted. The forecast reference feature extraction unit is configured to extract sales data of 5-10 clothing models that are similar to the clothing models to be predicted in terms of color, style, thickness, and length as reference features for forecasting. The sales feature extraction unit is configured to extract the overall sales data of the sub-category and the major category to which the clothing models to be predicted belong as the overall sales features of such commodities.

[0193] Specifically, the data cleaning and data repair subsystem is configured to: identify distorted sales data in clothing sales data, and perform data cleaning and data repair. The data cleaning and data repair subsystem includes an over-limit data identification and elimination unit, a jump data identification and elimination unit, an inventory abnormal data identification and elimination unit, a sales time series curve generation unit, a sales stage division unit, and an abnormal point sales data correction unit. The over-limit data identification and elimination unit is configured to set upper and lower limits for clothing sales data, identify over-limit data, and eliminate the over-limit data. The jump data identification and elimination unit is configured to make a difference between the data at each moment and the previous and subsequent data respectively. When the absolute values ​​of the two differences are greater than the set threshold value, the value at this moment is identified as jump data and eliminated. The inventory abnormal data identification and elimination unit is configured to obtain clothing inventory data and theoretical sales volume data at the corresponding moment, perform data comparison, and when the clothing inventory data is 0, or the inventory data is less than the theoretical sales volume at the corresponding moment, the sales data at this moment is eliminated. The sales time series curve generation unit is configured to obtain non-overlimit sales data within a normal range and generate a sales time series curve. The sales stage division unit is configured to divide sales into four stages: introduction stage, growth stage, maturity stage, and decline stage according to the sales time series curve. The abnormal point sales data correction unit is configured to count and obtain the sales growth slope data values ​​k1, k2, k3 and k4 of each stage based on the respective data of the four stages; for each abnormal point, count the average sales data value s of each abnormal point in the week before; and multiply the slope data value ki of the life cycle of each abnormal point by the average sales data value s, and use the product as the correction value of the sales data of each abnormal point.

[0194] Specifically, the forecast algorithm model subsystem includes a forecast algorithm model training unit, a forecast algorithm model verification unit and an initial forecast result unit. The forecast algorithm model training unit is configured to: obtain year data n, split the historical data of n years into a training set and a test set of the forecast algorithm model, the data of the previous n-1 years is used as a test set, and the data of the most recent year is used as a verification set; the historical sales data of the clothing model number to be predicted, the historical sales data of similar models, and the historical sales data of the sub-category and the major category are input into the forecast algorithm model as features, and the sales results at the corresponding time of the next year's data are output as targets to perform the training of the forecast algorithm model. The forecast algorithm model verification unit is configured to: perform verification on the forecast algorithm model using the verification set data, calculate the root mean square error between the forecast result and the actual result, and output the forecast algorithm model when the error meets the requirements, and return to the forecast algorithm model training unit to perform the forecast algorithm model and parameter adjustment when the error does not meet the requirements, and retrain. The initial forecast result unit is configured to: obtain feature data of the most recent year, input it into the forecast algorithm model, perform forecasting on the sales results of the next year, and obtain initial forecast results.

[0195] Specifically, the temperature correction subsystem includes a temperature difference correction regression model establishment unit and an initial sales result correction unit. The temperature difference correction regression model establishment unit is configured to: obtain temperature data for one or more months before the current moment, and temperature data for the same period of 5-10 years in the future month, forecast the average weekly temperature for the future month, obtain historical temperature data and sales data, and establish a temperature difference correction regression model for the impact of temperature on sales. The initial sales result correction unit is configured to: obtain the established temperature difference correction regression model and the predicted temperature, and perform corrections on the initial sales results for the future week.

[0196] Specifically, the holiday and promotional activity correction subsystem includes a sales gain model establishment unit, a sales gain calculation unit, and an initial forecast result correction unit. The sales gain model establishment unit is configured to: obtain the activity content and sales data of historical holidays and promotional activities, and establish a sales gain model caused by holiday and promotional activity scenarios. The sales gain calculation unit is configured to: obtain holiday situation information and promotion strategy information within one or more days in the future and calculate the sales gain brought by holidays and promotional activities based on the sales gain model. The initial forecast result correction unit is configured to: obtain sales gain and correct the initial forecast result.

[0197] like Figures 2 to 9As shown, according to another aspect of the present invention, there is also provided an artificial intelligence-based clothing sales forecasting device, which includes a clothing sales data extraction device, a data cleaning and data repair device, a forecasting algorithm model device, a temperature correction device, and a holiday and promotion activity correction device.

[0198] Specifically, the clothing sales data extraction device includes a historical sales data extraction module for the clothing model to be predicted, a forecast reference feature extraction module, and a sales feature extraction module. The historical sales data extraction module for the clothing model to be predicted is used to determine whether the clothing model to be predicted has historical sales data. When the clothing model to be predicted has historical sales data, the historical sales data of the clothing model to be predicted is extracted, and the judgment is performed on whether the clothing model to be predicted is a new product of the year. When the clothing to be predicted is a new product launched that year, the historical sales data of the model most similar to the clothing model to be predicted is extracted. The forecast reference feature extraction module is used to extract the sales data of 5-10 clothing models that are similar to the clothing model to be predicted in color, style, thickness, and length as reference features for forecasting. The sales feature extraction module is used to extract the overall sales data of the sub-category and the major category to which the clothing model to be predicted belongs, as the overall sales features of this type of goods.

[0199] Specifically, the data cleaning and data repair device includes an over-limit data identification and elimination module, a jump data identification and elimination module, an inventory abnormal data identification and elimination module, a sales time series curve generation module, a sales stage division module, and an abnormal point sales data correction module. The over-limit data identification and elimination module is used to set upper and lower limits for clothing sales data, identify over-limit data, and eliminate the over-limit data. The jump data identification and elimination module is used to make a difference between the data at each moment and the previous and subsequent data respectively. When the absolute values ​​of the two differences are both greater than the set threshold, the value at this moment is identified as jump data and eliminated. The inventory abnormal data identification and elimination module is used to obtain clothing inventory data and theoretical sales volume data at the corresponding moment, perform data comparison, and when the clothing inventory data is 0, or the inventory data is less than the theoretical sales volume at the corresponding moment, the sales data at this moment is eliminated. The sales time series curve generation module is used to obtain non-over-limit sales data within the normal range and generate a sales time series curve. The sales stage division module is used to divide sales into four stages: introduction period, growth period, maturity period, and decline period according to the sales time series curve. The abnormal point sales data correction module is used to count and obtain the sales growth slope data values ​​k1, k2, k3 and k4 of each stage based on the respective data of the four stages; for each abnormal point, count the average sales data value s of each abnormal point in the week before; and multiply the slope data value ki of the life cycle of each abnormal point by the average sales data value s, and use the product as the correction value of the sales data of each abnormal point.

[0200] In a specific embodiment, Figure 2 As shown, sales curve A1 is the original sales data curve, and sales curve A2 is the sales data curve after data cleaning and data repair by the clothing sales data extraction device of the clothing sales forecasting device based on artificial intelligence. In this embodiment, since there are many scenes of insufficient inventory in the second half of the sales period, the sales data can be increased by different degrees through data repair, so as to more truly reflect the actual sales demand of the goods. In addition, some data with negative sales values ​​are also corrected.

[0201] Specifically, the forecasting algorithm model device includes a forecasting algorithm model training module, a forecasting algorithm model verification module and an initial forecasting result module. The forecasting algorithm model training module is used to: obtain year data n, split the historical data of n years into a training set and a test set of the forecasting algorithm model, the data of the previous n-1 years is used as a test set, and the data of the most recent year is used as a verification set; the historical sales data of the clothing model to be predicted, the historical sales data of similar models, and the historical sales data of the sub-category and the major category are input into the forecasting algorithm model as features, and the sales results at the corresponding time of the next year's data are output as targets to perform the training of the forecasting algorithm model. The forecasting algorithm model verification module is used to: use the verification set data to perform verification on the forecasting algorithm model, calculate the root mean square error between the forecast result and the actual result, and when the error meets the requirements, output the forecasting algorithm model; when the error does not meet the requirements, return to the forecasting algorithm model training module to perform the forecasting algorithm model and parameter adjustment, and retrain. The initial forecast result module is used to obtain the characteristic data of the most recent year, input it into the forecast algorithm model, perform forecasting on the sales results of the next year, and obtain the initial forecast result.

[0202] In a specific embodiment, Figures 3 to 5 As shown, the input characteristics of the forecasting algorithm of the forecasting algorithm model device are explained. Figure 3 For sales data of similar models, Figure 4 The sales data of the subcategory to which the model number to be predicted belongs. Figure 5 It is the sales data of the major categories of the model number to be predicted. Figure 3 Sales curve B1, Figure 4 The sales curve C1 and Figure 5 The sales curves D1 are all sales curves before the forecasting algorithm model device performs filtering and smoothing on the data. Figure 3 Sales curve B2, Figure 4 The sales curve C2 and Figure 5 The sales curves D2 are all sales curves after the forecasting algorithm model device performs filtering and smoothing on the data. Figures 3 to 5The input features include sales data of similar models to the model to be predicted (the algorithm model selects 5-10 models, preferably one model is displayed) and sales data of the subcategory and major category to which it belongs. Preferably, one of the women's long down jackets is used as the prediction object, and its subcategory is long down jacket and its major category is down jacket. Before forecasting, the data is aggregated in weeks. In addition, in order to improve the predictability of the data, the data is filtered and smoothed, and finally, the smoothed data on a weekly scale is used as the forecast target. In order to eliminate the influence of the order of magnitude differences of the numerical values ​​of different input features, in this embodiment, the characteristics of the forecast algorithm model device are normalized.

[0203] In a specific embodiment, when the forecasting algorithm model device trains the algorithm model, the learned label is the sales result of the product number to be predicted in the next year. Figures 3 to 5 The data year of the feature is the winter of 2018-2019, and the label is the sales data of the winter of 2019-2020, such as Figure 6 As shown. Among them, Figure 6 The sales curve E1 is the sales curve before the forecast algorithm model device performs filtering and smoothing on the data, and the sales curve E2 is the sales curve after the forecast algorithm model device performs filtering and smoothing on the data. In this preferred embodiment, after the forecast algorithm model device trains the model, it can be verified with the data of the most recent year. After the verification is passed, the model can be used to forecast future sales results to obtain an initial forecast result.

[0204] Specifically, the temperature correction device includes a temperature difference correction regression model establishment module and an initial sales result correction module. The temperature difference correction regression model establishment module is used to obtain the temperature data of one or more months before the current moment, and the temperature data of the same period of 5-10 years in the next month, forecast the average weekly temperature in the next month, obtain historical temperature data and sales data, and establish a temperature difference correction regression model for the impact of temperature on sales. The initial sales result correction module is used to obtain the established temperature difference correction regression model and the predicted temperature, and perform corrections on the initial sales results of the next week.

[0205] In a specific embodiment, first, the future temperature is predicted, and then the sales data is corrected for the temperature difference based on the comparison between the predicted temperature of the next year and the actual temperature of the previous year. Figure 7 As shown in Figure 2, the sales curves before and after temperature difference correction are as follows: Figure 8 shown. Figure 7 The comparison between the actual temperature and the predicted temperature in the adjacent years is explained. Figure 8 The comparison of sales forecast results before and after temperature correction is explained. Figure 7Curve F1 in the figure is the actual temperature in the winter of 2018-2019, and curve F2 is the predicted temperature in the winter of 2019-2020. Figure 8 The sales curve G1 in FIG. 1 is the initial forecast result, and the sales curve G2 is the result after the temperature is corrected by the temperature correction device.

[0206] Specifically, the holiday and promotional activity correction device includes a sales gain model establishment module, a sales gain calculation module and an initial forecast result correction module. The sales gain model establishment module is used to obtain the activity content and sales data of historical holidays and promotional activities, and establish a sales gain model caused by holiday and promotional activity scenarios. The sales gain calculation module is used to obtain holiday situation information and promotion strategy information within one or more days in the future, and calculate the sales gain brought by holidays and promotional activities based on the sales gain model. The initial forecast result correction module is used to obtain sales gain and correct the initial forecast result.

[0207] In a specific embodiment, for example, down jackets have a sales life cycle that is mainly in winter, and the covered holidays include New Year's Day and the Spring Festival. These two holidays usually occur at the same time as sales promotions, so when modeling down jacket sales, the impact of these two effects are superimposed. Modeling is based on the sales history data from two weeks before New Year's Day to two weeks after the Spring Festival, and the sales gain brought by holidays and promotions is obtained. The sales gain is then multiplied by the temperature-corrected result of the previous step, which is the forecast result of the holiday and promotion correction device, that is, the final forecast result of the artificial intelligence-based clothing sales forecasting device of the present invention. The sales curves before and after the forecast of the holiday and promotion correction device are as follows: Fig. 9 As shown in Figure 1, the sales forecast results before and after the correction of holidays and promotional activities are compared. Among them, the sales curve H1 is the result before the correction, and the sales curve H2 is the result after the correction.

[0208] It will be appreciated by those skilled in the art that the present invention has been described with reference to the flowcharts and / or block diagrams of the methods, systems and computer program products according to the present invention. Each box in the flowchart and / or block diagram, and the combination of boxes in the flowchart and / or block diagram, can obviously be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions (the instructions are passed through the processor of the computer or other programmable data processing device) generate a device for implementing the functions specified in one or more boxes of the flowchart and / or block diagram.

[0209] It should be understood by those skilled in the art that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments, and the embodiments of the present invention may be deformed or modified in any way without departing from the principles.

Claims

1. A clothing sales forecasting method based on artificial intelligence, characterized in that: The clothing sales forecasting method based on artificial intelligence comprises the following steps: Extract apparel sales data and perform data preparation; Identify distorted sales data in clothing sales data and perform data cleaning and data repair; Based on the sales data of cleaned clothing, an AI-based forecasting algorithm model is established to forecast the sales trend of the next sales cycle for a single store or a regional store group; Apply temperature corrections; and Implement holiday and promotional event corrections; The step of identifying distorted sales data in clothing sales data also includes the steps of: identifying and eliminating over-limit data; identifying and eliminating jump data; and identifying and eliminating abnormal clothing inventory data; Wherein, the step of identifying and eliminating over-limit data includes the steps of: setting upper and lower limits for clothing sales data, identifying over-limit data, and eliminating the over-limit data; wherein, the step of identifying and eliminating jump data includes the steps of: subtracting the data at each moment from the previous and subsequent data, and when the absolute values ​​of the two differences are both greater than the set thresholds, identifying the value at that moment as jump data and eliminating it; wherein, the step of identifying and eliminating abnormal clothing inventory data includes the steps of: obtaining clothing inventory data and theoretical sales volume data at the corresponding moment, performing data comparison, and when the clothing inventory data is 0, or the inventory data is less than the theoretical sales volume at the corresponding moment, eliminating the sales data at that moment; The data repair step further includes the steps of: obtaining non-overlimit sales data within a normal range and generating a sales time series curve; dividing sales into four stages, namely, an introduction stage, a growth stage, a mature stage, and a decline stage, according to the sales time series curve; based on the respective data of the four stages, respectively, counting and obtaining the sales growth slope data values ​​k1, k2, k3, and k4 of each stage; for each abnormal point, counting the average sales data value s of each abnormal point in the week before; and multiplying the slope data value ki of the life cycle of each abnormal point by the average sales data value s, and using the product as the correction value of the sales data of each abnormal point; In the forecasting algorithm model, the sales data of the target model of the single store or regional store group is obtained as a feature, and the sales data of similar models and the sales data of the sub-categories and major categories to which they belong are introduced; The step of performing temperature correction also includes the steps of: obtaining temperature data of one or more months before the current moment, and temperature data of the same period of 5-10 years in the future month, and performing a forecast of the weekly average temperature of the future month; The step of executing the holiday and promotion activity correction step further includes the steps of: obtaining holiday feature information or promotion activity strength information of the single store or regional store group, and correcting the original sales forecast result; Among them, the smoothed sales curve, short-term temperature impact, and forecast of holidays and promotional activities are first executed, and then local corrections are performed.

2. The clothing sales forecasting method based on artificial intelligence according to claim 1, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: Extract historical sales data of the clothing style to be predicted; Extract forecast reference features; as well as Extract the overall sales characteristics of the product.

3. The clothing sales forecasting method based on artificial intelligence according to claim 2, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: Execute a judgment on whether the clothing model to be predicted has historical sales data, and when the clothing model to be predicted has historical sales data, extract the historical sales data of the clothing model to be predicted; and Execute the judgment on whether the clothing model to be predicted is a new clothing launched in the current year. When the clothing model to be predicted is a new clothing launched in the current year, extract the historical sales data of the model most similar to the clothing model to be predicted as the historical sales data of the clothing model to be predicted.

4. The artificial intelligence-based clothing sales forecasting method as described in claim 2, wherein the extracted forecast reference features are sales data of 5-10 clothing models that are similar to the clothing model to be predicted in terms of color, style, thickness and length.

5. The artificial intelligence-based clothing sales prediction method as described in claim 2, wherein the extracted overall sales features of the goods are the overall sales data of the sub-category and the major category to which the clothing style to be predicted belongs.

6. The artificial intelligence-based clothing sales forecasting method as described in claim 1, wherein the distorted sales data includes data missing, over-limit anomalies, data jumps, or sales data distortion caused by insufficient inventory.

7. The clothing sales forecasting method based on artificial intelligence according to claim 1, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: Obtain the training set and test set of the forecast algorithm model; Execute the training of forecast algorithm model; Perform validation on forecast algorithm models; and Acquire feature data, input it into the forecasting algorithm model, and obtain initial forecasting results.

8. The clothing sales forecasting method based on artificial intelligence according to claim 7, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: Get year data n, split the historical data of n years into training set and test set of the forecast algorithm model, use the data of the first n-1 years as the test set, and the data of the most recent year as the validation set.

9. The clothing sales forecasting method based on artificial intelligence according to claim 7, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: The historical sales data of the clothing model to be predicted, the historical sales data of similar models, and the historical sales data of the subcategories and major categories are input into the forecasting algorithm model as features, and the sales results at the corresponding time of the next year's data are output as the target to execute the training of the forecasting algorithm model.

10. The clothing sales forecasting method based on artificial intelligence according to claim 7, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: The prediction algorithm model is verified using the validation set data, and the root mean square error between the prediction result and the actual result is calculated. When the error meets the requirements, the prediction algorithm model is output. When the error does not meet the requirements, the prediction algorithm model and parameter adjustment are returned and retrained.

11. The clothing sales forecasting method based on artificial intelligence according to claim 7, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: Obtain the characteristic data of the most recent year, input it into the forecast algorithm model, and obtain the initial forecast results.

12. The artificial intelligence-based clothing sales forecasting method as described in any one of claims 7 to 11, wherein the machine learning model is selected from an extreme gradient boosting tree model, a random forest model, a support vector machine model, and a neural network model.

13. The clothing sales forecasting method based on artificial intelligence according to claim 1, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: Obtain historical temperature data and perform average temperature forecasts for the coming weeks; Establish a temperature difference correction regression model for the impact of temperature on sales; as well as The established temperature difference correction regression model and the predicted temperature are obtained to correct the initial sales results for the next week.

14. The clothing sales forecasting method based on artificial intelligence according to claim 13, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: wherein, The algorithmic timing model is selected from the long short-term memory model, the gated recurrent unit, and the differential integrated moving average autoregressive model.

15. The clothing sales forecasting method based on artificial intelligence according to claim 13, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: Historical temperature data and sales data are obtained, and a temperature difference correction regression model of the impact of temperature on sales is established, where the algorithm model is selected from the support vector machine model, the Ridge regression model, and the Lasso regression model.

16. The clothing sales forecasting method based on artificial intelligence according to claim 1, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: Establish sales gain models caused by holidays and promotional activities; Calculate sales gains from holidays and promotions; and Capture sales gains and revise initial forecast results.

17. The clothing sales forecasting method based on artificial intelligence according to claim 16, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: Obtain the activity content and sales data of historical holidays and promotional activities, and establish a sales gain model caused by holiday and promotional activity scenarios.

18. The clothing sales forecasting method based on artificial intelligence according to claim 16, wherein the clothing sales forecasting method based on artificial intelligence further comprises the following steps: Obtain holiday information and promotion strategy information for the next one or more days, and calculate the sales gain brought by holidays and promotional activities based on the sales gain model.

19. A clothing sales forecasting device based on artificial intelligence, characterized in that: The artificial intelligence-based clothing sales forecasting device includes: a software application, a memory for storing the software application, and a processor for executing the software application, wherein each program of the software application correspondingly executes the steps in the artificial intelligence-based clothing sales forecasting method described in any one of claims 1 to 18.

20. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the clothing sales forecasting method based on artificial intelligence described in any one of claims 1 to 18 are performed.

21. An artificial intelligence-based clothing sales forecasting system, applying the artificial intelligence-based clothing sales forecasting method according to any one of claims 1 to 18, for forecasting clothing sales of a single store or a regional store group, characterized in that: The clothing sales forecasting system based on artificial intelligence includes: A clothing sales data extraction subsystem configured to: extract clothing sales data and perform data preparation; A data cleaning and data repair subsystem, which is configured to: identify distorted sales data in clothing sales data, and perform data cleaning and data repair; A forecasting algorithm model subsystem, which is configured to: establish a forecasting algorithm model based on artificial intelligence based on the sales data of the washed clothing, and forecast the sales trend of the next sales cycle; An air temperature correction subsystem configured to: perform air temperature correction; and A holiday and promotion activity correction subsystem, which is configured to perform holiday and promotion activity correction; Among them, the data cleaning and data repair subsystem includes a sales timing curve generation unit, a sales stage division unit and an abnormal point sales data correction unit, the sales timing curve generation unit is configured to: generate a sales timing curve; the sales stage division unit is configured to: divide the sales into multiple stages according to the sales timing curve; the abnormal point sales data correction unit is configured to: based on the data of each stage, perform sales data correction for each abnormal point; wherein the sales stage division unit is further configured to: divide the sales into four stages of introduction, growth, maturity and decline according to the sales timing curve; based on the respective data of the four stages, count and obtain the sales growth slope data values ​​k1, k2, k3 and k4 of each stage; for each abnormal point, count the average sales data value s of each abnormal point in the week before; and multiply the slope data value ki of the life cycle of each abnormal point by the average sales data value s, and use the product as the correction value of the sales data of each abnormal point.

22. The artificial intelligence-based clothing sales forecasting system according to claim 21, wherein the clothing sales data extraction subsystem further comprises a historical sales data extraction unit for clothing models to be predicted, a forecast reference feature extraction unit and a sales feature extraction unit, wherein the historical sales data extraction unit for clothing models to be predicted is configured to: extract historical sales data of clothing models to be predicted; the forecast reference feature extraction unit is configured to extract forecast reference features; and the sales feature extraction unit is configured to extract overall sales features of the product.

23. The artificial intelligence-based clothing sales forecasting system as described in claim 22, wherein the historical sales data extraction unit for the clothing model to be predicted is further configured to: determine whether the clothing model to be predicted has historical sales data, and when the clothing model to be predicted has historical sales data, extract the historical sales data of the clothing model to be predicted; and determine whether the clothing model to be predicted is a new product of the year, and when the clothing model to be predicted is a new product launched that year, extract the historical sales data of the model most similar to the clothing model to be predicted as the historical sales data of the clothing model to be predicted.

24. The artificial intelligence-based clothing sales forecasting system as described in claim 22, wherein the extracted forecast reference features are sales data of 5-10 clothing models that are similar to the clothing model to be predicted in terms of color, style, thickness and length.

25. The artificial intelligence-based clothing sales forecasting system as described in claim 22, wherein the extracted overall sales features of the goods are the overall sales data of the sub-category and the major category to which the clothing style to be predicted belongs.

26. The artificial intelligence-based clothing sales forecasting system as described in claim 21, wherein the distorted sales data includes data missing, over-limit anomalies, data jumps, or sales data distortion caused by insufficient inventory.

27. An artificial intelligence-based clothing sales forecasting system as described in claim 21, wherein the data cleaning and data repair subsystem includes an over-limit data identification and elimination unit, a jump data identification and elimination unit, and an abnormal inventory data identification and elimination unit, the over-limit data identification and elimination unit is configured to: identify and eliminate over-limit data; the jump data identification and elimination unit is configured to identify and eliminate jump data; the abnormal inventory data identification and elimination unit is configured to identify and eliminate abnormal clothing inventory data.

28. The artificial intelligence-based clothing sales forecasting system as described in claim 27, wherein the over-limit data identification and elimination unit is further configured to: set upper and lower limits for clothing sales data, identify over-limit data, and eliminate the over-limit data.

29. The artificial intelligence-based clothing sales forecasting system as described in claim 27, wherein the jump data identification and elimination unit is further configured to: subtract the data at each moment from the previous and next data respectively, and when the absolute values ​​of the two differences are both greater than a set threshold, identify the value at that moment as jump data and eliminate it.

30. The artificial intelligence-based clothing sales forecasting system as described in claim 27, wherein the abnormal inventory data identification and elimination unit is further configured to: obtain clothing inventory data and theoretical sales volume data at a corresponding moment, perform data comparison, and when the clothing inventory data is 0, or the inventory data is less than the theoretical sales volume at a corresponding moment, eliminate the sales data at that moment.

31. The artificial intelligence-based clothing sales forecasting system as described in claim 21, wherein the sales timing curve generation unit is further configured to: obtain non-overlimit sales data within a normal range and generate a sales timing curve.

32. The artificial intelligence-based clothing sales forecasting system as described in claim 21, wherein the forecasting algorithm model subsystem includes a forecasting algorithm model training unit, a forecasting algorithm model verification unit and an initial forecasting result unit, the forecasting algorithm model training unit is configured to: obtain a training set and a test set of the forecasting algorithm model and perform training of the forecasting algorithm model; the forecasting algorithm model verification unit is configured to: perform verification on the forecasting algorithm model; the initial forecasting result unit is configured to: obtain feature data, input it into the forecasting algorithm model, and obtain an initial forecasting result.

33. The artificial intelligence-based clothing sales forecasting system as described in claim 32, wherein the forecasting algorithm model training unit is further configured to: obtain year data n, split the historical data of n years into a training set and a test set of the forecasting algorithm model, use the data of the previous n-1 years as the test set, and use the data of the most recent year as the verification set.

34. The artificial intelligence-based clothing sales forecasting system as described in claim 32, wherein the forecasting algorithm model training unit is further configured as follows: the historical sales data of the clothing model to be predicted, the historical sales data of similar models, and the historical sales data of the corresponding subcategories and major categories are input into the forecasting algorithm model as features, and the sales results at the corresponding moment of the next year's data are output as targets to execute the training of the forecasting algorithm model.

35. The artificial intelligence-based clothing sales forecasting system as described in claim 32, wherein the forecast algorithm model verification unit is further configured to: use the verification set data to verify the forecast algorithm model, calculate the root mean square error between the forecast result and the actual result, and when the error meets the requirements, output the forecast algorithm model; when the error does not meet the requirements, return to execute the forecast algorithm model and adjust the parameters, and retrain.

36. The artificial intelligence-based clothing sales forecasting system as described in claim 32, wherein the initial forecast result unit is further configured to: obtain feature data of the most recent year, input it into the forecast algorithm model, and obtain initial forecast results.

37. An artificial intelligence-based clothing sales forecasting system as described in any one of claims 32 to 36, wherein the machine learning model is selected from an extreme gradient boosting tree model, a random forest model, a support vector machine model, and a neural network model.

38. The clothing sales forecasting system based on artificial intelligence according to claim 21, wherein the temperature correction subsystem comprises a temperature difference correction regression model building unit and an initial sales result correction unit, and the temperature difference correction regression model building unit is configured to: obtain historical temperature data and perform future weekly average temperature forecast; And establish a temperature difference correction regression model for the impact of temperature on sales; The initial sales result correction unit is configured to obtain the established temperature difference correction regression model and the predicted temperature, and perform correction on the initial sales results for the next week.

39. The clothing sales forecasting system based on artificial intelligence according to claim 38, wherein the temperature difference correction regression model building unit is further configured to: obtain temperature data of one or more months before the current moment, and temperature data of the same period of 5-10 years in the future month, and perform a forecast on the weekly average temperature of the future month, wherein: The algorithmic timing model is selected from the long short-term memory model, the gated recurrent unit, and the differential integrated moving average autoregressive model.

40. The artificial intelligence-based clothing sales forecasting system as described in claim 38, wherein the temperature difference correction regression model establishment unit is further configured to: obtain historical temperature data and sales data, and establish a temperature difference correction regression model of the impact of temperature on sales, wherein the algorithm model is selected from a support vector machine model, a Ridge regression model, and a Lasso regression model.

41. The artificial intelligence-based clothing sales forecasting system as described in claim 21, wherein the holiday and promotional activity correction subsystem includes a sales gain model building unit, a sales gain calculation unit and an initial forecast result correction unit, the sales gain model building unit is configured to: build a sales gain model caused by holidays and promotional activity scenarios; the sales gain calculation unit is configured to: calculate the sales gain brought by holidays and promotional activities; the initial forecast result correction unit is configured to: obtain sales gain and correct the initial forecast result.

42. The artificial intelligence-based clothing sales forecasting system as described in claim 41, wherein the sales gain model building unit is further configured to: obtain activity content and sales data of historical holidays and promotional activities, and build a sales gain model caused by holiday and promotional activity scenarios.

43. The artificial intelligence-based clothing sales forecasting system as described in claim 41, wherein the sales gain calculation unit is further configured to: obtain holiday situation information and promotion strategy information within one or more days in the future and calculate the sales gain brought by holidays and promotional activities based on the sales gain model.

44. An artificial intelligence-based clothing sales forecasting device, applying the artificial intelligence-based clothing sales forecasting method described in any one of claims 1 to 18 and the artificial intelligence-based clothing sales forecasting system described in any one of claims 21 to 43, for forecasting clothing sales of a single store or a regional store group, characterized in that: The clothing sales forecasting device based on artificial intelligence includes a clothing sales data extraction device, a data cleaning and data repair device, a forecasting algorithm model device, a temperature correction device, and a holiday and promotion activity correction device, wherein the clothing sales data extraction device is used to extract clothing sales data and perform data preparation; the data cleaning and data repair device is used to identify distorted sales data in clothing sales data, and perform data cleaning and data repair; the forecasting algorithm model device is used to establish an artificial intelligence-based forecasting algorithm model based on the cleaned clothing sales data, and forecast the sales trend of the next sales cycle; the temperature correction device is used to perform temperature correction; the holidays and promotion activities The correction device is used to perform holiday and promotional activity corrections; wherein the clothing sales data extraction device also includes a historical sales data extraction module for clothing models to be predicted, a forecast reference feature extraction module and a sales feature extraction module, wherein the historical sales data extraction module for clothing models to be predicted is used to extract the historical sales data of the clothing models to be predicted; the forecast reference feature extraction module is used to extract forecast reference features; the sales feature extraction module is used to extract the overall sales features of the goods; wherein the historical sales data extraction module for clothing models to be predicted is also used to: determine whether the clothing models to be predicted have historical sales data, and when the clothing models to be predicted have historical sales data, extract the historical sales data of the clothing models to be predicted. historical sales data; and executing the judgment whether the clothing model to be predicted is a new clothing product launched in the current year. When the clothing model to be predicted is a new clothing product launched in the current year, the historical sales data of the model most similar to the clothing model to be predicted is extracted as the historical sales data of the clothing model to be predicted; wherein the extracted forecast reference features are the sales data of 5-10 clothing models that are similar to the clothing model to be predicted in terms of color, style, thickness and length; wherein the extracted overall sales features of the goods are the overall sales data of the sub-category and the major category to which the clothing model to be predicted belongs; wherein the distorted sales data include data missing, over-limit abnormalities, data jumps or sales data distortion caused by insufficient inventory; wherein the data cleaning and data repair device includes an over-limit abnormality A data identification and elimination module, a jump data identification and elimination module, and an abnormal inventory data identification and elimination module, wherein the over-limit data identification and elimination module is used to: identify and eliminate over-limit data; the jump data identification and elimination module is used to identify and eliminate jump data; the abnormal inventory data identification and elimination module is used to identify and eliminate abnormal clothing inventory data; wherein the over-limit data identification and elimination module is also used to: set upper and lower limits for clothing sales data, identify over-limit data, and eliminate the over-limit data; wherein the jump data identification and elimination module is also used to: make a difference between the data at each moment and the previous and subsequent data respectively, and when the absolute values ​​of the two differences are both greater than the set threshold value, the value at that moment is identified as jump data and eliminated;The abnormal inventory data identification and elimination module is further used to: obtain clothing inventory data and theoretical sales volume data at a corresponding time, perform data comparison, and when the clothing inventory data is 0, or the inventory data is less than the theoretical sales volume at a corresponding time, eliminate the sales data at that time; The data cleaning and data repairing device further comprises a sales timing curve generating module, a sales stage dividing module and an abnormal point sales data correction module, wherein the sales timing curve generating module is used to generate a sales timing curve; the sales stage dividing module is used to divide the sales into multiple stages according to the sales timing curve; the abnormal point sales data correction module is used to perform sales data correction for each abnormal point based on the data of each stage; wherein the sales timing curve generating module is also used to obtain non-over-limit sales data within a normal range to generate a sales timing curve; wherein the sales stage dividing module is also used to divide the sales into four stages: introduction stage, growth stage, maturity stage and decline stage according to the sales timing curve; based on the respective data of the four stages, respectively, statistics and obtain the sales growth slope data values ​​k1, k2, k3 and k4 of each stage; for each abnormal point, statistics the average sales data value s of each abnormal point in the week before; And multiply the slope data value ki of the life cycle where each abnormal point is located by the average sales data value s, and the product is used as the correction value of the sales data of each abnormal point.

45. The artificial intelligence-based clothing sales forecasting device as described in claim 44, wherein the forecast algorithm model device includes a forecast algorithm model training module, a forecast algorithm model verification module and an initial forecast result module, the forecast algorithm model training module is used to: obtain the training set and test set of the forecast algorithm model and perform the training of the forecast algorithm model; the forecast algorithm model verification module is used to: perform verification on the forecast algorithm model; the initial forecast result module is used to: obtain feature data, input it into the forecast algorithm model, and obtain an initial forecast result.

46. ​​The artificial intelligence-based clothing sales forecasting device as described in claim 45, wherein the forecast algorithm model training module is also used to: obtain year data n, split the historical data of n years into a training set and a test set of the forecast algorithm model, use the data of the previous n-1 years as the test set, and use the data of the most recent year as the verification set.

47. The artificial intelligence-based clothing sales forecasting device as described in claim 45, wherein the forecast algorithm model training module is also used for: the historical sales data of the clothing model to be predicted, the historical sales data of similar models, and the historical sales data of the corresponding subcategories and major categories are input into the forecast algorithm model as features, and the sales results at the corresponding time of the next year's data are output as targets to execute the training of the forecast algorithm model.

48. The artificial intelligence-based clothing sales forecasting device as described in claim 45, wherein the forecast algorithm model verification module is also used to: use the verification set data to verify the forecast algorithm model, and calculate the root mean square error between the forecast result and the actual result. When the error meets the requirements, the forecast algorithm model is output; when the error does not meet the requirements, the forecast algorithm model is returned to be executed and the parameters are adjusted, and retrained.

49. The artificial intelligence-based clothing sales forecasting device as described in claim 45, wherein the initial forecast result module is also used to: obtain feature data of the most recent year, input it into the forecast algorithm model, and obtain initial forecast results.

50. An artificial intelligence-based clothing sales forecasting device as described in any one of claims 45 to 49, wherein the machine learning model is selected from an extreme gradient boosting tree model, a random forest model, a support vector machine model, and a neural network model.

51. The clothing sales forecasting device based on artificial intelligence according to claim 44, wherein the temperature correction device comprises a temperature difference correction regression model building module and an initial sales result correction module, and the temperature difference correction regression model building module is used to: obtain historical temperature data and perform future weekly average temperature forecast; And establish a temperature difference correction regression model for the impact of temperature on sales; The initial sales result correction module is used to obtain the established temperature difference correction regression model and the predicted temperature, and perform corrections on the initial sales results for the next week.

52. The clothing sales forecasting device based on artificial intelligence according to claim 51, wherein the temperature difference correction regression model building module is further used to: obtain temperature data of one or more months before the current moment, and temperature data of the same period of 5-10 years in the next month, and perform forecasting on the weekly average temperature of the next month, wherein: The algorithmic timing model is selected from the long short-term memory model, the gated recurrent module, and the differential integrated moving average autoregressive model.

53. The artificial intelligence-based clothing sales forecasting device as described in claim 51, wherein the temperature difference correction regression model establishment module is also used to: obtain historical temperature data and sales data, and establish a temperature difference correction regression model of the impact of temperature on sales, wherein the algorithm model is selected from a support vector machine model, a Ridge regression model, and a Lasso regression model.

54. The artificial intelligence-based clothing sales forecasting device as described in claim 44, wherein the holiday and promotional activity correction device includes a sales gain model establishment module, a sales gain calculation module and an initial forecast result correction module, the sales gain model establishment module is used to: establish a sales gain model caused by holidays and promotional activity scenarios; the sales gain calculation module is used to: calculate the sales gain brought by holidays and promotional activities; the initial forecast result correction module is used to: obtain sales gain and correct the initial forecast result.

55. The artificial intelligence-based clothing sales forecasting device as described in claim 54, wherein the sales gain model building module is also used to: obtain the activity content and sales data of historical holidays and promotional activities, and establish a sales gain model caused by holidays and promotional activity scenarios.

56. The artificial intelligence-based clothing sales forecasting device as described in claim 54, wherein the sales gain calculation module is also used to: obtain holiday situation information and promotion strategy information within one or more days in the future and calculate the sales gain brought by holidays and promotional activities based on the sales gain model.

Images