A plastic chemical e-commerce management platform and management method based on data analysis

Abstract

The application discloses a kind of based on data analysis's chemical plastic electric business management platform and management method, it is related to electric business management technical field.The method includes: obtaining the multidimensional data of chemical plastic product and forms standardization dataset;Standardization dataset is analyzed by LSTM-GARCH prediction model, and the predicted market data is output, including: long-term change rule of chemical plastic product market is obtained by analyzing conventional data by LSTM model;By GARCH model analysis fluctuation data, judge whether there is a sudden event influence;According to the analysis result, the predicted market data is output and supply chain management decision is generated.The application realizes the intelligentization and dynamic of chemical plastic supply chain prediction system by applying LSTM-GARCH model, deeply fusing chemical plastic market long-term rule and sudden event response mechanism, so that the system can capture market changes in real time, optimize foreign trade management strategy, enhance the international competitiveness of enterprises.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of e-commerce management, and particularly relates to a plastic chemical e-commerce management platform and method based on data analysis. BACKGROUND

[0002] The plastic chemical product supply chain is challenged by large market demand fluctuations and high prediction difficulty, which leads to difficulties in production planning and inventory management, and low supply chain efficiency. Traditional prediction methods are difficult to accurately grasp market changes, especially for volatile data types, and the prediction accuracy is insufficient. At the same time, different types of data have large feature differences, and a single model cannot be considered. In addition, supply chain decisions involve multiple links, and how to effectively convert prediction results into specific production and inventory strategies is also a big problem. The lack of interpretability of the prediction results makes it difficult for decision makers to develop reasonable supply chain adjustment programs. The regional differences are large, and the demand and cost distribution of different regions are different, and how to develop regional strategies is also a difficult problem.

[0003] These factors comprehensively lead to the complexity of the optimization of the plastic chemical product supply chain, and an optimization method of the supply chain is needed, which can comprehensively consider multi-dimensional data, consider the characteristics of different types of data, provide interpretable prediction results and support regional decision-making, to cope with market fluctuations and improve the overall operation efficiency of the supply chain. SUMMARY

[0004] The purpose of the application is to provide a plastic chemical e-commerce management platform and method based on data analysis, which can be realized by the following technical solutions:

[0005] In a first aspect, the application provides a plastic chemical e-commerce management method based on data analysis, including the following steps:

[0006] Obtain multi-dimensional data of plastic chemical products on e-commerce platforms, the multi-dimensional data including regular data reflecting market trends and fluctuation data reflecting sudden events;

[0007] Preprocess the multi-dimensional data to form a standardized data set;

[0008] Analyze the standardized data set by a pre-trained LSTM-GARCH prediction model to output predicted market data, specifically including:

[0009] Input the standardized data set into the LSTM-GARCH prediction model; the LSTM-GARCH model includes a GARCH model and an LSTM model;

[0010] Analyze the regular data by the LSTM model in the LSTM-GARCH prediction model to obtain the long-term change rule of the plastic chemical product market;

[0011] analyzing the volatility data by the GARCH model in the LSTM-GARCH prediction model to determine whether there is an impact of a sudden event;

[0012] outputting the predicted market data according to the analysis result;

[0013] generating supply chain management decisions according to the predicted market data;

[0014] The LSTM model has a long-term trend capturing function, and the GARCH model has a volatility risk quantifying function.

[0015] The predicted market data is represented as:

[0016] y t = ·LSTM( )+ ·GARCH( );

[0017] The weight of the LSTM model is represented as:

[0018] The weight of the GARCH model is represented as, and is 0 when the volatility data does not have abnormal fluctuations.

[0019] The predicted result output by the LSTM model is represented as:

[0020] The predicted result output by the GARCH model is represented as.

[0021] Preferably, the multi-dimensional data of the chemical and plastic product on the e-commerce platform is obtained, including:

[0022] The multi-dimensional data is collected through an e-commerce program interface;

[0023] The multi-dimensional data is classified into regular data and volatility data, wherein the regular data includes sales data, raw material demand data, and inventory change data of the chemical and plastic product; and the volatility data includes crude oil price data, exchange rate data, and tariff data;

[0024] The regular data and the volatility data are stored separately for subsequent calling.

[0025] Preferably, the multi-dimensional data is preprocessed to form a standardized data set, including:

[0026] The multi-dimensional data is data cleaned to remove outliers;

[0027] filling missing values in the multi-dimensional data;

[0028] performing unit standardization operation on the multi-dimensional data to unify data format;

[0029] generating the standardized data set and storing it for calling by the LSTM-GARCH prediction model, ensuring consistency of the standardized data set in time zone and unit.

[0030] Preferably, the analysis of the standardized data set by the pre-trained LSTM-GARCH prediction model comprises:

[0031] If no abnormal fluctuation of the fluctuation data is detected, the LSTM model is used to predict the regular data, and the predicted second regular data is output;

[0032] If abnormal fluctuation of the fluctuation data is detected, the GARCH model is used to predict the regular data to obtain first regular data, and the first regular data and the second regular data are calibrated to obtain predicted third regular data;

[0033] The final predicted market data is generated according to the third regular data or the second regular data.

[0034] Preferably, the analysis of the fluctuation data by the GARCH model in the LSTM-GARCH prediction model comprises:

[0035] Obtaining the current data fluctuation trend when the fluctuation data occurs abnormal fluctuation;

[0036] Obtaining similar sudden events in historical sudden events that have similarity exceeding a preset threshold with the current data fluctuation trend;

[0037] Respectively extracting feature values of the fluctuation data that occurs abnormal fluctuation and feature values of historical fluctuation data corresponding to the similar sudden events, and matching the two types of extracted feature values, if the features match, obtaining the reference parameters of the similar sudden events;

[0038] Based on the reference parameters, inputting the obtained regular data into the GARCH model to obtain the volatility rate of the regular data under the influence of the fluctuation data that occurs abnormal fluctuation, and the predicted first regular data;

[0039] Based on the volatility rate, judging the risk level of the current sudden event.

[0040] Preferably, the first regular data is expressed as: ; wherein, represents a mean value corresponding to the regular data; represents a residual term.

[0041] Preferably, the volatility of the current regular data is represented as:

[0042]

[0043] wherein, represents a constant term; represents a previous residual square; represents a previous conditional variance; represents an ARCH term coefficient; is a GARCH term coefficient; represents a lag order of the residual square; represents a lag order of the conditional variance.

[0044] Preferably, in the LSTM-GARCH prediction model, the GARCH model is trained by using transfer learning, comprising:

[0045] Obtain historical transaction data of the plastic product and perform event screening, and use event data when abnormal fluctuations occur in crude oil prices, exchange rates and tariffs as training data samples, label the events that occur fluctuations and establish an event database;

[0046] Extract features from historical fluctuation data of different types of events, including extreme point number, volatility and trend slope;

[0047] Pre-train the GARCH model according to the fluctuation data characteristics of different types of events to obtain baseline parameters;

[0048] When detecting that the fluctuation data has an abnormal fluctuation, real-time match similar events by using the baseline parameters and perform parameter transfer;

[0049] According to the GARCH model, the result is predicted, the historical transaction data is used to calibrate and fit the prediction result, the baseline parameters are further optimized, and the training of the GARCH model is completed; the event database is updated by using the detected new event.

[0050] Preferably, the second regular data is represented as:

[0051] wherein, represents the regular data at the current time; represents the regular data at the previous time.

[0052] In a second aspect, the embodiments of the present application provide a plastic e-commerce management platform based on data analysis, comprising:​​

[0053] The data acquisition module is configured to acquire multi-dimensional data of the plastic product on the e-commerce platform, wherein the multi-dimensional data includes regular data reflecting market trends and fluctuation data reflecting sudden events.

[0054] The data processing module is configured to preprocess the multi-dimensional data to form a standardized data set.

[0055] The data analysis module is configured to analyze the standardized data set by using a pre-trained LSTM-GARCH prediction model to output predicted market data, and specifically includes: inputting the standardized data set into the LSTM-GARCH prediction model; the LSTM-GARCH model includes a GARCH model and an LSTM model; analyzing the regular data by using the LSTM model in the LSTM-GARCH prediction model to obtain long-term change rules of the plastic product market; analyzing the fluctuation data by using the GARCH model in the LSTM-GARCH prediction model to determine whether there is an influence of a sudden event; and outputting the predicted market data according to the analysis result.

[0056] The decision generation module is configured to generate a supply chain management decision according to the predicted market data.

[0057] The application has the following beneficial effects:

[0058] (1) The application uses the LSTM-GARCH model to deeply integrate the long-term rules of the plastic market and the response mechanism of sudden events, realizes the intelligentization and dynamicization of the plastic supply chain prediction system, enables the system to capture market changes in real time, optimizes foreign trade management strategies, and enhances the international competitiveness of enterprises.

[0059] (2) The application acquires multi-dimensional historical data and preprocesses the data, respectively uses a generalized autoregressive conditional heteroscedasticity model (GARCH model) and a long short-term memory network model (LSTM model) based on sample migration learning to analyze and predict fluctuation type data and regular type data, and then obtains comprehensive prediction data by weighted fitting. Based on the prediction result, sales, demand and inventory prediction are generated, market fluctuation trends are analyzed and the adjustment direction of the supply chain is determined, and then production planning and inventory management suggestions are made. Finally, the prediction data and the optimization strategy are displayed through a visual interface, a space-time heat map is generated to determine the regional demand and cost distribution, and support is provided for supply chain decision-making. The application can effectively respond to market fluctuations, optimize production planning and inventory management, and improve the efficiency of supply chain operation. BRIEF DESCRIPTION OF DRAWINGS

[0060] In order to better understand and implement the technical scheme of the present application, the following will describe the technical scheme of the present application in detail with reference to the accompanying drawings.

[0061] In order to better understand and implement the technical scheme of the present application, the following will describe the technical scheme of the present application in detail with reference to the accompanying drawings.Figure 1 A step flow chart of a data analysis-based chemical and plastic e-commerce management method provided by an embodiment of the present application is provided.

[0062] Figure 2 A step flow chart of analyzing volatility data by a GARCH model provided by an embodiment of the present application is provided.

[0063] Figure 3 A step flow chart of training a GARCH model by using transfer learning provided by an embodiment of the present application is provided.

[0064] Figure 4 A structural schematic diagram of a data analysis-based chemical and plastic e-commerce management platform provided by an embodiment of the present application is provided. DETAILED DESCRIPTION

[0065] To further illustrate the technical means and effects adopted by the present application to achieve the predetermined inventive purpose, exemplary embodiments will be described in detail, which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of methods and systems consistent with some aspects of the present application as detailed in the appended claims.

[0066] The terms used in the present application are merely for the purpose of describing specific embodiments and are not intended to limit the present application. The singular forms "a," "an," and "the" used in the present application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to any or all possible combinations of one or more associated listed items.

[0067] The specific embodiments, features and effects according to the present application are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0068] Embodiment 1

[0069] Please refer to Figure 1 The present application provides a data analysis-based chemical and plastic e-commerce management method, which comprises the following steps:

[0070] S101, obtaining multi-dimensional data of chemical and plastic products on an e-commerce platform, wherein the multi-dimensional data comprises regular data reflecting market trends and volatility data reflecting sudden events;

[0071] S102, preprocessing the multi-dimensional data to form a standardized data set;

[0072] S103, analyzing the standardized data set by the pre-trained LSTM-GARCH prediction model, and outputting predicted market data, specifically including:

[0073] inputting the standardized data set into the LSTM-GARCH prediction model; the LSTM-GARCH model includes a GARCH model and an LSTM model;

[0074] analyzing the regular data by the LSTM model in the LSTM-GARCH prediction model to obtain long-term change rules of the plastic product market;

[0075] analyzing the fluctuation data by the GARCH model in the LSTM-GARCH prediction model to determine whether there is an impact of a sudden event;

[0076] outputting the predicted market data according to the analysis result;

[0077] S104, generating supply chain management decisions according to the predicted market data;

[0078] wherein the LSTM model has a long-term trend capturing function; and the GARCH model has a fluctuation risk quantification function;

[0079] wherein the predicted market data is represented as:

[0080] y t = ·LSTM( )+ ·GARCH( );

[0081] LSTM represents the weight of the LSTM model;

[0082] GARCH represents the weight of the GARCH model, and is 0 when the fluctuation data does not have abnormal fluctuation;

[0083] LSTM represents the prediction result output by the LSTM model;

[0084] GARCH represents the prediction result output by the GARCH model.

[0085] Specifically, in order to better analyze the data of the plastic product market, which is easily affected by external events such as crude oil prices, exchange rate changes, and tariffs, the application adopts an LSTM (main) -GARCH (auxiliary) model for data prediction and analysis. The LSTM model is suitable for processing relationships in long sequence data and can be used to capture long-term trend rules in cross-border trade of plastic products, but cannot quantify the impact of sudden events, so it is used for plastic market data analysis without sudden events. The GARCH model performs stably in volatility prediction tasks and can be used to quantify sudden volatility risk, i.e., data volatility caused by sudden events. The application creatively combines the LSTM model and the GARCH model, and when a sudden event occurs, the GARCH model is introduced to weight and fit the benchmark predicted by the LSTM model. Based on this, the application defines two types of data: volatility data and regular data. Volatility data refers to data that can reflect the occurrence of events that will impact the plastic product market, such as tariff adjustments, financial crises, and the like, such as crude oil price data, exchange rate data, and tariff data. That is, when a sudden event occurs, data such as crude oil price data, exchange rate data, and tariff data will exhibit significant volatility. Regular data refers to market data generated during cross-border trade of plastic products, such as sales data, raw material demand data, and inventory change data. Volatility data is used to detect whether a sudden event has occurred, and regular data is used to reflect market changes and analyze market trends. The application uses the LSTM model to predict and analyze market data without sudden events, predicts and outputs future regular data (the weight of the GARCH model is 0 at this time), and when a sudden event is detected, the GARCH model is introduced to fit and correct the output of the LSTM model, to generate more accurate prediction results, optimize foreign trade management strategies, and assist enterprise decision-makers in optimizing the supply chain, so as to accurately grasp the changes in the plastic e-commerce foreign trade market.

[0086] It should be noted that the LSTM (Long Short-Term Memory) model of the application can capture long-term and short-term dependencies and is suitable for processing complex nonlinear time series data, especially when the data has seasonality, trend, and periodicity. The GARCH (Generalized Autoregressive Conditional Heteroskedasticity) model captures the change of volatility over time in time series by modeling conditional heteroskedasticity, and is mainly used to model the volatility of time series, especially in the financial field, to help analyze and predict market volatility (such as stock market, exchange rate, etc.).

[0087] It can be understood that, in the present application, the long time series data includes sales data of chemical and plastic products, raw material demand data, inventory change data, and customer behavior data; and the short time series data includes crude oil price data, exchange rate data, and tariff data.

[0088] In the step S101 described above, the big data is collected in the manner of an e-commerce program interface (API), and multi-dimensional data of a specific platform is quickly and efficiently acquired. Since the collected data may have time zones, non-uniform units, missing values, and abnormal values, data preprocessing is performed based on data cleaning, abnormal value processing, missing value filling, and unit standardization.

[0089] In an embodiment, the multi-dimensional data of the chemical and plastic products on the e-commerce platform is acquired, including:

[0090] The multi-dimensional data is collected through an e-commerce program interface;

[0091] The multi-dimensional data is classified into regular data and fluctuation data, wherein the regular data includes sales data of chemical and plastic products, raw material demand data, and inventory change data; and the fluctuation data includes crude oil price data, exchange rate data, and tariff data;

[0092] The regular data and the fluctuation data are stored separately for subsequent calling.

[0093] In an embodiment provided by the present application, the multi-dimensional data is preprocessed to form a standardized data set, including:

[0094] The multi-dimensional data is cleaned to remove abnormal values;

[0095] Missing values in the multi-dimensional data are filled and processed;

[0096] The multi-dimensional data is subjected to unit standardization operation to unify the data format;

[0097] The standardized data set is generated and stored for calling by the LSTM-GARCH prediction model, ensuring the consistency of the standardized data set in time zone and unit.

[0098] In an embodiment provided by the present application, the standardized data set is analyzed by a pre-trained LSTM-GARCH prediction model, including:

[0099] If no abnormal fluctuation of the fluctuation data is detected, the regular data is predicted by the LSTM model, and a predicted second regular data is output;

[0100] If the fluctuation data is detected to be abnormally fluctuated, the GARCH model is used to predict the regular data to obtain first regular data, and the first regular data and the second regular data are calibrated to obtain predicted third regular data;

[0101] The final predicted market data is generated according to the third regular data or the second regular data.

[0102] As shown in the embodiment provided in the present application, the GARCH model in the LSTM-GARCH prediction model is used to analyze the fluctuation data, including: Figure 2

[0103] The current data fluctuation trend when the fluctuation data is abnormally fluctuated is obtained;

[0104] In the historical sudden events, a similar sudden event with a similarity to the current data fluctuation trend exceeding a preset threshold is obtained;

[0105] The characteristic values of the fluctuation data abnormally fluctuated and the characteristic values of the historical fluctuation data corresponding to the similar sudden event are extracted respectively, and the two types of extracted characteristic values are matched, if the characteristics are matched, the reference parameters of the similar sudden event are obtained;

[0106] Based on the reference parameters, the obtained regular data is input into the GARCH model to obtain the volatility rate of the regular data under the influence of the fluctuation data abnormally fluctuated and the predicted first regular data;

[0107] Based on the volatility rate, the risk level of the current sudden event is judged.

[0108] As shown in the embodiment provided in the present application, in the LSTM-GARCH prediction model, the GARCH model is trained by using transfer learning, including: Figure 3 The historical transaction data of plastic products are obtained and sudden event screening is performed, the event data when the crude oil price, exchange rate and tariff are abnormally fluctuated are taken as training data samples, the fluctuation of the sudden event is labeled and a sudden event database is established;

[0109] The historical fluctuation data of different types of sudden events are extracted, including the number of extreme points, the volatility rate and the trend slope;

[0110] According to the fluctuation data characteristics of different types of sudden events, the GARCH model is pre-trained to obtain a reference parameter set;

[0111]

[0112] ​​When abnormal fluctuations are detected in the fluctuation data, similar sudden events are matched in real time using the benchmark parameter set and parameter migration is performed.

[0113] The results are predicted based on the GARCH model, and the prediction results are calibrated and fitted using the historical transaction data to further optimize the baseline parameters until the training of the GARCH model is completed; the sudden event database is updated using the detection of new sudden events.

[0114] Specifically, this application collects regular and volatile data of cross-border trade in chemical and plastic products through e-commerce program interfaces to obtain a multi-dimensional raw dataset. The raw dataset undergoes data cleaning, outlier handling, missing value imputation, and unit standardization to obtain a preprocessed dataset. Volatility data, including crude oil price data, exchange rate data, and tariff data, is extracted from the preprocessed dataset to obtain a volatility dataset. If a daily fluctuation in volatility data exceeds 5% (low), 10% (medium), or 20% (high), the risk level is determined based on the magnitude of the fluctuation, resulting in the characteristic value and risk level of the volatility event. Based on the characteristic value of the volatility event, historical similar events' volatility data are matched to obtain the corresponding historical benchmark parameters, resulting in the benchmark parameter set required for transfer learning. A sample transfer learning method is used to train a generalized autoregressive conditional heteroscedasticity model (GARCH model) using the benchmark parameter set, resulting in a trained volatility prediction model. Using the trained volatility prediction model, current market regular and volatile data are input to predict the impact of volatile data on regular data, yielding the volatility impact prediction result. Based on the predicted impact of fluctuations and the calibration fit using historical data, the baseline parameters of the generalized autoregressive conditional heteroscedasticity (GHP) model are optimized to obtain the optimized parameter set. Using the optimized parameter set, the GHP model is rerun to predict the future trends of conventional data types, yielding the final prediction results.

[0115] like Figure 3 As shown, in one embodiment provided in this application, regular type data is extracted from the preprocessed dataset, and an LSTM model is obtained by training a Long Short-Term Memory network model. The model captures long-term trend patterns for regular type data during periods without fluctuation events, and predicts the regular trend by inputting current market data.

[0116] Specifically, the regular type data of the non-volatility event period is extracted from the preprocessed data set to obtain a training data sample. The training data sample is trained by a long short-term memory network model (LSTM model) to capture the trend rule of the regular type data of the non-volatility event period. The current market regular data is input according to the trend rule captured by training to obtain predicted regular data output by the LSTM model. The history record of the volatility type data is obtained, and the abnormal event of the single-day price fluctuation exceeding 5% is detected to determine whether there is a volatility influence. If the volatility influence is detected, the volatility variance is predicted by the GARCH model and the adjusted regular data is output; if there is no volatility influence, the predicted regular data output by the LSTM model is directly used. The regular data output by the LSTM model and the regular data output by the GARCH model are fitted to obtain a comprehensive prediction result.

[0117] In another embodiment provided in the application, the regular trend prediction result and the volatility influence prediction result are weighted and fitted, a preset weight distribution rule is adopted, if the volatility type data is detected to be abnormal, the weight of the GARCH model is increased, otherwise the weight is distributed to the LSTM model to obtain comprehensive prediction data.

[0118] The volatility type data includes crude oil price data, exchange rate data, and tariff data. The data is processed to unify time zones, standardize units, and eliminate missing values and outliers, thereby obtaining a standardized volatility type data set. According to the standardized volatility type data set, the single-day price change is calculated. If the single-day price change of crude oil is greater than 5%, 10%, or 20%, or the exchange rate or tariff data reaches a corresponding threshold, a low, medium, or high volatility anomaly is determined, and the volatility anomaly level is determined. According to the volatility anomaly level, a historical event database with similar volatility characteristics is matched to obtain the baseline parameters of the corresponding historical event, and a parameter set for the GARCH model is obtained. The parameter set is input into the pre-trained GARCH model to predict the mean and residual term of the conventional data under the influence of volatility, and the volatility prediction result is obtained. The conventional data of the chemical and plastic products in the past 5-10 years without volatility events is obtained, and the LSTM model is trained to capture the trend of the cross-border trade conventional data, and the trained model is obtained. The current and previous conventional type data is input into the trained LSTM model to predict the conventional data without volatility events, and the conventional trend prediction result is obtained. According to whether the volatility type data has an anomaly, if a volatility anomaly is detected, the GARCH model weight is set to a high preset value, and the LSTM model weight is set to a low value, otherwise, the LSTM model weight is set to 1, and the GARCH model weight is set to 0, and the weight distribution rule is determined. The conventional trend prediction result of the LSTM model and the volatility prediction result of the GARCH model are weighted and fitted according to the determined weight distribution rule to obtain the comprehensive prediction data. The comprehensive prediction data is calibrated and fitted by historical data to optimize the baseline parameters of the GARCH model and the LSTM model, and the updated model parameters are obtained.

[0119] In an embodiment provided in the present application, the first conventional data is represented as: ; wherein, represents the mean value corresponding to the current conventional data; represents the residual term.

[0120] In an embodiment provided in the present application, the volatility rate of the current conventional data is represented as:

[0121] ;

[0122] wherein, represents the constant term; represents the previous residual square; represents the previous conditional variance; represents the ARCH term coefficient; represents the GARCH term coefficient; represents the lag order of the residual square; represents the lag order of the conditional variance.

[0123] Specifically, the embodiment adopts a sample transfer learning method to train a generalized autoregressive conditional heteroscedasticity model (GARCH model), and analyzes the situation of fluctuations in crude oil prices, exchange rates, and tariffs when a sudden event causes volatility type data to change. For example: detecting that the daily price of crude oil rises or falls by more than 5% low, 10% medium, and 20% high, different types of values are different. According to the degree of abnormality, the risk level is judged, and the predicted regular data in the future stage is output.

[0124] Specifically, when the crude oil price, exchange rate, and tariff fluctuate, the influence of the fluctuation data on the regular data is predicted. When the fluctuation data is detected to be abnormal, the characteristic value of the fluctuation data is matched with the historical sudden event similar to the fluctuation, the benchmark parameter of the historical sudden event is input according to the benchmark parameter of the historical sudden event, the fluctuation variance of the current sudden event is predicted, the risk level is judged, and the change trend of the regular data under the influence of the sudden event is output.

[0125] In short, the present application determines whether the current sudden event is similar to the sudden event that has occurred in the historical data through the GARCH model. If they are similar, the benchmark parameter of the sudden event in the historical data is used as the benchmark parameter of the current sudden event to predict the risk level of the current sudden event.

[0126] In an embodiment provided by the present application, the LSTM model is trained by using a long short-term memory network, which comprises:

[0127] According to the historical transaction data, the historical regular data of the plastic product without a sudden event is selected as the training data sample;

[0128] The LSTM model is trained based on the training data sample to capture the trend rule of the regular data of the plastic product without a sudden event, and an optimized LSTM model is obtained.

[0129] In an embodiment provided by the present application, the second regular data is represented as:

[0130] ; wherein, represents the regular data at the current time; represents the regular data at the previous time.

[0131] Specifically, the embodiment adopts a long short-term memory network (LSTM) model, and trains the LSTM model according to the historical regular data to capture the trend rule of the regular type data of cross-border trade of the plastic product without a sudden event. The predicted regular data is output according to the input of the regular data of the market in recent years.

[0132] Specifically, the conventional data of the plastic product in the period without fluctuation events in the historical data is used as the training data sample to train the LSTM model, the trend of the conventional type data of the cross-border trade of the plastic product in the period without fluctuation events is captured, the current conventional data of the dynamic market is input, the predicted conventional data is output according to the captured trend.

[0133] In another embodiment of the present application, the sales data prediction, the raw material demand data prediction and the inventory change data prediction of the plastic product are generated according to the comprehensive prediction data, and a prediction data set is obtained. By analyzing the sales data, the raw material demand data and the inventory change data in the prediction data set, the market demand fluctuation trend is calculated, and the supply chain adjustment direction is determined.

[0134] The production plan is also proposed by analyzing the sales data, the raw material demand data and the inventory change data in the prediction data set, such as increasing production when the sales are good and the inventory is sufficient, and decreasing production when the sales are poor, and the inventory management suggestion, such as supplementing the inventory when the raw material supply is sufficient and the inventory is insufficient. A visual interface is also provided to generate a space-time heat map. The regional demand is displayed according to the amount of sales data of the plastic product in different regions, and the material cost is displayed according to the amount of raw material demand data, which assists the enterprise decision maker to optimize the supply chain. The fluctuation trend of the market demand is analyzed according to the prediction data, the adjustment direction of the supply chain management is determined, and the preliminary decision basis is obtained.

[0135] In summary, the present application applies the LSTM-GARCH model, deeply integrates the long-term rules of the plastic market and the response mechanism of the sudden events, realizes the intelligentization and the dynamicization of the plastic supply chain prediction system, enables the system to capture the market changes in real time, optimizes the foreign trade management strategy, and enhances the international competitiveness of the enterprise.

[0136] Embodiment 2

[0137] Please refer to Figure 4 The present application embodiment provides a plastic e-commerce management platform based on data analysis, which applies the plastic e-commerce management method as described above, comprising:

[0138] The data acquisition module is used for acquiring the historical transaction data of the plastic product on the e-commerce management platform, and performing a preprocessing operation on the historical transaction data;

[0139] The model training module is used for training the GARCH model based on the preprocessed historical transaction data by using the transfer learning, and training the LSTM model by using the long short-term memory network;

[0140] The data analysis module: using the trained GARCH model to analyze data, specifically: analyzing the risk level when the emergency occurs, outputting the predicted first regular data; using the trained LSTM model to analyze data, specifically: obtaining the current regular data of the current dynamic market and inputting it into the LSTM model, outputting the predicted second regular data;

[0141] The data fitting module: used for fitting the first regular data and the second regular data to obtain the LSTM-GARCH model, outputting the predicted third regular data;

[0142] The model prediction module: generating the supply chain prediction result of the plastic product based on the third regular data predicted by the LSTM-GARCH model; based on the supply chain prediction result, generating the space-time heat map using the visual interface, used for providing auxiliary decision to optimize the supply chain.

[0143] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in a certain embodiment can be referred to the related description of other embodiments.

[0144] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is taken as an example, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or software. In addition, the specific name of each functional unit and module is only for easy distinction, and does not limit the protection scope of the application. The specific working process of the units and modules in the system can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0145] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0146] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application in any form. Although the present application has been disclosed with the preferred embodiments as above, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content to obtain equivalent embodiments with equivalent changes, as long as the changes or modifications do not deviate from the technical solution of the present application. Any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application still falls within the scope of the technical solution of the present application.

Claims

1. A data analysis-based plastic e-commerce management method, characterized in that: The method comprises the following steps: obtaining multi-dimensional data of chemical plastic products on an e-commerce platform, the multi-dimensional data including regular data reflecting market trends and fluctuation data reflecting sudden events; preprocessing the multi-dimensional data to form a standardized data set; analyzing the standardized data set by a pre-trained LSTM-GARCH prediction model, and outputting predicted market data, specifically including: inputting the standardized data set into the LSTM-GARCH prediction model; the LSTM-GARCH model includes a GARCH model and an LSTM model; analyzing the regular data by the LSTM model in the LSTM-GARCH prediction model to obtain the long-term change rule of the chemical plastic product market; analyzing the fluctuation data by the GARCH model in the LSTM-GARCH prediction model to determine whether there is a sudden event influence; outputting the predicted market data according to the analysis result; generating supply chain management decisions according to the predicted market data; wherein the LSTM model has a long-term trend capturing function; the GARCH model has a fluctuation risk quantification function; wherein the predicted market data is represented as: ； representing the weights of the LSTM model; a weight representing a GARCH model and being 0 when the volatility data does not exhibit an abnormal fluctuation; represents the prediction result output by the LSTM model; represents the prediction result of the GARCH model output; the multi-dimensional data of chemical plastic products on the e-commerce platform includes: collecting the multi-dimensional data through an e-commerce program interface; classifying the multi-dimensional data into regular data and fluctuation data, wherein the regular data includes sales data, raw material demand data, and inventory change data of chemical plastic products; the fluctuation data includes crude oil price data, exchange rate data, and tariff data; storing the regular data and the fluctuation data separately for subsequent calling; the analysis of the standardized data set by the pre-trained LSTM-GARCH prediction model includes: if no abnormal fluctuation of the fluctuation data is detected, predicting the regular data by the LSTM model and outputting predicted second regular data; if abnormal fluctuation of the fluctuation data is detected, first predicting the regular data by the GARCH model to obtain first regular data, then calibrating the first regular data and the second regular data obtained by predicting the regular data by the LSTM model to obtain predicted third regular data; generating the final predicted market data according to the third regular data or the second regular data; the analysis of the fluctuation data by the GARCH model in the LSTM-GARCH prediction model includes: obtaining the current data fluctuation trend when the fluctuation data occurs abnormal fluctuation; obtaining similar sudden events in historical sudden events that have a similarity to the current data fluctuation trend exceeding a preset threshold; extracting feature values of the fluctuation data that occurs abnormal fluctuation and feature values of historical fluctuation data corresponding to the similar sudden events, and matching the extracted two types of feature values, if the features match, obtaining the reference parameters of the similar sudden events; Based on the benchmark parameters, the obtained conventional data is input into the GARCH model to obtain the volatility rate of the conventional data under the influence of the volatility data of abnormal fluctuations, and the first predicted conventional data; Based on the volatility rate, the risk level of the current sudden event is judged.

2. The method of claim 1, wherein: The preprocessing of the multi-dimensional data to form a standardized data set includes: Data cleaning is performed on the multi-dimensional data to remove outliers; The missing values in the multi-dimensional data are filled in; Unit standardization operation is performed on the multi-dimensional data to unify the data format; The standardized data set is generated and stored for calling by the LSTM-GARCH prediction model, ensuring the consistency of the standardized data set in time zone and unit.

3. The method of claim 1, wherein: The first regular data is represented as: ; wherein, represents the mean value corresponding to the regular data; represents the residual term.

4. The method of claim 1, wherein: The volatility is expressed as: ; wherein represents a constant term; represents a previous residual square; represents a previous conditional variance; represents an ARCH term coefficient; is a GARCH term coefficient; p represents a lag order of the residual square; and q represents a lag order of the conditional variance.

5. The method of claim 1, wherein: In the LSTM-GARCH prediction model, the GARCH model is trained using transfer learning, including: Obtain the historical transaction data of the plastic product and perform sudden event screening, and use the event data when the abnormal fluctuations of crude oil price, exchange rate and tariff occur as training data samples, label the sudden events that occur fluctuations and establish a sudden event database; Feature extraction is performed on the historical volatility data of different types of sudden events, including extreme point number, volatility rate and trend slope; According to the volatility data characteristics of different types of sudden events, the GARCH model is pre-trained to obtain a set of benchmark parameters; When detecting that the volatility data has abnormal fluctuations, real-time matching of similar sudden events is performed through the set of benchmark parameters and parameter transfer is performed; According to the GARCH model, the result is predicted, the prediction result is calibrated and fitted using the historical transaction data, the benchmark parameters are further optimized, and the training of the GARCH model is completed; the sudden event database is updated by detecting new sudden events.

6. The method of claim 1, wherein: The second regular data is represented as: ; wherein, represents the regular data at the current time point; represents the regular data at the previous time point.

7. A data analysis-based plastic e-commerce management platform, applying the plastic e-commerce management platform according to any one of claims 1-6 A business management method characterized by: Comprising: Data acquisition module: used for acquiring multi-dimensional data of plastic products on e-commerce platforms, the multi-dimensional data including conventional data reflecting market trends and volatility data reflecting sudden events; Data processing module: used for preprocessing the multi-dimensional data to form a standardized data set; Data analysis module: used for analyzing the standardized data set through a pre-trained LSTM-GARCH prediction model to output predicted market data, specifically including: inputting the standardized data set into the LSTM-GARCH prediction model; The LSTM-GARCH model includes a GARCH model and an LSTM model; the long-term change rule of the plastic product market is obtained by analyzing the conventional data through the LSTM model in the LSTM-GARCH prediction model; whether there is a sudden event influence is judged by analyzing the volatility data through the GARCH model in the LSTM-GARCH prediction model; the predicted market data is output according to the analysis result; Decision generation module: generates supply chain management decisions according to the predicted market data.

Images