A method and system for building a supplier supply data model based on big data

Abstract

The present invention discloses a method and system for constructing a supplier supply data model based on big data, and specifically relates to the field of data processing technology: by acquiring and organizing the structured and unstructured data of suppliers in real time, the model is updated regularly using a sliding window mechanism, and supplier performance is tracked dynamically. The method classifies the training data, focuses on evaluating the impact of uncommon parameters on the model prediction results, and evaluates the timeliness of the model by comprehensively analyzing the fluctuations of uncommon parameters and the accuracy of the model's dynamic tracking. According to the timeliness evaluation results, the prediction results are divided into high timeliness, medium timeliness and low timeliness, and corresponding optimization adjustments are made; through further analysis and optimization of the medium timeliness prediction results, the prediction accuracy and timeliness of the model can be improved, helping enterprises to better manage suppliers and enhance the stability and risk resistance of the supply chain.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and system for constructing a supplier supply data model based on big data. Background Art

[0002] The construction of a supplier supply data model based on big data refers to the use of big data technology to collect, organize and analyze relevant data of suppliers, so as to establish a comprehensive and dynamic model to help enterprises understand the various suppliers in the supply chain and their supply capabilities. In this model, enterprises can evaluate key indicators such as timeliness of supply, quality stability, cost-effectiveness, etc. of different suppliers, and even predict future supply risks or opportunities. This model based on big data technology can not only help enterprises optimize supplier management and selection, but also provide data support for the long-term strategic planning of the supply chain, and improve the overall efficiency and risk resistance of the supply chain. However, the supply chain is a highly dynamic system affected by many factors. If the model is not updated in time to reflect these changes in real time, it may lead to inaccurate predictions. At the same time, if the model cannot be updated according to the latest information, the enterprise may not be able to identify high-risk suppliers in time, resulting in the failure of the supply chain risk management system. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for building a supplier supply data model based on big data to address the shortcomings of the background technology.

[0004] In order to achieve the above object, the present invention provides the following technical solution: a method for constructing a supplier supply data model based on big data, comprising the following steps:

[0005] S1: Acquire supplier-related structured and unstructured data from several sources in real time, clean and integrate the collected data, and divide the data into training and test sets;

[0006] S2: Classify the data in the training set into common parameters and uncommon parameters, adjust the values ​​of uncommon parameters, determine the fluctuation range of the model output, and evaluate the influence of uncommon parameters on the model prediction results;

[0007] S3: Use the sliding window mechanism to regularly update the model's prediction results, determine the impact of uncommon parameters on supplier performance over time, and evaluate the accuracy of the model's tracking dynamics;

[0008] S4: Comprehensively analyze the impact of uncommon parameters on model prediction results and the accuracy of model tracking dynamics to evaluate the timeliness of model prediction results;

[0009] S5: Based on the evaluation results, the timeliness of the model prediction results is divided into high timeliness prediction results, medium timeliness prediction results and low timeliness prediction results, and the model is adjusted accordingly;

[0010] S6: For medium-term prediction results, further analyze the timeliness of the model prediction results within a fixed time period, and continuously optimize the model based on the analysis results.

[0011] Preferably, in S2, after determining the fluctuation range of the uncommon parameters on the model output, an uncommon interference index is generated to evaluate the influence of the uncommon parameters on the model prediction results. The uncommon interference index is obtained as follows:

[0012] Identify the uncommon parameters in the model and number them, labeling the market volatility index in the supply chain model as x 1 , Supplier financial health status is marked as x 2 , and construct a dataset of uncommon parameters {x 1 ,x 2 ,...,x i ,...,x n}; For each uncommon parameter x i , calculate its The marginal contribution in, for different feature subsets S, f(S) is the prediction result of the model using only feature subset S, f(S∪{x i}) Use S and x for the model i The prediction results of the features are calculated, and the difference in the predicted values ​​f(S∪{x i})-f(S), that is, parameter x i The marginal contribution in this subset is calculated by weighting the marginal contributions of all subsets S to calculate the value of each uncommon parameter x. i The uncommon interference index is calculated as follows: n is the total number of uncommon parameters in the model, and KN is the uncommon interference index.

[0013] Preferably, in S3, a logistics cost volatility index is generated according to the changing trend of the impact of uncommon parameters on supplier performance over time, and the accuracy of the model tracking dynamics is evaluated. The method for obtaining the logistics cost volatility index is:

[0014] Set the size of the sliding window W, and calculate the average value of the total logistics cost in each time window t. The expression is: n t represents the number of data entries in time window t, LC t represents the average logistics cost in time window t, TC iRefers to the transportation costs paid by the supplier to deliver the product to the customer's designated location. i is the oil price cost, WC i TC is the storage fee incurred by the supplier during the delivery process. O ther i It is the fee that needs to be paid when importing / exporting; the change value of logistics cost in each window compared with the previous window is calculated, that is, the logistics cost volatility is calculated, and the calculation formula is: LCVI t is the logistics cost volatility, LC t-1 It is the logistics cost of the time window t-1. The logistics cost volatility index is obtained by taking the weighted average sum of all logistics cost volatility rates generated within the total time window W.

[0015] Preferably, in S4, the influence of uncommon parameters on the model prediction results and the accuracy of the model tracking dynamics are comprehensively analyzed to evaluate the timeliness of the model prediction results;

[0016] The uncommon interference index and the logistics cost fluctuation index are normalized, and the timeliness analysis coefficient of the model prediction results is calculated by using the normalized uncommon interference index and the logistics cost fluctuation index.

[0017] Preferably, in S5, the timeliness of the model prediction results is divided into high timeliness prediction results, medium timeliness prediction results and low timeliness prediction results;

[0018] The obtained timeliness analysis coefficient of the model prediction result is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the timeliness analysis coefficient of the model prediction result is compared with the first standard threshold and the second standard threshold respectively;

[0019] If the timeliness analysis coefficient of the model prediction result is greater than the second standard threshold, it means that the timeliness of the model prediction result is high. At this time, a high timeliness signal is generated, and the timeliness of the model prediction result is classified as a high timeliness prediction result. The model responds quickly and accurately to dynamic changes and does not need to be adjusted.

[0020] If the timeliness analysis coefficient of the model prediction result is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the timeliness of the model prediction result is general. At this time, a medium timeliness signal is generated, and the timeliness of the model prediction result is classified as medium timeliness prediction result. The model has lag and needs to be optimized.

[0021] If the timeliness analysis coefficient of the model prediction result is less than the first standard threshold, it means that the timeliness of the model prediction result is low. At this time, a low timeliness signal is generated, and the timeliness of the model prediction result is classified as a low timeliness prediction result, indicating that the model has a large lag and needs to be adjusted and improved immediately.

[0022] Preferably, in S6, for the medium-time prediction results, the timeliness of the model prediction results within a fixed time period is further analyzed;

[0023] If the timeliness analysis coefficient of the model prediction results generated within a fixed time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the timeliness of the model prediction results is divided into medium timeliness prediction results, and the timeliness analysis coefficients of the model prediction results generated within the subsequent fixed time period after optimization and adjustment that are greater than or equal to the first standard threshold and less than or equal to the second standard threshold are collected, and a corresponding data set is established, and the mean and standard deviation of the data set are calculated. After analyzing the data set, the timeliness of the model prediction results is further analyzed, and the model is continuously optimized according to the analysis results.

[0024] Preferably, if the mean of the timeliness analysis coefficients in the data set is greater than or equal to the reference threshold of the timeliness analysis coefficient mean, and the standard deviation of the timeliness analysis coefficient is less than the reference threshold of the standard deviation of the timeliness analysis coefficient, it indicates that the timeliness prediction of the model is generally high and stable, and the model performance is stable and accurate, and no adjustment is required;

[0025] If the mean of the timeliness analysis coefficient is greater than or equal to the reference threshold of the mean of the timeliness analysis coefficient, and the standard deviation of the timeliness analysis coefficient is greater than or equal to the reference threshold of the standard deviation of the timeliness analysis coefficient, it means that the overall timeliness of the model is high but unstable, and the fluctuation of the timeliness prediction results is large. The model needs to be optimized to reduce the fluctuation;

[0026] If the mean of the timeliness analysis coefficient is less than the reference threshold of the timeliness analysis coefficient mean, and the standard deviation of the timeliness analysis coefficient is greater than or equal to the reference threshold of the timeliness analysis coefficient standard deviation, it means that the overall timeliness of the model is low and the volatility is large, and it needs to be redesigned;

[0027] If the mean of the timeliness analysis coefficient is less than the reference threshold of the mean of the timeliness analysis coefficient, and the standard deviation of the timeliness analysis coefficient is less than the reference threshold of the standard deviation of the timeliness analysis coefficient, the model has poor timeliness but is stable, and the prediction ability needs to be improved.

[0028] The present invention also provides a supplier supply data model construction system based on big data, including a data acquisition module, an interference impact analysis module, a dynamic prediction module, a timeliness analysis module, a timeliness classification module and an optimization module;

[0029] Data collection module: It obtains structured and unstructured data related to suppliers from several sources in real time, cleans and integrates the collected data, and divides the data into training sets and test sets;

[0030] Interference impact analysis module: classifies the data in the training set into common parameters and uncommon parameters, adjusts the values ​​of uncommon parameters, determines the fluctuation range of the model output, and evaluates the impact of uncommon parameters on the model prediction results;

[0031] Dynamic prediction module: Uses a sliding window mechanism to regularly update the model's prediction results, determine the changing trend of the impact of uncommon parameters on supplier performance over time, and evaluate the accuracy of the model's tracking dynamics;

[0032] Timeliness analysis module: comprehensively analyzes the impact of uncommon parameters on model prediction results and the accuracy of model tracking dynamics, and evaluates the timeliness of model prediction results;

[0033] Timeliness classification module: Based on the evaluation results, the timeliness of the model prediction results is divided into high timeliness prediction results, medium timeliness prediction results and low timeliness prediction results, and the model is adjusted accordingly;

[0034] Optimization module: For medium-term prediction results, further analyze the timeliness of the model prediction results within a fixed time period, and continuously optimize the model based on the analysis results.

[0035] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0036] 1. The present invention obtains structured and unstructured data related to suppliers in real time, cleans and integrates them, and then uses a sliding window mechanism to regularly update the model, dynamically analyzing the impact of uncommon parameters on supplier performance. This method can comprehensively evaluate the timeliness of the model prediction results by generating uncommon interference indexes and logistics cost volatility indexes, and classify the model prediction results into high, medium, and low timeliness according to the timeliness analysis coefficient, ensuring that the model can promptly reflect changes in the supply chain.

[0037] 2. The present invention can further improve the stability and accuracy of the model by continuously optimizing the model, especially for the medium-term prediction results, by analyzing the timeliness performance within a fixed time period and collecting subsequent data for evaluation. Adjusting the model accordingly according to the timeliness analysis results will help reduce forecast fluctuations and improve the model's response speed to dynamic changes, ultimately improving the company's decision-making efficiency in supply chain management, reducing risks, and ensuring the long-term stability and efficient operation of the supply chain. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0039] Figure 1 The figure is a flow chart of the method of the present invention.

[0040] Figure 2 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] Example 1, please refer to Figure 1 As shown, the method for constructing a supplier supply data model based on big data described in this embodiment includes the following steps:

[0043] S1: Acquire supplier-related structured and unstructured data from several sources in real time, clean and integrate the collected data, and divide the data into training and test sets;

[0044] S2: Classify the data in the training set into common parameters and uncommon parameters, adjust the values ​​of uncommon parameters, determine the fluctuation range of the model output, and evaluate the influence of uncommon parameters on the model prediction results;

[0045] S3: Use the sliding window mechanism to regularly update the model's prediction results, determine the impact of uncommon parameters on supplier performance over time, and evaluate the accuracy of the model's tracking dynamics;

[0046] S4: Comprehensively analyze the impact of uncommon parameters on model prediction results and the accuracy of model tracking dynamics to evaluate the timeliness of model prediction results;

[0047] S5: Based on the evaluation results, the timeliness of the model prediction results is divided into high timeliness prediction results, medium timeliness prediction results and low timeliness prediction results, and the model is adjusted accordingly;

[0048] S6: For medium-term prediction results, further analyze the timeliness of the model prediction results within a fixed time period, and continuously optimize the model based on the analysis results.

[0049] Among them, structured and unstructured data related to suppliers are obtained from several sources in real time, and after cleaning and integrating the collected data, the data is divided into training sets and test sets;

[0050] Structured data refers to data with a fixed format, usually coming from the company's internal system or supplier management system, including: ERP system: provides suppliers' historical order records, delivery time, invoice and payment information, etc. Supply chain management system: records suppliers' procurement, inventory management, transportation and other data. Quality management system (QMS): stores quality inspection reports and non-conforming rates of products delivered by suppliers. Financial system: provides suppliers' financial data, such as credit ratings, contract prices and payment cycles.

[0051] Unstructured data sources: Unstructured data has no fixed format and may come from external or non-traditional sources. This data may be very useful for predicting supplier performance. Common unstructured data sources include: Social media and news: Monitor social media comments, news events, or market sentiment related to suppliers to understand potential supplier risks or opportunities. Email and contract documents: Analyze key information in email communications and contract terms through natural language processing (NLP) technology. Industry reports and market trends: Obtain external market analysis reports to understand the impact of macroeconomic and industry changes on suppliers.

[0052] Check if there are missing values ​​in the data set and take appropriate measures to fill or delete them. For example: For a small number of missing values, they can be filled by interpolation (such as mean value filling) or inference methods based on similar data. For a large amount of missing data, it may be necessary to re-acquire the data or consider the reliability of the data source. For data from multiple sources, duplicate records may appear, and it is necessary to deduplicate through key fields such as ID and timestamp to ensure that each record is unique. After data cleaning is completed, integration is the process of unifying data from different sources into an analytical framework. Since different systems may use different data structures or terminology, special attention should be paid to the following aspects when integrating: Standardize data from different sources to ensure that their meanings are consistent. For example, convert qualitative descriptions such as "excellent", "good", and "poor" in supplier ratings into a unified score representation.

[0053] After completing the data cleaning and integration, the data is divided into a training set and a test set, specifically: usually the data is divided in a ratio of 80% training set and 20% test set. However, depending on the amount of data and the specific application scenario, the ratio can be adjusted appropriately, such as 70:30 or 90:10. If there are multiple categories in the data set (for example, different risk levels of suppliers), stratified sampling can be used to ensure that the proportion of each category in the training set and the test set is consistent with the overall data set. This ensures that the model learns the data characteristics of each category during the training process, and the performance of the model is not affected by data imbalance. If the data contains a time dimension (such as the historical performance data of the supplier), ensure that the time order is not disrupted when dividing. A common practice is to use early data as a training set and later data as a test set to evaluate the model's predictive ability in the future.

[0054] S2: Classify the data in the training set into common parameters and uncommon parameters, adjust the values ​​of uncommon parameters, determine the fluctuation range of the model output, and evaluate the influence of uncommon parameters on the model prediction results.

[0055] Classify the features in the training set and identify common and uncommon parameters;

[0056] Common parameters are usually used as key input parameters in most supply chain models. These parameters directly affect the core aspects of supply chain operations. Common parameters include: Supplier delivery time: indicates whether the supplier delivers on time, which is usually an important indicator to measure supplier performance. Product quality score: a quality assessment based on historical data, reflecting the quality of the products or services provided by the supplier. Historical order data: the supplier's past performance record helps predict future performance capabilities. Cost and pricing data: the price provided by the supplier directly affects the overall cost of the supply chain.

[0057] Uncommon parameters: These parameters are not commonly used, but in certain circumstances may significantly affect the accuracy of model output or supply chain decisions. Uncommon parameters are usually complex or difficult to obtain, involving external environment or hidden correlation factors. Examples of uncommon parameters include: Market volatility index: Fluctuations in the global raw material or financial markets may have an impact on the supplier's ability to perform. Supplier's financial health: The supplier's financial statements, debt situation, etc. will affect its long-term supply ability. Natural disasters and emergencies: Natural disasters or other uncontrollable emergencies in the supplier's location.

[0058] According to the business logic or historical data analysis, set a reasonable adjustment range for uncommon parameters. For example, the market volatility index can be set according to the range of changes in the past five years. Select different values ​​from the set adjustment range to make small progressive adjustments to uncommon parameters. For example, the market volatility index can start from 1.0 and be adjusted to 1.05, 1.10, 1.15, 1.20, or adjusted downward to 0.95, 0.90, etc. After each adjustment of uncommon parameters, run the model and record the output results. Calculate the change in model output before and after each adjustment of uncommon parameters. For example, if the supplier reliability score drops from 85 to 80 when the market volatility index is adjusted from 1.0 to 1.1, the change is (85-80) / 85=5.88%. Compare the effects of different uncommon parameter adjustments on model output and identify the parameters that have a greater impact on model output. Visualize the sensitivity results with charts and use bar charts or radar charts to show the sensitivity of each parameter to facilitate the identification of uncommon parameters that have a greater impact on model prediction results.

[0059] After determining the fluctuation range of the model output caused by uncommon parameters, an uncommon interference index is generated to evaluate the influence of uncommon parameters on the model prediction results. The method for obtaining the uncommon interference index is as follows:

[0060] Identify uncommon parameters in the model and number them, such as labeling the market volatility index in the supply chain model as x 1 , Supplier financial health status is marked as x 2 , and construct a dataset of uncommon parameters {x 1 ,x 2 ,...,x i ,...,x n}; For each uncommon parameter x i , calculate its The marginal contribution in, for different feature subsets S, f(S) is the prediction result of the model using only feature subset S, f(S∪{x i}) Use S and x for the model i The prediction results of the features are calculated, and the difference in the predicted values ​​f(S∪{x i})-f(S), that is, parameter x i The marginal contribution in this subset is calculated by weighting the marginal contributions of all subsets S to calculate the value of each uncommon parameter x. i The uncommon interference index is calculated as follows: n is the total number of uncommon parameters in the model, and KN is the uncommon interference index.

[0061] When the uncommon interference index is large, it means that the uncommon parameter has a greater impact on the model prediction results and will significantly affect the model output. For example, when the uncommon interference index of the market volatility index or the supplier's financial health is large, it means that changes in these parameters will cause significant fluctuations in the model's predicted results (such as supplier ratings or supply chain disruption risks).

[0062] When the uncommon interference index is small, it means that the uncommon parameter has little impact on the model prediction results and the model is not very sensitive to it. For example, when the uncommon interference index of a specific indicator of a certain supplier is small, it means that even if these parameters fluctuate greatly, the model's prediction results will not change significantly. The uncommon parameter has a weak effect in the model prediction, and its change will not significantly affect the overall prediction results.

[0063] S3: Use the sliding window mechanism to regularly update the model’s prediction results, determine the changing trend of the impact of uncommon parameters on supplier performance over time, and evaluate the accuracy of the model’s tracking dynamics.

[0064] Sliding windows are a time series analysis technique that extracts data within a fixed time period and continuously moves the time window forward to capture dynamic changes in data. The data in each window is used to train or update the model to help analyze changes in supplier performance over time. The size of the window depends on business needs and the frequency of data updates. Windows can be set by day, week, month, quarter, etc. For example, if supplier data is updated monthly, each month can be set as a window. After each window slides, the earliest time period data is removed and the latest time period data is added. In this way, the model can continuously adjust the forecast results with the latest data to adapt to changes in the market and supplier performance.

[0065] In each time window, the model is retrained or updated based on the latest supplier data. The latest changes in uncommon parameters are reflected in the model through the sliding window to ensure that the model can capture the changing trends of supplier performance in a timely manner. The frequency of window sliding should be determined based on the frequency of supplier data updates. If the financial health of the supplier is updated quarterly, the model update frequency can be set to once a quarter. For market fluctuations that affect the supply chain more frequently, updates can be made monthly or less frequently.

[0066] The logistics cost volatility index is generated based on the changing trend of the impact of uncommon parameters on supplier performance over time, and the accuracy of the model tracking dynamics is evaluated. The method for obtaining the logistics cost volatility index is as follows:

[0067] Set the size of the sliding window W, for example, each month or each quarter as a time window, and calculate the average value of the total logistics cost in each time window t. The expression is: nt represents the number of data entries in time window t (i.e., the number of shipping orders from the supplier), LC t represents the average logistics cost in time window t, TC i Refers to the transportation costs paid by the supplier to deliver the product to the customer's designated location, including air, sea, land and other costs. i is the oil price cost, WC i TC is the storage fee incurred by the supplier during the delivery process. O ther i It is the fee that needs to be paid when importing / exporting; calculate the relative change of logistics cost in each window compared with the previous window, that is, calculate the logistics cost volatility, the calculation formula is: LCVI t is the logistics cost volatility, LC t-1 It is the logistics cost of the time window t-1. The logistics cost volatility index is obtained by taking the weighted average sum of all logistics cost volatility rates generated within the total time window W.

[0068] When the logistics cost volatility index is large, it means that the logistics costs (such as transportation costs, oil prices, storage costs, etc.) have changed dramatically within the current time window. Drastic fluctuations in costs may lead to large fluctuations in suppliers' supply capabilities, delivery times, cost control, etc. If logistics costs fluctuate greatly and the model fails to accurately reflect these changes, the prediction error may increase. For example, the model may underestimate the impact of rising transportation costs on suppliers' on-time delivery, resulting in a greater deviation between the forecast and actual performance. At this point, the accuracy of the model's tracking dynamics is reduced and needs to be readjusted or optimized.

[0069] When the logistics cost volatility index is small, it means that the logistics cost is relatively stable in the current time window, the logistics cost factors in the supply chain do not change much, and the supplier's delivery time, fees, etc. are less affected by external costs. Since the logistics cost fluctuates less, the model does not need to adjust parameters or algorithms frequently, and the prediction error will be relatively small. In this case, the model's dynamic tracking ability is usually stronger, because the supplier's performance is more predictable, and the model can capture these patterns well.

[0070] S4: Comprehensively analyze the impact of uncommon parameters on the model prediction results and the accuracy of the model tracking dynamics to evaluate the timeliness of the model prediction results.

[0071] The uncommon interference index and the logistics cost fluctuation index are normalized, and the timeliness analysis coefficient of the model prediction results is calculated by using the normalized uncommon interference index and the logistics cost fluctuation index.

[0072] For example, the present invention can use the following formula to calculate the timeliness analysis coefficient of the model prediction result, and the calculation expression is: Where LK is the timeliness analysis coefficient of the model prediction results, KN ​​is the uncommon interference index, AS is the logistics cost fluctuation index, and a 1 、a 2 is the proportional coefficient of the uncommon interference index and the logistics cost fluctuation index, and a 2 >a 1 >0.

[0073] S5: Based on the evaluation results, the timeliness of the model prediction results is divided into high-timeliness prediction results, medium-timeliness prediction results and low-timeliness prediction results, and the model is adjusted accordingly.

[0074] The obtained timeliness analysis coefficient of the model prediction result is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the timeliness analysis coefficient of the model prediction result is compared with the first standard threshold and the second standard threshold respectively;

[0075] If the timeliness analysis coefficient of the model prediction result is greater than the second standard threshold, it means that the timeliness of the model prediction result is high. At this time, a high timeliness signal is generated, and the timeliness of the model prediction result is classified as a high timeliness prediction result. The model responds quickly and accurately to dynamic changes and does not need to be adjusted.

[0076] If the timeliness analysis coefficient of the model prediction result is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the timeliness of the model prediction result is general. At this time, a medium timeliness signal is generated, and the timeliness of the model prediction result is classified as medium timeliness prediction result. The model has lag, but the overall performance is good and needs to be optimized.

[0077] If the timeliness analysis coefficient of the model prediction result is less than the first standard threshold, it means that the timeliness of the model prediction result is low. At this time, a low timeliness signal is generated, and the timeliness of the model prediction result is classified as a low timeliness prediction result, indicating that the model has a large lag and needs to be adjusted and improved immediately.

[0078] Adjustment strategies for time-sensitive forecast results: Improve the model's response speed and accuracy to dynamic changes, reduce lag effects, and achieve a higher level of model timeliness. Consider more complex time series models or dynamic forecasting models, such as long short-term memory networks (LSTMs), ARIMA models, etc., to improve the model's ability to capture dynamic changes in time. For the sliding window mechanism, it may be necessary to reduce the window size and shorten the time span of each forecast so that the model can respond to supplier changes more quickly. Analyze the source of the lag effect, add lag features to historical data, and use these lag features to reduce the lag of model predictions. For example, increase the consideration of supplier performance data for the past few time periods to enable the model to better capture trends. If data is not updated frequently enough, it may cause model prediction lags. Increasing the frequency of data collection and processing can shorten the model's response time.

[0079] Adjustment strategies for low timeliness prediction results: Significantly improve the response speed of the model, reduce the prediction lag, and ensure that the model can capture dynamic changes. If the timeliness is very low, it means that the existing model may not be able to handle the dynamic changes in data. You can consider reselecting the algorithm, such as selecting an algorithm that can handle non-stationary data or more complex time series features (such as Prophet, time series network in deep learning). Low timeliness may be due to the model's failure to obtain the latest supply chain data. Introducing real-time data sources, such as logistics tracking systems, real-time market price data, etc., can enhance the model's perception of the latest changes. Adjust the model's prediction time span and shorten the time period of each prediction to reduce lag. If the model performs poorly for long-term predictions, you can try phased predictions or short-term predictions. If some uncommon parameters (such as emergencies, market fluctuations) affect the timeliness of the model, adjust the weights of these features in the model to ensure that the model pays more attention to these dynamic changes.

[0080] S6: For medium-term prediction results, further analyze the timeliness of the model prediction results within a fixed time period, and continuously optimize the model based on the analysis results.

[0081] If the timeliness analysis coefficient of the model prediction results generated within a fixed time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the timeliness of the model prediction results is divided into medium timeliness prediction results, and the timeliness analysis coefficients of the model prediction results generated within the subsequent fixed time period after optimization and adjustment that are greater than or equal to the first standard threshold and less than or equal to the second standard threshold are collected, and a corresponding data set is established, and the mean and standard deviation of the data set are calculated. After analyzing the data set, the timeliness of the model prediction results is further analyzed, and the model is continuously optimized according to the analysis results.

[0082] If the mean of the timeliness analysis coefficient in the data set is greater than or equal to the reference threshold of the timeliness analysis coefficient mean, and the standard deviation of the timeliness analysis coefficient is less than the reference threshold of the timeliness analysis coefficient standard deviation, it means that the timeliness prediction of the model is generally high and stable, the model performs well, and no major adjustments are needed. It may only need fine-tuning to further optimize the accuracy;

[0083] If the mean of the timeliness analysis coefficient is greater than or equal to the reference threshold of the mean of the timeliness analysis coefficient, and the standard deviation of the timeliness analysis coefficient is greater than or equal to the reference threshold of the standard deviation of the timeliness analysis coefficient, it means that the overall timeliness of the model is high but unstable, and the fluctuation of the timeliness prediction results is large. The model needs to be optimized to reduce the fluctuation, focusing on the timeliness of the unstable time period and improving the consistency of the model;

[0084] If the mean of the timeliness analysis coefficient is less than the reference threshold of the mean of the timeliness analysis coefficient, and the standard deviation of the timeliness analysis coefficient is greater than or equal to the reference threshold of the standard deviation of the timeliness analysis coefficient, it means that the overall timeliness of the model is low and the volatility is large. The model needs major adjustments and may need to be redesigned or a new algorithm may be used to improve the overall timeliness and consistency of the model.

[0085] If the mean of the timeliness analysis coefficient is less than the reference threshold of the mean of the timeliness analysis coefficient, and the standard deviation of the timeliness analysis coefficient is less than the reference threshold of the standard deviation of the timeliness analysis coefficient, the model has low timeliness and small fluctuations, and the timeliness results are stable but poor. The model needs to be significantly optimized, especially in improving the overall prediction timeliness. More dynamic features can be introduced or the data source can be updated to ensure that the model can better track changes.

[0086] Model optimization specifically includes: By collecting more supplier data, especially diversified historical data, the model can capture more patterns and reduce forecast fluctuations caused by insufficient data. Enhance the extraction and processing of important features, especially sensitivity analysis of uncommon parameters. Eliminate irrelevant features that contribute little to model predictions or have more noise to reduce unnecessary fluctuations. Adjust the size of the sliding window according to the periodicity and timeliness of data fluctuations so that the model can adapt to dynamic changes more flexibly. Use ensemble learning methods (such as random forests or XGBoost) to combine the results of multiple models to reduce the instability of single model predictions and improve the stability of overall predictions.

[0087] In this embodiment, structured and unstructured data are acquired from multiple sources in real time, and after cleaning and integration, they are divided into training sets and test sets. Next, the data in the training set is classified into common parameters and uncommon parameters, and the fluctuation range and degree of influence on the model output are evaluated by adjusting the values ​​of uncommon parameters. Using the sliding window mechanism, the model prediction results are updated regularly, the dynamic changes of uncommon parameters on supplier performance are tracked, and the dynamic accuracy of the model is evaluated. Then, the impact of uncommon parameters and the dynamic tracking accuracy of the model are comprehensively analyzed to evaluate the timeliness of the model prediction results. According to the evaluation, the timeliness is divided into three levels: high, medium, and low, and the model is adjusted accordingly. Finally, for the medium timeliness prediction results, the timeliness performance within a fixed time period is further analyzed, and the model is continuously optimized to improve the prediction accuracy.

[0088] Example 2, please refer to Figure 2 As shown, the supplier supply data model construction system based on big data described in this embodiment includes a data acquisition module, an interference impact analysis module, a dynamic prediction module, a timeliness analysis module, a timeliness classification module and an optimization module;

[0089] Data collection module: It obtains structured and unstructured data related to suppliers from several sources in real time, cleans and integrates the collected data, and divides the data into training sets and test sets;

[0090] Interference impact analysis module: classifies the data in the training set into common parameters and uncommon parameters, adjusts the values ​​of uncommon parameters, determines the fluctuation range of the model output, and evaluates the impact of uncommon parameters on the model prediction results;

[0091] Dynamic prediction module: Uses a sliding window mechanism to regularly update the model's prediction results, determine the changing trend of the impact of uncommon parameters on supplier performance over time, and evaluate the accuracy of the model's tracking dynamics;

[0092] Timeliness analysis module: comprehensively analyzes the impact of uncommon parameters on model prediction results and the accuracy of model tracking dynamics, and evaluates the timeliness of model prediction results;

[0093] Timeliness classification module: Based on the evaluation results, the timeliness of the model prediction results is divided into high timeliness prediction results, medium timeliness prediction results and low timeliness prediction results, and the model is adjusted accordingly;

[0094] Optimization module: For medium-term prediction results, further analyze the timeliness of the model prediction results within a fixed time period, and continuously optimize the model based on the analysis results.

[0095] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0096] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0097] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0098] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0099] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0100] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A method for constructing a supplier supply data model based on big data, characterized by: The steps include: S1: Acquire supplier-related structured and unstructured data from several sources in real time, clean and integrate the collected data, and divide the data into training and test sets; S2: Classify the data in the training set into common parameters and uncommon parameters, adjust the values ​​of uncommon parameters, determine the fluctuation range of the model output, and evaluate the influence of uncommon parameters on the model prediction results; The step S2 specifically includes: generating an uncommon interference index after determining the fluctuation range of the uncommon parameters on the model output, and evaluating the influence of the uncommon parameters on the model prediction results. The method for obtaining the uncommon interference index is: Determine the uncommon parameters in the model and number them, mark the market volatility index in the supply chain model as x1, the supplier financial health status as x2, and construct a data set of uncommon parameters {x1, x2, ..., x i ,...,x n }; For each uncommon parameter x i , calculate its The marginal contribution in , where X is a data set with uncommon parameters, for different feature subsets S, f(S) is the prediction result of the model using only feature subset S, f(S∪{x i }) Use S and x for the model i The prediction results of the features are calculated, and the difference in the predicted values ​​f(S∪{x i })-f(S), that is, parameter x i The marginal contribution in this subset is calculated by weighting the marginal contributions of all subsets S to calculate the value of each uncommon parameter x. i The uncommon interference index is calculated as: n is the total number of uncommon parameters in the model, KN is the uncommon interference index; S3: Use the sliding window mechanism to regularly update the model's prediction results, determine the impact of uncommon parameters on supplier performance over time, and evaluate the accuracy of the model's tracking dynamics; The step S3 specifically includes: generating a logistics cost fluctuation index according to the changing trend of the impact of uncommon parameters on supplier performance over time, and evaluating the accuracy of the model in tracking dynamics. The method for obtaining the logistics cost fluctuation index is: Set the size of the sliding window W, and calculate the average value of the total logistics cost in each time window t. The expression is: n t represents the number of data entries in time window t, LC t represents the average logistics cost in time window t, TC i Refers to the transportation costs paid by the supplier to deliver the product to the customer's designated location. i is the oil price cost, WC i TC is the storage fee incurred by the supplier during the delivery process. O ther i It is the fee that needs to be paid when importing / exporting; the change value of logistics cost in each window compared with the previous window is calculated, that is, the logistics cost volatility is calculated, and the calculation formula is: LCVI t is the logistics cost volatility, LC t-1 is the logistics cost of the time window t-1. The logistics cost volatility index is obtained by taking the weighted average sum of all logistics cost volatility rates generated within the total time window W; S4: Comprehensively analyze the impact of uncommon parameters on the model prediction results and the accuracy of the model tracking dynamics to evaluate the timeliness of the model prediction results; The step S4 specifically includes: comprehensively analyzing the influence of uncommon parameters on the model prediction results and the accuracy of the model tracking dynamics, and evaluating the timeliness of the model prediction results; The uncommon interference index and the logistics cost fluctuation index are normalized, and the timeliness analysis coefficient of the prediction result of the model is calculated by the normalized uncommon interference index and the logistics cost fluctuation index; S5: Based on the evaluation results, the timeliness of the model prediction results is divided into high timeliness prediction results, medium timeliness prediction results and low timeliness prediction results, and the model is adjusted accordingly; S6: For medium-term prediction results, further analyze the timeliness of the model prediction results within a fixed time period, and continuously optimize the model based on the analysis results; The step S6 specifically includes: for the medium-time prediction results, further analyzing the timeliness of the model prediction results within a fixed time period; If the timeliness analysis coefficient of the model prediction result generated within a fixed time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the timeliness of the model prediction result is divided into a medium timeliness prediction result, and the timeliness analysis coefficients of the model prediction results generated within the subsequent fixed time period after optimization and adjustment that are greater than or equal to the first standard threshold and less than or equal to the second standard threshold are collected, and a corresponding data set is established, and the mean and standard deviation of the data set are calculated. After analyzing the data set, the timeliness of the model prediction results is further analyzed, and the model is continuously optimized according to the analysis results; If the mean of the timeliness analysis coefficient in the data set is greater than or equal to the reference threshold of the timeliness analysis coefficient mean, and the standard deviation of the timeliness analysis coefficient is less than the reference threshold of the timeliness analysis coefficient standard deviation, it means that the timeliness prediction of the model is generally high and stable, and the model performance is stable and accurate, and no adjustment is required; If the mean of the timeliness analysis coefficient is greater than or equal to the reference threshold of the mean of the timeliness analysis coefficient, and the standard deviation of the timeliness analysis coefficient is greater than or equal to the reference threshold of the standard deviation of the timeliness analysis coefficient, it means that the overall timeliness of the model is high but unstable, and the fluctuation of the timeliness prediction results is large. The model needs to be optimized to reduce the fluctuation; If the mean of the timeliness analysis coefficient is less than the reference threshold of the timeliness analysis coefficient mean, and the standard deviation of the timeliness analysis coefficient is greater than or equal to the reference threshold of the timeliness analysis coefficient standard deviation, it means that the overall timeliness of the model is low and the volatility is large, and it needs to be redesigned; If the mean of the timeliness analysis coefficient is less than the reference threshold of the mean of the timeliness analysis coefficient, and the standard deviation of the timeliness analysis coefficient is less than the reference threshold of the standard deviation of the timeliness analysis coefficient, the model has poor timeliness but is stable, and the prediction ability needs to be improved.

2. The method for constructing a supplier supply data model based on big data according to claim 1, characterized in that: In S5, the timeliness of the model prediction results is divided into high timeliness prediction results, medium timeliness prediction results and low timeliness prediction results; The obtained timeliness analysis coefficient of the model prediction result is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the timeliness analysis coefficient of the model prediction result is compared with the first standard threshold and the second standard threshold respectively; If the timeliness analysis coefficient of the model prediction result is greater than the second standard threshold, it means that the timeliness of the model prediction result is high. At this time, a high timeliness signal is generated, and the timeliness of the model prediction result is classified as a high timeliness prediction result. The model responds quickly and accurately to dynamic changes and does not need to be adjusted. If the timeliness analysis coefficient of the model prediction result is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the timeliness of the model prediction result is general. At this time, a medium timeliness signal is generated, and the timeliness of the model prediction result is classified as medium timeliness prediction result. The model has lag and needs to be optimized. If the timeliness analysis coefficient of the model prediction result is less than the first standard threshold, it means that the timeliness of the model prediction result is low. At this time, a low timeliness signal is generated, and the timeliness of the model prediction result is classified as a low timeliness prediction result, indicating that the model has a large lag and needs to be adjusted and improved immediately.

3. A supplier supply data model construction system based on big data, used to implement a supplier supply data model construction method based on big data as described in any one of claims 1-2, characterized in that: It includes data acquisition module, interference impact analysis module, dynamic prediction module, timeliness analysis module, timeliness classification module and optimization module; Data collection module: It obtains structured and unstructured data related to suppliers from several sources in real time, cleans and integrates the collected data, and divides the data into training sets and test sets; Interference impact analysis module: classifies the data in the training set into common parameters and uncommon parameters, adjusts the values ​​of uncommon parameters, determines the fluctuation range of the model output, and evaluates the impact of uncommon parameters on the model prediction results; Dynamic prediction module: Uses a sliding window mechanism to regularly update the model's prediction results, determine the changing trend of the impact of uncommon parameters on supplier performance over time, and evaluate the accuracy of the model's tracking dynamics; Timeliness analysis module: comprehensively analyzes the impact of uncommon parameters on model prediction results and the accuracy of model tracking dynamics, and evaluates the timeliness of model prediction results; Timeliness classification module: Based on the evaluation results, the timeliness of the model prediction results is divided into high timeliness prediction results, medium timeliness prediction results and low timeliness prediction results, and the model is adjusted accordingly; Optimization module: For medium-term prediction results, further analyze the timeliness of the model prediction results within a fixed time period, and continuously optimize the model based on the analysis results.

Images