Material purchasing data processing method and system based on safety risk identification and prediction

Abstract

The invention discloses a material purchasing data processing method and system based on safety risk identification and prediction. The method comprises the following steps: material purchasing data acquisition and preprocessing; material purchasing data feature extraction and optimization; constructing a multi-task learning model; training and optimizing a multi-task learning model; material purchasing data security risk identification and dynamic evaluation; according to the method, multiple risks in material purchasing are predicted based on the multi-task learning model, multi-task learning is optimized through the self-adaptive weighted loss function, the coverage range of risk identification is enlarged, the prediction precision is improved, the accuracy and the real-time performance of material purchasing risk prediction are remarkably improved, and the risk prediction efficiency is improved. And real-time early warning and business strategy suggestions are provided, so that the intelligent level of material purchasing management is improved, the prevention and control capability of enterprises on material purchasing risks is enhanced, and the uncertainty of supply chain operation is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and particularly relates to a method and system for processing material procurement data based on security risk identification and prediction. Background Art

[0002] With the acceleration of the globalization of the supply chain and the increasing uncertainty of the market environment, the complexity of material procurement management has been increasing day by day. Enterprises face various risks in the procurement process, such as supplier defaults, price fluctuations, delivery delays, inventory shortages, and material quality problems. These problems not only affect the execution of the procurement plan, but may also have an adverse impact on production operations, financial conditions, and market competitiveness.

[0003] The existing procurement risk management methods mainly rely on manual experience and historical data analysis, lacking real-time and comprehensiveness. The existing methods are usually based on fixed rules or statistical models, and it is difficult to fully integrate multi-dimensional data such as procurement orders, supplier information, market conditions, and inventory status, resulting in limited ability to identify complex procurement risks. Moreover, due to the dynamic changes in the procurement environment, the existing methods often can only conduct post-event analysis, and it is difficult to predict potential risks in a timely manner, lacking intelligent decision-making support when dealing with emergencies. Therefore, we propose a method and system for processing material procurement data based on security risk identification and prediction. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for processing material procurement data based on security risk identification and prediction to solve the problems mentioned in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A method for processing material procurement data based on security risk identification and prediction, including the following steps:

[0006] A. Acquisition and preprocessing of material procurement data:

[0007] Collect raw material procurement data from multiple material procurement data sources, and preprocess the collected raw material procurement data;

[0008] B. Feature extraction and optimization of material procurement data:

[0009] Extract material procurement risk features and time series features from the preprocessed raw material procurement data, generate interaction features of inventory turnover rate and market price fluctuations, screen out a feature set for predicting the security risk of material procurement data, optimize the screened feature set, construct high-order features based on the business requirements of material procurement, and generate an optimized feature set;

[0010] C. Construction of a multi-task learning model:

[0011] The multi-layer perceptron is used as the shared feature extraction layer, and the optimized feature set generated in the prediction of the safety risk of material procurement data is learned through a multi-layer fully connected network. According to the task of predicting the safety risk of material procurement data, multiple specific output layers for the task of predicting the safety risk of material procurement data are constructed;

[0012] D. Training and optimization of the multi-task learning model:

[0013] The multi-task learning model is trained and optimized through loss function optimization, hyperparameter optimization, deep learning optimization algorithms, and optimization of the model generalization ability;

[0014] E. Identification and dynamic assessment of the safety risk of material procurement data:

[0015] Based on the multi-task learning model, the safety risks in material procurement are identified and dynamically assessed through material procurement data;

[0016] F. Generation of early warning information for material procurement data and countermeasures:

[0017] Multi-category early warning information for material procurement data is generated based on the results of the identification and assessment of the safety risk of material procurement data, and corresponding countermeasures are provided according to the multi-category early warning information for material procurement data.

[0018] Preferably, in step A, the specific process of collecting and preprocessing the material procurement data is as follows:

[0019] A1. Collect the original material procurement data from multiple material procurement data sources for subsequent analysis and preprocessing of the original material procurement data. The original material procurement data includes the procurement quantity, procurement price, supplier reputation, inventory turnover rate, market price fluctuation, and supply and demand relationship;

[0020] A2. Clean the collected original material procurement data, remove outliers and duplicate values, and use the interpolation method to fill in the missing values. Then, perform unified standardization processing on the original material procurement data with different dimensions, and optimize the quality of the original material procurement data through denoising technology to eliminate potential errors.

[0021] Preferably, in step B, the specific process of feature extraction and optimization of the material procurement data is as follows:

[0022] B1. Extract the material procurement risk features and time series features from the preprocessed original material procurement data, and generate the interaction features of the inventory turnover rate and market price fluctuation to reflect the comprehensive impact of inventory pressure and market dynamics. Capture the long-term trend and periodic fluctuations of the original material procurement data according to the time series features, analyze and process the material procurement risk features, time series features, and interaction features, and screen out the feature set for predicting the safety risk of the material procurement data;

[0023] B2. Optimize the selected feature set, construct high-order features based on the material procurement business requirements, generate an optimized feature set, and provide an input for the multi-task learning model;

[0024] The high-order features include non-linear combined features of inventory turnover rate and supplier reputation.

[0025] Preferably, in step C, the specific process of constructing the multi-task learning model is as follows:

[0026] C1. Construction of the shared feature extraction layer:

[0027] Use a multi-layer perceptron as the shared feature extraction layer, and learn the optimized feature set generated in the prediction of material procurement data security risks through a multi-layer fully connected network. The shared features include purchase orders, supplier information, inventory data, market conditions, and historical purchase records;

[0028] The output of the shared feature extraction layer is non-linearly mapped through the ReLU activation function. The output formula of the shared feature extraction layer is shown in formula (1):

[0029] H l = ReLU (W l H l-1 + b l ) (1);

[0030] Among them, H l represents the output vector of the l-th layer of the shared feature extraction layer;

[0031] W l represents the weight matrix of the l-th layer of the shared feature extraction layer;

[0032] b l represents the bias term of the l-th layer of the shared feature extraction layer;

[0033] ReLU() represents the activation function, and its output is defined as: ReLU(x) = max(0, x);

[0034] C2. Construction of the task-specific output layer:

[0035] According to the supplier risk prediction task, delivery delay risk prediction task, market condition risk prediction task, price fluctuation range prediction task, and material quality risk prediction task, construct a supplier risk prediction task-specific output layer, a delivery delay risk prediction task-specific output layer, a market condition risk prediction task-specific output layer, a price fluctuation range prediction task-specific output layer, and a material quality risk prediction task-specific output layer respectively.

[0036] Preferably, in step C2,

[0037] Specific output layer for supplier risk prediction task: Based on the historical performance records and supply stability characteristics of suppliers, output the risk probability of supplier default. The output formula of the specific output layer for the supplier risk prediction task is shown in Formula (2):

[0038] (2);

[0039] where z 违约 represents the raw score of the specific output layer for the supplier risk prediction task;

[0040] P 违约 represents the risk probability of supplier default. The output value range is [0, 1]. An output value close to 1 indicates a high risk probability of supplier default, and an output value close to 0 indicates a low risk probability of supplier default;

[0041] Then the calculation formula of z 违约 is shown in Formula (3):

[0042] z 违约 = W 违约 H L + b 违约 (3);

[0043] where W 违约 represents the weight matrix of the specific output layer for the supplier risk prediction task;

[0044] b 违约 represents the bias vector of the specific output layer for the supplier risk prediction task;

[0045] H L represents the output vector of the L-th layer of the shared feature extraction layer;

[0046] Specific output layer for delivery delay risk prediction task: Combining the delivery cycle and supplier delivery on-time rate characteristics, output the risk probability of delivery delay. The output formula of the specific output layer for the delivery delay risk prediction task is shown in Formula (4):

[0047] (4);

[0048] where z 延误 represents the raw score of the specific output layer for the delivery delay risk prediction task;

[0049] P 延误 represents the risk probability of delivery delay. The output value range is [0, 1]. An output value close to 1 indicates a high risk probability of delivery delay, and an output value close to 0 indicates a low risk probability of delivery delay;

[0050] Then the calculation formula of z 延误 is shown in Formula (5):

[0051] z 延误 =W 延误 H L +b 延误 (5);

[0052] Among them, W 延误 represents the weight matrix of the specific output layer of the delivery delay risk prediction task;

[0053] b 延误 represents the bias vector of the specific output layer of the delivery delay risk prediction task;

[0054] Specific output layer of the market condition risk prediction task: According to the supply-demand relationship and market trend characteristics, output the market condition risk level. Then, the output formula of the specific output layer of the market condition risk prediction task is as shown in formula (6):

[0055] (6);

[0056] Among them, i represents the current risk level category;

[0057] j represents all risk level categories;

[0058] P 市场风险 (i)represents the prediction probability of the i-th category;

[0059] Softmax represents a function that converts scores into a market condition risk probability distribution. The output value is the probability distribution of multiple risk levels, and the sum is 1;

[0060] z i represents the original score of the current i-th category, that is, the linear combination result corresponding to the i-th category of the specific output layer of the market condition risk prediction task;

[0061] Specific output layer of the price fluctuation range prediction task: Based on the historical market price fluctuations and real-time supply-demand relationship characteristics, use a linear activation function to predict continuous variables, and output the predicted value of the price fluctuation range. Then, the output formula of the specific output layer of the price fluctuation range prediction task is as shown in formula (7):

[0062] (7);

[0063] Among them, represents the price fluctuation range predicted by the specific output layer of the price fluctuation range prediction task, and the output predicted value is a continuous numerical value;

[0064] H L represents the output vector of the L-th layer of the shared feature extraction layer;

[0065] W 价格It represents the weight matrix of the specific output layer for the price fluctuation amplitude prediction task;

[0066] b 价格 It represents the bias vector of the specific output layer for the price fluctuation amplitude prediction task;

[0067] Specific output layer for the material quality risk prediction task: Based on the material quality feedback records and supplier qualification characteristics, it outputs the occurrence probability of material quality problems. The output formula of the specific output layer for the material quality risk prediction task is as shown in formula (8):

[0068] (8);

[0069] Among them, P 质量风险 represents the risk probability of material quality problems. The output value range is [0, 1]. An output value close to 1 indicates a high risk probability of material quality problems, and an output value close to 0 indicates a low risk probability of material quality problems;

[0070] z 质量风险 represents the raw score of the specific output layer for the material quality risk prediction task;

[0071] Then z 质量风险 is calculated as shown in formula (9):

[0072] z 质量风险 = W 质量风险 H L + b 质量风险 (9);

[0073] W 质量风险 represents the weight matrix of the specific output layer for the material quality risk prediction task;

[0074] b 质量风险 represents the bias vector of the specific output layer for the material quality risk prediction task.

[0075] Preferably, in step D, training and optimization of the multi-task learning model: The specific process of training and optimizing the multi-task learning model through loss function optimization, hyperparameter optimization, deep learning optimization algorithms, and model generalization ability optimization is as follows:

[0076] D1. Loss function optimization: For the specific output layers of multiple material procurement data security risk prediction tasks, based on the loss function, by weighted summing the loss values of each specific output layer for the material procurement data security risk prediction task, the multi-task learning model is optimized. The formula of the loss function is as shown in formula (10):

[0077] (10);

[0078] Among them, N represents the total number of tasks;

[0079] represents the loss function of the i-th task;

[0080] represents the task weight;

[0081] and respectively represent the predicted value and the true value of the i-th task;

[0082] D2. Hyperparameter Optimization: Combining the importance of safety risks and data distribution in material procurement, the hyperparameters of the multi-task learning model are tuned through grid search, including the learning rate, task weight and batch size;

[0083] D3. Deep Learning Optimization Algorithm: The parameters of the multi-task learning model are updated using a deep learning optimization algorithm for learning dynamic features in material procurement data. The formula of the deep learning optimization algorithm is shown in Formula (11):

[0084] (11);

[0085] where and respectively represent the current and updated parameters of the multi-task learning model;

[0086] represents the learning rate;

[0087] represents the first-order momentum estimate of the gradient, smoothing the change of the gradient;

[0088] represents the second-order momentum estimate of the gradient, adjusting the learning rate;

[0089] represents a small constant to prevent division-by-zero errors;

[0090] D4. Model Generalization Ability Optimization: The adaptability of the multi-task learning model to the actual material procurement business is optimized through cross-validation and early stopping mechanism. Among them,

[0091] Cross-validation: The material procurement dataset is divided into a training set and a validation set to make the performance of the multi-task learning model consistent on different data subsets;

[0092] Early stopping mechanism: When the performance of the validation set no longer improves, the training process of the multi-task learning model is automatically stopped.

[0093] Preferably, in step E, the identification and dynamic assessment of the safety risks of material procurement data: Based on the multi-task learning model, the specific process of identifying and dynamically assessing the safety risks in material procurement through material procurement data is as follows:

[0094] E1. Risk level classification:

[0095] Segment the prediction results of each specific task output layer and divide them into three risk levels: low risk, medium risk, and high risk;

[0096] E2. Risk level assessment and determination:

[0097] Supplier risk prediction and assessment: The probability value predicted by the specific output layer of the supplier risk prediction task is 0 - 1, and it is divided into: 0 - 0.5; 0.5 - 0.8; 0.8 - 1;

[0098] Among them, 0 - 0.5 indicates low risk;

[0099] 0.5 - 0.8 indicates medium risk;

[0100] 0.8 - 1 indicates high risk;

[0101] Delivery delay risk prediction and assessment: The probability value predicted by the specific output layer of the delivery delay risk prediction task is 0 - 1, and it is divided into: 0 - 0.4; 0.4 - 0.7; 0.7 - 1;

[0102] Among them, 0 - 0.4 indicates low risk;

[0103] 0.4 - 0.7 indicates medium risk;

[0104] 0.7 - 1 indicates high risk;

[0105] Market condition risk prediction and assessment: The classification probability predicted by the specific output layer of the market condition risk prediction task, where the category with the highest probability is the risk level category for assessment and determination;

[0106] Price fluctuation range prediction and assessment: The percentage of the price fluctuation range predicted by the specific output layer of the price fluctuation range prediction task, and it is divided into: less than or equal to 5%; 5 - 10%; greater than or equal to 10%;

[0107] Among them, less than or equal to 5% indicates low risk;

[0108] 5 - 10% indicates medium risk;

[0109] Greater than or equal to 10% indicates high risk;

[0110] Material quality risk prediction and assessment: The probability value predicted by the specific output layer of the material quality risk prediction task is between 0 and 1, which is divided into: 0 - 0.5; 0.5 - 0.7; 0.7 - 1;

[0111] Among them, 0 - 0.5 indicates low risk;

[0112] 0.5 - 0.7 indicates medium risk;

[0113] 0.7 - 1 indicates high risk.

[0114] Preferably, in step F, the early warning information generation and response strategy for material procurement data: The specific process of generating multi - category material procurement data early warning information based on the results of material procurement data security risk identification and assessment and providing corresponding response strategies is as follows:

[0115] F1. Generation of material procurement data early warning information, specifically including:

[0116] Supplier risk early warning: Generate early warning information including supplier name, default probability, and recommended supplier replacement;

[0117] Delivery delay risk early warning: Generate early warning information including order number, delivery delay probability, and recommended delivery cycle adjustment;

[0118] Market condition risk early warning: Generate a market dynamic analysis report, clarify low, medium, and high risk levels and their corresponding suggestions;

[0119] Price fluctuation early warning: Generate price adjustment suggestions and provide decision - making support for locking prices or adjusting purchase quantities;

[0120] Material quality risk early warning: Generate a probability prediction of quality problems and provide corresponding supplier audit suggestions or quality control plans;

[0121] F2. Provide response strategies according to the early warning risk level:

[0122] Low risk: It is recommended to maintain the existing material procurement plan;

[0123] Medium risk: It is recommended to review the supplier's reputation, inventory status, or quality inspection records and pay attention to market dynamics;

[0124] High risk: It is recommended to immediately take emergency measures, such as replacing suppliers, adjusting the procurement plan, implementing more strict quality inspection processes, or expanding inventory reserves.

[0125] A material procurement data processing system based on security risk identification and prediction, which is applied to a material procurement data processing method based on security risk identification and prediction, includes:

[0126] Data acquisition module: It is used to collect the original material procurement data from multiple material procurement data sources, including procurement quantity, procurement price, supplier reputation, inventory turnover rate, market price fluctuations, and supply and demand relationships;

[0127] Data processing module: It is used to clean, denoise, and normalize the collected original material procurement data, extract material procurement risk characteristics from the preprocessed original material procurement data, screen out the feature set for predicting the material procurement data security risk, optimize the screened feature set, construct high-order features based on the material procurement business requirements, and generate an optimized feature set;

[0128] Multi-task learning model module: It identifies and predicts multiple material procurement security risks through a shared feature extraction layer and multiple task-specific output layers;

[0129] Risk output module: It is used to evaluate the risk of real-time material procurement data using the trained multi-task learning model, output the risk level or risk value, generate early warning information, and provide corresponding countermeasures or measures.

[0130] The present invention utilizes multi-source material procurement data, including purchase orders, supplier information, inventory data, market conditions, and historical procurement records, etc., to construct an intelligent multi-task learning model. Based on the multi-task learning framework of deep learning, the present invention adopts a shared feature extraction layer to capture the global data pattern, and combines task-specific output layers to respectively predict supplier default risk, price fluctuation risk, delivery delay risk, market condition change risk, and material quality risk. By optimizing multi-task learning through an adaptive weighted loss function, the coverage of risk identification and the prediction accuracy are improved, significantly enhancing the accuracy and real-time performance of material procurement risk prediction. The aim is to solve multiple risks such as supplier default, price fluctuation, delivery delay, inventory shortage, and material quality problems during the procurement process, and provide real-time early warning and business strategy suggestions to improve the intelligent level of material procurement management, enhance the enterprise's ability to prevent and control material procurement risks, and reduce the uncertainty of supply chain operation;

[0131] Compared with the traditional method that relies on manual experience and historical data, the present invention has the following significant advantages:

[0132] (1) Improve the risk prediction accuracy: By optimizing feature engineering and the multi-task learning model, the present invention can accurately identify multiple procurement risks such as supplier default, price fluctuation, and delivery delay, and the risk identification accuracy is increased by 8 - 10%;

[0133] (2) Early warning of potential risks: By real-time analyzing procurement data, the present invention can provide early warning 3 - 5 days before high-risk events occur, helping enterprises take countermeasures in advance and reducing the risk of supply chain interruption and financial losses;

[0134] (3) Provide intelligent decision-making support: Based on the risk assessment results, the present invention can generate accurate early warning information and provide specific coping strategies according to different risk levels, such as supplier replacement, procurement quantity adjustment, etc., optimizing the procurement decision-making process. BRIEF DESCRIPTION OF THE DRAWINGS

[0135] Figure 1 It is a schematic diagram of the method steps of the present invention;

[0136] Figure 2 It is a comparison chart of the risk prediction accuracy rates between the method of the present invention and the traditional method. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0137] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0138] Please refer to Figure 1 , a method for processing material procurement data based on security risk identification and prediction provided by the present invention includes the following steps:

[0139] A. Collection and preprocessing of material procurement data:

[0140] Collect the original material procurement data from multiple material procurement data sources and preprocess the collected original material procurement data;

[0141] The specific process is as follows:

[0142] A1. Collect the original material procurement data from multiple material procurement data sources for subsequent analysis and preprocessing of the original material procurement data. The original material procurement data includes procurement quantity, procurement price, supplier reputation, inventory turnover rate, market price fluctuation, and supply and demand relationship. The original material procurement data is shown in Table 1 below:

[0143] Table 1:

[0144]

[0145] A2. Clean the collected original material procurement data, remove outliers and duplicate values, and use the interpolation method to fill in the missing values. Then, perform unified standardization processing on the original material procurement data with different dimensions, and optimize the quality of the original material procurement data through denoising technology to eliminate potential errors;

[0146] Data preprocessing aims to improve data quality and ensure its consistency. The present invention adopts methods such as data cleaning, missing value imputation, normalization, and noise reduction:

[0147] Data cleaning is mainly used to remove duplicate data and detect outliers. For example, if the credit score of a certain supplier exceeds the reasonable range of 0 - 100, it will be excluded or corrected;

[0148] Missing value imputation uses methods such as mean interpolation or nearest neighbor imputation to ensure data integrity;

[0149] Normalization processing uses Min - Max scaling to map the data to the interval [0, 1] to reduce the impact of numerical scale differences;

[0150] B. Feature extraction and optimization of material procurement data:

[0151] Extract material procurement risk features and time - series features from the pre - processed original material procurement data, generate interaction features of inventory turnover rate and market price fluctuations, screen out the feature set for predicting the security risk of material procurement data, optimize the screened feature set, construct high - order features based on the business requirements of material procurement, and generate an optimized feature set;

[0152] The specific process is as follows:

[0153] B1. Extract material procurement risk features and time - series features from the pre - processed original material procurement data, generate interaction features of inventory turnover rate and market price fluctuations to reflect the comprehensive impact of inventory pressure and market dynamics, capture the long - term trend and periodic fluctuations of the original material procurement data according to the time - series features, and analyze and process the material procurement risk features, time - series features, and interaction features to screen out the feature set for predicting the security risk of material procurement data;

[0154] B2. Optimize the screened feature set, construct high - order features based on the business requirements of material procurement, generate an optimized feature set, and provide input for the multi - task learning model;

[0155] The high - order features include non - linear combination features of inventory turnover rate and supplier reputation;

[0156] The feature extraction strategy of the present invention includes basic feature, interaction feature, and high - order feature optimization;

[0157] The basic features include purchase quantity, purchase price, inventory turnover rate, supplier reputation, etc., and these features can be directly obtained from the dataset;

[0158] Interaction features reflect deeper risk patterns by calculating the relationships between variables. For example, the present invention constructs interaction features between inventory turnover rate and market price fluctuations to measure the impact of the market environment on inventory risk. The calculation formula is as follows:

[0159] ;

[0160] In terms of high-order feature optimization, the present invention generates non-linear combined features of inventory turnover rate and supplier reputation based on the requirements of material procurement business to enhance the ability of the multi-task learning model to capture complex risk patterns. For example, the method of exponential transformation is used to calculate this feature:

[0161] ;

[0162] Among them, is an adjustment coefficient that controls the weight of the impact of inventory turnover rate on supplier reputation;

[0163] C. Construction of multi-task learning model:

[0164] A multi-layer perceptron is used as the shared feature extraction layer, and the optimized feature set generated in the prediction of material procurement data security risk is learned through a multi-layer fully connected network. According to the material procurement data security risk prediction task, multiple specific output layers for material procurement data security risk prediction are constructed;

[0165] The specific process is as follows:

[0166] C1. Construction of the shared feature extraction layer:

[0167] A multi-layer perceptron is used as the shared feature extraction layer, and the optimized feature set generated in the prediction of material procurement data security risk is learned through a multi-layer fully connected network. The shared features include purchase orders, supplier information, inventory data, market conditions, and historical purchase records;

[0168] The output of the shared feature extraction layer is non-linearly mapped through the ReLU activation function. The output formula of the shared feature extraction layer is shown in formula (1):

[0169] H l =ReLU (W l H l-1 +b l ) (1);

[0170] Among them, H l represents the output vector of the l-th layer of the shared feature extraction layer;

[0171] W l represents the weight matrix of the l-th layer of the shared feature extraction layer;

[0172] b l represents the bias term of the l-th layer of the shared feature extraction layer;

[0173] ReLU() represents the activation function, and its output is defined as: ReLU(x) = max(0, x);

[0174] The input data of the multi-task learning model includes purchase orders, supplier information, inventory data, market conditions, and historical purchase records, etc. After data preprocessing and feature engineering, a high-dimensional feature set is formed. These features are input into the shared feature extraction layer to learn the global procurement risk pattern. The shared feature extraction layer adopts a multi-layer perceptron (MLP) structure, which includes multiple fully connected layers and uses the ReLU activation function to enhance the non-linear modeling ability;

[0175] C2. Construction of the task-specific output layer:

[0176] According to the supplier risk prediction task, delivery delay risk prediction task, market condition risk prediction task, price fluctuation range prediction task, and material quality risk prediction task, a task-specific output layer for supplier risk prediction, a task-specific output layer for delivery delay risk prediction, a task-specific output layer for market condition risk prediction, a task-specific output layer for price fluctuation range prediction, and a task-specific output layer for material quality risk prediction are constructed respectively;

[0177] Task-specific output layer for supplier risk prediction: Based on the historical performance records and supply stability characteristics of the supplier, output the risk probability of supplier default. The output formula of the task-specific output layer for supplier risk prediction is as shown in formula (2):

[0178] (2);

[0179] where z 违约 represents the raw score of the task-specific output layer for supplier risk prediction;

[0180] P 违约 represents the risk probability of supplier default, and the output value range is [0, 1]. An output value close to 1 indicates a high risk probability of supplier default, and an output value close to 0 indicates a low risk probability of supplier default;

[0181] Then the calculation formula of z 违约 is as shown in formula (3):

[0182] z 违约 =W 违约 H L +b 违约 (3);

[0183] where W 违约Represents the weight matrix of the specific output layer for the supplier risk prediction task;

[0184] b 违约 Represents the bias vector of the specific output layer for the supplier risk prediction task;

[0185] H L Represents the output vector of the L-th layer of the shared feature extraction layer;

[0186] Specific output layer for the delivery delay risk prediction task: Combining the delivery cycle and supplier delivery on-time rate features, it outputs the risk probability of delivery delay. Then, the output formula of the specific output layer for the delivery delay risk prediction task is shown in Formula (4):

[0187] (4);

[0188] where, z 延误 Represents the raw score of the specific output layer for the delivery delay risk prediction task;

[0189] P 延误 Represents the risk probability of delivery delay. The output value range is [0, 1]. An output value close to 1 indicates a high risk probability of delivery delay, and an output value close to 0 indicates a low risk probability of delivery delay;

[0190] Then, the calculation formula of z 延误 is shown in Formula (5):

[0191] z 延误 = W 延误 H L + b 延误 (5);

[0192] where, W 延误 Represents the weight matrix of the specific output layer for the delivery delay risk prediction task;

[0193] b 延误 Represents the bias vector of the specific output layer for the delivery delay risk prediction task;

[0194] Specific output layer for the market condition risk prediction task: Based on the supply-demand relationship and market trend features, it outputs the market condition risk level. Then, the output formula of the specific output layer for the market condition risk prediction task is shown in Formula (6):

[0195] (6);

[0196] where, i represents the current risk level category;

[0197] j represents all risk level categories;

[0198] P 市场风险(i) represents the predicted probability of the i-th category;

[0199] Softmax represents a function that converts scores into a probability distribution of market condition risks. The output values are the probability distributions of multiple risk levels, and the sum is 1. The market condition risks are divided into low risk (0 - 0.3), medium risk (0.3 - 0.7), and high risk (0.7 - 1.0);

[0200] z i represents the original score of the current i-th category, that is, the linear combination result corresponding to the i-th category of the specific output layer of the market condition risk prediction task;

[0201] Specific output layer of the price fluctuation amplitude prediction task: Based on the historical market price fluctuations and real-time supply and demand relationship characteristics, it outputs the predicted value of the price fluctuation amplitude. Then, the output formula of the specific output layer of the price fluctuation amplitude prediction task is as shown in formula (7):

[0202] (7);

[0203] where, represents the price fluctuation amplitude predicted by the specific output layer of the price fluctuation amplitude prediction task, and the output predicted value is a continuous numerical value;

[0204] H L represents the output vector of the L-th layer of the shared feature extraction layer;

[0205] W 价格 represents the weight matrix of the specific output layer of the price fluctuation amplitude prediction task;

[0206] b 价格 represents the bias vector of the specific output layer of the price fluctuation amplitude prediction task;

[0207] In the specific output layer of the price fluctuation amplitude prediction task, a linear activation function is adopted without introducing any non-linear transformation. Therefore, the original score of the specific output layer of the price fluctuation amplitude prediction task is the predicted price fluctuation amplitude. So corresponds directly to z 价格 without additional processing, which conforms to the standard modeling method of continuous variable regression prediction;

[0208] Specific output layer of the material quality risk prediction task: Based on the material quality feedback records and supplier qualification characteristics, it outputs the occurrence probability of material quality problems. Then, the output formula of the specific output layer of the material quality risk prediction task is as shown in formula (8):

[0209] (8);

[0210] where, P 质量风险Represents the risk probability of problems occurring in the quality of materials. The output value range is [0, 1]. An output value close to 1 indicates a high risk probability of problems occurring in the quality of materials, and an output value close to 0 indicates a low risk probability of problems occurring in the quality of materials. If P 质量风险 > 0.7, it is recommended to strengthen quality inspection or replace the supplier;

[0211] z 质量风险 Represents the original score of the specific output layer of the material quality risk prediction task;

[0212] Then z 质量风险 The calculation formula of is as shown in formula (9):

[0213] z 质量风险 = W 质量风险 H L + b 质量风险 (9);

[0214] W 质量风险 Represents the weight matrix of the specific output layer of the material quality risk prediction task;

[0215] b 质量风险 Represents the bias vector of the specific output layer of the material quality risk prediction task;

[0216] D. Multi-task learning model training and optimization:

[0217] The multi-task learning model is trained and optimized through loss function optimization, hyperparameter optimization, deep learning optimization algorithms, and model generalization ability optimization;

[0218] The specific process is as follows:

[0219] D1. Loss function optimization: For the specific output layers of multiple material procurement data security risk prediction tasks, based on the loss function, by weighted summing the loss values of each specific output layer of the material procurement data security risk prediction tasks, the multi-task learning model is optimized. Then the formula of the loss function is as shown in formula (10):

[0220] (10);

[0221] Among them, N represents the total number of tasks;

[0222] Represents the loss function of the i-th task;

[0223] Represents the task weight;

[0224] and respectively represent the predicted value and the true value of the i-th task;

[0225] D2. Hyperparameter Optimization: Combining the importance of safety risks in material procurement and data distribution, the hyperparameters of the multi-task learning model are tuned through grid search, including the learning rate, task weights and batch size;

[0226] Learning Rate: Used to control the step size of parameter updates, set to 0.001. Being too large may lead to non-convergence, while being too small results in slow convergence. In this invention, an exponential decay strategy is used in combination with the training process to achieve dynamic adjustment and ensure stable training of the model;

[0227] Batch Size: That is, the number of samples used in each training, set to 64. The batch size affects the training speed and the generalization performance of the model. In this invention, a medium scale is selected to achieve a balance between efficiency and effect;

[0228] D3. Deep Learning Optimization Algorithm: The parameters of the multi-task learning model are updated using a deep learning optimization algorithm for learning dynamic features in material procurement data. The formula of the deep learning optimization algorithm is as shown in formula (11):

[0229] (11);

[0230] Among them, and respectively represent the current and updated parameters of the multi-task learning model;

[0231] represents the learning rate;

[0232] represents the first-order momentum estimate of the gradient, smoothing the change of the gradient;

[0233] represents the second-order momentum estimate of the gradient, adjusting the learning rate;

[0234] represents a small constant to prevent division by zero errors;

[0235] D4. Model Generalization Ability Optimization: The adaptability of the multi-task learning model to the actual material procurement business is optimized through cross-validation and early stopping mechanism. Among them,

[0236] Cross-validation: The material procurement data set is divided into a training set and a validation set to make the performance of the multi-task learning model consistent on different data subsets and improve stability;

[0237] Early stopping mechanism: When the performance of the validation set no longer improves, the training process of the multi-task learning model is automatically stopped to prevent overfitting;

[0238] Data augmentation: Perturb the historical material procurement data, such as simulating market price fluctuations and changes in supplier performance capabilities, to enhance the robustness of the multi-task learning model to abnormal situations;

[0239] E. Identification and dynamic assessment of material procurement data security risks:

[0240] Based on the multi-task learning model, identify and dynamically assess the security risks in material procurement through material procurement data;

[0241] The specific process is as follows:

[0242] E1. Risk level classification:

[0243] Segment the prediction results of each specific task output layer into three risk levels: low risk, medium risk, and high risk;

[0244] E2. Risk level assessment and determination:

[0245] Supplier risk prediction and assessment: The probability value predicted by the specific output layer of the supplier risk prediction task is 0 - 1, and it is divided into: 0 - 0.5; 0.5 - 0.8; 0.8 - 1;

[0246] Among them, 0 - 0.5 indicates low risk;

[0247] 0.5 - 0.8 indicates medium risk;

[0248] 0.8 - 1 indicates high risk;

[0249] Delivery delay risk prediction and assessment: The probability value predicted by the specific output layer of the delivery delay risk prediction task is 0 - 1, and it is divided into: 0 - 0.4; 0.4 - 0.7; 0.7 - 1;

[0250] Among them, 0 - 0.4 indicates low risk;

[0251] 0.4 - 0.7 indicates medium risk;

[0252] 0.7 - 1 indicates high risk;

[0253] Market condition risk prediction and assessment: The classification probability predicted by the specific output layer of the market condition risk prediction task. Among them, the category with the highest probability is the risk level category for assessment and determination;

[0254] Price fluctuation range prediction and assessment: The percentage of the price fluctuation range predicted by the specific output layer of the price fluctuation range prediction task, and it is divided into: less than or equal to 5%; 5 - 10%; greater than or equal to 10%;

[0255] Among them, less than or equal to 5% indicates low risk;

[0256] 5 - 10% indicates medium risk;

[0257] Greater than or equal to 10% indicates high risk;

[0258] Material quality risk prediction and assessment: The probability value predicted by the specific output layer of the material quality risk prediction task is between 0 and 1, and it is divided into: 0 - 0.5; 0.5 - 0.7; 0.7 - 1;

[0259] Among them, 0 - 0.5 indicates low risk;

[0260] 0.5 - 0.7 indicates medium risk;

[0261] 0.7 - 1 indicates high risk;

[0262] The overall process is as follows:

[0263] Data input: The system receives the latest purchase orders, supplier information, market conditions, inventory data, etc., and after pre - processing, converts them into feature vectors.

[0264] Model calculation: Use the shared feature extraction layer to extract global patterns and pass the data into the task - specific output layer for risk prediction.

[0265] Risk score calculation: Different types of risks use corresponding activation functions to calculate risk scores or probabilities, and determine the risk level according to the preset threshold.

[0266] Risk level classification: According to business requirements, risks are divided into low risk, medium risk, and high risk;

[0267] As shown in Table 2, the risk score calculation results and corresponding risk levels based on purchase orders PO1001, PO1002, and PO1003:

[0268] Table 2:

[0269]

[0270] F. Generation of early warning information for material procurement data and countermeasures:

[0271] Generate multi - category early warning information for material procurement data based on the results of material procurement data security risk identification and assessment, and provide corresponding countermeasures according to the multi - category early warning information for material procurement data;

[0272] The specific process is as follows:

[0273] F1. Generation of early warning information for material procurement data, specifically including:

[0274] Supplier risk early warning: Generate early warning information including supplier name, default probability, and recommended supplier replacement;

[0275] Delivery delay risk warning: Generate warning information including order number, delivery delay probability, and recommended adjusted delivery cycle;

[0276] Market condition risk warning: Generate a market dynamics analysis report, clarify low, medium, and high risk levels and their corresponding recommendations;

[0277] Price fluctuation warning: Generate price adjustment recommendations and provide decision-making support for locking prices or adjusting purchase quantities;

[0278] Material quality risk warning: Generate a probability prediction of quality problems and provide corresponding supplier audit recommendations or quality control plans;

[0279] As shown in Table 3, generate warning information according to the risk levels in Table 2:

[0280] Table 3:

[0281]

[0282] F2. Provide coping strategies according to the risk warning levels:

[0283] Low risk: It is recommended to maintain the existing material procurement plan;

[0284] Medium risk: It is recommended to review the supplier reputation, inventory status, or quality inspection records, and pay attention to market dynamics;

[0285] High risk: It is recommended to immediately take emergency measures, such as replacing suppliers, adjusting procurement plans, implementing more stringent quality inspection processes, or expanding inventory reserves;

[0286] As shown in Table 4, recommend coping strategies according to the warning information and evaluation results:

[0287] Table 4:

[0288] .

[0289] The material procurement data processing system based on security risk identification and prediction provided by the present invention is applied to a material procurement data processing method based on security risk identification and prediction, including:

[0290] Data acquisition module: Used to collect original material procurement data from multiple material procurement data sources, including purchase quantity, purchase price, supplier reputation, inventory turnover rate, market price fluctuation, and supply and demand relationship;

[0291] Data processing module: It is used to clean, denoise, and normalize the original material procurement data collected, extract the material procurement risk features from the preprocessed original material procurement data, screen out the feature set for predicting the material procurement data security risk, optimize the screened feature set, construct high-order features based on the material procurement business requirements, and generate an optimized feature set;

[0292] Multi-task learning model module: It identifies and predicts multiple material procurement security risks through a shared feature extraction layer and multiple task-specific output layers;

[0293] Risk output module: It is used to conduct risk assessment on real-time material procurement data by using the trained multi-task learning model, output the risk level or risk value, generate early warning information, and provide corresponding countermeasures or measures.

[0294] The present invention utilizes multi-source material procurement data, including purchase orders, supplier information, inventory data, market conditions, and historical purchase records, etc., to construct an intelligent multi-task learning model. Based on the multi-task learning framework of deep learning, the present invention adopts a shared feature extraction layer to capture the global data pattern, and combines task-specific output layers to respectively predict supplier default risk, price fluctuation risk, delivery delay risk, market condition change risk, and material quality risk. By optimizing multi-task learning through an adaptive weighted loss function, the coverage of risk identification and the prediction accuracy are improved, significantly enhancing the accuracy and real-time performance of material procurement risk prediction. The purpose is to solve multiple risks such as supplier default, price fluctuation, delivery delay, inventory shortage, and material quality problems during the procurement process, and provide real-time early warning and business strategy suggestions to improve the intelligent level of material procurement management, enhance the enterprise's prevention and control ability of material procurement risks, and reduce the uncertainty of supply chain operation;

[0295] As Figure 2 shown, compared with the traditional method that relies on manual experience and historical data, the present invention has the following significant advantages:

[0296] (1) Improve the risk prediction accuracy: By optimizing the feature engineering and multi-task learning model, the present invention can accurately identify multiple procurement risks such as supplier default, price fluctuation, and delivery delay, and the risk identification accuracy is increased by 8 - 10%;

[0297] (2) Early warning of potential risks: By analyzing the procurement data in real time, the present invention can provide early warning 3 - 5 days before high-risk events occur, helping enterprises take countermeasures in advance and reducing the risk of supply chain interruption and financial losses;

[0298] (3) Provide intelligent decision-making support: Based on the risk assessment results, the present invention can generate accurate early warning information and provide specific coping strategies according to different risk levels, such as supplier replacement, purchase quantity adjustment, etc., optimizing the procurement decision-making process.

[0299] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for processing material procurement data based on security risk identification and prediction, characterized in that It includes the following steps: A. Material procurement data collection and preprocessing: Collect the original material procurement data from multiple material procurement data sources and preprocess the collected original material procurement data; B. Material procurement data feature extraction and optimization: Extract the material procurement risk features and time series features from the preprocessed original material procurement data, generate the interaction features of inventory turnover rate and market price fluctuation, screen out the feature set for predicting the security risk of material procurement data, optimize the screened feature set, construct high-order features based on the material procurement business requirements, and generate the optimized feature set; C. Multi-task learning model construction: Use a multi-layer perceptron as the shared feature extraction layer, learn the optimized feature set generated in the prediction of the security risk of material procurement data through a multi-layer fully connected network, and construct multiple specific output layers for the prediction tasks of the security risk of material procurement data according to the prediction tasks of the security risk of material procurement data; D. Multi-task learning model training and optimization: Train and optimize the multi-task learning model through loss function optimization, hyperparameter optimization, deep learning optimization algorithms, and model generalization ability optimization; E. Identification and dynamic assessment of the security risk of material procurement data: Based on the multi-task learning model, identify and dynamically assess the security risks in material procurement through material procurement data; F. Generation of early warning information for material procurement data and countermeasures: Generate multi-category early warning information for material procurement data according to the results of the identification and assessment of the security risk of material procurement data, and provide corresponding countermeasures according to the multi-category early warning information for material procurement data.

2. The method for processing material procurement data based on security risk identification and prediction according to claim 1, characterized in that: In step A, the specific process of material procurement data collection and preprocessing is as follows: A1. Collect the original material procurement data from multiple material procurement data sources for subsequent analysis and preprocessing of the original material procurement data. The original material procurement data includes procurement quantity, procurement price, supplier reputation, inventory turnover rate, market price fluctuation, and supply-demand relationship; A2. Clean the collected original material procurement data, remove outliers and duplicate values, fill in the missing values using the interpolation method, then perform unified standardization processing on the original material procurement data with different dimensions, and optimize the quality of the original material procurement data through denoising technology to eliminate potential errors.

3. The method for processing material procurement data based on security risk identification and prediction according to claim 2, wherein: In step B, the specific process of material procurement data feature extraction and optimization is as follows: B1. Extract the material procurement risk features and time series features from the preprocessed original material procurement data, generate the interaction features of inventory turnover rate and market price fluctuation to reflect the comprehensive impact of inventory pressure and market dynamics, capture the long-term trend and periodic fluctuations of the original material procurement data according to the time series features, analyze and process the material procurement risk features, time series features, and interaction features, and screen out the feature set for predicting the security risk of material procurement data; B2. Optimize the screened feature set, construct high-order features based on the material procurement business requirements, and generate the optimized feature set to provide input for the multi-task learning model; The high-order features include the non-linear combined features of inventory turnover rate and supplier reputation.

4. A method for processing material procurement data based on safety risk identification and prediction according to claim 3, characterized in that: In step C, the specific process of constructing the multi-task learning model is as follows: C1. Construction of the shared feature extraction layer: A multi-layer perceptron is used as the shared feature extraction layer to learn the optimized feature set generated in the prediction of the safety risk of material procurement data through a multi-layer fully-connected network. The shared features include purchase orders, supplier information, inventory data, market conditions, and historical purchase records; The output of the shared feature extraction layer is non-linearly mapped through the ReLU activation function. The output formula of the shared feature extraction layer is shown in formula (1): H l = ReLU(W l H l-1 + b l ) (1); Among them, H l represents the output vector of the l-th layer of the shared feature extraction layer; W l denotes the weight matrix of the l-th layer of the shared feature extraction layer; b l Denotes the bias term of the l-th layer of the shared feature extraction layer; ReLU() represents the activation function, and its output is defined as: ReLU(x) = max(0, x); C2. Construction of the task-specific output layer: According to the supplier risk prediction task, delivery delay risk prediction task, market condition risk prediction task, price fluctuation range prediction task, and material quality risk prediction task, a supplier risk prediction task-specific output layer, a delivery delay risk prediction task-specific output layer, a market condition risk prediction task-specific output layer, a price fluctuation range prediction task-specific output layer, and a material quality risk prediction task-specific output layer are respectively constructed.

5. A method for processing material procurement data based on safety risk identification and prediction according to claim 4, characterized in that: In step C2, Supplier risk prediction task-specific output layer: Based on the historical performance records and supply stability characteristics of the supplier, the risk probability of supplier default is output. The output formula of the supplier risk prediction task-specific output layer is shown in formula (2): （2）； Among them, z 违约 represents the raw score of the specific output layer of the supplier risk prediction task; P 违约 represents the risk probability of supplier default. The output value range is [0, 1]. An output value close to 1 indicates a high risk probability of supplier default, and an output value close to 0 indicates a low risk probability of supplier default; Then z 违约 is calculated by the formula shown in Formula (3) as follows: z 违约 =W 违约 H L +b 违约 (3); Among them, W 违约 represents the weight matrix of the specific output layer of the supplier risk prediction task; b 违约 Represents the bias vector of the specific output layer for the supplier risk prediction task; H L represents the output vector of the L-th layer of the shared feature extraction layer; Delivery delay risk prediction task-specific output layer: Combining the delivery cycle and supplier delivery on-time rate characteristics, the risk probability of delivery delay is output. The output formula of the delivery delay risk prediction task-specific output layer is shown in formula (4): （4）； Among them, z 延误 represents the original score of the specific output layer of the delivery delay risk prediction task; P 延误 Indicates the risk probability of delivery delay. The output value ranges from [0, 1]. An output value close to 1 indicates a high risk probability of delivery delay, and an output value close to 0 indicates a low risk probability of delivery delay; Then z 延误 is calculated by the formula shown in formula (5): z 延误 =W 延误 H L +b 延误 (5); Among them, W 延误 represents the weight matrix of the specific output layer for the delivery delay risk prediction task; b 延误 Denotes the bias vector of the specific output layer for the delivery delay risk prediction task; Market condition risk prediction task-specific output layer: According to the supply-demand relationship and market trend characteristics, the market condition risk level is output. The output formula of the market condition risk prediction task-specific output layer is shown in formula (6): （6）； where, i represents the current risk level category; j represents all risk level categories; P 市场风险 (i) represents the predicted probability of the i-th category; Softmax represents the function that converts the scores into the probability distribution of the market condition risk. The output value is the probability distribution of multiple risk levels, and the sum is 1; z i Represents the original score of the current category i, that is, the linear combination result corresponding to the i-th category of the specific output layer of the market condition risk prediction task; Price fluctuation range prediction task-specific output layer: Based on the historical market price fluctuations and real-time supply-demand relationship characteristics, a linear activation function is used to predict continuous variables, and the predicted value of the price fluctuation range is output. The output formula of the price fluctuation range prediction task-specific output layer is shown in formula (7): （7）； Among them, represents the price fluctuation range predicted by the specific output layer of the price fluctuation range prediction task, and the output prediction value is a continuous numerical value; H L represents the output vector of the L-th shared feature extraction layer; W 价格 Represents the weight matrix of the specific output layer for the price fluctuation amplitude prediction task; b 价格 Represents the bias vector of the specific output layer for the price fluctuation amplitude prediction task; Material quality risk prediction task-specific output layer: Based on the material quality feedback records and supplier qualification characteristics, the occurrence probability of material quality problems is output. The output formula of the material quality risk prediction task-specific output layer is shown in formula (8): （8）； Among them, P 质量风险 represents the risk probability of material quality problems. The output value range is [0, 1]. An output value close to 1 indicates a high risk probability of material quality problems, and an output value close to 0 indicates a low risk probability of material quality problems; z 质量风险 Represents the raw score of the specific output layer for the material quality risk prediction task; Then z 质量风险 is calculated by the formula shown in Formula (9) as follows: z 质量风险 = W 质量风险 H L +b 质量风险 (9); W 质量风险 represents the weight matrix of the specific output layer of the material quality risk prediction task; b 质量风险 Represents the bias vector of the specific output layer of the material quality risk prediction task.

6. The method for processing material procurement data based on security risk identification and prediction according to claim 5, wherein: In step D, training and optimization of the multi-task learning model: The specific process of training and optimizing the multi-task learning model through loss function optimization, hyperparameter optimization, deep learning optimization algorithms, and model generalization ability optimization is as follows: D1. Loss function optimization: For the specific output layer of multiple material procurement data security risk prediction tasks, based on the loss function, by weighted summing the loss values of each specific output layer of the material procurement data security risk prediction tasks, the multi-task learning model is optimized. The formula of the loss function is shown in Formula (10): （10）； Where N represents the total number of tasks; Denote the loss function of the i-th task; Indicates the task weight; and represent the predicted value and the true value of the i-th task, respectively; D2. Hyperparameter Optimization: Combining the importance of safety risks in material procurement and data distribution, the hyperparameters of the multi-task learning model are tuned through grid search, including the learning rate, task weights and batch size; D3. Deep learning optimization algorithm: The parameters of the multi-task learning model are updated using the deep learning optimization algorithm for learning the dynamic features in the material procurement data. The formula of the deep learning optimization algorithm is shown in Formula (11): （11）； Among them, and respectively represent the current and updated multi-task learning model parameters; represents the learning rate; Represents the first-order momentum estimate of the gradient, smoothing the change in the gradient; Represents the second-order momentum estimate of the gradient and adjusts the learning rate; Represents a small constant to prevent division-by-zero errors; D4. Optimization of model generalization ability: The adaptability of the multi-task learning model to the actual material procurement business is optimized through cross-validation and early stopping mechanism. Among them, Cross-validation: The material procurement data set is divided into a training set and a validation set to make the performance of the multi-task learning model consistent on different data subsets; Early stopping mechanism: When the performance of the validation set no longer improves, the training process of the multi-task learning model is automatically stopped.

7. A method for processing material procurement data based on safety risk identification and prediction according to claim 6, characterized in that: In step E, the identification and dynamic assessment of material procurement data security risks: Based on the multi-task learning model, the specific process of identifying and dynamically assessing the security risks in material procurement through material procurement data is as follows: E1. Risk level classification: The prediction results of each specific task output layer are segmented into three risk levels: low risk, medium risk, and high risk; E2. Risk level assessment and determination: Supplier risk prediction assessment: The probability value predicted by the specific output layer of the supplier risk prediction task is 0 - 1, which is divided into: 0 - 0.5; 0.5 - 0.8; 0.8-1； Among them, 0 - 0.5 represents low risk; 0.5 - 0.8 represents medium risk; 0.8 - 1 represents high risk; Delivery delay risk prediction assessment: The probability value predicted by the specific output layer of the delivery delay risk prediction task is 0 - 1, which is divided into: 0 - 0.4; 0.4 - 0.7; 0.7 - 1; Among them, 0 - 0.4 represents low risk; 0.4 - 0.7 represents medium risk; 0.7 - 1 represents high risk; Market condition risk prediction assessment: The classification probability predicted by the specific output layer of the market condition risk prediction task, among which the category with the highest probability is the risk level category of the assessment and determination; Price fluctuation range prediction assessment: The percentage of the price fluctuation range predicted by the specific output layer of the price fluctuation range prediction task, which is divided into: less than or equal to 5%; 5 - 10%; greater than or equal to 10%; Among them, less than or equal to 5% represents low risk; 5 - 10% represents medium risk; Greater than or equal to 10% represents high risk; Material quality risk prediction assessment: The probability value predicted by the specific output layer of the material quality risk prediction task is 0 - 1, which is divided into: 0 - 0.5; 0.5 - 0.7; 0.7 - 1; Among them, 0 - 0.5 represents low risk; 0.5 - 0.7 represents medium risk; 0.7 - 1 represents high risk.

8. A method for processing material procurement data based on security risk identification and prediction according to claim 7, characterized in that: In step F, the generation of material procurement data warning information and countermeasures: According to the results of the identification and assessment of material procurement data security risks, multi-category material procurement data warning information is generated, and the corresponding countermeasures are provided according to the multi-category material procurement data warning information. The specific process is as follows: F1. Generation of early warning information for material procurement data, specifically including: Supplier risk early warning: Generate early warning information including supplier name, default probability, and recommended supplier replacement; Delivery delay risk early warning: Generate early warning information including order number, delivery delay probability, and recommended delivery cycle adjustment; Market condition risk early warning: Generate a market dynamic analysis report, clarify low, medium, and high risk levels and their corresponding recommendations; Price fluctuation early warning: Generate price adjustment recommendations and provide decision-making support for locking prices or adjusting procurement quantities; Material quality risk early warning: Generate a probability prediction of quality problems and provide corresponding supplier audit recommendations or quality control plans; F2. Provide coping strategies according to the risk early warning level: Low risk: It is recommended to maintain the existing material procurement plan; Medium risk: It is recommended to review the supplier reputation, inventory status, or quality inspection records and pay attention to market dynamics; High risk: It is recommended to immediately take emergency measures, such as replacing suppliers, adjusting procurement plans, implementing stricter quality inspection processes, or expanding inventory reserves.

9. A material procurement data processing system based on security risk identification and prediction, which is applied to a material procurement data processing method based on security risk identification and prediction as described in claim 8, and is characterized in that, Including: Data acquisition module: Used to collect original material procurement data from multiple material procurement data sources, including procurement quantity, procurement price, supplier reputation, inventory turnover rate, market price fluctuations, and supply and demand relationships; Data processing module: Used to clean, denoise, and normalize the collected original material procurement data, extract material procurement risk characteristics from the preprocessed original material procurement data, screen out the feature set for predicting material procurement data security risks, optimize the screened feature set, and construct high-order features based on material procurement business requirements to generate an optimized feature set; Multi-task learning model module: Identify and predict various material procurement security risks through a shared feature extraction layer and multiple task-specific output layers; Risk output module: Used to evaluate the risks of real-time material procurement data using the trained multi-task learning model, output risk levels or risk values, generate early warning information, and provide corresponding coping strategies or measures.

Images