Purchase efficiency optimization management method and system based on artificial intelligence

Abstract

The invention relates to the technical field of supply chain management, and discloses a purchase efficiency optimization management method and system based on artificial intelligence, and the system comprises a data fusion module, an AI decision center, and an execution monitoring module. Integrating market dynamics, supplier qualification and historical transaction features in real time through a multi-source heterogeneous data fusion module to construct a structured purchase feature matrix; generating a dynamic purchase portrait model based on a deep reinforcement learning algorithm to accurately predict a supply and demand fluctuation trend; a multi-algorithm collaborative decision engine is innovatively designed to dynamically select an optimal purchasing strategy in combination with a meta-learning controller; an isolation forest algorithm is adopted to realize abnormal order dual-threshold monitoring and intelligent fusing; a purchase performance process is automatically triggered through an intelligent contract execution engine; and finally, realizing full-process tampering-free evidence storage based on a block chain double-chain architecture. According to the method, the three problems of procurement decision lag, supplier risk response lagging and cross-regional coordination obstacle are solved, and the enterprise supply chain intelligent decision normal form is comprehensively reconstructed.

Description

Technical Field

[0001] This invention relates to the field of supply chain management technology, specifically to a procurement efficiency optimization management method and system based on artificial intelligence. Background Technology

[0002] Supply chain management technology refers to the systematic technologies and tools used in coordinating the flow of products from raw material procurement to the end consumer. It aims to optimize processes, improve efficiency, and enhance supply chain resilience and responsiveness. The AI-based procurement efficiency optimization management method and system is an intelligent decision-making system that integrates deep learning and blockchain technologies. This system's core engine is responsible for integrating multi-source heterogeneous procurement data to generate dynamic decision-making strategies. The procurement profile model built using deep learning algorithms can accurately predict supply and demand fluctuations, ensuring that procurement strategies are synchronized with market dynamics in real time. The blockchain evidence storage module guarantees the immutability of the entire process for audit and traceability, providing technical support for procurement compliance.

[0003] Currently, due to the complex interactions involving multiple stakeholders and stages in enterprise procurement activities, the static historical data relied upon for automated procurement decisions cannot respond in real time to sudden market price fluctuations. If raw material prices rise abnormally and suppliers suddenly cut off supplies, procurement strategies may fail, leading to a surge in procurement costs and supply chain disruptions. At the same time, when conducting supplier risk assessments, existing systems cannot dynamically capture changes in supplier qualifications and may overlook signals of new administrative penalties or capacity reductions for suppliers, causing high-risk suppliers to enter the procurement process. Furthermore, there is a lack of emergency circuit breaker mechanisms after cooperation risks are exposed. In cross-regional collaborative procurement, the data isolation of subsidiaries creates information silos, making it impossible to achieve privacy-preserving collaborative computing under the federated learning framework. This leads to a decrease in the accuracy of the global procurement optimization model, further affecting the cost control effect of bulk material centralized procurement.

[0004] Therefore, this paper proposes an artificial intelligence-based procurement efficiency optimization management method and system to address the aforementioned technical bottlenecks. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a procurement efficiency optimization management method and system based on artificial intelligence, which solves the problems mentioned in the background.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a procurement efficiency optimization management method and system based on artificial intelligence, wherein the method includes the following steps: S1. Collect multi-source heterogeneous procurement data, including historical order data of procurement demanders, supplier qualification information, market dynamic data and contract texts; S2. Perform multimodal feature extraction processing on the procurement data to generate a structured procurement feature matrix and a semantic association graph; S3. Based on deep reinforcement learning, construct a dynamic procurement profile to generate a demand-side behavior prediction model and a supplier risk assessment model. S4. Based on the procurement feature matrix and semantic association graph, the optimal procurement strategy is matched through a multi-algorithm collaborative decision engine to generate an initial procurement decision scheme; S5. Based on real-time market feedback data, dynamically optimize and iterate the initial procurement decision scheme to generate an adaptive procurement strategy set; S6. Employ an anomaly detection algorithm to monitor the procurement process in real time, and trigger strategy replanning when compliance deviations from cost exceed the threshold. S7. Input the optimized procurement strategy set into the smart contract execution engine to automatically generate purchase orders, supplier allocation plans and performance tracking instructions; S8. Record the data of the entire process decision nodes through the blockchain evidence storage module to generate an immutable procurement audit traceability chain.

[0007] Preferably, S1 includes: S11. Synchronize historical purchase order data from the ERP system via API interface, and extract characteristics such as order amount, delivery cycle, and number of defaults. S12. Use web crawlers to collect supplier business information, patent databases and public opinion data to build a full-dimensional supplier information pool. S13. Use OCR recognition technology to analyze the scanned contract documents, extract key fields such as payment terms and liability for breach of contract, and store them in a quantitative format.

[0008] Preferably, S2 includes: S21. Use the BiLSTM-CRF model to extract entity relationships from text data and generate semantic triples of supplier-product-qualification. S22. Apply wavelet transform to decompose the trend term and periodic term of time-series order data to generate a procurement demand fluctuation coefficient matrix. S23. Construct a multi-entity association graph using a graph convolutional network, where the node weight calculation formula is:

[0009] The log function here is used to prevent excessively high weights when the nodes are too large, maintaining numerical stability. All variables are dimensionless, and all terms in the formula are dimensionless. This represents the overall weight value of node v. These are the text feature weight coefficients. Let be the associated edge of node v. For nodes The set of all directly connected edges. To calculate associated edges Textual keyness, For node degree = AND The number of directly connected nodes. For risk weighting coefficients, External risk rating.

[0010] Preferably, S3 includes: S31. Construct a dual-channel neural network: the demand channel inputs historical procurement frequency and budget deviation rate characteristics, and the supply channel inputs capacity utilization rate and quality inspection pass rate characteristics. S32. The PPO reinforcement learning algorithm is used to dynamically update the profile weights, and the reward function is defined as follows:

[0011] in, For the instant reward of time step t, Weighting cost savings, To achieve cost savings, As a risk penalty weight, For comprehensive risk assessment, For compliance reward weighting, To achieve compliance and standards.

[0012] Preferably, S4 includes: S41. Establish a multi-algorithm strategy library, including combinatorial auction algorithm, genetic optimization algorithm, and collaborative filtering recommendation algorithm; S42. By selecting algorithm combinations through the meta-learning controller, the controller output satisfies: Cost variance normalization:

[0013] Time term normalization:

[0014] The formula becomes:

[0015] in, For the parameters of the meta controller, The actual procurement cost, To predict procurement costs, This is the time penalty coefficient. The time required to generate a decision.

[0016] Preferably, S5 includes: S51. Constructing a market volatility response model:

[0017] in, This represents the change in market conditions. ∈[−1,1], negative values ​​indicate a decrease in demand, positive values ​​indicate an increase in demand, and absolute values ​​indicate intensity. For Long Short-Term Memory (LSTM) networks, For price volatility, For the gradient of demand changes, The competition coefficient; S52, when When the value is greater than δ, the strategy re-optimization module is activated to recalculate the procurement batches and supplier ratios.

[0018] Preferably, S6 includes: S61. Set up a dual-threshold monitoring mechanism: Cost threshold:

[0019] in, The threshold for cost anomalies. The historical average cost Cost standard deviation; Compliance threshold:

[0020] in, As a compliance risk threshold, This represents the maximum compliance risk value. S62. Select a standardization method for the input features according to their type. Numerical features are standardized using Z-score, and proportional features are normalized using Min-Max. All features are ultimately standardized to the [0,1] interval. The isolated forest algorithm is used to detect abnormal orders. The abnormality judgment condition is:

[0021] Feature anomaly scores are converted to [0,1] using the Sigmoid function, where, For the comprehensive score of abnormality, The number of feature dimensions, For the index of the feature dimension, This is the feature deviation function, with an output range of [0,1]. Let i be the specific value of the i-th feature. To determine the anomaly threshold, the input features are standardized using Z-score. ,limited The range is [0,1].

[0022] Preferably, S7 includes: S71, Smart Contract Engine executes automatically: The bank's payment interface is invoked when the payment conditions are triggered. The penalty calculation module will be activated when delivery is delayed. When quality fails to meet standards, alternative supplier sourcing is activated. S72. Generate a 3D performance dashboard: including a cost execution heatmap, supplier performance radar chart, and risk warning matrix.

[0023] Preferably, the method includes the following steps: S111. Deploy local training modules on multiple procurement nodes, and train an initial supplier evaluation model on each node based on private procurement data. S112. Homomorphic encryption technology is used to encrypt the gradient parameters of the local model to generate encrypted gradient vectors; S113. Transmit the encrypted gradient vector to the central decision-making node through a secure aggregation channel; S114. Execute the gradient-weighted fusion algorithm at the central decision node to generate global model update parameters; S115. Synchronize the global update parameters to each procurement node to update the local model; S116. Generate a cross-regional joint procurement strategy based on the updated local model.

[0024] Preferably, the method includes the following steps: S121. Generate data fingerprints at key nodes in the procurement decision-making process, including demand submission, supplier selection, contract signing, and performance and delivery. S122. Pack the data fingerprints into data blocks according to the time sequence; S123. Adopt a dual-chain storage architecture: store the original data in a private chain and store the hash value of key operations in a public chain; S124. Verify the consistency of data between the private blockchain and the public blockchain using zero-knowledge proof technology; S125. When an audit request is triggered, extract the hash value index from the public chain and decrypt the original data from the private chain. S126. Generate a visual audit trajectory map, marking abnormal decision nodes and related responsible persons.

[0025] Compared with existing technologies, this invention provides a procurement efficiency optimization management method and system based on artificial intelligence, which has the following beneficial effects: 1. In this invention, by setting up a multi-source data fusion module, when executing automated procurement decisions, a structured feature matrix is ​​constructed by integrating real-time market dynamics and historical procurement characteristics to ensure the spatiotemporal consistency of multi-dimensional procurement data. At the same time, market fluctuation signals are synchronously input into the dynamic optimizer, which can perceive the risk of supply disruption due to abnormal raw material prices in real time, ensure that the procurement strategy responds to emergencies in a timely manner, avoid the surge in procurement costs and supply chain disruptions caused by data lag, and improve the anti-interference capability of the procurement plan.

[0026] 2. In this invention, by setting up a risk dynamic profiling system, when evaluating suppliers, a real-time risk assessment model is generated by capturing signals of changes in business registration information and production capacity fluctuations. This model automatically identifies suppliers with new administrative penalties or deterioration in their qualifications, enabling the system to preemptively block high-risk suppliers from entering the market. Furthermore, when the risk exceeds a threshold, the system automatically triggers a circuit breaker mechanism to terminate cooperation, reducing the performance risk caused by supplier negligence and ensuring compliance and controllability throughout the procurement process.

[0027] 3. In this invention, by setting up a federated collaborative decision-making engine, when making cross-regional group purchases, a distributed model training architecture is established to ensure that privacy data does not leave the domain, and supports the sharing of procurement behavior pattern characteristics among subsidiaries. This enables the system to both integrate global procurement rules to optimize centralized procurement costs and eliminate the risk of sensitive data leakage, thus solving the problem of reduced optimization accuracy caused by information silos and improving the cost control effect of bulk material procurement. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the architecture of the procurement efficiency optimization management method and system based on artificial intelligence of the present invention; Figure 2 The flowchart shows the procurement efficiency optimization management method and system based on artificial intelligence of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Please see Figure 1-2 The AI-based procurement efficiency optimization management method and system includes the following steps: S1, collecting multi-source heterogeneous procurement data, including historical order data of procurement demanders, supplier qualification information, market dynamic data and contract texts; S2. Perform multimodal feature extraction processing on the procurement data to generate a structured procurement feature matrix and semantic association graph; S3. Based on federated reinforcement learning, a dynamic procurement profile is constructed. Each node trains the basic model locally, and the central node optimizes the strategy through deep reinforcement learning. That is, the central node integrates gradients through a secure aggregation mechanism, updates the global model, and generates a demand-side behavior prediction model and a supplier risk assessment model. S4. Based on the procurement feature matrix and semantic association graph, the optimal procurement strategy is matched through a multi-algorithm collaborative decision engine to generate an initial procurement decision scheme. S5. Dynamically optimize and iterate the initial procurement decision plan based on real-time market feedback data to generate an adaptive procurement strategy set; S6. Employ an anomaly detection algorithm to monitor the procurement process in real time, and trigger strategy replanning when compliance deviations from cost exceed the threshold. S7. Input the optimized procurement strategy set into the smart contract execution engine to automatically generate purchase orders, supplier allocation plans and performance tracking instructions; S8. Record the data of the entire process decision nodes through the blockchain evidence storage module to generate an immutable procurement audit traceability chain; S11. Synchronize historical purchase order data from the ERP system via API interface, and extract characteristics such as order amount, delivery cycle, and number of defaults. S12. Use web crawlers to collect supplier business information, patent databases and public opinion data to build a full-dimensional supplier information pool. S13. Use OCR recognition technology to analyze the scanned contract documents, extract key fields such as payment terms and liability for breach of contract, and store them in a quantitative format. S21. Use the BiLSTM-CRF model to extract entity relationships from text data and generate semantic triples of supplier-product-qualification. S22. Apply wavelet transform to decompose the trend term and periodic term of time-series order data to generate a procurement demand fluctuation coefficient matrix. S23. Construct a multi-entity association graph using a graph convolutional network, where the node weight calculation formula is:

[0031] The log function here is used to prevent excessively high weights when the nodes are too large, maintaining numerical stability. All variables are dimensionless, and all terms in the formula are dimensionless. This represents the overall weight value of node v. These are the text feature weight coefficients. Let be the associated edge of node v. For nodes The set of all directly connected edges. To calculate associated edges Textual keyness, For node degree = AND The number of directly connected nodes. For risk weighting coefficients, External risk rating; S31. Construct a dual-channel neural network: the demand channel inputs historical procurement frequency and budget deviation rate characteristics, and the supply channel inputs capacity utilization rate and quality inspection pass rate characteristics. S32. The PPO reinforcement learning algorithm is used to dynamically update the profile weights, and the reward function is defined as follows:

[0032] in, For the instant reward of time step t, Weighting cost savings, To achieve cost savings, As a risk penalty weight, For comprehensive risk assessment, For compliance reward weighting, To achieve compliance standards; S41. Establish a multi-algorithm strategy library, including combinatorial auction algorithm, genetic optimization algorithm, and collaborative filtering recommendation algorithm; S42. By selecting algorithm combinations through the meta-learning controller, the controller output satisfies: Cost variance normalization:

[0033] Time term normalization:

[0034] The formula becomes:

[0035] in, For the parameters of the meta controller, The actual procurement cost, To predict procurement costs, This is the time penalty coefficient. The time required to generate a decision; S51. Constructing a market volatility response model:

[0036] in, This represents the change in market conditions. ∈[−1,1], negative values ​​indicate a decrease in demand, positive values ​​indicate an increase in demand, and absolute values ​​indicate intensity. For Long Short-Term Memory (LSTM) networks, For price volatility, For the gradient of demand changes, The competition coefficient; S52, when When the value is greater than δ, the strategy re-optimization module is activated to recalculate the procurement batches and supplier ratios. S61. Set up a dual-threshold monitoring mechanism: Cost threshold:

[0037] in, The threshold for cost anomalies. The historical average cost Cost standard deviation; Compliance threshold:

[0038] in, As a compliance risk threshold, This represents the maximum compliance risk value. S62. Select a standardization method for the input features according to their type. Numerical features are standardized using Z-score, and proportional features are normalized using Min-Max. All features are ultimately standardized to the [0,1] interval. The isolated forest algorithm is used to detect abnormal orders. The abnormality judgment condition is:

[0039] Feature anomaly scores are converted to [0,1] using the Sigmoid function, where, For the comprehensive score of abnormality, The number of feature dimensions, For the index of the feature dimension, This is the feature deviation function, with an output range of [0,1]. Let i be the specific value of the i-th feature. To determine the anomaly threshold, the input features are standardized using Z-score. ,limited The range is [0,1]; S71, Smart Contract Engine executes automatically: The bank's payment interface is invoked when the payment conditions are triggered. The penalty calculation module will be activated when delivery is delayed. When quality fails to meet standards, alternative supplier sourcing is activated. S72. Generate a 3D performance dashboard: including a cost execution heatmap, supplier performance radar chart, and risk warning matrix; S111. Deploy local training modules on multiple procurement nodes, and train an initial supplier evaluation model on each node based on private procurement data. S112. Homomorphic encryption technology is used to encrypt the gradient parameters of the local model to generate encrypted gradient vectors; S113. Transmit the encrypted gradient vector to the central decision-making node through a secure aggregation channel; S114. Execute the gradient-weighted fusion algorithm at the central decision node to generate global model update parameters; S115. Synchronize the global update parameters to each procurement node to update the local model; S116. Generate a cross-regional joint procurement strategy based on the updated local model; S121. Generate data fingerprints at key nodes in the procurement decision-making process, including demand submission, supplier selection, contract signing, and performance and delivery. S122. Pack the data fingerprints into data blocks according to the time sequence; S123. Adopt a dual-chain storage architecture: store the original data in a private chain and store the hash value of key operations in a public chain; S124. Verify the consistency of data between the private blockchain and the public blockchain using zero-knowledge proof technology; S125. When an audit request is triggered, extract the hash value index from the public chain and decrypt the original data from the private chain. S126. Generate a visual audit trajectory map, marking abnormal decision nodes and related responsible persons. Specific Implementation

[0040] Example 1: Application of Dynamic Procurement Hub in Multi-Source Data Scenarios During the implementation of this system, a multinational manufacturing company's South China branch suddenly received an urgent order to purchase 20 tons of special alloy. The traditional procurement process would require manual price comparison for more than 3 days. At this time, the system's data fusion module collected multiple key data in real time: it obtained the material's procurement records for the past six months through the ERP interface, showing that the highest procurement price was 52,000 yuan / ton; it called the commodity exchange API to capture the current nickel price rising by 18,485 yuan / ton in a single day, saving $70,000 compared to the budget; the blockchain evidence storage module fully recorded the decision-making path during the price fluctuation period; and the audit showed that the smart contract automatically activated the penalty clause when the supplier failed to deliver the first batch on time, protecting the company from the risk of supply chain disruption.

[0041] Example 2: Practical Verification of Supplier Risk Circuit Breaker Mechanism During a centralized procurement of medical polymer materials by a medical equipment group, the system's supplier risk assessment model detected a significant update: two new administrative penalty records were added to the business registration database of a major supplier in the early morning, involving environmental violations with penalties exceeding 2 million yuan. Simultaneously, a red typhoon warning was issued for the area where the supplier's factory was located. The supplier's risk rating (Rv) in the dynamic procurement profile weight matrix jumped from 0.35 to 0.82. At 10:00 AM that day, when the procurement specialist submitted an order for 5 tons of materials from this supplier, the anomaly detection module calculated the risk index (AnomalyScore) in real time to be 0.78, exceeding the threshold (τ) of 0.7. The system immediately triggered a three-level response mechanism: first, it froze the order signing process and sent a risk warning report to the procurement committee; second, it activated the alternative supplier sourcing program, matching two alternative manufacturers with the same specifications of products within one hour using a semantic association graph; and finally, the smart contract engine automatically generated a split procurement plan, splitting the original order into two parts: 3 tons and 2 tons. The results showed that the typhoon caused the original supplier's factory to shut down for two weeks, while the new plan ensured the timely delivery of raw materials, avoiding losses due to production line shutdowns.

[0042] Example 3: Cost Optimization of Federated Collaboration in Group Procurement When a retail group implemented cross-regional joint procurement, the system deployed a federated learning architecture connecting five regional centers. The North China center needed to procure cold chain transportation equipment. In the traditional model, it could only rely on local data to form limited solutions. In this system, after each branch trained a supplier evaluation model locally, they only shared gradient parameters to the central decision pool. The federated collaborative decision engine identified a key pattern: the equipment maintenance cost data of the South China center showed that the annual maintenance cost of a certain brand was as low as 8% of the purchase price, but this information did not appear in the North China supplier evaluation database. After the meta-learning controller dynamically integrated this parameter, it recalculated the full life cycle cost model and output a new decision solution: the weight of the original preferred supplier was reduced, and the three-year maintenance cost was included in the price comparison system. In the end, the North China center chose a supplier with a price 5% higher but excellent maintenance record, and the actual cost over two years was reduced. The entire decision-making process strictly protected data privacy. The original procurement data of each center was kept on the local private chain, and the public chain only stored the hash value of key operations for audit verification, achieving global procurement optimization, cost reduction and zero risk of data leakage.

[0043] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI-based procurement efficiency optimization management system, characterized by: Comprise: Data fusion module: integrate multi-source procurement data and generate spatio-temporal aligned feature warehouse; AI decision hub: contains dynamic portrait construction unit, strategy matching engine, real-time optimizer; Execution monitoring module: deploy smart contract executor, blockchain storage node, multi-terminal visual dashboard; Wherein the AI decision hub adopts micro-service architecture, each component communicates through gRPC protocol, and the decision delay is less than 200ms.

2. The artificial intelligence-based procurement efficiency optimization management system according to claim 1, characterized in that: The blockchain storage module adopts double-chain structure: Private chain stores original procurement data, using Paillier homomorphic encryption; Public chain stores key operation hash value, and realizes verifiable privacy through zk-SNARK; The system supports federated learning mode, and each procurement demand side trains portrait model locally and shares gradient parameters.

3. An artificial intelligence-based procurement efficiency optimization management method, characterized by, The method comprises the following steps: S1, collecting multi-source heterogeneous procurement data, including historical order data of procurement demand side, qualification information of supplier, market dynamic data and contract text; S2, multi-modal feature extraction processing is performed on the procurement data to generate structured procurement feature matrix and semantic association graph; S3, based on federated reinforcement learning, a dynamic procurement portrait is constructed, each node locally trains a basic model, the central node optimizes the strategy through deep reinforcement learning, and demand side behavior prediction model and supplier risk assessment model are generated; S4, according to the procurement feature matrix and semantic association graph, the optimal procurement strategy is matched through multi-algorithm collaborative decision engine to generate the initial procurement decision scheme; S5, based on real-time market feedback data, the initial procurement decision scheme is dynamically optimized and iterated to generate adaptive procurement strategy set; S6, abnormal detection algorithm is used to monitor the procurement process in real time, and strategy re-planning is triggered when compliance deviation cost exceeds threshold value; S7, the optimized procurement strategy set is input into the intelligent contract execution engine to automatically generate procurement order, supplier allocation scheme and performance tracking instruction; S8, record the data of decision nodes in the whole process through the blockchain storage module to generate an unalterable procurement audit traceability chain. 4.The AI-based procurement efficiency optimization management method of claim 3, wherein: The S1 comprises: S11, synchronizing the historical procurement order data of ERP system through API interface, extracting order amount, delivery period and default times features; S12, using web crawler to capture supplier business information, patent database and public opinion data, and constructing supplier full-dimensional information pool; S13, analyzing contract scans by OCR recognition technology, extracting payment terms and default liability key fields and vectorizing storage. 5.The AI-based procurement efficiency optimization management method of claim 3, characterized by: The S2 comprises: S21, using BiLSTM-CRF model for entity relation extraction on text data to generate semantic triple of supplier-product-qualification; S22, applying wavelet transform to decompose trend item and periodic item of time series order data to generate procurement demand fluctuation coefficient matrix; S23, constructing multi-entity association graph through graph convolution network, wherein the node weight calculation formula is: Here the log function is used to prevent the weight too high when the node degree is too large, to maintain numerical stability, all variables are dimensionless, and each term in the formula is dimensionless, wherein, is the comprehensive weight value of node v, is the text feature weight coefficient, is the associated edge of node v, is the set of all edges directly connected to node , is the text key degree of the associated edge , is the node degree = the number of nodes directly connected to , is the risk weight coefficient, is the external risk rating. 6.The AI-based procurement efficiency optimization management method of claim 3, characterized in that: The S3 comprises: S31, constructing double-channel neural network: demand channel input historical procurement frequency and budget deviation rate features, and supply channel input capacity utilization rate and quality inspection pass rate features; S32, the PPO reinforcement learning algorithm is used to dynamically update the image weight, and the reward function is defined as: wherein, is an immediate reward for time step t, is a cost savings weight, is a cost savings rate, is a risk penalty weight, is a risk composite score, is a compliance reward weight, is a compliance threshold. 7.The AI-based procurement efficiency optimization management method of claim 3, characterized by: The S4 includes: S41, establish a multi-algorithm strategy library: including combination auction algorithm, genetic optimization algorithm, collaborative filtering recommendation algorithm; S42, select algorithm combination through meta-learning controller, and the controller output satisfies: Cost difference item normalization: Time item normalization: The formula becomes: wherein, is a meta-controller parameter, is an actual procurement cost, is a predicted procurement cost, is a time penalty coefficient, is a decision generation time consumption. 8.The AI-based procurement efficiency optimization management method of claim 3, characterized by: The S5 includes: S51, construct a market fluctuation response model: wherein, is the market state change amount, ∈[−1,1], a negative value indicates a decrease in demand, a positive value indicates an increase in demand, and an absolute value indicates intensity, is a long short-term memory network, is the price volatility rate, is the demand change gradient, is the competition coefficient; S52, when > delta, the strategy re-optimization module recalculates the procurement lot and supplier mix. 9.The AI-based procurement efficiency optimization management method of claim 3, characterized by: The S6 includes: S61, set a double threshold monitoring mechanism: cost threshold: wherein, is a cost anomaly threshold, is a historical cost mean, is a cost standard deviation; Compliance threshold: wherein, is a compliance risk threshold, is a maximum compliance risk value; S62, select the standardization method according to the type of input features, the numerical features adopt Z-score standardization, the proportional features adopt Min-Max normalization, all features are finally unified to the [0, 1] interval, and the isolation forest algorithm is used to detect abnormal orders, and the abnormal judgment condition is: The feature anomaly score is converted to [0, 1] by a Sigmoid function, where, is the anomaly comprehensive score, is the number of feature dimensions, is the index of the feature dimension, is the feature deviation function, and the output value range is [0, 1], is the specific value of the i-th feature, is the anomaly determination threshold, and the input feature is Z-score standardized: , limited the value range is [0, 1]. 10.The AI-based procurement efficiency optimization management method of claim 3, wherein: The S7 includes: S71, the smart contract engine automatically executes: Call bank payment interface when triggering payment conditions; Start the penalty calculation module when the delivery is delayed; Activate the alternative supplier sourcing when the quality is not up to standard; S72, generate a three-dimensional performance board: including cost execution heat map, supplier performance radar chart, risk early warning matrix.

Images