Artificial Intelligence-Based Procurement Efficiency Optimization Management Method and System

Abstract

The present invention discloses a method and system for optimizing procurement efficiency based on artificial intelligence. The method includes data itemized collection, data preprocessing, supply chain relationship modeling, procurement strategy optimization, and procurement efficiency optimization. The present invention relates to the technical field of power fuel procurement management, specifically referring to a method and system for optimizing procurement efficiency based on artificial intelligence. The present invention adopts a comprehensive intelligent method of cross-domain data fusion combined with supply chain modeling and procurement strategy optimization, realizing real-time dynamic optimization of procurement efficiency optimization and management; adopts a dynamically heterogeneous improved graph neural network method for supply chain relationship modeling, optimizing the quantification calculation of supply chain state update and risk conduction, and enhancing the intelligent dimension of procurement efficiency management; adopts a multi-objective reward improved multi-layer strategy reinforcement learning method for procurement strategy optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power fuel procurement management, and specifically to an artificial intelligence-based procurement efficiency optimization management method and system. Background Art

[0002] The AI-based procurement efficiency optimization management method refers to the use of AI technologies such as machine learning and deep learning to intelligently analyze, model and make decisions on the entire procurement process data in order to achieve a modern management method of reducing procurement costs, improving supply stability and automating processes. Its core function is to automatically identify procurement rules, predict market changes, optimize supplier selection and procurement strategies through intelligent algorithms, thereby significantly improving the accuracy and timeliness of procurement decisions, reducing the impact of human factors, and being able to dynamically adapt to market fluctuations and supply chain risks. In key areas such as the power industry, this method can also effectively balance multiple goals such as economy, reliability and environmental protection, creating significant cost reduction and efficiency improvement value for enterprises. Compared with traditional methods, AI-driven procurement optimization has outstanding advantages such as real-time response, continuous learning and multi-objective collaboration, and has become an important direction for the digital transformation of corporate supply chains.

[0003] However, in the existing procurement efficiency optimization management methods, there is a problem that the existing power industry fuel procurement mostly adopts a static evaluation system, and the static system is inevitably out of touch with the dynamic market, which leads to decision lags and lacks effective application in extreme scenarios. Therefore, the traditional method relies on fixed-cycle evaluation and cannot adapt to the sharp fluctuations in fuel prices in the power industry. At the same time, production data and supply chain data are separated, emergency procurement delays are long, and manual simulation tests are difficult to cover complex risk scenarios. Technical problems; in the existing supply chain relationship modeling methods, there are existing supplier evaluations, which are mainly based on historical transaction data, but the relevant data of the real-time coal consumption rate changes of the units are not reflected in the traditional methods. At the same time, the traditional graph model cannot quantify the impact of risk events on procurement demand, and the physical relationship between coal quality and power generation efficiency is difficult to represent. Technical problems; in the existing procurement strategy optimization methods, there is the traditional reinforcement learning procurement strategy optimization, which fails to make a comprehensive decision analysis for complex procurement processes and various procurement risks, carbon emission indicators and market fluctuations, which leads to the technical problem that the existing methods are difficult to combine the results of supply chain relationship modeling for multi-directional comprehensive procurement decisions. Summary of the invention

[0004] In view of the above situation, to overcome the defects of the prior art, the present invention provides an artificial intelligence-based procurement efficiency optimization management method and system. In the existing procurement efficiency optimization management methods, the existing power industry fuel procurement mostly adopts a static evaluation system, and it is inevitable that the static system is prone to being out of touch with the dynamic market, resulting in lagging decisions. Moreover, there is a lack of effective application in extreme scenarios. Therefore, traditional methods rely on fixed-cycle evaluations and cannot adapt to the drastic fluctuations in fuel prices in the power industry. At the same time, production data and supply chain data are fragmented, the emergency procurement delay time is long, and it is difficult for artificial simulation tests to cover complex risk scenarios. This solution creatively adopts a comprehensive intelligent method that combines cross-domain data fusion with supply chain modeling and procurement strategy optimization to achieve real-time dynamic optimization of procurement efficiency optimization and management. In the existing supply chain relationship modeling methods, the existing supplier evaluations are mainly based on historical transaction data, but the relevant data on the real-time coal consumption rate changes of the units are not reflected in traditional methods. At the same time, traditional graph models cannot quantify the impact of risk events on procurement requirements, and it is difficult to characterize the physical relationship between coal quality and power generation efficiency. This solution creatively adopts a dynamically heterogeneous improved graph neural network method for supply chain relationship modeling, optimizes the state update of the supply chain and the quantitative calculation of risk conduction, and improves the intelligent dimension of procurement efficiency management. In the existing procurement strategy optimization methods, the traditional reinforcement learning procurement strategy optimization fails to make comprehensive decision-making analyses for complex procurement processes, various procurement risks, carbon emission indicators, and market fluctuations, resulting in the difficulty of the existing methods to conduct multi-directional comprehensive procurement decisions in combination with the results of supply chain relationship modeling. This solution creatively adopts a multi-objective reward-improved multi-layer strategy reinforcement learning method for procurement strategy optimization. By combining the improvement of state parameters, the improvement of rewards, and the improvement of perturbation control, and then building a basic multi-layer reinforcement learning model, the analysis efficiency of procurement strategies is improved, and the overall usability and intelligence of procurement efficiency management are also improved.

[0005] The technical solutions adopted by the present invention are as follows: The artificial intelligence-based procurement efficiency optimization management method provided by the present invention includes the following steps:

[0006] Step S1: Sub-item data collection;

[0007] Step S2: Data preprocessing;

[0008] Step S3: Supply chain relationship modeling;

[0009] Step S4: Procurement strategy optimization;

[0010] Step S5: Procurement efficiency optimization.

[0011] Further, in step S1, the sub-item data collection is used to collect the original dataset required for optimizing procurement efficiency. Specifically, through classified data collection, the original data for optimizing procurement efficiency management is obtained;

[0012] The classified data collection includes production data collection, supply chain data collection, and market data collection;

[0013] The original data for optimizing procurement efficiency management includes production data, supply chain data, and market data.

[0014] Further, in step S2, the data preprocessing is used to preprocess the original data. Specifically, based on the original data for optimizing procurement efficiency management, through data augmentation and transformation, data preprocessing is carried out to obtain the data for optimizing procurement efficiency, including the following steps:

[0015] Step S21: Original data mapping. Specifically, data mapping is carried out on the data in the original data for optimizing procurement efficiency management to obtain classified mapping data; the classified mapping data includes production mapping data, supply chain mapping data, and market mapping data;

[0016] Step S22: Production data preprocessing. Specifically, feature enhancement processing is carried out on the production mapping data. Through the sliding window feature enhancement operation, feature enhancement is carried out on the load rate and coal consumption rate data in the production mapping data to obtain the load efficiency feature and the coal consumption trend feature, and by statistically calculating the fault frequency of the fault event feature data, the fault risk score feature is obtained;

[0017] Step S23: Supply chain data preprocessing. Specifically, feature enhancement processing is carried out on the supply chain mapping data. By reconstructing the time series structure of the coal calorific value detection information in the basic features of the supplier, the supplier quality fluctuation feature is constructed, and by calculating the default probability of the supplier historical default record information in the basic features of the supplier, the default probability distribution feature is obtained, and by calculating the transportation efficiency of the supply coal transmission sensing information in the basic features of the supplier, the supply transportation efficiency feature is obtained;

[0018] Step S24: Market data preprocessing. Specifically, feature enhancement processing is carried out on the market mapping data. By decomposing the trend and calculating the volatility of the futures price data, the coal price trend feature is obtained, and by performing linear regression calculation on the carbon price data, the carbon price weight factor is obtained, and the carbon price weight factor is used as the carbon price feature;

[0019] Step S25: Feature fusion. Specifically, enhanced feature data is obtained through the preprocessing of the production data, the preprocessing of the supply chain data, and the preprocessing of the market data. And procurement efficiency optimization data is obtained by successively performing missing value filling, outlier processing, timestamp synchronization, high-dimensional feature compression, and feature labeling.

[0020] The procurement efficiency optimization data includes optimized load efficiency features, optimized coal consumption trend features, optimized fault risk scoring features, optimized supplier quality fluctuation features, optimized default probability distribution features, optimized supply transportation efficiency features, optimized coal price trend features, and optimized carbon price features.

[0021] Further, in step S3, the supply chain relationship modeling is used to digitally model and analyze the supply chain relationship of the materials to be purchased and evaluate the risks. Specifically, based on the procurement efficiency optimization data, a graph neural network method with dynamic heterogeneous improvement is used to perform supply chain relationship modeling to obtain supply chain relationship analysis reference data, including the following steps:

[0022] Step S31: Construction of the supply chain heterogeneous graph. Specifically, based on the procurement efficiency optimization data, a procurement supply chain heterogeneous graph is constructed, and supply chain heterogeneous graph data is obtained by defining node types, node attributes, and edge types.

[0023] The node types include supplier nodes, material nodes, and logistics nodes.

[0024] The node attributes include production capacity attributes, supplier credit rating attributes, coal inventory attributes, logistics node throughput capacity attributes, and transportation delay rate attributes.

[0025] The edge types include supply relationship edges, logistics path edges, and risk conduction edges.

[0026] The supply relationship edges are specifically established based on historical transaction frequencies; the logistics path edges are specifically established based on the actual transportation path and real-time transport capacity; the risk conduction edges are specifically used to simulate the risk cascading effect between nodes, including default and delay effects.

[0027] Step S32: Dynamic evolution and reconstruction of the graph structure. Specifically, a dynamic evolution mechanism is constructed for graph structure reconstruction. Specifically, the node state is updated once every hour, and the graph structure topology is reconstructed once every day to reflect the real-time changes of entity states and the impacts and changes of supply relationships.

[0028] Step S33: Hierarchical attention improvement, specifically constructing a three-layer hierarchical attention mechanism, including a cross-modal feature fusion attention layer, a spatio-temporal attention layer, and a risk conduction attention layer, for graph information extraction, and through the risk conduction attention layer, conducting risk cascade conduction modeling in the procurement process to calculate risk propagation probability reference data;

[0029] The cross-modal feature fusion attention layer is used to integrate node data and edge data, and through constructing a standard long short-term memory neural network, conduct temporal node feature processing to obtain relevance attention weights;

[0030] The spatio-temporal attention layer is used to introduce a time window attention weight function for supply chain temporal characteristic analysis, and through weighted calculation, obtain spatio-temporal attention weights;

[0031] The risk conduction attention layer is used to comprehensively calculate risk propagation probability by combining relevance attention weights and spatio-temporal attention weights to obtain risk propagation probability reference data, and the calculation formula is:

[0032] ;

[0033] In the formula, P risk is the risk propagation probability reference data, is the risk source node, is the conduction target node, sig(·) is the S-shaped activation function, K is the total number of intermediate nodes, k is the intermediate node index, used to represent the intermediate node between the risk source node and the conduction target node, is the relevance attention weight between the risk source node and the k-th intermediate node, is the spatio-temporal attention weight between the k-th intermediate node and the conduction target node, t is the time index, is the perturbation influence function, specifically using a standard Gaussian function to simulate the perturbation influence, is the reference value of material quality deviation;

[0034] Step S34: Risk propagation modeling, specifically through the risk propagation probability reference data and the supply chain heterogeneous graph data, conduct supply chain node risk simulation modeling to obtain procurement risk modeling reference data;

[0035] Step S35: Coal quality decomposition modeling, specifically based on the coal consumption trend characteristics and load efficiency characteristics in the procurement efficiency optimization data, conduct coal quality decomposition modeling to construct coal quality three-dimensional tensor modeling data and coal quality optimization reference data;

[0036] Step S36: Supply chain relationship modeling, specifically, through the construction of the supply chain heterogeneous graph, the reconstruction of the graph structure dynamic evolution, the improvement of the hierarchical attention, the modeling of the risk propagation, and the modeling of the coal quality decomposition, conduct supply chain relationship modeling to obtain supply chain relationship analysis reference data;

[0037] The supply chain relationship analysis reference data includes a supplier procurement risk assessment tensor, a supplier risk conduction probability matrix, and a supplier comprehensive evaluation value;

[0038] The supplier procurement risk assessment tensor includes credit risk assessment, delivery risk assessment, combustion quality risk assessment, price assessment, and emergency procurement quality assessment.

[0039] Further, in step S4, the procurement strategy optimization is used to calculate the optimal procurement strategy. Specifically, based on the supply chain relationship analysis reference data and the procurement efficiency optimization data, adopt a multi-objective reward-improved multi-layer policy reinforcement learning method to conduct procurement strategy optimization to obtain procurement optimal strategy reference data, including the following steps:

[0040] Step S41: Improvement of the state parameter space, specifically, construct the state parameter space of reinforcement learning and conduct system state vector improvement to obtain improved state parameter space data, including coal inventory status, unit operation status, and carbon emission quota balance status;

[0041] Step S42: Multi-layer policy network modeling, specifically, construct a meta-policy layer, a procurement dynamic optimization layer, and an intelligent allocation execution layer in sequence, and through the multi-layer policy network modeling, conduct multi-layer policy reinforcement learning to obtain an initial procurement strategy;

[0042] The meta-policy layer is used to handle the formulation of the annual framework agreement and optimize the long-term resource allocation. Specifically, adopt a meta-policy generator based on the transformer structure to generate a meta-procurement strategy;

[0043] The procurement dynamic optimization layer is used to control the dynamic optimization of the monthly procurement rhythm and supply structure. Specifically, adopt the double-delayed deep deterministic policy gradient algorithm for optimization calculation;

[0044] The intelligent allocation execution layer is used to intelligently allocate daily or real-time orders and balance the procurement cost and inventory pressure;

[0045] Step S43: Improvement of the multi-objective reward function, specifically, through the introduction of procurement cost, delivery deviation, emission index, and emergency procurement indicator function, conduct improvement of the multi-objective reward function to obtain an improved multi-objective reward function. The calculation formula is:

[0046] ;

[0047] In the formula, R t is an improved multi-objective reward function, is the purchase cost weight, C t is the purchase cost value, C max is the maximum purchase budget value, is the delivery bias weight, is the delivery deviation time, is the mean delivery deviation, is the emission index weight, m ​​is the emission element index, including sulfur dioxide emission, nitrogen oxide emission and dust emission, E m is the emission index parameter, is the maximum emission limit, is the emergency purchase instruction weight, is the emergency purchase indicator function value, which is used to indicate whether it is an emergency purchase;

[0048] Step S44: disturbance injection improvement, specifically, by introducing a disturbance factor control function, disturbance injection improvement is performed to obtain a disturbance injection function, the calculation formula of which is:

[0049] ;

[0050] Where roll(·) is the disturbance injection function, t is the time index, and g is the disturbance control parameter, which is used to control the degree of disturbance injection. By controlling the value of the disturbance injection function, disturbance simulation is performed, including market price fluctuation disturbance, transportation congestion disturbance, and supply chain disruption disturbance.

[0051] The specific calculation formula of the market price fluctuation disturbance is: ,in, is the historical statistical parameter of market volatility; the specific calculation formula of the transportation congestion disturbance is: ; The specific calculation formula for the supply chain disruption disturbance is: ,in, is the Bernoulli distribution function;

[0052] Step S45: Procurement strategy optimization, specifically, performing phased optimization of procurement strategy optimization through the state parameter space improvement, the multi-layer strategy network modeling, the multi-objective reward function improvement and the disturbance injection improvement, to obtain the optimal procurement strategy reference data;

[0053] The optimal procurement strategy reference data includes a multi-objective procurement strategy vector, a strategy enhancement reference function, and an emergency procurement plan reference data;

[0054] The phased optimization includes an initial stabilization phase, a disturbance introduction phase, and a multi-objective optimization phase;

[0055] The initial stable stage specifically involves improving based on the state parameter space and modeling through the multi-layer policy network, performing standard reinforcement learning to optimize the procurement strategy, and obtaining an initial procurement plan.

[0056] The perturbation introduction stage specifically involves, based on the initial procurement plan, improving according to the perturbation injection, introducing simulations of price fluctuations, transportation congestion, and supply interruptions, and performing perturbation introduction to improve the procurement strategy optimization to obtain a perturbation-resistant procurement plan.

[0057] The multi-objective optimization stage specifically involves, based on the perturbation-resistant procurement plan, improving according to the multi-objective reward function, and performing procurement strategy optimization with multi-objective optimization to obtain reference data for the optimal procurement strategy.

[0058] Furthermore, in step S5, the procurement efficiency optimization is used to construct an intelligent evaluation method to verify the optimized procurement strategy and improve the procurement efficiency. Specifically, based on the reference data for the optimal procurement strategy, multi-dimensional procurement strategy evaluation indicators are constructed and the procurement strategy effect is evaluated. By selecting the procurement strategy with the best evaluation effect, the procurement efficiency is optimized to obtain a comprehensive optimization reference plan for the procurement efficiency.

[0059] The multi-dimensional procurement strategy evaluation indicators include economic indicators, reliability indicators, environmental protection indicators, and agility indicators.

[0060] The economic indicator specifically refers to the deviation rate between the actual cost and the budget cost; the reliability indicator specifically refers to the emergency procurement frequency; the environmental protection indicator specifically refers to the emission quota utilization rate; the agility indicator specifically refers to the response time of the procurement strategy.

[0061] The procurement efficiency optimization management system based on artificial intelligence provided by the present invention includes a data source processing module, a supply relationship modeling module, a procurement strategy optimization module, and a comprehensive efficiency optimization module.

[0062] The data source processing module is used for data itemized collection and data preprocessing. Through data itemized collection and data preprocessing, procurement efficiency optimization data is obtained, and the procurement efficiency optimization data is sent to the supply relationship modeling module and the procurement strategy optimization module.

[0063] The supply relationship modeling module is used for supply chain relationship modeling. Through supply chain relationship modeling, reference data for supply chain relationship analysis is obtained, and the reference data for supply chain relationship analysis is sent to the procurement strategy optimization module.

[0064] The procurement strategy optimization module is used for procurement strategy optimization. Through procurement strategy optimization, reference data for the optimal procurement strategy is obtained, and the reference data for the optimal procurement strategy is sent to the comprehensive efficiency optimization module.

[0065] The comprehensive efficiency optimization module is used for optimizing the procurement efficiency. Through the optimization of the procurement efficiency, a comprehensive optimization reference plan for the procurement efficiency is obtained.

[0066] The beneficial effects achieved by the present invention using the above solution are as follows:

[0067] (1) In the existing procurement efficiency optimization management method, there is a problem that the existing power industry fuel procurement mostly adopts a static evaluation system. However, it is inevitable that the static system is prone to being out of touch with the dynamic market, which in turn leads to decision-making lag, and there is a lack of effective application in extreme scenarios. Therefore, the traditional method relies on fixed-cycle evaluation and cannot adapt to the drastic fluctuations in fuel prices in the power industry. At the same time, production data and supply chain data are fragmented, the emergency procurement delay time is long, and it is difficult for manual simulation tests to cover complex risk scenarios. This solution creatively adopts a comprehensive intelligent method that combines cross-domain data fusion with supply chain modeling and procurement strategy optimization, realizing real-time dynamic optimization of procurement efficiency optimization and management;

[0068] (2) In the existing supply chain relationship modeling method, there is a problem that the existing supplier evaluation is mainly based on historical transaction data, but the relevant data on the real-time coal consumption rate change of the unit is not reflected in the traditional method. At the same time, the traditional graph model cannot quantify the impact of risk events on procurement demand, and it is difficult to characterize the physical relationship between coal quality and power generation efficiency. This solution creatively adopts a dynamically heterogeneous improved graph neural network method to conduct supply chain relationship modeling, optimizing the state update of the supply chain and the quantitative calculation of risk conduction, and enhancing the intelligent dimension of procurement efficiency management;

[0069] (3) In the existing procurement strategy optimization method, there is a problem that the traditional reinforcement learning procurement strategy optimization fails to make comprehensive decision-making analysis for complex procurement processes, various procurement risks, carbon emission indicators, and market fluctuations, which in turn leads to the difficulty of the existing method to conduct multi-directional comprehensive procurement decisions in combination with the results of supply chain relationship modeling. This solution creatively adopts a multi-objective reward-improved multi-layer strategy reinforcement learning method to optimize the procurement strategy. Through the combination of state parameter improvement, reward improvement, and perturbation control improvement, and then building a multi-layer reinforcement learning basic model, the analysis efficiency of the procurement strategy is improved, and the overall usability and intelligence of procurement efficiency management are also improved. Brief Description of the Drawings

[0070] Figure 1 It is a schematic flow chart of the procurement efficiency optimization management method based on artificial intelligence provided by the present invention;

[0071] Figure 2 It is a schematic diagram of the procurement efficiency optimization management system based on artificial intelligence provided by the present invention;

[0072] Figure 3 It is a schematic flowchart of data preprocessing in step S2;

[0073] Figure 4 It is a schematic flowchart of supply chain relationship modeling in step S3;

[0074] Figure 5 It is a schematic flowchart of procurement strategy optimization in step S4.

[0075] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. Detailed implementation manners

[0076] 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 of the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0077] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be understood as a limitation to the present invention.

[0078] Example 1, refer to Figure 1 , the procurement efficiency optimization management method based on artificial intelligence provided by the present invention, the method includes the following steps:

[0079] Step S1: Data itemized collection;

[0080] Step S2: Data preprocessing;

[0081] Step S3: Supply chain relationship modeling;

[0082] Step S4: Procurement strategy optimization;

[0083] Step S5: Procurement efficiency optimization.

[0084] By performing the above operations, in the existing procurement efficiency optimization management method, there is a problem that the existing power industry fuel procurement mostly adopts a static evaluation system. However, it is inevitable that the static system is prone to being out of touch with the dynamic market, which in turn leads to lagging decisions and a lack of effective application in extreme scenarios. Therefore, the traditional method relies on fixed-cycle evaluation and cannot adapt to the drastic fluctuations in fuel prices in the power industry. At the same time, production data and supply chain data are fragmented, the emergency procurement delay time is long, and it is difficult for manual simulation tests to cover complex risk scenarios. This solution creatively adopts a comprehensive intelligent method that combines cross-domain data fusion, supply chain modeling, and procurement strategy optimization, achieving real-time dynamic optimization of procurement efficiency optimization and management.

[0085] Example 2. Refer to Figure 1 and Figure 2 In step S1, the data itemized collection is used to collect the original data set required for procurement efficiency optimization. Specifically, through classified data collection, the original data for procurement efficiency optimization management is obtained;

[0086] The classified data collection includes production data collection, supply chain data collection, and market data collection;

[0087] The production data collection includes unit load rate collection, coal consumption rate collection, and equipment failure log collection;

[0088] The supply chain data collection includes coal inventory collection, basic information collection of spare part suppliers, and transportation schedule data collection;

[0089] The basic information collection of spare part suppliers includes coal calorific value detection report collection, supplier historical default record collection, and supply coal transportation sensing data collection;

[0090] The market data collection includes thermal coal futures price collection and carbon emission rights trading data collection;

[0091] The original data for procurement efficiency optimization management includes production data, supply chain data, and market data;

[0092] The production data, with the collection frequency set to the minute level, includes time series data and event log data;

[0093] The supply chain data, with the collection frequency set to the hour level, includes supplier information text, structured tables, and transportation geographic information;

[0094] The market data, with the collection frequency set to the day level, includes multivariate time series data.

[0095] Example 3. Refer to Figure 1 、 Figure 2 and Figure 3, this embodiment is based on the above embodiment. In step S2, the data preprocessing is used to preprocess the original data. Specifically, the original data is optimized and managed according to the procurement efficiency, and through data augmentation and transformation, data preprocessing is carried out to obtain procurement efficiency optimization data, including the following steps:

[0096] Step S21: Original data mapping. Specifically, data mapping is performed on the data in the original data for procurement efficiency optimization management to obtain classified mapping data; the classified mapping data includes production mapping data, supply chain mapping data, and market mapping data;

[0097] The production mapping data includes load rate, coal consumption rate, and fault event feature data; the fault event feature data includes timestamp, equipment number, and fault type;

[0098] The supply chain mapping data includes inventory, supply transportation trajectory, and basic supplier characteristics;

[0099] The basic supplier characteristics include coal calorific value detection information, supplier historical default record information, and supply coal transportation sensing information;

[0100] The market mapping data includes futures price data and carbon price data;

[0101] Step S22: Production data preprocessing. Specifically, feature enhancement processing is performed on the production mapping data. Through sliding window feature enhancement operation, feature enhancement is performed on the load rate and coal consumption rate data in the production mapping data to obtain load efficiency features and coal consumption trend features, and by performing fault frequency statistics on the fault event feature data, a fault risk score feature is obtained;

[0102] Step S23: Supply chain data preprocessing. Specifically, feature enhancement processing is performed on the supply chain mapping data. By reconstructing the time series structure of the coal calorific value detection information in the basic supplier characteristics, a supplier quality fluctuation feature is constructed, and by calculating the default probability of the supplier historical default record information in the basic supplier characteristics, a default probability distribution feature is obtained. By calculating the transportation efficiency of the supply coal transportation sensing information in the basic supplier characteristics, a supply transportation efficiency feature is obtained;

[0103] The supplier quality fluctuation feature includes coal calorific value, sulfur content, and ash content;

[0104] The supply transportation efficiency feature is specifically calculated by dividing the actual transported volume by the actual transport capacity;

[0105] Step S24: Preprocess market data, specifically perform feature enhancement processing on the market mapping data. By performing trend decomposition and volatility calculation on the futures price data, obtain coal price trend features, and perform linear regression calculation on the carbon price data to obtain a carbon price weight factor, and use the carbon price weight factor as a carbon price feature;

[0106] Step S25: Feature fusion, specifically obtain enhanced feature data through the production data preprocessing, the supply chain data preprocessing, and the market data preprocessing, and obtain procurement efficiency optimization data by sequentially performing missing value filling, outlier handling, timestamp synchronization, high-dimensional feature compression, and feature labeling;

[0107] The procurement efficiency optimization data includes optimized load efficiency features, optimized coal consumption trend features, optimized fault risk scoring features, optimized supplier quality fluctuation features, optimized default probability distribution features, optimized supply transportation efficiency features, optimized coal price trend features, and optimized carbon price features. The calculation formula is:

[0108] ;

[0109] In the formula, X is the procurement efficiency optimization data, e load is the optimized load efficiency feature, y coal is the optimized coal consumption trend feature, p fault is the fault risk scoring feature, Q t is the optimized supplier quality fluctuation feature, P de is the optimized default probability distribution feature, n tr is the optimized supply transportation efficiency feature, S coal is the optimized coal price trend feature, C car is the optimized carbon price feature.

[0110] Example 4, refer to Figure 1 、 Figure 2 and Figure 4 , based on the above example, in step S3, the supply chain relationship modeling is used to digitally model and analyze the supply chain relationship of the materials to be purchased and evaluate risks. Specifically, based on the procurement efficiency optimization data, use the dynamically heterogeneous improved graph neural network method to perform supply chain relationship modeling to obtain supply chain relationship analysis reference data, including the following steps:

[0111] Step S31: Construct a supply chain heterogeneous graph, specifically construct a procurement supply chain heterogeneous graph based on the procurement efficiency optimization data, and obtain supply chain heterogeneous graph data by defining node types, node attributes, and edge types. The calculation formula is:

[0112] ;

[0113] Wherein, G t is the heterogeneous graph data of the supply chain, V t is the set of nodes, which is used to represent coal suppliers, spare part manufacturers and coal hub locations, and E t is the set of edges, which is used to represent supply transportation videos and procurement relationships, and U t is the coal price trend vector, which is used to represent the coal price trend and carbon price weight;

[0114] The node types include supplier nodes, material nodes and logistics nodes;

[0115] The node attributes include production capacity attributes, supplier credit rating attributes, coal inventory attributes, logistics node throughput capacity attributes and transportation delay rate attributes;

[0116] The edge types include supply relationship edges, logistics path edges and risk conduction edges;

[0117] The supply relationship edge is specifically established based on historical transaction frequencies; the logistics path edge is specifically established based on the actual transportation path and real-time transport capacity; the risk conduction edge is specifically used to simulate the risk cascade effect between nodes, including default and delay effects;

[0118] Step S32: Dynamic evolution reconstruction of the graph structure, specifically constructing a dynamic evolution mechanism for graph structure reconstruction, specifically updating the node state once every hour and performing a graph structure topology reconstruction once every day, for reflecting the real-time changes of entity states and the influence and changes of supply relationships;

[0119] Step S33: Hierarchical attention improvement, specifically constructing a three-layer hierarchical attention mechanism, including a cross-modal feature fusion attention layer, a spatio-temporal attention layer and a risk conduction attention layer, for graph information extraction, and through the risk conduction attention layer, performing risk cascade conduction modeling in the procurement process to calculate the risk propagation probability reference data;

[0120] The cross-modal feature fusion attention layer is used to integrate node data and edge data, and through constructing a standard long short-term memory neural network, perform time-series node feature processing to obtain the correlation attention weight, and the calculation formula is:

[0121] ;

[0122] Wherein, a ij is the correlation attention weight, softmax(·) is the classifier function, sig(·) is the S-shaped activation function, W l is the cross-modal feature fusion attention layer weight, h i is the hidden state output by the node data through the standard long short-term memory neural network, i is the node index, hj is the hidden state output by the edge data through the standard long short-term memory neural network, j is the edge index, and a l is a learnable coefficient;

[0123] The spatio-temporal attention layer is used to introduce a time window attention weight function for analyzing the temporal characteristics of the supply chain, and through weighted calculation, obtain spatio-temporal attention weights. The calculation formula is:

[0124] ;

[0125] In the formula, b t is the spatio-temporal attention weight, exp(·) is the natural base function, f(·) is the time window attention weight function, and the specific calculation formula is , where is the time period variable, w1 is the first time window attention weight, w2 is the second time window attention weight, t0 is the time series fluctuation time index used to represent the occurrence time of abnormal events or risk events in the time series data, t is the time index, T is the total number of time series fluctuations, and t1 is the second time series fluctuation time index;

[0126] The risk conduction attention layer is used to comprehensively calculate the risk propagation probability by combining the relevance attention weight and the spatio-temporal attention weight, and obtain risk propagation probability reference data. The calculation formula is:

[0127] ;

[0128] In the formula, P risk is the risk propagation probability reference data, is the risk source node, is the conduction target node, sig(·) is the S-shaped activation function, K is the total number of intermediate nodes, and k is the intermediate node index used to represent the intermediate node between the risk source node and the conduction target node, is the relevance attention weight between the risk source node and the k-th intermediate node, is the spatio-temporal attention weight between the k-th intermediate node and the conduction target node, t is the time index, is the perturbation influence function, and specifically uses the standard Gaussian function to simulate the perturbation influence, is the reference value of the material quality deviation;

[0129] Step S34: Risk propagation modeling, specifically, based on the risk propagation probability reference data and the supply chain heterogeneous graph data, perform supply chain node risk simulation modeling to obtain procurement risk modeling reference data;

[0130] Step S35: Coal quality decomposition modeling. Specifically, based on the coal consumption trend characteristics and load efficiency characteristics in the optimized procurement efficiency data, coal quality decomposition modeling is carried out to construct coal quality three-dimensional tensor modeling data and coal quality optimization reference data. The calculation formula is as follows:

[0131] ;

[0132] In the formula, X c is the coal quality three-dimensional tensor modeling data, P f is the combustion efficiency reference data, which is specifically calculated by weighting the coal consumption trend characteristics and load efficiency characteristics. 1U is the supplier dimension data, 2V is the mining area source dimension data, and 3W is the transportation mode dimension data. By performing target optimization decomposition on the coal quality three-dimensional tensor modeling data, coal quality optimization reference data is obtained;

[0133] For the target optimization decomposition, specifically, Tucker decomposition is used for calculation. The calculation formula is as follows:

[0134] ;

[0135] In the formula, F target is the objective function of the target optimization decomposition, min is the function to find the minimum value, X c is the coal quality three-dimensional tensor modeling data, Core is the coal quality core vector obtained through Tucker decomposition, U is the characteristic factor matrix of the supplier dimension, V is the characteristic factor matrix of the mining area source dimension, and W is the characteristic factor matrix of the transportation mode dimension. is the Frobenius norm operator, which is used to measure the tensor error after decomposition. is the regularization coefficient, tr(·) is the regularization control function, and Core T is the transpose of the coal quality core vector;

[0136] Step S36: Supply chain relationship modeling. Specifically, through the construction of the supply chain heterogeneous graph, the reconstruction of the graph structure dynamic evolution, the improvement of the hierarchical attention, the modeling of the risk propagation, and the coal quality decomposition modeling, supply chain relationship modeling is carried out to obtain supply chain relationship analysis reference data;

[0137] The supply chain relationship analysis reference data includes the supplier procurement risk assessment tensor, the supplier risk conduction probability matrix, and the supplier comprehensive evaluation value;

[0138] The supplier procurement risk assessment tensor includes credit risk assessment, delivery risk assessment, combustion quality risk assessment, price assessment, and emergency procurement quality assessment.

[0139] By performing the above operations, in the existing supply chain relationship modeling methods, there are technical problems in the existing supplier evaluation, which is mainly based on historical transaction data, but the relevant data on the real-time coal consumption rate of the unit is not reflected in the traditional methods. At the same time, the traditional graph model cannot quantify the impact of risk events on procurement requirements, and it is difficult to characterize the physical relationship between coal quality and power generation efficiency. This solution creatively uses the dynamically heterogeneous improved graph neural network method to model the supply chain relationship, optimizes the state update of the supply chain and the quantitative calculation of risk conduction, and improves the intelligent dimension of procurement efficiency management.

[0140] Example 5, refer to Figure 1 、 Figure 2 and Figure 5 , in step S4, the procurement strategy optimization is used to calculate the optimal procurement strategy. Specifically, based on the supply chain relationship analysis reference data and the procurement efficiency optimization data, a multi-objective reward-improved multi-layer policy reinforcement learning method is used to optimize the procurement strategy, and the procurement optimal strategy reference data is obtained, including the following steps:

[0141] Step S41: Improvement of the state parameter space. Specifically, the state parameter space of reinforcement learning is constructed and the system state vector is improved to obtain the improved state parameter space data, including the coal inventory state, the unit operation state, and the carbon emission quota balance state;

[0142] Step S42: Multi-layer policy network modeling. Specifically, the meta-policy layer, the procurement dynamic optimization layer, and the intelligent allocation execution layer are constructed in sequence, and through the multi-layer policy network modeling, multi-layer policy reinforcement learning is performed to obtain the initial procurement strategy;

[0143] The meta-policy layer is used to handle the formulation of the annual framework agreement and optimize the long-term resource allocation. Specifically, a meta-policy generator based on the transformer structure is used to generate the meta-purchase strategy, and the calculation formula is:

[0144] ;

[0145] In the formula, is the meta-purchase strategy parameter group, softmax(·) is the classifier function, W Q is the meta-policy generation weight, K is the key vector, V is the value vector, and d is the dimension value of the key vector;

[0146] The procurement dynamic optimization layer is used to control the dynamic optimization of the monthly procurement rhythm and supply structure. Specifically, the double-delayed deep deterministic policy gradient algorithm is used for optimization calculation, and the calculation formula is:

[0147] ;

[0148] In the formula, at is the specific procurement action adopted at the current time t, s t is the improved state parameter space data, u(·) is the policy network function using the double-delay deep deterministic policy gradient algorithm. The policy network constructs five hidden layers to optimize the non-linear fitting ability, are the parameters of the policy network, is the overall added random Gaussian noise, are the Gaussian noise distribution parameters;

[0149] The intelligent allocation execution layer is used to intelligently allocate daily or real-time orders and balance the procurement cost and inventory pressure. The calculation formula is:

[0150] ;

[0151] In the formula, P pi is the short-term procurement order decision result of intelligent allocation execution, max is the maximum value function, E[·] is the expectation calculation function, p n is the expected revenue probability of the nth short-term procurement order decision strategy, and n is the short-term procurement order decision strategy index, is the probability likelihood function value of obtaining the return reward r under the parameter , is the parameter under the historical data D. The posterior distribution is specifically calculated using Bayesian update;

[0152] Step S43: Improvement of the multi-objective reward function. Specifically, by introducing the procurement cost, delivery deviation, emission index, and emergency procurement indicator function, the multi-objective reward function is improved to obtain the improved multi-objective reward function. The calculation formula is:

[0153] ;

[0154] In the formula, R t is the improved multi-objective reward function, is the procurement cost weight, C t is the procurement cost value, C max is the maximum procurement budget value, is the delivery deviation weight, is the delivery deviation time, is the delivery deviation mean, is the emission index weight, m is the emission element index, including sulfur dioxide emission, nitrogen oxide emission, and dust emission, E m is the emission index parameter, is the maximum emission limit, is the emergency procurement indicator weight, is the emergency purchase indicator function value, which is used to indicate whether it is an emergency purchase;

[0155] Preferably, the purchase cost weight The specific value of is 0.45, the delivery deviation weight The specific value of is 0.25, and the emission index weight The specific value is 0.2, and the emergency purchase indication weight The value of is 0.1;

[0156] Step S44: disturbance injection improvement, specifically, by introducing a disturbance factor control function, disturbance injection improvement is performed to obtain a disturbance injection function, the calculation formula of which is:

[0157] ;

[0158] Where roll(·) is the disturbance injection function, t is the time index, and g is the disturbance control parameter, which is used to control the degree of disturbance injection. By controlling the value of the disturbance injection function, disturbance simulation is performed, including market price fluctuation disturbance, transportation congestion disturbance, and supply chain disruption disturbance.

[0159] The specific calculation formula of the market price fluctuation disturbance is: ,in, is the historical statistical parameter of market volatility; the specific calculation formula of the transportation congestion disturbance is: ; The specific calculation formula for the supply chain disruption disturbance is: ,in, is the Bernoulli distribution function;

[0160] Preferably, the specific value of the disturbance control parameter is 0.05;

[0161] Step S45: Procurement strategy optimization, specifically, performing phased optimization of procurement strategy optimization through the state parameter space improvement, the multi-layer strategy network modeling, the multi-objective reward function improvement and the disturbance injection improvement, to obtain the optimal procurement strategy reference data;

[0162] The optimal procurement strategy reference data includes a multi-objective procurement strategy vector, a strategy enhancement reference function, and an emergency procurement plan reference data;

[0163] The phased optimization includes an initial stabilization phase, a disturbance introduction phase, and a multi-objective optimization phase;

[0164] The initial stabilization stage is specifically to optimize the standard reinforcement learning procurement strategy based on the state parameter space improvement and multi-layer strategy network modeling to obtain an initial procurement plan;

[0165] In the disturbance introduction stage, specifically, based on the initial procurement plan, improvements are made according to the disturbance injection, price fluctuations, transportation congestion, and supply interruption simulations are introduced, and the procurement strategy optimization with disturbance introduction improvements is carried out to obtain a disturbance-resistant procurement plan.

[0166] In the multi-objective optimization stage, specifically, based on the disturbance-resistant procurement plan, improvements are made according to the multi-objective reward function, and the procurement strategy optimization with multi-objective optimization is carried out to obtain the reference data of the optimal procurement strategy.

[0167] By performing the above operations, in the existing procurement strategy optimization methods, there is a problem that traditional reinforcement learning procurement strategy optimization fails to make comprehensive decision-making analyses for complex procurement processes, various procurement risks, carbon emission indicators, and market fluctuations, resulting in the difficulty of the existing methods to conduct multi-directional comprehensive procurement decisions in combination with the results of supply chain relationship modeling. This solution creatively adopts a multi-objective reward-improved multi-layer strategy reinforcement learning method to optimize the procurement strategy. Through the combination of state parameter improvement, reward improvement, and disturbance control improvement, and then building a basic multi-layer reinforcement learning model, the analysis efficiency of the procurement strategy is improved, and the usability and intelligence of the overall procurement efficiency management are also improved.

[0168] Example Six, refer to Figure 1 and Figure 2 In this example, based on the above example, in step S5, the procurement efficiency optimization is used to construct an intelligent evaluation method to verify the optimized procurement strategy and improve the procurement efficiency. Specifically, based on the reference data of the optimal procurement strategy, multi-dimensional procurement strategy evaluation indicators are constructed and the procurement strategy effect is evaluated. By selecting the procurement strategy with the best evaluation effect, the procurement efficiency is optimized to obtain a comprehensive reference plan for procurement efficiency optimization.

[0169] The multi-dimensional procurement strategy evaluation indicators include economic indicators, reliability indicators, environmental protection indicators, and agility indicators.

[0170] The economic indicator specifically refers to the deviation rate of the actual cost from the budget cost; the reliability indicator specifically refers to the emergency procurement frequency; the environmental protection indicator specifically refers to the emission quota utilization rate; the agility indicator specifically refers to the response time of the procurement strategy.

[0171] Example Seven, refer to Figure 1 and Figure 2 In this example, based on the above example, the procurement efficiency optimization management system based on artificial intelligence provided by the present invention includes a data source processing module, a supply relationship modeling module, a procurement strategy optimization module, and a comprehensive efficiency optimization module.

[0172] The data source processing module is used for itemized data collection and data preprocessing. Through itemized data collection and data preprocessing, optimized procurement efficiency data is obtained, and the optimized procurement efficiency data is sent to the supply relationship modeling module and the procurement strategy optimization module;

[0173] The supply relationship modeling module is used for supply chain relationship modeling. Through supply chain relationship modeling, reference data for supply chain relationship analysis is obtained, and the reference data for supply chain relationship analysis is sent to the procurement strategy optimization module;

[0174] The procurement strategy optimization module is used for procurement strategy optimization. Through procurement strategy optimization, reference data for the optimal procurement strategy is obtained, and the reference data for the optimal procurement strategy is sent to the comprehensive efficiency optimization module;

[0175] The comprehensive efficiency optimization module is used for procurement efficiency optimization. Through procurement efficiency optimization, a reference solution for comprehensive optimization of procurement efficiency is obtained.

[0176] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0177] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

[0178] The above describes the present invention and its implementation manners. Such a description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. An artificial intelligence-based procurement efficiency optimization management method, characterized in that: The method includes the following steps: Step S1: Data itemized collection to obtain the original data for optimizing procurement efficiency management; Step S2: Data preprocessing, through data augmentation and transformation, to perform data preprocessing and obtain the data for optimizing procurement efficiency; Step S3: Supply chain relationship modeling, using the method of dynamically heterogeneous improved graph neural network to conduct supply chain relationship modeling and obtain the reference data for supply chain relationship analysis, including the following steps: Step S31: Construction of a supply chain heterogeneous graph; Step S32: Dynamic evolution reconstruction of the graph structure; Step S33: Hierarchical attention improvement, constructing a three-layer hierarchical attention mechanism, including a cross-modal feature fusion attention layer, a spatio-temporal attention layer, and a risk conduction attention layer, to extract graph information; Step S34: Risk propagation modeling; Step S35: Coal quality decomposition modeling; Step S36: Supply chain relationship modeling; The cross-modal feature fusion attention layer is used to integrate node data and edge data, and through constructing a standard long short-term memory neural network, to process the temporal node features and obtain the correlation attention weights; the spatio-temporal attention layer is used to introduce a time window attention weight function for analyzing the temporal characteristics of the supply chain, and through weighted calculation, to obtain the spatio-temporal attention weights; the risk conduction attention layer is used to comprehensively calculate the risk propagation probability by combining the correlation attention weights and the spatio-temporal attention weights to obtain the reference data for the risk propagation probability; Step S4: Procurement strategy optimization, using the method of multi-objective reward improved multi-layer policy reinforcement learning to optimize the procurement strategy and obtain the reference data for the optimal procurement strategy, including the following steps: Step S41: Improvement of the state parameter space; Step S42: Multi-layer policy network modeling, successively constructing a meta-policy layer, a procurement dynamic optimization layer, and an intelligent allocation execution layer, and through the multi-layer policy network modeling, to conduct multi-layer policy reinforcement learning; Step S43: Improvement of the multi-objective reward function, by introducing the procurement cost, delivery deviation, emission index, and emergency procurement indicator function, to improve the multi-objective reward function; Step S44: Improvement of perturbation injection; Step S45: Procurement strategy optimization; In Step S42, the meta-policy layer is used to handle the formulation of the annual framework agreement and optimize the long-term resource allocation, specifically using a meta-policy generator based on the transformer structure to generate the meta-procurement strategy; the procurement dynamic optimization layer is used to control the dynamic optimization of the monthly procurement rhythm and supply structure, specifically using the double delayed deep deterministic policy gradient algorithm for optimization calculation; the intelligent allocation execution layer is used to intelligently allocate the daily or real-time orders and balance the procurement cost and inventory pressure; Step S5: Optimization of procurement efficiency to obtain a comprehensive optimization reference plan for procurement efficiency.

2. The procurement efficiency optimization management method based on artificial intelligence according to claim 1, characterized in that: In Step S1, the data itemized collection is used to collect the original data set required for optimizing procurement efficiency, specifically by classifying data collection to obtain the original data for optimizing procurement efficiency management; The classified data collection includes production data collection, supply chain data collection, and market data collection; the original data for optimizing procurement efficiency management includes production data, supply chain data, and market data.

3. The procurement efficiency optimization management method based on artificial intelligence according to claim 2, wherein: In step S2, the data preprocessing is used to preprocess the original data. Specifically, the original data is optimized and managed according to the procurement efficiency. Through data augmentation and transformation, data preprocessing is performed to obtain procurement efficiency optimization data, including the following steps: Step S21: Original data mapping. Specifically, data mapping is performed on the data in the original data for procurement efficiency optimization management to obtain classified mapping data. The classified mapping data includes production mapping data, supply chain mapping data, and market mapping data. The production mapping data includes load rate, coal consumption rate, and fault event characteristic data. The fault event characteristic data includes timestamp, equipment number, and fault type. The supply chain mapping data includes inventory, supply transportation trajectory, and basic supplier characteristics. The basic supplier characteristics include coal calorific value detection information, supplier historical default record information, and supply coal transportation sensing information. The market mapping data includes futures price data and carbon price data. Step S22: Production data preprocessing. Specifically, feature enhancement processing is performed on the production mapping data. Through sliding window feature enhancement operations, feature enhancement is performed on the load rate and coal consumption rate data in the production mapping data to obtain load efficiency features and coal consumption trend features. And by performing fault frequency statistics on the fault event characteristic data, a fault risk score feature is obtained. Step S23: Supply chain data preprocessing. Specifically, feature enhancement processing is performed on the supply chain mapping data. By reconstructing the time series structure of the coal calorific value detection information in the basic supplier characteristics, a supplier quality fluctuation feature is constructed. And by calculating the default probability of the supplier historical default record information in the basic supplier characteristics, a default probability distribution feature is obtained. By calculating the transportation efficiency of the supply coal transportation sensing information in the basic supplier characteristics, a supply transportation efficiency feature is obtained. Step S24: Market data preprocessing. Specifically, feature enhancement processing is performed on the market mapping data. By performing trend decomposition and volatility calculation on the futures price data, a coal price trend feature is obtained. And by performing linear regression calculation on the carbon price data, a carbon price weight factor is obtained, and the carbon price weight factor is used as the carbon price feature. Step S25: Feature fusion. Specifically, through the production data preprocessing, the supply chain data preprocessing, and the market data preprocessing, enhanced feature data is obtained. And by sequentially performing missing value filling, outlier processing, timestamp synchronization, high-dimensional feature compression, and feature labeling, procurement efficiency optimization data is obtained. The procurement efficiency optimization data includes optimized load efficiency features, optimized coal consumption trend features, optimized fault risk score features, optimized supplier quality fluctuation features, optimized default probability distribution features, optimized supply transportation efficiency features, optimized coal price trend features, and optimized carbon price features.

4. The procurement efficiency optimization management method based on artificial intelligence according to claim 3, characterized in that: In step S3, the supply chain relationship modeling is used to digitally model and analyze the supply chain relationship of the materials to be purchased and evaluate risks. Specifically, based on the procurement efficiency optimization data, a graph neural network method with dynamic heterogeneous improvement is used to conduct supply chain relationship modeling to obtain supply chain relationship analysis reference data, including the following steps: Step S31: Construction of a supply chain heterogeneous graph. Specifically, based on the procurement efficiency optimization data, a procurement supply chain heterogeneous graph is constructed, and by defining node types, node attributes, and edge types, supply chain heterogeneous graph data is obtained; The node types include supplier nodes, material nodes, and logistics nodes; The node attributes include production capacity attributes, supplier credit rating attributes, coal inventory attributes, logistics node throughput capacity attributes, and transportation delay rate attributes; The edge types include supply relationship edges, logistics path edges, and risk conduction edges; Step S32: Dynamic evolution reconstruction of the graph structure. Specifically, a dynamic evolution mechanism is constructed for graph structure reconstruction. Specifically, the node status is updated once every hour, and the graph structure topology is reconstructed once every day to reflect the real-time changes of entity status and the influence and changes of supply relationships; Step S33: Hierarchical attention improvement. Specifically, a three-layer hierarchical attention mechanism is constructed, including a cross-modal feature fusion attention layer, a spatio-temporal attention layer, and a risk conduction attention layer, to extract graph information, and through the risk conduction attention layer, a risk cascade conduction model of the procurement process is established, and risk propagation probability reference data is calculated; Step S34: Risk propagation modeling. Specifically, based on the risk propagation probability reference data and the supply chain heterogeneous graph data, a supply chain node risk simulation model is established to obtain procurement risk modeling reference data; Step S35: Coal quality decomposition modeling. Specifically, based on the coal consumption trend characteristics and load efficiency characteristics in the procurement efficiency optimization data, coal quality decomposition modeling is carried out to construct coal quality three-dimensional tensor modeling data and coal quality optimization reference data; Step S36: Supply chain relationship modeling. Specifically, through the construction of the supply chain heterogeneous graph, the dynamic evolution reconstruction of the graph structure, the hierarchical attention improvement, the risk propagation modeling, and the coal quality decomposition modeling, supply chain relationship modeling is carried out to obtain supply chain relationship analysis reference data; The supply chain relationship analysis reference data includes a supplier procurement risk assessment tensor, a supplier risk conduction probability matrix, and a supplier comprehensive evaluation value; The supplier procurement risk assessment tensor includes credit risk assessment, delivery risk assessment, combustion quality risk assessment, price assessment, and emergency procurement quality assessment.

5. The procurement efficiency optimization management method based on artificial intelligence according to claim 4, characterized in that: In step S33, the calculation formula for the risk propagation probability reference data is: Where P risk is the reference data of risk propagation probability, v i′ is the risk source node, v j′ is the conduction target node, sig(·) is the S-shaped activation function, K is the total number of intermediate nodes, k is the intermediate node index, which is used to represent the intermediate nodes in the risk source node and the conduction target node, a i′k is the correlation attention weight between the risk source node and the k-th intermediate node, b kj′，t is the spatio-temporal attention weight between the k-th intermediate node and the conduction target node, t is the time index, is the perturbation influence function, and the standard Gaussian function is specifically used to simulate the perturbation influence. Δq is the reference value of the material quality deviation.

6. The method for optimizing and managing procurement efficiency based on artificial intelligence according to claim 5, wherein: In step S4, the procurement strategy optimization is used to calculate the optimal procurement strategy. Specifically, based on the supply chain relationship analysis reference data and the procurement efficiency optimization data, a multi-objective reward-improved multi-layer policy reinforcement learning method is used to optimize the procurement strategy to obtain procurement optimal strategy reference data, including the following steps: Step S41: improving the state parameter space, specifically constructing a state parameter space for reinforcement learning and improving the system state vector to obtain improved state parameter space data, including coal inventory status, unit operation status, and carbon emission quota balance status; Step S42: multi-layer strategy network modeling, specifically constructing a meta-strategy layer, a procurement dynamic optimization layer, and an intelligent allocation execution layer in sequence, and performing multi-layer strategy reinforcement learning through the multi-layer strategy network modeling to obtain an initial procurement strategy; Step S43: Improvement of the multi-objective reward function, specifically, by introducing procurement cost, delivery deviation, emission index and emergency procurement indicator function, the multi-objective reward function is improved to obtain an improved multi-objective reward function, and the calculation formula is: where R t is the improved multi - objective reward function, ω1 is the procurement cost weight, C t is the procurement cost value, C max is the maximum procurement budget value, ω2 is the delivery deviation weight, ΔD is the delivery deviation time, σD is the mean of delivery deviation, ω3 is the emission index weight, m is the emission element index, including sulfur dioxide emission, nitrogen oxide emission and dust emission, E m is the emission index parameter, is the maximum emission limit, ω4 is the emergency procurement indication weight, I Y is the value of the emergency procurement indication function, used to indicate whether it is an emergency procurement indication judgment; Step S44: disturbance injection improvement, specifically, by introducing a disturbance factor control function, disturbance injection improvement is performed to obtain a disturbance injection function, the calculation formula of which is: roll(t) = 1 - e -gt ; Where roll(·) is the disturbance injection function, t is the time index, and g is the disturbance control parameter, which is used to control the degree of disturbance injection. By controlling the value of the disturbance injection function, disturbance simulation is performed, including market price fluctuation disturbance, transportation congestion disturbance, and supply chain disruption disturbance. The specific calculation formula for the disturbance of market price fluctuations is where is the historical statistical parameter of market volatility; the specific calculation formula for the transportation congestion disturbance is 0.1roll(t); the specific calculation formula for the supply chain interruption disturbance is Bernoulli(p = 0.01roll(t)), where Bernoulli(·) is the Bernoulli distribution function; Step S45: Procurement strategy optimization, specifically, performing phased optimization of procurement strategy optimization through the state parameter space improvement, the multi-layer strategy network modeling, the multi-objective reward function improvement and the disturbance injection improvement, to obtain the optimal procurement strategy reference data; The optimal procurement strategy reference data includes a multi-objective procurement strategy vector, a strategy enhancement reference function, and an emergency procurement plan reference data; The phased optimization includes an initial stabilization phase, a disturbance introduction phase, and a multi-objective optimization phase; The initial stabilization stage is specifically to optimize the standard reinforcement learning procurement strategy based on the state parameter space improvement and multi-layer strategy network modeling to obtain an initial procurement plan; The disturbance introduction stage is specifically to introduce price fluctuations, transportation congestion and supply interruption simulations based on the disturbance injection improvement on the basis of the initial procurement plan, and optimize the procurement strategy for disturbance introduction improvement to obtain an anti-disturbance procurement plan; The multi-objective optimization stage is specifically to optimize the procurement strategy of multi-objective optimization based on the anti-disturbance procurement plan and the multi-objective reward function, so as to obtain the reference data of the optimal procurement strategy.

7. The procurement efficiency optimization management method based on artificial intelligence according to claim 6, characterized in that: In step S5, the procurement efficiency optimization is used to construct an intelligent evaluation method to verify the optimization of procurement strategies and improve procurement efficiency. Specifically, based on the optimal procurement strategy reference data, a multi-dimensional procurement strategy evaluation index is constructed and the procurement strategy effect evaluation is performed. By selecting the procurement strategy with the best evaluation effect, procurement efficiency is optimized to obtain a comprehensive optimization reference plan for procurement efficiency. The multi-dimensional procurement strategy evaluation indicators include economic indicators, reliability indicators, environmental indicators and agility indicators; The economic indicator specifically refers to the deviation rate between the actual cost and the budgeted cost; the reliability indicator specifically refers to the emergency procurement frequency; the environmental protection indicator specifically refers to the emission quota utilization rate; the agility indicator specifically refers to the time taken for the procurement strategy to respond.

8. An artificial intelligence-based procurement efficiency optimization management system for implementing the artificial intelligence-based procurement efficiency optimization management method according to any one of claims 1-7, characterized in that: It includes a data source processing module, a supply relationship modeling module, a procurement strategy optimization module, and a comprehensive efficiency optimization module.

9. The procurement efficiency optimization management system based on artificial intelligence according to claim 8, characterized in that: The data source processing module is used for itemized data collection and data preprocessing. Through itemized data collection and data preprocessing, procurement efficiency optimization data is obtained, and the procurement efficiency optimization data is sent to the supply relationship modeling module and the procurement strategy optimization module. The supply relationship modeling module is used for supply chain relationship modeling. Through supply chain relationship modeling, supply chain relationship analysis reference data is obtained, and the supply chain relationship analysis reference data is sent to the procurement strategy optimization module. The procurement strategy optimization module is used for procurement strategy optimization. Through procurement strategy optimization, procurement optimal strategy reference data is obtained, and the procurement optimal strategy reference data is sent to the comprehensive efficiency optimization module. The comprehensive efficiency optimization module is used for procurement efficiency optimization. Through procurement efficiency optimization, a comprehensive optimization reference plan for procurement efficiency is obtained.

Images