Enterprise industry chain multi-dimensional evaluation system and method based on artificial intelligence

Abstract

The invention relates to the field of enterprise industry chain evaluation, and discloses an artificial intelligence-based enterprise industry chain multi-dimensional evaluation system and method, and the system comprises a data collection module which is used for obtaining industry chain data from a structured database, an unstructured text and a third-party platform; the data fusion module is used for constructing a dynamically updated industrial chain knowledge graph through a natural language processing technology; the multi-dimensional evaluation module comprises a financial health evaluation sub-module, a supply chain toughness evaluation sub-module, an innovation power evaluation sub-module and a carbon emission evaluation sub-module, and each sub-module generates quantitative scores by adopting an LSTM neural network, a PageRank algorithm and a space-time diagram convolutional network model and weights the quantitative scores to obtain comprehensive scores; the risk early warning module is used for simulating a triple conduction effect of an external event on an industrial chain based on a graph neural network and generating a risk index and a coping strategy; and the visual interaction interface dynamically displays the evaluation result. According to the method, the problems of data splitting and poor dynamic adaptability in a traditional evaluation method are solved, and accurate evaluation of the industrial chain is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of enterprise industrial chain evaluation, specifically to an enterprise industrial chain multi-dimensional evaluation system and method based on artificial intelligence. BACKGROUND

[0002] Enterprise industrial chain evaluation is an important basis for investment decision-making, risk management, and policy-making. However, existing industrial chain evaluation techniques still have many limitations, resulting in evaluation results that are not comprehensive, accurate, and real-time.

[0003] For example: single data source, lack of multi-modal fusion; because traditional industrial chain evaluation systems mainly rely on structured data such as enterprise financial statements and supply chain transaction records, some systems introduce patent data as a supplement; however, financial data only reflects the historical performance of the enterprise and cannot capture real-time operational status such as capacity utilization and supply chain disruption risks, so structured data has limitations. There is also an error in the evaluation of emerging technology fields; because technology leadership evaluation is usually based on the number of patent citations or a simple PageRank algorithm, new materials, artificial intelligence, and other technologies often span multiple industrial chains, such as graphene being used in both energy and electronics, but traditional methods rely on a single classification label, resulting in evaluation bias, such as misclassification of cross-border technologies.

[0004] Therefore, there is an urgent need for an enterprise industrial chain multi-dimensional evaluation system and method based on artificial intelligence. SUMMARY

[0005] The purpose of the present application is to provide an enterprise industrial chain multi-dimensional evaluation system and method based on artificial intelligence, which solves the above technical problems.

[0006] The purpose of the present application can be achieved through the following technical solutions: An enterprise industrial chain multi-dimensional evaluation system based on artificial intelligence, comprising: A data acquisition module for acquiring enterprise industrial chain related data from structured databases, unstructured text, and third-party data platforms; A data fusion module that extracts entity relationships in text through natural language processing techniques and constructs a dynamically updated industrial chain knowledge graph; A multi-dimensional evaluation module, including a financial health evaluation sub-module, a supply chain resilience evaluation sub-module, an innovation evaluation sub-module, and a carbon emission evaluation sub-module, each sub-module generates a quantitative score using an evaluation model, and then obtains a comprehensive score for the enterprise by weighted summation; A risk warning module that simulates the transmission of external events to the industrial chain in real time based on graph neural networks; A visual interactive interface that dynamically displays industrial chain evaluation results and risk traceability paths.

[0007] As a further technical solution, the data acquisition module specifically includes: a structured data unit connected to enterprise financial report databases, patent databases, and supply chain transaction record libraries; an unstructured data unit that collects news public opinion and social media data through crawler technology and performs sentiment analysis using a BERT model; a cross-modal data unit that integrates satellite remote sensing images and financial report data to cross-verify the actual operating status of the enterprise.

[0008] As a further technical solution, the multi-dimensional evaluation module includes: the financial health evaluation submodule uses an LSTM neural network to predict enterprise cash flow and outputs interpretable SHAP values; the supply chain resilience evaluation submodule generates a resilience score by calculating node concentration indicators and alternative supplier coverage indicators; the innovation evaluation submodule quantifies technological leadership based on the influence diffusion algorithm of the patent citation network; the carbon emission evaluation submodule calculates the carbon footprint distribution through a spatio-temporal graph convolution network based on enterprise production energy consumption data and logistics trajectory data, and generates a carbon emission evaluation value in combination with industry benchmark data.

[0009] As a further technical solution, the risk early warning module includes: an event transmission simulation unit that constructs an industry chain dependency matrix based on a dynamic graph neural network, and calculates the triple transmission effects of external shock events through supplier networks, financial chain networks, and technology dependency networks in real time; a risk quantification unit that obtains an enterprise comprehensive risk index by weighted summation of the following indicators: a node vulnerability index based on a knowledge graph to calculate the feature vector centrality and k-core decomposition value of the enterprise node; a transmission time delay index that uses time series graph embedding technology to predict the time threshold of risk transmission to the target enterprise and normalizes the result; an impact intensity index obtained through an impact degree prediction model; an emergency deduction unit that selects pre-set response strategies according to the risk level of the enterprise after comprehensive risk index evaluation; the risk level includes high risk, medium risk, and low risk.

[0010] As a further technical solution, the construction method of the knowledge graph includes: establishing a three-layer topological network of enterprises-suppliers-customers through joint entity recognition and relationship extraction technology; updating equity changes and supply chain contract changes every hour; embedding industry classification codes to enable cross-industry chain comparative analysis.

[0011] As a further technical solution, the node concentration index is calculated by the Hefen-Dal-Hirschman index, and the formula is , is the transaction amount of the th supplier, is the total procurement amount of the enterprise; The alternative supplier coverage index is: the ratio of the number of alternative suppliers with equivalent capacity to the number of core material species, that is, coverage = total number of alternative suppliers / number of core material species.

[0012] As a further technical solution, the influence diffusion algorithm is specifically: Take the number of patent citations as the initial weight, and calculate the influence score of each patent node by PageRank algorithm iteration ; The iteration formula is: ; Wherein, is the damping coefficient, is the set of all patents cited by the patent , is the total number of citations of the patent , is the patent node currently being evaluated, is the patent in the set , is the influence score of the patent , is the total number of patents.

[0013] As a further technical solution, the innovation evaluation sub-module adopts the following way for the technical positioning problem of the new material industry: S1, dynamic industry chain mapping model: Construct a ternary knowledge graph of patent-technology-industry, determine the industry chain membership degree of the patent by calculating the semantic similarity between the patent and the technology node, and combining the technology-industry correlation weight obtained by training the historical technology transfer data; Wherein the semantic similarity is calculated by using a pre-trained language model, and the minimum threshold value is set for the number of matched technology nodes; S2, cross-field influence correction: When it is detected that the patent has multi-industry chain characteristics, based on the influence score of each patent node, combined with the entropy value of its technology field distribution, a new material industry specific adjustment coefficient is introduced to correct the score of the cross-border patent; S3, rapid positioning optimization: A real-time industry classification system is established, the patent technology features are extracted, and similarity comparison is performed with the industry chain feature library, special processing procedures are started for patents with a matching degree lower than a set threshold, including automatic generation of temporary nodes, expert verification and knowledge graph updating.

[0014] A multi-dimensional evaluation method for an enterprise industry chain based on artificial intelligence, which is implemented based on the multi-dimensional evaluation system for an enterprise industry chain based on artificial intelligence.

[0015] The beneficial effects of the present application are: (1) The present application breaks through the limitations of data fragmentation and static in traditional industry chain evaluation by multi-source data fusion and dynamic knowledge graph construction, not only integrates structured financial and transaction data, unstructured public opinion information, but also includes cross-modal satellite remote sensing data, extracts entity relationships through natural language processing technology and constructs a real-time updated industry chain knowledge graph, realizes the deep integration of data from dispersion to correlation, and links the multi-dimensional evaluation module with financial health, supply chain resilience, innovation and carbon emission dimensions, generates quantitative scores combined with artificial intelligence models, solves the one-sidedness problem of traditional evaluation relying on single data or static indicators, makes the evaluation results more in line with the actual operation of the industry chain, and provides a comprehensive and real-time reference for enterprise decision-making.

[0016] (2) The present application improves the accuracy and forward-looking nature of industry chain risk early warning through triple network conduction simulation and multi-index risk quantification, the risk early warning module based on dynamic graph neural network can capture the conduction effect of external events in the supplier network, the capital chain network and the technology dependence network in real time, breaking through the limitations of traditional risk assessment which only focuses on the influence of a single link; through multi-dimensional weighting of node vulnerability, conduction time delay and impact intensity, a comprehensive risk index is generated and matched with a graded response strategy, which not only reveals the path and intensity of risk conduction, but also provides targeted solutions, solving the problem of lack of dynamic conduction simulation and precise strategy support in traditional early warning, effectively enhancing the initiative of enterprises in responding to industry chain risks.

[0017] (3) The present application solves the problem of inaccurate evaluation in emerging technology fields by cross-domain adaptive design of the innovation evaluation sub-module, realizes the accurate quantification of multi-industry chain patent technology, constructs a patent-technology-industry ternary knowledge graph, and determines the patent membership degree combined with semantic similarity and technology correlation weight; for cross-border patents, introduce the field intersection coefficient and entropy value dynamic correction influence score, and process low matching degree patents through real-time classification optimization mechanism, solve the cross-border technology misjudgment problem caused by the dependence of traditional methods on single classification label, especially suitable for the field of new materials spanning multiple industry chains, more accurately quantify the technology leadership, and provide a scientific framework for the technology evaluation of emerging industries. BRIEF DESCRIPTION OF DRAWINGS

[0018] The application will be further described below with reference to the drawings.

[0019] Figure 1 The system logic diagram of the application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the application will be apparently and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the application.

[0021] Please refer to Figure 1 The application is an enterprise industry chain multi-dimensional evaluation system based on artificial intelligence, which comprises: A data acquisition module is configured to acquire enterprise industry chain related data from a structured database, unstructured text and a third-party data platform. A data fusion module is configured to extract entity relationships in the text through natural language processing technology and construct a dynamically updated industry chain knowledge graph. The entity and relationship types include: entities include enterprises, suppliers, customers, patents, technologies and materials; relationships include transactions (such as enterprise A purchasing material C from supplier B), equity (such as enterprise A holding 30% of enterprise B), patent citation (such as patent P1 citing patent P2), technology dependence (such as enterprise A relying on technology T1 to produce products). A multi-dimensional evaluation module comprises a financial health evaluation sub-module, a supply chain resilience evaluation sub-module, an innovation evaluation sub-module and a carbon emission evaluation sub-module. Each sub-module generates a quantitative score using an evaluation model, and then obtains a comprehensive score of the enterprise by weighted summation. A risk early warning module is configured to simulate the conduction influence of external events on the industry chain in real time based on a graph neural network. A visual interactive interface is configured to dynamically display the industry chain evaluation results and the risk traceability path.

[0022] In one embodiment, the dynamic display content comprises: an industry chain knowledge graph, the node size is weighted according to the comprehensive score, and the color is marked according to the risk level: red = high risk, yellow = medium risk, green = low risk; a risk traceability path, the arrow thickness represents the conduction strength, and the arrow color gradient represents the conduction time; a sub-module score trend chart, the data of the last 12 months, supporting clicking to view details.

[0023] The data acquisition module specifically comprises: Structured data unit, connecting enterprise financial report database, patent database and supply chain transaction record database; Specifically including enterprise financial report database (fields: revenue, net profit, asset-liability ratio, operating cash flow), patent database (fields: patent number, application date, publication date, abstract, claims, cited patent list), supply chain transaction record database (fields: transaction parties enterprise ID, transaction amount, transaction time, material type, delivery period); Unstructured data unit, collecting news public opinion and social media data through crawler technology, and performing sentiment analysis using BERT model; BERT model sentiment analysis process: text segmentation → word vector encoding → sentiment polarity classification (positive / negative / neutral, classification threshold: positive ≥ 0.6, negative ≤ 0.3, neutral for neutral); Cross-modal data unit, integrating satellite remote sensing images and financial report data to cross-verify enterprise actual operation status. Satellite remote sensing images are used to verify the actual operation status of the enterprise, for example: through the night light intensity of the factory area (reflecting the production rate), the parking density of the logistics park (reflecting the transportation activity), and the cross comparison with the capacity utilization rate and the revenue growth rate in the financial report, if the deviation exceeds the threshold value such as 15%, the data verification mechanism is triggered.

[0024] In the multi-dimensional evaluation module: The financial health evaluation sub-module uses LSTM neural network to predict enterprise cash flow and outputs interpretable SHAP value; The input features of the long short-term memory network LSTM neural network are: past 36 months of monthly data, such as monthly cash flow, accounts receivable turnover rate, inventory turnover rate, etc., and the output is the cash flow prediction value of the next 12 months; SHAP value is used to explain the prediction result; For example: if the SHAP value of accounts receivable turnover rate is positive and has the largest absolute value, it means that this indicator has the highest contribution to cash flow growth; The supply chain resilience evaluation sub-module generates resilience score by calculating node concentration index and alternative supplier coverage index; The innovation evaluation sub-module quantifies technological leadership based on influence diffusion algorithm of patent citation network; The carbon emission evaluation sub-module calculates carbon footprint distribution based on enterprise production energy consumption data and logistics trajectory data through spatio-temporal graph convolution network, and generates carbon emission evaluation value combined with industry benchmark data.

[0025] In one embodiment, the input data of the spatio-temporal graph convolution network (ST-GCN) includes: production energy consumption (monthly electricity / water / coal consumption of each factory); logistics trajectory (GPS positioning, load, fuel type of transportation vehicle); The model output is the carbon footprint of each link of the enterprise (production link / transportation link / storage link).

[0026] In one embodiment, the specific acquisition process of the carbon emission evaluation value is: The industry benchmark data is derived from the annual carbon emission average of the sub-industry published by the National Bureau of Statistics (divided into three levels of large / middle / small according to the enterprise revenue size), and the carbon emission evaluation value is calculated by the ratio of the enterprise carbon footprint to the benchmark value of the same scale industry; the evaluation value = enterprise carbon footprint / industry benchmark value, the ratio > 1.2 is over standard, 0.8-1.2 is up to standard, and <0.8 is excellent. The enterprise size division of the industry benchmark data: large enterprises (revenue ≥ 2 billion yuan), medium-sized enterprises (2 billion ≤ revenue < 20 billion), and small enterprises (revenue < 2 billion).

[0027] The risk early warning module comprises: An event transmission simulation unit constructs an industry chain dependence relation matrix based on a dynamic graph neural network, and calculates in real time the triple transmission effects of external impact events through a supplier network, a capital chain network and a technology dependence network; As one embodiment, the supplier network transmission effect is quantified by the supplier interruption probability x transaction dependence (transaction amount proportion); wherein the supplier interruption probability = calculated based on the historical interruption times / total cooperation times, and the transaction dependence = the transaction amount of the supplier / total procurement amount of the enterprise; The capital chain network transmission effect is quantified by the accounts payable default probability x capital chain path length inverse; the path length refers to the number of capital flow transfer loops from the risk source enterprise to the target enterprise, such as A→B→C, and the path length of C to A is 2; The technology dependence network transmission effect is quantified by the core technology supply interruption probability x patent cross-dependence (co-citation coefficient); the patent cross-dependence = the number of co-cited patents of the two enterprises / √(total number of patents cited by enterprise A x total number of patents cited by enterprise B); The triple transmission effect is the weighted sum of the above three effects; the weights are 0.4, 0.3 and 0.3 respectively, and are calibrated based on the historical risk transmission contribution degree; A risk quantification unit obtains an enterprise comprehensive risk index by weighted sum of the following indexes: A node vulnerability index is obtained based on the feature vector center degree and k-core decomposition value of the enterprise node calculated by the knowledge graph; In one embodiment, the node vulnerability index = 0.5 x feature vector center degree + 0.5 x k-core decomposition value, the higher the feature vector center degree, the more core the node is in the network; the higher the k-core decomposition value, the more core the node is in the network, both of which are normalized to [0, 1]; A transmission delay index is obtained by predicting the time threshold of risk transmission to the target enterprise by using time sequence diagram embedding technology and normalizing the processing; For example, the upstream core supplier of the target automobile enterprise stops work due to the epidemic: Historical data shows that the transmission time of the shutdown event of the automotive industry supplier ranges from =2 days, =14 days; The timing chart embedding model predicts that the risk will be transmitted from the supplier to the enterprise in 5 days; Normalization calculation: =(5−2) / (14−2)=3 / 12=0.25, that is, the transmission delay index is 0.25; It means that the risk transmission is fast, and the response needs to be started within 2-5 days.

[0028] Impact intensity index: obtained by the impact degree prediction model; In one embodiment, the impact degree prediction model is a gradient boosting regression tree, the input features include impact event types such as natural disasters / policy changes, initial impact range of the event, network distance between the enterprise and the event source, and the output value is a quantitative impact score of 0-100, 100 being the maximum impact, which is normalized (current quantitative impact score / industry maximum possible impact value) and used as the impact intensity index; The emergency deduction unit selects the pre-set response strategy according to the risk level of the enterprise after the comprehensive risk index assessment; the risk level includes high risk, medium risk and low risk.

[0029] In one embodiment, the corresponding relationship between the risk level and the response strategy is: High risk (enterprise comprehensive risk index> 0.7): start emergency supply chain plan (such as enable backup supplier, freeze non-core expenditure), trigger top management decision-making process; Medium risk (0.3< enterprise comprehensive risk index≤0.7): strengthen key node monitoring (such as daily update of supplier inventory), start capital reserve warning; Low risk (enterprise comprehensive risk index≤0.3): push risk prompt report, optimize supplier cooperation terms; The response strategy is generated based on the reinforcement learning model of historical risk cases, and manual adjustment is supported.

[0030] The construction method of the knowledge graph includes: Through the joint entity recognition and relationship extraction technology, a three-layer topological network of enterprise-supplier-customer is established; the joint annotation model is BERT+pointer network model; Input layer: word segmentation and coding of text; such as encoding the procurement agreement into a vector through BERT to capture the semantic association between procurement and supplier-enterprise; Entity recognition layer: the entity boundary is marked by the start pointer and end pointer of the pointer network; such as marking the start position and end position of Shenzhen Electronic Component Factory, and classifying it as a supplier; Relation extraction layer: share the semantic vector of BERT, and classify the identified entity pairs; such as Shanghai Longying and Shenzhen Electronic Component Factory; output procurement relationship, cooperation relationship and other labels, and extract relationship details such as transaction frequency and material type through an additional classifier; Update equity change and supply chain contract change data every hour; scan equity change announcements and supply chain contract systems every hour, and if new / terminated contracts or equity changes ≥5% are detected, automatically update the node attributes and relationship weights of the knowledge graph; Embed industry classification codes to enable cross-industry chain comparison and analysis.

[0031] The node concentration index is calculated using the Herfindahl-Hirschman index, with the formula , is the transaction amount of the th supplier, is the total procurement amount of the enterprise; The supplier concentration index is used to measure the concentration of suppliers, which is essentially the sum of market shares squared; if the enterprise relies on only one supplier (n = 1), then , the concentration is the highest (the resilience is the worst); if tends to infinity and each supplier share is equal, then ≈0, the concentration is the lowest (the resilience is the strongest); The alternative supplier coverage index is: the ratio of the number of alternative suppliers with equivalent capacity to the number of core material types, i.e. coverage = total number of alternative suppliers / core material type number.

[0032] After standardizing the node concentration index and alternative supplier coverage index, the dependency risk score = 50×(1− ); the coping capacity score = 50× (coverage, 50); the resilience score = dependency risk score + coping capacity score.

[0033] In one embodiment, the definition criteria for alternative suppliers with equivalent capacity are: the average capacity in the past three years ≥80% of the target supplier; the quality inspection pass rate of the same material ≥95%, consistent with the target supplier); the delivery cycle deviation ≤3 days.

[0034] The influence diffusion algorithm is specifically: Using the number of patent citations as the initial weight, the influence score of each patent node is calculated by iterating the PageRank algorithm ; The iteration formula is: ; wherein, ​is the damping coefficient, usually takes 0.85, indicating the probability of the user continuing to browse the next patent, is the total number of patents cited by , is the total number of patents cited by , is the patent node currently being evaluated, is the patent in the set , is the influence score of the patent , is the total number of patents. Because the higher the influence of a patent, the higher the influence of the patents citing it and the more patents citing it, the higher the score of the patent; by aggregating the scores of all patents under the enterprise, the aggregated way can be selected by the conventional weighted sum, which can quantify the technological leadership of the enterprise in the industry chain.

[0035] The core logic of the iterative formula is that the patents cited by high-influence patents have higher influence themselves; the initial weight: the more times a patent is cited, the higher the initial ; iterative logic: the influence of each patent is evenly distributed to the patents it cites, i.e. , is the total number of citations of ; the damping coefficient , by default 0.85: indicates the probability of random citation of a patent, is the base influence of each patent, avoiding the influence of uncited patents being 0.

[0036] The innovation evaluation submodule adopts the following methods for the technical positioning problem of the new material industry: S1, dynamic industry chain mapping model: A ternary knowledge graph of patent-technology-industry is constructed, the industry chain membership degree of a patent is determined by calculating the semantic similarity between the patent and the technology node and combining the technology-industry association weight trained by historical technology transfer data; the semantic similarity is calculated by a pre-trained language model, and the minimum threshold is set for the number of matched technology nodes; S2, cross-field influence correction: When it is detected that a patent has multi-industry chain characteristics, based on the influence score of each patent node, the entropy value of its technology field distribution is combined for dynamic adjustment, and a new material industry specific adjustment coefficient is introduced to correct the score of the cross-border patent; S3, rapid positioning optimization: Establish a real-time industry classification system. By extracting patent technology features and comparing them with the industry chain feature database, special processing procedures are initiated for patents with a matching degree lower than a set threshold, including automatic generation of temporary nodes, expert verification, and knowledge graph updates.

[0037] The innovation assessment submodule addresses the technological positioning issue in the new materials industry by employing the following improved algorithm: S1, Dynamic Supply Chain Mapping Model: Construct a knowledge graph of the patent-technology-industry ternary set; Calculate new material patents using the following formula. Supply chain affiliation : ; in: Patent With technology nodes semantic similarity, through The model calculations yielded the results. Indicates technology node With the industrial chain The association weights are obtained through training on historical technology transfer data; the above formula is an extension of the weighted cosine similarity, used to calculate patent... With the industrial chain The membership degree, ranging from [-1, 1], with higher values ​​indicating stronger membership; the numerator term... , indicating patent With technology nodes semantic similarity Technology-industry correlation weight The weighted sum reflects the overall strength of the patent's indirect connection to the industrial chain through technology; denominator This is a normalization term to avoid result bias caused by an excessive number of technical nodes; The threshold for the number of technical nodes. ≥5; Ensure that at least 5 technical nodes support membership determination to avoid the random impact of a single technical node; S2, Cross-Domain Influence Correction: When a patent is detected to simultaneously satisfy multiple industry chain characteristics, a domain cross-correlation coefficient is introduced to correct the original PageRank score: This formula is used to correct the influence of cross-industry patents. The core principle is that the stronger the cross-domain nature, the greater the score correction. in: Patent The original PageRank score calculated by the PageRank algorithm; The new material industry adjustment factor, the default value is 0.3; for controlling the amplitude of cross-domain correction, to avoid excessive amplification of the influence of cross-domain patents; The Shannon entropy representing the distribution of patent technology fields; for calculating the degree of disorder of the distribution of patent technology fields, the formula is The probability that the patent belongs to the field The higher the entropy value, the stronger the cross-domain nature of the patent, such as patents involving both materials and electronics having a higher entropy value than single-domain patents; S3, rapid positioning optimization: Establish a new material technology radar chart, which is updated in real time by the following steps: Extract the technology keyword set K in the patent claim; Calculate the Jaccard similarity with each industry chain technology feature library; When the maximum Jaccard similarity is less than the threshold value θ, perform the following operations: Automatically generate a temporary industry chain node; Start the expert verification process; Update the knowledge graph based on the verification results.

[0038] An enterprise industry chain multi-dimensional evaluation method based on artificial intelligence, which is implemented based on the enterprise industry chain multi-dimensional evaluation system based on artificial intelligence.

[0039] It should be noted that the calculation formula and the parameters involved in the calculation in the present application are all pre-processed by dimensionless processing, and the process of dimensionless processing is known in the industry and will not be described here.

[0040] The above has described one embodiment of the present application in detail, but the content described is only the preferred embodiment of the present application, and cannot be considered as limiting the scope of the implementation of the present application. Any equivalent changes and improvements made within the scope of the present application shall still belong to the scope of the patent coverage of the present application.

Claims

1. A multi-dimensional enterprise supply chain evaluation system based on artificial intelligence, characterized in that, include: The data acquisition module is used to obtain enterprise supply chain-related data from structured databases, unstructured text, and third-party data platforms. The data fusion module extracts entity relationships from text using natural language processing technology and constructs a dynamically updated industry chain knowledge graph. The multi-dimensional assessment module includes a financial health assessment sub-module, a supply chain resilience assessment sub-module, an innovation assessment sub-module, and a carbon emission assessment sub-module. Each sub-module uses an assessment model to generate a quantitative score, which is then weighted and summed to obtain the company's comprehensive score. The risk warning module uses graph neural networks to simulate the transmission impact of external events on the industrial chain in real time. A visual interactive interface dynamically displays the results of the supply chain assessment and the path to trace the source of risks.

2. The multi-dimensional enterprise supply chain evaluation system based on artificial intelligence according to claim 1, characterized in that, The data acquisition module specifically includes: Structured data units connect corporate financial statement databases, patent databases, and supply chain transaction record databases; Unstructured data units are collected from news sentiment and social media data through web crawling technology, and sentiment analysis is performed using the BERT model; Cross-modal data units integrate satellite remote sensing images with financial statement data to cross-validate the actual operational status of enterprises.

3. The multi-dimensional enterprise supply chain evaluation system based on artificial intelligence according to claim 2, characterized in that, In the multi-dimensional evaluation module: The financial health assessment submodule uses an LSTM neural network to predict corporate cash flow and outputs an interpretable SHAP value. The supply chain resilience assessment submodule generates a resilience score by calculating the node concentration index and the alternative supplier coverage index. The innovation assessment submodule quantifies the leading edge of technology based on the influence diffusion algorithm of the patent citation network. The carbon emission assessment submodule calculates the carbon footprint distribution based on enterprise production energy consumption data and logistics trajectory data, and generates carbon emission assessment values ​​by combining industry benchmark data.

4. The multi-dimensional enterprise supply chain evaluation system based on artificial intelligence according to claim 3, characterized in that, The risk warning module includes: The event transmission simulation unit constructs a supply chain dependency matrix based on a dynamic graph neural network, and calculates in real time the triple transmission effect of external shock events through the supplier network, the capital chain network, and the technology dependence network. The risk quantification unit calculates the enterprise's overall risk index by weighting and summing the following indicators: The node vulnerability index is derived by calculating the feature vector centrality and k-kernel decomposition value of enterprise nodes based on the knowledge graph. Transmission Delay Index: Calculated by predicting the time threshold for risk transmission to the target enterprise using time series graph embedding technology and then normalizing the result. Impact Intensity Index: Derived from an impact degree prediction model; The emergency simulation unit selects a pre-set response strategy based on the risk level assessed by the enterprise's comprehensive risk index; the risk levels include high risk, medium risk, and low risk.

5. The multi-dimensional enterprise supply chain evaluation system based on artificial intelligence according to claim 4, characterized in that, The methods for constructing the knowledge graph include: By combining entity recognition and relationship extraction technologies, a three-layer topology network of enterprise-supplier-customer is established; Data on equity changes and supply chain contract changes is updated hourly. Embed industry classification codes to enable cross-industry chain comparative analysis.

6. The multi-dimensional enterprise supply chain evaluation system based on artificial intelligence according to claim 3, characterized in that, The node concentration index is calculated using the Herfindahl-Hirschman index, with the following formula: , For the first The transaction amount of each supplier This represents the company's total procurement amount; The alternative supplier coverage index is the ratio of the number of alternative suppliers with equivalent production capacity to the number of core material types, i.e., coverage = total number of alternative suppliers / number of core material types.

7. The multi-dimensional enterprise supply chain evaluation system based on artificial intelligence according to claim 3, characterized in that, The influence diffusion algorithm is specifically as follows: Using the number of patent citations as the initial weight, the influence score of each patent node is iteratively calculated using the PageRank algorithm. ; The iterative formula is: ; in, The damping coefficient is... For citing patents All patent collections, For patent Total number of citations, For the current patent node being evaluated, For set Patents in China For patent Influence score, This represents the total number of patents.

8. The multi-dimensional enterprise supply chain evaluation system based on artificial intelligence according to claim 7, characterized in that, The innovation assessment submodule addresses the technological positioning of the new materials industry using the following methods: S1, Dynamic Supply Chain Mapping Model: A knowledge graph of patent-technology-industry triples is constructed. By calculating the semantic similarity between patents and technology nodes and combining the technology-industry association weights obtained by training with historical technology transfer data, the industrial chain membership of patents is determined. The semantic similarity is calculated using a pre-trained language model, and a minimum threshold is set for the number of matching technology nodes. S2, Cross-Domain Influence Correction: When a patent is detected to have multi-industry chain characteristics, the influence score is based on the influence score of each patent node. The score of cross-border patents is adjusted dynamically by combining the entropy value of their technical field distribution and introducing a specific adjustment coefficient for the new materials industry. S3, Fast Positioning Optimization: Establish a real-time industry classification system. By extracting patent technology features and comparing them with the industry chain feature database, special processing procedures are initiated for patents with a matching degree lower than a set threshold, including automatic generation of temporary nodes, expert verification, and knowledge graph updates.

9. A multi-dimensional evaluation method for enterprise supply chain based on artificial intelligence, characterized in that, This method is implemented based on the AI-based multi-dimensional enterprise supply chain evaluation system described in any one of claims 1-8.

Images