Enterprise intelligent diagnosis method, system and equipment based on large model and medium

Abstract

The invention provides an enterprise intelligent diagnosis method, system and device based on a large model and a medium, and belongs to the technical field of enterprise diagnos.The method comprises the steps that enterprise heterogeneous data are collected through a multi-source data interface, cleaning, feature extraction and cross-modal alignment fusion are carried out, and enterprise real-time data are obtained; constructing a knowledge graph through a graph attention network on the basis of an industry index to which an enterprise belongs, and updating association weights among nodes at regular time; inputting enterprise real-time data into the pre-trained multi-modal large model for preliminary analysis, and outputting a risk thermodynamic diagram; key abnormal indexes are identified from the risk thermodynamic diagram, sub-graphs related to the key abnormal indexes are extracted from the knowledge graph, structured prompt words are generated from the sub-graphs, then the structured prompt words and standardized enterprise real-time data are jointly input into a multi-modal large model for joint reasoning analysis, and then a visual diagnosis report is generated. Accurate identification and intelligent diagnosis of enterprise risks are realized, and decision-making efficiency is improved.

Description

Technical Field

[0001] This application belongs to the field of enterprise diagnosis technology, and specifically relates to an enterprise intelligent diagnosis method, system, device and medium based on a large model. Background Art

[0002] With the acceleration of digital transformation, the data generated during enterprise operation has grown rapidly, covering multiple dimensions such as finance, production, sales, supply chain, and customer interaction. These data usually have the characteristics of multi-source heterogeneity. How to quickly and accurately identify enterprise operation risks from complex-structured data and give executable diagnosis suggestions has become an important requirement for enterprise intelligent decision-making.

[0003] Traditional enterprise diagnosis methods first rely on expert experience to discover problems through manual statistics and report analysis. This method is not only inefficient and difficult to handle massive enterprise data, but also relies on human subjectivity and is prone to missing hidden management. Secondly, traditional diagnosis methods perform risk detection based on predefined rules, cannot adapt to dynamic business changes, and are difficult to handle multi-index collaborative anomalies. One type of machine learning-based enterprise diagnosis method is a single-modal analysis paradigm that separately models specific types of data, which is prone to ignoring cross-modal management and resulting in one-sided diagnosis. On the other hand, the static knowledge graph method that constructs a graph of enterprise entity relationships for auxiliary analysis has a lag in updating node and edge weights and cannot reflect business dynamics in real time.

[0004] In addition, there is currently no unified standardization method for multi-source heterogeneous data in enterprise operation, resulting in feature alignment errors. Traditional knowledge graphs cannot update node weights in real time and are difficult to capture sudden risks. Summary of the Invention

[0005] In a first aspect, an embodiment of this application provides an enterprise intelligent diagnosis method based on a large model, including the following steps: S1. Collect enterprise heterogeneous data through a multi-source data interface, and clean, extract features, and perform cross-modal alignment and fusion on the collected enterprise heterogeneous data to obtain standardized enterprise real-time data; S2. Based on enterprise industry indicators, combined with the standardized enterprise real-time data, construct a knowledge graph with enterprise entities, business indicators, and industry standards as nodes through a graph attention network, and regularly update the association weights between nodes according to the enterprise real-time data; S3. Input the standardized enterprise real-time data into a pre-trained multi-modal large model for preliminary analysis, and output a risk heat map containing abnormal indicators and the probabilities of abnormal indicators; S4. Identify key abnormal indicators from the risk heat map, extract sub-graphs related to the key abnormal indicators from the knowledge graph, and then jointly input the sub-graphs after generating structured prompt words and the standardized enterprise real-time data into the multi-modal large model for joint reasoning analysis; S5. Convert the combined reasoning analysis results into an executable diagnostic report and display it to enterprise users through a visual interaction interface.

[0006] Further, the specific steps of step S1 are as follows: S11. Obtain enterprise financial data, sales data, inventory data, and human resource data through the enterprise ERP system interface, obtain production manufacturing data through the enterprise MES system interface, and obtain customer interaction data through the enterprise CRM system interface; S12. Crawl operation data from each business process of the enterprise through an RPA crawler tool, and obtain data from each sensor through the IoT device protocol interface; S13. Clean the collected data, remove duplicate, incorrect, incomplete, and abnormal data records, and supplement missing values; S14. Perform normalization and standardization processing on the cleaned data to complete dimension unification, and then perform feature screening based on SHAP values; S15. Map the data after feature screening to a unified dimensional space and align it based on timestamps to obtain standardized enterprise real-time data.

[0007] Further, the specific steps of step S2 are as follows: S21. Extract enterprise entity nodes from the standardized enterprise real-time data through a named entity recognition algorithm, load indicator metadata from the industry benchmark library of the enterprise's affiliated industry as business indicator nodes, and extract clause entities and constraint conditions from the enterprise industry standard documents as industry standard nodes to complete node initialization; S22. Construct a knowledge graph through a graph attention network: Perform GRU encoding on the standardized enterprise real-time data to generate feature vectors corresponding to the nodes; Set the initial value of the edge weights between nodes according to industry rules; S23. Dynamically update the nodes and edge weights of the knowledge graph when the update condition is triggered.

[0008] Further, the specific steps of step S23 are as follows: S231. When a new enterprise entity or abnormal indicator is detected in the standardized enterprise real-time data, automatically create a node for the knowledge graph and initialize the embedding vector of the node:

[0009] Among them, is the entity feature, is the embedding of similar nodes; S232. For existing nodes in the knowledge graph, if there is no associated data in the standardized real-time enterprise data for N consecutive days, they are marked as dormant nodes, and the dormant nodes are stored offline; S233. Update the knowledge graph according to the set update period, and use the temporal graph convolutional network to adjust the edge weights between nodes:

[0010] Among them, is the embedding of node i at time t, is the amount of data change between nodes.

[0011] Furthermore, the specific steps of step S3 are as follows: S31. Divide the standardized real-time enterprise data into structured data, text data, and temporal data; S32. Map the categorical features in the structured data into dense vectors of a set dimension through an embedding algorithm, and extract local pattern features from the numerical features in the structured data using a 1D convolutional neural network; S33. Divide the text data into tabular text and plain text, process the tabular text using the LayoutXLM model to obtain the table structure and content semantics, and use the RoBERTa-wwm model to extract the text semantic features from the plain text, and extract the token vectors from the Chinese text; S34. Use the Informer model to capture long-term dependencies for the sensor temporal data in the temporal data, and use the N-HiTS model to decompose the trend term and periodic term features for the business temporal data in the temporal data; S35. Calculate the correlation weights between modalities for the features extracted from the structured data, text data, and temporal data through the cross-modal attention mechanism algorithm:

[0012] Among them, is the query vector of modality i, is the key vector of modality j, is the dimensionality scaling factor; S36. Use the gated fusion algorithm to generate joint features for the feature vectors of the structured data, text data, and temporal data:

[0013]

[0014] Among them, represents the gating value, σ(*) is the activation function, is the trainable parameter matrix, is the structured data feature vector, is the feature vector of text data, is the feature vector of time series data, is the cross-modal feature vector after fusion; S37. Calculate the anomaly probability of each business metric through a multi-layer perceptron model:

[0015]

[0016] wherein, is the anomaly probability of the i-th metric, , is the weight matrix of the first layer of the multi-layer perceptron model, is the weight matrix of the second layer of the multi-layer perceptron model, is the bias vector of the first layer of the multi-layer perceptron model, is the bias value of the second layer of the multi-layer perceptron model, d is the dimension of the input features, m is the number of hidden units in the first layer, and 1 is the dimension of the output; Determine the abnormal business metrics according to the anomaly probabilities of each business metric, and generate an anomaly probability vector from the abnormal business metrics and the corresponding anomaly probabilities; S38. Use a deconvolution network to map the anomaly probability vector to the dimension of the original standardized enterprise real-time data to generate a pixel-level risk heat map; S39. Display the pixel-level risk heat map in layers.

[0017] Furthermore, the specific steps of step S4 are as follows: S41. The specific steps to identify key abnormal metrics from the risk heat map are as follows: Pre-set a dynamic probability threshold, and determine a business metric that meets any of the following conditions as a key abnormal metric: A business metric whose anomaly probability exceeds the dynamic probability threshold; A business metric that affects more than N business departments; The increase rate of the anomaly probability in M consecutive detection cycles exceeds the preset amplitude threshold; S42. The specific steps to extract the sub-graph related to the key abnormal metric from the knowledge graph: Starting from the key abnormal metric node, perform N-layer deep search in the knowledge graph: where the first layer is the node directly connected to the key abnormal metric node, and the subsequent layer is the node connected to the previous layer node; The search priority of the nodes in each layer is calculated by the edge weight and SHAP value:

[0018] wherein, is the weight between node i and node j, SHAP is the feature screening result, which indicates the contribution of the feature to the model output. and is an adjustable weight coefficient used to balance the influence of edge weight and SHAP value; S43. The specific steps of generating structured prompt words from the subgraph are as follows: Pre-build a prompt word template and fill the relevant nodes and relationships in the searched subgraph into the prompt word template, thereby converting the subgraph into a structured prompt word as a natural language description; S44. The specific steps for inputting structured prompt words and standardized real-time enterprise data into a multimodal large model for joint reasoning analysis are as follows: Construct a multimodal large model of hierarchical hybrid coding, including parallel text coding layer, table coding layer, time series coding layer and data fusion layer; The structured prompt words are concatenated with the structured data in the standardized real-time data and then input into the text encoding layer of the multimodal large model to generate a structured query vector. The structured prompt words are concatenated with the standardized text data in real-time data and then input into the table encoding layer of the multimodal large model to generate a text query vector. Input the time series prompt words and the time series data in the standardized real-time data into the time series encoding layer of the multimodal large model to generate a time series query vector; The structured query vector, text query vector, and temporal query vector are input into the data fusion layer, and the joint reasoning analysis results are generated through cross-modal attention fusion and gated feature fusion.

[0019] Furthermore, the specific steps of step S5 are as follows: S51. Use the T5 model to compress the text of the joint reasoning analysis results, generate an executive summary containing key anomaly indicators, root cause analysis, and risk prediction, and highlight key data in the executive summary; S52. Output a hierarchical risk heat map and set up a drill-down interface to connect to standardized enterprise real-time data and knowledge graphs; S53. In response to the user's click operation on the risk heat map, the corresponding standardized enterprise real-time data and related subgraphs in the knowledge graph are displayed in a linked manner, and the similarity ranking results of the related case nodes are displayed.

[0020] In a second aspect, an embodiment of the present application further provides an enterprise intelligent diagnosis system based on a large model, comprising: The data collection and processing module is used to collect heterogeneous enterprise data through multi-source data interfaces, and clean, extract features, and align and fuse the collected heterogeneous enterprise data across modalities to obtain standardized real-time enterprise data; A knowledge graph construction and update module, which is used to construct a knowledge graph with enterprise entities, business indicators, and industry standards as nodes based on industry indicators of an enterprise, combined with standardized real-time enterprise data, through a graph attention network, and regularly update the association weights between nodes according to the real-time enterprise data; A risk heat map generation module, which is used to input the standardized real-time enterprise data into a pre-trained multi-modal large model for preliminary analysis, and output a risk heat map containing abnormal indicators and the probabilities of abnormal indicators; An inference analysis module, which is used to identify key abnormal indicators from the risk heat map, extract a sub-graph related to the key abnormal indicators from the knowledge graph, and then generate structured prompt words from the sub-graph and jointly input them into the multi-modal large model with the standardized real-time enterprise data for joint inference analysis; A diagnosis display module, which is used to convert the joint inference analysis result into an executable diagnosis report and display it to enterprise users through a visual interactive interface.

[0021] Thirdly, an embodiment of the present application further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the enterprise intelligent diagnosis method based on a large model as described in the first aspect are implemented.

[0022] Fourthly, an embodiment of the present application further provides a storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the enterprise intelligent diagnosis method based on a large model as described in the first aspect are implemented.

[0023] From the above technical solutions, it can be seen that the present application has the following advantages: In the enterprise intelligent diagnosis method, system, device, and medium based on a large model provided by the present application, through multi-source data collection and processing, dynamic knowledge graph construction, multi-modal depth analysis, and visual display, the accurate identification, intelligent analysis, and real-time monitoring of enterprise operation risks are realized; not only the accuracy, efficiency, and operability of diagnosis are improved, but also the labor cost and the possibility of risk omission are reduced, providing support for enterprise decision-making and enhancing the digital transformation level of the enterprise. Description of the Drawings

[0024] In order to more clearly illustrate the technical solutions of the present application, the drawings required to be used in the description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0025] Figure 1 It is a schematic flowchart of the enterprise intelligent diagnosis method based on a large model of the present invention.

[0026] Figure 2 This is a schematic diagram of the enterprise intelligent diagnosis system based on large models of the present invention. Detailed implementation manners

[0027] In the following, the specific steps of the enterprise intelligent diagnosis method based on large models will be described in detail, and various embodiments of the present disclosure will be described more comprehensively. The present disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of the present disclosure to the specific embodiments disclosed herein, but the present disclosure should be understood to cover all adjustments, equivalents, and / or alternative solutions that fall within the spirit and scope of the various embodiments of the present disclosure.

[0028] Exemplarily, with the acceleration of the digital transformation process, enterprises have accumulated a large amount of multi-source heterogeneous data in the operation process, covering multiple aspects such as finance, production, sales, supply chain, and customer interaction. Facing these complex data, enterprises urgently need a method that can quickly and accurately identify operation risks and provide practical diagnosis suggestions to meet the needs of intelligent decision-making.

[0029] Traditional diagnosis methods have obvious limitations. On the one hand, the manual statistics and report analysis methods relying on expert experience are not only inefficient, difficult to process large-scale data, but also prone to missing potential problems due to manual subjective judgment. On the other hand, the risk detection methods based on fixed rules lack flexibility and are difficult to adapt to the rapid changes in business. Especially when multiple indicators are abnormally coordinated, they often seem powerless. Existing machine learning methods are usually limited to the analysis of single data types, easily ignoring the associations between cross-modal data, resulting in one-sided diagnosis results. At the same time, due to the untimely update of node and edge weights in the static knowledge graph method, it cannot reflect business dynamics in real time and is difficult to capture sudden risks.

[0030] In addition, when enterprises process multi-source heterogeneous data, they lack a unified standardization method, resulting in errors in feature alignment. Traditional knowledge graph technologies cannot update node weights in real time and are difficult to adapt to the rapid changes in the business environment, thus affecting the timeliness and accuracy of risk identification.

[0031] In view of the above problems, this embodiment provides an enterprise intelligent diagnosis method based on large models, which realizes the accurate identification and intelligent diagnosis of enterprise risks and improves the decision-making efficiency through multi-source data fusion, dynamic knowledge graph, and multi-modal analysis.

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0033] Please refer to Figure 1 The figure shows a flowchart of an enterprise intelligent diagnosis method based on a large model in a specific embodiment. The method includes the following steps: S1. Collect enterprise heterogeneous data through a multi-source data interface, and clean, extract features, and perform cross-modal alignment and fusion on the collected enterprise heterogeneous data to obtain standardized enterprise real-time data; It should be noted that collecting enterprise heterogeneous data through a multi-source data interface, and performing cleaning, feature extraction, and cross-modal alignment and fusion can integrate the data scattered in various enterprise systems into a unified standard format, improve data quality, reduce data dimensions, and provide a data basis for subsequent analysis; S2. Based on the industry indicators of the enterprise, combined with the standardized enterprise real-time data, construct a knowledge graph with enterprise entities, business indicators, and industry standards as nodes through a graph attention network, and regularly update the association weights between nodes according to the enterprise real-time data; It should be noted that based on the industry indicators of the enterprise and the standardized real-time data, constructing a knowledge graph with enterprise entities, business indicators, and industry standards as nodes through a graph attention network can intuitively display the complex relationships and business logics within the enterprise; regularly updating the association weights between nodes according to the enterprise real-time data ensures that the knowledge graph can reflect the changes in the enterprise operation state in real time, providing the latest structured information for subsequent risk analysis and diagnosis; using enterprise entities, business indicators, and industry standards as graph nodes facilitates in-depth analysis of various relationships and influencing factors in enterprise operation, providing semantic information for identifying anomalies and risks; S3. Input the standardized enterprise real-time data into a pre-trained multi-modal large model for preliminary analysis, and output a risk heat map containing abnormal indicators and the probabilities of abnormal indicators; It should be noted that inputting the standardized real-time data into a pre-trained multi-modal large model for preliminary analysis can make full use of the model's processing ability for various data types to identify potential patterns and abnormal information in the data; outputting a risk heat map containing abnormal indicators and the probabilities of abnormal indicators presents the risk distribution in enterprise operation in an intuitive visual way, assisting in quickly locating the problem area; through the comprehensive analysis of multi-modal data, potential abnormal indicators are initially identified, providing a direction for in-depth diagnosis and analysis; S4. Identify key abnormal indicators from the risk heat map, extract the sub-graphs related to the key abnormal indicators from the knowledge graph, and then generate structured prompt words from the sub-graphs and jointly input them into the multi-modal large model together with the standardized real-time enterprise data for joint reasoning and analysis; It should be noted that identifying key abnormal indicators from the risk heat map enables the identification of the problem points that have the greatest impact on enterprise operations, improving the pertinence and efficiency of diagnosis; extracting the sub-graphs related to the key abnormal indicators from the knowledge graph and generating structured prompt words facilitate the in-depth analysis of the background and associated factors of the abnormal indicators, providing rich context information for subsequent reasoning and analysis; jointly inputting the structured prompt words and the standardized real-time data into the multi-modal large model for joint reasoning and analysis combines the data-driven and knowledge-guided methods, enabling a more comprehensive and accurate analysis of the causes and impacts of the abnormal indicators and providing in-depth diagnostic results; S5. Convert the joint reasoning and analysis results into an executable diagnostic report and display it to enterprise users through a visual interaction interface; It should be noted that converting the joint reasoning and analysis results into an executable diagnostic report provides the enterprise with clear and specific improvement suggestions and action plans; displaying the diagnostic report through a visual interaction interface makes complex data and analysis results intuitive, facilitating enterprise users to quickly obtain key information and enhancing the user experience.

[0034] In this embodiment, enterprise heterogeneous data is collected and standardized through a multi-source data interface to improve data quality. A knowledge graph is constructed by combining industry indicators of the enterprise, and the association weights between nodes are dynamically updated to reflect the enterprise operation status in real time. Preliminary analysis is performed through a pre-trained multi-modal large model to generate a risk heat map, visually displaying enterprise abnormal indicators and probabilities. By integrating the structured prompt words generated from the knowledge graph sub-graphs and the standardized real-time data, joint reasoning and analysis are carried out to improve the accuracy and depth of diagnosis. The joint reasoning and analysis results are converted into an executable diagnostic report and displayed through a visual interaction interface, facilitating enterprise users to understand and use the diagnostic results.

[0035] Furthermore, as a refinement and extension of the specific implementation manner of the above embodiment, to fully illustrate the specific implementation process in this embodiment, another enterprise intelligent diagnosis method based on a large model is provided. This method includes the following steps: S1. Collect enterprise heterogeneous data through a multi-source data interface, and perform cleaning, feature extraction, and cross-modal alignment and fusion on the collected enterprise heterogeneous data to obtain standardized real-time enterprise data. The specific steps of step S1 are as follows: S11. Obtain enterprise financial data, sales data, inventory data, and human resources data through the enterprise ERP system interface, obtain production manufacturing data through the enterprise MES system interface, and obtain customer interaction data through the enterprise CRM system interface; Exemplarily, enterprise financial data includes data of core financial indicators such as balance sheets, income statements, and cash flow statements that can show the financial health of the enterprise. Enterprise sales data includes data such as sales volume, sales amount, sales area distribution, and customer purchase behavior of product lines that help the enterprise grasp market dynamics and sales trends; enterprise inventory data includes data such as the quantity, turnover rate, and inventory cost of raw material inventory, work-in-progress inventory, and finished product inventory that help the enterprise optimize inventory management; human resources data includes data such as employee attendance, performance evaluation, staff turnover rate, and human resource allocation of each department; S12. Crawl operation data from each business process of the enterprise through the RPA crawler tool, and obtain data of each sensor through the IoT device protocol interface; Exemplarily, sensor data includes temperature data, humidity data, and pressure data; S13. Clean the collected data, remove duplicate, incorrect, incomplete, and abnormal data records, and supplement missing values; S14. Perform normalization and standardization processing on the cleaned data to complete the unification of measurement units, and then perform feature screening based on SHAP values; Specifically, the normalization processing adopts the Min-Max normalization method to scale the data to the interval [0,1]. The specific formula is as follows:

[0036] The standardization processing adopts the Z-score standardization method to convert the data into a distribution with a mean of 0 and a standard deviation of 1. The specific formula is as follows:

[0037] Among them, μ is the mean of the data, and σ is the standard deviation of the data; The feature screening based on SHAP values is as follows: Take each piece of collected data as a feature, and use SHAP values to analyze the contribution degree of each feature to the output of the pre-trained model; by training a simple machine learning model (such as linear regression or decision tree), calculate the SHAP value of each feature, and sort the features according to the absolute value of the SHAP value, and select the top K features with the greatest contribution to the model output; The specific formula is as follows:

[0038] Among them, is the SHAP value of feature i, is the set of all features, is a subset of certain features, is the feature subset of the model prediction value; S15. Map the data with completed feature screening to a unified dimensional space and align it based on the timestamp to obtain standardized enterprise real-time data; Specifically, map data from different sources (such as financial data, sales data, inventory data, etc.) to a unified multi-dimensional space; for example, map the "sales amount", "cost", "profit" indicators in financial data, the "sales volume", "customer purchase behavior" indicators in sales data, and the "inventory turnover rate", "inventory cost" indicators in inventory data to a space with "time", "business type", and "indicator value" as dimensions; S2. Based on the enterprise's industry indicators, combined with the standardized enterprise real-time data, construct a knowledge graph with enterprise entities, business indicators, and industry standards as nodes through a graph attention network, and regularly update the association weights between nodes according to the enterprise real-time data; the specific steps of step S2 are as follows: S21. Extract enterprise entity nodes from the standardized enterprise real-time data through a named entity recognition algorithm, load indicator metadata from the industry benchmark library of the enterprise's industry as business indicator nodes, and extract clause entities and constraint conditions from the enterprise industry standard documents as industry standard nodes to complete node initialization; Exemplarily, enterprise entity nodes can extract departments, production lines, and equipment numbers obtained from the original ERP system or MES system interface from the standardized enterprise real-time data and generate a unique URI identifier; S22. Construct a knowledge graph through a graph attention network: Perform GRU encoding on the standardized enterprise real-time data to generate feature vectors corresponding to the nodes; Set the initial value of the edge weights between nodes according to industry rules; Exemplarily, the weight of the association between the Finance Department and the Purchasing Department as enterprise entity nodes is set to 0.8; S23. When the update condition is triggered, dynamically update the nodes and edge weights of the knowledge graph; The specific steps of step S23 are as follows: S231. When a new enterprise entity or indicator anomaly is detected in the standardized enterprise real-time data, automatically create a node for the knowledge graph and initialize the embedding vector of the node:

[0039] Among them, is the entity feature, is the embedding of the same type of nodes; S232. If there is no associated data in the standardized real-time enterprise data for N consecutive days for the existing nodes in the knowledge graph, mark them as dormant nodes and store the dormant nodes offline; S233. Update the knowledge graph according to the set update period, and use the temporal graph convolutional network to adjust the edge weights between nodes:

[0040] where is the embedding of node i at time t, is the data change amount between nodes; S3. Input the standardized real-time enterprise data into a pre-trained multi-modal large model for preliminary analysis, and output a risk heat map containing abnormal indicators and the probabilities of abnormal indicators; The specific steps of step S3 are as follows: S31. Divide the standardized real-time enterprise data into structured data, text data, and temporal data; S32. Map the categorical features in the structured data into dense vectors of a set dimension through an embedding algorithm, and extract local pattern features from the numerical features in the structured data using a 1D convolutional neural network; Specifically, the 1D convolutional neural network is a neural network architecture for processing sequential data (such as time series, signals, etc.), which uses a one-dimensional convolutional kernel to slide on the sequence and calculates the dot product of the convolutional kernel and the sequence segment to extract local features; Exemplarily, taking the enterprise sales data (monthly recorded sales amount) as an example, the specific data format is as follows: [0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0] These data have been processed by Min-Max normalization and are in the range of [0,1]; Through the convolution operation of the 1D convolutional neural network, the convolutional kernel slides on the sequence and calculates the dot product of the convolutional kernel and the sequence segment. Taking a convolutional kernel of size 3 and a stride of 1 as an example, the weights of the convolutional kernel can be randomly initialized, for example: [0.2,0.5,0.3] The calculation process of the convolution operation is as follows: The first convolution: 0.1×0.2 + 0.2×0.5 + 0.3×0.3 = 0.23 The second convolution: 0.2×0.2 + 0.3×0.5 + 0.4×0.3 = 0.31 And so on until the entire sequence is covered; Through convolutional operations, 1D CNN can extract local patterns in a sequence; for example, in sales data, convolutional operations can identify local features such as upward trends, downward trends, or periodic changes in sales amounts; S33. Divide the text data into tabular text and plain text. Process the tabular text with the LayoutXLM model to obtain the table structure and content semantics, while use the RoBERTa-wwm model to extract the text semantic features for plain text, and extract the token vectors for Chinese text; It should be noted that the LayoutXLM model is a pre-trained multi-modal model for processing tabular text, which can capture both the structural information and content semantics of the table; Take the tabular text containing sales data shown in Table 1 as an example: Table 1

[0041] Use the LayoutXLM model to capture the row-column relationships of the table 4×3 and the field semantics of product name, sales volume, and sales amount; It should be noted that RoBERTa-wwm is a pre-trained Chinese language model for processing plain text data, which can extract the semantic features of the text; S34. Use the Informer model to capture long-term dependencies for the sensor time series data in the time series data, and use the N-HiTS model to decompose the trend term and periodic term features for the business time series data in the time series data; S35. Calculate the correlation weights between modalities for the features extracted from structured data, text data, and time series data through the cross-modal attention mechanism algorithm:

[0042] where, is the query vector of modality i, is the key vector of modality j, is the dimensional scaling factor; S36. Use the gated fusion algorithm to generate joint features for the feature vectors of structured data, text data, and time series data:

[0043]

[0044] where, represents the gating value, which is used to determine the mixing ratio of structured data and text data, σ(*) is the activation function, is the trainable parameter matrix, is the structured data feature vector, which comes from the processing output of 1D CNN for numerical features, is the feature vector of text data, which is the embedding of the [CLS] token from the RoBERTa model. is the feature vector of time series data, which is the mean of the final hidden states from the Informer model. is the fused cross-modal feature vector for subsequent anomaly detection tasks. S37. Calculate the anomaly probability of each business metric through a multi-layer perceptron model:

[0045]

[0046] where is the anomaly probability of the i-th metric. , , is the weight matrix of the first layer of the multi-layer perceptron model. is the weight matrix of the second layer of the multi-layer perceptron model. is the bias vector of the first layer of the multi-layer perceptron model. is the bias value of the second layer of the multi-layer perceptron model, d is the dimension of the input features, m is the number of hidden units in the first layer, and 1 is the dimension of the output. Determine the abnormal business metrics based on the anomaly probabilities of each business metric, and generate an anomaly probability vector with the abnormal business metrics and their corresponding anomaly probabilities. S38. Use a deconvolution network to map the anomaly probability vector to the dimension of the original standardized enterprise real-time data, generating a pixel-level risk heat map. Specifically, perform a deconvolution operation on the anomaly probability vector:

[0047] where is the anomaly probability vector, K is the number of business metrics. is the matrix constructed from the original standardized enterprise real-time data.

[0048] where is the upsampling operation that interpolates P from K dimensions to the original data dimension m×n, and AttentionMask(X) is a binary mask generated based on the validity of the original data. represents element-wise multiplication. The upsampling operation interpolates P from K dimensions to m×n dimensions using bilinear interpolation:

[0049] where is the abnormal probability of the kth business indicator, u and v are interpolation coordinates, indicating the position of the kth element in P in the m×n matrix, is the upsampled matrix The (i,j)th element in ; The binary mask is based on Generate a binary mask:

[0050] in, is the indicator function, when 1 if yes, 0 otherwise. is the original data matrix The (i,j)th element in ; The upsampled matrix Z and attention mask AttentionMask( X ) Element-wise multiplication suppresses the heat value of the invalid area:

[0051] Finally, the wind heat map matrix is output ; S39. Displaying the pixel-level risk heat map in layers; For example, the first-level risk heat map shows an overview of department-level anomalies, and the second-level risk heat map shows the degree of deviation of specific indicators; S4. Identify key abnormal indicators from the risk heat map, extract subgraphs related to the key abnormal indicators from the knowledge graph, generate structured prompt words from the subgraphs, and then input them into the multimodal large model together with standardized real-time enterprise data for joint reasoning and analysis. The specific steps of step S4 are as follows: S41. The specific steps for identifying key abnormal indicators from the risk heat map are as follows: Dynamic probability thresholds are pre-set to determine business indicators that meet any of the following conditions as key abnormal indicators: Business indicators whose abnormal probability exceeds the dynamic probability threshold; Affect business indicators of more than N business departments; The abnormal probability increases for M consecutive detection cycles exceeding the preset amplitude threshold; For example, the dynamic probability threshold can be set to , Represents the average value of all current abnormal probabilities, Represents the standard deviation of all current abnormal probabilities; Business metrics that can affect more than three business objectives can be set as key exception metrics, and business metrics with an increase in the abnormal probability exceeding 15% in two consecutive detection periods can be set as abnormal business metrics; S42. Specific steps to extract the subgraph related to the key exception metric from the knowledge graph: Starting from the key exception metric node, perform a search of N layers deep in the knowledge graph: Among them, the first layer is the nodes directly connected to the key exception metric node, and the subsequent layer is the nodes connected to the nodes in the previous layer; The search priority of the nodes in each layer is calculated by the edge weight and SHAP value:

[0052] Among them, is the weight between node i and node j, and SHAP is the result of feature screening, indicating the contribution of the feature to the model output, and are adjustable weight coefficients used to balance the influence of the edge weight and SHAP value. For example, takes 0.6, takes 0.4; S43. Specific steps to generate structured prompts from the subgraph are as follows: Pre-construct a prompt template, and fill in the relevant nodes and relationships in the searched subgraph into the prompt template, so as to convert the subgraph into a structured prompt described in natural language; S44. Specific steps to jointly input the structured prompt and the standardized enterprise real-time data into the multimodal large model for joint reasoning and analysis are as follows: Construct a multimodal large model with hierarchical hybrid encoding, including a parallel text encoding layer, table encoding layer, time series encoding layer, and data fusion layer; Concatenate the structured prompt and the structured data in the standardized real-time data and input them into the text encoding layer of the multimodal large model to generate a structured query vector; Concatenate the structured prompt and the text data in the standardized real-time data and input them into the table encoding layer of the multimodal large model to generate a text query vector; Input the time series prompt and the time series data in the standardized real-time data into the time series encoding layer of the multimodal large model to generate a time series query vector; Specifically, in each encoding layer, the query vector Q is jointly generated by the prompt and the data:

[0053] Among them, is the hidden state of the prompt, is the data feature; Input the structured query vector, text query vector, and temporal query vector into the data fusion layer, and generate a joint inference analysis result through cross-modal attention fusion and gated feature fusion; Specifically, the inter-modal attention weights are calculated by the following formula:

[0054] where, is the query vector of modality i (such as the text query vector), is the key vector of modality j (such as the temporal key vector), is the vector dimension scaling factor (for example, d k = 64); Use gated dynamic weighting for the features of each modality:

[0055] where, is the feature row of modality i, is the global average pooling vector of the prompt words; Extract key entities from the prompt words for root cause analysis of anomalies:

[0056] where, is the weight matrix used to calculate the probability distribution of key entities, is the bias vector used to adjust the probability distribution; Use a temporal convolutional network to predict the risk within time T:

[0057] [[ID=e5]]where TCN is a temporal convolutional network used to capture long-term dependencies in time series data, is the temporal feature representing the feature representation of time series data; Retrieve case nodes in the knowledge graph whose similarity with the current fused cross-modal feature vector is greater than the threshold, select the top-N case nodes, and generate output prompt words; Output the output prompt words, root cause analysis result, and risk propagation prediction result as a structured joint inference analysis result; S5. Convert the joint inference analysis result into an executable diagnostic report and display it to enterprise users through a visual interaction interface; The specific steps of step S5 are as follows: S51. Use the T5 model to compress the text of the joint inference analysis result, generate an executive summary containing key anomaly indicators, root cause analysis, and risk prediction, and highlight the key data in the executive summary; For example, key data such as abnormal probability exceeding dynamic thresholds or indicators affecting multiple departments; S52. Output a hierarchical risk heat map and set up a drill-down interface to connect to standardized enterprise real-time data and knowledge graphs; S53. In response to the user's click operation on the risk heat map, the corresponding standardized enterprise real-time data and related subgraphs in the knowledge graph are displayed in a linked manner, and the similarity ranking results of the related case nodes are displayed.

[0058] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0059] like Figure 2 As shown, the following is an embodiment of the enterprise intelligent diagnosis system based on a big model provided by the embodiment of the present disclosure. This system and the enterprise intelligent diagnosis method based on a big model in the above-mentioned embodiments belong to the same inventive concept. For details not fully described in the embodiment of the enterprise intelligent diagnosis system based on a big model, please refer to the embodiment of the above-mentioned enterprise intelligent diagnosis method based on a big model.

[0060] The system includes: The data collection and processing module is used to collect heterogeneous enterprise data through multi-source data interfaces, and clean, extract features, and align and fuse the collected heterogeneous enterprise data across modalities to obtain standardized real-time enterprise data; The knowledge graph construction and update module is used to build a knowledge graph with enterprise entities, business indicators and industry standards as nodes based on the industry indicators of the enterprise and standardized real-time enterprise data through the graph attention network, and regularly update the association weights between nodes according to the real-time enterprise data; The risk heat map generation module is used to input standardized real-time enterprise data into a pre-trained multimodal large model for preliminary analysis and output a risk heat map containing abnormal indicators and abnormal indicator probabilities; The reasoning analysis module is used to identify key abnormal indicators from the risk heat map and extract subgraphs related to the key abnormal indicators from the knowledge graph. The subgraphs are then converted into structured prompt words and input into a multimodal large model together with standardized real-time enterprise data for joint reasoning analysis. The diagnosis display module is used to convert the joint reasoning analysis results into executable diagnosis reports and display them to enterprise users through a visual interactive interface.

[0061] Through the interactive collaboration of the data acquisition and processing module, the knowledge graph construction and update module, the risk heat map generation module, the reasoning and analysis module, and the diagnosis and display module, this embodiment realizes the accurate identification and intelligent diagnosis of enterprise risks, and improves the decision-making efficiency.

[0062] The enterprise intelligent diagnosis method based on the large model provided by the embodiment of the present application can be applied to an electronic device. Those skilled in the art can understand that the structure of the electronic device involved in the embodiment of the present invention does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements. In the embodiment of the present invention, the electronic device includes, but is not limited to, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as, a personal digital processor, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the embodiments of the present application described herein and / or claimed.

[0063] The electronic device may include a processor, an external memory interface, an internal memory, a universal serial bus (USB) interface, a charging management module, a power management module, a battery, a wireless communication module, an audio module, a speaker, a microphone, a sensor module, a key, a camera, a display screen, and a SIM card interface, etc.

[0064] It can be understood that the structure schematically shown in the embodiment of the present application does not constitute a specific limitation on the electronic device. In other embodiments of the present application, the electronic device may include more or fewer components than shown in the figure, or combine some components, or split some components, or have different component arrangements. The components shown in the figure may be implemented in hardware, software, or a combination of software and hardware.

[0065] The processor may include one or more processing units. For example, the processor may include a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, a memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural-network processing unit (NPU), etc. Among them, different processing units may be independent devices or integrated in one or more processors.

[0066] Among them, the processor may be the nerve center and command center of the electronic device. The controller may generate operation control signals according to the instruction operation code and timing signals to complete the control of fetching and executing instructions.

[0067] A memory may also be set in the processor for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can save the instructions or data that the processor has just used or recycled. If the processor needs to use the instruction or data again, it can directly call it from this memory. This avoids repeated accesses, reduces the waiting time of the processor, and thus improves the system efficiency.

[0068] The above-mentioned electronic device implements the technical solution of the enterprise intelligent diagnosis method based on a large model in the present application, which collects enterprise heterogeneous data through a multi-source data interface, cleans, extracts features, and performs cross-modal alignment and fusion on the collected enterprise heterogeneous data to obtain standardized enterprise real-time data; based on the industry indicators of the enterprise, combined with the standardized enterprise real-time data, constructs a knowledge graph with enterprise entities, business indicators, and industry standards as nodes through a graph attention network, and periodically updates the association weights between nodes according to the enterprise real-time data; inputs the standardized enterprise real-time data into a pre-trained multi-modal large model for preliminary analysis, and outputs a risk heat map containing abnormal indicators and the probabilities of abnormal indicators; identifies key abnormal indicators from the risk heat map, extracts sub-graphs related to the key abnormal indicators from the knowledge graph, and then generates structured prompt words from the sub-graphs and jointly inputs them into the multi-modal large model with the standardized enterprise real-time data for joint reasoning analysis; converts the joint reasoning analysis results into an executable diagnostic report and displays it to enterprise users through a visual interaction interface, achieving the beneficial effect of accurately identifying and intelligently diagnosing enterprise risks through multi-source data fusion, dynamic knowledge graph, and multi-modal analysis, and improving decision-making efficiency.

[0069] In the storage medium provided by this application, there is a program product that can implement the enterprise intelligent diagnosis method based on large models.

[0070] The enterprise intelligent diagnosis method based on large models includes: collecting enterprise heterogeneous data through a multi-source data interface, and cleaning, feature extracting, and cross-modal alignment and fusion of the collected enterprise heterogeneous data to obtain standardized enterprise real-time data; based on the industry indicators to which the enterprise belongs, combining with the standardized enterprise real-time data, constructing a knowledge graph with enterprise entities, business indicators, and industry standards as nodes through a graph attention network, and regularly updating the association weights between nodes according to the enterprise real-time data; inputting the standardized enterprise real-time data into a pre-trained multi-modal large model for preliminary analysis, and outputting a risk heat map containing abnormal indicators and the probabilities of abnormal indicators; identifying key abnormal indicators from the risk heat map, extracting sub-graphs related to the key abnormal indicators from the knowledge graph, and then jointly inputting the sub-graphs generated into structured prompt words and the standardized enterprise real-time data into the multi-modal large model for joint reasoning and analysis; converting the joint reasoning and analysis results into an executable diagnostic report, and displaying it to enterprise users through a visual interaction interface.

[0071] In some possible implementation manners, the enterprise intelligent diagnosis method based on large models of the present disclosure can be implemented in the form of a program product, which includes program code. When the program product runs on a terminal device, the program code is used to cause the terminal device to execute the steps according to various exemplary embodiments of the present disclosure described in the above "Exemplary Method" section of this specification.

[0072] The storage medium of the present disclosure can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0073] The foregoing description of the disclosed embodiments enables those skilled in the art to practice or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An enterprise intelligent diagnosis method based on a large model, characterized in that, It includes the following steps: S1. Collect enterprise heterogeneous data through a multi-source data interface, clean, extract features, and perform cross-modal alignment and fusion on the collected enterprise heterogeneous data to obtain standardized enterprise real-time data; S2. Based on the industry indicators of the enterprise, combined with the standardized enterprise real-time data, construct a knowledge graph with enterprise entities, business indicators, and industry standards as nodes through a graph attention network, and regularly update the association weights between nodes according to the enterprise real-time data; S3. Input the standardized enterprise real-time data into a pre-trained multi-modal large model for preliminary analysis, and output a risk heat map containing abnormal indicators and the probabilities of abnormal indicators; S4. Identify key abnormal indicators from the risk heat map, extract sub-graphs related to the key abnormal indicators from the knowledge graph, and then generate structured prompt words from the sub-graphs and jointly input them into the multi-modal large model with the standardized enterprise real-time data for joint reasoning analysis; S5. Convert the results of the joint reasoning analysis into an executable diagnostic report and display it to enterprise users through a visual interaction interface.

2. The enterprise intelligent diagnosis method based on the large model according to claim 1, wherein The specific steps of step S1 are as follows: S11. Obtain enterprise financial data, sales data, inventory data, and human resource data through the enterprise ERP system interface, obtain production manufacturing data through the enterprise MES system interface, and obtain customer interaction data through the enterprise CRM system interface; S12. Crawl operation data from each business process of the enterprise through an RPA crawler tool, and obtain sensor data through the IoT device protocol interface; S13. Clean the collected data, remove duplicate, incorrect, incomplete, and abnormal data records, and supplement missing values; S14. Perform normalization and standardization processing on the cleaned data to complete the unification of dimensions, and then perform feature screening based on SHAP values; S15. Map the data after feature screening to a unified dimensional space and align it based on the timestamp to obtain standardized enterprise real-time data.

3. The enterprise intelligent diagnosis method based on a large model according to claim 2, wherein The specific steps of step S2 are as follows: S21. Extract enterprise entity nodes from the standardized enterprise real-time data through a named entity recognition algorithm, load index metadata from the industry benchmark library of the enterprise's industry as business indicator nodes, and extract clause entities and constraint conditions from the enterprise industry standard document as industry standard nodes to complete node initialization; S22. Construct a knowledge graph through a graph attention network: Perform GRU encoding on the standardized enterprise real-time data to generate feature vectors corresponding to the nodes; Set the initial value of the edge weights between nodes according to industry rules; S23. Dynamically update the nodes and edge weights of the knowledge graph when the update condition is triggered.

4. The enterprise intelligent diagnosis method based on large models according to claim 3, wherein, The specific steps of step S23 are as follows: S231. When a new enterprise entity or indicator anomaly is detected in the standardized enterprise real-time data, automatically create a node for the knowledge graph and initialize the embedding vector of the node: Among them, is an entity feature, is a homogeneous node embedding; S232. For existing nodes in the knowledge graph, if there is no associated data in the standardized enterprise real-time data for N consecutive days, mark them as dormant nodes and store the dormant nodes offline; S233. Update the knowledge graph according to the set update period, and use the temporal graph convolutional network to adjust the edge weights between nodes: Among them, is the embedding of node i at time t, is the amount of data change between nodes.

5. The enterprise intelligent diagnosis method based on a large model according to claim 3, characterized in that, The specific steps of step S3 are as follows: S31. Divide the standardized real-time enterprise data into structured data, text data, and temporal data; S32. Map the categorical features in the structured data into dense vectors of a set dimension through an embedding algorithm, and extract local pattern features from the numerical features in the structured data using a 1D convolutional neural network; S33. Divide the text data into tabular text and plain text, process the tabular text using the LayoutXLM model to obtain the table structure and content semantics, and use the RoBERTa-wwm model to extract text semantic features from the plain text, and extract token vectors from Chinese text; S34. Use the Informer model to capture long-term dependencies for the sensor temporal data in the temporal data, and use the N-HiTS model to decompose the trend term and periodic term features for the business temporal data in the temporal data; S35. Calculate the correlation weights between modalities for the features extracted from the structured data, text data, and temporal data through a cross-modal attention mechanism algorithm: Among them, is the query vector of modality i, is the key vector of modality j, is the dimensional scaling factor; S37. Use a gated fusion algorithm to generate joint features for the feature vectors of the structured data, text data, and temporal data: Among them, represents the gating value, and σ(*) is the activation function, is the trainable parameter matrix, is the structured data feature vector, is the text data feature vector, is the time series data feature vector, is the fused cross-modal feature vector; S38. Calculate the anomaly probability of each business metric through a multi-layer perceptron model: where, is the anomaly probability of the i-th indicator, , is the weight matrix of the first layer of the multi-layer perceptron model, is the weight matrix of the second layer of the multi-layer perceptron model, is the bias vector of the first layer of the multi-layer perceptron model, is the bias value of the second layer of the multi-layer perceptron model, d is the dimension of the input feature, m is the number of hidden units in the first layer, and 1 is the dimension of the output; Determine the abnormal business metrics according to the anomaly probabilities of each business metric, and generate an anomaly probability vector with the abnormal business metrics and the corresponding anomaly probabilities; S39. Use a transposed convolutional network to map the anomaly probability vector to the dimension of the original standardized real-time enterprise data to generate a pixel-level risk heat map; S40. Display the pixel-level risk heat map in layers.

6. The enterprise intelligent diagnosis method based on a large model according to claim 5, characterized in that The specific steps of step S4 are as follows: The specific steps to identify key abnormal metrics from the risk heat map are as follows: Pre-set a dynamic probability threshold, and determine a business metric as a key abnormal metric if it meets any of the following conditions: A business metric with an anomaly probability exceeding the dynamic probability threshold; A business metric that affects more than N business departments; The rising amplitude of the anomaly probability in M consecutive detection cycles exceeds the preset amplitude threshold; S42. The specific steps to extract the subgraph related to the key abnormal metric from the knowledge graph: Starting from the key abnormal metric node, perform a search of N layers deep in the knowledge graph: Where the first layer is the nodes directly connected to the key abnormal metric node, and the subsequent layer is the nodes connected to the nodes in the previous layer; The search priority of the nodes in each layer is calculated by the edge weight and SHAP value: Among them, is the weight between node i and node j. SHAP is the feature screening result, indicating the contribution degree of features to the model output. and are adjustable weight coefficients, used to balance the influence of edge weights and SHAP values; S43. The specific steps to generate structured prompt words from the subgraph are as follows: Pre-construct a prompt word template, and fill the relevant nodes and relationships in the searched subgraph into the prompt word template, so as to convert the subgraph into structured prompt words described as natural language; S44. The specific steps to jointly input the structured prompt words and the standardized real-time enterprise data into a multi-modal large model for joint reasoning and analysis are as follows: Construct a multi-modal large model with hierarchical hybrid encoding, including parallel text encoding layer, table encoding layer, temporal encoding layer, and data fusion layer; The structured prompt words are concatenated with the structured data in the standardized real-time data and then input into the text encoding layer of the multimodal large model to generate a structured query vector. The structured prompt words are concatenated with the standardized text data in real-time data and then input into the table encoding layer of the multimodal large model to generate a text query vector. Input the time series prompt words and the time series data in the standardized real-time data into the time series encoding layer of the multimodal large model to generate a time series query vector; The structured query vector, text query vector, and temporal query vector are input into the data fusion layer, and the joint reasoning analysis results are generated through cross-modal attention fusion and gated feature fusion.

7. The enterprise intelligent diagnosis method based on a large model according to claim 6, wherein, The specific steps of step S5 are as follows: S51. Use the T5 model to compress the text of the joint reasoning analysis results, generate an executive summary containing key anomaly indicators, root cause analysis, and risk prediction, and highlight key data in the executive summary; S52. Output a hierarchical risk heat map and set up a drill-down interface to connect to standardized enterprise real-time data and knowledge graphs; S53. In response to the user's click operation on the risk heat map, the corresponding standardized enterprise real-time data and related subgraphs in the knowledge graph are displayed in a linked manner, and the similarity ranking results of the related case nodes are displayed.

8. An enterprise intelligent diagnosis system based on a large model, characterized in that, include: The data collection and processing module is used to collect heterogeneous enterprise data through multi-source data interfaces, and clean, extract features, and align and fuse the collected heterogeneous enterprise data across modalities to obtain standardized real-time enterprise data; The knowledge graph construction and update module is used to build a knowledge graph with enterprise entities, business indicators and industry standards as nodes based on the industry indicators of the enterprise and standardized real-time enterprise data through the graph attention network, and regularly update the association weights between nodes according to the real-time enterprise data; The risk heat map generation module is used to input standardized real-time enterprise data into a pre-trained multimodal large model for preliminary analysis and output a risk heat map containing abnormal indicators and abnormal indicator probabilities; The reasoning analysis module is used to identify key abnormal indicators from the risk heat map and extract subgraphs related to the key abnormal indicators from the knowledge graph. The subgraphs are then converted into structured prompt words and input into a multimodal large model together with standardized real-time enterprise data for joint reasoning analysis. The diagnosis display module is used to convert the joint reasoning analysis results into executable diagnosis reports and display them to enterprise users through a visual interactive interface.

9. An electronic device, characterized in that The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method implements the steps of the enterprise intelligent diagnosis method based on a large model as claimed in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, the steps of the enterprise intelligent diagnosis method based on a large model as claimed in any one of claims 1 to 7 are implemented.

Images