Large model technology government affair intelligent management system and method

Abstract

The invention belongs to the technical field of information management, and discloses an intelligent management system and method for large model technology government affairs, and the system comprises four layers of architectures of data collection and access, processing and storage, intelligent analysis and decision, and user interaction and application. Through federated learning and large model fusion governance data, a business process is optimized based on a knowledge graph and a large model, and intelligent service interaction is realized by applying multi-modal interaction and the large model. Experiments prove that the system and the method can significantly improve the business handling efficiency, enhance the decision accuracy, improve the user satisfaction, provide effective technical support for intelligent transformation of government affair management, and assist in improving the government service efficiency and decision scientificity.

Description

Technical Field

[0001] The present invention relates to the technical field of information management, and particularly relates to a government affairs intelligent management system and method based on large model technology. Background Art

[0002] With the development of the digital age, a vast amount of data has been generated and accumulated in government affairs management, covering multiple fields such as people's livelihood, economy, and environment. However, there are many pain points in current government affairs management. On the one hand, data of each department is scattered, and the phenomenon of information silos is serious, resulting in difficulties in data sharing and collaboration. For example, it is difficult for the population information of the civil affairs department to interact with the tax payment information of the tax department in real time, affecting the efficiency of policy formulation and implementation. On the other hand, traditional government affairs management relies on manual experience and simple data statistical analysis, and it is difficult to mine potential value from a vast amount of complex data. When facing emergencies or complex policy formulation, decision-making lacks accurate data support and it is difficult to respond quickly and effectively.

[0003] Most existing government affairs management systems are designed for specific business processes, and their functions are limited to simple data entry, query, and report generation, lacking the ability to deeply integrate and analyze multi-source data. For example, some local government affairs approval systems can only handle a single approval process and cannot associate other relevant business data for comprehensive evaluation. Although some government affairs systems have introduced simple data mining techniques, in the face of the diversity, complexity, and real-time nature of government affairs data, these techniques are difficult to achieve a comprehensive and in-depth analysis of the data. When dealing with cross-departmental and cross-field data, traditional analysis methods cannot effectively discover the internal connections between data and cannot provide strong support for government affairs decision-making. Traditional government affairs management systems mainly rely on manual judgment in the decision-making link and lack intelligent decision-making support tools. In key tasks such as formulating policies and planning resource allocation, it is impossible to use data-driven intelligent algorithms to quickly generate multiple decision-making options and evaluate their feasibility and effects, resulting in low decision-making efficiency and insufficient scientific nature. Large model technology, with its powerful language understanding, generation, and knowledge reasoning capabilities, brings new opportunities for government affairs intelligent management. By learning a vast amount of government affairs data, large models can understand complex government affairs business logics, realize the fusion analysis of multi-source data, and provide accurate data analysis and intelligent suggestions for government affairs decision-making. For example, by analyzing people's livelihood data using large models, the needs of the public can be accurately grasped, providing a scientific basis for policy formulation; in emergency management, large models can quickly analyze various data, predict the development trend of disasters, and assist in formulating emergency response plans. Therefore, developing a government affairs intelligent management system and method based on large model technology has important practical significance and application value. Summary of the Invention

[0004] The object of the present invention is a government affairs intelligent management system based on large model technology, including: Data collection and access layer, deploying multiple data collection interfaces to achieve docking with the internal systems of each government department and external data sources, and adopting data encryption and desensitization technologies to ensure the security and privacy of data during collection and transmission; Data processing and storage layer, performing preprocessing operations such as cleaning, conversion, and integration on the collected data, using distributed storage technology to store massive government affairs data, establishing data indexes, and using data warehouse technology for multi-dimensional analysis and management; Intelligent analysis and decision-making layer, based on a data governance and analysis platform that integrates federated learning and large models, an intelligent optimization algorithm for government affairs business processes that combines knowledge graphs and large models, and an intelligent interactive system for government affairs services that combines multi-modal interaction and large models, deeply analyzing and mining government affairs data to provide intelligent support for government affairs decision-making, and also including a model training and management module; User interaction and application layer, providing a friendly human-computer interaction interface for government affairs staff and the public. Government affairs staff can perform operations such as business processing, data analysis, and decision-making. The public can obtain government affairs services and provide feedback through websites, mobile applications, etc.

[0005] Furthermore, the data governance and analysis platform that integrates federated learning and large models includes: Federated learning architecture construction module, building a federated learning framework. Each government department acts as an independent node, and security protocols such as homomorphic encryption are used for the exchange and aggregation of model parameters; Large model adaptation and optimization module, selecting a pre-trained language model based on the Transformer architecture and using a supervised fine-tuning training with a government affairs domain-specific corpus; Data governance and analysis process module, each department node performs cleaning, annotation, and feature engineering on local data and then inputs it into the large model for local training. After the training is completed, the model parameters are uploaded to the federated learning server, and the server updates the global model through a secure aggregation algorithm and distributes it to each node.

[0006] Furthermore, the intelligent management system for government affairs using large model technology is characterized in that the intelligent optimization algorithm for government affairs business processes based on knowledge graphs and large models includes: Government affairs knowledge graph construction module, collecting knowledge such as policies, regulations, business processes, and organizational structures in the government affairs domain, constructing a knowledge graph through entity extraction, relationship extraction, and semantic annotation technologies, and storing it in a graph database; Business process modeling and optimization module, abstracting government affairs business processes into directed graphs, using large models to analyze the business process knowledge graph, and mining bottlenecks and optimization points in the process; Intelligent decision-making support module, according to the government affairs business status, searching for relevant information in the knowledge graph, and combining large models to provide the best business processing path and decision-making suggestions for staff.

[0007] Furthermore, the large model technology-based government affairs intelligent management system is characterized in that the government affairs service intelligent interaction system based on multimodal interaction and large model includes: A multimodal data collection and fusion module, which integrates technologies such as speech recognition, natural language processing, and image recognition to realize the collection and fusion of multimodal interaction data such as user speech, text, and images; A large model-driven intelligent response module, which uses the large model to understand user needs, retrieves relevant knowledge in the large model, generates accurate and easy-to-understand answers, and provides personalized services according to the user's historical interaction records; A service quality evaluation and feedback module, which evaluates service quality through methods such as user satisfaction surveys and interaction data analysis, and adjusts the large model parameters and interaction strategies according to the evaluation results.

[0008] Furthermore, the present invention also provides a government affairs intelligent management method based on large model technology, including the following steps: A data collection and access step, which uses the deployed data collection interface to connect to the internal systems and external data sources of each government department, and collects and transmits data using data encryption and desensitization technologies; A data processing and storage step, which preprocesses the collected data such as cleaning, transformation, and integration, stores the data using distributed storage technology, and establishes a data index and a data warehouse; An intelligent analysis and decision-making step, based on a data governance and analysis platform that integrates federated learning and large model, an intelligent optimization algorithm for government affairs business processes that combines knowledge graph and large model, and a government affairs service intelligent interaction system that combines multimodal interaction and large model, deeply analyzes and mines government affairs data to provide intelligent support for government affairs decision-making; A user interaction and application step, through the human-computer interaction interface, government affairs staff perform operations such as business processing, data analysis, and decision-making, and the public obtains government affairs services and gives feedback.

[0009] Furthermore, in the step of the data governance and analysis platform that integrates federated learning and large model, when each department node cleans the local data, duplicate data is removed, incorrect data is corrected, and missing values are filled; when performing annotation and feature engineering processing, the key features of the data are extracted; when training the large model, the learning rate is set to 0.001, the number of training rounds is 10 rounds, and the optimizer selects AdamW.

[0010] Furthermore, in the step of the intelligent optimization algorithm for government affairs business processes that combines knowledge graph and large model, when collecting government affairs domain knowledge, natural language processing technology is used to extract entities and relationships from policy and regulation texts, business process documents, etc.; after constructing the knowledge graph, the government affairs business process is abstracted into a directed graph and input into the large model for analysis.

[0011] Furthermore, in the steps of the government affairs service intelligent interaction system based on multi-modal interaction and large models, when collecting multi-modal data, users interact with the system through voice, text, uploading pictures, etc.; when the large model generates response content, relevant content is retrieved from the policy and regulation knowledge graph, etc.

[0012] Furthermore, the human-computer interaction interface design of the user interaction and application layer follows the principles of simplicity and usability, uses visual charts to display government affairs data and business processes, and sets operation buttons to facilitate users to perform business operations and data queries.

[0013] Furthermore, the data collection and access layer supports docking with at least three different types of internal systems of government departments and two external data sources to ensure the diversity and comprehensiveness of data.

[0014] Beneficial effects: Precise progress monitoring: The progress prediction algorithm based on the spatio-temporal attention mechanism and variational autoencoder can accurately predict the project progress, discover potential progress risks in advance, and provide timely and effective decision-making support for project managers.

[0015] Efficient task scheduling: The task scheduling algorithm of the dynamic task network and reinforcement learning realizes the intelligent scheduling and optimization of tasks, improves the project execution efficiency, and ensures the timely completion of the project.

[0016] Optimized resource allocation: The resource optimization allocation algorithm of multi-agent collaboration and game theory can realize the optimal allocation of resources according to project requirements and resource status, improve resource utilization rate, and reduce project costs.

[0017] Intelligent risk warning: Through the real-time monitoring and analysis of the project progress and resource usage, combined with the uncertainty estimation of the prediction results by the variational autoencoder, the risk factors in the project can be discovered in time and warnings can be issued to help project managers formulate countermeasures in advance. Description of the drawings

[0018] Figure 1 System principle flowchart; Specific implementation manners

[0019] The following further describes in detail the implementation manners of the present invention in conjunction with the drawings and embodiments. The following embodiments are used to illustrate the present invention, but cannot be used to limit the scope of the present invention.

[0020] Embodiment 1 The objective of the present invention is to provide a government affairs intelligent management system and method based on large model technology. By integrating multi-source government affairs data and applying advanced artificial intelligence algorithms and large model technology, it realizes the intelligent management of government affairs processes, in-depth analysis and mining of data, and intelligent auxiliary support for decision-making, improves the efficiency of government affairs management, optimizes the quality of government services, and enhances the scientificity and accuracy of government decision-making.

[0021] Data collection and access layer: Deploy a variety of data collection interfaces to support docking with the internal systems of each government department and external data sources (such as Internet public opinion data, third-party statistical data). Adopt technologies such as data encryption and desensitization to ensure the security and privacy of data during collection and transmission. For example, obtain the business data of each department through API interfaces and use the SSL / TLS protocol for data transmission encryption.

[0022] Data processing and storage layer: Perform preprocessing operations such as cleaning, transformation, and integration on the collected data to remove noise and outliers and unify the data format. Adopt distributed storage technology, such as the Hadoop Distributed File System (HDFS), to store massive government affairs data and establish a data index to improve data query efficiency. At the same time, use data warehouse technology to perform multidimensional analysis and management of data.

[0023] Intelligent analysis and decision-making layer: Based on the data governance and analysis platform integrating federated learning and large models, the intelligent optimization algorithm for government affairs business processes of knowledge graphs and large models, and the intelligent interactive system for government services of multimodal interaction and large models, deeply analyze and mine government affairs data to provide intelligent support for government affairs decision-making. This layer also includes a model training and management module responsible for the training, optimization, and update of large models.

[0024] User interaction and application layer: Provide a friendly human-computer interaction interface for government affairs staff and the public. Government affairs staff can perform operations such as business processing, data analysis, and decision-making through this interface; the public can interact with the system through methods such as websites and mobile applications to obtain government services and give feedback. The interface design follows the principles of simplicity and usability to improve user operation efficiency.

[0025] Data governance and analysis platform integrating federated learning and large models Construction of the federated learning architecture: In view of the sensitivity and department isolation of government affairs data, build a federated learning framework. Each government department acts as an independent node, and through an encryption mechanism and security protocol, exchanges and aggregates model parameters without sharing the original data. For example, use homomorphic encryption technology to encrypt the transmitted parameters to ensure the security of data during transmission.

[0026] Large Model Adaptation and Optimization: Select a large model suitable for government data processing, such as a pre-trained language model based on the Transformer architecture, and fine-tune it according to the characteristics of the government domain. Utilize the professional corpus in the government domain, including policies, regulations, government reports, etc., to conduct supervised fine-tuning training on the large model, enabling it to better understand and process government data.

[0027] Data Governance and Analysis Process: Each department node first cleans, annotates, and performs feature engineering on local data, and then inputs the processed data into the large model for local training. After the training is completed, each node uploads the model parameters to the federated learning server, and the server updates the global model through a secure aggregation algorithm. The updated global model is then sent back to each node for the analysis and decision support of government data, such as policy effect evaluation, risk prediction, etc.

[0028] Intelligent Optimization Algorithm for Government Affairs Business Processes Based on Knowledge Graph and Large Model Construction of Government Affairs Knowledge Graph: Collect and integrate various types of knowledge in the government domain, including information such as policies, regulations, business processes, and organizational structures, to construct a comprehensive government affairs knowledge graph. Through technologies such as entity extraction, relationship extraction, and semantic annotation, transform government affairs knowledge into structured graph data and store it in a graph database for convenient knowledge query and reasoning.

[0029] Business Process Modeling and Optimization: Abstract government affairs business processes into directed graphs, where nodes represent business links and edges represent the logical relationships between links. Use the large model to analyze the business process knowledge graph and identify bottlenecks and optimization points in the process. For example, through the reasoning ability of the large model, it is found that some approval links can be processed in parallel, thus shortening the business handling time.

[0030] Intelligent Decision Support: When government affairs staff handle business, the system searches for relevant information in the knowledge graph according to the current business status and, combined with the intelligent recommendation function of the large model, provides the staff with the best business handling path and decision-making suggestions. For example, when handling enterprise administrative approvals, the system recommends appropriate approval processes and required materials based on information such as enterprise type and business scope.

[0031] Intelligent Interaction System for Government Affairs Services Based on Multimodal Interaction and Large Model Multimodal Data Collection and Fusion: Integrate technologies such as speech recognition, natural language processing, and image recognition to achieve the collection and fusion of user multimodal interaction data. For example, users can interact with the system through speech, text, or uploading pictures, and the system fuses different modal data to accurately understand user needs.

[0032] Large Model-Driven Intelligent Response: Leveraging the powerful language understanding and generation capabilities of large models to intelligently answer users' questions. The system first parses and semantically understands the data input by users, then retrieves relevant knowledge in the large model to generate accurate and easy-to-understand answers. At the same time, the system can provide personalized services based on users' historical interaction records and preferences.

[0033] Service Quality Assessment and Feedback: Evaluate the service quality of the intelligent interaction system for government services through methods such as user satisfaction surveys and interactive data analysis. According to the evaluation results, timely adjust the parameters of the large model and interaction strategies to continuously improve the service level and user experience of the system.

[0034] Deploy federated learning nodes within each government department, configure data collection interfaces, connect to the internal business systems of the department, and obtain raw government data, such as population information, economic data, environmental monitoring data, etc.

[0035] Clean and preprocess the collected data to remove duplicate data, correct incorrect data, and fill in missing values. For example, use data cleaning tools to process incorrect names and duplicate records in population information according to data quality rules.

[0036] Perform annotation and feature engineering on the preprocessed data to extract key features from the data. For example, quantify features such as GDP and industrial structure in economic data.

[0037] Select a suitable large model, such as the BERT model, and perform fine-tuning training on each department node using local annotated data. Set training parameters, such as a learning rate of 0.001, 10 training epochs, and select the AdamW optimizer.

[0038] After the training of each department node is completed, upload the model parameters to the federated learning server through an encrypted channel. The server uses a secure aggregation algorithm, such as the FedAvg algorithm, to aggregate the parameters of each node and update the global model.

[0039] The federated learning server distributes the updated global model to each department node, and each node uses the global model to analyze local government data, such as predicting the implementation effect of policies and evaluating social risks.

[0040] Collect various types of knowledge in the government affairs field, including policy and regulation texts, business process documents, organizational information, etc., and use natural language processing techniques to perform entity extraction and relationship extraction. For example, extract entities such as policy names, scope of application, and implementation time from policy and regulation texts, as well as the association relationships between entities.

[0041] Convert the extracted knowledge into a knowledge graph and store it in the Neo4j graph database. Construct the nodes and edges of the knowledge graph. For example, use business processes as nodes and the sequence between processes as edges to establish a business process knowledge graph.

[0042] Abstract the government affairs business process as a directed graph and input it into a large model for analysis. Through the understanding and reasoning of the knowledge graph, the large model discovers optimization points in the business process. For example, some approval processes can be merged or simplified.

[0043] When government affairs staff handle business, the system queries relevant information in the knowledge graph according to the current business status and combines the intelligent recommendation function of the large model to provide the staff with an optimized business processing path and operation suggestions. For example, when handling the enterprise tax declaration business, the system recommends the optimal declaration process and precautions according to the enterprise type and declaration data.

[0044] Integrate multimodal interaction components such as speech recognition, natural language processing, and image recognition in the government affairs service website and mobile applications. For example, users can input questions by voice, and the system uses speech recognition technology to convert the voice into text.

[0045] After the user inputs a question, the system first performs multimodal fusion processing on the input data, integrates information such as voice, text, and images, and accurately understands the user's needs. For example, when the user uploads a photo of an ID card and asks about the process of handling a certain business, the system obtains the ID card information through image recognition and conducts comprehensive analysis in combination with the text question.

[0046] The system inputs the user's needs into the large model, and the large model generates answer content based on the pre-trained knowledge and fine-tuned government affairs domain knowledge. For example, for a user's consultation about a certain policy, the large model retrieves relevant content from the policy and regulation knowledge graph and generates a detailed policy interpretation and application guidance.

[0047] The system feeds back the answer content to the user and records the user's interaction history and satisfaction evaluation. By analyzing the user's interaction data and satisfaction feedback, continuously optimize the parameters and interaction strategies of the large model to improve the service quality of the system. For example, if the user is not satisfied with an answer, the system analyzes the reason, adjusts the model parameters, and improves the answer content.

[0048] Select multiple government affairs departments as experimental objects, including the civil affairs department, tax department, environmental protection department, etc., and deploy the government affairs intelligent management system proposed by the present invention in these departments.

[0049] Divide the experimental departments into an experimental group and a control group. The experimental group adopts the system and method of the present invention, and the control group adopts the traditional government affairs management method.

[0050] Determine the experimental indicators, including business handling efficiency, decision-making accuracy, user satisfaction, etc. Business handling efficiency is measured by counting the business handling time and the degree of process simplification; decision-making accuracy is evaluated by comparing the matching degree between the decision result and the actual effect; user satisfaction collects user feedback through questionnaires and online evaluations.

[0051] Business handling efficiency: The average business handling time of the experimental group was shortened by 35%, and the degree of process simplification was increased by 40%. For example, in the enterprise administrative approval business, which originally required separate approvals from multiple departments and had a cumbersome process, after adopting the system of the present invention, through intelligent optimization of the business process, parallel processing of some approval links was achieved, greatly shortening the approval time.

[0052] Decision-making accuracy: The decision-making accuracy of the experimental group was increased by 28% compared with the control group, and it could more accurately predict the implementation effect of policies and social risks. For example, when formulating environmental protection policies, through in-depth analysis of multi-source data such as environmental monitoring data and economic data, and using the prediction ability of the large model, the policies formulated by the experimental group were more in line with the actual situation, effectively reducing the environmental pollution risk.

[0053] User satisfaction: Through questionnaires and online evaluations, the user satisfaction of the experimental group reached more than 85%, while that of the control group was only about 60%. Users feedback that the intelligent interaction function of the system is convenient and fast, can accurately answer questions, and provide personalized services, greatly improving the user experience.

[0054] The experimental results show that the government affairs intelligent management system and method based on large model technology proposed by the present invention can significantly improve the government affairs management efficiency, enhance the accuracy and scientificity of decision-making, and improve user satisfaction.

[0055] Example 2 Initial modeling of the blockchain ledger: Build a consortium blockchain architecture and use the Practical Byzantine Fault Tolerance (PBFT) consensus algorithm to construct the blockchain network. Define the block structure of the blockchain. The block is divided into a block header and a block body. The block header records information such as the timestamp, the hash value of the previous block, the hash value of the current block, and the Merkle root; the block body is used to store the hash value of government affairs data. With the PBFT algorithm, each node can reach a consensus on the generation of new blocks, ensuring the consistency and security of the ledger.

[0056] Data Encryption and Storage Modeling: Select the AES (Advanced Encryption Standard) symmetric encryption algorithm to encrypt government affairs data. For each data file, generate a random encryption key and use this key to encrypt the data. Store the encrypted data in the distributed storage system of the blockchain node. At the same time, use the recipient's public key to perform RSA asymmetric encryption on the encryption key and store it in the smart contract of the blockchain. Only the authorized recipient can decrypt and obtain the original encryption key using the private key, and then decrypt the data.

[0057] Smart Contract Modeling: Write smart contracts using the Solidity language and set access rights and sharing rules for data. For example, create a smart contract called DataAccess. There is a mapping table accessControl in the contract to store the access rights of each department to different data. When a department applies to access data, the smart contract calls the accessControl mapping table to verify the rights of the requester. If the rights verification passes, the smart contract returns the corresponding encrypted data hash value, and the requester obtains the data from the blockchain storage based on the hash value.

[0058] Large Model Analysis Modeling: Select a large model based on the Transformer architecture, such as BERT. When analyzing on-chain data, first decrypt the encrypted data obtained from the blockchain, and then perform text preprocessing, including operations such as word segmentation and stop word removal. Then convert the preprocessed data into an input format that the large model can accept, such as word vector representation. Through supervised fine-tuning training on financial data in the government affairs field, let the large model master the characteristics and laws of financial data. When predicting the trend of fiscal revenue and expenditure, input historical financial data into the fine-tuned large model, and the model gives the prediction result.

[0059] Large Model Migration and Fine-tuning Modeling: Select the BERT model pre-trained on a large-scale general corpus (such as Wikipedia) for migration. Define a new classifier layer and connect the output of the BERT model as the input to this classifier layer. In the government affairs field, collect text data such as policies, regulations, and service guides, and divide them into training sets, validation sets, and test sets. Use the training set to fine-tune the migrated model, set a small learning rate, train for 5 rounds, and select the Adam optimizer. During the fine-tuning process, update the model parameters through the backpropagation algorithm to enable the model to better understand the semantics related to government affairs operations.

[0060] User Behavior Data Modeling: Design a user behavior data collection system to record users' operation behaviors on the government service platform, such as clicks, queries, submissions, etc. Represent the user behavior data as time series data, where each time step contains information such as user ID, operation type, operation object, etc. Use a Recurrent Neural Network (RNN) or its variant, the Long Short-Term Memory Network (LSTM), to model the user behavior data. Input the user behavior time series data into the LSTM network, and the LSTM network learns the time-dependent relationships of user behaviors and outputs user behavior feature vectors.

[0061] Personalized Recommendation Modeling: Combine the understanding of government affairs texts by the large model and the user behavior feature vectors for personalized recommendations. Use the collaborative filtering algorithm to construct a rating matrix of users - service items. For each user, calculate the user's interest level in different service items based on their behavior feature vectors and the large model's understanding of service items. When a user accesses the government service platform, sort the service items according to the interest level and recommend the top N most interesting service items to the user.

[0062] State Space Modeling: Define the state space, which is a multi-dimensional vector covering the available quantities of government affairs resources, such as the number of human resources, material inventories, financial balances, as well as the priority level of tasks, represented by integers from 0 - 10, where 10 represents the highest priority, and the urgency level of tasks, represented by floating-point numbers from 0 - 1, where 1 represents the most urgent, and other information.

[0063] Action Space Modeling: Define the action space, which contains various resource allocation strategies. For example, an action can be to allocate a certain amount of human, material, and financial resources to a certain task.

[0064] Reward Function Modeling: Design the reward function, which is determined based on factors such as the completion of tasks and the utilization efficiency of resources. If a task is completed on time, a positive reward is given, and the reward value is related to the task priority; if there is resource waste, that is, the actual amount of resources used is greater than the reasonable demand, a negative reward is given, and the negative reward value is related to the amount of wasted resources.

[0065] Reinforcement Learning Algorithm Modeling: Use the Deep Q-Network (DQN) algorithm to solve. Construct a Q-network, with the state space as the input and the Q-values of each action as the output. Utilize the experience replay mechanism to store the states, actions, rewards, and next states generated by the agent's interaction with the environment in the experience pool. Regularly randomly sample a batch of data from the experience pool for training the Q-network. After continuous iterative training, the Q-network learns the optimal resource allocation strategy, and the agent can select the optimal action according to different states to obtain the maximum cumulative reward.

[0066] Sensor Data Acquisition and Transmission Modeling: Deploy Internet of Things sensors in key urban areas. For traffic flow sensors, use geomagnetic sensors or video image sensors; for air quality sensors, use multi-parameter gas sensors; for sensors monitoring the operating status of water conservancy facilities, use pressure sensors, water level sensors, etc. Use the MQTT (Message Queuing Telemetry Transport) protocol for data transmission, and define the message format of sensor data, including information such as sensor ID, timestamp, measurement value, etc. After encapsulating the collected data according to the message format, the sensor sends it to the MQTT server through the MQTT client, and then the server forwards the data to the data processing center.

[0067] Data Preprocessing Modeling: In the data processing center, perform denoising and normalization on the collected sensor data. For noisy data, use median filtering or Kalman filtering algorithms to remove it. For example, for traffic flow data, use median filtering to remove sudden outliers. Normalization is to map the data to the interval of 0 - 1.

[0068] Large Model Prediction Modeling: Select the Gated Recurrent Unit (GRU), a variant based on the Recurrent Neural Network (RNN), to build a prediction model. Divide the preprocessed data into training sets and test sets according to the time series. During training, input the training set data into the GRU model, set a certain number of hidden layer neurons, set a relatively small learning rate, train for 10 rounds, and select the Adam optimizer. The GRU model learns the time series features and patterns of the data to predict future traffic flow, air quality and other indicators.

[0069] Early Warning Mechanism Modeling: Set early warning rules and thresholds according to historical data and business requirements. For example, for traffic congestion early warning, when the predicted traffic flow exceeds a certain proportion of the maximum road capacity, trigger the early warning. Define the format of the early warning information, including early warning type, early warning time, early warning location, early warning details, etc. When the data predicted by the large model reaches the early warning threshold, send early warning information to relevant departments and the public through the SMS gateway or APP push service.

Claims

1. A large-scale model technology government affairs intelligent management system, characterized in that: include: The data collection and access layer deploys a variety of data collection interfaces to achieve docking with the internal systems of various government departments and external data sources, and uses data encryption and desensitization technology to ensure the security and privacy of data during collection and transmission; The data processing and storage layer cleans, converts, and integrates the collected data for preprocessing operations, uses distributed storage technology to store massive government data, establishes data indexes, and uses data warehouse technology for multi-dimensional analysis and management; Intelligent analysis and decision-making layer, based on the data governance and analysis platform integrating federated learning and big models, the intelligent optimization algorithm of government business processes based on knowledge graphs and big models, and the intelligent interaction system of government services based on multimodal interaction and big models, which conducts in-depth analysis and mining of government data, provides intelligent support for government decision-making, and also includes model training and management modules; The user interaction and application layer provides a friendly human-computer interaction interface for government staff and the public. Government staff can perform business processing, data analysis, decision-making and other operations, and the public can obtain government services and feedback through websites, mobile applications, etc.

2. According to the large model technology government affairs intelligent management system described in claim 1, it is characterized in that: The data governance and analysis platform integrating federated learning and large models also includes: The federated learning architecture building module builds a federated learning framework. Each government department acts as an independent node and uses security protocols such as homomorphic encryption to exchange and aggregate model parameters. The large model adaptation and optimization module selects a pre-trained language model based on the Transformer architecture and uses a professional corpus in the government affairs field for supervised fine-tuning training; In the data governance and analysis process module, each department node cleans, annotates, and performs feature engineering on local data, and then inputs the data into the large model for local training. After the training is completed, the model parameters are uploaded to the federated learning server. The server updates the global model through a secure aggregation algorithm and sends it to each node.

3. According to the large model technology government affairs intelligent management system described in claim 1, it is characterized in that: The intelligent optimization algorithm for government affairs business processes based on knowledge graph and big model includes: The government affairs knowledge graph construction module collects policies and regulations, business processes, and organizational knowledge in the government affairs field, constructs a knowledge graph through entity extraction, relationship extraction, and semantic annotation, and stores it in a graph database; The business process modeling and optimization module abstracts the government business process into a directed graph, uses a large model to analyze the business process knowledge graph, and discovers bottlenecks and optimization points in the process; The intelligent decision support module searches for relevant information in the knowledge graph based on the status of government affairs, and combines the big model to provide staff with the best business processing path and decision-making recommendations.

4. According to the large model technology government affairs intelligent management system described in claim 1, it is characterized in that: The government service intelligent interaction system based on multimodal interaction and large model includes: Multimodal data collection and fusion module integrates speech recognition, natural language processing, and image recognition to realize the collection and fusion of multimodal interactive data such as user speech, text, and images; The intelligent response module driven by the big model uses the big model to understand user needs, retrieve relevant knowledge in the big model, generate accurate and easy-to-understand answers, and provide personalized services based on the user's historical interaction records; The service quality evaluation and feedback module evaluates service quality through user satisfaction surveys and interactive data analysis, and adjusts large model parameters and interactive strategies based on the evaluation results.

5. A method for intelligent government affairs management based on the large model technology system of claim 1, characterized in that: The following steps are involved: Data collection and access steps: Use the deployed data collection interface to connect the internal systems of various government departments and external data sources, and use data encryption and desensitization technology to collect and transmit data; Data processing and storage steps: cleaning, conversion, and integration preprocessing of collected data, using distributed storage technology to store data, and establishing data indexes and data warehouses; Intelligent analysis and decision-making steps: Based on the data governance and analysis platform that integrates federated learning and big models, the intelligent optimization algorithm of government business processes based on knowledge graphs and big models, and the intelligent interaction system of government services based on multimodal interaction and big models, the government data is deeply analyzed and mined to provide intelligent support for government decision-making; User interaction and application steps: Through the human-computer interaction interface, government staff conduct business processing, data analysis, and decision-making operations, and the public obtains government services and provides feedback.

6. According to claim 5, the method for intelligent management of government affairs using large model technology is characterized in that: In the data governance and analysis platform step of the integration of federated learning and big models, when each department node cleans the local data, it removes duplicate data, corrects erroneous data, and fills in missing values; when performing labeling and feature engineering, it extracts key data features; when training the big model, the learning rate is set to 0.001, the number of training rounds is 10, and the optimizer is selected as AdamW.

7. According to claim 5, the method for intelligent management of government affairs using large model technology is characterized in that: In the steps of the intelligent optimization algorithm for government affairs business processes based on knowledge graphs and big models, when collecting knowledge in the government affairs field, natural language processing technology is used to extract entities and relationships from policy and regulatory texts, business process documents, etc.; after constructing the knowledge graph, the government affairs business process is abstracted into a directed graph and input into the big model for analysis.

8. According to claim 5, the method for intelligent management of government affairs using large model technology is characterized in that: In the steps of the intelligent interactive system for government services based on multimodal interaction and big models, when multimodal data is collected, users interact with the system through voice, text, and uploading pictures; when the big model generates answer content, relevant content is retrieved from the policy and regulatory knowledge graph.

9. The large-scale model technology government affairs intelligent management system according to claim 1 is characterized in that: The design of the human-computer interaction interface of the user interaction and application layer follows the principles of simplicity and ease of use, uses visual charts to display government data and business processes, and sets operation buttons to facilitate users to perform business operations and data queries.

10. The large-scale model technology government affairs intelligent management system according to claim 1 is characterized in that: The data collection and access layer supports connection with at least three different types of internal systems of government departments and two external data sources to ensure the diversity and comprehensiveness of the data.

Images