Government affair internet of things data assetization and intelligent early warning method and system based on deep learning and blockchain

Abstract

The application belongs to the technical field of data processing, and particularly relates to a government affair Internet of Things data assetization and intelligent early warning method and system based on deep learning and block chain. The method comprises the following steps: integrating government affair Internet of Things equipment data and associated auxiliary data, and completing data collection after preprocessing; adopting a double-engine parallel processing mode, and realizing intelligent early warning of government affair equipment anomalies by relying on a self-adaptive fusion convolution recursive neural network; simultaneously, through block chain right confirmation, multi-dimensional quality evaluation and scenario-based value quantification, data assetization conversion is completed; high-quality data assets are standardized and encapsulated, and are pushed to government affair demanders through a precise matching algorithm; asset trusted circulation is realized by adopting federal learning and differential privacy technology, and the model is continuously optimized in combination with feedback data. The upgrading from single device monitoring and early warning to data assetization utilization is realized, and the early warning accuracy and government affair collaborative governance efficiency are significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to a method and system for the assetization and intelligent early warning of government IoT data based on deep learning and blockchain. Background Technology

[0002] With the deepening of smart government construction, the number of IoT devices deployed in government scenarios such as urban governance, environmental protection, water supply networks, and building energy consumption has exploded, generating massive amounts of IoT sensing data. Existing technologies based on deep learning-based automatic early warning methods for monitoring IoT sensing devices have achieved the collection, processing, and anomaly warning of data from individual sensors, effectively reducing equipment failure rates and maintenance costs. However, existing technologies only focus on the monitoring and early warning functions of the devices themselves and have not solved the more core problem in government scenarios: how to transform scattered and isolated IoT data into government data assets that can be identified, circulated, quantified, and reused across departments.

[0003] Currently, government IoT data management faces three major pain points in practical applications. First, data is scattered across different government departments, creating severe data silos. Raw data from single sensors lacks clear business relevance, lacks independent asset value, and cannot be directly utilized by other departments. Second, a comprehensive government data assetization system has not yet been established. Data ownership is ambiguous, and value lacks unified quantitative standards, resulting in a lack of effective matching and connection mechanisms between data providers and users. Third, government data has extremely high security and compliance requirements. Risks such as privacy leaks and data tampering are highly likely during cross-departmental circulation. Existing early warning technologies are disconnected from data assetization, failing to form a comprehensive management system encompassing early warning, asset creation, and circulation. Some assetization solutions geared towards general applications fail to deeply integrate with the business characteristics and security requirements of government affairs, making it difficult to directly implement them while ensuring early warning accuracy, assetization feasibility, and data compliance. Summary of the Invention

[0004] To overcome the problems of existing technologies, such as fragmented and isolated government IoT data, unclear data ownership, ambiguous value quantification, lack of matching with demanders, poor circulation security, and a serious disconnect between early warning and assetization, this invention provides a method and system for government IoT data assetization and intelligent early warning based on deep learning and blockchain. This upgrade from single-device monitoring and early warning to data asset utilization significantly improves the accuracy of early warning and the efficiency of collaborative governance in government.

[0005] The technical solution of this application specifically includes: According to one aspect of this application, a method for the assetization and intelligent early warning of government IoT data based on deep learning and blockchain is provided, including: By integrating IoT sensing device data and related auxiliary data through government-dedicated IoT interfaces and government cloud storage, the integrated data is preprocessed to obtain standardized data. Based on standardized data, the intelligent early warning engine and the assetization engine processes data simultaneously. The intelligent early warning engine uses an adaptive fusion convolutional recurrent neural network to output device health index and graded early warning signals. The assetization engine receives early warning signals and standardized data, first confirms the data source, department, and collection time through a blockchain smart contract and generates a certificate of authenticity, then scores the data stream based on four dimensions: completeness, accuracy, timeliness, and consistency to select high-quality data streams, and finally calculates the data asset value of the high-quality data streams using either an early warning value quantification formula or a collaborative value quantification formula according to their application scenarios. High-quality data streams, early warning results, quality scores, value quantification results, and blockchain-based evidence are packaged into government data asset packages and digital asset ID cards are generated. At the same time, the cosine similarity algorithm is used to calculate the matching degree between the feature vector of the asset package and the feature vector of the demand side. When the matching degree reaches a preset threshold, the asset package is automatically pushed to the corresponding demand side. Demand users access the analysis results of data asset packages through federated learning and differential privacy technology, achieving data usability without visibility, and the access process is recorded through blockchain; at the same time, based on user feedback and newly added government data, the intelligent early warning model and asset value quantification model are continuously optimized using the time-enhanced loss function.

[0006] As a further option of the method of the present invention, the associated auxiliary data includes equipment operation and maintenance records, fault history data, government policy requirements, and lists of departmental responsibilities; Among them, the government policy requirements cover the standards for data asset entry into the table and the specifications for cross-departmental sharing. The list of departmental responsibilities includes the functional codes, business scope descriptions and cross-departmental collaborative items of each government department. Related auxiliary data is collected through the government business system interface using a dual mechanism that combines scheduled batch synchronization and event-triggered real-time synchronization. Event-triggered real-time synchronization is executed immediately when government policy documents are updated and released or when departmental responsibility adjustment notices are issued.

[0007] As a further option of the method of the present invention, the data preprocessing employs an outlier detection algorithm based on time window weighting, including: An outlier detection algorithm based on time windows is used to clean, correct, and standardize the collected data, remove noise, and fill in missing values. The preprocessing formula is as follows: ; ; in, These are standardized outlier metrics. This represents the average of the data from the previous moment. The standard deviation of the data at the previous time step. The corrected data values. This is a correction factor.

[0008] As a further option of the method of the present invention, in the intelligent early warning engine processing, the convolutional layer of the adaptive fusion convolutional recurrent neural network includes three cascaded one-dimensional convolutional layers. The first convolutional layer uses 64 convolutional kernels of size 3, the second convolutional layer uses 128 convolutional kernels of size 3, and the third convolutional layer uses 64 convolutional kernels of size 3. Each convolutional operation is followed by a batch normalization layer and a modified linear unit activation function to extract the spatial correlation features of sensor data in the channel dimension.

[0009] As a further option of the method of the present invention, in the intelligent early warning engine processing, the long short-term memory recurrent network of the adaptive fusion convolutional recurrent neural network adopts a two-layer stacked structure, with each layer having a hidden state dimension of 128 dimensions; the long short-term memory unit models the temporal dependency relationship through a forget gate, an input gate, and an output gate, wherein the forget gate controls the retention ratio of the unit state at the previous time step, and the input gate controls the writing of new information, as shown in the following formula: Forgotten Gate: ; Input Gate: , ; Cell status update: ; Output gate: , ; in, , , These are the forget gate, input gate, and output gate, respectively. This represents the current state of the cell. This refers to the cell state at the previous moment. This represents the current state of the candidate cell. It is the Sigmoid activation function. This is the weight matrix. For bias terms, The state was hidden in the previous moment. Hide the current state. Enter the current time. This is element-wise multiplication.

[0010] As a further option of the method of the present invention, in the intelligent early warning engine processing, a dynamic gating mechanism is introduced between the convolutional neural network and the long short-term memory recurrent network, and the dynamic gating vector... according to Calculation, where For dynamic gating values, The state was hidden in the previous moment. Input features for the current time step. For the gated weight matrix, As a bias term, the original spatial feature vector and The gated feature vector is obtained after element-wise multiplication.

[0011] As a further option of the method of the present invention, in the assetization processing engine, the data stream is scored from four dimensions: completeness, accuracy, timeliness, and consistency, and the overall quality score is used to determine the quality score. Calculate using the following formula: ;in, The total score for data quality. For completeness score, To score for accuracy, For timeliness, For consistency score, These are the weighting coefficients, and the sum of the four equals 1.

[0012] As a further option of the method of the present invention, in the assetization processing engine, the formula for quantifying the value of early warning categories is: ,in For the value of early warning data assets, To avoid the average total loss from a single equipment failure, The percentage improvement in early warning accuracy compared to traditional manual monitoring methods; the formula for quantifying the value of collaborative methods is... ,in For collaborative data assets, The total cost saved through cross-departmental collaboration, For data reuse rate.

[0013] As a further option of the method of the present invention, the cosine similarity algorithm calculates the matching degree between the asset package feature vector A and the demand side's demand feature vector B. Perform as follows: ; in, Let A be the feature vector of data asset A. Let n be the demand feature vector of demander B, and n be the total dimension of the feature vector.

[0014] As a further option of the method of the present invention, the differential privacy technique is implemented by providing the original analysis results... To achieve privacy protection, Laplace noise is added. The processed output is... ;in, For data after differential privacy processing, For raw data assets, For privacy budget parameters, It is Laplace noise.

[0015] As a further option of the method of the present invention, the time-enhancing loss function is calculated as follows: ; in, The loss value. This is the actual value. For predicted values, These are weighting coefficients. This represents the difference between the prediction and the actual result.

[0016] This application also provides a government IoT data assetization and intelligent early warning system based on deep learning and blockchain, the system including: The data collection and preprocessing module is used to integrate data from IoT sensing devices and related auxiliary data through a dedicated government IoT interface and government cloud storage, and to preprocess the integrated data to obtain standardized data. The dual-engine parallel processing module is used to simultaneously execute intelligent early warning engine processing and assetization processing engine processing based on standardized data. The intelligent early warning engine processing uses an adaptive fusion convolutional recurrent neural network to output device health index and graded early warning signals. The assetization processing engine processing receives early warning signals and standardized data, first confirms the data source, department of ownership, and collection time through blockchain smart contracts and generates storage certificates, then scores the data streams in terms of quality from four dimensions: completeness, accuracy, timeliness, and consistency to filter out high-quality data streams, and finally calculates the data asset value of high-quality data streams according to their application scenarios using either an early warning value quantification formula or a collaborative value quantification formula. The asset encapsulation and matching module is used to encapsulate high-quality data streams, early warning results, quality scores, value quantification results, and blockchain-based evidence certificates into government data asset packages and generate digital asset ID cards. At the same time, it uses a cosine similarity algorithm to calculate the matching degree between the asset package feature vector and the demand feature vector of the demand party. When the matching degree reaches a preset threshold, the asset package is automatically pushed to the corresponding demand party. The Trusted Circulation and Model Optimization module supports demanders in accessing the analysis results of data asset packages through federated learning and differential privacy technologies, achieving data availability without visibility, and recording the access and circulation process through the blockchain; at the same time, based on demander feedback and newly added government data, the module continuously optimizes the intelligent early warning model and asset value quantification model using the time-enhanced loss function.

[0017] The beneficial effects of this application are as follows: Building upon the applicant's deep learning early warning technology from last year, this approach reduces the difficulty of research and development and writing, while simultaneously upgrading the technology from single-device early warning to an integrated approach of early warning and assetization. This addresses the core issue of government data assetization that is not covered by existing technologies, ensuring technological continuity with the applicant's existing patents while also creating a clear differentiation from existing government patents. By using dual-engine parallel processing, the original deep learning model's advantage of accurate early warning is retained, while a new asset-based processing module is added. This clearly solves the core questions of how to calculate equipment data as an asset, who needs it, and how much it is worth. It clarifies that raw data from a single sensor is not an asset, but a reusable data stream that has been processed and combined with early warning results is an asset. The demand side is internal departments of the government system, and the value is reflected in collaborative cost reduction and improved governance efficiency. By combining blockchain technology to achieve data ownership confirmation and notarization, and by combining federated learning and differential privacy technology to achieve trusted data circulation, it fully complies with the requirements for government data security and compliance, avoids problems such as data leakage and tampering, and is compatible with the latest government policies such as the "Twenty Articles on Data" and data asset entry into tables.

[0018] The technical solution is tailored to practical government scenarios, clarifies the correspondence between data suppliers and demanders, and provides specific algorithm formulas and operation steps. It can be directly applied to government departments such as urban management, environmental protection, water resources, and government affairs management, and is highly practical with a high probability of authorization. Through a continuous model optimization mechanism, the system adapts to the dynamic changes in government scenarios, ensuring the accuracy of early warnings and the quantification of asset value. This enables closed-loop management of data, early warnings, assets, circulation, and optimization, breaking down data silos in government affairs and improving the level of refinement in government governance. Attached Figure Description

[0019] Figure 1 This is the overall flowchart; Figure 2 This is a flowchart of the dual-engine parallel processing. Figure 3 This is a flowchart of the data assetization process. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The core theoretical foundation of this invention is built upon the theories of deep learning temporal feature extraction and spatial feature fusion, blockchain distributed ledger and smart contract notarization, federated learning and differential privacy data protection, and government data asset valuation. By constructing an adaptive fusion convolutional recurrent neural network, the spatial feature extraction capability of convolutional neural networks is integrated with the temporal dependency capture capability of long short-term memory recurrent networks. A dynamic gating mechanism is introduced to optimize feature selection, enabling accurate identification and tiered early warning of the status of government IoT devices. A data ownership confirmation and notarization traceability system is constructed based on blockchain smart contracts to ensure clear data asset ownership and traceable flow. A four-dimensional quality assessment model (completeness, accuracy, timeliness, consistency) and a dual-scenario value quantification model (early warning and collaborative scenarios) are established to transform scattered sensor data into measurable and tableable government data assets. A cosine similarity matching algorithm is used to achieve precise matching between data assets and government demanders. Federated learning and differential privacy technologies are combined to achieve trusted circulation of data assets that are usable but invisible. Finally, a time-enhanced loss function drives continuous model optimization, forming a complete closed-loop method from secure data collection, dual-engine parallel processing, asset encapsulation and matching to trusted circulation and model optimization.

[0022] The specific embodiments of the present invention will be described in detail below.

[0023] Example 1: Please see Figure 1 This document illustrates an overall flowchart of a method for the assetization and intelligent early warning of government IoT data based on deep learning and blockchain, according to an embodiment of the present invention. The method includes: S1: Secure collection of government IoT data, integrating IoT sensing device data and related auxiliary data, achieving secure collection through government-specific IoT interfaces and government cloud storage, and using a time window-weighted outlier detection algorithm for data preprocessing; S2: Dual-engine parallel processing. The intelligent early warning engine uses an adaptive fusion convolutional recurrent neural network to perform device status recognition and anomaly warning, while the asset processing engine simultaneously performs data ownership confirmation, quality assessment and value quantification. S3: Data asset encapsulation and demand matching. The processing results are encapsulated into standardized government data asset packages and digital asset ID cards are generated. The cosine similarity algorithm is used to achieve accurate matching between assets and demanders. S4: Trusted circulation of data assets and model optimization. Federated learning and differential privacy technologies are used to achieve secure access to data assets. Time-enhanced loss functions are used to continuously optimize the early warning model and the asset value quantification model.

[0024] The specific plan is as follows: In a method for assetizing and intelligently warning government IoT data based on deep learning and blockchain, S1 integrates two types of data sources: IoT sensing device data and government business-related auxiliary data. It uses a time window-weighted outlier detection algorithm to clean, correct, and standardize the raw data. It also achieves secure collection of data throughout the entire process through a dedicated government IoT interface and a government cloud storage architecture, providing a high-quality and standardized data foundation for subsequent dual-engine parallel processing.

[0025] S11: Data collection from IoT sensing devices.

[0026] In this embodiment, raw data generated by various sensors in government scenarios is collected in real time through a dedicated government IoT gateway. Each sensor device in a government scenario is connected to the dedicated government IoT gateway via a aggregation node. The aggregation node has edge computing capabilities and can perform preliminary analog-to-digital conversion, digital filtering, and timestamping of the raw sensor signals. Each connected sensor device is assigned a globally unique device identifier.

[0027] The data collection from IoT sensing devices covers four typical government application scenarios: urban management, environmental protection, water affairs, and government affairs management.

[0028] In one possible implementation of this embodiment, the urban management scenario collects data on the tilt angle, vibration acceleration, displacement, and opening / closing status of the smart manhole cover at a sampling frequency of 1Hz; the environmental protection scenario collects air quality data such as PM2.5 concentration, PM10 concentration, sulfur dioxide concentration, and ozone concentration, as well as water quality data such as dissolved oxygen, ammonia nitrogen, and turbidity, with an air quality data sampling frequency of 0.1Hz and a water quality data sampling frequency of once every 15 minutes; the water affairs scenario collects pipeline pressure, instantaneous flow rate, and cumulative flow rate data at a sampling frequency of 1Hz; and the government affairs management scenario collects energy consumption data such as building indoor temperature, relative humidity, three-phase voltage of the power distribution cabinet, and operating current of the air conditioning system at a sampling frequency of 0.2Hz.

[0029] S12: Synchronize associated auxiliary data.

[0030] In this embodiment, government business data associated with IoT sensing devices are collected synchronously through the standard data interface of the government business system.

[0031] In one possible implementation of this embodiment, the associated auxiliary data includes four categories: Equipment operation and maintenance records include the equipment installation date, the date of the most recent maintenance, the type of maintenance, the list of replaced parts, the personnel responsible for the inspection, and the inspection results. Historical fault data, including fault occurrence time, fault type, fault severity level, handling measures, recovery time, and impact assessment; Government policy requirements for data include the data asset entry standard document number and clause summary, the cross-departmental sharing specification document number and scope of application, and the data classification and hierarchical management method version number; The data on departmental responsibilities includes the unified social credit code, functional code, business scope description, and list of cross-departmental collaborative matters for each government department.

[0032] The synchronization of related auxiliary data adopts a dual mechanism that combines scheduled batch synchronization and event-triggered real-time synchronization. Scheduled batch synchronization is executed once every natural hour by default, while event-triggered real-time synchronization is triggered immediately when policy documents are updated or departmental responsibilities are adjusted.

[0033] S13: Data preprocessing.

[0034] In this embodiment, data preprocessing is performed on the raw IoT sensing device data collected in S11 and S12. The core algorithm is an outlier detection algorithm based on time window weighting.

[0035] In one possible implementation of this embodiment, a time-window-weighted outlier detection algorithm is used to clean, correct, and standardize the collected raw data, remove noise, and fill in missing values ​​to ensure data quality. The preprocessing formula is as follows: ; ; in, These are standardized outlier metrics. This represents the average of the data from the previous moment. The standard deviation of the data at the previous time step. The corrected data values. This is a correction factor, with a value range of 0.1-0.3, adapted to the characteristics of government data.

[0036] In a method for the assetization and intelligent early warning of government IoT data based on deep learning and blockchain, S2 synchronously distributes the standardized data preprocessed by S1 to the intelligent early warning engine and the assetization processing engine, with the two engines executing in parallel at the logical level. For example... Figure 2 As shown, the dual-engine parallel processing flow is as follows: the preprocessed data is input into the intelligent early warning engine and the assetization processing engine respectively; the intelligent early warning engine uses the AF-CRNN model and dynamic gating mechanism to realize feature extraction, time series processing, and anomaly early warning, and outputs the early warning result; the assetization processing engine uses data ownership confirmation, quality assessment, and value quantification to output the assetization processing result; the output results of the two engines are used together for subsequent asset encapsulation.

[0037] S21: Intelligent Early Warning Engine.

[0038] In this embodiment, the intelligent early warning engine adopts an adaptive fusion convolutional recurrent neural network. The model structure consists of three cascaded parts: a convolutional neural network feature extraction module, a dynamic gating fusion module, and a long short-term memory recurrent network time-series processing module.

[0039] S211: Convolutional neural networks extract spatial features.

[0040] The preprocessed multidimensional sensor standardized data from S1 is input into the intelligent early warning engine in the form of a three-dimensional tensor. The convolutional neural network feature extraction module performs a one-dimensional convolution operation on the input data to extract spatial correlation feature patterns between different sensor channels at the same time. The convolution operation is calculated as follows: the convolution filter f is set as a weight matrix of size k×k. For a k×k window region in the input data x with position (i,j) as the upper left corner, the convolution output... It is equal to the sum of the product of the filter f with the corresponding elements within the window region, element by element. Each convolutional kernel can learn and recognize a specific spatial correlation pattern of sensor data.

[0041] The convolutional neural network module consists of three cascaded convolutional layers: the first convolutional layer uses 64 kernels of size 3; the second convolutional layer uses 128 kernels of size 3 to extract higher-order combined features from the 64-dimensional feature space output by the first layer; and the third convolutional layer uses 64 kernels of size 3 to compress the high-dimensional features back to 64 dimensions. Each convolutional operation is followed by a batch normalization layer and a modified linear unit activation function.

[0042] S212: Temporal feature processing of long short-term memory recurrent networks.

[0043] The spatial feature sequences extracted by the convolutional neural network are sequentially input into the long short-term memory recurrent network in time step order to model temporal dependencies. The long short-term memory network effectively captures the device state evolution patterns spanning long time spans in government IoT data through three gating mechanisms: forget gate, input gate, and output gate.

[0044] Forgot the goalkeeper's hidden state from the previous moment With current input After concatenation and weight matrix Multiplication plus bias The output control vector is activated by the Sigmoid activation function. Control the unit state of the previous moment The proportion of information retained. The input gate is split into two parallel paths: one path passes through the Sigmoid activation function and outputs... The position is determined, and the other path outputs the candidate state via the hyperbolic tangent activation function. Provides new information. Cell status. The system is updated jointly by the forget gate (filtering out the old state) and the input gate (writing in the new information). The output gate is updated based on the updated cell state. Calculate the hidden state at the current time. Two layers of long short-term memory units are stacked. The first layer captures short-term local temporal patterns, while the second layer abstracts longer-term trend changes.

[0045] In one possible implementation of this embodiment, the temporal characteristics of the sensor data are processed by an RNN, and the LSTM unit formula is as follows: Forgotten Gate: ; Input Gate: , ; Cell status update: ; Output gate: , ; in, , , These are the forget gate, input gate, and output gate, respectively. It is the Sigmoid activation function. This is the weight matrix. For bias terms, The state was hidden in the previous moment. Hide the current state. Enter the current time. This is element-wise multiplication.

[0046] S213: Execution of dynamic gating mechanism.

[0047] A dynamic gating mechanism is introduced between the convolutional neural network and the long short-term memory network to adaptively select features flowing to the temporal processing module. The input for calculating the dynamic gating value includes the hidden state of the previous time step. and the spatial feature vector at the current time step The two are concatenated and then combined with the gating weight matrix. Multiplication plus bias The output gate vector is activated by the Sigmoid activation function. . The calculation formula is: ; in, For dynamic gating values, The state was hidden in the previous moment. Input features for the current time step. For the gated weight matrix, This is the bias term. The original spatial feature vector... and After element-wise multiplication, the gated feature vector is obtained and used as the actual input to the Long Short-Term Memory network at the current time step. The dynamic gating mechanism enables the model to dynamically adjust the information flow path according to the context. When the sensor is subjected to occasional interference, the gating value of the corresponding channel is automatically reduced to ensure the stability of the time series prediction.

[0048] S22: The asset processing engine performs data ownership confirmation, quality assessment, and value quantification.

[0049] In this embodiment, the assetization processing engine and the intelligent early warning engine are started in parallel. The blockchain infrastructure of the assetization processing engine adopts a government consortium blockchain architecture, and the network node roles are divided into four categories: data supplier nodes, rights confirmation nodes, demand-side nodes, and regulatory nodes. The data rights confirmation smart contract is deployed to the rights confirmation nodes during the engine initialization phase. Figure 3 As shown, the data assetization process is as follows: pre-processed data → blockchain ownership confirmation → quality assessment → value quantification → asset encapsulation.

[0050] S221: Data ownership confirmation and generation of blockchain-based evidence certificates.

[0051] Once S1 completes the preprocessing and storage of a batch of data, the data ownership confirmation process is automatically triggered. Key metadata such as data source, timestamp, processing flow, and data type are extracted from the preprocessed data batch. The data provider node uses its blockchain account private key to perform elliptic curve digital signature on the metadata and sends the signed ownership confirmation request message to the ownership confirmation node. The ownership confirmation node's smart contract automatically verifies the signature, concatenates the data provider account address, device identifier, timestamp, processing flow hash, and current block height into a string, and applies the SHA-256 hash algorithm to calculate the notarization hash value, which serves as the unique digital fingerprint of this batch of data assets on the blockchain. Simultaneously, a globally unique notarization certificate number is generated. The smart contract constructs the ownership confirmation transaction, and after reaching a consensus through the consortium blockchain mechanism, it is written into the blockchain distributed ledger. The notarization certificate number, notarization hash value, and on-chain transaction hash value are returned to the data provider node for archiving and storage.

[0052] S222: Execution of the four-dimensional data asset quality assessment model.

[0053] Once data ownership is confirmed, the data stream enters the quality assessment phase, where a comprehensive quality score is given based on four dimensions: completeness, accuracy, timeliness, and consistency. The completeness dimension calculates the ratio of valid data points within the assessment window to the theoretically required number, introducing a penalty mechanism for consecutive missing data: any data point missing more than five consecutively is deducted double. The accuracy dimension, when a reference benchmark is available, is represented by the average relative error between the sensor's measured value and the standard value; otherwise, it is estimated based on the sensor's factory accuracy level and the time since the most recent calibration. The timeliness dimension is scored in segments based on the delay from data acquisition to storage: ≤10s receives 100 points, 10s to 30s linearly decreases to 90 points, 30s to 60s decreases to 75 points, 60s to 300s decreases to 60 points, and >300s is fixed at 30 points. The consistency dimension establishes a verification rule base, comparing the logical consistency of sensor data with related data such as maintenance records; violations of the rules result in point deductions.

[0054] In one possible implementation of this embodiment, the four-dimensional scores are weighted to synthesize a comprehensive quality score. The evaluation formula is as follows: ; in, The total score for data quality (0-100 points) is given. For completeness score, To score for accuracy, For timeliness, For consistency score, These are weighting coefficients, and the sum of the four factors equals 1. For example, for environmental air quality monitoring data, the timeliness weight γ is 0.4, and α, β, and δ are each 0.2; for water supply network pressure monitoring data, accuracy and completeness have higher weights, with α at 0.3, β at 0.3, γ at 0.2, and δ at 0.2. A quality pass score is set at 60 points. Data that meets the standard enters the value quantification stage, while data that does not meet the standard is only used for routine monitoring.

[0055] S223: Implementation of the quantitative model for the value of government data assets.

[0056] High-quality data streams enter the value quantification stage, and the system provides two types of value quantification formulas: early warning and collaborative data. Early warning data assets are suitable for equipment fault early warning scenarios, and their value... The calculation formula is: ; in, Value of early warning data assets; To avoid the average comprehensive loss amount for a single equipment failure, the accounting scope includes the direct costs of equipment repair and replacement, compensation for personal injury and property damage caused by the accident to third parties, the conversion of traffic congestion efficiency loss, and the administrative costs of government emergency response. The accounting scope is determined by each government department based on historical failure records and financial settlement data. The percentage improvement in early warning accuracy compared to traditional manual monitoring methods after implementing this method is expressed as a decimal.

[0057] Collaborative data assets are suitable for cross-departmental collaborative governance scenarios. The formula for calculating the value of collaborative data assets is as follows: ; in, For collaborative data assets, The total cost saved through cross-departmental collaboration is jointly determined by each collaborating department based on business processes and historical collaborative project final audit data; η is the data reuse rate, which is the ratio of the number of demanding departments actually called within the statistical period to the total number of potential demanding departments, and is automatically counted by the number of asset call transactions on the blockchain.

[0058] S224: Aggregation and output of assetization processing results.

[0059] The assetization processing engine aggregates and integrates the rights confirmation and storage information, four-dimensional quality assessment results, value quantification results, and applicable scenario tags into a structured assetization processing result object, which is then transmitted to S3 along with the early warning results output by the intelligent early warning engine.

[0060] In a method for the assetization and intelligent early warning of government IoT data based on deep learning and blockchain, S3 standardizes and encapsulates the early warning results and assetization processing results output by the parallel processing of the S2 dual engines, generates a government data asset package in a unified format and assigns a unique digital asset ID. At the same time, the cosine similarity matching algorithm is used to achieve accurate matching and automatic push of asset packages with demanders.

[0061] S31: Standardized packaging of government data asset packages and generation of digital asset ID cards.

[0062] In this embodiment, the system encapsulates the assetization processing results output by S224 and the early warning results output by S21 into a standardized government data asset package. The encapsulation format adopts JSON format, which is fully compatible with the existing government information system data exchange standards. The top-level structure of the asset package contains five logical blocks: the asset identification block, which contains the digital asset ID number and generation timestamp; the data content block, which contains high-quality sensor standardized data streams, equipment health index, and early warning level codes; the quality and value block, which contains the comprehensive quality score, detailed scores for each dimension, the value quantification result amount, and the type of quantification formula used; the evidence storage and security block, which contains the blockchain evidence storage certificate number, evidence storage hash value, and on-chain transaction hash value; and the scenario and demand block, which contains an array of applicable government scenario tags and a list of recommended demanding departments.

[0063] In one possible implementation of this embodiment, a globally unique digital asset ID number is assigned to each asset package. The ID number structure adopts a multi-segment hierarchical coding rule: version number field, 6-digit administrative division code, 3-digit department code, 2-digit asset category code, where JG represents urban management manhole cover monitoring, HB represents environmental protection monitoring, SW represents water supply network monitoring, JN represents government affairs energy consumption, 8-digit generation date and 4-digit daily serial number, and each field is separated by hyphens.

[0064] S32: Demand-side matching and asset delivery.

[0065] Maintain a government data asset requirement database, recording the requirement scenarios of various government departments. Each entry in the database includes fields such as the requesting department code, requirement scenario name, detailed description text, required data type, required early warning signal type, time granularity, and spatial scope requirements. The system employs a vector space model based on term frequency-inverse document frequency weighting to transform the text description of each requirement scenario and the scenario label text of each data asset package into requirement feature vector B and data asset feature vector A, respectively.

[0066] For each newly generated data asset package, iterate through and calculate the cosine similarity between its feature vector A and all requirement feature vectors B in the requirement database. Cosine similarity. The calculation formula is: ; in, The degree of matching between data asset A and demander B. Let A be the feature vector of data asset A. For demander B, the data asset package is automatically pushed to the corresponding demander when the matching degree is ≥0.8. n is the total dimension of the feature vector.

[0067] In one possible implementation of this embodiment, the preset matching score push threshold is 0.8: if If the value is ≥0.8, a successful match is determined, and the JSON content of the asset package, the digital asset ID number, and the summary information are automatically and encryptedly pushed to the business system interface of the requesting department through the government data sharing and exchange channel; if If the value is less than 0.8, it will be temporarily stored in the waiting-to-match observation queue. After each push is completed, the system records the matching log and sends a confirmation request to the requester. The returned records are included in the matching algorithm optimization dataset.

[0068] In a method for the assetization and intelligent early warning of government IoT data based on deep learning and blockchain, S4 provides secure and reliable data asset access services to the demand side after S3 completes the asset package push. It adopts a federated learning architecture and differential privacy technology to make data assets available but invisible, while continuously optimizing the early warning model and asset value quantification model by using a time-enhanced loss function.

[0069] S41: Trusted access to data assets based on federated learning and differential privacy.

[0070] In this embodiment, the requester's access to the asset package analysis results employs a security scheme combining a federated learning horizontal architecture with differential privacy noise injection. The federated learning architecture deploys client components on each participating node in the government cloud. The data provider holds complete local raw data and deep learning models, while the requester only initiates query requests and receives returned results. The coordination server is responsible for routing query tasks but does not access the raw data in plaintext.

[0071] In one possible implementation of this embodiment, differential privacy protection processing is applied to the output results before the data provider returns the analysis results to the requester. The processed output results... Equal to the original analysis results Adding a noise term, which is ε multiplied by a Laplace random variable following a location parameter of 0 and a scale parameter of 1 / δ, the differential privacy processing formula is as follows: ; in, For data after differential privacy processing, For raw data assets, This is a privacy budget parameter that controls the balance between the strength of privacy protection and data utility. The default configuration range is 0.1 to 0.5. It is Laplace noise; In one possible implementation of this embodiment, the request-side invocation process is as follows: The requester submits an invocation request through the federated learning query interface, specifying the asset package number and analysis result type; the coordination server queries the blockchain to verify access authorization; after successful verification, the request is routed to the data provider node; the provider node performs analysis and calculation locally, applies differential privacy noise, and returns a blurred result through a secure channel. The requester's account address, asset package number, invocation timestamp, and analysis result type summary for each invocation are recorded in the blockchain flow log via a smart contract for auditing by regulatory nodes. The log does not store the specific numerical values ​​of the analysis results.

[0072] S42: The time-augmented loss function drives continuous model optimization.

[0073] In this embodiment, online incremental optimization of the model is triggered periodically based on feedback from demanders, changes in government policies, and newly labeled data. The model optimization employs a time-enhanced loss function, adding a time-weighted mechanism to the traditional prediction error loss, making the model more attentive to the distribution characteristics of recent samples. Time-enhanced loss function The calculation method is as follows: ; in, The loss value. This is the actual value. For predicted values, These are weighting coefficients. This represents the difference between the prediction and the actual result.

[0074] The total loss is obtained by summing the weighted errors of all samples. The optimizer uses an adaptive moment estimation algorithm with an initial learning rate of 0.001 and a batch size of 64. After each optimization, the warning accuracy is evaluated on an independent validation set. The new model parameters are hot-swapped to the online service only when the accuracy improves by more than 0.5 percentage points to prevent performance degradation.

[0075] Thus, this embodiment fully implements the closed-loop method of the entire process from S1 secure data collection, S2 dual-engine parallel processing, S3 asset encapsulation and matching to S4 trusted circulation and model optimization, realizing the technical upgrade of government IoT data from single monitoring and early warning to early warning-assessment-circulation integration.

[0076] Example 2: Assetization and intelligent early warning of air quality data by environmental protection departments; S1. Data Collection: Collect air quality sensor data (PM2.5, PM10, sulfur dioxide concentration, etc.) from environmental protection departments, as well as related data such as air quality monitoring records and collaborative policies between environmental protection and health commissions; preprocess the data using the TWA-OUD algorithm to remove noise and store it after standardization. S2, Dual-engine processing: S21, Intelligent Early Warning Engine: Processes air quality time-series data through the AF-CRNN model, sets dust exceeding warning thresholds, and outputs air quality anomaly warning signals and control suggestions.

[0077] S22. Asset Processing Engine: After blockchain confirmation of rights, high-quality data is selected based on timeliness (real-time updates ≤ 5 minutes) and accuracy. The value of assets is calculated using a collaborative value quantification formula (if cross-departmental collaboration costs are reduced by 100,000 yuan and data reuse rate is 80%, then the asset value = 100,000 × 80% = 80,000 yuan).

[0078] S3. Asset Packaging and Matching: After packaging the asset package, it is pushed to the demand side (Health Commission, Education Bureau) for disease early warning and school outdoor activity arrangement.

[0079] S4. Reliable Circulation and Optimization: Based on user feedback, optimize the model's warning threshold and value quantification coefficient to adapt to air quality fluctuations caused by seasonal changes.

[0080] Comparison of existing technologies with the specific technologies of this solution: I. Detailed Comparison Table of Existing Technologies and the Technical Solution of This Invention; II. Experimental data for the core algorithm; The experiment uses real-world government application scenarios as the test environment, selecting four typical government scenarios: urban management manhole cover monitoring, environmental air quality monitoring, water pipe network monitoring, and government building energy consumption monitoring. A total of 1,500 sets of raw data from government IoT devices were collected, including 1,050 sets of normal data and 450 sets of abnormal data. The abnormal data covers various common abnormal scenarios in government, such as equipment failure, data anomalies, and environmental anomalies. The core performance of the applicant's previous patent model and this solution model were compared, and key indicators such as data asset matching efficiency and data circulation security were tested to verify the technological advancement, practicality, and security of this solution. All experimental data are based on real government equipment and are reproducible and verifiable. The specific experimental setup and data are as follows.

[0081] (a) Basic experimental setup; 1. Experimental Environment: This experimental environment is deployed on a government-dedicated cloud server. This server is configured with an 8-core CPU, 16GB of memory, and runs on a Kylin V10 SP3 ARM64-bit operating system. It adopts the Kunpeng ARM processor architecture and is equipped with a 500GB hard drive. The backend is developed using Java 8 and built on the Spring Boot 2.7.18 and MyBatis 3.5.16 frameworks. It integrates the TensorFlow 2.8 deep learning framework and the Hyperledger Fabric 2.4 blockchain framework. The database uses PolarDB O engine 14 and is configured with a dedicated 16 cores, 64GB of RAM, and a 2000GB hard drive.

[0082] 2. Test data sources: Real-time data were collected continuously for 30 days from 200 manhole cover monitoring points of the urban management department, 50 air quality monitoring stations of the environmental protection department, 80 pipeline monitoring points of the water affairs department, and energy consumption monitoring equipment of 30 buildings of the government affairs administration bureau in a certain city. The data were pre-processed and used for the experiment.

[0083] 3. Comparison objects: TWA-DCNN+AF-CRNN, and the optimized AF-CRNN model of this solution.

[0084] 4. Evaluation indicators: early warning accuracy, missed reporting rate, false alarm rate, response time; matching accuracy, response time; data circulation security, and traceability accuracy.

[0085] (II) Experimental data on the early warning performance of the AF-CRNN model; The AF-CRNN model in this solution significantly outperforms prior patented models in terms of early warning performance, primarily due to two major improvements: First, the dynamic gating mechanism is optimized by adjusting the gating weight matrix parameters, enhancing the model's ability to capture temporal anomalies in government equipment and enabling rapid identification of sudden equipment failures. Second, the TELF time-augmented loss function is adopted, which reduces model overfitting and improves the model's generalization ability by differentially weighting historical fault data and real-time monitoring data, effectively lowering the false negative and false positive rates. Simultaneously, the model's early warning response time is consistently controlled within 3 seconds, meeting the timeliness requirements of government emergency response. This provides timely and accurate early warning information to government departments, facilitating rapid handling of equipment anomalies and reducing losses caused by failures. (III) Experimental data on data asset matching efficiency; This solution employs a cosine similarity algorithm to achieve precise matching between data assets and demanders. Compared to existing general matching algorithms, it offers three major advantages: First, it boasts a short matching response time, enabling rapid matching and delivery of assets and demands, meeting the efficient office needs of government departments. Second, it offers high matching accuracy, with no significant mismatches, ensuring that the delivered assets precisely match the business needs of each department. Third, it achieves high data reuse, breaking down data silos in government affairs, enhancing the utilization value of government data assets, and resolving issues such as ambiguous matching, low reuse rates, and inconvenient access in existing technologies. Furthermore, the asset access success rate remains above 99.5%, indicating that the asset encapsulation format is compatible with existing government business systems and can be implemented without large-scale modifications, further demonstrating the practicality and advanced nature of this solution.

[0086] (iv) Experimental data on data circulation security; This solution employs a triple security mechanism of federated learning, differential privacy, and blockchain to achieve trusted circulation of government data assets. Experimental data shows a data leakage rate of 0% and a traceability accuracy rate of ≥99.8%. Specifically, federated learning ensures data is usable but not visible, preventing original data from leaving the department and thus avoiding data leakage; differential privacy technology dynamically adjusts the privacy budget to perturb the original data, further enhancing data security; and blockchain records the entire circulation process, enabling traceability and immutability, ensuring compliance with regulations such as the "Government Data Security Management Measures." This solution addresses the insecurity and lack of traceability in existing government data circulation technologies, providing a secure guarantee for the circulation of government data assets.

[0087] Example 3: A government IoT data assetization and intelligent early warning system based on deep learning and blockchain, the system includes: The data collection and preprocessing module is used to integrate data from IoT sensing devices and related auxiliary data through a dedicated government IoT interface and government cloud storage, and to preprocess the integrated data to obtain standardized data. The dual-engine parallel processing module is used to simultaneously execute intelligent early warning engine processing and assetization processing engine processing based on standardized data. The intelligent early warning engine processing uses an adaptive fusion convolutional recurrent neural network to output device health index and graded early warning signals. The assetization processing engine processing receives early warning signals and standardized data, first confirms the data source, department of ownership, and collection time through blockchain smart contracts and generates storage certificates, then scores the data streams in terms of quality from four dimensions: completeness, accuracy, timeliness, and consistency to filter out high-quality data streams, and finally calculates the data asset value of high-quality data streams according to their application scenarios using either an early warning value quantification formula or a collaborative value quantification formula. The asset encapsulation and matching module is used to encapsulate high-quality data streams, early warning results, quality scores, value quantification results, and blockchain-based evidence certificates into government data asset packages and generate digital asset ID cards. At the same time, it uses a cosine similarity algorithm to calculate the matching degree between the asset package feature vector and the demand feature vector of the demand party. When the matching degree reaches a preset threshold, the asset package is automatically pushed to the corresponding demand party. The Trusted Circulation and Model Optimization module supports demanders in accessing the analysis results of data asset packages through federated learning and differential privacy technologies, achieving data availability without visibility, and recording the access and circulation process through the blockchain; at the same time, based on demander feedback and newly added government data, the module continuously optimizes the intelligent early warning model and asset value quantification model using the time-enhanced loss function.

[0088] Those skilled in the art will understand that the embodiments of this application are provided as methods, systems, or computer program products. Therefore, this application takes the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application takes the form of a computer program product implemented on one or more computer storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer program code. The solutions in the embodiments of this application are implemented using various computer languages, exemplified by the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0089] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, are implemented by computer program instructions. These computer program instructions are provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams.

[0090] These computer program instructions are also stored in a computer read-memory that can direct a computer or other programmed data processing device to operate in a particular manner, such that the instructions stored in the computer read-memory produce an article of manufacture including instruction means that implement the functions specified in the flowchart or multiple flowcharts and / or block diagram blocks or multiple block diagrams.

[0091] These computer program instructions are also loaded onto a computer or other programming data processing device to cause a series of operational steps to be performed on the computer or other programming device to produce a computer-implemented process, such that the instructions, which execute on the computer or other programming device, provide steps for implementing the functions specified in the flowchart flow or multiple flows and / or the block diagram blocks or multiple blocks.

[0092] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0093] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for assetizing and intelligently warning government IoT data based on deep learning and blockchain, characterized in that, include: By integrating IoT sensing device data and related auxiliary data through government-dedicated IoT interfaces and government cloud storage, the integrated data is preprocessed to obtain standardized data. Based on standardized data, the intelligent early warning engine and the assetization engine processes data simultaneously. The intelligent early warning engine uses an adaptive fusion convolutional recurrent neural network to output device health index and graded early warning signals. The assetization engine receives early warning signals and standardized data, first confirms the data source, department, and collection time through a blockchain smart contract and generates a certificate of authenticity, then scores the data stream based on four dimensions: completeness, accuracy, timeliness, and consistency to select high-quality data streams, and finally calculates the data asset value of the high-quality data streams using either an early warning value quantification formula or a collaborative value quantification formula according to their application scenarios. High-quality data streams, early warning results, quality scores, value quantification results, and blockchain-based evidence are packaged into government data asset packages and digital asset ID cards are generated. At the same time, the cosine similarity algorithm is used to calculate the matching degree between the feature vector of the asset package and the feature vector of the demand side. When the matching degree reaches a preset threshold, the asset package is automatically pushed to the corresponding demand side. The demand side uses federated learning and differential privacy technology to access the analysis results of the data asset package, enabling secure access to the data assets. The access process is recorded through the blockchain. At the same time, based on the demand side's feedback and the addition of new government data, the intelligent early warning model and the asset value quantification model are continuously optimized using the time-enhanced loss function.

2. The method for assetizing and intelligently warning government IoT data based on deep learning and blockchain according to claim 1, characterized in that, The associated auxiliary data includes equipment operation and maintenance records, historical fault data, government policy requirements, and lists of departmental responsibilities. Among them, the government policy requirements cover the standards for data asset entry into the table and the specifications for cross-departmental sharing. The list of departmental responsibilities includes the functional codes, business scope descriptions and cross-departmental collaborative items of each government department. Related auxiliary data is collected through the government business system interface using a dual mechanism that combines scheduled batch synchronization and event-triggered real-time synchronization. Event-triggered real-time synchronization is executed immediately when government policy documents are updated and released or when departmental responsibility adjustment notices are issued.

3. The method for assetizing and intelligently warning government IoT data based on deep learning and blockchain according to claim 2, characterized in that, The data preprocessing employs a time-window-weighted outlier detection algorithm, including: An outlier detection algorithm based on time windows is used to clean, correct, and standardize the collected data, remove noise, and fill in missing values. The preprocessing formula is as follows: ； ； in, These are standardized outlier metrics. This represents the average of the data from the previous moment. The standard deviation of the data at the previous time step. The corrected data values. This is a correction factor.

4. The method for assetizing and intelligently warning government IoT data based on deep learning and blockchain according to claim 1, characterized in that, In the intelligent early warning engine processing, the convolutional layer of the adaptive fusion convolutional recurrent neural network contains three cascaded one-dimensional convolutional layers. The first convolutional layer uses 64 convolutional kernels of size 3, the second convolutional layer uses 128 convolutional kernels of size 3, and the third convolutional layer uses 64 convolutional kernels of size 3. Each convolutional operation is followed by a batch normalization layer and a modified linear unit activation function to extract the spatial correlation features of sensor data in the channel dimension.

5. The method for assetizing and intelligently warning government IoT data based on deep learning and blockchain according to claim 4, characterized in that, In the intelligent early warning engine processing, the Long Short-Term Memory (LSTM) recurrent network of the adaptively fused convolutional recurrent neural network adopts a two-layer stacked structure, with each layer having a hidden state dimension of 128 dimensions. The LSM units model temporal dependencies through forget gates, input gates, and output gates. The forget gate controls the retention ratio of the unit state at the previous time step, and the input gate controls the writing of new information, as shown in the following formula: Forgotten Gate: ; Input Gate: , ; Cell status update: ; Output gate: , ; in, , , These are the forget gate, input gate, and output gate, respectively. This represents the current state of the cell. This refers to the cell state at the previous moment. This represents the current state of the candidate cell. It is the Sigmoid activation function. This is the weight matrix. For bias terms, The state was hidden in the previous moment. Hide the current state. Enter the current time. This is element-wise multiplication.

6. The method for assetizing and intelligently warning government IoT data based on deep learning and blockchain according to claim 5, characterized in that, In the intelligent early warning engine processing, a dynamic gating mechanism is introduced between the convolutional neural network and the long short-term memory recurrent network, and the dynamic gating vector... according to Calculation, where For dynamic gating values, The state was hidden in the previous moment. Input features for the current time step. For the gated weight matrix, As a bias term, the original spatial feature vector and The gated feature vector is obtained after element-wise multiplication.

7. The method for assetizing and intelligently warning government IoT data based on deep learning and blockchain according to claim 1, characterized in that, In the assetization processing engine, the data stream is scored based on four dimensions: completeness, accuracy, timeliness, and consistency. The overall quality score is then used to determine the final quality score. Calculate using the following formula: ;in, The total score for data quality. For completeness score, To score for accuracy, For timeliness, For consistency score, These are the weighting coefficients, and the sum of the four equals 1.

8. The method for assetizing and intelligently warning government IoT data based on deep learning and blockchain according to claim 7, characterized in that, In the asset processing engine, the formula for quantifying the value of early warning categories is as follows: ,in For the value of early warning data assets, To avoid the average total loss amount from a single equipment failure, The percentage improvement in early warning accuracy compared to traditional manual monitoring methods; the formula for quantifying the value of collaborative methods is... ,in For collaborative data assets, The total cost saved through cross-departmental collaboration, For data reuse rate.

9. The method for assetizing and intelligently warning government IoT data based on deep learning and blockchain according to claim 1, characterized in that, The cosine similarity algorithm calculates the matching degree between the asset package feature vector A and the demand side's demand feature vector B. Perform as follows: ； in, Let A be the feature vector of data asset A. Let n be the demand feature vector of demander B, and n be the total dimension of the feature vector.

10. The method for assetizing and intelligently warning government IoT data based on deep learning and blockchain according to claim 9, characterized in that, The differential privacy technology provides privacy through the original analysis results. To achieve privacy protection, Laplace noise is added. The processed output is... ;in, For data after differential privacy processing, For raw data assets, For privacy budget parameters, It is Laplace noise.

11. The method for assetizing and intelligently warning government IoT data based on deep learning and blockchain according to claim 1, characterized in that, The time-enhancing loss function is calculated as follows: ； in, This is the loss value. This is the actual value. For predicted values, These are weighting coefficients. This represents the difference between the prediction and the actual result.

12. The government IoT data assetization and intelligent early warning system based on deep learning and blockchain according to any one of claims 1-11, characterized in that the system include: The data collection and preprocessing module is used to integrate data from IoT sensing devices and related auxiliary data through a dedicated government IoT interface and government cloud storage, and to preprocess the integrated data to obtain standardized data. The dual-engine parallel processing module is used to simultaneously execute intelligent early warning engine processing and asset-based processing engine processing based on standardized data; The intelligent early warning engine processes data using an adaptive fusion convolutional recurrent neural network to output device health index and graded early warning signals. The asset processing engine processes the received early warning signals and standardized data. First, it uses a blockchain smart contract to confirm the data source, the department to which it belongs, and the time of collection and generate a certificate of authenticity. Then, it scores the data streams in terms of quality from four dimensions: completeness, accuracy, timeliness, and consistency to select high-quality data streams. Finally, it calculates the data asset value of high-quality data streams by using either an early warning value quantification formula or a collaborative value quantification formula based on their application scenarios. The asset encapsulation and matching module is used to encapsulate high-quality data streams, early warning results, quality scores, value quantification results, and blockchain-based evidence certificates into government data asset packages and generate digital asset ID cards. At the same time, it uses a cosine similarity algorithm to calculate the matching degree between the asset package feature vector and the demand feature vector of the demand party. When the matching degree reaches a preset threshold, the asset package is automatically pushed to the corresponding demand party. The Trusted Circulation and Model Optimization module supports demanders in accessing the analysis results of data asset packages through federated learning and differential privacy technologies, achieving data availability without visibility, and recording the access and circulation process through the blockchain; at the same time, based on demander feedback and newly added government data, the module continuously optimizes the intelligent early warning model and asset value quantification model using the time-enhanced loss function.

Images