Method and system for collecting and tracing weight measurement data based on the Internet of Things

Abstract

The present invention provides a method and system for collecting and tracing weight measurement data based on the Internet of Things (IoT). This system involves embedding RFID tags in weights to collect environmental parameters and weight data. This system uses a deep learning algorithm to establish a mapping relationship between environmental parameters and weight, perform multi-dimensional environmental compensation and thermal expansion coefficient correction, and utilizes blockchain technology to construct an adaptive traceability chain structure to achieve secure storage and full-process traceability of measurement data. This method improves the accuracy of weight measurement, enables trusted traceability of measurement data, and ensures the security of the measurement process.

Description

Technical Field

[0001] The present invention relates to Internet of Things technology, and in particular to a method and system for collecting and tracing weight measurement data based on the Internet of Things. Background Art

[0002] Currently, weight measurement management has begun to use technical means such as electronic tags and sensors to collect data, and store and manage data through information systems. However, the existing weight measurement data collection and management methods still have the following shortcomings:

[0003] The existing measurement data acquisition system lacks the ability to comprehensively monitor and analyze environmental parameters, and is unable to accurately assess the combined impact of environmental factors such as temperature, humidity, and air pressure on the measurement accuracy of weights, which affects the reliability of measurement results.

[0004] Traditional data management methods make it difficult to ensure the authenticity and integrity of measurement data, and are prone to data tampering or loss. It is impossible to establish a reliable measurement traceability system, which is not conducive to the quality traceability and responsibility determination of measurement results.

[0005] The existing system mainly relies on experience to evaluate the service life of weights, lacks scientific prediction methods based on historical data, and is unable to detect the performance degradation of weights in a timely manner, which increases the risk of failure of measuring instruments and affects the reliability of measurement work. Summary of the Invention

[0006] The embodiments of the present invention provide a weight measurement data collection and traceability management method and system based on the Internet of Things, which can solve the problems in the prior art.

[0007] According to a first aspect of the embodiments of the present invention,

[0008] Provides a method for collecting and tracing weight measurement data based on the Internet of Things, including:

[0009] An RFID electronic tag storing the weight identity information is implanted into the weight body, and temperature, humidity, and air pressure parameters are collected through an environmental parameter sensor. The weight data is collected through a weight data acquisition module equipped with a high-precision strain gauge. The weight identity information, environmental parameters, weight data, and measurement standard instrument information are time-stamped to form the measurement raw data.

[0010] The raw measurement data is sent to an edge computing server through an IoT communication module with a data encryption function for outlier detection and data smoothing preprocessing, and the preprocessed data is transmitted to a cloud data processing server. The cloud data processing server establishes a mapping relationship between environmental parameters and weights based on a deep learning algorithm, performs multi-dimensional environmental compensation on the weight data according to the coupling relationship between temperature parameters, humidity parameters, and air pressure parameters, corrects the thermal expansion coefficient in combination with the characteristics of the weight material, generates measurement calibration data, and predicts the wear rate of the weight based on historical measurement data.

[0011] Build a secure isolation architecture, encrypt the metrology calibration data and predicted wear rate based on the secure isolation architecture, build an adaptive traceability chain structure including a basic chain and an extended chain, write the metrology calibration data and wear rate prediction results into the blockchain network through a hierarchical consensus mechanism to form a traceability information chain for the entire metrology process; establish an access control strategy based on behavioral characteristics, trigger hierarchical response measures according to anomaly scores, and provide traceability analysis results to user terminals.

[0012] In an optional embodiment,

[0013] The cloud data processing server establishes a mapping relationship between environmental parameters and weights based on a deep learning algorithm, performs multi-dimensional environmental compensation on weight data according to the coupling relationship between temperature parameters, humidity parameters, and air pressure parameters, and corrects the thermal expansion coefficient in combination with the material properties of the weights. The steps of generating metrological calibration data include:

[0014] A multi-branch attention network structure is constructed using a deep learning algorithm. The multi-branch attention network structure includes a dual-stream architecture of time stream and spatial stream, and adaptively fuses multi-scale features to obtain mapping results.

[0015] Based on the mapping results, Kalman filtering and wavelet transform are used to compensate for the spatiotemporal dimensions of the environmental parameters, and a nonlinear coupling relationship between temperature parameters, humidity parameters and air pressure parameters is constructed, and physical dimension compensation is performed in combination with material thermodynamic calculations and molecular dynamics simulations. The multi-dimensional environmental compensation also includes intelligent dimension compensation: a heterogeneous graph neural network is constructed to model the interaction relationship between environmental parameters, and temperature parameters, humidity parameters and air pressure parameters are used as graph network nodes. The dynamic correlation between parameters is captured through the attention edge convolution layer, and jump connections are introduced to retain historical interaction information. A deep reinforcement learning model with a dual reward mechanism is designed to dynamically adjust the compensation strategy. The dual reward mechanism includes an environmental stability reward and a measurement accuracy reward, and a policy gradient algorithm based on an advantage function is used to optimize the compensation action. A hybrid architecture of transfer learning and multi-task learning is constructed, and the common features of the environmental parameters are extracted through a shared representation layer. The distribution differences under different working conditions are eliminated by adaptive feature alignment. The features are expressed in a fine-grained manner in combination with a multi-head attention mechanism to complete the optimization of the compensation strategy.

[0016] Multi-scale thermal expansion correction is performed on the compensated data. At the lattice scale, a stress-strain tensor is established to describe the thermal deformation characteristics and calculate the contribution of atomic vibration to volume change. At the microscale, thermal expansion correlation characteristics are determined through dynamic simulation of grain boundary slip effects and dislocation motion analysis. At the macroscale, a multi-scale thermodynamic equation system is established to couple elastic deformation and thermal expansion.

[0017] The corrected data is input into the calibration data generation module, outlier detection and filtering denoising are performed on the data, and adaptive weighted fusion of deep learning prediction results, multi-dimensional environmental compensation results and thermal expansion correction results is performed to generate metrological calibration data.

[0018] In an optional embodiment,

[0019] A multi-branch attention network structure is constructed using a deep learning algorithm. The multi-branch attention network structure includes a dual-stream architecture of time stream and spatial stream. The steps of adaptively fusing multi-scale features to obtain a mapping result include:

[0020] A multi-branch attention network structure is constructed using DenseNet as the backbone network. The multi-branch attention network structure includes a temporal stream branch and a spatial stream branch. The multi-branch attention network structure is provided with a branch interaction module. The branch interaction module exchanges information between branches through a cross-attention mechanism, constructs a feature map similarity matrix to calculate attention weights, and controls information flow through a gating mechanism.

[0021] The time stream branch extracts time series features through causal convolution, sets a time step attention mechanism to calculate the weights of the time series features, and models the time series features through the LSTM layer; the time stream branch sets a time series feature processing module, the time series feature processing module constructs a time series multi-scale feature pyramid, and processes the features through the time series attention pooling layer;

[0022] The spatial stream branch extracts spatial features through a spatial attention convolution layer and sets a deformable convolution to deform the convolution kernel; the spatial stream branch sets a spatial relationship modeling module, which calculates the correlation relationship of spatial features through a graph attention network and introduces position encoding to mark the spatial position of features;

[0023] Adaptively fusing the features of the temporal stream branch and the spatial stream branch, including: unifying feature dimensions through a feature transformation network and aligning feature resolutions using a learnable upsampling module; constructing a dynamic weight prediction network to calculate feature fusion weights, and obtaining the final mapping result through an attention-guided feature aggregation module;

[0024] A multi-task loss function with adaptive weights is used to optimize network training. The multi-task loss function includes the Focal L1 loss of the main task and the feature consistency loss and temporal smoothness loss of the auxiliary tasks. The loss weights are calculated through an adaptive weight learning method based on task uncertainty. A staged training strategy is adopted to train the temporal stream branch, the spatial stream branch, the branch interaction module and the feature fusion module in sequence, and finally end-to-end optimization is performed.

[0025] In an optional embodiment,

[0026] The steps for predicting the wear rate of a weight based on historical metrology data include:

[0027] The historical metrological data were preprocessed using sliding median filtering to extract the quality change rate, environmental parameters, usage intensity and surface roughness features and then normalized to obtain a standardized feature sequence.

[0028] Inputting the standardized feature sequence into a bidirectional LSTM network, extracting bidirectional features through forward propagation and backward propagation, setting a time step calculation module in the bidirectional LSTM network to quantify the degree of temporal correlation, and performing a weighted combination of the bidirectional features and the degree of temporal correlation to obtain a deep feature representation of weight wear;

[0029] A multi-task learning framework is constructed based on the deep feature representation, and the multi-task learning framework includes a wear pattern classification branch and a rate prediction branch; the wear pattern classification branch divides wear into normal wear, accelerated wear and abnormal wear through a fully connected network, and adopts focal loss for optimization; the rate prediction branch predicts the wear rate in the future time period through a residual fully connected structure; a prediction reliability evaluation module is constructed in the multi-task learning framework, the mean and variance of the prediction results are output through a neural network, and the negative log-likelihood loss function is adopted for optimization; an adaptive correction mechanism is constructed based on the prediction reliability evaluation result, when the prediction uncertainty exceeds the set threshold, the data sampling density is increased and the prediction time span is shortened, the prediction deviation is calculated based on the latest measurement data, and the prediction result is corrected online by updating the correction coefficient through the exponential sliding average.

[0030] In an optional embodiment,

[0031] The steps of constructing a secure isolation architecture, encrypting the metrology calibration data and predicted wear rate based on the secure isolation architecture, building an adaptive traceability chain structure including a basic chain and an extended chain, and writing the metrology calibration data and wear rate prediction results into the blockchain network through a layered consensus mechanism to form a traceability information chain for the entire metrology process include:

[0032] Construct a secure isolation architecture consisting of a hardware root of trust layer, a trusted execution environment layer, and a secure processing layer. The hardware root of trust layer uses a trusted platform module for key storage and integrity measurement. The trusted execution environment layer divides the secure world and the normal world and performs key management. The secure processing layer is equipped with an independent data processing environment.

[0033] Inputting metrology calibration data and predicted wear rate into the secure isolation architecture for encryption, and constructing an adaptive traceability chain structure based on the encrypted data. The adaptive traceability chain structure includes a base chain for storing metrology parameters and metrology calibration data, and an extended chain for storing wear rate prediction results and metrology scenario data. The cross-chain data of the base chain and the extended chain are maintained consistent through an inter-chain data reference mechanism;

[0034] A hierarchical consensus mechanism is designed for the adaptive traceability chain structure. A practical Byzantine fault-tolerant algorithm with dynamic view switching is introduced to achieve consensus processing on the measurement traceability benchmark nodes. Proof-of-stake consensus is used to confirm data on the measurement data collection nodes. A consensus switching mechanism is used to achieve compatibility with different consensus algorithms.

[0035] The metrology calibration data is standardized and quality assessed, and a spatiotemporal feature extraction algorithm is used to obtain metrology process characteristic parameters. A metrology process correlation matrix is established by calculating the temporal correlation and spatial similarity between the metrology process characteristic parameters and the predicted wear rate. Data fusion operations are performed based on a smart contract rule engine to generate traceability information.

[0036] The traceability information is written into the blockchain network, and a secure storage and trusted traceability information chain for the entire measurement process information is established based on the adaptive traceability chain structure and hierarchical consensus mechanism.

[0037] In an optional embodiment,

[0038] The steps of implementing consensus processing on the measurement traceability benchmark nodes by introducing a practical Byzantine fault-tolerant algorithm with dynamic view switching, confirming data on the measurement data collection nodes through proof-of-stake consensus, and adopting a consensus switching mechanism to achieve compatibility with different consensus algorithms include:

[0039] A practical Byzantine fault-tolerant algorithm for dynamic view switching is designed for measurement traceability reference nodes. The algorithm calculates a view activity metric based on the number of messages, consensus time, and view stability. Based on this view activity metric, it triggers view switching and updates the master node. When a view switch message exceeding twice the fault tolerance limit plus one is received, the measurement traceability reference node enters a new view.

[0040] Calculate the stake value of the metering data collection node based on data quality, response time and storage capacity, and use the proof-of-stake consensus to determine the verification node based on the stake value and complete data confirmation;

[0041] Standardize the encoding of the state information of the Practical Byzantine Fault Tolerance algorithm and the Proof of Stake consensus, organize the state data using a Merkle tree structure, and construct a unified state description format that includes block height, state root, and consensus proof. Construct cross-consensus messages based on the unified state description format, use a multi-level message confirmation mechanism to transmit the cross-consensus messages, perform retransmission control and timeout detection on the cross-consensus messages, authenticate the source of the cross-consensus messages, and verify the legitimacy of the state mapping and the state transition proof.

[0042] Build a state version chain and set checkpoints, monitor consensus efficiency indicators, node response status, and state consistency deviation, and trigger consensus switching when the consensus efficiency indicator falls below a preset threshold, or the node response times out, or the state consistency deviation exceeds the limit. Broadcast the switching proposal to the nodes in the network and collect node confirmation information, suspend the current consensus mechanism, and activate the target consensus mechanism;

[0043] The consensus result of the measurement traceability reference node is used as the trust basis for the measurement data collection node to perform proof-of-stake consensus through the unified state description format. The data confirmed by the measurement data collection node is fed back to the measurement traceability reference node through the cross-consensus message. The state consistency of the heterogeneous consensus is maintained based on the state version chain, and a consensus switching mechanism is used to adjust between the practical Byzantine fault tolerance algorithm and the proof-of-stake consensus.

[0044] In an optional embodiment,

[0045] The steps for establishing an access control policy based on behavioral characteristics, triggering graded response measures based on anomaly scores, and providing traceability analysis results to user terminals include:

[0046] Obtaining identity features and behavioral features of user access behavior, wherein the identity features include static user identification, dynamic authentication information, and environmental attributes; and the behavioral features include temporal behavioral features, operational behavioral features, and data interaction features. Using an autoencoder, the identity features and behavioral features are dimensionality reduced to obtain a feature vector. The matching degree between the feature vector and the historical baseline is calculated based on weighted cosine similarity to obtain a real-time trust score.

[0047] Dividing the access control level based on the real-time trust score, when the real-time trust score is in a first trust range, executing a basic access control policy to allow access with original permissions; when the real-time trust score is in a second trust range, executing an enhanced access control policy to perform secondary authentication and limit access frequency and data scope; when the real-time trust score is in a third trust range, executing a strict access control policy to allow only read-only operations and require multi-factor authentication;

[0048] Dynamically adjust the control strength of the access control policy based on the changing trend of the real-time trust score during the continuous observation period, and generate an access control result including user identity information, access behavior characteristics and control policy;

[0049] Calculate an anomaly score based on the access control result and establish a hierarchical response mechanism for abnormal behavior. When abnormal behavior occurs, trigger a warning-level response, a blocking-level response, or a locking-level response based on the anomaly score. The warning-level response includes recording the abnormal event and initiating feature resampling. The blocking-level response includes suspending session permissions and requiring identity reauthentication. The locking-level response includes freezing access permissions and triggering a security audit.

[0050] Build a correlation map of user operation resources, record the occurrence time, abnormal type, response measures and disposal results of abnormal behavior, establish an abnormal behavior feature library, and push traceability analysis results including metering data query, calibration cycle reminder, wear warning and maintenance suggestions to the user terminal in real time according to the level of the abnormal score.

[0051] According to a second aspect of the embodiments of the present invention,

[0052] Provides an IoT-based weight measurement data collection and traceability management system, including:

[0053] The first unit is used to implant an RFID electronic tag storing the weight identity information into the weight body, collect temperature parameters, humidity parameters and air pressure parameters through an environmental parameter sensor, and collect weight data through a weight data acquisition module equipped with a high-precision strain gauge; the weight identity information, environmental parameters, weight data and measurement standard instrument information are timestamped to form the measurement raw data;

[0054] The second unit is used to send the raw measurement data to the edge computing server through an Internet of Things communication module with a data encryption function for preprocessing of outlier detection and data smoothing, and transmit the preprocessed data to the cloud data processing server. The cloud data processing server establishes a mapping relationship between environmental parameters and weights based on a deep learning algorithm, performs multi-dimensional environmental compensation on the weight data according to the coupling relationship between temperature parameters, humidity parameters and air pressure parameters, corrects the thermal expansion coefficient in combination with the characteristics of the weight material, generates measurement calibration data, and predicts the wear rate of the weight based on historical measurement data;

[0055] The third unit is used to build a secure isolation architecture, encrypt the metrology calibration data and predicted wear rate based on the secure isolation architecture, build an adaptive traceability chain structure including a basic chain and an extended chain, write the metrology calibration data and wear rate prediction results into the blockchain network through a hierarchical consensus mechanism to form a traceability information chain for the entire metrology process; establish an access control strategy based on behavioral characteristics, trigger hierarchical response measures according to anomaly scores, and provide traceability analysis results to user terminals.

[0056] According to a third aspect of the embodiments of the present invention,

[0057] An electronic device is provided, comprising:

[0058] processor;

[0059] a memory for storing processor-executable instructions;

[0060] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0061] According to a fourth aspect of the embodiments of the present invention,

[0062] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0063] The present invention realizes the automatic collection and real-time monitoring of weight measurement data by implanting RFID electronic tags in weights and combining multiple environmental sensors to collect data, thereby improving the accuracy and efficiency of data collection, and ensuring the authenticity and traceability of data through timestamp marking.

[0064] The present invention uses a deep learning algorithm to establish a mapping relationship between environmental parameters and weights. It improves the accuracy of measurement data through multi-dimensional environmental compensation and thermal expansion coefficient correction, and can predict the wear rate of weights, providing a scientific basis for the maintenance and replacement of weights, thereby extending the service life of weights.

[0065] The present invention constructs a secure isolation architecture and adaptive traceability chain structure based on blockchain technology, ensures the security and non-tamperability of measurement data through a layered consensus mechanism, and establishes an access control strategy based on behavioral characteristics, realizing trusted traceability of the entire measurement process and improving the standardization and reliability of measurement management. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a flow chart of a method for collecting and tracing weight measurement data based on the Internet of Things according to an embodiment of the present invention;

[0067] Figure 2 Multi-level access control flow chart based on behavioral characteristics;

[0068] Figure 3 This is a structural diagram of a weight measurement data collection and traceability management system based on the Internet of Things according to an embodiment of the present invention. DETAILED DESCRIPTION

[0069] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0070] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0071] Figure 1 FIG is a flow chart of a method for collecting and tracing weight measurement data based on the Internet of Things according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0072] An RFID electronic tag storing the weight identity information is implanted into the weight body, and temperature, humidity, and air pressure parameters are collected through an environmental parameter sensor. The weight data is collected through a weight data acquisition module equipped with a high-precision strain gauge. The weight identity information, environmental parameters, weight data, and measurement standard instrument information are time-stamped to form the measurement raw data.

[0073] The raw measurement data is sent to an edge computing server through an IoT communication module with a data encryption function for outlier detection and data smoothing preprocessing, and the preprocessed data is transmitted to a cloud data processing server. The cloud data processing server establishes a mapping relationship between environmental parameters and weights based on a deep learning algorithm, performs multi-dimensional environmental compensation on the weight data according to the coupling relationship between temperature parameters, humidity parameters, and air pressure parameters, corrects the thermal expansion coefficient in combination with the characteristics of the weight material, generates measurement calibration data, and predicts the wear rate of the weight based on historical measurement data.

[0074] Build a secure isolation architecture, encrypt the metrology calibration data and predicted wear rate based on the secure isolation architecture, build an adaptive traceability chain structure including a basic chain and an extended chain, write the metrology calibration data and wear rate prediction results into the blockchain network through a hierarchical consensus mechanism to form a traceability information chain for the entire metrology process; establish an access control strategy based on behavioral characteristics, trigger hierarchical response measures according to anomaly scores, and provide traceability analysis results to user terminals.

[0075] In an optional embodiment,

[0076] The cloud data processing server establishes a mapping relationship between environmental parameters and weights based on a deep learning algorithm, performs multi-dimensional environmental compensation on weight data according to the coupling relationship between temperature parameters, humidity parameters, and air pressure parameters, and corrects the thermal expansion coefficient in combination with the material properties of the weights. The steps of generating metrological calibration data include:

[0077] A multi-branch attention network structure is constructed using a deep learning algorithm. The multi-branch attention network structure includes a dual-stream architecture of time stream and spatial stream, and adaptively fuses multi-scale features to obtain mapping results.

[0078] Based on the mapping results, Kalman filtering and wavelet transform are used to compensate for the spatiotemporal dimensions of the environmental parameters, and a nonlinear coupling relationship between temperature parameters, humidity parameters and air pressure parameters is constructed, and physical dimension compensation is performed in combination with material thermodynamic calculations and molecular dynamics simulations. The multi-dimensional environmental compensation also includes intelligent dimension compensation: a heterogeneous graph neural network is constructed to model the interaction relationship between environmental parameters, and temperature parameters, humidity parameters and air pressure parameters are used as graph network nodes. The dynamic correlation between parameters is captured through the attention edge convolution layer, and jump connections are introduced to retain historical interaction information. A deep reinforcement learning model with a dual reward mechanism is designed to dynamically adjust the compensation strategy. The dual reward mechanism includes an environmental stability reward and a measurement accuracy reward, and a policy gradient algorithm based on an advantage function is used to optimize the compensation action. A hybrid architecture of transfer learning and multi-task learning is constructed, and the common features of the environmental parameters are extracted through a shared representation layer. The distribution differences under different working conditions are eliminated by adaptive feature alignment. The features are expressed in a fine-grained manner in combination with a multi-head attention mechanism to complete the optimization of the compensation strategy.

[0079] Multi-scale thermal expansion correction is performed on the compensated data. At the lattice scale, a stress-strain tensor is established to describe the thermal deformation characteristics and calculate the contribution of atomic vibration to volume change. At the microscale, thermal expansion correlation characteristics are determined through dynamic simulation of grain boundary slip effects and dislocation motion analysis. At the macroscale, a multi-scale thermodynamic equation system is established to couple elastic deformation and thermal expansion.

[0080] The corrected data is input into the calibration data generation module, outlier detection and filtering denoising are performed on the data, and adaptive weighted fusion of deep learning prediction results, multi-dimensional environmental compensation results and thermal expansion correction results is performed to generate metrological calibration data.

[0081] For example, the cloud-based data processing server first obtains real-time data on the temperature, humidity, and air pressure parameters of the weight measurement environment. A deep learning algorithm is used to construct a multi-branch attention network, which includes a dual-stream architecture of time and space streams. The time stream branch uses a long short-term memory network to process the temporal variation characteristics of environmental parameters. The number of network units is set to 128, and the time window length is 24 hours. The spatial stream branch uses a three-dimensional convolutional neural network to extract the spatial distribution characteristics of environmental parameters. The convolution kernel size is 3×3×3, the stride is 1, and the padding method is SAME. The features of the two branches are fused using an adaptive weighting method, and the weight coefficients are dynamically updated using the backpropagation algorithm.

[0082] The obtained mapping results are subjected to multi-dimensional environmental compensation. First, the environmental parameters are denoised using a Kalman filter, with the diagonal elements of the observation noise covariance matrix set to 0.01. Wavelet transform is then used for multi-scale decomposition, with the db4 wavelet basis function selected and three decomposition layers. On this basis, a heterogeneous graph neural network is constructed, with temperature, humidity, and air pressure parameters used as graph network nodes, with a node feature dimension of 64. Dynamic correlation features between parameters are extracted using two layers of graph attention convolutional layers, with the number of attention heads set to 8 and the output feature dimension of each layer being 32. Residual connections are introduced to preserve historical interaction information, with a skip connection step size of 2.

[0083] A deep reinforcement learning model based on a dual reward mechanism was designed. The environmental stability reward is calculated based on the magnitude of parameter fluctuations, and the measurement accuracy reward is determined based on the deviation from the standard value. An actor-critic network architecture is used, with the actor network outputting compensatory actions and the critic network evaluating the value of those actions. The network has a three-layer fully connected architecture, with 256, 128, and 64 hidden neurons, respectively, and a Reluctant Unit (ReLU) activation function.

[0084] In the thermal expansion correction stage, the stress-strain relationship model is first established at the lattice scale. Taking the copper-based weight material as an example, the influence of atomic vibration on volume change is calculated by molecular dynamics simulation at room temperature of 293K. The simulation time step is set to 1 femtosecond and the total simulation time is 100 picoseconds. At the microscale, the dislocation dynamics method is used to analyze the grain boundary slip effect and calculate the dislocation density evolution law. The initial dislocation density is set to 10 8 / m 2 A multi-scale coupling model is established at the macro scale to calculate elastic deformation and thermal expansion in a unified manner.

[0085] Finally, during the calibration data generation phase, the isolation forest algorithm was used for outlier detection, with a contamination threshold of 0.1. A Butterworth low-pass filter was used for signal denoising, with a cutoff frequency of 0.1 Hz. A weighted fusion of the deep learning prediction results, environmental compensation results, and thermal expansion correction results was performed, with weight coefficients determined through cross-validation to be 0.4, 0.35, and 0.25, respectively.

[0086] The present invention realizes the precise mapping of environmental parameters and weight by constructing a multi-branch attention network. The dual-stream architecture can simultaneously capture the temporal variation characteristics and spatial distribution characteristics, thereby improving the prediction accuracy of the model. The adaptive feature fusion mechanism can dynamically adjust the feature weights according to different working conditions, thereby enhancing the generalization ability of the model. The multi-dimensional environmental compensation strategy comprehensively considers the compensation effects of the spatiotemporal dimension and the intelligent dimension. The heterogeneous graph neural network effectively models the complex interaction relationship between environmental parameters. The reinforcement learning model of the dual reward mechanism can adaptively optimize the compensation strategy, thereby improving the accuracy and robustness of environmental compensation. The multi-scale thermal expansion correction method realizes full-scale analysis from the lattice scale to the macroscopic scale, accurately describes the thermal deformation characteristics of the material, and combines molecular dynamics simulation and dislocation dynamics analysis to deeply reveal the microscopic mechanism, thereby improving the scientificity and reliability of the thermal expansion coefficient correction. Finally, through outlier detection and adaptive weighted fusion, the accuracy and stability of the calibration data are ensured.

[0087] In an optional embodiment,

[0088] A multi-branch attention network structure is constructed using a deep learning algorithm. The multi-branch attention network structure includes a dual-stream architecture of time stream and spatial stream. The steps of adaptively fusing multi-scale features to obtain a mapping result include:

[0089] A multi-branch attention network structure is constructed using DenseNet as the backbone network. The multi-branch attention network structure includes a temporal stream branch and a spatial stream branch. The multi-branch attention network structure is provided with a branch interaction module. The branch interaction module exchanges information between branches through a cross-attention mechanism, constructs a feature map similarity matrix to calculate attention weights, and controls information flow through a gating mechanism.

[0090] The time stream branch extracts time series features through causal convolution, sets a time step attention mechanism to calculate the weights of the time series features, and models the time series features through the LSTM layer; the time stream branch sets a time series feature processing module, the time series feature processing module constructs a time series multi-scale feature pyramid, and processes the features through the time series attention pooling layer;

[0091] The spatial stream branch extracts spatial features through a spatial attention convolution layer and sets a deformable convolution to deform the convolution kernel; the spatial stream branch sets a spatial relationship modeling module, which calculates the correlation relationship of spatial features through a graph attention network and introduces position encoding to mark the spatial position of features;

[0092] Adaptively fusing the features of the temporal stream branch and the spatial stream branch, including: unifying feature dimensions through a feature transformation network and aligning feature resolutions using a learnable upsampling module; constructing a dynamic weight prediction network to calculate feature fusion weights, and obtaining the final mapping result through an attention-guided feature aggregation module;

[0093] A multi-task loss function with adaptive weights is used to optimize network training. The multi-task loss function includes the Focal L1 loss of the main task and the feature consistency loss and temporal smoothness loss of the auxiliary tasks. The loss weights are calculated through an adaptive weight learning method based on task uncertainty. A staged training strategy is adopted to train the temporal stream branch, the spatial stream branch, the branch interaction module and the feature fusion module in sequence, and finally end-to-end optimization is performed.

[0094] For example, we first constructed the basic architecture of a multi-branch attention network structure. We chose DenseNet-121 as the backbone network, which contains four dense blocks, each with 6, 12, 24, and 16 densely connected layers, respectively. Based on this backbone network, we constructed the temporal and spatial branches.

[0095] In the time stream branch, causal convolution is used to extract temporal features. This causal convolution uses a 3×1 one-dimensional convolution kernel with a stride of 1 and causal padding. The input sequence length is set to 16 frames, and the feature dimension is 256. The time-step attention mechanism weights each time step by calculating its importance score. Specifically, the feature vector of each time step is passed through two fully connected layers to obtain an attention score, which is then normalized using softmax to obtain a weight coefficient. A two-layer LSTM network is then used for temporal modeling, with a hidden layer dimension of 512.

[0096] The temporal feature processing module constructs a three-layer temporal feature pyramid with scale ratios of 1:2:4. Each layer of features is processed through temporal attention pooling with pooling windows of 1, 2, and 4, respectively, to obtain a multi-scale temporal representation.

[0097] In the spatial stream branch, the spatial attention convolution layer uses a 3×3 convolution kernel with 256 channels. The offset of the deformable convolution is predicted by an independent convolution branch. The spatial relationship modeling module first divides the feature map into a 7×7 grid, with each grid cell representing a node in the graph. An 8-head attention mechanism is used to calculate the correlation strength between nodes, with an attention head dimension of 32. Positional encoding uses sine-cosine encoding, with the encoding dimension being the same as the feature dimension.

[0098] In the branch interaction module, a cross-attention mechanism calculates a similarity matrix between the feature maps of two branches. Each element of the similarity matrix represents the degree of match between the corresponding position of the source feature map and the target feature map. The gating mechanism is implemented by a two-layer perceptron, adaptively adjusting the information flow ratio based on the input features.

[0099] The feature fusion stage first unifies the feature dimension to 256 using 1×1 convolution. The learnable upsampling module uses a deconvolution layer with a stride of 2 and a kernel size of 4. The dynamic weight prediction network consists of three convolutional layers, with the same number of output channels as the number of features. The attention-guided feature aggregation module calculates the attention map between features and performs weighted fusion based on it.

[0100] For network training, the Focal L1 loss parameters for the main task, alpha, were set to 2, and gamma to 0.25. The feature consistency loss used cosine similarity to measure the similarity of features between two branches. The temporal smoothness loss calculated the differences in features between adjacent time steps. Adaptive weight learning determined the weight coefficients based on the gradient norm of each task.

[0101] The training strategy consists of three phases: the first phase pre-trains the temporal and spatial branches separately, with each branch trained for 50 epochs; the second phase incorporates a branch interaction module and trains for 30 epochs; the third phase introduces a feature fusion module and performs end-to-end training for 20 epochs. The initial learning rate is set to 0.001, and a cosine annealing strategy is used for adjustment.

[0102] Through the design of a dual-stream architecture and a multi-branch attention mechanism, the present invention can fully capture the temporal dynamic features and spatial structural features of the input data, improving the feature expression capability and model performance. The branch interaction module and the adaptive feature fusion strategy achieve the effective fusion of spatiotemporal features, enhancing the model's ability to understand complex scenes; the use of multi-scale feature extraction and an adaptive fusion mechanism improves the model's adaptability to targets of different scales and the robustness of feature expression; the introduction of deformable convolution and graph attention networks improves the model's ability to model spatial geometric transformations; the multi-task learning framework and adaptive weight optimization strategy achieve stable convergence of model training and improve the model's generalization capability. The phased training strategy ensures the full optimization of each module, ultimately achieving end-to-end performance improvement.

[0103] In an optional embodiment,

[0104] The steps for predicting the wear rate of a weight based on historical metrology data include:

[0105] The historical metrological data were preprocessed using sliding median filtering to extract the quality change rate, environmental parameters, usage intensity and surface roughness features and then normalized to obtain a standardized feature sequence.

[0106] Inputting the standardized feature sequence into a bidirectional LSTM network, extracting bidirectional features through forward propagation and backward propagation, setting a time step calculation module in the bidirectional LSTM network to quantify the degree of temporal correlation, and performing a weighted combination of the bidirectional features and the degree of temporal correlation to obtain a deep feature representation of weight wear;

[0107] A multi-task learning framework is constructed based on the deep feature representation, and the multi-task learning framework includes a wear pattern classification branch and a rate prediction branch; the wear pattern classification branch divides wear into normal wear, accelerated wear and abnormal wear through a fully connected network, and adopts focal loss for optimization; the rate prediction branch predicts the wear rate in the future time period through a residual fully connected structure; a prediction reliability evaluation module is constructed in the multi-task learning framework, the mean and variance of the prediction results are output through a neural network, and the negative log-likelihood loss function is adopted for optimization; an adaptive correction mechanism is constructed based on the prediction reliability evaluation result, when the prediction uncertainty exceeds the set threshold, the data sampling density is increased and the prediction time span is shortened, the prediction deviation is calculated based on the latest measurement data, and the prediction result is corrected online by updating the correction coefficient through the exponential sliding average.

[0108] For example, historical metrology data is first preprocessed. The raw quality data is processed using a sliding median filter with a sliding window size of seven data points, effectively removing outliers and noise. Features are extracted from the processed data, including quality change rate (percentage of quality loss per month), environmental parameters (temperature, humidity, air pressure, etc.), usage intensity (number of uses per month), and surface roughness. All features are normalized to a value between 0 and 1.

[0109] Next, a bidirectional LSTM network was constructed for feature extraction. The network consists of two bidirectional LSTM layers, each with 128 neurons. The forward LSTM layer extracts time series features from the past to the present, while the reverse LSTM layer extracts features from the present to the past. A time step calculation module was set up, and an attention mechanism was used to calculate the degree of correlation between data at different time steps. Specifically, the data from the past 12 months was used as the input sequence, and the correlation weight between each time step and the current prediction moment was calculated. The bidirectional features were multiplied and superimposed with the correlation weights to obtain a deep feature representation of weight wear.

[0110] A multi-task learning framework was then constructed. The wear pattern classification branch uses a three-layer fully connected network with 256, 128, and 3 neurons, respectively. The softmax function outputs the probabilities of the three wear patterns. The rate prediction branch uses a residual structure consisting of two parallel fully connected layers. The outputs are summed to obtain the final prediction value. The prediction reliability assessment module also uses a fully connected network structure and outputs the mean and variance of the prediction results. When the prediction variance exceeds 0.1, an adaptive correction mechanism is triggered.

[0111] In practical applications, consider a set of 1kg standard weights. Raw mass data indicates a 0.5mg mass loss over a year. Sliding median filtering yields a smooth mass change curve. Environmental parameter records indicate an ambient temperature of 20-25°C, relative humidity of 45-55%, and atmospheric pressure of 98-102kPa. Usage intensity averages 50 times per month, with a surface roughness Ra of 0.4μm. Inputting this data into the trained model, the wear rate over the next three months is predicted to be 0.03mg / month, with a prediction variance of 0.08, indicating a reliable prediction. Wear pattern classification results indicate normal wear, with a confidence level of 0.92.

[0112] The present invention achieves high-quality data preprocessing through sliding median filtering and feature normalization, effectively reduces noise interference, and improves the accuracy and stability of subsequent predictions; adopts a bidirectional LSTM network combined with a time step calculation module to fully utilize the temporal information of historical data, accurately captures the dynamic change law of weight wear, and enhances the feature extraction ability of the model; based on a multi-task learning framework, it realizes the coordinated optimization of wear pattern classification and rate prediction, and introduces prediction reliability evaluation and adaptive correction mechanism, which significantly improves the accuracy and reliability of the prediction results, providing strong support for weight service life assessment and maintenance decision-making.

[0113] In an optional embodiment,

[0114] The steps of constructing a secure isolation architecture, encrypting the metrology calibration data and predicted wear rate based on the secure isolation architecture, building an adaptive traceability chain structure including a basic chain and an extended chain, and writing the metrology calibration data and wear rate prediction results into the blockchain network through a layered consensus mechanism to form a traceability information chain for the entire metrology process include:

[0115] Construct a secure isolation architecture consisting of a hardware root of trust layer, a trusted execution environment layer, and a secure processing layer. The hardware root of trust layer uses a trusted platform module for key storage and integrity measurement. The trusted execution environment layer divides the secure world and the normal world and performs key management. The secure processing layer is equipped with an independent data processing environment.

[0116] Inputting metrology calibration data and predicted wear rate into the secure isolation architecture for encryption, and constructing an adaptive traceability chain structure based on the encrypted data. The adaptive traceability chain structure includes a base chain for storing metrology parameters and metrology calibration data, and an extended chain for storing wear rate prediction results and metrology scenario data. The cross-chain data of the base chain and the extended chain are maintained consistent through an inter-chain data reference mechanism;

[0117] A hierarchical consensus mechanism is designed for the adaptive traceability chain structure. A practical Byzantine fault-tolerant algorithm with dynamic view switching is introduced to achieve consensus processing on the measurement traceability benchmark nodes. Proof-of-stake consensus is used to confirm data on the measurement data collection nodes. A consensus switching mechanism is used to achieve compatibility with different consensus algorithms.

[0118] The metrology calibration data is standardized and quality assessed, and a spatiotemporal feature extraction algorithm is used to obtain metrology process characteristic parameters. A metrology process correlation matrix is established by calculating the temporal correlation and spatial similarity between the metrology process characteristic parameters and the predicted wear rate. Data fusion operations are performed based on a smart contract rule engine to generate traceability information.

[0119] The traceability information is written into the blockchain network, and a secure storage and trusted traceability information chain for the entire measurement process information is established based on the adaptive traceability chain structure and hierarchical consensus mechanism.

[0120] For example, a three-layer security isolation architecture is first constructed. The hardware root of trust layer uses a Trusted Platform Module (TPM) chip to store keys and achieves trusted boot by measuring the integrity of the boot process and operating environment. The Trusted Execution Environment layer uses ARM TrustZone technology to divide the secure world into a normal world and a secure world. The secure world is responsible for key management and sensitive data processing, while the normal world runs general applications. The secure processing layer establishes an independent data processing environment and uses memory isolation technology to ensure data processing security.

[0121] Input metrological calibration data is encrypted. For example, in the case of measuring instrument calibration, environmental parameters such as temperature and humidity, along with calibration data, are encrypted using the AES-256 algorithm and stored in the base chain. Predicted wear rate data is homomorphically encrypted and stored in the extended chain. The base chain and extended chain are linked via hash pointers to maintain cross-chain data consistency.

[0122] In the consensus mechanism design, the measurement traceability benchmark node uses a dynamic view PBFT algorithm (Practical Byzantine Fault Tolerance). For temperature measurement, for example, when a node anomaly is detected, fault tolerance is achieved by dynamically adjusting the view membership. The measurement data collection node uses a PoS consensus mechanism, determining block weight based on node reputation. The two consensus mechanisms are compatible through a layered design.

[0123] In terms of metrological data processing, calibration data is normalized and data quality assessed. Taking pressure measurement as an example, time-series characteristic parameters such as pressure and temperature are extracted during the measurement process. The temporal correlation and spatial similarity between these characteristic parameters and the predicted wear rate are calculated to generate a correlation matrix. Based on the smart contract rule engine, the associated data is integrated to form complete traceability information.

[0124] Finally, the traceability information is written into the blockchain network. Distributed storage technology is used to store data in shards on different nodes. A consensus mechanism ensures data consistency and enables trusted traceability throughout the entire measurement process.

[0125] The present invention realizes the secure storage and trusted processing of measurement data through a three-layer security isolation architecture, effectively prevents data leakage and tampering, and ensures the integrity and confidentiality of measurement data; based on the adaptive traceability chain structure and layered consensus mechanism, it realizes the trusted storage and efficient consensus of measurement data, improves the fault tolerance and scalability of the system, and ensures the reliability of traceability information; uses the smart contract rule engine to integrate and analyze the measurement data, establishes the correlation between the measurement process characteristics and wear prediction, improves the accuracy and availability of measurement traceability, and provides a reliable basis for the management of measuring instruments.

[0126] In an optional embodiment,

[0127] The steps of implementing consensus processing on the measurement traceability benchmark nodes by introducing a practical Byzantine fault-tolerant algorithm with dynamic view switching, confirming data on the measurement data collection nodes through proof-of-stake consensus, and adopting a consensus switching mechanism to achieve compatibility with different consensus algorithms include:

[0128] A practical Byzantine fault-tolerant algorithm for dynamic view switching is designed for measurement traceability reference nodes. The algorithm calculates a view activity metric based on the number of messages, consensus time, and view stability. Based on this view activity metric, it triggers view switching and updates the master node. When a view switch message exceeding twice the fault tolerance limit plus one is received, the measurement traceability reference node enters a new view.

[0129] Calculate the stake value of the metering data collection node based on data quality, response time and storage capacity, and use the proof-of-stake consensus to determine the verification node based on the stake value and complete data confirmation;

[0130] Standardize the encoding of the state information of the Practical Byzantine Fault Tolerance algorithm and the Proof of Stake consensus, organize the state data using a Merkle tree structure, and construct a unified state description format that includes block height, state root, and consensus proof. Construct cross-consensus messages based on the unified state description format, use a multi-level message confirmation mechanism to transmit the cross-consensus messages, perform retransmission control and timeout detection on the cross-consensus messages, authenticate the source of the cross-consensus messages, and verify the legitimacy of the state mapping and the state transition proof.

[0131] Build a state version chain and set checkpoints, monitor consensus efficiency indicators, node response status, and state consistency deviation, and trigger consensus switching when the consensus efficiency indicator falls below a preset threshold, or the node response times out, or the state consistency deviation exceeds the limit. Broadcast the switching proposal to the nodes in the network and collect node confirmation information, suspend the current consensus mechanism, and activate the target consensus mechanism;

[0132] The consensus result of the measurement traceability reference node is used as the trust basis for the measurement data collection node to perform proof-of-stake consensus through the unified state description format. The data confirmed by the measurement data collection node is fed back to the measurement traceability reference node through the cross-consensus message. The state consistency of the heterogeneous consensus is maintained based on the state version chain, and a consensus switching mechanism is used to adjust between the practical Byzantine fault tolerance algorithm and the proof-of-stake consensus.

[0133] For example, a practical Byzantine fault-tolerant algorithm for dynamic view switching first achieves consensus on traceability benchmark nodes by calculating a view liveness metric. This view liveness metric comprises three dimensions: message count, consensus time, and view stability assessment. Message count is measured by recording the total number of prepare and confirm messages transmitted between nodes within each view. When the message count exceeds a preset threshold, it indicates excessive communication overhead for the current view. Consensus time is calculated by measuring the time interval from proposal generation to consensus. If it exceeds three seconds, consensus efficiency is considered low. View stability is assessed based on view switch frequency. If the number of view switches exceeds five within ten minutes, the network is considered unstable. Based on the combined evaluation of these metrics, a value below 0.6 triggers the view switch process. During a view switch, a node broadcasts a digitally signed switch request. Upon receiving a switch request equal to twice the fault tolerance limit plus one, it enters the new view.

[0134] The Proof-of-Stake consensus mechanism manages metering data collection nodes. First, node stake is calculated, encompassing three dimensions: data quality score, average response time, and available storage capacity. The data quality score is based on data integrity checks and outlier detection results, ranging from 0 to 100. Response time measures the average latency for a node to process a request, and must not exceed 200 milliseconds. Storage capacity examines a node's data storage capabilities, requiring a minimum of 100GB of reserved space. The scores for these three dimensions are weighted to create a composite stake, with weights of 0.4, 0.3, and 0.3, respectively. The top 30% of nodes by stake are selected as validating nodes to participate in consensus.

[0135] To enable interoperability among heterogeneous consensuses, a unified state description format is adopted. Consensus state information is encoded as a fixed-length byte array, consisting of the block height, state root hash, and consensus proof signature. State data is stored using a Merkle tree, with leaf nodes storing specific state values and intermediate nodes storing child node hash values. Cross-consensus message transmission utilizes a three-phase confirmation mechanism: the sender first broadcasts a preliminary message, then sends a commit message after receiving more than half of the confirmations, and finally waits for the recipient's completion confirmation. During message transmission, heartbeat detection is used to detect timeouts and retransmissions, with a timeout threshold set to 1 second.

[0136] A state version chain is constructed to track the consensus transition process. A state version is generated for each consensus cycle, recording information such as the consensus type, participating nodes, and state snapshots. A checkpoint interval is set to 100 blocks, and state consistency is verified at checkpoints. Three metrics are monitored: consensus efficiency (transactions per second), node response time, and state consistency deviation. A consensus transition is triggered when consensus efficiency falls below 1000 TPS, when more than 20% of nodes experience response timeouts, or when state consistency deviation exceeds 5%. During the transition, the current consensus is suspended for 30 seconds until the new consensus mechanism is established and stable.

[0137] The consensus results of the measurement traceability reference node are transmitted to the measurement data collection node in a unified state format, serving as the trusted basis for its proof-of-stake consensus. The measurement data confirmed by the data collection node is also fed back to the reference node via cross-consensus messages. The state version chain records state evolution throughout the entire process, ensuring state consistency across heterogeneous consensuses. A consensus switching mechanism enables dynamic adjustment between the two consensus algorithms.

[0138] The present invention improves the credibility and consensus efficiency of metering data through a two-layer consensus architecture that combines dynamic view switching and proof of stake, and solves the problem that traditional single consensus mechanisms are difficult to balance security and performance; adopts a unified state description format and a cross-consensus message mechanism to achieve interoperability and state synchronization between heterogeneous consensuses, overcoming the obstacles to information exchange between different consensus algorithms; based on the design of the state version chain and the consensus switching mechanism, the system can dynamically adjust the consensus strategy according to the operating status, enhance the adaptability and reliability of the system, and ensure the continuity and consistency of metering data.

[0139] In an optional embodiment,

[0140] The steps for establishing an access control policy based on behavioral characteristics, triggering graded response measures based on anomaly scores, and providing traceability analysis results to user terminals include:

[0141] Obtaining identity features and behavioral features of user access behavior, wherein the identity features include static user identification, dynamic authentication information, and environmental attributes; and the behavioral features include temporal behavioral features, operational behavioral features, and data interaction features. Using an autoencoder, the identity features and behavioral features are dimensionality reduced to obtain a feature vector. The matching degree between the feature vector and the historical baseline is calculated based on weighted cosine similarity to obtain a real-time trust score.

[0142] Dividing the access control level based on the real-time trust score, when the real-time trust score is in a first trust range, executing a basic access control policy to allow access with original permissions; when the real-time trust score is in a second trust range, executing an enhanced access control policy to perform secondary authentication and limit access frequency and data scope; when the real-time trust score is in a third trust range, executing a strict access control policy to allow only read-only operations and require multi-factor authentication;

[0143] Dynamically adjust the control strength of the access control policy based on the changing trend of the real-time trust score during the continuous observation period, and generate an access control result including user identity information, access behavior characteristics and control policy;

[0144] Calculate an anomaly score based on the access control result and establish a hierarchical response mechanism for abnormal behavior. When abnormal behavior occurs, trigger a warning-level response, a blocking-level response, or a locking-level response based on the anomaly score. The warning-level response includes recording the abnormal event and initiating feature resampling. The blocking-level response includes suspending session permissions and requiring identity reauthentication. The locking-level response includes freezing access permissions and triggering a security audit.

[0145] Build a correlation map of user operation resources, record the occurrence time, abnormal type, response measures and disposal results of abnormal behavior, establish an abnormal behavior feature library, and push traceability analysis results including metering data query, calibration cycle reminder, wear warning and maintenance suggestions to the user terminal in real time according to the level of the abnormal score.

[0146] For example, Figure 2 As shown in the flowchart for multi-level access control based on behavioral characteristics, user access behavior data must first be collected. Identity characteristics collected include static identification information such as user account, login credentials, IP address, and device fingerprint; authentication information such as dynamic passwords and biometrics; and environmental attributes such as time and location. Behavioral characteristics collected cover temporal characteristics such as the time interval and operation sequence of user operations; operational characteristics such as command type and parameter values; and interaction characteristics such as data access volume and data flow direction.

[0147] During the feature processing phase, an autoencoder is used for feature dimensionality reduction. The input layer receives the original feature vector, compresses the high-dimensional features into a low-dimensional latent space through a multi-layer encoder, and then restores and reconstructs the features through a decoder. For example, in a user login scenario, the original input features include 20 dimensions, such as login time, location, and device. These are compressed into an 8-dimensional feature vector using a 64-32-16-8 encoder structure. The degree of match between the feature vector and the user's historical behavior baseline is calculated using weighted cosine similarity, resulting in a real-time trust score between 0 and 1.

[0148] Access control policies are graded based on trust scores. When the trust score is between 0.8 and 1.0, basic access control is implemented, allowing users to access system resources normally. When the score is between 0.5 and 0.8, enhanced access control is triggered, requiring users to perform secondary authentication such as SMS verification codes. Access frequency is limited to no more than 10 times per minute, and data queries are restricted to the user's department. When the score is below 0.5, strict access control is activated, allowing only read-only data operations and requiring multi-factor authentication such as facial recognition.

[0149] The system continuously monitors changes in the trust score within a 30-minute observation period. If the score shows a downward trend, access control strength is increased accordingly; if the score stabilizes and rebounds, restrictions are relaxed appropriately. Ultimately, access control results are generated that include user identification, behavioral characteristics, and control policies.

[0150] The tiered response mechanism for abnormal behavior calculates an anomaly score based on access control results. A score between 0.3 and 0.5 triggers a warning-level response, logging the abnormal event and initiating feature resampling. A score between 0.5 and 0.8 triggers a blocking-level response, suspending the current session and requiring identity reauthentication. A score exceeding 0.8 triggers a lock-level response, freezing the account and conducting a security audit.

[0151] Tracing analysis is built on a user-operation-resource correlation graph. It records information such as the time, type, and response to abnormal behavior, and builds an abnormality feature library for matching similar cases. Based on the anomaly score, analysis results are pushed to the user terminal, including the timeline of the abnormal operation, the scope of impact, and recommended actions.

[0152] Through multi-dimensional feature analysis and adaptive access control, the present invention improves the accuracy and timeliness of abnormal behavior detection, effectively preventing internal threats and data leakage risks; based on a hierarchical response mechanism, it achieves precise management and control, maintaining system availability to the greatest extent while ensuring security, improving user experience and work efficiency; real-time feedback on traceability analysis results helps users understand and improve operational behavior, while providing decision-making support for security managers, promoting the improvement of security awareness and the optimization of management efficiency.

[0153] Figure 3 FIG is a structural diagram of a weight measurement data collection and traceability management system based on the Internet of Things according to an embodiment of the present invention. Figure 3 As shown, the system includes:

[0154] The first unit is used to implant an RFID electronic tag storing the weight identity information into the weight body, collect temperature parameters, humidity parameters and air pressure parameters through an environmental parameter sensor, and collect weight data through a weight data acquisition module equipped with a high-precision strain gauge; the weight identity information, environmental parameters, weight data and measurement standard instrument information are timestamped to form the measurement raw data;

[0155] The second unit is used to send the raw measurement data to the edge computing server through an Internet of Things communication module with a data encryption function for preprocessing of outlier detection and data smoothing, and transmit the preprocessed data to the cloud data processing server. The cloud data processing server establishes a mapping relationship between environmental parameters and weights based on a deep learning algorithm, performs multi-dimensional environmental compensation on the weight data according to the coupling relationship between temperature parameters, humidity parameters and air pressure parameters, corrects the thermal expansion coefficient in combination with the characteristics of the weight material, generates measurement calibration data, and predicts the wear rate of the weight based on historical measurement data;

[0156] The third unit is used to build a secure isolation architecture, encrypt the metrology calibration data and predicted wear rate based on the secure isolation architecture, build an adaptive traceability chain structure including a basic chain and an extended chain, write the metrology calibration data and wear rate prediction results into the blockchain network through a hierarchical consensus mechanism to form a traceability information chain for the entire metrology process; establish an access control strategy based on behavioral characteristics, trigger hierarchical response measures according to anomaly scores, and provide traceability analysis results to user terminals.

[0157] According to a third aspect of the embodiments of the present invention,

[0158] An electronic device is provided, comprising:

[0159] processor;

[0160] a memory for storing processor-executable instructions;

[0161] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0162] According to a fourth aspect of the embodiments of the present invention,

[0163] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0164] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. The method for collecting and tracing weight measurement data based on the Internet of Things is characterized by: include: An RFID electronic tag storing the identity information of the weight is implanted into the weight body, and the temperature parameter, humidity parameter and air pressure parameter are collected through the environmental parameter sensor to collect the weight data; The weight identity information, environmental parameters, weight data and measurement standard instrument information are time-stamped to form measurement raw data; The original measurement data is sent to the edge computing server through the Internet of Things communication module for data smoothing preprocessing, and the preprocessed data is transmitted to the cloud data processing server. The cloud data processing server establishes a mapping relationship between environmental parameters and weights based on a deep learning algorithm, performs multi-dimensional environmental compensation on the weight data according to the coupling relationship between temperature parameters, humidity parameters and air pressure parameters, and corrects the thermal expansion coefficient in combination with the material characteristics of the weight to generate measurement calibration data, including: using a deep learning algorithm to construct a multi-branch attention network structure, the multi-branch attention network structure includes a dual-stream architecture of time stream and space stream, and adaptively fuses multi-scale features to obtain mapping results; based on the mapping results, Kalman filtering and wavelet transform are used to compensate for the spatiotemporal dimensions of environmental parameters, construct a nonlinear coupling relationship between temperature parameters, humidity parameters and air pressure parameters, and combine material thermodynamics calculation and molecular dynamics simulation to perform physical dimension compensation; the multi-dimensional environmental compensation also includes intelligent dimension compensation: constructing a heterogeneous graph neural network to model the interactive relationship between environmental parameters, taking temperature parameters, humidity parameters and air pressure parameters as graph network nodes, capturing the dynamic correlation between parameters through the attention edge convolution layer, and introducing Jump connections preserve historical interaction information; a deep reinforcement learning model with a dual reward mechanism is designed to dynamically adjust the compensation strategy. The dual reward mechanism includes environmental stability rewards and measurement accuracy rewards, and a policy gradient algorithm based on the advantage function is used to optimize the compensation action. A hybrid architecture of transfer learning and multi-task learning is constructed. The common features of environmental parameters are extracted through a shared representation layer, and adaptive feature alignment is used to eliminate distribution differences under different working conditions. A multi-head attention mechanism is combined to perform fine-grained feature expression to optimize the compensation strategy. Multi-scale thermal expansion correction is performed on the compensated data. A stress-strain tensor is established at the lattice scale to describe the thermal deformation characteristics and calculate the contribution of atomic vibration to volume change. At the microscale, thermal expansion correlation characteristics are determined through dynamic simulation of grain boundary slip effects and dislocation motion analysis. At the macroscale, a multi-scale thermodynamic equation system is established to couple elastic deformation and thermal expansion calculations. The corrected data is input into the calibration data generation module, and the data is subjected to outlier detection and filtering denoising. The deep learning prediction results, multi-dimensional environmental compensation results, and thermal expansion correction results are adaptively weighted fused to generate metrological calibration data. The wear rate of the weight is also predicted based on historical metrological data. Build a secure isolation architecture, encrypt the metrology calibration data and predicted wear rate based on the secure isolation architecture, build an adaptive traceability chain structure including a basic chain and an extended chain, write the metrology calibration data and wear rate prediction results into the blockchain network through a hierarchical consensus mechanism to form a traceability information chain for the entire metrology process; establish an access control strategy based on behavioral characteristics, trigger hierarchical response measures according to anomaly scores, and provide traceability analysis results to user terminals.

2. The method according to claim 1, characterized in that A multi-branch attention network structure is constructed using a deep learning algorithm. The multi-branch attention network structure includes a dual-stream architecture of time stream and spatial stream. The steps of adaptively fusing multi-scale features to obtain a mapping result include: DenseNet is used as the backbone network to construct a multi-branch attention network structure, which includes a temporal stream branch and a spatial stream branch; The multi-branch attention network structure sets up a branch interaction module, which exchanges information between branches through a cross-attention mechanism, constructs a feature map similarity matrix to calculate attention weights, and controls information flow through a gating mechanism; The time stream branch extracts time series features through causal convolution, sets a time step attention mechanism to calculate the weights of the time series features, and models the time series features through the LSTM layer; the time stream branch sets a time series feature processing module, the time series feature processing module constructs a time series multi-scale feature pyramid, and processes the features through the time series attention pooling layer; The spatial stream branch extracts spatial features through a spatial attention convolution layer and sets a deformable convolution to deform the convolution kernel; the spatial stream branch sets a spatial relationship modeling module, which calculates the correlation relationship of spatial features through a graph attention network and introduces position encoding to mark the spatial position of features; Adaptively fusing the features of the temporal stream branch and the spatial stream branch, including: unifying feature dimensions through a feature transformation network and aligning feature resolutions using a learnable upsampling module; Build a dynamic weight prediction network to calculate feature fusion weights, and obtain the final mapping result through the attention-guided feature aggregation module; A multi-task loss function with adaptive weights is used to optimize network training. The multi-task loss function includes the Focal L1 loss of the main task and the feature consistency loss and temporal smoothness loss of the auxiliary tasks. The loss weights are calculated through an adaptive weight learning method based on task uncertainty. A staged training strategy is adopted to train the temporal stream branch, the spatial stream branch, the branch interaction module and the feature fusion module in sequence, and finally end-to-end optimization is performed.

3. The method according to claim 1, characterized in that The steps for predicting the wear rate of a weight based on historical metrology data include: The historical metrological data were preprocessed using sliding median filtering to extract the quality change rate, environmental parameters, usage intensity and surface roughness features and then normalized to obtain a standardized feature sequence. Inputting the standardized feature sequence into a bidirectional LSTM network, extracting bidirectional features through forward propagation and backward propagation, setting a time step calculation module in the bidirectional LSTM network to quantify the degree of temporal correlation, and performing a weighted combination of the bidirectional features and the degree of temporal correlation to obtain a deep feature representation of weight wear; A multi-task learning framework is constructed based on the deep feature representation, and the multi-task learning framework includes a wear pattern classification branch and a rate prediction branch; the wear pattern classification branch divides wear into normal wear, accelerated wear and abnormal wear through a fully connected network, and adopts focal loss for optimization; the rate prediction branch predicts the wear rate in the future time period through a residual fully connected structure; a prediction reliability evaluation module is constructed in the multi-task learning framework, the mean and variance of the prediction results are output through a neural network, and the negative log-likelihood loss function is adopted for optimization; an adaptive correction mechanism is constructed based on the prediction reliability evaluation result, when the prediction uncertainty exceeds the set threshold, the data sampling density is increased and the prediction time span is shortened, the prediction deviation is calculated based on the latest measurement data, and the prediction result is corrected online by updating the correction coefficient through the exponential sliding average.

4. The method according to claim 1, wherein The steps of constructing a secure isolation architecture, encrypting the metrology calibration data and predicted wear rate based on the secure isolation architecture, building an adaptive traceability chain structure including a basic chain and an extended chain, and writing the metrology calibration data and wear rate prediction results into the blockchain network through a layered consensus mechanism to form a traceability information chain for the entire metrology process include: Construct a secure isolation architecture consisting of a hardware root of trust layer, a trusted execution environment layer, and a secure processing layer. The hardware root of trust layer uses a trusted platform module for key storage and integrity measurement. The trusted execution environment layer divides the secure world and the normal world and performs key management. The secure processing layer is equipped with an independent data processing environment. Inputting metrology calibration data and predicted wear rate into the secure isolation architecture for encryption, and constructing an adaptive traceability chain structure based on the encrypted data. The adaptive traceability chain structure includes a base chain for storing metrology parameters and metrology calibration data, and an extended chain for storing wear rate prediction results and metrology scenario data. The cross-chain data of the base chain and the extended chain are maintained consistent through an inter-chain data reference mechanism; A hierarchical consensus mechanism is designed for the adaptive traceability chain structure. A practical Byzantine fault-tolerant algorithm with dynamic view switching is introduced to achieve consensus processing on the measurement traceability benchmark nodes. Proof-of-stake consensus is used to confirm data on the measurement data collection nodes. A consensus switching mechanism is used to achieve compatibility with different consensus algorithms. The metrology calibration data is standardized and quality assessed, and a spatiotemporal feature extraction algorithm is used to obtain metrology process characteristic parameters. A metrology process correlation matrix is established by calculating the temporal correlation and spatial similarity between the metrology process characteristic parameters and the predicted wear rate. Data fusion operations are performed based on a smart contract rule engine to generate traceability information. The traceability information is written into the blockchain network, and a secure storage and trusted traceability information chain for the entire measurement process information is established based on the adaptive traceability chain structure and hierarchical consensus mechanism.

5. The method according to claim 4, characterized in that The steps of implementing consensus processing on the measurement traceability benchmark nodes by introducing a practical Byzantine fault-tolerant algorithm with dynamic view switching, confirming data on the measurement data collection nodes through proof-of-stake consensus, and adopting a consensus switching mechanism to achieve compatibility with different consensus algorithms include: A practical Byzantine fault-tolerant algorithm for dynamic view switching is designed for measurement traceability reference nodes. The algorithm calculates a view activity metric based on the number of messages, consensus time, and view stability. Based on this view activity metric, it triggers view switching and updates the master node. When a view switch message exceeding twice the fault tolerance limit plus one is received, the measurement traceability reference node enters a new view. Calculate the stake value of the metering data collection node based on data quality, response time and storage capacity, and use the proof-of-stake consensus to determine the verification node based on the stake value and complete data confirmation; Standardize the encoding of the state information of the Practical Byzantine Fault Tolerance algorithm and the Proof of Stake consensus, organize the state data using a Merkle tree structure, and construct a unified state description format that includes block height, state root, and consensus proof. Construct cross-consensus messages based on the unified state description format, use a multi-level message confirmation mechanism to transmit the cross-consensus messages, perform retransmission control and timeout detection on the cross-consensus messages, authenticate the source of the cross-consensus messages, and verify the legitimacy of the state mapping and the state transition proof. Build a state version chain and set checkpoints, monitor consensus efficiency indicators, node response status, and state consistency deviation, and trigger consensus switching when the consensus efficiency indicator falls below a preset threshold, or the node response times out, or the state consistency deviation exceeds the limit. Broadcast the switching proposal to the nodes in the network and collect node confirmation information, suspend the current consensus mechanism, and activate the target consensus mechanism; The consensus result of the measurement traceability reference node is used as the trust basis for the measurement data collection node to perform proof-of-stake consensus through the unified state description format. The data confirmed by the measurement data collection node is fed back to the measurement traceability reference node through the cross-consensus message. The state consistency of the heterogeneous consensus is maintained based on the state version chain, and a consensus switching mechanism is used to adjust between the practical Byzantine fault tolerance algorithm and the proof-of-stake consensus.

6. The method according to claim 1, characterized in that The steps for establishing an access control policy based on behavioral characteristics, triggering graded response measures based on anomaly scores, and providing traceability analysis results to user terminals include: Obtaining identity features and behavioral features of user access behavior, wherein the identity features include static user identification, dynamic authentication information, and environmental attributes; and the behavioral features include temporal behavioral features, operational behavioral features, and data interaction features. Using an autoencoder, the identity features and behavioral features are dimensionality reduced to obtain a feature vector. The matching degree between the feature vector and the historical baseline is calculated based on weighted cosine similarity to obtain a real-time trust score. Dividing the access control level based on the real-time trust score, when the real-time trust score is in a first trust range, executing a basic access control policy to allow access with original permissions; when the real-time trust score is in a second trust range, executing an enhanced access control policy to perform secondary authentication and limit access frequency and data scope; when the real-time trust score is in a third trust range, executing a strict access control policy to allow only read-only operations and require multi-factor authentication; Dynamically adjust the control strength of the access control policy based on the changing trend of the real-time trust score during the continuous observation period, and generate an access control result including user identity information, access behavior characteristics and control policy; Calculate an anomaly score based on the access control result and establish a hierarchical response mechanism for abnormal behavior. When abnormal behavior occurs, trigger a warning-level response, a blocking-level response, or a locking-level response based on the anomaly score. The warning-level response includes recording the abnormal event and initiating feature resampling. The blocking-level response includes suspending session permissions and requiring identity reauthentication. The locking-level response includes freezing access permissions and triggering a security audit. Build a correlation map of user operation resources, record the occurrence time, abnormal type, response measures and disposal results of abnormal behavior, establish an abnormal behavior feature library, and push traceability analysis results including metering data query, calibration cycle reminder, wear warning and maintenance suggestions to the user terminal in real time according to the level of the abnormal score.

7. A weight measurement data collection and traceability management system based on the Internet of Things, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to implant an RFID electronic tag storing the weight identity information into the weight body, collect temperature parameters, humidity parameters and air pressure parameters through an environmental parameter sensor, and collect weight data through a weight data acquisition module equipped with a high-precision strain gauge; the weight identity information, environmental parameters, weight data and measurement standard instrument information are timestamped to form the measurement raw data; The second unit is used to send the raw measurement data to the edge computing server through an Internet of Things communication module with a data encryption function for preprocessing of outlier detection and data smoothing, and transmit the preprocessed data to the cloud data processing server. The cloud data processing server establishes a mapping relationship between environmental parameters and weights based on a deep learning algorithm, performs multi-dimensional environmental compensation on the weight data according to the coupling relationship between temperature parameters, humidity parameters and air pressure parameters, corrects the thermal expansion coefficient in combination with the characteristics of the weight material, generates measurement calibration data, and predicts the wear rate of the weight based on historical measurement data; The third unit is used to build a secure isolation architecture, encrypt the metrology calibration data and predicted wear rate based on the secure isolation architecture, build an adaptive traceability chain structure including a basic chain and an extended chain, write the metrology calibration data and wear rate prediction results into the blockchain network through a hierarchical consensus mechanism to form a traceability information chain for the entire metrology process; establish an access control strategy based on behavioral characteristics, trigger hierarchical response measures according to anomaly scores, and provide traceability analysis results to user terminals.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images