Digital twinborn simulation method for automatic verification system of electric energy meter

Abstract

The invention provides a digital twinborn simulation method for an automatic verification system of an electric energy meter, and belongs to the technical field of electric energy meters, and the method comprises the steps: S1, collecting the multi-modal data of the electric energy meter in real time; s2, based on a multi-scale convolutional neural network, deep features of the multi-modal signal data are extracted, and a digital twinborn model is constructed; s3, operating the digital twin model in the virtual environment, simulating the verification process of the electric energy meter, and monitoring the equipment state in real time; and S4, inputting the implicit state representation into the graph convolutional network, constructing a fault relation network of the electric energy meter verification system, analyzing fault causes through the graph convolutional network, obtaining a comprehensive analysis report based on output of fault prediction and diagnosis, and optimizing an electric energy meter verification process by using a multi-arm tiger machine model. The digital twinning technology is fully utilized, the actual operation state of the electric energy meter is simulated in the virtual environment, real-time synchronization of the physical entity and the virtual model is achieved, and the intelligent level and the management efficiency of the system are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric energy meters, and in particular relates to a digital twin simulation method for an automatic calibration system of an electric energy meter. Background Art

[0002] With the development of smart grid and Internet of Things technology, the calibration and maintenance of electric energy meters as important equipment for power measurement is particularly important. The accuracy and reliability of the electric energy meter calibration system are directly related to the fairness of power measurement and the protection of user rights. The development of smart grid requires that electric energy meters can collect, process and transmit power data in real time, and perform accurate measurement and monitoring.

[0003] The traditional electric energy meter calibration system mainly relies on manual operation and simple automation equipment, which is difficult to meet the needs of modern power systems for efficient, accurate and intelligent calibration. The traditional method has the following technical problems: the traditional calibration lacks the ability to efficiently process real-time data and extract deep features, resulting in the inability to ensure data quality and consistency during the calibration process, and the inability to effectively capture the dynamic characteristics and complex fault modes of time series data, resulting in low accuracy in fault prediction and diagnosis, and the inability to take preventive measures in a timely manner, reducing the reliability of the system, lacking intelligent means, and being unable to dynamically adjust the calibration strategy according to real-time data, resulting in a lack of flexibility and adaptability in the calibration process. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, a digital twin simulation method of an electric energy meter automatic calibration system is provided.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] This technical solution proposes a digital twin simulation method for an electric energy meter automatic verification system, comprising the following steps:

[0007] S1: Based on sensors and IoT devices, multimodal data of electric energy meters are collected in real time and preprocessed;

[0008] S2: Based on a multi-scale convolutional neural network, the deep features of multimodal signal data are extracted. Based on the cyclic cross-fusion mechanism, the data features of different modes are cyclically cross-recombined, the recombined features are fused, and the mixed Gaussian process is used to achieve weighted fusion of multimodal data to build a digital twin model.

[0009] S3: Run the digital twin model in a virtual environment, simulate the calibration process of the electric energy meter, monitor the equipment status in real time, use the fused features as the input of the fault prediction model, and perform fault prediction and diagnosis based on the long short-term memory network and conditional random field;

[0010] S4: Input the implicit state representation into the graph convolutional network, construct the fault relationship network of the electric energy meter calibration system, analyze the fault cause through the graph convolutional network, obtain a comprehensive analysis report based on the output of fault prediction and diagnosis, and use the multi-armed bandit model to optimize the electric energy meter calibration process.

[0011] Preferably, in S1, the multimodal data includes detection data, equipment status data and environmental parameter data, and the preprocessing includes data cleaning, denoising and normalization, by standardizing the data to a range between 0 and 1.

[0012] Preferably, in S2, the extracted deep features are expressed as:

[0013]

[0014] In the formula, F(t) represents the features extracted from the multimodal data at time t, and D clean (t) is the data collected at time t after preprocessing, W k and b k They represent the weight and bias of the k-th convolution kernel respectively, and σ represents the activation function.

[0015] Preferably, based on the cyclic cross-fusion mechanism, the data features of different modalities are cyclically cross-recombined, and the formula is as follows:

[0016]

[0017] Where: F rec (t) represents the characteristics of the reorganized data collected at time t, T is a collection cycle, α 1t , α 2t and α 3t They represent the weight parameters of cyclic cross fusion, F(t-1) represents the features of the data collected at the previous moment, ⊙ represents the Hadamard product, and ∈ is a constant;

[0018] The reorganized features are fused and the mixed Gaussian process is used to achieve weighted fusion of multimodal data. The formula is as follows:

[0019]

[0020] In the formula, F fused represents the fused features, w i represents the weight of the i-th Gaussian component, represents Gaussian distribution, μ i ,∑ i denote the mean and covariance matrix of the i-th Gaussian component respectively.

[0021] Preferably, the fused features are input into a calibration model for calibration, and the digital twin model is calibrated by piecewise linear regression, and the calibration process includes multiple rounds of simulation and actual data comparison;

[0022] Use the fusion feature F fused As input, the operating status of the electric energy meter is predicted by the piecewise linear regression model, and the predicted output is obtained:

[0023]

[0024] In the formula, y pred represents the predicted output, a j and b j Represent the coefficients of the j-th linear regression, F fused,j Represents the jth fused feature.

[0025] Preferably, the calibration loss function is obtained by calculating the error between the predicted output and the actual output:

[0026] L cal =Σ(y pred -y real ) 2 (5)

[0027] Where, L cal represents the calibration loss function, y real Indicates the actual output;

[0028] Using the gradient descent algorithm to minimize the calibration loss function, the formula for the updated digital twin model parameters is as follows:

[0029]

[0030] In the formula, θ and θ' represent the parameters of the digital twin model before and after the update, η represents the learning rate, Represents the gradient of the calibration loss function with respect to the digital twin model parameters.

[0031] Preferably, in S3: running the calibrated digital twin model in a virtual environment, simulating the calibration process of the electric energy meter, monitoring the device status in real time, and using the fused feature F fused As the input of the fault prediction model, the input features are processed by the long short-term memory network to obtain the implicit state representation h at each moment t , the formula is as follows:

[0032] h t =LSTM(F fused (t),h t-1 ) (7)

[0033] In the formula, LSTM represents the calculation process of the long short-term memory network.

[0034] Preferably, the conditional random field model is combined to calculate the probability of a fault occurrence according to the implicit state, and the formula is as follows:

[0035]

[0036] Where P fault (t) represents the probability of failure at time t, Z represents the normalization factor, θ p is the weight of the univariate feature function, θ ij is the weight of the bivariate characteristic function;

[0037] f p (t,h t ) is a univariate feature, indicating that at time t, based on the implicit state h t The features of the pth dimension;

[0038] f pq It is a bivariate feature, indicating that at time t, based on the implicit state h t The interaction between the p-th dimension and the q-th dimension in .

[0039] Preferably, in S4, the implicit state is represented by h t Input into the graph convolutional network, the inter-layer propagation formula of the graph convolutional network is as follows:

[0040]

[0041] In the formula, H (l) represents the feature representation of the lth convolutional layer, H (0) =h t represents the initial input features, A u represents the adjacency matrix of the u-th graph, represents the weight matrix of the u-th graph in the l-th convolutional layer, U represents the number of graphs, and σ represents the activation function;

[0042] Through the calculation of the multi-layer graph convolutional network, the feature representation H of each node is obtained (L) , where L represents the number of layers of the graph convolutional network;

[0043] According to the output of the last layer, the status of each node is classified and the cause of the failure is inferred. The probability calculation formula of the cause of the failure is as follows:

[0044]

[0045] Where P cause (v) represents the failure cause probability of node v, is the feature representation of node v at the Lth layer.

[0046] Preferably, the outputs of fault prediction and diagnosis are combined to form a comprehensive status assessment report, which includes the probability of fault occurrence and possible fault causes at the current moment, and the comprehensive status assessment report serves as an input for calibration process optimization;

[0047] Based on the reinforcement learning algorithm, the multi-armed bandit model is used to optimize the verification process and parameter configuration. The optimization strategy is as follows:

[0048]

[0049] In the formula, S new is the optimized verification strategy, S is the current verification strategy, is the expected value operator, which represents the expectation of reward;

[0050] γ is the discount factor, ranging from 0≤γ≤1, which indicates the discount rate of future rewards;

[0051] rp(t) is the comprehensive status evaluation report at time t, and R(t|rp(t)) is the reward value based on the comprehensive status evaluation report at time t.

[0052] Compared with the prior art, the present invention has the following advantages:

[0053] 1. This application collects multimodal data of the electric energy meter and its calibration system in real time through sensors and Internet of Things devices, which can comprehensively and accurately obtain the operating information of the electric energy meter. It uses a multi-scale convolutional neural network to extract local and global information in the data, and can quickly and efficiently process large-scale data and extract deep-level features, providing a high-quality data foundation for subsequent analysis and processing.

[0054] 2. This application introduces a cyclic cross-fusion mechanism to cyclically cross-reorganize and fuse data features of different modalities, enhance the correlation between different modalities, make feature fusion more complex and sophisticated, and thus improve the efficiency and accuracy of verification; adopt a mixed Gaussian process to realize the weighted fusion of multimodal data, and comprehensively process the data through Gaussian distribution to obtain a more accurate and stable feature representation, thereby improving the accuracy of the model.

[0055] 3. This application calibrates the digital twin model through piecewise linear regression, uses fused features as input, and calibrates the model parameters through multiple rounds of simulation and actual data comparison to make it better match the actual situation and ensure the accuracy of the model; by running the calibrated digital twin model in a virtual environment, simulating the calibration process of the electric energy meter, monitoring the equipment status in real time, and promptly discovering and predicting possible faults, the accuracy of the calibration can be further improved.

[0056] 4. This application uses conditional random fields combined with long short-term memory networks to perform fault prediction and diagnosis. Combined with the CRF model, the probability of fault occurrence is calculated based on the implicit state. By predicting the probability of future faults, measures can be taken in advance to avoid the occurrence of faults and improve the reliability and safety of the system. Through the calculation of multi-layer graph convolutional networks, the root cause of the fault can be effectively analyzed, and the calibration process of the electric energy meter can be effectively diagnosed and processed. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0058] Figure 1 It is the overall flow chart of the present invention. DETAILED DESCRIPTION

[0059] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0060] like Figure 1 As shown, this embodiment proposes a digital twin simulation method for an electric energy meter automatic verification system, comprising the following steps:

[0061] S1: Based on sensors and IoT devices, multimodal data of electric energy meters are collected in real time, and multimodal data are preprocessed to ensure data quality and consistency;

[0062] S2: Based on a multi-scale convolutional neural network, the deep features of multimodal signal data are extracted. Based on the cyclic cross-fusion mechanism, the data features of different modes are cyclically cross-recombined, the recombined features are fused, and the mixed Gaussian process is used to achieve weighted fusion of multimodal data to build a digital twin model.

[0063] S3: Run the digital twin model in a virtual environment, simulate the calibration process of the electric energy meter, monitor the equipment status in real time, use the fused features as the input of the fault prediction model, and perform fault prediction and diagnosis based on the long short-term memory network and conditional random field;

[0064] S4: Input the implicit state representation into the graph convolutional network, construct the fault relationship network of the electric energy meter calibration system, analyze the fault cause through the graph convolutional network, obtain a comprehensive analysis report based on the output of fault prediction and diagnosis, and use the multi-armed bandit model to optimize the electric energy meter calibration process.

[0065] Make full use of digital twin technology to simulate the actual operating status of the electric energy meter in a virtual environment, realize real-time synchronization between the physical entity and the virtual model, effectively improve the intelligence level and management efficiency of the system, and realize the full life cycle management of the electric energy meter calibration process through virtual simulation technology.

[0066] In S1, multimodal data includes detection data, equipment status data and environmental parameter data. The detection data includes voltage, current, power, frequency, etc. The preprocessing includes data cleaning, denoising and normalization. The data is standardized so that its range is between 0 and 1.

[0067] In S2, the multi-scale convolutional neural network model can capture local and global information in the data, and the extracted deep features are represented as follows:

[0068]

[0069] In the formula, F(t) represents the features extracted from the multimodal data at time t, and D clean (t) is the data collected at time t after preprocessing, W k and b k They represent the weight and bias of the k-th convolution kernel respectively, and σ represents the activation function; by extracting the deep features of the data, it is convenient for subsequent analysis and processing.

[0070] Based on the cyclic cross fusion mechanism (RCMF), the data features of different modalities are cyclically cross-recombined so that the information of different modalities can fully interact and fuse. The formula is as follows:

[0071]

[0072] Where: F rec (t) represents the characteristics of the reorganized data collected at time t, T is a collection cycle, α 1t , α 2t and α 3t They represent the weight parameters of cyclic cross fusion, F(t-1) represents the features of the data collected at the previous moment, ⊙ represents the Hadamard product, and ∈ is a constant to prevent division by zero errors;

[0073] By enhancing the correlation between different modalities, feature fusion is made more complex and sophisticated, improving the expressive power of the digital twin model.

[0074] The reorganized features are fused and the mixed Gaussian process (MGP) is used to achieve weighted fusion of multimodal data to further improve the accuracy and reliability of the features. The formula is as follows:

[0075]

[0076] In the formula, F fused represents the fused features, w i represents the weight of the i-th Gaussian component, represents Gaussian distribution, μ i ,Σ i Represent the mean and covariance matrix of the i-th Gaussian component respectively. The features of different modes are processed comprehensively to obtain a more accurate and stable feature representation.

[0077] The fused features are input into the calibration model for calibration. The digital twin model is calibrated through piecewise linear regression (PLR). The calibration process includes multiple rounds of simulation and actual data comparison.

[0078] Use the fusion feature F fused As input, the operating status of the energy meter, such as voltage, current, power, frequency and other key parameters, is predicted through the piecewise linear regression model to obtain the predicted output:

[0079]

[0080] In the formula, y pred represents the predicted output, a j and b j Represent the coefficients of the j-th linear regression, F fused,j Represents the jth fused feature;

[0081] By calculating the error between the predicted output and the actual output, the calibration loss function is obtained:

[0082] L cal =Σ(y pred -y real ) 2 (5)

[0083] Where, L cal represents the calibration loss function, y real Indicates the actual output;

[0084] In order to adjust the digital twin model parameters, such as the model weights, bias terms, regression coefficients, etc., the gradient descent algorithm is used to minimize the calibration loss function. The formula for the updated digital twin model parameters is as follows:

[0085]

[0086] In the formula, θ and θ' represent the parameters of the digital twin model before and after the update, η represents the learning rate, Represents the gradient of the calibration loss function relative to the parameters of the digital twin model. The model parameters are adjusted through calibration to better match the actual situation and ensure the accuracy of the model.

[0087] The specific calibration process can be described as follows: First, the fused feature F fused Input into the piecewise linear regression model and calculate the predicted output y pred Then, by calculating the predicted output and the actual output y real The mean square error of the calibration loss function L is obtained cal ,Then, the digital twin model parameters θ are adjusted through the gradient descent algorithm to minimize the calibration loss function. The above process is repeated until the calibration loss function converges to the minimum value.

[0088] In S3: Run the calibrated digital twin model in the virtual environment, simulate the calibration process of the electric energy meter, monitor the device status in real time, and use the fused feature F fused As the input of the fault prediction model, in order to capture the dynamic characteristics of time series data, the input features are processed through the long short-term memory network to obtain the implicit state representation h at each moment t , the formula is as follows:

[0089] h t =LSTM(F fused (t),h t-1 ) (7)

[0090] In the formula, LSTM represents the calculation process of the long short-term memory network;

[0091] Combined with the conditional random field model, the probability of failure is calculated according to the implicit state. The formula is as follows:

[0092]

[0093] Where P fault (t) represents the probability of failure at time t, Z represents the normalization factor, θ p is the weight of the univariate feature function, θ ij is the weight of the bivariate characteristic function;

[0094] f p (t,h t ) is a univariate feature, indicating that at time t, based on the implicit state h t The features of the pth dimension;

[0095] f pq It is a bivariate feature, indicating that at time t, based on the implicit state h t The interaction between the pth dimension and the qth dimension in the model. By predicting the probability of future failures, measures can be taken in advance to avoid failures.

[0096] In S4, the hidden state is represented by h tInput into the graph convolutional network to build the fault relationship network of the electric energy meter verification system and infer the possible cause of the fault. The inter-layer propagation formula of the graph convolutional network is as follows:

[0097]

[0098] In the formula, H (l) represents the feature representation of the lth convolutional layer, H (0) =h t represents the initial input features, A u represents the adjacency matrix of the u-th graph, represents the weight matrix of the u-th graph in the l-th convolutional layer, U represents the number of graphs, and σ represents the activation function. By extracting the features of the graph structure, the root cause of the fault is analyzed, so that effective diagnosis and processing can be carried out.

[0099] Through the calculation of the multi-layer graph convolutional network, the feature representation H of each node is obtained (L) , where L represents the number of layers of the graph convolutional network;

[0100] According to the output of the last layer, the status of each node is classified and the cause of the failure is inferred. The probability calculation formula of the cause of the failure is as follows:

[0101]

[0102] Where P cause (v) represents the failure cause probability of node v, is the feature representation of node v at layer L;

[0103] The outputs of fault prediction and diagnosis are combined to form a comprehensive status assessment report, which includes the probability of fault occurrence and possible causes of faults at the current moment. The comprehensive status assessment report serves as the input for calibration process optimization. The calibration process optimization algorithm uses the comprehensive status assessment report to adjust the calibration strategy to maximize detection efficiency and accuracy and minimize the risk of faults.

[0104] Based on the reinforcement learning algorithm, the multi-armed bandit model is used to optimize the verification process and parameter configuration. The optimization strategy is as follows:

[0105]

[0106] In the formula, S new is the optimized verification strategy, S is the current verification strategy, is the expected value operator, which represents the expectation of reward;

[0107] γ is the discount factor, ranging from 0≤γ≤1, which indicates the discount rate of future rewards;

[0108] rp(t) is the comprehensive status evaluation report at time t, and R(t|rp(t)) is the reward value based on the comprehensive status evaluation report at time t.

[0109] Based on the optimization results, the verification strategy is adjusted to optimize the verification process, and the new strategy will be applied to the next round of detection and monitoring.

[0110] By collecting multimodal data of the electric energy meter and its calibration system in real time through sensors and IoT devices, the operation information of the electric energy meter can be obtained comprehensively and accurately. The multi-scale convolutional neural network is used to extract local and global information in the data. It can process large-scale data quickly and efficiently, extract deep-level features, and provide a high-quality data foundation for subsequent analysis and processing.

[0111] By introducing a cyclic cross-fusion mechanism, data features of different modalities are cyclically cross-reorganized and fused, the correlation between different modalities is enhanced, and the feature fusion is made more complex and sophisticated, thereby improving the verification efficiency and accuracy; a mixed Gaussian process is used to realize the weighted fusion of multimodal data, and the data is comprehensively processed through Gaussian distribution to obtain a more accurate and stable feature representation, thereby improving the accuracy of the model.

[0112] The digital twin model is calibrated through piecewise linear regression, using fused features as input. Through multiple rounds of simulation and actual data comparison, the model parameters are calibrated to better match the actual situation and ensure the accuracy of the model. By running the calibrated digital twin model in a virtual environment, simulating the calibration process of the electricity meter, monitoring the equipment status in real time, and promptly discovering and predicting possible faults, the accuracy of the calibration can be further improved.

[0113] Fault prediction and diagnosis are performed through conditional random fields combined with long short-term memory networks. Combined with the CRF model, the probability of fault occurrence is calculated according to the implicit state. By predicting the probability of future faults, measures can be taken in advance to avoid faults and improve the reliability and safety of the system. Through the calculation of multi-layer graph convolutional networks, the root cause of the fault can be effectively analyzed, and the calibration process of the electric energy meter can be effectively diagnosed and processed.

[0114] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A digital twin simulation method for an electric energy meter automatic verification system, characterized in that: The following steps are involved: S1: Based on sensors and IoT devices, multimodal data of electric energy meters are collected in real time and preprocessed; S2: Based on a multi-scale convolutional neural network, the deep features of multimodal signal data are extracted. Based on the cyclic cross-fusion mechanism, the data features of different modes are cyclically cross-recombined, the recombined features are fused, and the mixed Gaussian process is used to achieve weighted fusion of multimodal data to build a digital twin model. S3: Run the digital twin model in a virtual environment, simulate the calibration process of the electric energy meter, monitor the equipment status in real time, use the fused features as the input of the fault prediction model, and perform fault prediction and diagnosis based on the long short-term memory network and conditional random field; S4: Input the implicit state representation into the graph convolutional network, construct the fault relationship network of the electric energy meter calibration system, analyze the fault cause through the graph convolutional network, obtain a comprehensive analysis report based on the output of fault prediction and diagnosis, and use the multi-armed bandit model to optimize the electric energy meter calibration process.

2. The digital twin simulation method of an electric energy meter automatic verification system according to claim 1 is characterized in that: In S1, the multimodal data includes detection data, equipment status data and environmental parameter data, and the preprocessing includes data cleaning, denoising and normalization, by standardizing the data to make it range between 0 and 1.

3. The digital twin simulation method of an electric energy meter automatic verification system according to claim 1 is characterized in that: In S2, the extracted deep features are expressed as: In the formula, F(t) represents the features extracted from the multimodal data at time t, and D clean (t) is the data collected at time t after preprocessing, W k and b k They represent the weight and bias of the k-th convolution kernel respectively, and σ represents the activation function.

4. The digital twin simulation method of an electric energy meter automatic verification system according to claim 1 is characterized in that: Based on the cyclic cross-fusion mechanism, the data features of different modalities are cyclically cross-recombined. The formula is as follows: In the formula, F rec (t) represents the characteristics of the reorganized data collected at time t, T is a collection cycle, α 1t , α 2t and α 3t They represent the weight parameters of cyclic cross fusion, F(t-1) represents the features of the data collected at the previous moment, ⊙ represents the Hadamard product, and ∈ is a constant; The reorganized features are fused and the mixed Gaussian process is used to achieve weighted fusion of multimodal data. The formula is as follows: In the formula, F fused represents the fused features, w i represents the weight of the i-th Gaussian component, represents Gaussian distribution, μ i ,∑ i denote the mean and covariance matrix of the i-th Gaussian component respectively.

5. The digital twin simulation method of the electric energy meter automatic verification system according to claim 4 is characterized in that: The fused features are input into the calibration model for calibration. The digital twin model is calibrated through piecewise linear regression. The calibration process includes multiple rounds of simulation and actual data comparison. Use the fusion feature F fused As input, the operating status of the electric energy meter is predicted by the piecewise linear regression model, and the predicted output is obtained: In the formula, y pred represents the predicted output, a j and b j Represent the coefficients of the j-th linear regression, F fused,j Represents the jth fused feature.

6. The digital twin simulation method of the electric energy meter automatic verification system according to claim 5 is characterized in that: By calculating the error between the predicted output and the actual output, the calibration loss function is obtained: L cal =∑(y pred -y real ) 2 (5) In the formula, L cal represents the calibration loss function, y real Indicates the actual output; Using the gradient descent algorithm to minimize the calibration loss function, the formula for the updated digital twin model parameters is as follows: In the formula, θ and θ' represent the parameters of the digital twin model before and after the update, η represents the learning rate, Represents the gradient of the calibration loss function with respect to the digital twin model parameters.

7. The digital twin simulation method of an electric energy meter automatic verification system according to claim 1 is characterized in that: In S3: the calibrated digital twin model is run in a virtual environment to simulate the calibration process of the electric energy meter, monitor the device status in real time, and use the fused feature F fused As the input of the fault prediction model, the input features are processed by the long short-term memory network to obtain the implicit state representation h at each moment t , the formula is as follows: h t =LSTM(F fused (t),h t-1 ) (7) In the formula, LSTM represents the calculation process of the long short-term memory network.

8. The digital twin simulation method of the electric energy meter automatic verification system according to claim 7 is characterized in that: Combined with the conditional random field model, the probability of failure is calculated according to the implicit state. The formula is as follows: Where P fault (t) represents the probability of failure at time t, Z represents the normalization factor, θ p is the weight of the univariate feature function, θ ij is the weight of the bivariate characteristic function; f p (t,h t ) is a univariate feature, indicating that at time t, based on the implicit state h t The features of the pth dimension; f pq It is a bivariate feature, indicating that at time t, based on the implicit state h t The interaction between the p-th dimension and the q-th dimension in .

9. The digital twin simulation method of an electric energy meter automatic verification system according to claim 1, characterized in that: In S4, the implicit state is represented by h t Input into the graph convolutional network, the inter-layer propagation formula of the graph convolutional network is as follows: In the formula, H (l) represents the feature representation of the lth convolutional layer, H (0) =h t represents the initial input features, A u represents the adjacency matrix of the u-th graph, represents the weight matrix of the u-th graph in the l-th convolutional layer, U represents the number of graphs, and σ represents the activation function; Through the calculation of the multi-layer graph convolutional network, the feature representation H of each node is obtained (L) , where L represents the number of layers of the graph convolutional network; According to the output of the last layer, the status of each node is classified and the cause of the failure is inferred. The probability calculation formula of the cause of the failure is as follows: Where P cause (v) represents the failure cause probability of node v, is the feature representation of node v at the Lth layer.

10. The digital twin simulation method of an electric energy meter automatic verification system according to claim 9, characterized in that: Combine the output of fault prediction and diagnosis to form a comprehensive status assessment report, which includes the current probability of fault occurrence and possible causes of faults. The comprehensive status assessment report serves as the input for calibration process optimization. Based on the reinforcement learning algorithm, the multi-armed bandit model is used to optimize the verification process and parameter configuration. The optimization strategy is as follows: In the formula, S new is the optimized verification strategy, S is the current verification strategy, is the expected value operator, which represents the expectation of reward; γ is the discount factor, ranging from 0≤γ≤1, which indicates the discount rate of future rewards; rp(t) is the comprehensive status evaluation report at time t, and R(t|rp(t)) is the reward value based on the comprehensive status evaluation report at time t.

Images