Intelligent Electric Energy Meter Fault Prediction Method Based on Multimodal Sensor Fusion

Abstract

The present invention provides an intelligent electricity meter fault prediction method based on multi-modal sensor fusion, including: collecting raw data of multi-modal sensors and preprocessing to obtain a standardized data stream; establishing a sensor health state model and monitoring the working state of sensors; extracting features and performing pattern recognition on the standardized data to generate a state feature vector of the electricity meter; combining the sensor health assessment results and the state feature vector of the electricity meter, and applying a two-layer causal inference framework to realize the separation and identification of sensor faults and electricity meter faults; quantifying multi-source uncertainties and evaluating the diagnostic reliability to generate a fault assessment report. The present invention can effectively distinguish between sensor faults and electricity meter faults, improve the accuracy of fault prediction, reduce the false alarm rate, and optimize maintenance decisions.

Description

Technical Field

[0001] The present invention relates to a fault prediction method, in particular to an intelligent electric energy meter fault prediction method based on multi-modal sensor fusion. Background Art

[0002] With the rapid development of the smart grid, as a key terminal device, the intelligent electric energy meter undertakes various functions such as electric energy metering, data acquisition, and power consumption information interaction. Faults in the electric energy meter not only affect the accuracy of electric energy metering and the fairness of electricity bill settlement, but may also lead to local abnormal operation of the distribution network, reduce power supply reliability, and even cause safety accidents. Real-time monitoring and fault prediction of the operating state of the intelligent electric energy meter are of great significance for improving the safety of power grid operation, reducing economic losses, and optimizing the allocation of operation and maintenance resources. Especially in the context of large-scale deployment of the smart grid, the active fault prediction technology based on sensor fusion can significantly reduce the failure rate, improve the power supply reliability index, and at the same time reduce the maintenance cost of power grid enterprises and the power outage loss of users.

[0003] Currently, there are mainly three technical routes for intelligent electric energy meter fault prediction: the threshold monitoring method based on a single sensor, the statistical feature analysis method based on time-series data, and the fault diagnosis method based on a model. The threshold monitoring method judges the operating state of the electric energy meter by setting safety limits for parameters such as voltage, current, and temperature, and is simple and has a small amount of calculation. The statistical feature analysis method uses the statistical characteristics of the operating data of the electric energy meter, such as mean, variance, kurtosis, etc., and combines methods such as fuzzy logic or decision tree to identify abnormal patterns, improving the detection accuracy. The fault diagnosis method based on a model establishes a mathematical model for the normal operation of the electric energy meter and predicts potential faults through residual analysis, with relatively high accuracy. In recent years, some researchers have tried to use machine learning methods to process single or simple combined sensor data for fault prediction, such as the error prediction of electric energy meters based on BP neural networks, the component aging monitoring based on support vector machines, etc., which have also improved the prediction performance.

[0004] The key technical problem faced by existing methods is that it is difficult to effectively distinguish between the faults of the sensors themselves and the actual faults of the electricity meters. In practical applications, the faults of the sensors themselves, such as the drift of voltage sensors, the slow response of current sensors, or the decrease in the sensitivity of temperature sensors, often highly resemble the actual faults, such as the aging of the circuit board of the electricity meter and the abnormality of the metering chip, in terms of data performance. This similarity causes the system to be unable to accurately determine the true source of the abnormal data, resulting in a false alarm rate of about 30%-45%, a large number of maintenance personnel being dispatched in vain, and a serious waste of resources. For example, the zero drift of the current sensor may be misjudged as a fault in the sampling circuit of the electricity meter, and the decrease in the sensitivity of the temperature sensor may be misjudged as an abnormal heating of the electricity meter. Most existing technologies use single redundancy or simple threshold judgment to determine sensor faults, lacking a joint evaluation mechanism for the health status of sensors and the fault status of electricity meters, and unable to achieve accurate tracing of the fault source. In addition, when multiple sensors work together, the cross-impact of fault effects becomes more complex, further increasing the difficulty of identifying the fault source. Summary of the Invention

[0005] The object of the invention is to provide an intelligent electricity meter fault prediction method based on multi-modal sensor fusion, in order to solve one of the above technical problems existing in the prior art.

[0006] Technical solution: An intelligent electricity meter fault prediction method based on multi-modal sensor fusion includes the following steps:

[0007] Collect the original data of multi-modal sensors and preprocess it to obtain standard data;

[0008] Based on the standard data, establish a sensor health status model and monitor the working status of the sensors to obtain the sensor health assessment result;

[0009] Extract features and perform pattern recognition on the standard data to generate an electricity meter status feature vector;

[0010] Combine the sensor health assessment result and the electricity meter status feature vector, and execute double-layer fault source separation and identification through a double-layer causal graph reasoning module to output the fault type and confidence level;

[0011] Based on the fault type and confidence level, quantify multi-source uncertainties and evaluate the reliability to generate a fault assessment report.

[0012] Preferably, the double-layer causal graph reasoning module includes:

[0013] The upper layer module is used to represent the causal relationship between sensor faults and measurement abnormalities,

[0014] The lower layer module is used to represent the causal relationship between electricity meter faults and physical parameter abnormalities,

[0015] Inter-layer connection, used to represent the observational relationship between measurement anomalies and physical parameter anomalies,

[0016] During operation, infer measurement anomalies from possible sensor failures from top to bottom, infer potential failures from observed anomalies from bottom to top, propagate probability information in the double-layer causal graph through a message passing mechanism, and output the probability distribution of the fault source;

[0017] Preferably, the process of double-layer fault source separation and identification includes:

[0018] Read the observed evidence set as the input of the double-layer causal graph inference module to obtain the probability distribution of the fault source;

[0019] Among them, the observed evidence set includes the sensor health assessment result, the electricity meter status feature vector, and the current measurement data;

[0020] Regard the fault sources with probabilities higher than the threshold in the probability distribution of the fault source as candidate fault sources;

[0021] Construct three competing hypotheses of sensor failure, electricity meter failure, and hybrid failure, apply the Bayesian model selection method to evaluate the posterior probabilities of each hypothesis, and output the optimal fault hypothesis and the set of candidate fault sources.

[0022] Preferably, the process of double-layer fault source separation and identification further includes:

[0023] For each candidate fault in the set of candidate fault sources, decompose the original signal into a sensor fault component and an electricity meter fault component, and reconstruct the pure electricity meter status signal;

[0024] Combined with the physical working principle of the electricity meter, verify whether the pure electricity meter status signal conforms to the physical consistency law, and output the fault verification result;

[0025] Combined with the sensor health assessment result, the electricity meter status feature vector, and the fault verification result, perform hierarchical identification of sensor faults and electricity meter faults through a dynamic decision tree to obtain the final fault type determination and fault confidence.

[0026] Preferably, the steps of reconstructing the pure electricity meter status signal include:

[0027] Construct a collaborative signal decomposition model for distinguishing different manifestation forms of sensor fault characteristics and electricity meter fault characteristics;

[0028] Construct a feature basis function library for sensor faults and electricity meter faults respectively to form a dual-mode adaptive basis function library;

[0029] Based on the collaborative decomposition model and the dual-mode adaptive basis function library, decompose the original signal into a sensor fault component and an electricity meter fault component;

[0030] Calculate the contribution ratios of the sensor fault component and the energy meter fault component to the original signal, and generate an energy distribution map for evaluating the relative importance components of the two types of faults;

[0031] According to the weight analysis of the component energy distribution map, eliminate the sensor fault component, adjust the separation threshold, and combine with the normal signal baseline to reconstruct the pure energy meter status signal.

[0032] Preferably, the steps of verifying whether the pure energy meter status signal conforms to the physical consistency law and outputting the fault verification result include:

[0033] Based on the working principle of the energy meter, establish an energy meter physical constraint model set including electrical models, thermal models, and mechanical models, and form a physically quantifiable verification constraint equation set;

[0034] Substitute the pure energy meter status signal into the physical constraint equation set, calculate the physical constraint satisfaction degree, and generate a physical consistency score;

[0035] Adjust the parameters and thresholds in the physical constraint equation set according to the current working condition type to make the constraint test adapt to the current working condition, and output the working condition adaptive constraint test result;

[0036] For the fault characteristics identified in the pure energy meter status signal, check whether they conform to the physical manifestation laws of the corresponding fault types, and generate a physical rationality score for the fault characteristics;

[0037] Fuse the above outputs (physical consistency score, working condition adaptive constraint test result, and physical rationality score of fault characteristics) to generate a fault verification result, including the verification passed status and credibility assessment.

[0038] Preferably, the process of establishing a sensor health status model and monitoring the working status of the sensor to obtain the sensor health assessment result includes:

[0039] Based on the historical normal operation data in the standard data, establish a sensor normal response characteristic model;

[0040] Compare the current sensor real-time data with the sensor normal response characteristic model, detect the sensor characteristic drift, and generate a sensor characteristic drift index;

[0041] Perform frequency domain analysis on the standard data, extract the sensor frequency domain characteristics, and construct a sensor frequency domain feature vector;

[0042] Perform multi-sensor cross-verification, analyze the physical correlation between different sensor data, and generate a sensor credibility score;

[0043] Fuse the above outputs (sensor characteristic drift index, sensor frequency domain feature vector, and sensor credibility score) to construct a sensor health status model and output the sensor health assessment results, including the sensor health index and the abnormal type identification result.

[0044] Preferably, the process of generating the sensor credibility score includes:

[0045] Analyze the physical dependence relationships and associations among the sensor data in the standard data to obtain a physical constraint relationship matrix;

[0046] Based on the physical constraint relationship matrix, perform a consistency check on the standard data, calculate the residual values of each physical constraint equation, and form a sensor consistency residual vector;

[0047] Decompose the sensor consistency residual vector into the contributions of each sensor and output the sensor abnormal contribution metric value;

[0048] Identify the subsets of sensors that can verify each other, construct a sensor redundancy group, and form a sensor group mapping table;

[0049] For each group in the sensor group mapping table, extract the complementary information of the sensors within the group and construct a sensor mutual information matrix;

[0050] Fuse the sensor abnormal contribution metric value and the sensor mutual information matrix, and calculate and output the sensor credibility score through a credibility scoring function.

[0051] Preferably, the process of forming the sensor group mapping table is specifically as follows:

[0052] Calculate the physical correlation distance matrix between all pairs of sensors in the standard data;

[0053] Based on the physical constraint relationship matrix, determine the initial clustering centers, perform principal component analysis on the physical constraint relationship matrix, select the sensors with the largest contribution to each principal component as candidate centers, and obtain an initial clustering center set;

[0054] Use the initial clustering center set as the starting clustering centers, assign each sensor to the nearest center, and reselect the new center within each group that can minimize the total distance to obtain an initial partition of the sensor groups;

[0055] Calculate the internal physical redundancy of each sensor group, adjust the critical groups, and output the optimized sensor groups;

[0056] Correspond each sensor group to the physical constraint relationship matrix, confirm that the sensors within the group form a meaningful physical association unit, and obtain a sensor group mapping table with physical significance.

[0057] Preferably, the steps of generating the state feature vector of the electric energy meter include:

[0058] Perform multi-scale wavelet packet decomposition on the standard data, extract time-domain, frequency-domain, and time-frequency joint features, and construct an initial multi-scale feature set;

[0059] Utilize the electrical characteristics and thermal model of the electric energy meter to enhance the initial multi-scale feature set with physical knowledge and generate a physically enhanced feature set;

[0060] Apply a feature optimization algorithm and dimensionality reduction processing to the physically enhanced feature set to form an optimized electric energy meter feature vector;

[0061] According to the working condition information of the electric energy meter, perform dynamic weight adjustment on the optimized electric energy meter feature vector and output a working condition adaptive feature vector;

[0062] Identify the working condition adaptive feature vector to generate a state feature vector of the electric energy meter and a preliminary fault type determination result.

[0063] Beneficial effects: The present invention can effectively distinguish between sensor faults and electric energy meter faults, improve the accuracy of fault prediction, reduce the false alarm rate, and optimize the maintenance decision. The relevant technical effects will be described in detail in combination with the embodiments. Description of the Drawings

[0064] Figure 1 is the flowchart of the present invention.

[0065] Figure 2 is the flowchart of the first embodiment of the present invention for performing double-layer fault source separation and identification.

[0066] Figure 3 is the flowchart of the second embodiment of the present invention for performing double-layer fault source separation and identification.

[0067] Figure 4 is the flowchart of the present invention for reconstructing a pure electric energy meter state signal. Detailed Embodiments

[0068] According to one aspect of the present application, the steps of quantifying multi-source uncertainty and evaluating reliability based on the fault type and confidence level to generate a fault assessment report include:

[0069] Based on the sensor health assessment result and historical calibration data, establish a sensor measurement error model to quantify the uncertainty range of each sensor data point;

[0070] Utilize the Monte Carlo method to perform uncertainty propagation analysis on the feature extraction process and the fault diagnosis algorithm, and calculate the model inference uncertainty index;

[0071] Analyze the influence of factors such as environmental temperature, humidity, and electromagnetic interference on sensor data and diagnostic results, and establish an environmental impact sensitivity model;

[0072] Fuse the sensor measurement error model, the model inference uncertainty index, and the environmental impact sensitivity model to construct a comprehensive multi-source uncertainty evaluation framework, and calculate the confidence interval and risk level of the fault diagnosis result;

[0073] Based on the final fault type determination, fault confidence, and the comprehensive multi-source uncertainty evaluation results, generate a fault evaluation report including the fault type, location, severity, confidence interval, and reliability score.

[0074] According to one aspect of the present application, the steps of dynamically adjusting the weights of the optimized electric energy meter feature vectors according to the operating conditions information of the electric energy meter and outputting the operating condition adaptive feature vectors include:

[0075] For each feature in the optimized electric energy meter feature vector, calculate its ability to distinguish faults from normal states under various operating conditions, and construct a feature-operating condition sensitivity matrix;

[0076] Based on the operating parameters of the electric energy meter, design an operating condition pattern recognizer to classify the current operating condition into predefined categories, and output the current operating condition type and operating condition confidence;

[0077] For each operating condition type, construct an adaptive evaluation function of feature importance based on the feature-operating condition sensitivity matrix, and combine the historical accuracy rate of fault diagnosis and the environmental adaptability factor to output the operating condition adaptive weight function;

[0078] Apply the operating condition adaptive weight function to rank the importance of the optimized electric energy meter feature vectors, design a threshold-based feature automatic selection algorithm, and output the operating condition optimized feature subset;

[0079] Perform weighted combination of the features in the operating condition optimized feature subset, dynamically adjust the combination coefficient according to the current operating condition type and operating condition confidence, and generate the operating condition adaptive feature vector.

[0080] According to one aspect of the present application, the steps of using the physical working principle of the electric energy meter to design a physical consistency checker to verify whether the pure electric energy meter state signal conforms to the physical law and output the fault verification result include:

[0081] Based on the working principle of the electric energy meter, establish a set of electric energy meter physical constraint models including electrical models, thermal models, and mechanical models to form a physically quantifiable verification constraint equation system;

[0082] Substitute the pure electric energy meter state signal into the physical constraint equation system, calculate the physical constraint satisfaction degree, and generate a physical consistency score;

[0083] Dynamically adjust the parameters and thresholds in the physical constraint equation set according to the current working condition type, so that the constraint test adapts to different working condition conditions, and output the working condition adaptive constraint test result;

[0084] For the fault characteristics identified in the pure electric energy meter status signal, check whether they conform to the physical manifestation laws of the corresponding fault types, and generate a physical rationality score for the fault characteristics;

[0085] Fuse the physical consistency score, the working condition adaptive constraint test result and the physical rationality score of the fault characteristics to generate the final fault verification result, including the verification pass status and the confidence level assessment.

[0086] According to one aspect of the present application, the steps of designing a constraint projection algorithm to decompose the sensor consistency residual vector into the contributions of each sensor and output the sensor abnormal contribution metric value include:

[0087] Based on the physical constraint relationship matrix and the current sensor measurement values, calculate the partial derivative of each constraint equation with respect to each sensor, and construct a sensitivity matrix;

[0088] Construct a weighted pseudo-inverse projection matrix, project the sensor consistency residual vector into the sensor space, and estimate the sensor measurement error vector;

[0089] Calculate the standardized score of each sensor error, apply the improved CUSUM algorithm to accumulate the scores, and output the cumulative abnormal score of each sensor;

[0090] Set an adaptive threshold, identify abnormal sensors, calculate the abnormal contribution degree, and output a preliminary set of abnormal sensors;

[0091] Re-evaluate the residuals in an iterative manner, recalculate the residuals based on the currently identified subset of normal sensors, perform residual projection, identify new abnormal sensors until the residuals are small enough or no new abnormalities are detected, and output the final sensor abnormal contribution metric value.

[0092] According to one aspect of the present application, the steps of designing a causal inference algorithm with two-way propagation to infer possible fault sources through a message passing mechanism between the upper and lower layers and output the fault source probability distribution include:

[0093] Map the data in the observation evidence set to the corresponding nodes of the sensor-electric energy meter double-layer causal graph, and initialize the probability distribution of the evidence nodes;

[0094] Perform top-down inference, starting from the potential fault nodes in the sensor layer, calculate its impact on the measurement abnormal nodes through the calibrated causal model parameters, and generate the top-down propagation result;

[0095] Perform bottom-up reasoning, starting from the observed abnormal nodes, calculate the probabilities of possible electricity meter fault nodes through Bayesian backward reasoning, and generate the bottom-up propagation results;

[0096] Design a cross-layer message passing mechanism to integrate the top-down propagation results and the bottom-up propagation results, handle conflicts between layers, and calculate the integrated fault probability;

[0097] Update the node probabilities in an iterative manner until convergence or the maximum number of iterations is reached, and output the final probability distribution of the fault sources.

[0098] According to one aspect of the present application, the steps of constructing three competitive hypotheses of sensor fault, electricity meter fault, and hybrid fault for high-probability candidate fault sources, evaluating the posterior probabilities, and outputting the optimal fault hypothesis and the set of candidate fault sources include:

[0099] Based on the fault source probability distribution, construct three competitive fault hypothesis models: Hypothesis 1 is a pure sensor fault, Hypothesis 2 is a pure electricity meter fault, and Hypothesis 3 is a hybrid fault of sensors and electricity meters;

[0100] For each hypothesis model, calculate the likelihood value of its explaining the observed data based on the sensor-electricity meter double-layer causal graph;

[0101] Introduce a model complexity penalty term, adjust the hypothesis model according to the Occam's razor principle to avoid overly complex explanations, and calculate the adjusted hypothesis score;

[0102] Apply the Bayesian factor to evaluate the relative support between hypotheses and calculate the posterior probability of each hypothesis;

[0103] Select the hypothesis with the highest posterior probability as the optimal fault hypothesis, and extract the high-probability fault sources from this hypothesis to form a set of candidate fault sources.

[0104] According to one aspect of the present application, the steps of decomposing the original signal into sensor fault components and electricity meter fault components by applying a signal decomposition algorithm with sparse constraints based on the collaborative decomposition model and the dual-mode adaptive basis function library include:

[0105] Construct an optimization objective function, including a data fitting term, a sparse constraint term, and a physical consistency constraint term, to form a constrained optimization objective function;

[0106] Design differential sparse constraints based on the characteristics of sensor faults and electricity meter faults, impose stronger sparsity constraints on the sensor fault components, and determine the sparse regularization parameter;

[0107] Introduce the electricity meter physical model constraint to ensure that the decomposition result conforms to the electricity meter working principle, and construct a decomposition model with physical constraints;

[0108] Apply the accelerated proximal gradient descent algorithm to solve the optimization objective function, iteratively optimize the coefficients of the sensor fault component and the energy meter fault component, and obtain the preliminary decomposition result;

[0109] According to the signal decomposition quality and the satisfaction of physical constraints, adaptively adjust the regularization parameter to balance the decomposition accuracy and physical rationality, and output the final sensor fault component and energy meter fault component.

[0110] According to one aspect of the present application, the steps of calculating the contribution ratio of the sensor fault component and the energy meter fault component to the original signal, generating a component energy distribution map, and evaluating the relative importance of the two types of faults include:

[0111] Calculate the energy values of the sensor fault component and the energy meter fault component in the time domain and frequency domain to quantify the absolute intensity of each component;

[0112] Analyze the proportion of the energy of each component in the total energy of the original signal to determine the relative contribution ratio of the sensor fault and the energy meter fault;

[0113] Study the distribution characteristics of each component in the time-frequency domain, identify the time localization and frequency characteristics of the fault component, and form the time-frequency domain component characteristics;

[0114] Combined with the fault type knowledge base, evaluate the influence degree of different components on the system performance, and calculate the fault importance index;

[0115] Fuse the energy value, contribution ratio, time-frequency characteristics and importance index to generate a multi-dimensional component energy distribution map for guiding the subsequent pure signal reconstruction process.

[0116] Aiming at the problem that it is difficult to distinguish between the sensor's own faults and the actual faults of the energy meter, the existing technologies mainly use single redundancy or simple threshold to judge sensor faults, lacking a joint evaluation mechanism, resulting in a high false alarm rate (30 - 45%) and waste of maintenance resources. Therefore, in this application, through the Sensor Characteristic Deviation Analysis (SCDA) method, a normal working characteristic model of the sensor is constructed and the deviation is detected. However, after applying SCDA, it is found that when multiple sensors work together, the fault effects cross-influence and it is difficult to clarify the fault source. Therefore, through the Cooperative Fault Source Separation Algorithm (CFSA), a fault probability propagation model based on Bayesian network and a multi-sensor collaborative state evaluation method are designed. In order to improve the robustness under dynamic working conditions and multi-source uncertainties. Through the Double-layer Adaptive Fault Identification Framework (DAFIF), a hierarchical fault identification structure and a cross-layer constraint mechanism are constructed, and a recursive filter is introduced to improve the robustness. Therefore, in this solution, CFSA and DAFIF are fused to form a complete framework, including a characteristic modeling layer, a deviation analysis layer, a double-layer identification layer and an uncertainty processing layer, effectively solving the key technical problem of separating and identifying sensor faults and energy meter faults.

[0117] Case 1

[0118] S1: Multi-modal Sensor Data Acquisition and Preprocessing

[0119] Read the raw data of the multi-modal sensor of the electricity meter, including voltage sensor data, current sensor data, temperature sensor data, vibration sensor data, and acoustic sensor data, synchronize them through the timestamp alignment algorithm, and obtain the time-aligned multi-modal sensor data.

[0120] Apply the adaptive wavelet threshold denoising algorithm to the time-aligned multi-modal sensor data, identify and mark the sudden outliers, and perform smoothing processing using local median filtering, and output the denoised sensor data.

[0121] For the denoised sensor data, according to the range and physical characteristics of each sensor, perform adaptive Z-score standardization, map the data of different physical quantities to a unified metric space, and obtain the standardized multi-modal sensor data stream.

[0122] Analyze the working state characteristics of the electricity meter in the standardized multi-modal sensor data stream, identify different working conditions such as steady-state operation, load change, and startup process, and divide the data stream into data segments with working condition labels as the basis for subsequent analysis.

[0123] S2: Sensor Health State Modeling and Monitoring

[0124] Based on the historical normal operation data in the standardized multi-modal sensor data stream, establish a sensor normal response characteristic model for each type of sensor, including core parameters such as sensitivity, linearity, response time, and noise level.

[0125] Compare the current sensor real-time data with the sensor normal response characteristic model, and use the cumulative sum control chart (CUSUM) method to detect sensor characteristic drift, and output the sensor characteristic drift index.

[0126] Perform short-time Fourier transform (STFT) on the standardized multi-modal sensor data stream, extract the sensor signal spectrum characteristics, construct the sensor frequency domain feature vector, and identify the sensor abnormal mode by comparing with the normal spectrum template.

[0127] Design a multi-sensor cross-validation algorithm, analyze the physical correlation between different sensor data, verify the sensor data consistency based on the law of conservation of energy and physical constraint conditions, and generate a sensor credibility score.

[0128] Fuse the sensor characteristic drift index, the analysis result of the sensor frequency domain feature vector, and the sensor credibility score, and use the improved evidence theory to construct a sensor health state model, and output the health index of each sensor and the recognition result of the abnormal type.

[0129] S3: Fault Feature Extraction and Pattern Recognition of Electric Energy Meters

[0130] Perform multi-scale wavelet packet decomposition on the standardized multi-modal sensor data stream, extract time-domain, frequency-domain, and time-frequency joint features, and construct an initial multi-scale feature set.

[0131] Utilize the electrical characteristics and thermal model of the electric energy meter to enhance the physical knowledge of the initial multi-scale feature set, calculate physical features such as power factor, harmonic distortion, and phase shift, and generate a physically enhanced feature set.

[0132] Apply the recursive feature elimination algorithm combined with principal component analysis to the physically enhanced feature set, remove redundant and low-information features, retain the most discriminative feature subset, and form an optimized electric energy meter feature vector.

[0133] According to the data segment information marked by the working conditions, design a feature importance evaluation function for different working conditions, dynamically adjust the feature weights, and output a working condition adaptive feature vector.

[0134] Use the improved support vector machine algorithm to classify the working condition adaptive feature vector, identify the normal state and potential fault modes, and generate an electric energy meter state feature vector and a preliminary fault type determination result.

[0135] S4: Double-layer Fault Source Separation and Identification

[0136] Construct a Bayesian network model to express the conditional dependence relationship between the sensor health state model and the electric energy meter state feature vector, quantify the influence degree of sensor anomalies on the electric energy meter fault judgment, and output a fault correlation matrix.

[0137] Design a double-layer causal inference framework, transform the fault correlation matrix into a causal graph structure, and determine whether the observed anomaly is due to sensor faults or actual electric energy meter faults through reverse inference, generating a set of candidate fault sources.

[0138] For each candidate fault in the set of candidate fault sources, apply the collaborative signal decomposition algorithm to decompose the original signal into a sensor fault component and an electric energy meter fault component, and reconstruct the pure electric energy meter state signal after removing the sensor fault.

[0139] Utilize the physical working principle of the electric energy meter and the internal constraint relationship between multi-sensor data to design a physical consistency checker to verify whether the separated pure electric energy meter state signal conforms to physical laws, and output a fault verification result.

[0140] Combine the sensor health index, the electric energy meter state feature vector, and the fault verification result, and adopt a dynamic decision tree structure to realize the hierarchical identification of sensor faults and electric energy meter faults, and output the final fault type determination and fault confidence.

[0141] S5: Multi-source Uncertainty Quantification and Reliability Assessment

[0142] Based on the sensor health index and historical calibration data, establish a sensor measurement error model to quantify the uncertainty range of each sensor data point.

[0143] Using the Monte Carlo method, perform uncertainty propagation analysis on the feature extraction process and fault diagnosis algorithm, and calculate the model inference uncertainty index.

[0144] Analyze the influence of factors such as environmental temperature, humidity, and electromagnetic interference on sensor data and diagnosis results, establish an environmental impact sensitivity model, and evaluate the interference degree of environmental conditions on fault judgment.

[0145] Fuse the sensor measurement error model, model inference uncertainty index, and environmental impact sensitivity model to construct a multi-source uncertainty comprehensive evaluation framework, and calculate the confidence interval and risk level of the fault diagnosis result.

[0146] Based on the final fault type determination, fault confidence, and multi-source uncertainty comprehensive evaluation results, generate a fault assessment report including fault type, location, severity, confidence interval, and reliability score.

[0147] In this embodiment, the physical correlation between multiple sensors is used for mutual inspection to realize the assessment of the health status of the sensors themselves. Through a double-layer causal inference framework and a collaborative signal decomposition algorithm, the effective separation of sensor faults and electricity meter faults is achieved. Based on the working condition-based feature adaptive selection mechanism, the problem of feature sensitivity change under different working conditions is solved. Based on the physical constraint-based fault verification method, it is ensured that the fault separation result conforms to the physical working principle of the electricity meter.

[0148] In this embodiment, the process of sensor cross-validation and self-diagnosis to obtain a credibility score is specifically as follows:

[0149] Analyze the physical dependence relationship between the data of each sensor in the standardized multi-modal sensor data stream, establish a sensor association constraint equation set, including: power balance constraint (voltage × current = power), thermodynamic constraint (relationship between power consumption and temperature rise), electromechanical coupling constraint (relationship between vibration and electrical parameters), etc., and output the physical constraint relationship matrix.

[0150] Based on the physical constraint relationship matrix, perform a consistency check on the standardized multi-modal sensor data stream, calculate the residual values of each physical constraint equation, form a sensor consistency residual vector, and quantify the coordination degree of the data between sensors.

[0151] Adopt improved clustering analysis to identify subsets of sensors that can verify each other, construct sensor redundancy groups, where the sensor data within each group can corroborate or be derived from each other, and form a sensor group mapping table.

[0152] Design a constraint projection algorithm to decompose the sensor consistency residual vector into the contributions of each sensor, and achieve precise positioning of abnormal sensors. Specifically: construct a sensitivity matrix of the residuals to each sensor; perform weighted pseudo-inverse operations to calculate the contribution of each sensor to the total residuals; apply statistical hypothesis testing to identify sensors with significant residual contributions and output the abnormal contribution metric values of the sensors.

[0153] For each group in the sensor group mapping table, extract the complementary information of the sensors within the group, construct a sensor mutual information matrix, and quantify the information redundancy and complementarity between different sensors.

[0154] Fuse the sensor abnormal contribution metric values and the sensor mutual information matrix, and design a credibility scoring function, including: calculating the stability of the residual contribution of each sensor within multiple time windows; analyzing the consistency degree between the sensor and other sensors within its redundant group; considering the historical reliability record and environmental adaptability factors of the sensor, and finally outputting the credibility score of each sensor.

[0155] In this embodiment, for the comprehensive assessment of the sensor health status, calculate the sensor health index, specifically:

[0156] Integrate the sensor characteristic drift index, the analysis results of the sensor frequency domain feature vectors, and the sensor credibility score, construct a multi-dimensional sensor health feature space, and comprehensively characterize the sensor state.

[0157] Design an improved Dempster-Shafer evidence theory framework applicable to sensor health assessment. The main improvement points include: introducing a basic probability assignment function based on the physical characteristics of the sensors; designing a dynamic adjustment mechanism for the feature reliability weights; developing an optimized combination rule for conflicting evidence, and outputting an evidence fusion model.

[0158] Based on historical data and expert knowledge, establish a sensor typical fault mode feature library, including: response delay feature mode, zero drift feature mode, sensitivity degradation feature mode, non-linear distortion feature mode, abnormal noise level feature mode. Each mode contains time-domain and frequency-domain feature descriptors.

[0159] Apply the evidence fusion model to fuse the features in the multi-dimensional sensor health feature space, calculate the probability mass of each sensor belonging to various states, and deduce the probability distribution of the sensor health status.

[0160] Match the probability distribution of the sensor health status with the characteristic library of typical sensor fault modes to identify the most likely type of sensor abnormality. At the same time, calculate the normalized sensor health index based on the probability values of the health status, with a quantification range of 0 - 100, where 100 represents a completely healthy state.

[0161] In this embodiment, the process of generating the working condition adaptive feature vector based on the characteristic adaptive selection of the working condition is specifically as follows:

[0162] For each feature in the physical enhancement feature set, calculate its ability to distinguish between the faulty and normal states under various working condition conditions, and apply Fisher discriminant ratio and mutual information analysis to construct a feature - working condition sensitivity matrix.

[0163] Based on the operating parameters of the electric energy meter in the data segment marked with the working condition, design a working condition pattern recognizer to divide the current working condition into predefined categories (such as no - load, light - load, heavy - load, fluctuating load, etc.), and output the current working condition type and working condition confidence.

[0164] For each type of working condition, construct an adaptive evaluation function for feature importance: set the basic weight based on the feature - working condition sensitivity matrix; adjust the weight coefficient in combination with the historical accuracy rate of fault diagnosis; introduce an environmental adaptability factor to compensate for the impact of environmental changes on the reliability of features; and output the working condition adaptive weight function.

[0165] Apply the working condition adaptive weight function to rank the importance of the physical enhancement feature set, design a threshold - based feature automatic selection algorithm to dynamically determine the optimal feature subset, and at the same time consider the redundancy and complementarity between features, and output the working condition optimized feature subset.

[0166] Perform weighted combination of the weights of the features in the working condition optimized feature subset to construct a composite feature, and dynamically adjust the combination coefficient according to the current working condition type and working condition confidence, and finally generate the working condition adaptive feature vector.

[0167] In this embodiment, the process of tracing the fault source based on causal reasoning is specifically as follows:

[0168] Based on the fault correlation matrix and the knowledge of the electric energy meter system, construct a two - layer causal graph structure: the upper layer represents the causal relationship between sensor faults and measurement abnormalities; the lower layer represents the causal relationship between electric energy meter faults and physical parameter abnormalities; the inter - layer connection represents the observational relationship between measurement abnormalities and physical parameter abnormalities; and output the sensor - electric energy meter two - layer causal graph.

[0169] Use historical fault data and expert knowledge to initialize the conditional probability table in the sensor - electric energy meter two - layer causal graph, and then perform parameter optimization based on the reliably marked measured data, and output the calibrated causal model parameters.

[0170] Integrate the sensor health index, the state feature vector of the electricity meter, and the current measurement data to form an observation evidence set, which serves as the input for causal reasoning.

[0171] Develop a causal reasoning algorithm with two-way propagation: starting from possible sensor faults top-down, infer the possible measurement anomalies they may cause; starting from the observed anomalies bottom-up, infer the most likely potential faults; propagate probability information in the double-layer causal graph through a message passing mechanism; output the probability distribution of the fault sources.

[0172] For the candidate fault sources with relatively high probabilities in the probability distribution of the fault sources, construct competing hypotheses:

[0173] Hypothesis 1: All observed anomalies are due to sensor faults; Hypothesis 2: All observed anomalies are due to electricity meter faults; Hypothesis 3: Some of the observed anomalies are due to sensor faults and some are due to electricity meter faults; Apply the Bayesian model selection method to evaluate the posterior probabilities of each hypothesis, and output the optimal fault hypothesis and the set of candidate fault sources.

[0174] In this embodiment, the process of fault feature separation and reconstruction is specifically as follows:

[0175] Develop a collaborative signal decomposition model, considering the different manifestations of sensor fault features and electricity meter fault features: Sensor faults usually manifest as sudden jumps, drifts, or increased noise; Electricity meter faults usually follow specific physical evolution laws; Based on this difference, design a matrix decomposition framework with physical constraints, and output a collaborative decomposition mathematical model.

[0176] Construct a feature basis function library for sensor fault modes and electricity meter fault modes respectively: Sensor fault basis functions: including step functions, ramp functions, oscillation functions, etc.; Electricity meter fault basis functions: A family of feature functions designed based on physical models are combined to form a dual-mode adaptive basis function library.

[0177] Based on the collaborative decomposition mathematical model and the dual-mode adaptive basis function library, design a signal decomposition algorithm with sparse constraints: Construct an optimization objective function, including a data fitting term, a sparse constraint term, and a physical consistency constraint term; Use the accelerated proximal gradient descent algorithm to solve the optimization problem; Through the adaptive adjustment of the regularization parameter, balance the decomposition accuracy and physical rationality; Output the signal decomposition results, including the sensor fault component and the electricity meter fault component.

[0178] Calculate the contribution ratios of the sensor fault component and the electricity meter fault component to the original signal respectively, generate a component energy distribution map, and visually show which type of fault dominates.

[0179] Eliminate the sensor fault component, only retain the electricity meter fault component, and combine it with the normal signal baseline to reconstruct the pure electricity meter state signal, eliminating the interference introduced by sensor faults.

[0180] In this embodiment, the process of fault verification based on physical constraints includes:

[0181] Based on the working principle of the electric energy meter, establish a set of physical constraint models for the electric energy meter, including: an electrical model, specifically the relationships of voltage, current, power, and power factor; a thermal model, specifically the relationships of power consumption, temperature rise, and heat dissipation; a mechanical model, specifically the vibration transmission and resonance characteristics; and output a set of physical constraint equations that can be quantitatively verified.

[0182] Substitute the state signal of the pure electric energy meter into the physical constraint equations, calculate the degree of satisfaction of the physical constraints, generate a physical consistency score, and quantify the physical rationality of the reconstructed signal.

[0183] According to the current working condition type, dynamically adjust the parameters and thresholds in the physical constraint equations to make the constraint test adapt to different working condition conditions, and output the working condition adaptive constraint test results.

[0184] For the fault features identified in the fault components of the electric energy meter, check whether they conform to the physical manifestation laws of the corresponding fault types, and generate a physical rationality score for the fault features.

[0185] Fuse the physical consistency score, the working condition adaptive constraint test results, and the physical rationality score of the fault features to generate the final fault verification results, including the verification passed status and the credibility assessment.

[0186] In this embodiment, the process of identifying the redundant information sensor group is specifically as follows:

[0187] Introduce a distance metric enhanced by physical knowledge d_phys(S_i, S_j) = w_1*d_corr(S_i, S_j) + w_2*d_info(S_i, S_j) + w_3*d_constraint(S_i, S_j); where d_corr is the temporal correlation distance, d_info is the mutual information distance, and d_constraint is the physical constraint deviation distance

[0188] Use the physical constraint relationship matrix to guide the selection of the initial center: identify the subsets of sensors that are mutually related in the physical constraint equations; select a representative sensor from each associated subset as the initial clustering center; ensure that the initial clustering has physical meaning rather than pure statistical association;

[0189] Design a clustering number evaluation function based on physical redundancy CR(k) = λ_1*WSS(k) + λ_2*BSS(k) + λ_3*PRI(k); WSS is the within-group sum of squares, BSS is the between-group sum of squares, and PRI is the physical redundancy index; automatically determine the optimal number of groups by minimizing CR(k);

[0190] During operation, first read the standardized multi-modal sensor data stream (standard data) and the physical constraint relationship matrix; calculate the physical correlation distance matrix between all pairs of sensors: obtain the n×n sensor physical correlation distance matrix D.

[0191] Determine the initial clustering centers based on the physical constraint relationship matrix: perform principal component analysis on the physical constraint relationship matrix; select the sensors with the largest contribution to each principal component as candidate centers; apply the maximum-minimum distance criterion to optimize the set of candidate centers; obtain the initial clustering center set C_init.

[0192] Use the initial clustering center set C_init as the starting clustering center and perform iterations: (a) Assign each sensor to the nearest center; (b) Re-select a new center within each group that can minimize the total distance; (c) If the centers no longer change or reach the maximum number of iterations, stop; output the initial partition G_init of the sensor groups.

[0193] Calculate the internal physical redundancy PR_i for each group G_i; adjust the critical groups, merge the groups with low redundancy; split the groups that are too large and highly heterogeneous; verify whether each group meets the minimum redundancy requirement. Obtain the optimized sensor groups G_opt and the sensor group mapping table.

[0194] Correspond each group to the physical constraint relationship matrix; confirm that the sensors within each group form a meaningful physical association unit; adjust the groups that cannot be physically explained; output the sensor redundancy groups with physical meaning and the mapping table

[0195] In this embodiment, the abnormal sensor localization process based on constraint projection is specifically as follows:

[0196] Discretize the m constraint equations in the physical constraint relationship matrix into a time series form;

[0197] For time point t, the j-th constraint equation is expressed as: f_j(S_1(t), S_2(t),..., S_n(t)) = 0; where S_i(t) represents the measurement value of the i-th sensor at time t;

[0198] For each time point t and each constraint equation j, calculate the residual after substituting the actual measurement values: r_j(t) = f_j(S_1(t), S_2(t), ..., S_n(t)); Ideally, r_j(t) should be close to zero; Deviation from zero indicates a constraint violation; For each time window W, collect the residuals of all constraint equations at all time points within this window; Construct an m-dimensional residual vector r(W), where each element is the average residual of a constraint equation within window W; Finally, form the sensor consistency residual vector R = [r_1(W), r_2(W), ..., r_m(W)]^T;

[0199] The process of the constraint projection algorithm is as follows:

[0200] Obtain the physical constraint equations and the current sensor measurement values, and calculate the partial derivative of each constraint equation with respect to each sensor: J_ij = df_i / dS_j|current measurement point; Obtain the m×n sensitivity matrix J.

[0201] Construct the weighted pseudo-inverse projection matrix: P = (J^T·W·J)^(-1)·J^T·W; where W is the diagonal matrix of the reliability weights of the constraint equations;

[0202] Perform residual projection δS = P·R; where δS represents the estimated sensor measurement error vector;

[0203] Calculate the standardized score of each sensor error Z_i = |δS_i| / σ_i; where σ_i is the historical error standard deviation of the i-th sensor;

[0204] Apply the improved CUSUM algorithm to accumulate the scores C_i(t) = max(0, C_i(t-1) + Z_i(t) - k); where k is the sensitivity parameter; Obtain the cumulative anomaly scores C_i of each sensor.

[0205] Set the adaptive threshold T_i = α·σ_C_i + β; Identify the abnormal sensor. If C_i > T_i, then mark sensor i as abnormal; Calculate the anomaly contribution degree A_i = (C_i - T_i) / T_i if C_i > T_i; A_i = 0 if C_i ≤ T_i; The sensor anomaly contribution metric value A = [A_1, A_2, ..., A_n]^T.

[0206] Initialize all sensors to the normal state; repeat until the residual is small enough or no new anomalies are detected: (a) Recalculate the residual based on the current subset of normal sensors; (b) Perform residual projection to identify new anomalous sensors; (c) Update the set of sensor states; (d) Verify the remaining residual after removing the anomalous sensors; obtain the final sensor anomaly contribution metric value and the set of anomalous sensors.

[0207] In this embodiment, the process of residual vector decomposition is specifically as follows:

[0208] Group the m constraint equations according to the subsets of sensors involved; form k relatively independent constraint groups G_1, G_2, ..., G_k; each constraint group generates an independent residual subvector; for each constraint group G_i, construct a local sensitivity matrix J_i; perform local residual projection δS_i = P_i·R_i; merge the projection results of each group, considering the inter-group correlation;

[0209] Calculate the residual vector on multiple time scales (short-term, medium-term, long-term); the short-term residual focuses on sudden anomalies, and the long-term residual focuses on drift anomalies; fuse the results of multi-scale analysis to form a comprehensive anomaly assessment;

[0210] Introduce a sparsity regularization term, assuming that most sensors are working properly; add a smoothness constraint to suppress the oscillation of anomaly detection; optimize the objective function min ||R - J·δS|| 2 _W + λ 1 ||δS|| 1 + λ 2 ||D·δS|| 2 .

[0211] Case 2 includes the following steps:

[0212] In this embodiment, taking the three-phase intelligent electricity meters deployed in a certain intelligent distribution network as the research object, five types of sensors are configured: voltage sensors (range 0 - 300V, accuracy ±0.2%), current sensors (range 0 - 100A, accuracy ±0.5%), temperature sensors (range -20°C to 100°C, accuracy ±0.5°C), vibration sensors (range 0 - 20g, accuracy ±2%) and acoustic sensors (range 30 - 130dB, accuracy ±3dB).

[0213] The data acquisition system uses a 24-bit high-precision AD converter and is configured with a sampling frequency of 10 kHz. To ensure data synchronization, a GPS-based time synchronization strategy is adopted, and the maximum time deviation between sensors is measured to be 43 μs. For the original timestamp TS_original(i) of the i-th sensor, the following formula is used for alignment: TS_aligned(i) = TS_original(i) - ΔT(i); where ΔT(i) is the time deviation of the i-th sensor relative to the reference clock. After time alignment processing, the time consistency error of each sensor's data is controlled within ±5 μs.

[0214] The adaptive wavelet threshold denoising algorithm is applied to the collected raw data. For the voltage signal, the db4 wavelet basis function is used for 4-layer decomposition, and the threshold λ = σ·√(2·log(N)); where σ is the noise standard deviation, estimated by the median absolute deviation (MAD) of the highest frequency band coefficients: σ = MAD / 0.6745; N is the signal length. In actual measurement, the value range of σ is usually 0.15 - 0.38 V. For the detected burst outliers, 7-point local median filtering is used for processing x_filtered(i) = median{x(i-3), x(i-2), x(i-1), x(i), x(i+1), x(i+2), x(i+3)}; after processing, the signal-to-noise ratio is increased by an average of 8.3 dB.

[0215] Z-score normalization is performed on the denoised data of each sensor x_normalized = (x - μ) / σ; where μ and σ are the data mean and standard deviation respectively. The calculation is based on a 20-minute sliding window and updated every 10 minutes. For example, for the current sensor data, the μ and σ on a typical working day are 43.7 A and 12.5 A respectively.

[0216] By analyzing the voltage and current characteristics, the working state of the electricity meter is identified, and the normalized data is divided into different working condition segments. The working condition classification criteria are: light load: load power < 20% of the rated power; medium load: load power between 20% - 60% of the rated power; heavy load: load power > 60% of the rated power; fluctuating load: load power change rate > 10% / minute;

[0217] In the measured data, the proportions of each working condition are: light load 37.2%, medium load 42.1%, heavy load 15.6%, and fluctuating load 5.1%.

[0218] Based on historical data, establish the normal characteristic models of each sensor. Taking the current sensor as an example, its normal model includes: linearity deviation < 0.8%, response time < 125 ms, and noise level < 0.3% of full scale. The training dataset contains 30 days of normal operation data, with a total of approximately 4.32×10^6 sample points.

[0219] Adopt the cumulative sum control chart (CUSUM) method to monitor the sensor characteristic drift. For the jth sensor characteristic parameter pj, the calculation formula of its CUSUM statistic is S^+(n) = max{0, S^+(n - 1) + (x_n - μ_0 - k)} and S^-(n)= max{0, S^-(n - 1) + (μ_0 - k - x_n)}; where, μ_0 is the normal mean of the parameter, and k is the reference value (with a value of 0.5σ). When S^+ or S^- exceeds the threshold h (with a value of 5σ), a drift alarm is triggered. In actual measurement, the zero - drift detection sensitivity of the voltage sensor is 0.23% of full scale.

[0220] Perform short - time Fourier transform (STFT) on the standardized sensor data to extract spectral features. The transformation parameters are set as follows: the window function is the Hanning window, the window length is 1024 points, and the overlap rate is 50%. Construct the sensor frequency - domain feature vector FV_i, which includes the energy ratios of 5 frequency bands and 3 harmonic indexes. Taking the temperature sensor as an example, when operating normally, the energy ratio of the low - frequency band (0 - 0.1 Hz) is 87.3% ± 3.2%, and this index rises to 96.5% ± 1.8% when a response - sluggish fault occurs.

[0221] Establish the sensor association constraint equations, including: power balance constraint P = U·I·cosφ; thermodynamic constraint ΔT = k·P·t / C + T_ambient; electromechanical coupling constraint V_vib = α·I^2 + β;

[0222] Where, P is the active power (W), U is the voltage (V), I is the current (A), cosφ is the power factor, ΔT is the temperature rise (℃), k is the heat conduction coefficient (0.42℃·W^-1·min^-1), t is the time (min), C is the heat capacity, T_ambient is the ambient temperature (℃), V_vib is the vibration amplitude (g), and α and β are the device characteristic parameters (the measured values are 0.0012 g·A^-2 and 0.023 g respectively).

[0223] Calculate the residuals of each physical constraint equation to form a consistency residual vector. Taking the power balance constraint as an example, the residual r_power(t) = P_measured(t) - U_measured(t)·I_measured(t)·cosφ_measured(t); during normal operation, r_power fluctuates within the range of ±2.1%.

[0224] Construct a sensor distance matrix using an improved physical correlation distance metric d_phys(S_i, S_j) = 0.3·d_corr(S_i, S_j) + 0.3·d_info(S_i, S_j) + 0.4·d_constraint(S_i, S_j); where d_corr is the Pearson correlation coefficient distance, d_info is the mutual information distance, and d_constraint is the physical constraint deviation distance. Apply the improved K-medoids clustering algorithm to identify 3 main sensor groups from the measured data: Group 1: voltage, current, power (physical redundancy 0.86); Group 2: current, temperature, vibration (physical redundancy 0.73); Group 3: vibration, acoustics (physical redundancy 0.68);

[0225] Construct a sensitivity matrix J (dimension m×n, where m is the number of constraints and n is the number of sensors), and calculate the partial derivative of the i-th constraint with respect to the j-th sensor J_ij = df_i / dS_j| at the current measurement point; taking the partial derivative of the power balance constraint with respect to the current sensor as an example, J_power,I = U·cosφ. Perform residual projection δS = P·R = (J^T·W·J)^-1·J^T·W·R; where W is a diagonal matrix representing the reliability weights of the constraint equations (the weight of the power balance constraint is 0.85, and the weight of the thermodynamic constraint is 0.75).

[0226] Calculate the standardized anomaly score: Z_i = |δS_i| / σ_i; apply the improved CUSUM algorithm to accumulate the anomaly scores: C_i(t) = max{0, C_i(t - 1) + Z_i(t) - 0.5}; when C_i exceeds the threshold T_i (set to 4.2), mark sensor i as abnormal and calculate the anomaly contribution degree: A_i = (C_i - T_i) / T_i; in the measured data, the anomaly contribution degree A_I caused by the zero drift of the current sensor is 1.75.

[0227] Based on the anomaly contribution degree and group consistency, calculate the credibility score Trust_i = (1 - 0.6·A_i)·(0.7·C_i + 0.3·H_i); where C_i is the consistency measure with the redundant group, and H_i is the historical reliability factor. In actual measurement, the credibility scores of normal sensors are usually in the range of 0.85 - 0.98, while those of faulty sensors drop to 0.35 - 0.60.

[0228] Fuse the sensor characteristic drift index, frequency domain features, and credibility scores, and use the improved Dempster - Shafer evidence theory to construct a health state model.

[0229] For the state set Θ = {normal, slightly abnormal, severely abnormal} of sensor i, construct the basic probability assignment function m(·) based on each characteristic evidence. For example, for the characteristic drift index, m_drift({normal}) = exp(-κ·S_drift) when S_drift < T_drift, m_drift({slightly abnormal, severely abnormal}) = 1 - m_drift({normal}); where κ is the sensitivity parameter (taking the value 1.5), S_drift is the drift statistic, and T_drift is the threshold (taking the value 4.2).

[0230] Fuse multi - source evidence through the Dempster combination rule m_1,2(A) = [Σ_{B∩C = A} m_1(B)·m_2(C)] / [1 - Σ_{B∩C = ∅} m_1(B)·m_2(C)];

[0231] Finally, calculate the sensor health index Health_i = 100·m({normal}) + 60·m({slightly abnormal}) + 20·m({severely abnormal}); In actual measurement, the index of healthy sensors is usually between 85 - 100, slightly abnormal is between 60 - 85, and severely abnormal is below 60.

[0232] Perform multi - scale wavelet packet decomposition on the standardized sensor data, and use the sym6 wavelet basis function for 4 - layer decomposition. Extract the statistical features of each decomposition node, including energy, entropy, kurtosis, skewness, etc., to construct the initial feature set F_init (with a dimension of 120).

[0233] Calculate enhanced features using the physical model of the electricity meter, including: Power Factor PF = P / S = P / √(P^2+Q^2); Total Harmonic Distortion THD = √(Σ_{i=2}^n I_i^2) / I_1; Phase Shift φ = arccos(P / S); Load Unbalance LU = max(|I_a-I_avg|,|I_b-I_avg|,|I_c-I_avg|) / I_avg; The dimension of the enhanced feature set F_enhanced is 152.

[0234] Apply the Recursive Feature Elimination (RFE) algorithm combined with Principal Component Analysis (PCA) to optimize the feature set, including: Calculate the feature importance scores using a random forest; Iteratively remove features with importance below the threshold (0.01); Perform PCA on the remaining features and retain the principal components with an explained variance ratio of 95%; The dimension of the optimized feature vector F_opt is reduced to 28.

[0235] Calculate the Fisher Discriminant Ratio FDR_i,j = |μ_i,1 - μ_i,0|^2 / (σ_i,1^2 + σ_i,0^2) for each feature under different working conditions; where μ_i,1 and μ_i,0 are the means of feature i in faulty and normal samples respectively, and σ_i,1 and σ_i,0 are the corresponding standard deviations. Construct a feature - working condition sensitivity matrix S, for example: The FDR of the Total Harmonic Distortion (THD) feature under light load conditions is 1.82, and it increases to 4.37 under heavy load conditions.

[0236] Design a working condition identifier based on voltage, current, and power factor parameters. Use a Support Vector Machine (SVM) classifier with an RBF kernel, penalty parameter C = 10, and γ = 0.1. The measured working condition identification accuracy reaches 96.3%.

[0237] For working condition j, the adaptive weight calculation formula for feature i is w_i,j = S_i,j·(1 + α·Acc_i,j)·(1 - β·E_i,j); where S_i,j is the feature - working condition sensitivity, Acc_i,j is the historical diagnosis accuracy gain, E_i,j is the environmental interference sensitivity coefficient, and α and β are balance parameters (0.3 and 0.2 respectively).

[0238] Based on the adaptive weights, sort the features and select the feature subset with an accumulated importance of 90%. The number of features selected under different working conditions are: 16 for light load, 19 for medium load, 14 for heavy load, and 23 for fluctuating load.

[0239] Perform weighted fusion on the selected features F_adaptive = Σ_i w_i,j·F_i;

[0240] Smooth transition when working condition changes: w_i,t = (1-α)·w_i,prev + α·w_i,current; where α is the transition rate (value is 0.3).

[0241] The improved support vector machine (SVM) algorithm is used to identify the state of the electric energy meter. The kernel function selects RBF, and the grid search is used for parameter optimization. The optimal parameters are C=8.5 and γ=0.07. The adaptive feature vector F_adaptive of the working condition is classified, and the fault probability and preliminary type judgment are output. The training set contains 2500 normal samples and 1800 fault samples (covering 8 typical faults), and the cross-validation accuracy rate reaches 89.7%.

[0242] A Bayesian network model is constructed to express the conditional dependency between the health status of the sensor and the status of the electric energy meter. The network consists of three layers: sensor layer (5 nodes), observation layer (12 nodes) and electric energy meter layer (8 nodes). The conditional probability table (CPT) represents the fault propagation relationship, for example: the conditional probability that the current sensor fault leads to the abnormal harmonic detection P (harmonic abnormality | current sensor fault) = 0.83. The initial CPT is constructed based on expert knowledge and historical data, and the parameters are optimized by maximum likelihood estimation using 1200 sets of labeled data.

[0243] Generate a fault correlation matrix R, which represents the correlation between sensor fault and electric energy meter fault, with a value range of [0,1]. For example, the correlation coefficient between voltage sensor and power circuit fault is 0.87.

[0244] Construct a two-layer causal graph structure, including:

[0245] Upper layer: represents the causal relationship between 5 sensor nodes (S1-S5) and 12 measurement anomaly nodes (M1-M12);

[0246] Lower layer: represents the causal relationship between 8 power meter fault nodes (F1-F8) and 12 physical parameter abnormal nodes (P1-P12);

[0247] Inter-layer connection: indicates the correspondence between measurement abnormal nodes and physical parameter abnormal nodes.

[0248] Initialize the conditional probability table, the probability of sensor failure affecting measurement anomaly P(M_j=abnormal|S_i=fault) = exp(-d_ij / 1.5); where d_ij is the impact distance parameter. Optimize the parameters using the expectation maximization (EM) algorithm based on 800 sets of labeled data, and the average parameter convergence error is 0.053.

[0249] Integrate the sensor health index, the power meter status vector, and the current measurement data to form the observation evidence set E. For example, in a certain observation, the health index of the current sensor is 57.3, and the measured value of the power factor deviates from the normal range by 3.2 standard deviations.

[0250] Design a two-way message passing mechanism from top to bottom and from bottom to top: Initialize the probability distribution of each node; Top-down propagation: Infer measurement anomalies from sensor nodes \(m_{down}(M_j)=\sum_iP(M_j|S_i)\cdot P(S_i)\); Bottom-up propagation: Infer the fault source from the observed anomalies \(m_{up}(S_i)=P(S_i)\cdot\prod_j[P(M_j|S_i) / P(M_j)]^{I(M_j\in E)}\); Message integration: \(P_{new}(S_i)=\alpha\cdot P(S_i)\cdot\prod_jm_{j\rightarrow i}\) \(P_{new}(F_k)=\alpha\cdot P(F_k)\cdot\prod_jm_{j\rightarrow k}\); where \(\alpha\) is the normalization factor and \(I(\cdot)\) is the indicator function. Iterate until convergence (usually 3 - 5 iterations).

[0251] For the high-probability fault sources, construct three competing hypotheses: H1: All anomalies are due to sensor faults; H2: All anomalies are due to power meter faults; H3: Some anomalies are due to sensor faults and some are due to power meter faults;

[0252] Calculate the posterior probability of the hypothesis \(P(H_i|E)=P(E|H_i)\cdot P(H_i) / \sum_jP(E|H_j)\cdot P(H_j)\); where \(P(E|H_i)\) is the likelihood and \(P(H_i)\) is the prior probability (set to 0.35, 0.35, and 0.30 respectively). Select the hypothesis with the maximum posterior probability to generate the candidate fault source set C. In actual measurement, the fault sources with a probability greater than 0.25 can be included in the candidate set.

[0253] Design a decomposition model to decompose the original signal \(x(t)\), \(x(t)=x_{baseline}(t)+x_{sensor}(t)+x_{meter}(t)+\varepsilon(t)\); where \(x_{baseline}\) is the baseline signal, \(x_{sensor}\) is the sensor fault component, \(x_{meter}\) is the power meter fault component, and \(\varepsilon\) is the residual.

[0254] Construct a dual-mode adaptive basis function library. The sensor fault basis functions include step functions, ramp functions, exponential functions, etc., which are used to represent characteristics such as sensor drift and slow response. The power meter fault basis functions are characteristic functions designed based on the physical model of the power meter, which are used to represent characteristics such as circuit faults and component aging. Library size: 25 sensor fault basis functions and 32 power meter fault basis functions.

[0255] Construct the optimization objective function min ||x - x_baseline - D_s·α_s - D_m·α_m|| 2 + λ 1 ||α_s|| 1 + λ 2 ||α_m|| 1 + λ 3 ||Φ(α_m)|| 2 ; where D_s and D_m are the feature dictionary matrices of sensor faults and meter faults respectively, α_s and α_m are the corresponding sparse coefficient vectors, and Φ(·) is the physical consistency constraint function. Regularization parameter settings: λ 1 = 0.35, λ 2 = 0.25, λ 3 = 0.4. Solve using the accelerated proximal gradient descent algorithm, set the maximum number of iterations to 200, and the convergence threshold to 10^-4.

[0256] Calculate the energy proportion of each component E_sensor = ||D_s·α_s|| 2 / ||x - x_baseline|| 2 ; E_meter = ||D_m·α_m|| 2 / ||x - x_baseline|| 2 ; In actual measurement, for the zero - drift fault of the current sensor, the energy proportion of the sensor component reaches 87.3%; for the fault of the meter metering chip, the energy proportion of the meter component reaches 82.1%.

[0257] Remove the sensor fault component, reconstruct the pure meter state signal x_pure(t) = x_baseline(t) + x_meter(t); use an adaptive threshold (β = 0.2) to control the separation degree: when E_sensor / (E_sensor + E_meter) > 0.8, it is considered that the abnormality mainly comes from the sensor fault; when this ratio < 0.2, it is considered that the abnormality mainly comes from the meter fault.

[0258] Establish a set of physical constraint models for the meter, including electrical models: power balance constraint S 2 = P 2 + Q 2 ; thermal model: temperature rise ΔT = k·P·t / C; mechanical model: vibration - current relationship V_vib = f(I); a total of 9 core constraint equations are constructed.

[0259] Substitute the pure electric energy meter status signal into the physical constraint equations to calculate the constraint satisfaction degree \(C_i = \exp(-|r_i| / \tau_i)\); where \(r_i\) is the residual of the \(i\)-th constraint equation, and \(\tau_i\) is the tolerance parameter (the value range is 0.05 - 0.15).

[0260] The comprehensive physical consistency score \(C_{physics}=\sum_i w_i\cdot C_i / \sum_i w_i\); where \(w_i\) is the constraint importance weight. In actual measurement, the \(C_{physics}\) of normal signals is usually greater than 0.85, and the \(C_{physics}\) of abnormal signals drops to 0.35 - 0.75.

[0261] Adjust the constraint parameters and threshold according to the current working condition \(\tau_{i,j}=\tau_{i,base}\cdot(1 + \delta_{i,j})\); where \(\tau_{i,base}\) is the reference tolerance, and \(\delta_{i,j}\) is the adjustment coefficient of working condition \(j\) for constraint \(i\). For example, the adjustment coefficient of the power balance constraint under fluctuating load conditions is 0.3.

[0262] Verify the physical rationality of the fault characteristics and calculate the matching degree with the typical fault mode \(M_k=\sum_i w_i\cdot sim(f_i, p_{k,i}) / \sum_i w_i\); where \(sim(\cdot)\) is the similarity function, \(f_i\) is the observed feature, and \(p_{k,i}\) is the typical feature of fault mode \(k\).

[0263] Fuse the physical consistency score, the results of the working condition adaptive constraint test, and the physical rationality score of the fault characteristics to generate the final verification result \(V_{result}=\beta\) 1 \(\cdot C_{physics}+\beta\) 2 \(\cdot C_{adaptive}+\beta\) 3 \(\cdot M_{best}\); where \(\beta\) 1 \( = 0.4\), \(\beta\) 2 \( = 0.35\), \(\beta\) 3 \( = 0.25\), \(M_{best}\) is the matching degree of the best matching fault mode. When \(V_{result}\) is lower than the threshold \(T_V\) (set to 0.65), reject the current fault hypothesis; when \(V_{result}\) is higher than 0.85, the fault judgment has a high confidence level.

[0264] Based on the comprehensive sensor health assessment results, the state feature vector of the electricity meter, and the fault verification results, a dynamic decision tree structure is used to achieve fault separation and identification: According to the comparison between the sensor health index and the threshold \(T_H\) (set to 70), the sensor state is initially judged; according to the consistency between the state feature vector of the electricity meter and the model prediction, the probability of electricity meter failure is evaluated; considering the component energy distribution and the physical verification results, the final fault type is determined; calculate the fault confidence level: Conf = α·P(F) + (1 - α)·V_result, where α = 0.6; finally, the fault type determination and confidence level are output.

[0265] Based on the sensor health index and calibration data, a measurement error model \(\sigma_i=\sigma_{i,base}\cdot(1 + \gamma\cdot(100 - Health_i) / 100)\) is established; where, \(\sigma_{i,base}\) is the reference uncertainty, and \(\gamma\) is the sensitivity coefficient (taking a value of 2.5). In actual measurement, for the voltage sensor, \(\sigma_{base}\) is 0.2%, and when the health index is 60, the measurement uncertainty rises to 0.5%.

[0266] The Monte Carlo method is used to evaluate the uncertainty propagation. Add random perturbations that conform to the measurement uncertainty distribution to the input data, perform 1000 simulations, and calculate the statistical distribution characteristics of the model output.

[0267] The inference uncertainty index \(U_{model}=\sigma_{output} / \mu_{output}\); in actual measurement, under normal working conditions, \(U_{model}\) is about 0.12, and it rises to 0.28 when the sensor is abnormal.

[0268] Analyze the influence of environmental temperature, humidity, and electromagnetic interference on the diagnostic results. Establish an environmental impact sensitivity model \(S_{env}=[dF / dT, dF / dH, dF / dE]\); where, \(dF / dT\) is the sensitivity of the diagnostic result to temperature. The measured data shows that for every 10°C change in temperature, the average change in the diagnostic accuracy rate is 3.5 percentage points. \(d\) is the partial derivative symbol.

[0269] Fuse each uncertainty source to construct a comprehensive evaluation framework \(U_{total}=\sqrt{w 1 \cdot U_{sensor} 2 + w 2 \cdot U_{model} 2 + w 3 \cdot U_{env} 2 )}; where, the weight coefficients \(w 1 = 0.5\), \(w 2 = 0.3\), \(w 3 = 0.2\).

[0270] Calculate the confidence interval CI of the fault diagnosis result: CI = [F - k·U_total, F + k·U_total]; where k is the coverage factor (the value is 1.96, corresponding to a 95% confidence level), F is the fault diagnosis result, and U_total is the comprehensive uncertainty.

[0271] The risk level is divided into four levels: Low risk: U_total < 0.15, confidence level > 0.90; Medium - low risk: 0.15 ≤ U_total < 0.25, 0.80 < confidence level ≤ 0.90; Medium - high risk: 0.25 ≤ U_total < 0.35, 0.70 < confidence level ≤ 0.80; High risk: U_total ≥ 0.35, confidence level ≤ 0.70;

[0272] In the measured cases, the comprehensive uncertainty of the zero - drift fault of the current sensor is 0.17, corresponding to medium - low risk; while the comprehensive uncertainty of the metering chip fault of the electric energy meter is 0.23, also corresponding to medium - low risk.

[0273] Based on the fault type determination, fault confidence level, and comprehensive uncertainty evaluation results, generate a fault evaluation report, which contains the following key information. Fault type: Clearly distinguish between sensor faults and electric energy meter faults; Fault location: Specific to the component level; Fault severity: Divided into four levels (slight, medium, severe, critical); Confidence interval: 95% confidence interval of the fault quantitative parameter; Reliability score: 0 - 100, reflecting the reliability degree of the diagnosis result; Maintenance suggestion: Disposal suggestions based on the fault type and reliability score.

[0274] For example, the report content of a typical case is: Fault type: Zero - drift of the A - phase current sensor (sensor fault); Fault location: A - phase current sensor; Fault severity: Medium (offset value 2.7%, exceeding the allowable range of 1.5%); Confidence interval: Offset value 2.7% ± 0.46%; Reliability score: 87 (medium - high reliability); Maintenance suggestion: Calibrate or replace the A - phase current sensor during the next routine maintenance.

[0275] Select 100 intelligent electric energy meters in the intelligent distribution network of a provincial power company for experiments, including three - phase intelligent electric energy meters of different brands, models, and service years. The data acquisition period is 6 months, the sampling frequency is 10 kHz, the acquisition window is 30 s / time, and the interval is 10 minutes. A comprehensive data set containing 20,736,000 sample points is constructed, among which there are 2,500 groups of labeled fault samples, covering 12 typical fault modes.

[0276] Evaluate the performance of separating and identifying sensor faults and electricity meter faults. Select 1,200 groups of samples from the dataset (including 600 groups of sensor faults, 400 groups of electricity meter faults, and 200 groups of mixed faults) for testing. The main performance indicators are shown in Table 1:

[0277] Table 1 Fault Separation and Identification Performance

[0278] Fault type Number of samples Number of correct identifications Recognition rate (%) Average confidence level Sensor fault 600 558 93.0 0.86 Electric energy meter fault 400 356 89.0 0.82 Hybrid fault 200 165 82.5 0.78 Total 1,200 1,079 89.9 0.83

[0279] Compared with the traditional method, this method has significant advantages in separating and identifying the fault sources, as shown in Table 2:

[0280] Table 2 Comparison of Fault Source Identification Performance of Different Methods

[0281] Method Overall recognition rate (%) False alarm rate (%) Missed alarm rate (%) Threshold judgment method 61.7 42.3 15.2 SVM classification method 74.3 23.6 11.8 Bayesian network method 82.1 18.2 9.5 This method 89.9 8.6 7.2

[0282] To verify the stability of this method under different working conditions, test the fault identification performance under four typical working conditions. The results are shown in Table 3:

[0283] Table 3 Fault Identification Performance under Different Working Conditions

[0284] Operating condition type Recognition rate (%) Sensor fault recognition rate (%) Electric energy meter fault recognition rate (%) Light load 90.6 94.2 87.3 Medium load 91.8 93.6 89.5 Heavy load 88.5 91.2 84.7 Fluctuating load 86.2 89.5 82.3

[0285] The results show that this method performs best under light load and medium load conditions, and its performance slightly decreases under fluctuating load conditions, but generally maintains a high recognition accuracy, verifying the effectiveness of the working condition adaptive mechanism.

[0286] Verify the uncertainty assessment of the fault diagnosis results. Randomly select 300 groups of fault samples and compare the relationship between the predicted confidence interval and the actual error. The results show that 92.7% of the actual errors fall within the predicted 95% confidence interval, indicating the effectiveness of the uncertainty quantification method. The average uncertainty indicators for various types of faults are shown in Table 4:

[0287] Table 4 Uncertainty Indicators for Different Fault Types

[0288] Fault type Sensor uncertainty Model uncertainty Environmental uncertainty Comprehensive uncertainty Sensor zero drift 0.23 0.14 0.11 0.17 Sensor sensitivity degradation 0.25 0.18 0.13 0.19 Electric energy meter metering chip fault 0.18 0.28 0.15 0.23 Electric energy meter power supply circuit fault 0.16 0.25 0.17 0.21

[0289] Conduct an actual application test of this method in 10 distribution substations of a provincial power company for 3 months, covering 1,200 intelligent electricity meters. During the test period, the system issued 124 fault warnings in total, among which 113 accurate warnings were verified on-site, and the accuracy rate was 91.1%. Compared with the original system, the false alarm rate decreased from 42.3% to 8.9%, significantly reducing the ineffective dispatch of maintenance personnel.

[0290] The application effects are mainly reflected in the following aspects: improving the accuracy of fault prediction: from the original 62.5% to 91.1%; reducing maintenance costs: reducing by about 37.2%, saving about 215,000 yuan in maintenance costs annually; extending the service life of equipment: by detecting and handling potential faults in advance, extending the average service life of electric energy meters by about 1.2 years; improving power supply reliability: reducing the power outage time caused by faults related to intelligent electric energy meters by about 42.6%.

[0291] The experimental results show that this method has achieved an accuracy of 89.9% in the separation and identification of fault sources, reduced the false alarm rate from 42.3% to 8.6%, has strong working condition adaptability and uncertainty quantification ability, and can provide reliable decision-making support for power grid operation and maintenance management. The successful application of this method has important practical value for improving the safe and reliable operation of smart grids and reducing maintenance costs.

[0292] By constructing a two-layer causal structure of upper-layer sensor faults and lower-layer electricity meter faults and connecting the two layers through observation relationships, the effective separation and identification of sensor faults and electricity meter faults are achieved. This structure enables the system to consider both the health status of the sensors themselves and the status of the measured objects simultaneously, so that when facing fault signals, the true fault source can be accurately traced. Through the sensor group identification based on physical constraints and the constrained projection anomaly localization algorithm, the system can make full use of the physical correlation between sensors for mutual inspection and complementarity. This method not only relies on statistical correlation, but also introduces physical knowledge such as energy conservation and thermodynamics laws, making the sensor anomaly detection more reliable. By decomposing the mixed fault signal into sensor fault components and electricity meter fault components and combining with a dual-mode adaptive basis function library and physical constraint sparse decomposition, the system can distinguish which anomalies in the signal are caused by sensor faults and which are caused by real electricity meter faults. This separation ability greatly improves the accuracy of fault diagnosis. By constructing a feature-operating condition sensitivity matrix and an adaptive weight function, the importance of features is dynamically adjusted according to the operating conditions. Since the fault characteristics of the electricity meter vary greatly under different operating conditions, static feature selection often fails to adapt to this change. Through multi-operating condition constraint adaptive adjustment and verification of fault physical characteristics, the fault separation results are combined with the physical working principle of the electricity meter for secondary verification, effectively avoiding incorrect judgments that violate physical laws. Through the basic probability assignment function based on the physical characteristics of sensors and the conflict evidence optimization combination rule, the scientific fusion of multi-source uncertain information is achieved. The traditional Dempster-Shafer evidence theory often produces counterintuitive results when facing highly conflicting evidence, while the improved framework significantly improves the rationality of evidence fusion by introducing the weights of sensor physical characteristics and the conflict evidence processing mechanism. By quantifying the information redundancy and complementarity between different sensors, the system can utilize multi-sensor information more intelligently. Different from simple correlation analysis, the mutual information matrix considers the non-linear information dependence relationship between sensors and can capture more complex data association patterns. Through the top-down and bottom-up two-way message passing mechanism, the full information interaction between the sensor layer and the electricity meter layer is achieved. The traditional Bayesian network usually has only one-way reasoning and is difficult to handle both the process of "sensor faults affecting the observation results" and the process of "inferring the fault source based on the observation" simultaneously. The two-way propagation mechanism enables the system to reason about the fault source more comprehensively and coordinately, considering various possible causal paths. By weighted combination of the temporal correlation distance, mutual information distance and physical constraint deviation distance, a distance metric that can better reflect the true association between sensors is formed. The traditional distance metric is mainly based on statistical similarity, ignoring the physical association between sensors, resulting in clustering results lacking physical meaning. The distance metric enhanced by physical knowledge makes sensor clustering more interpretable and practical, and the constructed sensor groups conform to the physical working principle of the electricity meter.

[0293] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for predicting faults of smart electric energy meters based on multimodal sensor fusion, characterized in that: include: Collect and preprocess multi-modal sensor raw data to obtain standard data; Based on standard data, a sensor health status model is established and the sensor working status is monitored to obtain the sensor health assessment results; Perform feature extraction and pattern recognition on standard data to generate the state feature vector of the electric energy meter; Combining the sensor health assessment results and the energy meter state feature vector, the two-layer causal graph reasoning module is used to perform two-layer fault source separation and identification, and output the fault type and confidence level. Quantify multi-source uncertainty and evaluate reliability based on fault type and confidence level, and generate fault assessment report; The process of performing double-layer fault source separation and identification includes: Read the observation evidence set as the input of the two-layer causal graph reasoning module to obtain the probability distribution of the fault source; Among them, the observation evidence set includes sensor health assessment results, energy meter state feature vectors and current measurement data; The fault sources with probability higher than a threshold in the fault source probability distribution are taken as candidate fault sources; Construct three competitive hypotheses: sensor fault, meter fault and mixed fault, apply Bayesian model selection method to evaluate the posterior probability of each hypothesis, and output the optimal fault hypothesis and candidate fault source set; For each candidate fault in the candidate fault source set, the original signal is decomposed into a sensor fault component and an electric energy meter fault component, and a pure electric energy meter state signal is reconstructed; Combined with the physical working principle of the electric energy meter, verify whether the state signal of the pure electric energy meter conforms to the physical consistency law, and output the fault verification result; Combining the sensor health assessment results, the energy meter state feature vector and the fault verification results, a dynamic decision tree is used to perform hierarchical identification of sensor faults and energy meter faults to obtain the final fault type judgment and fault confidence.

2. The method according to claim 1, characterized in that The two-layer causal graph reasoning module includes: The upper layer module is used to represent the causal relationship between sensor failure and measurement anomaly. The lower module is used to represent the causal relationship between the power meter failure and the abnormal physical parameters. Interlayer connections are used to represent the observed relationship between measurement anomalies and physical parameter anomalies. When working, it infers measurement anomalies from possible sensor failures from top to bottom, and infers potential faults from observed anomalies from bottom to top. It propagates probability information in the two-layer causal graph through the message passing mechanism and outputs the probability distribution of fault sources.

3. The method according to claim 1, characterized in that The steps of reconstructing the pure electric energy meter status signal include: Construct a collaborative signal decomposition model for distinguishing different manifestations of sensor fault signatures and electric energy meter fault signatures; Construct characteristic basis function libraries for sensor faults and electric energy meter faults respectively, forming a dual-mode adaptive basis function library; Based on the collaborative decomposition model and dual-mode adaptive basis function library, the original signal is decomposed into sensor fault component and electric energy meter fault component; Calculate the contribution ratio of the sensor fault component and the electric energy meter fault component to the original signal, and generate an energy distribution diagram for evaluating the relative importance of the two types of fault components; According to the weight analysis of the component energy distribution diagram, the sensor fault component is eliminated, the separation threshold is adjusted, and the pure electricity meter status signal is reconstructed in combination with the normal signal baseline.

4. The method according to claim 3, characterized in that The steps of verifying whether the pure electric energy meter status signal conforms to the physical consistency law and outputting the fault verification result include: Based on the working principle of the electric energy meter, a set of physical constraint models of the electric energy meter including electrical model, thermal model and mechanical model is established to form a set of physical constraint equations that can be quantified and verified; Substitute the pure electric energy meter state signal into the physical constraint equation group, calculate the physical constraint satisfaction, and generate a physical consistency score; Adjust the parameters and thresholds in the physical constraint equation group according to the current working condition type, so that the constraint test adapts to the current working condition, and output the working condition adaptive constraint test results; For the fault features identified in the pure electric energy meter status signal, check whether they conform to the physical manifestation rules of the corresponding fault type and generate a physical rationality score for the fault features; The above outputs are integrated to generate fault verification results, including verification pass status and credibility assessment.

5. The method according to claim 1, characterized in that The process of establishing a sensor health status model and monitoring the sensor working status to obtain the sensor health assessment results includes: Based on the historical normal operation data in the standard data, a normal response characteristic model of the sensor is established; Compare the current sensor real-time data with the sensor normal response characteristic model, detect sensor characteristic drift, and generate sensor characteristic drift indicators; Perform frequency domain analysis on standard data, extract sensor frequency domain features, and construct sensor frequency domain feature vectors; Perform multi-sensor cross-validation to analyze the physical correlation between different sensor data and generate sensor credibility scores; The above outputs are integrated to build a sensor health status model and output sensor health assessment results, including sensor health index and abnormal type identification results.

6. The method according to claim 5, characterized in that The process of generating a sensor confidence score includes: Analyze the physical dependencies and associations between sensor data in the standard data to obtain a physical constraint relationship matrix; Based on the physical constraint relationship matrix, the consistency check is performed on the standard data, and the residual value of each physical constraint equation is calculated to form the sensor consistency residual vector; Decompose the sensor consistency residual vector into individual sensor contributions and output the sensor anomaly contribution metric; Identify subsets of sensors that can authenticate each other, build redundant sensor groups, and form a sensor group mapping table; For each group in the sensor group mapping table, the complementary information of the sensors in the group is extracted to construct a sensor mutual information matrix; The sensor anomaly contribution metric and sensor mutual information matrix are integrated, and the sensor credibility score is calculated and output through the credibility scoring function.

7. The method according to claim 5, characterized in that The process of forming the sensor group mapping table is as follows: Calculate the physical correlation distance matrix between all sensor pairs in the standard data; Determine the initial cluster center based on the physical constraint relationship matrix, perform principal component analysis on the physical constraint relationship matrix, select the sensor with the largest contribution of each principal component as the candidate center, and obtain the initial cluster center set; Using the initial cluster center set as the starting cluster center, each sensor is assigned to the nearest center, and a new center that minimizes the total distance is reselected in each group to obtain the initial division of the sensor group; Calculate the internal physical redundancy for each sensor group, adjust the critical group, and output the optimized sensor group; Each sensor group is matched with a physical constraint relationship matrix, and it is confirmed that the sensors in the group constitute a meaningful physical association unit, so as to obtain a sensor group mapping table with physical meaning.

8. The method according to claim 1, characterized in that The steps of generating the state feature vector of the electric energy meter include: Perform multi-scale wavelet packet decomposition on standard data, extract time domain, frequency domain and time-frequency joint features, and construct an initial multi-scale feature set; Using the electrical characteristics and thermal model of the electric energy meter, the initial multi-scale feature set is enhanced with physical knowledge to generate a physical enhanced feature set. Apply feature optimization algorithm and dimensionality reduction processing to the physical enhancement feature set to form an optimized electric energy meter feature vector; According to the working condition information of the electric energy meter, the optimized electric energy meter feature vector is dynamically weighted and the working condition adaptive feature vector is output; The adaptive characteristic vector of the working condition is identified to generate the characteristic vector of the electric energy meter state and the preliminary fault type determination result.

Images