An electric energy metering box fault monitoring method and system based on the Internet of Things

Abstract

The present application relates to the technical field of electrical fault monitoring, and proposes a kind of electric energy metering box fault monitoring method and system based on Internet of Things, including identifying the physical connection and logical dependency of each component in the box, configuring multi-dimensional self-check parameter index and monitoring weight;Different sampling frequency data is mapped to a unified time axis using a data modal alignment algorithm, generating a feature vector;Based on logical dependency, establish fault evolution topology atlas, and inject disturbance coefficient deviating from topological benchmark into feature vector;Dimension reduction is performed on the feature sequence, and trend curve and transient disturbance features are extracted using variational modal decomposition;Based on monitoring period factor and component failure correlation parameter, construct trigger time monitoring matrix, lock component abnormal conflict domain, calculate euclidean space deviation to locate target abnormal component block;Extract real-time state data, combine internal state characteristics to match health degree grading rules, and output diagnosis results.The present application significantly improves the accuracy, early warning timeliness and precision of metering box fault identification.

Description

Technical Field

[0001] This invention relates to the field of electrical fault monitoring technology, and in particular to a method and system for monitoring faults in electricity metering boxes based on the Internet of Things. Background Technology

[0002] Existing fault monitoring technologies for power metering boxes mostly focus on threshold comparison of single electrical parameters or state diagnosis of isolated components, lacking a holistic consideration of the physical connections and logical dependencies between multiple components within an IoT terminal. Under complex operating conditions, the fault propagation effects and coupling relationships between power supply modules, communication modules, sensor groups, and the main control unit often exhibit nonlinear evolution characteristics, making it difficult for traditional monitoring methods to capture the chain response mechanisms triggered by faults between components. Furthermore, the heterogeneity of data acquisition from various components leads to phase misalignment of multi-source information over time, making it impossible for existing methods to effectively align operating data from different sampling frequencies to construct a unified feature analysis benchmark. More importantly, traditional fault diagnosis often remains at the level of static threshold judgment, failing to establish a topological mapping relationship describing the dynamic evolution of equipment health status. This results in difficulty in identifying early, weak anomalies, and during fault location, it cannot distinguish between single component failure and systemic anomalies caused by multi-component interaction conflicts, ultimately leading to technical defects such as delayed early warning, location errors, and high false alarm and false negative rates. Summary of the Invention

[0003] To address the aforementioned shortcomings, the present invention aims to propose a fault monitoring method and system for electricity metering boxes based on the Internet of Things (IoT). The goal is to achieve a shift from single-parameter monitoring to multi-dimensional collaborative diagnosis by constructing a mapping of physical connections and logical dependencies between components and integrating heterogeneous data modal alignment and fault evolution topology analysis mechanisms.

[0004] To achieve this objective, the present invention adopts the following technical solution: A method for fault monitoring of electricity metering boxes based on the Internet of Things, comprising: Identify the physical connections and logical dependencies between the components of the Internet of Things (IoT) terminal in the power metering box, and configure the multi-dimensional self-testing parameter indicators and their monitoring weight values ​​corresponding to the components. Collect heterogeneous operating data of the components, and use a heterogeneous data modal alignment algorithm to map data with different sampling frequencies to a unified time axis to generate a state feature vector sequence sorted by weight values; A fault evolution topology map of the device itself is established based on logical dependencies, and the perturbation coefficient of the component operation data deviating from the topology baseline is injected into the corresponding state feature vector. An adaptive sliding window is used to reduce the dimensionality of the state feature vector sequence, and variational mode decomposition is used to extract the trend curves describing the performance evolution of the components and the transient perturbation features describing the abnormal interaction of the components. Based on the trigger time monitoring matrix, which includes monitoring cycle factors and component failure correlation parameters, the abnormal conflict domain of the component is locked, and the target abnormal component block is located by calculating the Euclidean space deviation between the state feature vector and the topology map. Extract real-time status data of the target abnormal component block, combine internal status characteristics with equipment health grading rules, output diagnostic results and failure probability, and trigger early warning.

[0005] Preferably, the component includes a power supply module, a communication module, a sensor group, and a main control unit; Identifying the physical connections and logical dependencies between the various components of the IoT terminal within the electricity metering box includes: Obtain the physical electrical connection paths between the power supply module, communication module, sensor group, and main control unit; Based on the functional collaboration logic between components, a star-shaped dependency model with the main control unit as the core is established, defining the support dependency relationship of the power supply module to each active component, as well as the data flow dependency relationship between the sensor group and the communication module to the main control unit. Extract the spatiotemporal association rules of the operating data of each component under normal working conditions, and define the interaction behaviors that violate the logical sequence or electrical topology constraints as logical dependency conflict items.

[0006] Preferably, the heterogeneous data modality alignment algorithm is used to map data with different sampling frequencies to a unified time axis, generating a state feature vector sequence sorted by weight values, including: The component index with the highest sampling frequency in the heterogeneous operation data is determined as the time reference axis. The power supply index and communication link index of the low-frequency sampling are supplemented by the linear interpolation algorithm to achieve phase synchronization and alignment of different modal data at discrete time points. By using the deviation standardization method and combining the historical extreme values ​​of the parameters of each component, the heterogeneous operating data covering electrical quantities, electromagnetic quantities, logic percentages and discrete error codes are mapped to a unified numerical range, generating a standardized heterogeneous feature set. Standardized heterogeneous features aligned at the same reference time point are extracted to construct a multidimensional initial feature vector. The dimensions of the initial feature vector are rearranged and weighted according to the monitoring weight value to generate a sequence of state feature vectors describing the real-time health status of the device.

[0007] Preferably, establishing a fault evolution topology map of the device itself based on logical dependencies includes: Based on the physical electrical performance connection path and the star-shaped dependency model, each component is mapped to a physical node in the graph, and a directed edge representing the fault propagation law is established between each physical node according to the data flow dependency relationship between the components. Extract the collaborative constraint rules among the standardized heterogeneous feature sets corresponding to each physical node under normal operating conditions, including calculating the linear correlation between power supply voltage fluctuations and main control load changes, as well as the timing consistency between sensor sampling frequency and communication retransmission frequency, and generate a topology function describing the hardware health benchmark. The real-time operating data of each physical node is compared with the topology function. Based on the statistical indicators of voltage deviation, signal strength ratio and watchdog reset frequency, the deviation of the real-time characteristic value of the physical node from the health benchmark is calculated. Based on the degree of deviation, state offset vectors that conform to the failure propagation effect are selected, and the state offset vectors are mapped to the corresponding directed edges to form a fault evolution topology map of the device itself. The selection conditions include the degree of deviation continuously exceeding a first preset threshold and reaching a first preset duration.

[0008] Preferably, injecting the perturbation coefficients of component operating data deviating from the topology baseline into the corresponding state feature vector includes: The standardized heterogeneous feature set collected in real time is input into the fault evolution topology map, the quantization residual between the measured feature value and the health benchmark is extracted, and the degree of difference between the quantization residual and the physical node triggering condition and hardware response is calculated. The quantized residual is compared with a preset fault mode library to verify whether the difference has logical consistency on the failure path of the corresponding physical node, so as to determine the effective disturbance item. According to the preset mapping rules, each effective perturbation term is converted into a normalized perturbation coefficient using an exponential function, and the perturbation coefficient is used as the corresponding dimension of the weight gain injection state feature vector. Feature enhancement is achieved by increasing the geometric distance between the abnormal dimension and the normal dimension in the multidimensional Euclidean space.

[0009] Preferably, the dimensionality reduction process yields the principal feature components that reflect the operational fluctuations of the component. Variational mode decomposition is used to extract trend curves describing component hardware fatigue and transient disturbance characteristics of component software conflicts, including: The modal decomposition layers and penalty factors are preset, and the principal feature components are iteratively decomposed to obtain eigenmode function components with different center frequencies. Low-frequency terms with center frequencies below a preset frequency threshold are identified from the intrinsic mode function components and the signals are reconstructed to obtain trend curves describing the slow evolution of the electrical performance parameters of each component over time, which are used to quantify the long-term degradation of the components. High-frequency pulse terms are extracted from the intrinsic mode function components. By calculating the correlation between the high-frequency energy distribution and the envelope spectrum, transient disturbance characteristics describing data flow abrupt changes or logical collisions of each component are identified and used to quantify the instantaneous conflict intensity of the components.

[0010] Preferably, based on a trigger time monitoring matrix including monitoring cycle factors and component failure correlation parameters, the component anomaly conflict domain is locked as follows: The component failure correlation parameters Satisfying the relation: ; in, Representation Component With components Component failure correlation parameters between and Representing components respectively With components In the Feature response vectors under each historical fault mode This indicates the total number of preset historical fault modes. Representation Component With components Topological logical distance in the star dependency model This represents the preset topology attenuation coefficient; Extract the slope and rate of change of the trend curve and the pulse frequency of the transient disturbance characteristics, and combine them with the service load fluctuation of the power metering box to calculate the dynamic monitoring period factor. Satisfying the relation: ; in, This indicates the rate of change of the slope of the trend curve within the current sliding window. Indicates the duration of observation The number of transient disturbance pulses identified internally. Indicates the current service load power. This indicates the rated maximum power consumption of the electricity metering box. , and These represent the preset adjustment weights used to balance the effects of trends, disturbances, and loads; Using each component as a row vector and the discrete trigger time as a column vector, the failure correlation parameter and the monitoring period factor are injected to construct a trigger time monitoring matrix. The trigger time monitoring matrix elements Satisfying the relation: ; in, Indicates the monitoring matrix at the th The first component, the first The trigger strength value at each moment, This represents the activation function. Representation Component At any moment The real-time disturbance amplitude, Representation and Components A set of physically connected adjacent components. Represents the sampling decision indication function; Identify the cluster of elements in the trigger time monitoring matrix whose trigger intensity value exceeds a preset threshold, determine the set of components that have abnormal state coupling, and lock them as the component abnormal conflict domain.

[0011] Preferably, the component block for locating target anomalies by calculating the Euclidean spatial deviation between the state feature vector and the topological map includes: Within the component abnormal conflict domain, the state feature vector corresponding to each component is retrieved, and the corresponding health benchmark vector is extracted from the fault evolution topology map. The two are then mapped to a unified multi-dimensional feature coordinate space. For each dimension of the multidimensional feature coordinate space, the absolute value of the difference between the real-time feature value and the baseline feature value is calculated, and then weighted and summed in combination with the monitoring weight value to obtain the deviation value describing the degree of deviation of each component from the normal operating condition. A preset deviation threshold is set, and the deviation values ​​of each component in the conflict domain are sorted and clustered. The smallest set of components with deviation values ​​exceeding the deviation threshold and having failure correlation is selected and located as the target abnormal component block.

[0012] Preferably, combining internal state characteristics with equipment health grading rules, outputting diagnostic results and failure probabilities, and triggering early warnings includes: Extract the real-time status data of the target abnormal component block, including operating current and voltage fluctuations, node communication latency, instruction execution success rate, and ambient temperature. The real-time status data is converted into feature components describing the current operating condition of the component and compared with the health benchmark in the fault evolution topology map. By identifying the proportion of the real-time status data exceeding the preset deviation threshold and the frequency of interaction abnormal features occurring within the sliding window, the corresponding health level is obtained. Based on the health level obtained from the matching, and combined with the position of the target abnormal component block in the topology map, the real-time failure probability of each component is calculated, and diagnostic results including the location of the failure, the degree of performance degradation, and the type of logical anomaly are output synchronously. Based on the severity of the diagnostic results, a warning signal with a differentiated alarm level is pushed to the monitoring terminal.

[0013] An Internet of Things (IoT) based power metering box fault monitoring system includes: The parameter configuration module is used to identify the physical connections and logical dependencies between the components of the Internet of Things terminal in the power metering box, and to configure the multi-dimensional self-test parameter indicators and their monitoring weight values ​​corresponding to the components. The data alignment module is used to collect heterogeneous operating data of the components, and use a heterogeneous data modal alignment algorithm to map data with different sampling frequencies to a unified time axis, generating a state feature vector sequence sorted by weight value; The topology building module is used to build a fault evolution topology map of the device itself based on logical dependencies, and to inject the perturbation coefficient of the component operation data deviating from the topology baseline into the corresponding state feature vector. The feature extraction module is used to perform dimensionality reduction on the state feature vector sequence using an adaptive sliding window, and to extract trend curves describing the performance evolution of components and transient perturbation features describing abnormal component interactions using variational mode decomposition. The conflict localization module is used to locate the abnormal conflict domain of the component based on the trigger time monitoring matrix, which includes the monitoring cycle factor and the correlation parameters of component failure. The module locates the target abnormal component block by calculating the Euclidean space deviation between the state feature vector and the topology map. The graded early warning module is used to extract real-time status data of the target abnormal component block, combine internal status characteristics with equipment health grading rules, output diagnostic results and failure probability, and trigger early warning.

[0014] One of the above technical solutions has the following advantages or beneficial effects: This invention establishes a benchmark framework reflecting the inherent constraints of energy supply and data interaction between components by identifying the physical connections and logical dependencies among the components of an IoT terminal within an energy metering box and configuring multi-dimensional self-testing parameters and their monitoring weights. This addresses the blind spot in fault propagation path identification caused by the lack of topological correlation considerations in traditional monitoring. By collecting heterogeneous operating data from each component, a heterogeneous data modal alignment algorithm is used to map data with different sampling frequencies to a unified time axis and generate a state feature vector sequence sorted by weight values. This avoids the heterogeneity barriers of multi-source data in phase and dimension, achieving a spatiotemporal consistent representation of cross-component operating states. Furthermore, by establishing a fault evolution topology map of the equipment itself based on logical dependencies and injecting the perturbation coefficients of component operating data deviating from the topological benchmark into the corresponding state feature vectors, the electrical connection characteristics at the physical level and the degree of abnormal deviation at the data level are organically integrated to construct a dynamic mapping model that reflects the propagation and diffusion laws of faults between components. Subsequently, an adaptive sliding window is used to process the state feature vector sequence... Dimensionality reduction is performed, and variational mode decomposition is used to extract trend curves describing component performance evolution and transient disturbance features describing component interaction anomalies. This separates the slowly accumulating degradation trend and sudden interaction conflict signals during equipment operation, overcoming the limitations of traditional methods in extracting early weak anomalies and transient impact features. By locking the component anomaly conflict domain based on a trigger time monitoring matrix including monitoring cycle factors and component failure correlation parameters, the target abnormal component block is located by calculating the Euclidean space deviation between the state feature vector and the topology map, achieving accurate tracing from macroscopic anomaly areas to microscopic fault root causes, avoiding the ambiguity of location under multi-component coupled faults. Finally, real-time status data of the target abnormal component block is extracted, and combined with internal status features to match equipment health grading rules, outputting diagnostic results and failure probabilities and triggering early warnings. This transforms abstract topology deviations into interpretable health levels and risk probabilities, enabling quantitative assessment and graded response of fault severity, and comprehensively improving the fault prediction capability, location accuracy, and operation and maintenance response efficiency of the power metering box under complex operating conditions. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 This is a flowchart of the IoT-based fault monitoring method for electricity metering boxes provided in an embodiment of the present invention; Figure 2This is a schematic diagram of the structure of the Internet of Things-based power metering box fault monitoring system provided in an embodiment of the present invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0018] In this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0019] A method for fault monitoring of electricity metering boxes based on the Internet of Things, such as Figure 1 As shown, a preferred embodiment of the present invention includes the following steps: S1: Identify the physical connections and logical dependencies between the components of the IoT terminal in the power metering box, and configure the multi-dimensional self-testing parameter indicators and their monitoring weight values ​​corresponding to the components. It should be noted that physical connection relationships refer to the actual electrical path topology between components, including the physical routing of power supply lines, the actual wiring topology of communication buses, and the physical connection paths of signal transmission media. This relationship is obtained by analyzing the circuit schematic and PCB layout file of the power metering box and is used to establish the physical basis for energy transfer and signal transmission between components. Logical dependency relationships refer to the collaborative constraints between components at the functional implementation level, including the order of data flow, the hierarchical relationship of function calls, and the timing requirements of state synchronization. This relationship is obtained by analyzing the system software architecture and task scheduling mechanism and is used to establish the logical basis for functional cooperation and information interaction between components. Multidimensional self-testing parameter indicators refer to... Physical or logical quantities used to quantify the operating status of components include the output voltage / current of the power supply module, the signal strength / bit error rate of the communication module, the sampling accuracy / response latency of the sensor group, and the CPU utilization / task execution status of the main control unit. These indicators are obtained through sensor data collection or system log recording and are used to construct a digital representation of the component status. The monitoring weight value refers to a quantitative coefficient assigned based on the importance of the component in the realization of system functions and the sensitivity of the failure to the overall system performance. This coefficient is determined by analyzing the criticality of component functions and the impact range of historical failures, and is used to differentiate the impact of different component anomalies on the overall health status of the system in subsequent analyses.

[0020] Understandably, by identifying the physical connections and logical dependencies between the components of the IoT terminal within the power metering box, a benchmark framework reflecting the inherent constraints of energy supply and data interaction among components is established. This step addresses the problem of traditional monitoring methods treating each component as an isolated entity, thus failing to capture fault propagation paths, by acquiring the underlying electrical connection mapping and the high-level logical control architecture. By configuring multi-dimensional self-testing parameters, a comprehensive monitoring dimension covering electrical performance, communication quality, sensing accuracy, and processing capabilities is established, enabling the system to possess multi-perspective state perception capabilities. By configuring monitoring weight values, differentiated monitoring benchmarks are constructed, giving key components undertaking core control functions a higher risk sensitivity coefficient. This allows for prioritizing state changes of high-risk components in subsequent feature extraction and anomaly localization, optimizing the allocation efficiency of monitoring resources, and providing a structured analytical foundation for heterogeneous data fusion and fault localization.

[0021] S2: Collect heterogeneous operating data of the components, and use the heterogeneous data modal alignment algorithm to map data with different sampling frequencies to a unified time axis to generate a state feature vector sequence sorted by weight value; It should be noted that heterogeneous operational data refers to a collection of monitoring data originating from different types of components, with different data formats, sampling frequencies, and physical dimensions. For example, analog voltage signals collected by sensor groups at a frequency of 100 Hz, discrete status codes recorded by the main control unit at a frequency of 10 Hz, and communication quality indicators statistically analyzed by the communication module at a frequency of 1 Hz. Heterogeneous data modal alignment algorithm refers to a data processing method used to avoid differences in time phase and numerical scale between multi-source data. It includes sub-processes such as time base unification, sampling frequency normalization, and numerical standardization. Its function is to enable data from different sources to have a unified spatiotemporal reference and comparable numerical scale. Unified time axis refers to a time coordinate system established based on the highest sampling frequency in the system, providing a discrete time scale standard as a common reference for multi-source data alignment. State feature vector refers to a multi-dimensional numerical vector composed of standardized operating parameters of each component arranged in a specific order at a specific time point, with each dimension corresponding to a self-check parameter indicator. Sorting refers to adjusting the dimensional order of the state feature vector according to preset monitoring weight values, so that the parameters of high-weight components are at the forefront of the vector.

[0022] Understandably, by collecting heterogeneous operational data from various components, raw information reflecting the multi-level operational status of the equipment can be obtained, solving the problem of the one-sidedness of information from a single data source. By using a heterogeneous data modal alignment algorithm to map data with different sampling frequencies to a unified time axis, and by determining a high-frequency time reference axis and using linear interpolation to complete low-frequency data, phase synchronization in the time dimension is achieved, avoiding false anomalies caused by sampling time misalignment. By generating a state feature vector sequence sorted by weight values, a spatiotemporally consistent equipment health characterization model is constructed, which transforms the operating parameters of different physical attributes into a numerical form that can be uniformly compared and calculated. Furthermore, the feature dimensions are prioritized according to the importance of the components, thereby focusing on the state changes of high-risk components in subsequent analysis and improving the sensitivity and response speed of fault detection.

[0023] S3: Establish a fault evolution topology map of the device itself based on logical dependencies, and inject the disturbance coefficient of the component operation data deviating from the topology baseline into the corresponding state feature vector; It should be noted that the fault evolution topology map refers to a knowledge model that describes the fault propagation path and evolution law between components in the form of a graph structure. Nodes represent physical components, directed edges represent the direction of fault propagation between components, and edge weights characterize the probability or intensity of fault propagation. This map establishes a structured carrier for fault propagation by mathematically representing physical connections and logical dependencies. The topology benchmark refers to the cooperative constraint relationship that the operating parameters of each component should satisfy under normal operating conditions, including the correlation law of electrical parameters and the synchronization requirements of time-series data. This benchmark is generated by statistically analyzing historical normal operating data and serves as a reference system for determining whether the real-time state is abnormal. The disturbance coefficient refers to a gain factor that quantitatively describes the degree to which the real-time operating state of a component deviates from the normal benchmark. This coefficient is obtained by calculating the degree of deviation between the real-time feature value and the topology benchmark and performing a nonlinear mapping, reflecting the amplification effect of the abnormal state on the overall health of the system. The injection operation refers to applying the disturbance coefficient as a weight multiplier to the corresponding dimension of the state feature vector, enhancing the distinguishability of abnormal features in multi-dimensional space through mathematical operations.

[0024] Understandably, by establishing a fault evolution topology map of the device itself based on logical dependencies, the physical connections and functional dependencies between components are transformed into a computable mathematical structure. By mapping physical connection paths and star-shaped dependency models into map nodes and directed edges, a structured carrier for fault propagation is established, solving the problem that traditional threshold monitoring cannot describe the fault chain propagation mechanism. By injecting the perturbation coefficient of the component's operating data deviating from the topology benchmark into the corresponding state feature vector, the local anomalies of the components are compared with the overall health benchmark of the system. By calculating and quantifying the residuals and verifying their logical consistency on the failure path, the degree of deviation is quantified and converted into nonlinear gain through exponential mapping. This widens the geometric distance between abnormal and normal states in the feature space, enhances the sensitivity of subsequent pattern recognition to weak anomalies, and realizes the transformation from static threshold judgment to dynamic correlation analysis.

[0025] S4: Adaptive sliding window is used to reduce the dimensionality of the state feature vector sequence, and variational mode decomposition is used to extract the trend curve describing the performance evolution of the component and the transient perturbation features describing the abnormal interaction of the component. It should be noted that adaptive sliding window refers to a time-series data segmentation method that can dynamically adjust its length based on data fluctuation characteristics. The window length can be adaptively increased or decreased according to data variance, information entropy, or rate of change. The window is extended when the data is stable to reduce computational overhead, and shortened when the data fluctuates to capture rapid changes. Principal features refer to low-dimensional feature representations extracted from high-dimensional state feature vectors using dimensionality reduction methods such as principal component analysis, which can explain the main variability of the data. Typically, the top few principal components that explain a cumulative variance reaching a preset proportion (e.g., 85%) are selected. Variational mode decomposition (VMD) refers to an adaptive information... The decomposition algorithm iteratively solves a variational problem to decompose a complex signal into several intrinsic mode functions (IMF) components with specific center frequencies and finite bandwidths. The trend curve refers to the smooth signal obtained by identifying and reconstructing IMF components with center frequencies below a preset frequency threshold, reflecting the long-term law of the slow evolution of component performance parameters over time, and its slope reflects the degradation rate. The transient disturbance feature refers to the high-frequency pulse signal obtained by extracting IMF components with center frequencies above a preset frequency threshold, reflecting sudden anomalies in the data interaction process between components, and its pulse energy and frequency reflect the intensity of the conflict.

[0026] Understandably, employing an adaptive sliding window to reduce the dimensionality of the state feature vector sequence, and utilizing variational mode decomposition to extract trend curves describing component performance evolution and transient disturbance features describing component interaction anomalies, aims to address the challenges of massive and real-time data flow during the operation of power metering boxes. Simultaneously, it separates long-term degradation trends and short-term sudden anomalies mixed in the equipment operation signals. The adaptive sliding window dynamically adjusts the analysis granularity, reducing computational frequency to save resources when data is stable and improving analysis accuracy to capture details when abnormal activity occurs. Principal component analysis extracts principal feature components, reducing data dimensionality and suppressing noise interference while retaining key information. Variational mode decomposition adaptively decomposes the principal feature components into eigenmode functions of different frequencies, separating components with different physical meanings from the mixed signal. Identifying low-frequency components reconstructs the trend curve, quantifying the long-term degradation degree of component hardware fatigue. Extracting high-frequency pulse terms identifies transient disturbance features, quantifying the instantaneous intensity of component software conflicts. This dual-track parallel feature extraction mechanism overcomes the limitations of single-time-scale analysis, improving the comprehensiveness and timeliness of fault diagnosis.

[0027] S5: Based on the trigger time monitoring matrix including the monitoring cycle factor and the component failure correlation parameter, the abnormal conflict domain of the component is locked, and the target abnormal component block is located by calculating the Euclidean space deviation between the state feature vector and the topology map. It should be noted that the monitoring cycle factor refers to a control parameter that dynamically adjusts the sampling or analysis frequency based on the equipment's operating status. This parameter is calculated by integrating the trend evolution speed, transient disturbance frequency, and workload level, reflecting the urgency of current monitoring needs. A larger parameter value indicates a higher monitoring density. The component failure correlation parameter refers to a statistical indicator that quantifies the degree of correlation between failures of different components. This parameter is obtained by calculating the cosine similarity of the component's characteristic response vector under historical failure modes and multiplying it by an exponential decay factor based on topological distance, reflecting the probability of mutual influence between component failures. The characteristic response vector refers to the change pattern vector of each monitoring parameter of the component under a specific failure mode. The change pattern vector is constructed by recording the magnitude and direction of the numerical changes of each monitoring indicator (such as voltage, current, temperature, and time delay) when the component experiences a specific failure, and is a multi-dimensional vector. ,in Indicates the first The changes in each monitoring parameter are used to quantify the behavioral characteristics of components under specific failure modes. The trigger moment monitoring matrix refers to a two-dimensional data structure constructed with components as row vectors and discrete trigger moments as column vectors. The values ​​of the matrix elements represent the abnormal trigger intensity of a specific component at a specific moment. This intensity is obtained by nonlinearly mapping the weighted perturbation amplitude of the failure correlation of adjacent components through an activation function. The component abnormal conflict domain refers to the set of multiple components with abnormal state coupling identified by analyzing the clustering patterns of elements exceeding a preset threshold in the trigger moment monitoring matrix in the component and time dimensions. The Euclidean space deviation refers to the weighted Euclidean distance between the real-time state feature vector and the health baseline vector in the fault evolution topology map in the multi-dimensional feature coordinate system. It is calculated by weighted summation of the absolute values ​​of the differences in each dimension and reflects the comprehensive degree of deviation of the component from normal operating conditions. The target abnormal component block refers to the smallest set of components with the highest degree of deviation and failure correlation selected by cluster analysis within the abnormal conflict domain.

[0028] Understandably, by constructing a trigger time monitoring matrix that integrates the monitoring cycle factor and component failure correlation parameters based on the trigger time monitoring matrix, a dynamic and adaptive monitoring triggering mechanism is established. By integrating trend evolution speed, transient impact frequency, and business load level to calculate the monitoring cycle factor, adaptive allocation of monitoring resources is achieved. By calculating and integrating the component failure correlation parameters that integrate historical failure mode similarity and topological distance attenuation, the probability intensity of fault propagation between components is quantified. By locking the component anomaly conflict domain and locating the target anomaly component block through Euclidean space deviation, the focus shifts from macroscopic system-level anomaly response to microscopic fault root cause location. By identifying matrix element clusters to determine the anomaly propagation range, and then by weighted distance calculation and cluster analysis to distinguish the dominant fault source from the affected secondary components within the conflict domain, a leap from fuzzy regional alarms to precise component location is achieved, improving the targeting and efficiency of fault repair.

[0029] S6: Extract real-time status data of the target abnormal component block, combine internal status features with equipment health grading rules, output diagnostic results and failure probability, and trigger early warning.

[0030] It should be noted that the real-time status data of the target abnormal component block refers to the detailed operating parameters collected from each component of the block after location is completed, including operating current and voltage fluctuations, node communication latency, command execution success rate, and environmental temperature sensing. This data is collected in real time through the sensors and monitoring interfaces of each component. Internal status characteristics refer to the abstract representation of the component's current operating conditions after feature engineering transformation of the real-time status data. This includes statistical characteristics (mean, variance, extreme values), frequency domain characteristics (spectral energy, dominant frequency components), and time domain characteristics (waveform factor, impulse factor). These characteristics are extracted using signal processing algorithms. Equipment health grading rules refer to the criteria for classifying equipment status into different levels based on the severity and scope of the fault. These typically include gradients such as normal, caution, abnormal, and dangerous. Each level is determined by setting specific judgment conditions. For example, the normal level requires all parameters to be within the rated range and the proportion of parameters exceeding the limit to be 0%. The health level allows individual parameters to exceed limits for short periods with an exceedance rate of <5%. The abnormal level has multiple parameters exceeding limits with an exceedance rate of 5%-20%. The danger level has critical parameters severely exceeding limits with an exceedance rate of >20%. The diagnostic result refers to a structured output containing information such as the location of the fault, the degree of performance degradation, and the type of logical anomaly. The failure probability refers to a quantitative indicator of the likelihood of a component failing within a specific timeframe in the future, calculated based on the health level and topology location. This probability is calculated by establishing a mapping relationship between the health level and the failure probability. For example, the normal level corresponds to a failure probability of 0-5%, the attention level corresponds to 5%-20%, the abnormal level corresponds to 20%-50%, and the danger level corresponds to 50%-100%. The probability is also weighted and adjusted based on the component's position weight in the topology graph (center node weight 1.5, edge node weight 1.0). The differentiated alarm level refers to the warning level distinguished according to the severity of the diagnostic result, such as prompt, warning, serious, and emergency.

[0031] Understandably, extracting real-time status data of the target abnormal component block, combining internal status characteristics with equipment health grading rules, outputting diagnostic results and failure probabilities, and triggering early warnings aims to build a complete closed-loop management mechanism from fault discovery to maintenance decision-making. By extracting detailed real-time data of the located component block, the original physical information of the equipment operation is obtained. By converting real-time status data into feature components and comparing them with health benchmarks, the over-limit ratio and anomaly frequency are calculated, discretizing continuous deviations into understandable equipment status levels. By combining the component's position in the topology map to calculate the failure probability, the amplification effect of critical node failures on the overall system risk is reflected. By outputting structured diagnostic results including location, degradation level, and anomaly type, specific maintenance guidance is provided to maintenance personnel. By pushing differentiated alarms based on severity, reasonable allocation of monitoring resources and priority ranking of response measures are achieved, ultimately realizing a shift from passive fault handling to proactive health management in the maintenance model.

[0032] Preferably, the component includes a power supply module, a communication module, a sensor group, and a main control unit; Identifying the physical connections and logical dependencies between the various components of the IoT terminal within the electricity metering box includes: Obtain the physical electrical connection paths between the power supply module, communication module, sensor group, and main control unit; Based on the functional collaboration logic between components, a star-shaped dependency model with the main control unit as the core is established, defining the support dependency relationship of the power supply module to each active component, as well as the data flow dependency relationship between the sensor group and the communication module to the main control unit. Extract the spatiotemporal association rules of the operating data of each component under normal working conditions, and define the interaction behaviors that violate the logical sequence or electrical topology constraints as logical dependency conflict items.

[0033] It should be noted that the power supply module refers to the functional unit that provides a stable power supply to the various electrical components within the energy metering box, including AC-DC conversion circuits, voltage regulator circuits, and power management chips, used to convert external AC power into the DC operating voltage required by each component; the communication module refers to the wireless or wired communication unit responsible for data interaction with external networks or monitoring centers, including 4G / 5G modules, WiFi chips, or power line carrier communication circuits, used to establish a data channel between the metering box and the cloud platform; the sensor group refers to the collection of detection elements used to collect physical quantities related to energy metering, including current transformers, voltage transformers, temperature sensors, and humidity sensors, used to convert electrical parameters of the physical world... The data is converted into a processable electrical signal; the main control unit refers to the central processing module that undertakes data processing and logic control functions, including a microprocessor, memory, and peripheral interface circuits, used to coordinate the work of various components and execute metering algorithms; the physical electrical performance connection path refers to the topology describing the actual physical transmission route of electrical energy and signals between components, including the direction of power lines, connection nodes of signal lines, and distribution of grounding points, which is obtained by analyzing the circuit schematic; the star-dependent model refers to the logical structure with the central node as the core and peripheral nodes connected to it through independent links, reflecting the centralized control relationship of the main control unit over other components, which is established by analyzing the system control architecture; data flow direction dependency Dependency refers to the unidirectional or bidirectional flow constraints of data from the acquisition end to the processing end and then to the transmission end, including the data acquisition flow from the sensor group to the main control unit and the data upload flow from the main control unit to the communication module; spatiotemporal correlation rules refer to the coordinated laws that the operating parameters of each component should follow in the time and space dimensions under normal operating conditions, including the temporal consistency of voltage fluctuations and load changes (established by statistically analyzing the response delay time distribution of voltage drops relative to load surges under normal operating conditions, such as voltage drops should occur within 10ms after load surges) and the clock synchronization of sensor sampling and main control processing (established by statistically analyzing the time interval distribution between the interruption of sampling and the data reading operation, such as data reading...). The sampling should be executed within 5ms after the sampling interruption. This rule is established by calculating the statistical distribution characteristics (mean, variance, confidence interval) of historical normal data. Logical dependency conflict items refer to abnormal interactive behaviors that violate the above-mentioned coordination rules, including misalignment between power fluctuation timing and load change timing (such as voltage drop leading load surge) and clock loss during data acquisition and processing (such as data reading delay exceeding 20ms). The marking of this conflict item is used to provide anomaly judgment benchmark for establishing the fault evolution topology map in step S3, that is, to provide specific violation rule examples for the deviation degree calculation and state offset vector identification in step S3, so that the topology map can identify anomalies based on clear logical violation conditions.

[0034] Understandably, by clarifying the component composition and refining the specific steps for identifying physical connections and logical dependencies, the abstract component association identification process is transformed into an operable engineering implementation path, solving the problem of the lack of specific implementation methods for the identification steps. A low-level electrical connection mapping is established by acquiring physical electrical performance connection paths; a high-level logical control architecture is established by constructing a star-topology dependency model; and a dynamic collaborative constraint benchmark is established by extracting spatiotemporal correlation rules. This benchmark is established by statistically analyzing the time-series correlation of each component's operating data under normal operating conditions, including calculating voltage-load timing constraints and acquisition-processing timing constraints. This three-layer progressive identification mechanism works together to achieve a panoramic depiction of the static connection relationships and dynamic interaction patterns between components, providing accurate structured prior knowledge for subsequent heterogeneous data alignment and fault evolution analysis. This avoids blind spots in fault propagation path identification caused by missing topology information, improving the completeness and accuracy of the monitoring framework.

[0035] Specifically, obtaining the physical electrical performance connection paths between the power supply module, communication module, sensor group, and main control unit can be achieved by analyzing the circuit schematic and PCB layout file of the power metering box, extracting the power supply network topology from the AC input terminal through the AC-DC conversion circuit to the power pins of each component, marking the data acquisition link from the sensor interface through the signal conditioning circuit to the ADC pin of the main control unit, recording the data transmission path from the main control unit communication interface through the drive circuit to the communication module, and establishing a physical connection matrix including node coordinates and connection relationships; establishing a star-shaped dependency model with the main control unit as the core based on the functional coordination logic between components can be achieved by analyzing the system software architecture and task scheduling mechanism, identifying the main control unit as the system... The central role of the clock source and control command initiator is established, creating a power management dependency relationship from the main control unit to the power supply module (e.g., the main control unit adjusts the power supply voltage via the I2C bus), a data acquisition and control dependency relationship from the main control unit to the sensor group (e.g., the main control unit configures sampling parameters via the SPI bus), and a data upload dependency relationship from the main control unit to the communication module (e.g., the main control unit triggers data transmission via the UART interface), forming a four-node star topology. The spatiotemporal correlation rules for extracting the operating data of each component under normal operating conditions can be achieved by statistically analyzing the operating data under normal operating conditions. Specifically, this includes: analyzing the response delay time distribution of power supply voltage fluctuations relative to load current changes, calculating the time difference sequence between voltage dips and load surges, and establishing the mean. and standard deviation The voltage-load timing constraint rule is set so that voltage dips should occur after load surges. The time interval between the sensor sampling completion interruption and the main control data reading operation is analyzed, and the time difference sequence from the sampling interruption to the data reading is calculated to establish the mean. The time interval between the sampling interruption and the data reading is calculated within a specified time period (e.g., if the statistical mean is 10ms and the standard deviation is 2ms, then the constraint is 8-12ms). and standard deviation Set the acquisition-processing timing constraint rule so that data reading should occur after the sampling interruption. The execution time is specified (e.g., if the statistical mean is 5ms and the standard deviation is 1ms, then the constraint is 3-7ms). Interactions that violate the above timing constraints are marked as logical dependency conflicts. This mark is used in step S3 as an abnormal condition to determine deviation from the topology baseline, i.e., when real-time monitoring shows a voltage drop preceding a load surge (time difference < 0) or a data read delay exceeding [a certain threshold]. When the time is greater than 7ms, the interaction behavior is marked as a logical dependency conflict item, which serves as the basis for constructing abnormal edges in the fault evolution topology graph.

[0036] Preferably, the heterogeneous data modality alignment algorithm is used to map data with different sampling frequencies to a unified time axis, generating a state feature vector sequence sorted by weight values, including: The component index with the highest sampling frequency in the heterogeneous operation data is determined as the time reference axis. The power supply index and communication link index of the low-frequency sampling are supplemented by the linear interpolation algorithm to achieve phase synchronization and alignment of different modal data at discrete time points. By using the deviation standardization method and combining the historical extreme values ​​of the parameters of each component, the heterogeneous operating data covering electrical quantities, electromagnetic quantities, logic percentages and discrete error codes are mapped to a unified numerical range, generating a standardized heterogeneous feature set. Standardized heterogeneous features aligned at the same reference time point are extracted to construct a multidimensional initial feature vector. The dimensions of the initial feature vector are rearranged and weighted according to the monitoring weight value to generate a sequence of state feature vectors describing the real-time health status of the device.

[0037] It should be noted that sampling frequency refers to the number of discrete samples of a physical quantity per unit time, reflecting the temporal resolution of the data; the time reference axis refers to a high-frequency time coordinate system used as a synchronization reference, providing a unified time scale standard; and the linear interpolation algorithm refers to a data completion method that estimates intermediate values ​​between known discrete data points using a linear function, employing a formula... ,in and For adjacent sampled values, and For adjacent sampling times, For interpolation timing, the algorithm is based on the assumption of linear change between two points, and calculates the interpolation point value proportionally to complete the low-frequency data onto the high-frequency time axis; phase synchronization alignment refers to the process of adjusting data from different components with different timestamps to the same time coordinate point, which is achieved by interpolating other low-frequency data based on the highest frequency time point; the deviation normalization method refers to a dimensionless processing technique that linearly transforms the data to the 0-1 interval by subtracting the minimum value from the original data and then dividing by the range (maximum value minus minimum value). ,in The original data values, and This method uses historical extreme values ​​of a parameter to avoid dimensional differences between different physical quantities, making data from different attributes such as voltage, current, and temperature comparable. Historical extreme values ​​refer to the maximum and minimum values ​​of a parameter within a statistical period (e.g., the past 7 or 30 days), serving as boundary references for standardization calculations and updated statistically through a sliding window. Electrical quantities refer to physical quantities describing the electrical characteristics of a circuit, including voltage, current, power, and resistance. Electromagnetic quantities refer to physical quantities describing the characteristics of electromagnetic fields, including magnetic field strength, electromagnetic radiation, and electromagnetic interference. Logical percentages refer to relative quantities describing the digital logic state or resource utilization ratio, such as CPU utilization and memory usage. Discrete error codes refer to coded identifiers representing specific fault or anomaly types, usually represented by integers for different error categories (e.g., 0 for no error, 1 for communication timeout, 2 for sampling failure, etc.). In standardization, their integer values ​​are mapped to the 0-1 range. Error codes 0-255 are mapped to 0-1, allowing them to participate in vector calculations as numerical features, reflecting the severity or frequency of errors. They can then be used in subsequent analyses for unified calculations and distance calculations with other continuous parameters. This transforms discrete classification information into continuous numerical features, facilitating quantitative analysis and similarity calculations in the feature space. Standardized heterogeneous feature sets refer to multi-source data sets after numerical normalization, avoiding dimensional differences in the original data. Multidimensional initial feature vectors refer to column or row vectors composed of standardized parameters of each component at a single time point. Reordering and weighted encapsulation refer to adjusting the order and magnitude of each dimension in the feature vector according to preset importance coefficients. First, the dimension order is adjusted based on monitoring weight values ​​(e.g., high-weight dimensions are placed first). Then, each dimension value is multiplied by its corresponding weight coefficient (e.g., dimension value × weight value), giving important parameters higher numerical weights. This naturally reflects the amplification effect of key component anomalies on the overall state assessment in subsequent spatial distance calculations.

[0038] Understandably, by identifying the component index with the highest sampling frequency in the heterogeneous operational data as the time reference axis, a unified time coordinate system is established to solve the problem of time asynchrony between multi-source data. By using linear interpolation algorithms to complete the power supply and communication link indices of low-frequency samples, and estimating intermediate values ​​between adjacent sampling points based on the linear variation assumption, phase synchronization and alignment of different modal data at discrete time points is achieved, avoiding false anomalies caused by sampling time misalignment. Furthermore, by using deviation standardization methods combined with the historical extreme values ​​of each component parameter index, heterogeneous operational data covering electrical quantities, electromagnetic quantities, logic percentages, and discrete error codes are mapped to a unified numerical value. Within the interval, to avoid dimensional differences, discrete error codes are used as numerical features in vector calculations to reflect the severity of errors, generating a standardized heterogeneous feature set. By extracting standardized heterogeneous features aligned at the same reference time point, a multidimensional initial feature vector is constructed. The dimensions of the initial feature vector are rearranged and weighted according to the monitoring weight values. The dimensional order is adjusted and multiplied by the weight coefficient to generate a sequence of state feature vectors describing the real-time health status of the equipment. The geometric structure of the feature vectors reflects the differences in component importance, thus naturally reflecting the amplification effect of key component anomalies on the overall status assessment in subsequent spatial distance calculations, improving the sensitivity and positioning accuracy of fault detection.

[0039] Specifically, determining the component index with the highest sampling frequency in the heterogeneous operation data as the time reference axis can be achieved by analyzing the data acquisition configuration of each component. This involves identifying that the sensor group collects current and voltage signals at 100Hz, the main control unit records operating status at 10Hz, and the communication module reports link quality at 1Hz. 100Hz is selected as the highest sampling frequency to establish the time reference axis, with a time scale interval of 10 milliseconds. A linear interpolation algorithm is then used to complete the low-frequency sampling power supply and communication link indices. For the 10Hz data from the main control unit, nine equally spaced interpolation points are inserted between every two adjacent sampling points, using the linear interpolation formula. To calculate interpolation point values, for 1Hz data from the communication module, 99 interpolation points are inserted between every two sampling points to achieve phase synchronization and alignment of different modal data at discrete time points. The heterogeneous operational data is mapped to a unified numerical range using a deviation normalization method. The maximum and minimum values ​​from the past 7 days of operational data are extracted as historical extreme values ​​for each parameter, and the current collected value is substituted into the formula. Mapping to the 0-1 range, handling electrical quantities (e.g., mapping a voltage range of 12V to 15V to 0-1), electromagnetic quantities (e.g., mapping a magnetic field strength of 0 to 100mG to 0-1), logical percentages (e.g., mapping CPU utilization of 0% to 100% to 0-1), and discrete error codes (e.g., linearly compressing error codes 0 to 255 to the 0-1 range, mapping error code 0 to 0, error code 255 to 1, and mapping intermediate values ​​proportionally); extracting aligned and standardized heterogeneous features at the same reference time point to construct a multi-dimensional initial feature vector; at each 10-millisecond time point, extracting the aligned standardized value of the supply voltage and the main control CPU... The occupancy rate standardization value, communication signal strength standardization value, and sensor temperature standardization value are arranged in a fixed order to form a four-dimensional vector. The dimensions of the initial feature vector are rearranged and weighted according to the monitoring weight values. Based on the monitoring weight values ​​(e.g., power supply module 0.30, main control unit 0.35, communication module 0.20, sensor group 0.15), the dimension order of the initial feature vector is adjusted to main control unit parameters, power supply module parameters, communication module parameters, and sensor group parameters. Each dimension value is multiplied by its corresponding weight coefficient (e.g., main control dimension value 0.45 × 0.35 = 0.1575) to generate a weighted state feature vector sequence.

[0040] Preferably, establishing a fault evolution topology map of the device itself based on logical dependencies includes: Based on the physical electrical performance connection path and the star-shaped dependency model, each component is mapped to a physical node in the graph, and a directed edge representing the fault propagation law is established between each physical node according to the data flow dependency relationship between the components. Extract the collaborative constraint rules among the standardized heterogeneous feature sets corresponding to each physical node under normal operating conditions, including calculating the linear correlation between power supply voltage fluctuations and main control load changes, as well as the timing consistency between sensor sampling frequency and communication retransmission frequency, and generate a topology function describing the hardware health benchmark. The real-time operating data of each physical node is compared with the topology function. Based on the statistical indicators of voltage deviation, signal strength ratio and watchdog reset frequency, the deviation of the real-time characteristic value of the physical node from the health benchmark is calculated. Based on the degree of deviation, state offset vectors that conform to the failure propagation effect are selected, and the state offset vectors are mapped to the corresponding directed edges to form a fault evolution topology map of the device itself. The selection conditions include the degree of deviation continuously exceeding a first preset threshold and reaching a first preset duration.

[0041] It should be noted that a physical node refers to an abstract vertex representing an actual hardware component in the fault evolution topology graph, carrying the component's state characteristics and fault mode information; a directed edge refers to a graph edge connecting two physical nodes and having directionality, representing the possible paths and probabilities of fault propagation from the source node to the target node; fault propagation laws refer to empirical knowledge or statistical laws describing the mechanism of fault propagation between components, including the propagation law of power supply faults to load components and the propagation law of communication faults to data processing components; cooperative constraint laws refer to the mathematical or logical timing relationships that should be satisfied between the operating parameters of different components under normal operating conditions, such as the linear correlation between voltage and load, and the clock synchronization between sampling and transmission; linear correlation refers to the quantitative index of the degree of linear association between two variables, using the Pearson correlation coefficient. Calculation, where and These are sample values ​​for voltage and load, respectively. and The value represents the sample mean, ranging from -1 to 1. A larger absolute value indicates a stronger linear correlation. Temporal consistency refers to the sequential or time interval constraints that two discrete events should satisfy on the time axis. A topological function is a mathematical mapping describing the healthy and collaborative relationships between components, established using function fitting methods. For example, regarding the relationship between power supply voltage and main control load, by collecting voltage data corresponding to different loads under historical normal operating conditions, a polynomial function is fitted using the least squares method. ,in For load rate, Voltage, coefficient By minimizing the sum of squared errors This function represents a baseline mapping relationship between voltage and load variation fitted based on historical normal data, used to predict the theoretical normal value of voltage under the current load; voltage deviation refers to the absolute or relative value of the difference between the actual supply voltage and the rated voltage; signal strength ratio refers to the ratio of the actual received signal strength to the maximum receivable signal strength; watchdog reset frequency refers to the number of times the watchdog timer triggers a system reset per unit time; deviation degree refers to the quantitative difference between the real-time characteristic value and the health baseline value, using the relative deviation formula. Calculation, where These are measured values. The predicted value of the topological function. The nominal value reflects the relative magnitude of the real-time state deviation from the health baseline; the state offset vector is a multi-dimensional vector describing the direction and magnitude of the component state deviation from the normal baseline; the failure propagation effect refers to the physical phenomenon of a fault spreading from one component to another and triggering a chain of failures; the first preset threshold is the boundary value for determining whether the degree of deviation constitutes an abnormal state, which is determined by analyzing the statistical relationship between the degree of deviation before the fault and the occurrence of the fault in historical fault data. For example, the 90th percentile of the degree of deviation before the fault in history is taken as the threshold (e.g., 0.2, or 20%). This threshold indicates that there is a high risk of failure when the deviation exceeds this value; the first preset duration is the minimum time length for the degree of deviation to continuously exceed the threshold, which is determined by analyzing the distribution of the duration of abnormality in historical fault data. For example, 50% of the average fault development time is taken as the duration (e.g., 10 seconds). This duration is used to filter transient noise and ensure that only persistent abnormalities are considered as valid fault propagation.

[0042] Understandably, by mapping each component to a physical node in a graph based on physical electrical performance connection paths and a star-dependent model, and establishing directed edges representing fault propagation patterns between physical nodes according to the data flow dependencies between components, a structured carrier for fault propagation is created. By extracting the collaborative constraint patterns between standardized heterogeneous feature sets corresponding to each physical node under normal operating conditions, quantifying the linear correlation between voltage and load by calculating the Pearson correlation coefficient, and establishing the temporal consistency of sampling and transmission through statistical time-series distribution, a topology function describing the hardware health baseline is generated. This function establishes a mathematical mapping relationship through polynomial fitting or regression analysis. By comparing the real-time operating data of each physical node with the topology function, the deviation of real-time feature values ​​from the health baseline (relative deviation) is calculated, and state offset vectors that conform to the failure propagation effect are selected to form a fault evolution topology graph of the equipment itself. This constructs a computable model that reflects the propagation path and evolution trend of faults between components, providing a dynamic topology analysis basis for subsequent disturbance coefficient injection and anomaly localization, and improving the ability to predict complex fault chain reactions.

[0043] Specifically, mapping each component to physical nodes in the graph can be done by abstracting the power supply module, main control unit, communication module, and sensor group as four component nodes in the graph based on the physical electrical performance connection path and star-dependent model. Directed edges representing fault propagation patterns are established between each physical node. These can be based on the current supply direction, creating directed edges from the power supply module to the main control unit, communication module, and sensor group, representing the propagation path of power supply performance fluctuations to downstream components. Directed edges can also be created based on the data flow direction, from the sensor group to the main control unit and from the main control unit to the communication module, representing the propagation path of component interaction anomalies along the data processing chain. The collaborative constraint patterns between the standardized heterogeneous feature sets corresponding to each physical node under normal operating conditions are extracted. This can be achieved by statistically analyzing historical normal data and calculating the Pearson correlation coefficient between the power supply voltage and the main control load. For example, collecting 1000 sets of voltage and load data under normal operating conditions can yield the correlation coefficient. This indicates that the voltage decreases with increasing load. The average time interval and its variance from sensor sampling completion to the start of communication module transmission are calculated. A topology function describing the hardware health baseline is generated, such as fitting the voltage-load relationship using the least squares method. (in For voltage, (For load rate), or by using a neural network to fit a more complex nonlinear relationship, this function represents the theoretical normal value of voltage under a given load; the real-time operating data of each physical node is compared with the features of the topology function, and for the voltage deviation value, calculation is performed. The system calculates the actual strength / maximum strength based on the signal strength ratio and the number of resets per unit time based on the watchdog reset frequency to obtain the degree of deviation relative to the health baseline. Based on the degree of deviation, state offset vectors that conform to the failure propagation effect are selected. The first preset threshold is set to 0.2 (determined by analyzing historical fault data statistics) and the first preset duration is 10 seconds (determined by analyzing fault development time statistics). When the degree of deviation of a component continues to exceed 0.2 and the duration exceeds 10 seconds, it is determined that the component has generated a state offset vector that conforms to the failure propagation effect. This vector is then mapped to the directed edge between the component and the affected component to form the fault evolution topology map of the device itself.

[0044] Preferably, injecting the perturbation coefficients of component operating data deviating from the topology baseline into the corresponding state feature vector includes: The standardized heterogeneous feature set collected in real time is input into the fault evolution topology map, the quantization residual between the measured feature value and the health benchmark is extracted, and the degree of difference between the quantization residual and the physical node triggering condition and hardware response is calculated. The quantized residual is compared with a preset fault mode library to verify whether the difference has logical consistency on the failure path of the corresponding physical node, so as to determine the effective disturbance item. According to the preset mapping rules, each effective perturbation term is converted into a normalized perturbation coefficient using an exponential function, and the perturbation coefficient is used as the corresponding dimension of the weight gain injection state feature vector. Feature enhancement is achieved by increasing the geometric distance between the abnormal dimension and the normal dimension in the multidimensional Euclidean space.

[0045] It should be noted that the quantization residual refers to the difference vector between the measured feature values ​​and the health baseline values ​​predicted by the topological function, expressed by the formula... The calculation reflects the direction and magnitude of the deviation between the actual state and the normal state; the trigger condition refers to the specific physical or logical conditions that lead to the occurrence of a fault or abnormal activation, such as the voltage drop threshold (below 90% of the rated value) or the temperature over-limit (above 80°C of the rated value), which is determined by analyzing the component specifications and historical fault data; the hardware response refers to the output change or state change of the hardware component under specific trigger conditions, such as the main control unit reducing its frequency when undervoltage occurs, or the system reducing its load or restarting when overtemperature occurs, which is identified by monitoring the component status register or output signal; the difference refers to the degree of inconsistency between the quantization residual and the trigger condition and hardware response levels, using the formula Calculation, where For actual observed residuals, The residual, expected based on physical models (such as Ohm's law and power formula), reflects the degree of deviation between the actual abnormal mode and the theoretical fault model. The fault mode library refers to a predefined set of feature descriptions for various fault modes, including parameter combination modes under different fault types (such as power fluctuations, load anomalies, communication interruptions, sensor failures, etc.). Each mode is stored in the form of a feature vector. ,in Indicates the first fault under this fault mode The typical variation range of each monitoring parameter is analyzed in this database through historical fault case analysis and expert knowledge construction. Logical consistency refers to the degree of conformity between the measured abnormal mode and the known fault mode in terms of propagation path and impact range, which is calculated by comparing the measured difference feature vector with the cosine similarity of each mode in the fault mode database. Verification is performed when the similarity exceeds a preset threshold (e.g., 0.8), indicating logical consistency. Valid perturbation terms refer to verified abnormal deviations that conform to the fault propagation logic and are noteworthy; these deviations simultaneously exceed the threshold and exhibit high similarity to known fault modes. The mapping rule refers to the mathematical transformation relationship that converts residual values ​​into perturbation coefficients, using an exponential function. Or Sigmoid form ,in , , This is the sensitivity adjustment coefficient. The quantized residual is determined by analyzing the correlation between the magnitude of the residual and the severity of the fault in historical data. This mapping converts the linear residual into a nonlinear perturbation coefficient, achieving significant amplification of large deviations and effective suppression of small fluctuations. The normalized perturbation coefficient refers to the gain coefficient mapped to the 0 to 1 interval after exponential transformation, reflecting the severity of the anomaly. The weight gain refers to the multiplicative enhancement factor applied to the feature vector dimension. The geometric distance refers to the Euclidean distance between two vector points in the multidimensional feature space. Feature enhancement refers to the process of adjusting feature values ​​to make abnormal features easier to distinguish in pattern recognition.

[0046] Understandably, by inputting the standardized heterogeneous feature set acquired in real time into the fault evolution topology map, the quantization residual between the measured feature values ​​and the health baseline is extracted. The degree of difference between the quantization residual and the physical node triggering conditions and hardware response is calculated. Abnormal patterns are identified by comparing the actual observed residual with the theoretically expected residual. By comparing the quantization residual with a preset fault mode library, the cosine similarity is calculated to verify whether the degree of difference has logical consistency on the failure path of the corresponding physical node, ensuring that only valid disturbance terms that conform to the known fault propagation law are retained, thus determining the valid disturbance terms. Based on the preset mapping rules, an exponential function is used... The sigmoid function or the number of perturbations transforms each effective perturbation term into a normalized perturbation coefficient, nonlinearly mapping the absolute value of the residual to the interval between 0 and 1. The perturbation coefficient is then injected as the weight gain into the corresponding dimension of the state feature vector. The perturbation coefficient is incorporated into the feature vector through multiplication operations (such as original value × (1 + perturbation coefficient) or original value × perturbation coefficient). By increasing the geometric distance between the abnormal dimension and the normal dimension in the multidimensional Euclidean space, feature enhancement is achieved. This increases the geometric distance between the abnormal state and the normal state in the feature space, enhancing the sensitivity of subsequent classifiers or detection algorithms to the identification of weak anomalies and improving the early warning capability of fault detection.

[0047] Specifically, the standardized heterogeneous feature set collected in real time is input into the fault evolution topology map, and the quantization residual between the measured feature values ​​and the health baseline is extracted. For example, the standardized value of the measured voltage of the power supply module is 0.25 (actual 10V, rated 12V), and the topology function prediction value is 0.30 (predicted 11V). The quantization residual is then calculated. Calculate the difference between the quantization residual and the physical node triggering conditions and hardware response; analyze whether the residual triggers the voltage drop threshold (e.g., standardized value < 0.167, i.e., 10V / 12V); check whether the main control unit exhibits a frequency reduction response (e.g., CPU frequency decrease); and calculate the difference between the actually observed voltage residual of -0.05 and the voltage drop expected based on Ohm's law (e.g., a voltage drop of 0.03 due to increased load). The quantized residual is compared with a pre-defined fault mode library, which stores power fluctuation modes (feature vectors [-0.1, 0.05, 0.02]), load anomaly modes (feature vectors [0.05, -0.1, 0.03]), etc. The cosine similarity between the current difference vector [0.02, ...] and each mode is calculated. If the similarity with the power fluctuation mode is 0.85, exceeding the threshold of 0.8, logical consistency is verified, and it is determined as a valid disturbance term. Based on the pre-defined mapping rules, an exponential function is used... (in (Determined through debugging with historical data), the effective perturbation term (absolute residual value 0.05) is converted into a normalized perturbation coefficient of 0.632, and the perturbation coefficient is used as the corresponding dimension of the weight gain injected into the state feature vector. The original value of this dimension, 0.25, is multiplied by (1+0.632) to obtain 0.408, or multiplied by 0.632 to obtain 0.158, depending on the specific enhancement strategy, to achieve feature enhancement of this dimension.

[0048] Preferably, the dimensionality reduction process yields the principal feature components that reflect the operational fluctuations of the component. Variational mode decomposition is used to extract trend curves describing component hardware fatigue and transient disturbance characteristics of component software conflicts, including: The modal decomposition layers and penalty factors are preset, and the principal feature components are iteratively decomposed to obtain eigenmode function components with different center frequencies. Low-frequency terms with center frequencies below a preset frequency threshold are identified from the intrinsic mode function components and the signals are reconstructed to obtain trend curves describing the slow evolution of the electrical performance parameters of each component over time, which are used to quantify the long-term degradation of the components. High-frequency pulse terms are extracted from the intrinsic mode function components. By calculating the correlation between the high-frequency energy distribution and the envelope spectrum, transient disturbance characteristics describing data flow abrupt changes or logical collisions of each component are identified and used to quantify the instantaneous conflict intensity of the components.

[0049] It should be noted that dimensionality reduction refers to mapping high-dimensional state feature vectors to a low-dimensional space using PCA, while retaining the main variation information; principal feature components refer to selecting the principal components with the highest variance explanation rate after PCA dimensionality reduction as the main feature dimensions reflecting the component's operational fluctuations. This is specifically defined through the following steps: constructing a data matrix. ( For the sample size, (For feature dimension), calculate the covariance matrix. Solve the characteristic equation Obtain eigenvalues and eigenvectors Select the largest eigenvalue Corresponding feature vector As the first principal component, i.e., the principal characteristic component, this component explains a variance of [missing information]. It reflects the direction of maximum variation in the original data and contains the main information about component operational fluctuations; Variational Mode Decomposition (VMD) refers to an adaptive signal decomposition algorithm that iteratively solves variational problems to decompose complex signals into several intrinsic mode functions (IMF) components with specific center frequencies; the number of mode decomposition layers. The number of predefined intrinsic mode functions (IMFs) determines the level of granularity of the decomposition. This is determined by analyzing the signal's spectral complexity. For example, if the decomposition includes four layers: trend, low-frequency disturbance, noise, and transients, then... Punishment factor The weighting parameter, which controls the bandwidth of each mode, affects the convergence speed and mode separation effect of the decomposition. It is determined through debugging and is usually set to 1000-5000. Iterative decomposition refers to the process of solving the variational problem using the alternating direction multiplier method (ADMM). Specifically, it includes: initializing the spectrum and center frequency of each mode, and alternately updating the frequency domain representation of each mode. and center frequency The Lagrange multipliers are updated through soft thresholding until the convergence condition is met (e.g., energy change between adjacent iterations < 0.001), yielding each IMF component. The intrinsic mode function component refers to the physically meaningful single-component signal obtained from the decomposition, with each component representing the oscillation mode of the original signal in a specific frequency band. The center frequency refers to the energy concentration location of the intrinsic mode function component in the frequency domain. The preset frequency threshold refers to the boundary frequency value that distinguishes between low-frequency trend terms and high-frequency transient terms, usually set according to signal characteristics and analysis requirements, such as 1Hz. The trend curve refers to the smooth signal reconstructed from the low-frequency intrinsic mode function components (center frequency < 1Hz), reflecting the long-term law of the slow evolution of component performance parameters over time. The long-term degradation degree refers to the cumulative degree of slow deterioration of component performance over time, such as capacitance decay and increased contact resistance, calculated by the slope of the trend curve within the time window. or curvature Quantification, and The trend curve represents the rate of change of performance parameters, where the slope reflects the rate of change and the curvature reflects the acceleration of change. Both together characterize the trend and severity of component performance degradation; a larger absolute slope indicates faster degradation, and a larger curvature indicates a more drastic change in the degradation rate. The high-frequency pulse term refers to the intrinsic mode function components with a center frequency higher than a preset threshold (e.g., 5Hz), containing sudden changes and transient impact information of the signal. The high-frequency energy distribution refers to the energy concentration or pulse intensity of the high-frequency components in the time domain. The envelope spectrum correlation refers to the degree of synchronization between the high-frequency component envelope and the trigger signal of a specific event. The cross-correlation function is a function that measures the similarity between two signals, using the formula... Calculate, where, For the high-frequency pulse term envelope, For known fault trigger signals, To address the time delay, the correspondence between transient disturbance characteristics and fault triggering time is identified by finding the peak position of the cross-correlation function. Transient disturbance characteristics refer to the abnormal features of data flow abrupt changes (such as the envelope peak aligning with the communication interruption time) or logical collisions (such as the envelope peak aligning with the interruption conflict log time) identified by calculating the cross-correlation function between the envelope and the fault triggering signal. These features are used to quantify the instantaneous conflict intensity of the components.

[0050] Understandably, dimensionality reduction is used to obtain the principal eigenvalues ​​reflecting component operational fluctuations. PCA is then used to calculate the covariance matrix and perform eigenvalue decomposition. The principal eigenvalues ​​corresponding to the largest eigenvalues ​​are selected to define the main direction of data variation, addressing the challenge of massive data volumes. Variational mode decomposition is used to extract trend curves describing component hardware fatigue and transient disturbance features of component software conflicts. The number of mode decomposition levels and penalty factors are preset, and the principal eigenvalues ​​are iteratively decomposed to obtain eigenmode function components with different center frequencies. Low-frequency terms with center frequencies below a preset frequency threshold are identified and their information is analyzed. The trend curve is reconstructed, and the long-term degradation degree is quantified by calculating the slope and curvature of the trend curve. The slope reflects the rate of performance degradation, and the curvature reflects the acceleration of degradation. Both together characterize the cumulative degree of hardware fatigue. By extracting high-frequency pulse terms and calculating the correlation between high-frequency energy distribution and envelope spectrum, the envelope is extracted through Hilbert transform, and the cross-correlation function with the fault trigger signal is calculated. Transient disturbance characteristics are identified by identifying cross-correlation peaks, enabling accurate diagnosis of fault nature. This can predict hardware lifespan degradation and capture logical collisions at the software level, providing data support for the formulation of differentiated operation and maintenance solutions.

[0051] Specifically, an adaptive sliding window is used to reduce the dimensionality of the state feature vector sequence. The preset window length is 100 sampling points, which is dynamically adjusted according to the data variance within the window. The principal feature components after dimensionality reduction are defined, a data matrix is ​​constructed, the covariance matrix is ​​calculated, and eigenvalue decomposition is performed. The first principal component corresponding to the largest eigenvalue is selected as the principal feature component. If the variance explanation rate of the first principal component is calculated to be 75%, this component is selected to represent the main variation in the original data. Variational mode decomposition is used, with a preset mode decomposition layer number K=4 (corresponding to the trend term, low-frequency disturbance term, mid-frequency noise term, and high-frequency transient term), and a preset penalty factor. The principal eigenfunction components are iteratively optimized by updating the modal spectrum and center frequency using the alternating direction multiplier method until the energy change between adjacent iterations is less than 0.001, thus obtaining the four intrinsic mode function components. to and its center frequency (e.g.) Center frequency 0.05Hz Center frequency 0.8Hz Center frequency 5Hz Center frequency 15Hz); identify low-frequency items with a center frequency lower than a preset frequency threshold (e.g., 1Hz). and The signal is reconstructed to obtain the trend curve. The trend curve is It means that among them The time variable reflects the evolution trajectory of component performance parameters over time, and the trend curve is calculated within the time window (assuming the starting time is...). The end time is slope or curvature Quantify the degree of long-term degradation; for example, a slope of -0.02 indicates a slow decline in performance, while a curvature of 0.005 indicates a gradually increasing degradation rate; extract high-frequency pulse terms (such as...). ,by (representing the original time-domain signal), the analytic signal is calculated using the Hilbert transform. ,in The imaginary unit, for Hilbert transform (i.e. and (convolution), extract the envelope. (representing the instantaneous amplitude of the signal), calculate the envelope and the known fault trigger signal (in terms of...). The cross-correlation function (e.g., the pulse sequence at the moment of communication interruption) represents... ,in The time delay represents the time offset between the envelope signal and the fault trigger signal, determined by identifying the peak position of the cross-correlation function (e.g., ...). The peak value is 0.85. The transient disturbance characteristics of the data stream change (such as the envelope peak value being aligned with the communication interruption time) or logic collision (such as the envelope peak value being aligned with the interruption conflict log time) are identified, and the transient conflict intensity is quantified by the envelope amplitude or pulse count.

[0052] Preferably, based on a trigger time monitoring matrix including monitoring cycle factors and component failure correlation parameters, the component anomaly conflict domain is locked as follows: The component failure correlation parameters Satisfying the relation: ; in, Representation Component With components Component failure correlation parameters between and Representing components respectively With components In the Feature response vectors under each historical fault mode This indicates the total number of preset historical fault modes. Representation Component With components Topological logical distance in the star dependency model This represents the preset topology attenuation coefficient; Extract the slope and rate of change of the trend curve and the pulse frequency of the transient disturbance characteristics, and combine them with the service load fluctuation of the power metering box to calculate the dynamic monitoring period factor. Satisfying the relation: ; in, This indicates the rate of change of the slope of the trend curve within the current sliding window. Indicates the duration of observation The number of transient disturbance pulses identified internally. Indicates the current service load power. This indicates the rated maximum power consumption of the electricity metering box. , and These represent the preset adjustment weights used to balance the effects of trends, disturbances, and loads; Using each component as a row vector and the discrete trigger time as a column vector, the failure correlation parameter and the monitoring period factor are injected to construct a trigger time monitoring matrix. The trigger time monitoring matrix elements Satisfying the relation: ; in, Indicates the monitoring matrix at the th The first component, the first The trigger strength value at each moment, This represents the activation function. Representation Component At any moment The real-time disturbance amplitude, Representation and Components A set of physically connected adjacent components. Represents the sampling decision indication function; Identify the cluster of elements in the trigger time monitoring matrix whose trigger intensity value exceeds a preset threshold, determine the set of components that have abnormal state coupling, and lock them as the component abnormal conflict domain.

[0053] It should be noted that the component failure correlation parameter is a statistical indicator that quantifies the degree of correlation between failures of two components, taking into account both historical failure mode similarity and topological proximity. and Representing components respectively With components In the The characteristic response vector under historical fault modes is constructed by recording the deviation of each monitored parameter of the component from the normal value when the historical fault occurs. For example, if the power supply module voltage drops by 20%, current fluctuates by 15%, and temperature rises by 10°C during a power fault, then the characteristic response vector is [-0.20, 0.15, 0.10]. This indicates the total number of preset historical fault modes. Representation Component With components Topological logical distance in a star dependency model (1 for direct connection, 2 for cross-level connection). This represents the preset topological attenuation coefficient (usually 0.5). The first part of the formula calculates the cosine similarity of the feature response vectors of the two components, while the second part uses exponential decay to reflect the physical law that the closer the components are, the stronger their correlation. The rate of change of the slope of the trend curve is an indicator describing the acceleration of component performance degradation, expressed by the formula... The calculation reflects the urgency of the deterioration; the transient disturbance pulse frequency refers to the number of high-frequency pulse events detected per unit time, calculated using the formula... Calculation, where For observation duration The number of pulses identified internally; the service load power refers to the power consumption corresponding to the current amount of electricity metering services processed by the electricity metering box; the rated maximum power consumption refers to the maximum power limit allowed by the design of the metering box; adjustment weight. , , This refers to the coefficient that balances the contributions of different factors. It is determined by analyzing resource constraints and monitoring accuracy requirements through system calculations. For example, it may be set as follows: , , To balance the effects of trends, disturbances, and load, the weights are determined through optimization algorithms or engineering experience to ensure that the monitoring cycle can capture rapid changes while avoiding oversampling; the trigger moment monitoring matrix is ​​a two-dimensional data structure constructed with components as rows and discrete trigger moments as columns, where matrix elements represent the abnormal trigger strength of a specific component at a specific moment; activation function This refers to a function that maps a linear weighted sum to a nonlinear output, expressed in the form of a sigmoid function. The input value is mapped to the interval between 0 and 1 to quantize the trigger strength. This function has smooth nonlinear characteristics and can compress any real number input to the (0,1) interval, reflecting the trigger probability. The real-time perturbation amplitude refers to the quantized value of the abnormality of the component at the current moment. The adjacent component set refers to the set of components that have a direct physical or logical connection with the target component. Sampling decision indication function. This refers to a Boolean function that determines whether to sample at the current moment based on a monitoring period factor, using modular arithmetic logic. ,in To monitor the periodicity factor, this function at time... satisfy Can be When the value is divisible, the output is 1 (indicating that sampling is triggered at that moment); otherwise, the output is 0 (indicating that sampling is skipped at that moment), thus realizing adaptive sampling decision based on the monitoring period factor. Element cluster refers to the set of matrix elements whose values ​​exceed the threshold and are spatially adjacent (in the same row or column). Abnormal state coupling refers to the phenomenon that multiple components simultaneously or successively experience abnormalities within a short period of time and there is a correlation and transmission.

[0054] Understandably, by calculating component failure correlation parameters that integrate historical failure mode similarity and topological distance attenuation, and by calculating the cosine similarity of the characteristic response vectors and combining it with the topological attenuation coefficient, the probability intensity of fault propagation between components is quantified, and a statistical correlation model for fault propagation is established. Furthermore, by extracting the slope variation rate of the trend curve and the pulse frequency of transient disturbance characteristics, and combining this with the business load fluctuations of the energy metering box, a dynamic monitoring cycle factor is calculated through weighted summation. This system enables adaptive allocation of monitoring resources, reducing sampling frequency to conserve resources when the state is stable and increasing sampling frequency to capture details when abnormal activity occurs. By constructing a trigger moment monitoring matrix, the trigger intensity is calculated using the Sigmoid activation function, and the sampling decision indicator function controls the sampling moment, storing the abnormal intensity of multiple components and multiple moments in a structured form. By identifying clusters of elements exceeding the threshold in the matrix, the system macroscopically locks down the set of components coupled with abnormal states, determines the scope of fault propagation and hotspot areas, and provides macroscopic screening results for subsequent precise positioning, significantly improving the ability to quickly screen abnormal areas in large-scale systems.

[0055] Specifically, the component failure correlation parameter is calculated, assuming the component and The feature response vectors under the five historical fault modes are as follows: and The calculated cosine similarity is 0.85, and the topological distance is... attenuation coefficient ,but Calculate the dynamic monitoring cycle factor, assuming the trend curve slope changes at a certain rate. Transient disturbance pulse frequency Business load rate Adjusting weights , , ,but This indicates that sampling should be performed at a period of approximately 4 seconds; a trigger moment monitoring matrix is ​​constructed, with each component as a row vector and the discrete trigger moments as column vectors, and the matrix elements... ,in Use the Sigmoid function. Modulo operation is used for judgment. Assuming that the weighted perturbation sum of adjacent components is 2.5 at a certain moment, it becomes 0.924 after Sigmoid mapping. The sampling condition is met at the current moment. If the trigger strength value is divisible by 4, then the trigger strength value at that moment is 0.924; identify the cluster of elements in the trigger moment monitoring matrix whose trigger strength value exceeds the preset threshold (such as 0.8), determine the set of components that have abnormal state coupling, and lock them as the component abnormal conflict domain.

[0056] Preferably, the component block for locating target anomalies by calculating the Euclidean spatial deviation between the state feature vector and the topological map includes: Within the component abnormal conflict domain, the state feature vector corresponding to each component is retrieved, and the corresponding health benchmark vector is extracted from the fault evolution topology map. The two are then mapped to a unified multi-dimensional feature coordinate space. For each dimension of the multidimensional feature coordinate space, the absolute value of the difference between the real-time feature value and the baseline feature value is calculated, and then weighted and summed in combination with the monitoring weight value to obtain the deviation value describing the degree of deviation of each component from the normal operating condition. A preset deviation threshold is set, and the deviation values ​​of each component in the conflict domain are sorted and clustered. The smallest set of components with deviation values ​​exceeding the deviation threshold and having failure correlation is selected and located as the target abnormal component block.

[0057] It should be noted that the component anomaly conflict domain refers to the set of multiple components with abnormal state coupling identified through the above steps, which serves as the search range for the localization operation; the health baseline vector refers to the standard vector defined in the fault evolution topology map, describing the values ​​of each parameter that should take under normal operating conditions of the component, and is calculated through the above topology function; the multidimensional feature coordinate space refers to the vector space constructed with each feature dimension as the coordinate axis, and the state of each component can be represented as a point in this space; the absolute value of the difference... This refers to the absolute mathematical value of the difference between real-time feature values ​​and health benchmark values ​​on the same dimension, reflecting the magnitude of deviation in that dimension; weighted summation refers to multiplying the absolute differences of each dimension by their corresponding monitoring weights and then summing them up using the formula. Calculation, where For the first The monitoring weight values ​​of each dimension yield a comprehensive deviation index. This process ensures that deviations in key parameters have a greater impact on the final deviation value. The deviation value is a scalar value describing the degree to which a component deviates from its normal operating condition, calculated using weighted Euclidean distance or weighted Manhattan distance. The deviation threshold is the boundary value for determining whether a component is seriously abnormal, determined by analyzing historical fault data, such as taking the 90th percentile of the deviation value before a historical fault. Cluster analysis is a statistical method for grouping components with similar deviation characteristics or high correlation into one class. Using the K-means algorithm, the optimal number of clusters is determined by calculating intra-cluster and inter-cluster distances, grouping components with similar deviation values ​​and strong failure correlation into one class. The minimum component set refers to the set containing the fewest components under the condition of meeting the anomaly judgment, reflecting the principle of accurate fault location. The target abnormal component block refers to the final identified set of components that require key monitoring or maintenance.

[0058] Understandably, by retrieving the state feature vectors corresponding to each component within the component anomaly conflict domain and extracting the corresponding health baseline vectors from the fault evolution topology map, both are mapped to a unified multi-dimensional feature coordinate space, establishing a comparative reference system between real-time and normal states. For each dimension of the multi-dimensional feature coordinate space, the absolute value of the difference between the real-time feature value and the baseline feature value is calculated, and weighted summation is performed in conjunction with monitoring weight values. By weighted summation of the absolute differences in each dimension, a deviation value describing the degree of deviation of each component from normal operating conditions is obtained, giving higher deviation weights to anomalies in key components. By setting a deviation threshold, the deviation values ​​of each component within the conflict domain are sorted and clustered. Components with similar deviation values ​​are clustered using the K-means algorithm, and the smallest set of components with deviation values ​​exceeding the deviation threshold and exhibiting failure correlation is selected and located as the target anomaly component block. This allows for accurate identification of the fault origin point in the complex anomaly propagation network, avoiding indiscriminate maintenance of all components in the entire conflict domain, and improving the accuracy of fault location and maintenance efficiency.

[0059] Specifically, within the component exception conflict domain, the state feature vectors corresponding to each component are retrieved. This vector contains multidimensional health parameters aligned and enhanced through the above steps, and also extracts the health baseline vectors that each component should exhibit under normal operating conditions from the fault evolution topology map. Mapping both to a unified multidimensional feature coordinate space; for each feature dimension in this multidimensional space Calculate real-time feature values With health benchmarks absolute value of the difference The absolute difference of each dimension Monitoring weight values ​​for the corresponding dimensions Multiply and sum by using the formula Obtain a comprehensive deviation value describing the overall deviation of each component from normal operating conditions; preset deviation threshold. (Determined through analysis of historical failure data statistics) The deviation values ​​of all components within the conflict domain are sorted by numerical value. A K-means clustering algorithm (setting the number of clusters to 2-3) is used to group components with similar deviation values ​​and strong failure correlation into one cluster. The optimal clustering is determined by calculating intra-cluster distance (e.g., intra-cluster deviation value variance) and inter-cluster distance (e.g., cluster center distance). Components with deviation values ​​exceeding a preset threshold are then selected. Furthermore, by using the smallest set of components with failure propagation correlation as the target abnormal component block, the source of the fault is narrowed down to a specific target component block, which greatly improves the accuracy of on-site maintenance personnel in replacing spare parts.

[0060] Preferably, combining internal state characteristics with equipment health grading rules, outputting diagnostic results and failure probabilities, and triggering early warnings includes: Extract the real-time status data of the target abnormal component block, including operating current and voltage fluctuations, node communication latency, instruction execution success rate, and ambient temperature. The real-time status data is converted into feature components describing the current operating condition of the component and compared with the health benchmark in the fault evolution topology map. By identifying the proportion of the real-time status data exceeding the preset deviation threshold and the frequency of interaction abnormal features occurring within the sliding window, the corresponding health level is obtained. Based on the health level obtained from the matching, and combined with the position of the target abnormal component block in the topology map, the real-time failure probability of each component is calculated, and diagnostic results including the location of the failure, the degree of performance degradation, and the type of logical anomaly are output synchronously. Based on the severity of the diagnostic results, a warning signal with a differentiated alarm level is pushed to the monitoring terminal.

[0061] It should be noted that real-time status data refers to the raw operating parameters collected from the set of located faulty components, including operating current and voltage fluctuations, node communication latency, command execution success rate, and environmental temperature sensing, collected through the sensors and monitoring interfaces of each component; feature components refer to the abstract representation of real-time status data after transformation by feature extraction algorithms, including statistical features (such as the variance of voltage fluctuations, the mean and maximum deviation of communication latency), frequency domain features (such as spectral energy distribution), and time domain features (such as waveform factors and impulse factors), extracted through mathematical calculations and signal processing; equipment health grading rules refer to classifying equipment based on the severity of the fault. The standby status is divided into several levels based on judgment criteria, including normal, caution, abnormal, and dangerous levels. Each level is determined by setting specific judgment conditions: the normal level requires all parameters to be within the rated range and the over-limit ratio is 0%; the caution level allows individual parameters to exceed the limit for a short time and the over-limit ratio is <5%; the abnormal level has multiple parameters exceeding the limit and the over-limit ratio is 5%-20%; the dangerous level has critical parameters severely exceeding the limit and the over-limit ratio is >20%. This rule is determined by analyzing the statistical relationship between the severity of historical faults and the parameter over-limit range; the over-limit ratio refers to the proportion of real-time status data points that exceed the preset normal operating range to the total number of data points, using the formula... Calculation, where For the number of data points exceeding the limit, The total number of data points; the frequency of occurrence refers to the number of times a specific abnormal event occurs per unit of time, expressed by the formula. Calculation, where This represents the number of abnormal events that occurred. The observation time is [not specified]. Real-time failure probability refers to the probability value predicted based on the current health level and topology location of a component within a specific future timeframe, expressed by the formula [not specified]. Calculation, where Based on the probability of failure (e.g., 15% for the attention level, 30% for the abnormal level, and 50% for the danger level). The topology location weights are set at 1.5 for central nodes such as the main control unit and 1.0 for edge nodes such as sensors. This probability is determined through historical fault data statistics and component importance analysis. The diagnostic results refer to the structured output containing information such as the location of the fault, the degree of performance degradation, and the type of logical anomaly. The differentiated alarm levels refer to the warning levels distinguished according to the severity, such as prompt, warning, serious, and emergency, which are achieved by setting different notification methods and response processes.

[0062] Understandably, by extracting real-time status data of the target abnormal component block, detailed fault characterization information is obtained; by converting the real-time status data into feature components describing the current operating condition of the component, and comparing it with the health benchmark in the fault evolution topology map, the over-limit ratio and abnormal frequency are calculated, and the corresponding health level is matched according to the preset health level classification rules, thus discretizing the continuous deviation degree into an understandable equipment status category; by combining the matched health level with the position of the target abnormal component block in the topology map, the real-time failure probability of each component is calculated by multiplying the basic probability with the position weight, reflecting the amplification effect of critical node failure on the overall system risk; by synchronously outputting diagnostic results including the fault location, performance degradation degree, and logical anomaly type, and pushing early warning signals with differentiated alarm levels to the monitoring terminal according to the severity of the diagnostic results, a closed-loop management mechanism from fault discovery to maintenance response is constructed to achieve quantitative assessment and graded response of fault severity.

[0063] Specifically, real-time status data of the target abnormal component block is extracted, including collecting operating current fluctuations through current sensors, voltage fluctuations through voltage sensors, node communication latency through the communication protocol stack, command execution success rate through task scheduling logs, and ambient temperature through temperature sensors. This real-time status data is then transformed into feature components describing the current operating condition of the component, including calculating the statistical variance and waveform factor of current and voltage fluctuations. (in The root mean square voltage (RMS) represents the effective value of the voltage, expressed by the formula... calculate, For the first The instantaneous voltage value at each sampling time. This represents the total number of data points within the sampling window. The average voltage is obtained through the formula. calculate, The waveform factor reflects the degree of voltage waveform distortion for a standard sine wave. The larger the deviation from this value, the more severe the waveform distortion. The average and maximum deviation of communication delay are calculated, the descent gradient of command execution success rate is calculated, and the temperature rise rate is calculated. Combining internal state characteristics with equipment health grading rules, the above feature components are compared with the health benchmark in the fault evolution topology map. The proportion of data points exceeding the preset normal operating range (e.g., voltage rated value ±10%) in real-time status data is identified to obtain the over-limit ratio. The frequency of occurrence of interactive abnormal features (e.g., communication timeout, command failure) within a specific observation window (e.g., 5 minutes) is statistically analyzed to obtain the occurrence frequency. Health levels are matched according to preset health grading rules: over-limit ratio <5% and frequency <3 times / minute is matched as a warning level; over-limit ratio 5%-20% and frequency 3-10 times / minute is matched as an abnormal level; over-limit ratio >20% and frequency >10 times / minute is matched as a danger level. The obtained health level, combined with the position of the target abnormal component block in the topology map, calculates the real-time failure probability of each component. For example, if the base failure probability is 30% when the health level is abnormal, and the target component block includes the main control unit (central node, weight 1.5) and the power supply module (critical node, weight 1.3), then the real-time failure probability of the main control unit is 30% × 1.5 = 45%, and the real-time failure probability of the power supply module is 30% × 1.3 = 39%. The system synchronously outputs diagnostic results including the location of the fault (e.g., the communication interface of the main control unit), the degree of performance degradation (e.g., moderate degradation), and the type of logical anomaly (e.g., communication timeout accompanied by a decrease in processing capacity). Based on the severity of the diagnostic results, the system pushes early warning signals with differentiated alarm levels to the monitoring terminal, such as sending a prompt message (green) for the attention level, a warning message (yellow) for the abnormal level, and triggering an emergency alarm (red) for the danger level.

[0064] An IoT-based fault monitoring system for electricity metering boxes, such as Figure 2 As shown, it includes: The parameter configuration module is used to identify the physical connections and logical dependencies between the components of the Internet of Things terminal in the power metering box, and to configure the multi-dimensional self-test parameter indicators and their monitoring weight values ​​corresponding to the components. The data alignment module is used to collect heterogeneous operating data of the components, and use a heterogeneous data modal alignment algorithm to map data with different sampling frequencies to a unified time axis, generating a state feature vector sequence sorted by weight value; The topology building module is used to build a fault evolution topology map of the device itself based on logical dependencies, and to inject the perturbation coefficient of the component operation data deviating from the topology baseline into the corresponding state feature vector. The feature extraction module is used to perform dimensionality reduction on the state feature vector sequence using an adaptive sliding window, and to extract trend curves describing the performance evolution of components and transient perturbation features describing abnormal component interactions using variational mode decomposition. The conflict localization module is used to locate the abnormal conflict domain of the component based on the trigger time monitoring matrix, which includes the monitoring cycle factor and the correlation parameters of component failure. The module locates the target abnormal component block by calculating the Euclidean space deviation between the state feature vector and the topology map. The graded early warning module is used to extract real-time status data of the target abnormal component block, combine internal status characteristics with equipment health grading rules, output diagnostic results and failure probability, and trigger early warning.

[0065] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0066] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for fault monitoring of electricity metering boxes based on the Internet of Things, characterized in that, include: Identify the physical connections and logical dependencies between the components of the Internet of Things (IoT) terminal in the power metering box, and configure the multi-dimensional self-testing parameter indicators and their monitoring weight values ​​corresponding to the components. Collect heterogeneous operating data of the components, and use a heterogeneous data modal alignment algorithm to map data with different sampling frequencies to a unified time axis to generate a state feature vector sequence sorted by weight values; A fault evolution topology map of the device itself is established based on logical dependencies, and the perturbation coefficient of the component operation data deviating from the topology baseline is injected into the corresponding state feature vector. An adaptive sliding window is used to reduce the dimensionality of the state feature vector sequence, and variational mode decomposition is used to extract the trend curves describing the performance evolution of the components and the transient perturbation features describing the abnormal interaction of the components. Based on the trigger time monitoring matrix, which includes monitoring cycle factors and component failure correlation parameters, the abnormal conflict domain of the component is locked, and the target abnormal component block is located by calculating the Euclidean space deviation between the state feature vector and the topology map. Extract real-time status data of the target abnormal component block, combine internal status characteristics with equipment health grading rules, output diagnostic results and failure probability, and trigger early warning.

2. The method for fault monitoring of electricity metering boxes based on the Internet of Things according to claim 1, characterized in that, The components include a power supply module, a communication module, a sensor group, and a main control unit; Identifying the physical connections and logical dependencies between the various components of the IoT terminal within the electricity metering box includes: Obtain the physical electrical connection paths between the power supply module, communication module, sensor group, and main control unit; Based on the functional collaboration logic between components, a star-shaped dependency model with the main control unit as the core is established, defining the support dependency relationship of the power supply module to each active component, as well as the data flow dependency relationship between the sensor group and the communication module to the main control unit. Extract the spatiotemporal association rules of the operating data of each component under normal working conditions, and define the interaction behaviors that violate the logical sequence or electrical topology constraints as logical dependency conflict items.

3. The method for fault monitoring of electricity metering boxes based on the Internet of Things according to claim 2, characterized in that, Using a heterogeneous data modal alignment algorithm, data with different sampling frequencies are mapped to a unified time axis, generating a sequence of state feature vectors sorted by weight values, including: The component index with the highest sampling frequency in the heterogeneous operation data is determined as the time reference axis. The power supply index and communication link index of the low-frequency sampling are supplemented by the linear interpolation algorithm to achieve phase synchronization and alignment of different modal data at discrete time points. By using the deviation standardization method and combining the historical extreme values ​​of the parameters of each component, the heterogeneous operating data covering electrical quantities, electromagnetic quantities, logic percentages and discrete error codes are mapped to a unified numerical range, generating a standardized heterogeneous feature set. Standardized heterogeneous features aligned at the same reference time point are extracted to construct a multidimensional initial feature vector. The dimensions of the initial feature vector are rearranged and weighted according to the monitoring weight value to generate a sequence of state feature vectors describing the real-time health status of the device.

4. The method for fault monitoring of electricity metering boxes based on the Internet of Things according to claim 3, characterized in that, The device's own fault evolution topology map, built based on logical dependencies, includes: Based on the physical electrical performance connection path and the star-shaped dependency model, each component is mapped to a physical node in the graph, and a directed edge representing the fault propagation law is established between each physical node according to the data flow dependency relationship between the components. Extract the collaborative constraint rules between the standardized heterogeneous feature sets corresponding to each physical node under normal operating conditions, including calculating the linear correlation between power supply voltage fluctuations and main control load changes, as well as the timing consistency between sensor sampling frequency and communication retransmission frequency, and generate a topology function describing the hardware health benchmark. The real-time operating data of each physical node is compared with the topology function. Based on the statistical indicators of voltage deviation, signal strength ratio and watchdog reset frequency, the deviation of the real-time characteristic value of the physical node from the health benchmark is calculated. Based on the degree of deviation, state offset vectors that conform to the failure propagation effect are selected, and the state offset vectors are mapped to the corresponding directed edges to form a fault evolution topology map of the device itself. The selection conditions include the degree of deviation continuously exceeding a first preset threshold and reaching a first preset duration.

5. The method for fault monitoring of electricity metering boxes based on the Internet of Things according to claim 4, characterized in that, Injecting the perturbation coefficients of component operating data that deviate from the topology baseline into the corresponding state feature vector includes: The standardized heterogeneous feature set collected in real time is input into the fault evolution topology map, the quantization residual between the measured feature value and the health benchmark is extracted, and the degree of difference of the quantization residual between the physical node triggering condition and the hardware response is calculated. The quantized residual is compared with a preset fault mode library to verify whether the difference has logical consistency on the failure path of the corresponding physical node, so as to determine the effective disturbance item. According to the preset mapping rules, each effective perturbation term is converted into a normalized perturbation coefficient using an exponential function, and the perturbation coefficient is used as the corresponding dimension of the weight gain injection state feature vector. Feature enhancement is achieved by increasing the geometric distance between the abnormal dimension and the normal dimension in the multidimensional Euclidean space.

6. The method for fault monitoring of electricity metering boxes based on the Internet of Things according to claim 1, characterized in that, The dimensionality reduction process yields the main feature components that reflect the operational fluctuations of the component. Variational mode decomposition is used to extract trend curves describing component hardware fatigue and transient disturbance characteristics of component software conflicts, including: The modal decomposition layers and penalty factors are preset, and the principal feature components are iteratively decomposed to obtain eigenmode function components with different center frequencies. Low-frequency terms with center frequencies below a preset frequency threshold are identified from the intrinsic mode function components and the signals are reconstructed to obtain trend curves describing the slow evolution of the electrical performance parameters of each component over time, which are used to quantify the long-term degradation of the components. High-frequency pulse terms are extracted from the intrinsic mode function components. By calculating the correlation between the high-frequency energy distribution and the envelope spectrum, transient disturbance characteristics describing data flow abrupt changes or logical collisions of each component are identified and used to quantify the instantaneous conflict intensity of the components.

7. The method for fault monitoring of electricity metering boxes based on the Internet of Things according to claim 2, characterized in that, Based on the trigger time monitoring matrix, which includes monitoring cycle factors and component failure correlation parameters, the component anomaly conflict domains are identified as follows: The component failure correlation parameters Satisfying the relation: ； in, Representation Component With components Component failure correlation parameters between them and Representing components respectively With components In the Feature response vectors under historical fault modes This indicates the total number of preset historical fault modes. Representation Component With components Topological logical distance in the star dependency model This represents the preset topology attenuation coefficient; Extract the slope and rate of change of the trend curve and the pulse frequency of the transient disturbance characteristics, and combine them with the service load fluctuation of the power metering box to calculate the dynamic monitoring period factor. Satisfying the relation: ； in, This indicates the rate of change of the slope of the trend curve within the current sliding window. Indicates the duration of observation The number of transient disturbance pulses identified internally. Indicates the current service load power. This indicates the rated maximum power consumption of the electricity metering box. , and These represent the preset adjustment weights used to balance the effects of trends, disturbances, and loads; Using each component as a row vector and the discrete trigger time as a column vector, the failure correlation parameter and the monitoring period factor are injected to construct a trigger time monitoring matrix. The trigger time monitoring matrix elements Satisfying the relation: ； in, Indicates the monitoring matrix at the th The first component, the first The trigger strength value at each moment, This represents the activation function. Representation Component At any moment The real-time disturbance amplitude, Representation and Components A set of physically connected adjacent components. Represents the sampling decision indication function; Identify the cluster of elements in the trigger time monitoring matrix whose trigger intensity value exceeds a preset threshold, determine the set of components that have abnormal state coupling, and lock them as the component abnormal conflict domain.

8. The method for fault monitoring of electricity metering boxes based on the Internet of Things according to claim 7, characterized in that, The component block for locating target anomalies by calculating the Euclidean spatial deviation between the state feature vector and the topological map includes: Within the component abnormal conflict domain, the state feature vector corresponding to each component is retrieved, and the corresponding health benchmark vector is extracted from the fault evolution topology map. The two are then mapped to a unified multi-dimensional feature coordinate space. For each dimension of the multidimensional feature coordinate space, the absolute value of the difference between the real-time feature value and the baseline feature value is calculated, and then weighted and summed in combination with the monitoring weight value to obtain the deviation value describing the degree of deviation of each component from the normal operating condition. A preset deviation threshold is set, and the deviation values ​​of each component in the conflict domain are sorted and clustered. The smallest set of components with deviation values ​​exceeding the deviation threshold and failure correlation is selected and located as the target abnormal component block.

9. The method for fault monitoring of electricity metering boxes based on the Internet of Things according to claim 1, characterized in that, By combining internal state characteristics with equipment health grading rules, the system outputs diagnostic results and failure probabilities, and triggers early warnings, including: Extract the real-time status data of the target abnormal component block, including operating current and voltage fluctuations, node communication latency, instruction execution success rate, and ambient temperature. The real-time status data is converted into feature components describing the current operating condition of the component and compared with the health benchmark in the fault evolution topology map. By identifying the proportion of the real-time status data exceeding the preset deviation threshold and the frequency of interaction abnormal features occurring within the sliding window, the corresponding health level is obtained. Based on the health level obtained from the matching, and combined with the position of the target abnormal component block in the topology map, the real-time failure probability of each component is calculated, and diagnostic results including the location of the failure, the degree of performance degradation, and the type of logical anomaly are output synchronously. Based on the severity of the diagnostic results, a warning signal with a differentiated alarm level is pushed to the monitoring terminal.

10. A fault monitoring system for electricity metering boxes based on the Internet of Things, characterized in that, include: The parameter configuration module is used to identify the physical connections and logical dependencies between the components of the Internet of Things terminal in the power metering box, and to configure the multi-dimensional self-test parameter indicators and their monitoring weight values ​​corresponding to the components. The data alignment module is used to collect heterogeneous operating data of the components, and use a heterogeneous data modal alignment algorithm to map data with different sampling frequencies to a unified time axis, generating a state feature vector sequence sorted by weight value; The topology building module is used to build a fault evolution topology map of the device itself based on logical dependencies, and to inject the perturbation coefficient of the component operation data deviating from the topology baseline into the corresponding state feature vector. The feature extraction module is used to perform dimensionality reduction on the state feature vector sequence using an adaptive sliding window, and to extract trend curves describing the performance evolution of components and transient perturbation features describing abnormal component interactions using variational mode decomposition. The conflict localization module is used to locate the abnormal conflict domain of the component based on the trigger time monitoring matrix, which includes the monitoring cycle factor and the component failure correlation parameter. The target abnormal component block is located by calculating the Euclidean space deviation between the state feature vector and the topology map. The graded early warning module is used to extract real-time status data of the target abnormal component block, combine internal status characteristics with equipment health grading rules, output diagnostic results and failure probability, and trigger early warning.

Images