Fault self-diagnosis system for weak current engineering electrical equipment

Abstract

The invention relates to the field of fault monitoring, and discloses a weak current engineering electrical equipment fault self-diagnosis system, which comprises a data acquisition module, a feature processing module, a fault analysis module and a fault early warning module. Specifically, the method comprises the steps of collecting operation time sequence monitoring data of node equipment; performing multi-scale time sequence analysis through wavelet packet transformation, and extracting time domain, frequency domain and entropy features to generate a multi-scale feature set; screening out a feature subset having direct causal association with the fault state based on a causal discovery algorithm; constructing a dual discrimination baseline containing a static threshold and a dynamic trend for the feature subset by using historical normal data; and carrying out parallel comparison and conformity calculation on the real-time characteristics and the base line, and judging whether the equipment has a dominant fault or a hidden gradual change fault according to a calculation result. According to the method, the problems of effectively extracting fault features from single-dimension data and identifying hidden gradual change faults are solved, and the reliability of operation and maintenance of a weak current system is improved.

Description

Technical Field

[0001] This invention relates to the field of fault monitoring, and more specifically to a self-diagnosis system for faults in low-voltage electrical equipment. Background Technology

[0002] With the increasing prevalence of intelligent buildings, the scale of low-voltage electrical engineering systems, such as security monitoring, access control, fire alarms, and information networks, is growing rapidly. The stable operation of core electrical equipment, such as cameras, video recorders, controllers, sensors, and switches, is crucial. However, these devices present some challenges in fault diagnosis; for example: At the data level, low-voltage equipment typically only provides single-dimensional operational sequence data such as current, voltage, and signal strength, resulting in low information density. Traditional fault diagnosis methods either directly set fixed thresholds based on raw data or employ simple statistical features such as mean and maximum values. These methods cannot deeply mine and construct discriminative features that characterize the health status and fault modes of equipment from limited single-dimensional data, leading to insensitivity to early anomalies and complex faults, and a high false negative rate.

[0003] At the diagnostic logic level, most existing methods rely on static threshold alarms. This approach can only detect explicit, sudden faults where data momentarily exceeds the threshold, such as a complete power outage. However, for latent, gradual faults caused by slow degradation of equipment performance, such as signal attenuation due to aging wiring or a slow increase in power consumption due to declining component performance, traditional methods cannot effectively monitor and provide early warnings because the data will not exceed the fixed threshold for a long time. By the time the fault finally manifests, it has often already caused a severe decline in system performance or loss of function. Summary of the Invention

[0004] The purpose of this invention is to provide a self-diagnosis system for faults in electrical equipment in low-voltage engineering, thereby solving at least one of the above-mentioned technical problems.

[0005] The objective of this invention can be achieved through the following technical solutions: A self-diagnostic system for faults in low-voltage electrical equipment includes: The data acquisition module is used to collect real-time runtime sequence monitoring data of each node device in the low-voltage system. The feature processing module includes: A multi-scale analysis unit is used to perform multi-scale time-series analysis on the time-series monitoring data and generate a multi-scale feature set. The causal filtering unit performs feature filtering on the multi-scale feature set based on the causal discovery algorithm to obtain a causal association feature set. The fault analysis module constructs a static threshold baseline and a dynamic trend baseline based on the data distribution and change patterns of the causal correlation feature set under historical normal conditions, forming a baseline for judging the health status of the equipment. The causal relationship features extracted in real time are then compared and their conformity is calculated with the device health status discrimination baseline. The fault warning module is used to determine the health status of the equipment and locate faulty or abnormal nodes based on the comparison results.

[0006] The node devices include at least one of the following: cameras, network video recorders, access control hosts, sensors of fire alarm systems, and switches and routers of integrated cabling systems.

[0007] As a further technical solution, the specific process of performing multi-scale time series analysis on time series monitoring data to generate a multi-scale feature set includes: The time-series monitoring data is decomposed into multiple sub-signal components of different frequency bands based on wavelet packet transform; For each of the sub-signal components, time-domain statistical features, frequency-domain features, and nonlinear entropy features are extracted respectively; The time-domain statistical features, frequency-domain features, and nonlinear entropy features corresponding to all sub-signal components are fused to generate the multi-scale feature set.

[0008] As a further technical solution, the time-domain statistical features include mean, variance, kurtosis, and skewness; the frequency-domain features include dominant frequency, spectral centroid, and sub-band energy ratio; and the nonlinear entropy features include approximate entropy and sample entropy.

[0009] As a further technical solution, the specific process of performing feature filtering on the multi-scale feature set based on the causal discovery algorithm to obtain the causal association feature set includes: Using each feature in the multi-scale feature set as a variable node, an initial completely undirected graph is constructed. Based on the Granger causality test or conditional independence test, the edges in the initial completely undirected graph are pruned to eliminate edges without direct causal dependencies, forming a causal network skeleton graph. Based on the causal direction determination rule, the direction of the edges in the causal network skeleton graph is inferred to form a directed causal graph; From the directed causal graph, feature variables that have a direct causal relationship with the category variables representing the equipment fault state are selected to form the causal association feature set.

[0010] As a further technical solution, the process of constructing a dynamic trend baseline includes: For each feature in the causal correlation feature set, calculate multiple trend indicator sequences for that feature within a continuous time window; then, combine the multiple trend indicators by weighted summation into a comprehensive trend indicator. Based on the comprehensive trend indicator sequence under historical normal conditions, the allowable range of the comprehensive trend indicator and the duration threshold of continuous deviation are determined to form a dynamic trend baseline.

[0011] As a further technical solution, the process of constructing a dynamic trend baseline includes: For each feature in the causal correlation feature set, the trend indicator sequence of the feature within a continuous time window is calculated. Based on the trend indicator sequence under historical normal conditions, the allowable range of the trend change rate and the duration threshold of continuous deviation are determined to form a dynamic trend baseline.

[0012] As a further technical solution, the comparison and conformity calculation performed by the fault analysis module specifically includes the following parallel processes: Threshold compliance calculation: The causal correlation feature value extracted in real time is compared with the normal fluctuation range of the corresponding feature in the static threshold baseline, and the deviation of the feature value is calculated. Trend conformity calculation: Calculate the trend index within a preset time window based on the real-time causal correlation feature value, and compare it with the corresponding trend allowable range and duration threshold in the dynamic trend baseline to determine whether the trend conforms to expectations.

[0013] As a further technical solution, when the fault analysis module compares and calculates the conformity between real-time features and the equipment health status discrimination baseline, it executes the following diagnostic rules: If the real-time causal correlation feature value exceeds the corresponding range of the static threshold baseline, then the corresponding node device is determined to have a manifest fault. If the real-time causal correlation feature value does not exceed the static threshold baseline range, but the trend of the real-time causal correlation feature value continues to deviate from the dynamic trend baseline, then the corresponding node device is determined to have a latent gradual failure.

[0014] As a further technical solution, the process of the fault early warning module locating faulty or abnormal nodes includes: The system has a pre-set mapping library of causal relationship features and equipment components or operating procedures; When a fault is determined, the mapping relationship library is queried based on the specific causal relationship features that trigger the explicit fault or the latent gradual fault, and at least one candidate abnormal equipment component or operating link is output as the location result.

[0015] The beneficial effects of this invention are: (1) This invention performs multi-scale decomposition of runtime sequence data through wavelet packet transform, extracts time-domain statistical features, frequency-domain features and nonlinear entropy features from the complete spectrum from low-frequency trends to high-frequency transients, and deeply fuses them to generate a highly discriminative multi-scale feature set, thereby deeply mining rich information representing the health status of equipment from limited data.

[0016] (2) Based on the causal discovery algorithm, the present invention filters massive features, constructs a directed causal graph and accurately identifies a subset of features that are directly causally related to the fault state of the equipment, thereby effectively eliminating the interference of irrelevant or indirectly related features and improving the pertinence and accuracy of subsequent diagnosis; at the same time, based on historical normal data, a static threshold baseline is set for causal features to capture explicit faults; and a dynamic trend baseline is constructed by analyzing the changing trend of features within a continuous time window, and the allowable range of the trend and the threshold for the duration of continuous deviation are set; thus, the real-time features and the dual baselines are compared and the conformity calculation is performed in parallel, realizing the effective judgment of the health status of the equipment; it can not only immediately determine explicit faults where the data exceeds the threshold, but also keenly identify implicit gradual faults where the data is within the normal range but its changing trend is continuously abnormal, thereby shifting the fault diagnosis from post-alarm to early warning, improving the initiative and reliability of the operation and maintenance of weak current equipment. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a schematic diagram of the logic of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 As shown, the present invention is a self-diagnostic system for faults in electrical equipment of low-voltage engineering, comprising: The data acquisition module is used to collect real-time runtime sequence monitoring data of each node device in the low-voltage system. The node device includes at least one of the following: The monitoring parameters for cameras and network video recorders in security video surveillance systems are: device operating current, network port transmit / receive byte rate, video encoding frame rate, and bitrate fluctuation.

[0021] Access control system: access controller and card reader. The monitored parameters are the power supply voltage of the controller motherboard, the radio frequency signal strength of the card reader, the drive current of the electric lock, and the opening / closing gate response time sequence.

[0022] Fire alarm system: smoke detectors, heat detectors and alarm control panel. Monitoring parameters include detector inspection pulse current value, communication response delay time sequence between the control panel and detectors, and alarm loop impedance value.

[0023] Structured cabling system: network switches and routers, with monitoring parameters including inbound / outbound traffic rates at each port, cyclic redundancy check error packet counts, and timing data of the device's central processing unit utilization.

[0024] By collecting multi-dimensional runtime sequence data in real time, a comprehensive and continuous data foundation is provided for subsequent feature extraction and fault diagnosis, improving the system's adaptability to complex operating conditions. Furthermore, monitoring parameters are customized for different types of low-voltage equipment to achieve refined data collection and provide domain-adaptive data support for fault diagnosis.

[0025] The feature processing module includes: A multi-scale analysis unit is used to perform multi-scale time-series analysis on the time-series monitoring data and generate a multi-scale feature set. The specific process includes: Wavelet packet transform employs the db4 wavelet basis function for three-level decomposition, decomposing the original signal into eight sub-signal components with different frequency bands, covering the complete spectrum from low-frequency trends to high-frequency transients; by extracting multi-scale features through wavelet packet transform, it overcomes the limitations of single-dimensional data and significantly improves the richness and discrimination ability of fault features. For each of the sub-signal components, time-domain statistical features, frequency-domain features, and nonlinear entropy features are extracted respectively; The time-domain statistical features include mean, variance, kurtosis, and skewness; Skewness, also known as the third-order normalized moment, is used to measure the asymmetry of the signal amplitude distribution; positive skewness indicates that the distribution has a long tail on the right and negative skewness indicates that the distribution has a long tail on the left.

[0026] Skewness, or fourth-order normalized moment, is used to measure the sharpness or flatness of a signal amplitude distribution; kurtosis means that the distribution has sharper peaks and thicker tails.

[0027] The frequency domain features include: the dominant frequency, which is the frequency at which the power spectral density is maximum, representing the frequency component with the most concentrated energy in the signal; and the spectral centroid, which is... ,in, The number of frequency components in the power spectrum is determined by the discrete Fourier transform parameters. For the first The frequency value corresponding to each frequency component. For frequency The power spectral density at a given location; this characteristic reflects the center of the energy distribution in the signal spectrum; the sub-band energy ratio, i.e. ,in , The first , The energy of each sub-band, and the change in this ratio can indicate the transfer of signal energy between different frequency bands, and is related to certain mechanical or electrical fault modes.

[0028] The nonlinear entropy features include: Approximate entropy indicates that the more complex and irregular the sequence, the better. Sample entropy, an improved algorithm of approximate entropy, is used to evaluate the complexity of time series data. It has the advantages of being insensitive to changes in data length and providing more consistent results. Similarly, a larger value indicates a more complex sequence. The specific expression is existing technology and will not be elaborated upon.

[0029] The nine features extracted from each sub-signal component—namely, four time-domain features, three frequency-domain features, and two entropy features—are concatenated in a predefined order to form a fixed-dimensional feature vector, thus obtaining the multi-scale feature set.

[0030] Taking the port ingress traffic rate time series data of a switch as an example, through multi-scale decomposition and multi-dimensional feature extraction, the original single-dimensional traffic rate data is transformed into multiple quantitative indicators that can respectively characterize its trend stability, burst mode, and energy distribution, greatly enriching the characterization information of the device status.

[0031] The causal screening unit performs feature screening on the multi-scale feature set based on the causal discovery algorithm to obtain a causal correlation feature set; irrelevant features are eliminated by the causal discovery algorithm, and a subset of features directly related to the fault are retained, which reduces the feature dimensionality and improves the efficiency and accuracy of the subsequent diagnostic model.

[0032] The specific process includes: Initial graph construction: Each feature in the multi-scale feature set is regarded as a node in the graph; a completely undirected graph is initialized, that is, it is assumed that there is an undirected edge between any two feature nodes, indicating that there may be a correlation. Conditional Independence Test and Graph Pruning: Partial correlation coefficients are used as the test statistic for conditional independence. For each pair of adjacent nodes in the graph, i.e., feature X and feature Y, given some other set of nodes Z, the partial correlation coefficient between X and Y is calculated. Then, hypothesis testing is performed, typically with a significance level of α = 0.05, to determine whether X and Y are independent given Z. If the test result supports independence, the edge connecting X and Y is deleted. This process starts with an empty condition set and gradually increases the size of the condition set until all adjacent node pairs are traversed, ultimately resulting in a sparse causal network skeleton graph.

[0033] Edge directionality inference: Based on the causal network skeleton graph, rules based on collision structure are applied to determine the direction of edges. The core rule is: if there are three nodes X, Y, and Z, where X is connected to Y, Y is connected to Z, but X and Z are not connected, and Y is not in the conditional independence set of X and Z, then the direction of the edge can be inferred to be X→Y←Z. Further, the direction propagation rule is applied to determine the direction of the remaining edges in the graph, ultimately forming a partially directed acyclic graph. Causal Relationship Feature Filtering: This directed acyclic graph is a causal graph; in the causal graph, a binary variable F representing the overall fault state of the equipment is defined or introduced; for example, normal = 0, fault = 1. Identify all feature nodes that directly point to the variable F from the causal graph.

[0034] Ultimately, the set of these identified features constitutes the causal relationship feature set. Causal graph modeling enables the causal interpretability of fault features, providing a theoretical basis for fault localization; causal screening ensures that diagnosis focuses on core features, avoiding noise interference. The dual baseline not only defines the absolute boundary of health but also defines the pattern of health changes, laying the foundation for core judgment criteria for capturing gradual anomalies.

[0035] The fault analysis module constructs a static threshold baseline and a dynamic trend baseline based on the data distribution and change patterns of the causal correlation feature set under historical normal conditions, forming a device health status discrimination baseline. The construction of the dual discrimination baseline not only retains the immediacy of traditional threshold detection, but also introduces dynamic trend analysis, which significantly improves the early identification capability of latent gradual faults.

[0036] The specific process for constructing the static threshold baseline includes: For each feature in the causal association feature set, the statistical distribution of that feature in historical normal state data is calculated. The distribution estimation method is as follows: a nonparametric fit is performed on the historical distribution of the feature values ​​using kernel density estimation. The expression for kernel density estimation is: ;in, For kernel functions, such as Gaussian kernels; >0 indicates the bandwidth parameter. These are historical sample values. For eigenvalues, bandwidth By Silverman's Rule of Thumb Automatic selection, among which This represents the sample standard deviation. Let be the probability density function.

[0037] Based on the fitted probability density function The corresponding cumulative distribution function is calculated by integration. The upper limit of normal fluctuation U is taken as The 100*(1-a / 2)% quantile, with the lower limit of normal fluctuation L taken as... The 100*(a / 2)% quantile, where 'a' is the significance level (usually 0.05), constitutes the static threshold baseline for this feature. Static thresholds are constructed based on the statistical distribution of historical normal data, exhibiting high robustness and adaptability, and reducing false alarm rates.

[0038] The process of constructing a dynamic trend baseline includes: Trend indicator calculation: For each causal relationship feature, in its time series data Slide a time window of fixed length w upwards. In each window... Within the window, a least-squares linear fit is performed on the feature value sequence, and the slope of the resulting fitted line is... As a quantitative indicator of the trend of feature changes within this window. This is the end time of the window.

[0039] Calculate the cumulative sum statistic of the eigenvalues ​​within this window relative to the long-term historical mean μ: ; c is the tolerance parameter, typically taken as 0.5 times the historical standard deviation. This represents the feature value at the current time point. This indicates the magnitude by which the accumulated eigenvalues ​​exceed the target value μ. This indicates the cumulative magnitude of the eigenvalue falling below the target value μ.

[0040] Trend baseline construction: Collect all trend indicator sequences, such as slope sequences, calculated using the above method during the long-term historical normal operation of the device. Accumulation and sequence and .

[0041] After standardizing each trend indicator, a dimensionless value is obtained. Then, calculate the moving average of each trend indicator over the entire historical normal period as the normal center of the trend, and calculate the moving standard deviation of each trend indicator as a measure of normal trend fluctuation. The standardized values ​​of the moving averages and moving standard deviations of each trend indicator are weighted and summed to form a composite trend indicator, i.e., the composite moving average. Composite moving standard deviation ; The permissible range of the composite trend indicator is defined as [ , ],in As a control parameter, it is typically set to 2. The duration threshold Tth for the continuous deviation is an empirical parameter, which can be set to M consecutive calculation windows. This interval constitutes the dynamic trend threshold baseline for this feature. The dynamic trend baseline can capture the trend changes of slow degradation in equipment performance, enabling early warning of latent faults.

[0042] The causal relationship features extracted in real time are then compared and their conformity is calculated with the equipment health status discrimination baseline. Through parallel threshold and trend conformity calculation, simultaneous diagnosis of explicit and implicit faults is achieved, thereby improving the diagnostic coverage.

[0043] Specifically, the following parallel processes are included: Threshold compliance calculation: This involves calculating the causal correlation feature values ​​extracted in real time. The deviation degree Dth of the feature value is calculated by comparing it with the normal fluctuation range [L, U] of the corresponding feature in the static threshold baseline. when When <L, then Dth = ;when When >U, then Dth = Otherwise, Dth = 0; Trend conformity calculation: Calculate the trend index within a preset time window based on the real-time causal correlation feature value, and compare it with the corresponding trend allowable range and duration threshold in the dynamic trend baseline to determine whether the trend conforms to expectations.

[0044] For example: Calculate the current trend indicator value based on the feature value sequence of the most recent s time points. and and the allowable range of the dynamic trend threshold baseline [ , The comparison is performed, and the duration of the out-of-range condition is checked to see if it satisfies the condition: Tv ≥ Tth; where Tv is the duration of continuous out-of-range condition. When the fault analysis module compares and calculates the conformity of real-time features with the equipment health status discrimination baseline, it executes the following diagnostic rules: If the real-time causal correlation feature value exceeds the corresponding range of the static threshold baseline, then the corresponding node device is determined to have a manifest fault. If the real-time causal correlation feature value does not exceed the static threshold baseline range, but the trend of the real-time causal correlation feature value continues to deviate from the dynamic trend baseline, then the corresponding node device is determined to have a latent gradual failure.

[0045] Based on the above parallel computing results, the following hierarchical diagnostic decision is executed: If Dth > 0, it is determined to be an explicit fault. If Dth = 0 and Tv ≥ Tth, it is determined to be a latent, gradual fault. By comparing dual baselines in parallel, the system achieves rapid response to acute faults and early warning of chronic faults, transforming the operation and maintenance mode from post-event remediation to pre-event prevention.

[0046] The fault warning module is used to determine the health status of the equipment and locate faulty or abnormal nodes based on the comparison results.

[0047] The specific process includes: The system has a pre-set mapping library of causal relationship features and equipment components or operating procedures; Structure and construction of the mapping relation library: The mapping database between causal association features and equipment components or operational processes is a structured knowledge base. Each mapping record contains at least the following fields: Feature identifier: uniquely identifies a causal relationship feature.

[0048] Description of physical meaning of feature: A textual description of the physical or operating state of the equipment reflected by this feature.

[0049] Associated device component: The name of the hardware component most likely to fail when this feature is abnormal.

[0050] Related operational links: When this feature is abnormal, the function or software operation link that is most likely to be affected.

[0051] Typical failure mode: The typical failure phenomenon corresponding to this abnormal feature.

[0052] Confidence coefficient: A value γ between 0 and 1, representing the reliability of the mapping relationship.

[0053] Recommended maintenance measures: Preliminary troubleshooting or maintenance operations recommended for this failure mode.

[0054] When a fault is identified, the mapping database is queried based on the specific causal relationship features that trigger the explicit or latent gradual fault, and at least one candidate abnormal equipment component or operating step is output as the location result. For example, when the fault warning module receives the diagnostic results from the fault analysis module, including the fault type and a list of specific triggering features, the location inference is initiated: Single-feature trigger location: If a fault is triggered by a single feature, the system directly queries the mapping database for the corresponding device component, operating procedure, and fault mode, outputting the location result. Because it's a single-feature trigger, the system directly outputs the aforementioned components and procedures as candidate location results, providing a clear direction for troubleshooting. This frees maintenance personnel from the uncertainty of inspecting the entire system, allowing them to focus directly on the most likely fault source and significantly shortening the mean time to repair (MTBL).

[0055] Multi-feature collaborative localization: If a fault is triggered simultaneously by R feature items, a Bayesian probabilistic inference mechanism is employed. Assuming there are m candidate fault components, the calculation is performed for each component. posterior probability : ; Where prior probability and likelihood It can be estimated from historical data or expert knowledge. The component with the highest posterior probability is selected as the localization result. In this process, the confidence coefficient γ in the mapping relation library can be used as the weight of prior information.

[0056] By combining a mapping database with Bayesian inference, precise fault location is achieved from the feature layer to the physical device layer, improving operational efficiency. Through the integration of a structured knowledge base and probabilistic reasoning, intelligent fault location and interpretation are realized, providing operational personnel with clear troubleshooting directions.

[0057] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A self-diagnostic system for faults in low-voltage electrical equipment, characterized in that, include: The data acquisition module is used to collect real-time runtime sequence monitoring data of each node device in the low-voltage system. The feature processing module includes: A multi-scale analysis unit is used to perform multi-scale time-series analysis on the time-series monitoring data and generate a multi-scale feature set. The causal filtering unit performs feature filtering on the multi-scale feature set based on the causal discovery algorithm to obtain a causal association feature set. The fault analysis module constructs a static threshold baseline and a dynamic trend baseline based on the data distribution and change patterns of the causal correlation feature set under historical normal conditions, forming a baseline for judging the health status of the equipment. The causal relationship features extracted in real time are then compared and their conformity is calculated with the device health status discrimination baseline. The fault warning module is used to determine the health status of the equipment and locate faulty or abnormal nodes based on the comparison results.

2. The fault self-diagnosis system for low-voltage electrical equipment according to claim 1, characterized in that, The specific process of performing multi-scale time series analysis on time series monitoring data to generate a multi-scale feature set includes: The time-series monitoring data is decomposed into multiple sub-signal components of different frequency bands based on wavelet packet transform; For each of the sub-signal components, time-domain statistical features, frequency-domain features, and nonlinear entropy features are extracted respectively; The time-domain statistical features, frequency-domain features, and nonlinear entropy features corresponding to all sub-signal components are fused to generate the multi-scale feature set.

3. The fault self-diagnosis system for low-voltage electrical equipment according to claim 2, characterized in that, The time-domain statistical features include mean, variance, kurtosis, and skewness; the frequency-domain features include dominant frequency, spectral centroid, and subband energy ratio; and the nonlinear entropy features include approximate entropy and sample entropy.

4. The fault self-diagnosis system for low-voltage electrical equipment according to claim 3, characterized in that, The specific process of using a causal discovery algorithm to filter features from the multi-scale feature set to obtain a causal association feature set includes: Each feature in the multi-scale feature set is used as a variable node to construct an initial completely undirected graph; Based on the Granger causality test or conditional independence test, the edges in the initial completely undirected graph are pruned to eliminate edges without direct causal dependencies, forming a causal network skeleton graph. Based on the causal direction determination rule, the direction of the edges in the causal network skeleton graph is inferred to form a directed causal graph; From the directed causal graph, feature variables that have a direct causal relationship with the category variables representing the equipment fault state are selected to form the causal association feature set.

5. The fault self-diagnosis system for electrical equipment in low-voltage engineering according to claim 4, characterized in that, The specific process for constructing the static threshold baseline includes: For each feature in the causal correlation feature set, the statistical distribution of the feature in historical normal state data is calculated, and the upper and lower limits of the normal fluctuation of the feature value are determined based on the preset confidence level to form a static threshold baseline.

6. The fault self-diagnosis system for electrical equipment in low-voltage engineering according to claim 4, characterized in that, The process of constructing a dynamic trend baseline includes: For each feature in the causal correlation feature set, calculate multiple trend indicator sequences for that feature within a continuous time window; then, combine the multiple trend indicators by weighted summation into a comprehensive trend indicator. Based on the comprehensive trend indicator sequence under historical normal conditions, the allowable range of the comprehensive trend indicator and the duration threshold of continuous deviation are determined to form a dynamic trend baseline.

7. The fault self-diagnosis system for electrical equipment in low-voltage engineering according to claim 1, characterized in that, The comparison and conformity calculation performed by the fault analysis module specifically includes the following parallel processes: Threshold compliance calculation: The causal correlation feature value extracted in real time is compared with the normal fluctuation range of the corresponding feature in the static threshold baseline, and the deviation of the feature value is calculated. Trend conformity calculation: Calculate the trend index within a preset time window based on the real-time causal correlation feature value, and compare it with the corresponding trend allowable range and duration threshold in the dynamic trend baseline to determine whether the trend conforms to expectations.

8. The fault self-diagnosis system for electrical equipment in low-voltage engineering according to claim 7, characterized in that, When the fault analysis module compares and calculates the conformity of real-time features with the equipment health status judgment baseline, it executes the following diagnostic rules: If the real-time causal correlation feature value exceeds the corresponding range of the static threshold baseline, then the corresponding node device is determined to have a manifest fault. If the real-time causal correlation feature value does not exceed the static threshold baseline range, but the trend of the real-time causal correlation feature value continues to deviate from the dynamic trend baseline, then the corresponding node device is determined to have a latent gradual failure.

9. The fault self-diagnosis system for electrical equipment in low-voltage engineering according to claim 8, characterized in that, The process by which the fault early warning module locates faulty or abnormal nodes includes: The system has a pre-set mapping library of causal relationship features and equipment components or operating procedures; When a fault is determined, the mapping relationship library is queried based on the specific causal relationship features that trigger the explicit fault or the latent gradual fault, and at least one candidate abnormal equipment component or operating link is output as the location result.

Images