Fault monitoring method and system for power supply equipment

Abstract

The invention discloses a fault monitoring method and system for power supply equipment, and the method comprises the following steps: 1, collecting the time-space correlation electric parameters of the power supply equipment in real time through an edge calculation node disposed at the side of the power supply equipment; the time-space correlation electric parameters comprise a steady state quantity, a transient state quantity and a zero sequence component, and marking microsecond-level precision timestamps on all the collected data; step 2, carrying out real-time analysis on a transient state quantity in the time-space correlation electric parameter to generate a transient state characteristic value; and step 3, performing real-time comparison on the steady-state quantity in the time-space correlation electric parameter and a preset dynamic threshold value, and generating a first-level early warning signal when the data exceed the threshold value. Through the implementation of the invention, the full-spectrum monitoring of the power supply equipment from the steady state abnormity to the transient state fault is realized. According to the scheme, the fault detection precision is remarkably improved, full-life-cycle health state monitoring from emergency limit exceeding to early trend abnormity is covered, and false alarm and missing alarm are effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of fault monitoring technology, and in particular to a fault monitoring method and system for power supply equipment. Background Technology

[0002] Power supply equipment is a crucial facility in a power system responsible for converting and distributing electrical energy generated by power plants to various users. It encompasses a wide range of equipment, from high-voltage transmission lines to low-voltage distribution networks. In modern power grids, power supply equipment mainly includes transformers, circuit breakers, switchgear, cables, and various protective devices. This equipment ensures the safe and stable transmission of electrical energy to end users, supporting the electricity needs of industrial production and daily life.

[0003] However, with the expansion of the power grid and the widespread application of power electronic equipment, higher demands are placed on the reliability and stability of power supply equipment. Any equipment failure can lead to large-scale power outages or damage to other related equipment, causing huge economic losses and social impacts. Therefore, how to effectively monitor the operating status of power supply equipment and promptly detect potential faults has become a significant challenge for the power industry.

[0004] Traditional fault monitoring methods have significant limitations: First, they often focus on monitoring the effective values ​​of basic electrical parameters such as voltage and current, while neglecting key indicators such as harmonic content, zero-sequence components, and phase angles, resulting in insufficient detection capabilities for early-stage latent faults such as arc faults and insulation aging. Second, the timestamp accuracy of monitoring data is usually only at the second level, which cannot support accurate phase relationship analysis and transient process capture, greatly reducing the accuracy of multi-parameter correlation diagnosis. Third, early warning mechanisms often rely on fixed thresholds, making it difficult to adapt to complex and changing operating conditions, and easily leading to false alarms or missed alarms. Finally, there is a lack of effective multi-dimensional data fusion and intelligent analysis methods, making it impossible to extract deep fault features from massive amounts of data and achieve accurate identification and prediction of fault modes.

[0005] In summary, there is an urgent need for a fault monitoring method and system for power supply equipment to address the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a fault monitoring method and system for power supply equipment, aiming to solve problems such as missing feature dimensions, insufficient analysis accuracy, rigid early warning mechanisms, and limited diagnostic capabilities.

[0007] On the one hand, this application provides a fault monitoring method for power supply equipment, comprising the following steps:

[0008] Step 1: Collect spatiotemporal correlated electrical parameters of the power supply equipment in real time through edge computing nodes deployed on the power supply equipment side; the spatiotemporal correlated electrical parameters include steady-state quantities, transient quantities and zero-sequence components, and add microsecond-level precision timestamps to all collected data;

[0009] Step 2: Perform real-time analysis on the transient quantities in the spatiotemporally correlated electrical parameters to generate transient characteristic values;

[0010] Step 3: Compare the steady-state quantities in the spatiotemporal correlated electrical parameters with the preset dynamic thresholds in real time, and generate a first-level warning signal when the data exceeds the thresholds;

[0011] Step 4: Store historical spatiotemporal correlated electrical parameters with timestamps of microsecond precision, and perform trend analysis based on their time series characteristics. When a sudden data change is detected or an abnormal trend is predicted, a secondary early warning signal is generated.

[0012] Step 5: Construct a fault diagnosis model, perform multi-dimensional fusion analysis on the spatiotemporal correlated electrical parameters or their extracted features, identify fault modes and generate three-level early warning signals;

[0013] Step 6: Integrate the Level 1, Level 2, and Level 3 early warning signals to generate tiered early warning information, which is then output to the human-machine interface and uploaded to the remote monitoring center via the communication module. Through microsecond-level precision spatiotemporal correlation electrical parameter acquisition, multi-dimensional feature fusion analysis, and a three-level tiered early warning mechanism, full-spectrum monitoring of power supply equipment, from steady-state anomalies to transient faults, is achieved. This solution significantly improves fault detection accuracy, covering the entire lifecycle health status monitoring from emergency over-limit to early trend anomalies, effectively reducing false alarms and missed alarms, and providing probabilistic diagnostic support for operation and maintenance decisions.

[0014] Furthermore, in step 2, the transient quantities in the spatiotemporally correlated electrical parameters are analyzed in real time to generate transient characteristic values. Specifically, this involves performing harmonic analysis based on the current and voltage waveform data aligned with the microsecond-level timestamps to generate harmonic characteristic values ​​that include the content and phase of each harmonic.

[0015] Furthermore, the harmonic analysis specifically includes:

[0016] Perform an FFT transform on the denoised current data to generate the current harmonic spectrum;

[0017] Perform an FFT transform on the denoised voltage data to generate the voltage harmonic spectrum;

[0018] Based on the phase information of the harmonic spectrum, a preliminary determination of the direction of the harmonic source is made.

[0019] Furthermore, in step 3, the dynamic threshold is dynamically adjusted by the cloud platform based on historical load data and environmental factors through a machine learning model and then distributed to the edge computing nodes.

[0020] Furthermore, the method for adjusting the dynamic threshold includes:

[0021] The cloud platform aggregates historical operational data from multiple edge nodes, using load rate, ambient temperature, and time as features, and voltage and current safe operating limits as labels, to train a random forest regression model.

[0022] The model parameters are distributed to each edge computing node, and the edge computing nodes calculate the dynamic threshold at the current moment based on the input features of the real-time operating conditions.

[0023] Furthermore, in step 4, detecting data mutations or predicting trend anomalies includes:

[0024] Calculate the approximate entropy value of current or voltage data within adjacent fixed-length sliding time windows;

[0025] When the rate of change of approximate entropy exceeds a set threshold, a level 2 warning signal is triggered.

[0026] Furthermore, step 5, performing multi-dimensional fusion analysis on the spatiotemporally correlated electrical parameters or their extracted features, includes the following steps:

[0027] Step 5.1: Feature extraction and alignment. Steady-state features, transient features, and zero-sequence features are extracted from the spatiotemporally correlated electrical parameters with microsecond-level timestamps. Based on the timestamps, all features are strictly aligned on the time axis to form a unified multidimensional feature vector.

[0028] Step 5.2: Feature fusion and dimensionality reduction. The multidimensional feature vector is input into the feature fusion network to calculate the correlation weights between different feature dimensions and output a fused high-dimensional feature tensor. Then, a dimensionality reduction algorithm is used to process the tensor to obtain a low-dimensional sensitive feature set for fault classification.

[0029] Step 5.3: Fault mode identification, input the low-dimensional sensitive feature set into the pre-trained fault classification model, and output the probability of the power supply equipment being in various fault modes;

[0030] Step 5.4: Early warning decision-making. Based on the fault probability and the preset early warning strategy, generate and output a three-level early warning signal.

[0031] Furthermore, fault monitoring methods also include:

[0032] Calculate the negative-sequence current and zero-sequence current imbalance of the system;

[0033] By combining harmonic content characteristics, a pre-trained gradient boosting decision tree model is used to diagnose abnormal load types or equipment insulation aging faults.

[0034] On the other hand, this application provides a fault monitoring system for power supply equipment, comprising:

[0035] Edge intelligent sensing terminals, deployed at the power supply equipment site, include a high-precision sensor array, signal conditioning circuit, embedded processor, and edge AI inference module, used to perform data acquisition, preprocessing, and local real-time analysis;

[0036] The cloud-edge collaborative analysis platform includes a cloud platform and the edge intelligent sensing terminal. The cloud platform is used to receive data from multiple terminals, perform big data mining, model training and dynamic threshold calculation, and distribute the model and threshold to the edge terminal. The edge terminal is used to perform real-time inference and early warning.

[0037] Human-computer interaction terminal, used to receive and visualize early warning information, historical curves and fault diagnosis reports;

[0038] The blockchain evidence storage module is used to upload key early warning events and fault recording data to the blockchain network for evidence storage.

[0039] Furthermore, the embedded processor in the edge intelligent sensing terminal is specifically used for:

[0040] Run a lightweight AI inference engine, load the fault classification model and / or trend prediction model issued by the cloud platform, and perform online analysis on the real-time data stream.

[0041] The substantial effects of this invention:

[0042] 1. This invention achieves full-spectrum monitoring of power supply equipment from steady-state anomalies to transient faults by acquiring spatiotemporally correlated electrical parameters with microsecond-level precision, performing multi-dimensional feature fusion analysis, and implementing a three-level hierarchical early warning mechanism. This solution significantly improves fault detection accuracy, covers the entire lifecycle health status monitoring from emergency over-limit to early trend anomalies, effectively reduces false alarms and missed alarms, and provides probabilistic diagnostic support for operation and maintenance decisions.

[0043] 2. In this invention, by performing harmonic analysis on transient quantities and generating harmonic feature values ​​containing phase information, it is possible not only to determine whether harmonics exceed the standard, but also to preliminarily determine the direction of harmonic sources, providing a key basis for power quality management and fault tracing. Furthermore, by using a cloud platform and a machine learning model to dynamically adjust the threshold and distribute it to the edge, the threshold can be adaptively adjusted according to changes in load, environment, and other operating conditions, overcoming the shortcomings of poor adaptability of fixed thresholds and significantly improving the accuracy and reliability of early warning.

[0044] 3. In this invention, by using approximate entropy and its rate of change as indicators for detecting trend mutations, it is extremely sensitive to random components and mutation anomalies in signals, and can effectively capture early and weak fault signs that are easily missed by traditional methods.

[0045] 4. In this invention, through a standardized process of feature extraction, fusion, dimensionality reduction and then pattern recognition, the effective fusion and utilization of multi-dimensional heterogeneous features is achieved, enabling the fault diagnosis model to learn deep and complex fault modes, which greatly improves the diagnostic accuracy and reliability of complex faults. Attached Figure Description

[0046] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of the method flow in Example 1.

[0048] Figure 2 This is a schematic diagram of real-time system fault monitoring in Example 2.

[0049] Figure 3 This is a schematic diagram of the system voltage harmonics display in Example 2.

[0050] Figure 4 This is a schematic diagram of the system current harmonics display in Example 2. Detailed Implementation

[0051] To facilitate understanding of the present invention, a more detailed description is provided below with reference to the accompanying drawings and specific embodiments. It should be noted that when an element is described as being "fixed to" another element, it can be directly on the other element, or one or more intermediate elements may exist between them. When an element is described as being "connected to" another element, it can be directly connected to the other element, or one or more intermediate elements may exist between them. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this specification are for illustrative purposes only.

[0052] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0053] Example 1:

[0054] Reference Figure 1 As shown, a fault monitoring method for power supply equipment includes the following steps:

[0055] Step 1: By deploying edge computing nodes on the power supply equipment side, the spatiotemporal correlated electrical parameters of the power supply equipment are collected in real time. The spatiotemporal correlated electrical parameters include steady-state quantities, transient quantities, and zero-sequence components. All collected data are timestamped with microsecond precision, constructing a complete "spatiotemporal correlated electrical parameter" dataset. This solves the problem of single feature dimensions in traditional solutions and provides a data foundation for full-spectrum monitoring from steady-state anomalies to transient faults. The microsecond-level timestamps ensure the accuracy of the phase relationship between different parameters, making harmonic analysis, vector calculation, and multi-parameter correlation diagnosis possible, and improving the depth and accuracy of fault analysis.

[0056] Step 2: Perform real-time analysis on transient quantities in spatiotemporally correlated electrical parameters to generate transient characteristic values; the generated transient characteristic values ​​provide a quantitative diagnostic basis for detecting early, latent, and waveform quality-related faults;

[0057] Step 3: Compare the steady-state quantities in the spatiotemporal correlated electrical parameters with the preset dynamic thresholds in real time, and generate a first-level warning signal when the data exceeds the thresholds;

[0058] Step 4: Store historical spatiotemporal correlated electrical parameters with timestamps of microsecond precision, and perform trend analysis based on their time series characteristics. When a sudden data change is detected or an abnormal trend is predicted, a secondary early warning signal is generated.

[0059] Step 5: Construct a fault diagnosis model, perform multi-dimensional fusion analysis on spatiotemporally correlated electrical parameters or their extracted features, identify fault modes and generate three-level early warning signals;

[0060] Step 6: Integrate Level 1, Level 2, and Level 3 early warning signals to generate tiered early warning information, which is then output to the human-machine interface and uploaded to the remote monitoring center via the communication module. This achieves full lifecycle health status monitoring of power supply equipment, from "emergency over-limit" to "early trend anomaly" and then to "fault mode identification." The tiered early warning mechanism has a fast response speed, covers all fault types, and significantly reduces false alarms and missed alarms. The final output tiered early warning information not only includes the alarm level but also the probabilistic diagnostic results of the fault mode, providing direct decision support for operation and maintenance personnel and significantly reducing troubleshooting time and manpower costs.

[0061] As one implementation method, steady-state quantities include three-phase voltage, three-phase current, and line voltage, which represent the basic parameters of the power system during normal operation and are used to judge steady-state anomalies such as overvoltage, overcurrent, and imbalance; they also include active power, reactive power, apparent power, and power factor, which characterize energy transmission and quality and are used to evaluate equipment load conditions and energy efficiency.

[0062] As one implementation method, transient quantities represent high-frequency components in a signal and are used to detect transient or early latent faults such as electric arcs, insulation aging, and power electronic equipment failures; including harmonic content characteristic values ​​(generated by FFT).

[0063] As one implementation method, the zero-sequence component is a direct and sensitive indicator of neutral point-related faults in a three-phase system, used to monitor three-phase imbalance, third harmonic superposition, and grounding faults, specifically the neutral current.

[0064] As one implementation method, the edge computing nodes use industrial-grade equipment with built-in BeiDou modules to ensure microsecond-level time synchronization accuracy and a data acquisition sampling rate of no less than 10kHz.

[0065] As one implementation method, in step 2, the transient quantities in the spatiotemporally correlated electrical parameters are analyzed in real time to generate transient characteristic values. Specifically, harmonic analysis is performed on the current and voltage waveform data aligned with microsecond-level timestamps to generate harmonic characteristic values ​​containing the content and phase of each harmonic.

[0066] As one implementation method, harmonic analysis specifically includes:

[0067] Perform an FFT transform on the denoised current data to generate the current harmonic spectrum;

[0068] Perform an FFT transform on the denoised voltage data to generate the voltage harmonic spectrum;

[0069] Based on the phase information of the harmonic spectrum, a preliminary determination of the direction of the harmonic source is made.

[0070] In one implementation, in step 3, the dynamic threshold is dynamically adjusted by the cloud platform based on historical load data and environmental factors through a machine learning model and then distributed to the edge computing nodes.

[0071] As one implementation method, the dynamic threshold adjustment method includes:

[0072] The cloud platform aggregates historical operational data from multiple edge nodes, using load rate, ambient temperature, and time as features, and voltage and current safe operating limits as labels, to train a random forest regression model.

[0073] The model parameters are distributed to each edge computing node, and the edge computing nodes calculate the dynamic threshold at the current moment based on the input features of the real-time operating conditions.

[0074] As one implementation method, step 4, detecting data mutations or predicting trend anomalies, includes:

[0075] Calculate the approximate entropy value of current or voltage data within adjacent fixed-length sliding time windows;

[0076] When the rate of change of approximate entropy exceeds a set threshold, a level 2 warning signal is triggered.

[0077] As one implementation method, step 5, which involves multi-dimensional fusion analysis of spatiotemporally correlated electrical parameters or their extracted features, includes the following steps:

[0078] Step 5.1: Source Extraction: Extract the following from the raw data with microsecond-level timestamps:

[0079] Steady-state characteristics, such as the effective values ​​of three-phase voltage and current, active / reactive / apparent power values, power factor, etc.

[0080] Transient characteristics, such as the amplitude and phase of each harmonic (3rd order) obtained by FFT, total harmonic distortion (THD), etc.

[0081] Zero-sequence characteristics, such as the effective value of the neutral current, the content of fundamental and third harmonic zero-sequence currents, etc.

[0082] Time alignment: Using microsecond-level timestamps as the sole time reference, the features extracted from different physical quantities are strictly aligned on the time axis to ensure that each data point represents the system state at the same moment. Finally, all features are combined into a unified multidimensional feature vector.

[0083] Step 5.2: Feature Fusion and Dimensionality Reduction. The multidimensional feature vectors obtained in Step 5.1 are input into a feature fusion network (a neural network using an attention mechanism). This network automatically calculates the importance (association weight) of different features to the current diagnostic task and performs weighted fusion accordingly, outputting a high-dimensional feature tensor that comprehensively reflects the system state. A dimensionality reduction algorithm (Principal Component Analysis, PCA) is used to process the fused high-dimensional feature tensor. While preserving key information to the maximum extent, the data is compressed into a low-dimensional space, resulting in a low-dimensional sensitive feature set. This effectively prevents model overfitting and improves subsequent computational efficiency.

[0084] Step 5.3: Fault Mode Recognition. The low-dimensional sensitive feature set obtained in Step 5.2 is input into the pre-trained fault classification model, which outputs the probability of the power supply equipment being in various fault modes. The fused and dimensionality-reduced features are then input into the classification model (neural network), which outputs a probability distribution, such as: [Motor bearing damage: 85%, Capacitor failure: 10%, Normal: 5%]. This achieves a leap from "alarm" to "diagnosis," not only identifying a problem but also determining the most likely problem in probabilistic terms, thus improving operational efficiency.

[0085] Step 5.4: Early Warning Decision. Based on the fault probabilities obtained in Step 5.3 and combined with the preset early warning strategy, a three-level early warning signal is generated and output. Decisions are executed based on the probability output. For example, a rule can be set: when the probability of any fault mode exceeds 80%, a three-level early warning is immediately generated; if the probability is between 50% and 80%, it is marked as "to be observed," and the monitoring frequency is increased. This makes the early warning logic more intelligent and flexible, reduces false alarms, and increases the reliability of the system.

[0086] As one implementation method, the feature fusion network in step S5.2 adopts an attention mechanism to enable the model to dynamically focus on the feature dimension most relevant to the current fault, and the dimensionality reduction algorithm selected is the t-SNE algorithm.

[0087] As one implementation method, the fault classification model in step S5.3 selects a one-dimensional convolutional neural network (1D-CNN) algorithm. The model is trained on the cloud platform using historical fault data and converted into a lightweight format (such as TensorFlowLite) and distributed to the edge nodes.

[0088] As one implementation method, the fault monitoring method further includes:

[0089] Calculate the negative-sequence current and zero-sequence current imbalance of the system;

[0090] By combining harmonic content characteristics, a pre-trained gradient boosting decision tree model is used to diagnose abnormal load types or equipment insulation aging faults.

[0091] Example 2:

[0092] Reference Figures 2-4 As shown, this embodiment is basically the same as embodiment 1, except that it provides a fault monitoring system for power supply equipment, including:

[0093] Edge intelligent sensing terminals, deployed at the power supply equipment site, include a high-precision sensor array, signal conditioning circuit, embedded processor, and edge AI inference module, used to perform data acquisition, preprocessing, and local real-time analysis;

[0094] The cloud-edge collaborative analysis platform includes a cloud platform and edge intelligent sensing terminals. The cloud platform is used to receive data from multiple terminals, perform big data mining, model training and dynamic threshold calculation, and distribute the models and thresholds to the edge terminals; the edge terminals are used to perform real-time inference and early warning.

[0095] Human-computer interaction terminal, used to receive and visualize early warning information, historical curves and fault diagnosis reports;

[0096] The blockchain evidence storage module is used to upload key early warning events and fault recording data to the blockchain network for evidence storage, ensuring the immutability and traceability of the data.

[0097] As one implementation, the high-precision sensor array includes voltage sensors with ±0.2% accuracy and current transformers with ±0.5% accuracy to ensure the accuracy of the raw data.

[0098] As one implementation method, in the cloud-edge collaborative analysis platform, the cloud platform adopts a Kubernetes-based microservice architecture to elastically support data access and model training tasks for a large number of edge terminals; the edge terminals and the cloud platform communicate using the MQTT protocol to ensure the reliability and real-time performance of data transmission.

[0099] As one implementation method, the human-computer interaction terminal provides a web-based graphical interface. Its warning-driven interface redirection logic is as follows: when a level three warning (harmonic exceedance) is received, the interface automatically redirects from the main dashboard to... Figure 3 , Figure 4 The harmonic spectrum analysis interface is shown. When a level 2 warning (trend abrupt change) is received, a historical curve comparison window will automatically pop up. When a level 1 warning (threshold exceeding the limit) is received, an audible and visual alarm will be triggered and the alarm information will be displayed in a prominent position on the interface.

[0100] As one implementation method, the embedded processor in the edge intelligent sensing terminal is specifically used for:

[0101] Run a lightweight AI inference engine, load fault classification models and / or trend prediction models distributed by the cloud platform, and perform online analysis of real-time data streams.

[0102] It should be noted that while the preferred embodiments of the present invention are provided in the specification and accompanying drawings, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are not intended to impose additional limitations on the content of the present invention; their purpose is to provide a more thorough and comprehensive understanding of the disclosure of the present invention. Furthermore, the above-described technical features can be combined with each other to form various embodiments not listed above, all of which are considered to be within the scope of the present invention specification. Moreover, those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A fault monitoring method for power supply equipment, characterized in that, Includes the following steps: Step 1: Collect the spatiotemporal correlation electrical parameters of the power supply equipment in real time through edge computing nodes deployed on the power supply equipment side; The spatiotemporal correlated electrical parameters include steady-state quantities, transient quantities, and zero-sequence components, and all collected data are timestamped with microsecond-level precision. Step 2: Perform real-time analysis on the transient quantities in the spatiotemporally correlated electrical parameters to generate transient characteristic values; Step 3: Compare the steady-state quantities in the spatiotemporal correlated electrical parameters with the preset dynamic thresholds in real time, and generate a first-level warning signal when the data exceeds the thresholds; Step 4: Store historical spatiotemporal correlated electrical parameters with timestamps of microsecond precision, and perform trend analysis based on their time series characteristics. When a sudden data change is detected or an abnormal trend is predicted, a secondary early warning signal is generated. Step 5: Construct a fault diagnosis model, perform multi-dimensional fusion analysis on the spatiotemporal correlated electrical parameters or their extracted features, identify fault modes and generate three-level early warning signals; Step 6: Integrate the Level 1, Level 2, and Level 3 early warning signals to generate graded early warning information, and output it to the human-machine interface and upload it to the remote monitoring center through the communication module.

2. The fault monitoring method for power supply equipment according to claim 1, characterized in that, In step 2, the transient quantities in the spatiotemporally correlated electrical parameters are analyzed in real time to generate transient characteristic values. Specifically, harmonic analysis is performed based on the current and voltage waveform data aligned with the microsecond-level timestamps to generate harmonic characteristic values ​​containing the content and phase of each harmonic.

3. The fault monitoring method for power supply equipment according to claim 2, characterized in that, The harmonic analysis specifically includes: Perform an FFT transform on the denoised current data to generate the current harmonic spectrum; Perform an FFT transform on the denoised voltage data to generate the voltage harmonic spectrum; Based on the phase information of the harmonic spectrum, a preliminary determination of the direction of the harmonic source is made.

4. The fault monitoring method for power supply equipment according to claim 1, characterized in that, In step 3, the dynamic threshold is dynamically adjusted by the cloud platform based on historical load data and environmental factors through a machine learning model and then distributed to the edge computing nodes.

5. The fault monitoring method for power supply equipment according to claim 4, characterized in that, The method for adjusting the dynamic threshold includes: The cloud platform aggregates historical operational data from multiple edge nodes, using load rate, ambient temperature, and time as features, and voltage and current safe operating limits as labels, to train a random forest regression model. The model parameters are distributed to each edge computing node, and the edge computing nodes calculate the dynamic threshold at the current moment based on the input features of the real-time operating conditions.

6. The fault monitoring method for power supply equipment according to claim 1, characterized in that, In step 4, detecting data mutations or predicting trend anomalies includes: Calculate the approximate entropy value of current or voltage data within adjacent fixed-length sliding time windows; When the rate of change of approximate entropy exceeds a set threshold, a level 2 warning signal is triggered.

7. The fault monitoring method for power supply equipment according to claim 1, characterized in that, Step 5, which involves multi-dimensional fusion analysis of the spatiotemporally correlated electrical parameters or their extracted features, includes the following steps: Step 5.1: Feature extraction and alignment. Steady-state features, transient features, and zero-sequence features are extracted from the spatiotemporally correlated electrical parameters with microsecond-level timestamps. Based on the timestamps, all features are strictly aligned on the time axis to form a unified multidimensional feature vector. Step 5.2: Feature fusion and dimensionality reduction. The multidimensional feature vector is input into the feature fusion network to calculate the correlation weights between different feature dimensions and output a fused high-dimensional feature tensor. Then, a dimensionality reduction algorithm is used to process the tensor to obtain a low-dimensional sensitive feature set for fault classification. Step 5.3: Fault mode identification, input the low-dimensional sensitive feature set into the pre-trained fault classification model, and output the probability of the power supply equipment being in various fault modes; Step 5.4: Early warning decision-making. Based on the fault probability and the preset early warning strategy, generate and output a three-level early warning signal.

8. The fault monitoring method for power supply equipment according to claim 1, characterized in that, Also includes: Calculate the negative-sequence current and zero-sequence current imbalance of the system; By combining harmonic content characteristics, a pre-trained gradient boosting decision tree model is used to diagnose abnormal load types or equipment insulation aging faults.

9. A fault monitoring system for power supply equipment, used to implement the fault monitoring method for power supply equipment as described in any one of claims 1-8, characterized in that, include: Edge intelligent sensing terminals, deployed at the power supply equipment site, include a high-precision sensor array, signal conditioning circuit, embedded processor, and edge AI inference module, used to perform data acquisition, preprocessing, and local real-time analysis; The cloud-edge collaborative analysis platform includes a cloud platform and the edge intelligent sensing terminal. The cloud platform is used to receive data from multiple terminals, perform big data mining, model training and dynamic threshold calculation, and distribute the model and threshold to the edge terminal. The edge terminal is used to perform real-time inference and early warning. Human-computer interaction terminal, used to receive and visualize early warning information, historical curves and fault diagnosis reports; The blockchain evidence storage module is used to upload key early warning events and fault recording data to the blockchain network for evidence storage.

10. A fault monitoring system for power supply equipment according to claim 9, characterized in that, The embedded processor in the edge intelligent sensing terminal is specifically used for: Run a lightweight AI inference engine, load the fault classification model and / or trend prediction model issued by the cloud platform, and perform online analysis on the real-time data stream.

Images