Intelligent detection and fault early warning method and system for ring main unit

Abstract

The invention relates to a ring main unit intelligent detection and fault early warning method and system. The method comprises the following steps: deploying an ultrahigh frequency sensor, an infrared thermal imager, an ultraviolet sensor and a high-frequency current transformer, and synchronously acquiring multi-modal data such as partial discharge, temperature distribution, corona discharge and grounding current of the ring main unit; carrying out feature extraction on the collected data, wherein the feature extraction comprises a time-frequency statistical feature and a temperature-load correlation feature of the partial discharge signal; the Dempster-Shafer evidence theory is utilized to carry out fusion calculation on the multi-source feature evidence, and confidence coefficients for insulation degradation, poor contact, arc faults and other fault types are obtained; arc fault prediction is carried out in combination with a long-short-term memory network model; and finally executing a grading early warning strategy according to the fault confidence. The problems of low detection reliability and fault early warning lag of a single sensor are solved, early accurate diagnosis and predictive maintenance of the insulation fault of the ring main unit are realized, and the operation safety and reliability of power equipment are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power equipment state monitoring and fault diagnosis, and particularly relates to an intelligent detection and fault early warning method and system for ring network boxes. BACKGROUND

[0002] A ring network box is a key device in a power distribution network system, and undertakes important functions such as power distribution, line control and protection. As a mainstream product of 10kV voltage level in China, the operation reliability of the ring network box is directly related to the safety and stability of the power grid. However, in the long-term operation process, the ring network box is affected by multiple factors such as power, heat, mechanical and environmental stress, and the internal insulation parts, conductive connections, circuit breaker mechanisms and other key components will gradually deteriorate, eventually leading to faults. Common fault types include insulation breakdown, poor contact overheating, arc fault and the like, which often cause equipment damage, power outages and even personal safety accidents.

[0003] At present, the state monitoring of the ring network box mainly has the following technical problems: Single detection means, insufficient reliability: the existing technology mainly relies on periodic inspection or online monitoring of a single parameter (such as temperature or partial discharge). The information obtained by a single sensor is limited, and is easily affected by electromagnetic interference and environmental changes, resulting in high false alarm rate and missed alarm rate, and it is difficult to comprehensively and accurately reflect the true health status of the equipment.

[0004] Fault diagnosis lags behind, and early warning capability is weak: the traditional method usually alarms when the fault characteristics are very obvious (such as a significant increase in temperature and a large amount of discharge), and lacks the ability to capture and correlate early and weak fault characteristics, so it cannot achieve early warning and prediction of faults, and misses the best opportunity for maintenance intervention.

[0005] Lack of intelligent analysis, relying on manual experience: most monitoring systems only complete data acquisition and simple threshold alarm, and lack algorithms for deep fusion and intelligent analysis of multi-source heterogeneous data, so the fault diagnosis is severely dependent on the personal experience of the operation and maintenance personnel, which is low in efficiency and poor in consistency.

[0006] Unable to identify complex faults: ring network box faults are often complex faults caused by multiple factors and mutual correlation (such as poor contact leading to overheating, which in turn accelerates insulation deterioration). The existing method is difficult to distinguish and identify such complex fault modes.

[0007] Therefore, there is an urgent need in the art for an intelligent monitoring scheme that can fuse multi-source information, achieve early and accurate diagnosis and predictive early warning, in order to improve the operation reliability and intelligent level of operation and maintenance of the ring network box. SUMMARY

[0008] The primary objective of this invention is to overcome the shortcomings of existing technologies and provide a method and system for intelligent detection and fault analysis of ring main units with high diagnostic accuracy and early warning time. Specific objectives include: It enables early and accurate detection of various faults in ring main units, such as insulation degradation, overheating, and arcing risks.

[0009] By fusing information from multiple sensors, the reliability and anti-interference ability of diagnostic results are significantly improved.

[0010] Establish a data-driven fault prediction model to realize the transformation from "post-event maintenance" to "predictive maintenance".

[0011] Provides clear tiered early warning and decision support to guide on-site operation and maintenance work.

[0012] To achieve the above objectives, the present invention adopts the following technical solution: A method for intelligent detection and fault early warning of ring network cages includes the following steps: Step S100: Simultaneously collect multimodal time-series data characterizing the operating status of the ring network box by using multiple heterogeneous sensors deployed inside the ring network box; the multimodal data includes at least: partial discharge ultra-high frequency signal, infrared temperature signal of key components, corona discharge ultraviolet signal and grounding wire high frequency current signal; Step S200: Process the multimodal time series data to extract multi-dimensional feature vectors related to insulation faults; the multi-dimensional feature vectors include time-frequency statistical features extracted from the partial discharge UHF signal, and correlation features based on load current and measured temperature rise. Step S300: Input the multi-dimensional feature vector into the fusion diagnostic model based on Dempster-Shafer evidence theory for calculation to obtain the confidence level for multiple preset fault types; the preset fault types include at least: insulation degradation, poor contact and arcing fault; Step S400: Execute a graded early warning strategy based on the confidence level of the fault type and output a diagnostic report.

[0013] The present invention further specifies that, in step S200, extracting time-frequency statistical features from the partial discharge ultra-high frequency signal specifically involves: The S-transform of the partial discharge ultra-high frequency signal is performed to obtain the time-frequency matrix; Statistical features, including skewness, kurtosis, and discharge repetition rate, are extracted from the time-frequency matrix.

[0014] The present invention further specifies that, in step S200, the correlation feature between the load current and the measured temperature rise is obtained through the following method: Based on the real-time collected load current I, and using the thermal model formula... Calculate the theoretical temperature rise ; Calculate the measured temperature rise With the theoretical temperature rise absolute difference ; The difference As the associated feature, I is the real-time monitored load current. The thermal resistance of the circuit. is the thermal time constant.

[0015] The present invention further includes step S300, which specifically includes: S301, Construct an identification framework that includes all preset fault types. ; S302, establish a basic probability allocation function for each sensor data source, and map the extracted feature vectors to the recognition framework. Reliability distribution of each proposition; S303 uses the Dempster combination rule to fuse the basic probability assignment functions from different sensor data sources. The calculation formula is as follows: for ,have: in, Assign values ​​to the basic probabilities of proposition A after fusion. and The base probabilities from different sources of evidence are assigned values, where K is the conflict factor and 1-K is the normalization factor.

[0016] The present invention further includes step S300, which further includes arc fault prediction sub-step S304: inputting the multi-dimensional feature vector sequence within a continuous time window into a pre-trained long short-term memory network prediction model, and outputting the probability value of arc fault occurring in a future period of time. The probability value can be used as an independent piece of evidence in the fusion process of the Dempster combination rule, or it can be compared with the fault confidence obtained after fusion.

[0017] The present invention further provides that the structure of the LSTM prediction model includes: The input layer is used to receive feature sequences with dimensions [time step, number of features]. At least one LSTM layer is used to learn time series dependencies; Dropout layers are used to prevent overfitting; The system consists of a fully connected layer and an output layer. The output layer uses the Sigmoid activation function to output an arc fault probability value between 0 and 1.

[0018] The present invention further specifies that the graded early warning strategy described in step S400 is as follows: When the confidence level of any fault type is within the first threshold range, a level one warning is triggered, local audible and visual alarms are activated, and the warning information is uploaded to the monitoring center. When the confidence level of any fault type is higher than a second threshold that is higher than the upper limit of the first threshold range, a second-level warning is triggered, and a circuit breaker tripping recommendation instruction is generated based on the first-level warning action.

[0019] The present invention also provides an intelligent detection and early warning system for implementing the method, characterized in that it comprises: The data acquisition module includes an ultra-high frequency (UHF) sensor, an infrared thermal imager, an ultraviolet sensor, and a high-frequency current transformer, all arranged within the ring network box. An edge computing gateway, with a built-in processor and memory, configured to perform the steps of the method described in claim 1; The cloud platform or monitoring center communicates with the edge computing gateway to receive, store, and display diagnostic results and early warning information.

[0020] The invention further provides that the edge computing gateway supports the IEC 61850 MMS communication protocol, enabling it to upload diagnostic results and early warning information to the cloud platform or monitoring center in a standard format.

[0021] The present invention further provides that the temperature resolution of the infrared thermal imager is not less than 0.1℃, and the sampling rate of the UHF sensor for the partial discharge ultra-high frequency signal is not less than 100MS / s.

[0022] Beneficial effects of the present invention: Compared with the prior art, the present invention has at least one of the following significant advantages: Diagnostic accuracy and reliability are greatly improved: Through multimodal sensing and information fusion technology, multi-angle and cross-verification of faults is achieved, effectively overcoming the limitations of single sensor monitoring and significantly reducing false alarms and missed alarms.

[0023] It possesses true early warning capabilities: it can capture and correlate subtle, early fault characteristics (such as slight discharge or abnormal temperature rise trends), and combine them with predictive models to issue warnings before the fault causes serious damage, providing a valuable time window for preventive maintenance.

[0024] High level of intelligence: By adopting DS evidence theory and LSTM deep learning algorithm, the system can simulate the comprehensive judgment process of human experts, and has the ability to handle uncertainty and learn the patterns of historical data, thus reducing the reliance on human experience.

[0025] It has diagnostic capabilities for complex faults: The DS evidence theory framework can handle the situation where multiple faults exist at the same time, give the credibility of each fault, and help to analyze the causal relationship between faults.

[0026] The output is actionable: the tiered early warning mechanism and specific maintenance suggestions enable the diagnostic results to directly guide on-site operation and maintenance work, improving operation and maintenance efficiency and economy. Attached Figure Description

[0027] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating an embodiment of the present invention. Detailed Implementation

[0028] The following will describe in detail the implementation of this application with reference to the accompanying drawings and embodiments, so that the implementation process of how this application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0029] The system hardware configuration in this invention is as follows: Sensing layer: Sensors are reasonably arranged in key parts of the ring network box.

[0030] UHF sensors are typically installed with built-in or external antennas to monitor partial discharge throughout the cabinet.

[0031] Miniature infrared thermal imagers are installed at critical connection points and circuit breaker locations to monitor their temperature in real time.

[0032] The ultraviolet sensor is installed near the observation window or in a suitable location inside the cabinet.

[0033] The HFCT is tightly fitted onto the grounding busbar of the ring network box.

[0034] In addition, conventional current transformers and voltage transformers need to be configured to obtain the load current.

[0035] Edge computing unit: An industrial-grade edge computing gateway is installed in the ring main unit's instrument room or a nearby intelligent control cabinet. This gateway should possess the following features: It uses a multi-core ARM or x86 processor with a clock speed of no less than 1GHz.

[0036] Abundant I / O interfaces: Supports analog signals, digital signals, RS485 / 232, Ethernet, etc., for connecting various sensors.

[0037] Sufficient storage space: Built-in storage unit of no less than 8GB for caching temporary data.

[0038] Reliable communication module: Supports 4G / 5G wireless communication and industrial Ethernet, and integrates the IEC 61850 MMS communication protocol stack for interaction with higher-level systems.

[0039] The core algorithm software of this invention runs in this edge computing unit.

[0040] Platform layer: Remote monitoring center or cloud platform, which receives data and diagnostic results uploaded from multiple ring network box edge computing units, and performs centralized display, storage, in-depth analysis and operation and maintenance management.

[0041] The software implementation of the method is as follows: Step S100: Synchronous acquisition of multimodal data Multiple heterogeneous sensors are deployed at key locations within the ring network enclosure to form a distributed sensor network. Multimodal time-series data characterizing the equipment's operating status are synchronously acquired at a set sampling frequency. The multimodal data includes at least: Partial discharge ultra-high frequency signal: Acquired by a UHF sensor installed inside the air chamber of the ring network box, with a center frequency typically ranging from 500MHz to 1.5GHz and a sampling rate of no less than 100MS / s, used to capture electromagnetic pulses generated by partial discharge occurring inside or on the surface of insulating materials.

[0042] Infrared temperature signals are acquired by infrared thermal imagers or infrared temperature sensors installed inside the cabinet and aimed at easily overheated areas such as circuit breaker contacts, busbar connection points, and cable joints. The thermal imager has a temperature resolution of no less than 0.1℃ and a data update frequency of no less than 1Hz, and is used to obtain absolute temperatures and temperature distribution images of key points.

[0043] Corona discharge ultraviolet signal: Acquired by an ultraviolet sensor, with a sensitive wavelength in the solar blind ultraviolet region (240-280nm), used to detect corona discharge phenomena caused by electric field concentration.

[0044] High-frequency current signal of grounding wire: Acquired by a high-frequency current transformer connected to the grounding lead of the ring network box. The bandwidth is usually 1Hz-30MHz. It is used to detect the current signal flowing into the grounding wire from partial discharge, as a supplement and verification of the UHF signal.

[0045] Step S200: Multi-dimensional fault feature extraction After performing signal preprocessing (such as noise reduction, filtering, and alignment) on the raw data collected in step S100, multi-dimensional feature vectors related to potential faults are extracted. The feature extraction includes: S201: Partial Discharge Feature Extraction: Time-frequency analysis of the UHF signal. Preferably, the S-transform (Stockwell Transform) algorithm is used to convert the one-dimensional time-domain signal into a time-frequency matrix. The formula for the S-transform is: Where x(t) is the original signal, τ is the time shift factor, and f is the frequency. From the obtained time-frequency matrix, the following statistical characteristics are further calculated: Discharge amplitude skewness, used to characterize the asymmetry of the discharge pulse amplitude distribution, is calculated using the following formula: in Let be the amplitude of the i-th discharge pulse. The average value of all discharge pulse amplitudes The standard deviation of the discharge pulse amplitude This represents the total number of discharge pulses within the statistical window. When its skewness... ≈ 0: The distribution is nearly symmetrical (e.g., normal distribution); skewness >0 (positive bias): The distribution has a long tail on the right side, and there are a few discharge pulses with extremely large amplitudes; bias Less than 0 (negative bias): The distribution has a long tail on the left and contains a large number of small-amplitude discharge pulses. When the internal bubble discharges, it usually shows positive bias with a small number of large-amplitude pulses; when the surface discharges, it may show negative bias with a large number of small-amplitude pulses.

[0046] Discharge amplitude kurtosis: Characterizes the steepness of the discharge pulse amplitude distribution, reflecting the sharpness of the pulse; its calculation formula is as follows: The subtraction of 3 is to make the kurtosis of the normal distribution zero.

[0047] Kudo ≈ 0: The distribution is as steep as the normal distribution; kurtosis >0: The distribution is steeper than a normal distribution; kurtosis <0 (low kurtosis and thin tail): The distribution is flatter than a normal distribution. When there is corona discharge at the tip: the kurtosis value is usually large, the distribution is concentrated, and it has a fat tail characteristic; multiple discharge sources inside the insulation: may lead to a decrease in kurtosis value and a flatter distribution. Discharge repetition rate: The number of discharge pulses exceeding a set threshold per unit time, calculated using the following formula: ;in To ensure that the amplitude exceeds a preset threshold within the measurement time T. The number of discharge pulses; For the statistical time window (usually 1 second or 1 power frequency cycle), the threshold setting method is as follows: Background noise method: ,in The mean of the background noise. The standard deviation of the background noise amplitude; or the fixed percentage method: take 10%-20% of the maximum discharge amplitude as the threshold; a sudden increase in the repetition rate: indicates that the insulation defect is developing rapidly; a high repetition rate: indicates that there is a continuous severe discharge.

[0048] Phase-resolved spectral characteristics: Analyze the distribution pattern of discharge at the power frequency phase and record the amplitude of each discharge pulse. and phase angle (relative to the zero-crossing point of the power frequency voltage) S202: Temperature-Load Correlation Feature Extraction: Establish a dynamic thermal circuit model and calculate the theoretical temperature rise. Theoretical Temperature Rise The calculation formula is as follows: Where I is the load current monitored in real time. The thermal resistance of the circuit (determined by the equipment structure and materials). The thermal time constant is used. The temperature rise measured by the infrared sensor is... and Compare and calculate abnormal temperature rise when If the temperature continues to exceed a set threshold (e.g., 15°C), a strongly correlated fault feature is generated.

[0049] S203: Ultraviolet and High-Frequency Current Feature Extraction: For ultraviolet signals, extract the photon count rate or ultraviolet pulse intensity; for HFCT signals, extract the amplitude, frequency and other features of the pulse current.

[0050] Step S300: Multi-source evidence fusion and intelligent diagnosis This step is the core of the invention. It inputs the multi-dimensional feature vector extracted in step S200 into a fusion diagnostic model based on Dempster-Shafer evidence theory (DS evidence theory), and combines it with a long short-term memory network (LSTM) for prediction, and finally outputs the confidence level of various faults.

[0051] S301: Constructing the recognition framework: Define the set of propositions to be recognized as the recognition framework. For example: ={F0: Normal, F1: Insulation Deterioration, F2: Poor Contact, F3: Floating Discharge, F4: Arc Fault}; S302: Establish a basic probability assignment function: Establish a basic probability assignment function for each sensor data source (evidence source). This function maps the numerical values ​​of the feature vector to a confidence assignment for each proposition (including single-point propositions and compound propositions) in the recognition frame Θ. For example, when UHF features strongly indicate insulation defects, the BPA function might assign: S303: DS Evidence Fusion: This involves pairwise fusion of BPA functions from different evidence sources (such as UHF, infrared, and ultraviolet) using Dempster's combination rules. The combination rules are as follows: for ,have: in, Assign values ​​to the basic probabilities of proposition A after fusion. and Basic probability values ​​are assigned to different sources of evidence, where K is the conflict factor, measuring the degree of conflict between pieces of evidence. 1-K is the normalization factor. Through multiple fusions, the final basic probability values ​​under the combined effect of all evidence are obtained.

[0052] S304: LSTM Arc Fault Prediction: In parallel, a multi-dimensional feature sequence (which may include UHF features, temperature change rate, etc.) spanning a continuous time window (e.g., the past 60 sampling moments) is fed into a pre-trained LSTM network. The LSTM model structure can be designed as follows: Input layer: Receives sequences with dimensions of [time step, number of features].

[0053] LSTM layer: at least one layer, with the number of neurons optionally being 64 or 128, used to learn long-term dependencies in time series.

[0054] Dropout layer: Prevents overfitting.

[0055] Fully connected layer: Maps the LSTM output to the final prediction.

[0056] Output layer: Using the Sigmoid activation function, the output is a scalar between 0 and 1, representing the probability of an arc fault occurring in the future.

[0057] The predicted probability Parc will be used as an independent, future-oriented piece of evidence in the evidence fusion process of step S303, or it will be used in a comprehensive judgment with the fused result.

[0058] Step S400: Fault Early Warning and Decision Support Based on the various fault confidence levels (i.e., the fused basic probability assignments) and arc fault prediction probabilities calculated in step S300, a graded early warning strategy is executed, and a diagnostic report is generated.

[0059] Tiered early warning strategy: Level 1 Warning: Triggered when the confidence level of any fault type falls within a first preset threshold range (e.g., 70% to 90%). The system activates the local audible and visual alarm and transmits the warning information (including fault type, confidence level, and timestamp) to the remote monitoring center via the communication network.

[0060] Level 2 warning: Triggered when the confidence level of any fault type exceeds a second preset threshold (e.g., 90%), or the predicted probability of an arc fault, Parc, exceeds a specific threshold (e.g., 0.9). In addition to fulfilling the Level 1 warning action, the system will also generate a clear circuit breaker tripping suggestion command on the human-machine interface. This command must be confirmed by maintenance personnel before execution, or it may be executed automatically under certain conditions to prevent the accident from escalating.

[0061] Diagnostic report generation: The system automatically generates a detailed report containing the following: Diagnostic conclusions: the most prevalent fault types and their confidence levels.

[0062] Summary of evidence: List the main sensor features and their values ​​that support this conclusion.

[0063] Maintenance recommendations include: "Insulation check recommended within 72 hours" and "Immediately check the tightness of the circuit breaker contacts."

[0064] The invention will now be further illustrated by taking the diagnosis of a potential fault of "insulation degradation accompanied by arc risk" as a specific example: Data Acquisition (S100): The system runs continuously, and each sensor acquires data synchronously. At a certain moment, the UHF sensor captures a series of discharge pulses, and the infrared thermal imager shows a slight abnormal increase in the contact temperature.

[0065] Feature extraction (S200): An S-transform of the UHF signal yielded a skewness of 1.5, a kurtosis of 8.0, and a discharge repetition rate of 150 times / second. These characteristics are consistent with typical features of air gap discharge within insulating materials.

[0066] Based on a real-time current of 1000A, the theoretical temperature rise should be 25K, but the actual measured temperature rise is 38K. (Although it did not exceed the 15K threshold, the trend is obvious.) Evidence fusion and diagnosis (S300): BPA allocation: UHF evidence sources: allocation based on their characteristics Infrared evidence source: This supports the increase in temperature rise caused by increased dielectric loss due to insulation degradation. There may be contact issues. .

[0067] DS fusion: Calculate the conflict factor K according to the combination rules, and then fuse the above BPA. The fused result is... The confidence level for "insulation degradation" is 78%.

[0068] LSTM prediction: Simultaneously, the feature sequence over a period of time is input into the LSTM model, and the output arc fault probability Parc=0.75 is obtained.

[0069] Early warning and decision-making (S400): Because the confidence level of insulation degradation is 78% (>70%), the system triggers a Level 1 warning. Simultaneously, the arc risk probability of 75% is also relatively high, and this is highlighted in the diagnostic report.

[0070] The system generated a report: "Warning: Insulation degradation signs detected (confidence 78%), with a risk of arc fault (probability 75%). It is recommended to conduct a power outage inspection of the busbar support insulators and circuit breaker insulation components in the near future." The local HMI displays a yellow alarm, and the information is uploaded to the monitoring center.

[0071] Algorithm parameter settings S-transform: The frequency range can be set to 0-500MHz, and the time window can be adjusted according to the signal characteristics.

[0072] DS Evidence Theory: The BPA function can be constructed based on an expert experience base or a machine learning classifier (such as a fuzzy inference system) to map the feature vector to a BPA.

[0073] LSTM model: The time step can be set to 60, and the number of features can be 5 (such as UHF amplitude, skewness, kurtosis, temperature, etc.). The model needs to be trained using a large amount of historical data (including data under normal and various failure states) for supervised learning in order to optimize its weight parameters.

[0074] In this invention: diagnostic accuracy and reliability are significantly improved: through multimodal sensing and information fusion technology, multi-angle and cross-verification of faults is achieved, effectively overcoming the limitations of single-sensor monitoring and significantly reducing false alarms and missed alarms. It possesses true early warning capabilities: it can capture and correlate weak, early fault characteristics (such as slight discharge or abnormal temperature rise trends), and combined with predictive models, issue early warnings before the fault causes serious damage, providing a valuable time window for preventative maintenance.

[0075] High level of intelligence: Employing DS evidence theory and LSTM deep learning algorithms, the system can simulate the comprehensive judgment process of human experts, possessing the ability to handle uncertainty and learn patterns from historical data, thus reducing reliance on human experience. Diagnostic capability for complex faults: The DS evidence theory framework effectively handles situations where multiple faults coexist, assigning credibility to each and aiding in the analysis of causal relationships between faults. Output results: A tiered early warning mechanism and specific maintenance recommendations enable diagnostic results to directly guide on-site operation and maintenance work, improving operational efficiency and cost-effectiveness.

[0076] As used in the specification and claims, certain terms refer to specific components. Those skilled in the art will understand that hardware manufacturers may use different names to refer to the same component. This specification and claims do not distinguish components based on differences in name, but rather on differences in function. The term "comprising" throughout the specification and claims is an open-ended term and should be interpreted as "comprising but not limited to." "Approximately" means that within an acceptable margin of error, those skilled in the art can solve the technical problem and substantially achieve the technical effect within a certain margin of error.

[0077] The foregoing description illustrates and describes several preferred embodiments of the present invention. However, as previously stated, it should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for intelligent detection and fault early warning of ring network boxes, characterized in that, Includes the following steps: Step S100: Multimodal time-series data characterizing the operating status of the ring network box are collected synchronously by multiple heterogeneous sensors deployed inside the ring network box. The multimodal data includes at least: partial discharge ultra-high frequency signals, infrared temperature signals of key components, corona discharge ultraviolet signals, and grounding wire high-frequency current signals; Step S200: Process the multimodal time series data to extract multi-dimensional feature vectors related to insulation faults; the multi-dimensional feature vectors include time-frequency statistical features extracted from the partial discharge UHF signal, and correlation features based on load current and measured temperature rise. Step S300: Input the multi-dimensional feature vector into the fusion diagnostic model based on Dempster-Shafer evidence theory for calculation to obtain the confidence level for multiple preset fault types; the preset fault types include at least: insulation degradation, poor contact and arcing fault; Step S400: Execute a graded early warning strategy based on the confidence level of the fault type and output a diagnostic report.

2. The intelligent detection and fault early warning method for ring network boxes according to claim 1, characterized in that, In step S200, extracting time-frequency statistical features from the partial discharge ultra-high frequency signal specifically involves: The S-transform of the partial discharge ultra-high frequency signal is performed to obtain the time-frequency matrix; Statistical features, including skewness, kurtosis, and discharge repetition rate, are extracted from the time-frequency matrix.

3. The intelligent detection and fault early warning method for ring network boxes according to claim 1, characterized in that, In step S200, the correlation characteristics between load current and measured temperature rise are obtained in the following way: Based on the real-time collected load current I, and using the thermal model formula... Calculate the theoretical temperature rise ; Calculate the measured temperature rise With the theoretical temperature rise absolute difference ; The difference As the associated feature, I is the real-time monitored load current. The thermal resistance of the circuit. is the thermal time constant.

4. The intelligent detection and fault early warning method for ring network boxes according to claim 1, characterized in that, Step S300 specifically includes: S301, Construct an identification framework that includes all preset fault types. ; S302, establish a basic probability allocation function for each sensor data source, and map the extracted feature vectors to the recognition framework. Reliability distribution of each proposition; S303 uses the Dempster combination rule to fuse the basic probability assignment functions from different sensor data sources. The calculation formula is as follows: Where m(A) is the basic probability assignment after the fusion of proposition A. and The base probabilities from different sources of evidence are assigned values, where K is the conflict factor and 1-K is the normalization factor.

5. The intelligent detection and fault early warning method for ring network boxes according to claim 4, characterized in that, Step S300 also includes the arc fault prediction sub-step S304: The multi-dimensional feature vector sequence within a continuous time window is input into a pre-trained long short-term memory network prediction model, which outputs the probability value of an arc fault occurring in the future. The probability value can be used as an independent piece of evidence in the fusion process of the Dempster combination rule, or it can be compared with the fault confidence obtained after fusion.

6. The intelligent detection and fault early warning method for ring network boxes according to claim 5, characterized in that, The Long Short-Term Memory Network prediction model structure includes: The input layer is used to receive feature sequences with dimensions [time step, number of features]; At least one LSTM layer is used to learn time series dependencies; Dropout layers are used to prevent overfitting; The system consists of a fully connected layer and an output layer. The output layer uses the Sigmoid activation function to output an arc fault probability value between 0 and 1.

7. The intelligent detection and fault early warning method for ring network boxes according to claim 1, characterized in that, The graded early warning strategy described in step S400 is as follows: When the confidence level of any fault type is within the first threshold range, a level one warning is triggered, local audible and visual alarms are activated, and the warning information is uploaded to the monitoring center. When the confidence level of any fault type is higher than a second threshold that is higher than the upper limit of the first threshold range, a second-level warning is triggered, and a circuit breaker tripping recommendation instruction is generated based on the first-level warning action.

8. An intelligent detection and early warning system for implementing the method according to any one of claims 1 to 7, characterized in that, include: The data acquisition module includes an ultra-high frequency (UHF) sensor, an infrared thermal imager, an ultraviolet sensor, and a high-frequency current transformer, all arranged within the ring network box. An edge computing gateway, with a built-in processor and memory, configured to perform the steps of the method as described in any one of claims 1 to 7; The cloud platform or monitoring center communicates with the edge computing gateway to receive, store, and display diagnostic results and early warning information.

9. The intelligent detection and early warning system according to claim 8, characterized in that, The edge computing gateway supports the IEC 61850 MMS communication protocol and can upload diagnostic results and early warning information to the cloud platform or monitoring center in a standard format.

10. The intelligent detection and early warning system according to claim 8, characterized in that, The infrared thermal imager has a temperature resolution of no less than 0.1℃, and the UHF sensor has a sampling rate of no less than 100MS / s for the partial discharge ultra-high frequency signal.

Images