Method and system for monitoring state of low-voltage power distribution network

Abstract

The invention discloses a low-voltage power distribution network state monitoring method, and belongs to the technical field of power system monitoring. According to the method, edge computing nodes are deployed in a power distribution site, and multi-phase current time sequence data, multi-phase voltage time sequence data and temperature data of a target line are synchronously collected in real time; performing parallel monitoring on the time series data based on a plurality of trigger criteria such as a current abrupt change rate criterion, a current effective value out-of-limit criterion, a voltage effective value out-of-limit criterion and a temperature change rate criterion, and when any one of the trigger criteria is met, triggering and storing a waveform data set in a time window before and after a trigger point; processing the waveform data set and the associated steady-state data, and extracting a waveform feature vector, an impedance feature vector and a temperature associated feature vector; and inputting the extracted feature vectors into a multi-criterion fusion diagnosis model deployed at an edge computing node, and carrying out fusion analysis on the multi-dimensional feature vectors to obtain a diagnosis result containing fault type classification. According to the method, real-time state monitoring and fault diagnosis of the low-voltage power distribution network can be realized, and the safety and reliability of a power distribution system are improved.

Description

Technical Field

[0001] This application relates to a method for monitoring the condition of a distribution network, and more specifically, to a method for monitoring the condition of a low-voltage distribution network based on edge computing nodes and multi-criteria fusion fault diagnosis. Background Technology

[0002] In power systems, the low-voltage distribution network, as a crucial link connecting the user side, directly affects power supply reliability and power quality. Currently, condition monitoring of the low-voltage distribution network mainly relies on various protection devices and monitoring terminals deployed in the field. Traditional methods widely employ fault detection techniques based on single electrical quantity thresholds, such as detecting overcurrent faults by monitoring whether the line current exceeds a set value, or detecting undervoltage events by detecting voltage dips. Furthermore, some monitoring devices with waveform recording capabilities trigger waveform recording when they detect sudden changes in electrical quantities, and upload the collected raw waveform data via network to a back-end analysis system located at the master station, where the master station performs further data processing and fault assessment.

[0003] Currently, distribution network fault monitoring technology mainly adopts distributed monitoring methods based on edge computing. Chinese patent application CN120632538A discloses a distribution network line fault analysis method based on edge computing. This method uses sensors distributed across nodes to collect real-time data on the distribution network's physical facilities and operation, and extracts the distribution network line layout anomaly and power deviation based on edge computing. Chinese patent application CN120214478A provides a distribution network fault monitoring system that uses edge computing for preliminary processing and feature extraction of multi-source data, and generates fault feature vectors using multi-source data fusion technology. Chinese patent application CN120539535A proposes a fault identification method based on low-voltage distribution network topology modeling and intelligent algorithms. This method collects voltage, current, and other data in real-time through IoT terminals and constructs a dataset using edge computing technology. Furthermore, Chinese patent application CN110646677A describes a method for identifying the topology and line impedance of a low-voltage distribution network in a distribution area, utilizing an edge computing terminal for waveform data sampling and topology identification. Chinese patent application CN117454234A discloses a fault identification method for county power grids based on cloud-edge collaboration, which uses edge computing nodes for fault identification preprocessing.

[0004] However, existing fault monitoring methods for low-voltage (below 400V) distribution networks still have significant shortcomings in dealing with the increasingly complex operating environment of distribution networks. First, traditional fault capture mechanisms typically rely on a single current or voltage threshold as a trigger condition. This approach struggles to comprehensively capture the complete transient process before and after a fault occurs, leading to the omission of many transient or high-resistance faults due to their unclear characteristics. Frequent unexplained tripping events in the field directly reflect this limitation. Second, in terms of data processing, uploading massive amounts of raw waveform data to the master station not only consumes significant communication bandwidth resources but also makes data analysis and processing entirely dependent on the master station, resulting in significant time delays in the entire diagnostic process and failing to meet the requirements for real-time and rapid fault response. Furthermore, existing methods often focus on simple threshold judgments of single electrical quantities, lacking the ability to deeply mine the rich features inherent in the waveform data itself. In particular, they fail to effectively integrate non-electrical quantity information such as temperature, which reflects the equipment's physical state, leading to a rather crude judgment of the fault's nature. Furthermore, the diagnostic logic of most current monitoring systems is relatively rigid, relying on preset values ​​and rules. When line characteristics change or new fault modes occur, the system is difficult to adapt and adjust itself. Its flexibility and scalability can no longer meet the needs of the development of smart distribution networks. Summary of the Invention

[0005] Traditional low-voltage distribution network fault diagnosis methods in existing technologies mainly rely on the single current threshold action of protection devices, which cannot record the transient process of faults, leading to frequent unexplained tripping. Existing monitoring devices with waveform recording functions upload all raw waveform data for post-analysis by the master station system. The massive data transmission of raw data consumes bandwidth, and the analysis is completed at the master station, resulting in poor real-time performance. There is a lack of ability to intelligently extract and fuse waveform data itself, and it is impossible to correlate non-electrical quantity information such as temperature. The diagnostic logic is rigid and difficult to adapt to different line characteristics and new fault modes, resulting in poor flexibility and scalability.

[0006] The technical problem this application aims to solve is that existing low-voltage distribution network fault diagnosis methods mainly rely on the single current threshold action of protection devices, which cannot record the transient process of the fault, leading to frequent "unexplained" tripping. Existing monitoring devices with waveform recording functions upload all raw waveform data for post-analysis by the master station system. The massive data transmission of raw data consumes bandwidth, and the analysis is completed at the master station, resulting in poor real-time performance. Furthermore, there is a lack of ability to intelligently extract and fuse the waveform data itself, making it impossible to correlate non-electrical quantity information such as temperature. The diagnostic logic is rigid, making it difficult to adapt to different line characteristics and new fault modes, resulting in poor flexibility and scalability.

[0007] The technical solution adopted by this application to solve its technical problem is: In a first aspect, this application provides a low-voltage distribution network condition monitoring method, the method being executed on an edge computing node deployed at the distribution site, comprising: Real-time synchronous acquisition of multi-phase current timing data, multi-phase voltage timing data, and temperature data of the target line; The time series data is monitored in parallel based on a variety of preset trigger criteria, including at least one of current mutation rate criteria and temperature change rate criteria. When any of the trigger criteria is met, a waveform dataset containing waveform data within the time window before and after the trigger point is triggered and saved. The waveform dataset and associated steady-state data are processed on the edge computing node to extract waveform feature vectors, impedance feature vectors, and temperature-related feature vectors. The waveform feature vector, impedance feature vector, and temperature-related feature vector are input into the multi-criteria fusion diagnostic model deployed on the edge computing node to obtain diagnostic results that include fault type classification.

[0008] As a further improvement to this application, the multiple triggering criteria include at least two of the following: current mutation rate criterion, current effective value exceeding limit criterion, voltage effective value exceeding limit criterion, and temperature change rate criterion.

[0009] As a further improvement of this application, the impedance feature vector includes: calculating the transient loop impedance during the fault based on the waveform dataset, and comparing the transient loop impedance with the normal state impedance.

[0010] As a further improvement of this application, the transient loop impedance during the fault is calculated based on the waveform dataset. Specifically, a preset cycle data after the fault occurs is selected, the fundamental component is extracted, the voltage and current of the fault loop are calculated, and the transient loop impedance is obtained. The calculation of transient loop impedance during a fault based on the waveform dataset further includes: extracting harmonic voltage and harmonic current under a preset characteristic frequency band, calculating harmonic transient loop impedance, and using the ratio of fundamental transient loop impedance to harmonic transient loop impedance as part of the impedance feature vector.

[0011] As a further improvement of this application, after obtaining the diagnostic results, the method further includes: if the diagnostic results include short-circuit faults, then evaluating the operating performance of the associated circuit breaker based on the waveform dataset.

[0012] As a further improvement of this application, evaluating the operating performance of the associated circuit breaker based on the waveform dataset includes: extracting the fault current curve and the circuit breaker operating time point from the waveform dataset, and comparing the fault current curve with the standard operating characteristic curve to generate a performance evaluation result.

[0013] As a further improvement of this application, the waveform feature vector includes at least one of the amplitude abrupt change rate, phase offset angle, and waveform distortion rate of each phase current and voltage extracted based on the comparison of cycles before and after the fault. The waveform distortion rate is the total harmonic distortion rate, and its calculation formula is as follows:

[0014] in This is the fundamental effective value. This is the effective value of the h-th harmonic.

[0015] As a further improvement of this application, the waveform feature vector includes total harmonic distortion (THD); the method further includes: identifying the fault nature as an arc fault or a metallic fault based on the value of the THD and its trend over time. The temperature-related feature vector includes a temperature-current coupling coefficient, which is used to characterize the degree of correlation between temperature changes and current changes, in order to distinguish between overload overheating faults and overheating faults due to poor contact of the equipment body.

[0016] As a further improvement to this application, a model optimization step is also included: forming a training dataset from diagnostic reports and corresponding confirmation results from multiple edge computing nodes, which is used to retrain and optimize the multi-criteria fusion diagnostic model, and then sending the optimized model parameters to the edge computing nodes.

[0017] Secondly, this application provides a low-voltage distribution network condition monitoring system, deployed in an edge computing node at the distribution site, comprising: The data acquisition and triggering module is used to acquire multi-phase current time-series data, multi-phase voltage time-series data and temperature data of the target line in real time and to monitor the time-series data in parallel based on a variety of preset triggering criteria. When any of the triggering criteria is met, the module triggers and saves a waveform dataset containing waveform data within the time window before and after the trigger point. The feature extraction module is used to process the waveform dataset and associated steady-state data on the edge computing node to extract waveform feature vectors, impedance feature vectors and temperature-related feature vectors. The intelligent diagnostic module, which embeds a multi-criteria fusion diagnostic model, is used to receive the waveform feature vector, impedance feature vector, and temperature-related feature vector, and output diagnostic results that include fault type classification.

[0018] The system also includes a circuit breaker performance evaluation module, which is used to evaluate the operating performance of the associated circuit breaker based on the waveform dataset when the diagnostic results output by the intelligent diagnostic module include short-circuit faults.

[0019] Compared with the prior art, this application has at least the following beneficial effects: This application deploys edge computing nodes at the power distribution site, bringing data acquisition, feature extraction, and intelligent diagnosis closer to the data source, thus changing the traditional model that relies on centralized processing at a master station. The edge nodes, based on real-time synchronous acquisition of multi-phase current, voltage, and temperature data, perform parallel monitoring based on multiple triggering criteria. Once an anomaly is detected, it immediately triggers and saves a complete dataset of waveforms before and after the fault, ensuring comprehensive capture of various fault transient processes and effectively solving the problems of missed detections and unclear fault causes associated with traditional single-criteria methods.

[0020] This application performs deep processing on waveform datasets directly at the edge, extracting multi-dimensional feature vectors containing waveform features, impedance features, and temperature correlation features. This transforms massive amounts of raw data into high-value-density feature information, significantly reducing the amount of data that needs to be transmitted and saving communication bandwidth resources. Simultaneously, the multi-dimensional feature vectors are input into a multi-criteria fusion diagnostic model deployed at the edge nodes for real-time inference, achieving edge-side closed-loop fault diagnosis. This reduces the diagnosis time from hours in traditional main station analysis to seconds / minutes, significantly improving the real-time performance of fault response.

[0021] This application constructs a multi-dimensional feature system by integrating electrical and non-electrical quantity characteristics, encompassing amplitude abrupt changes, phase shifts, waveform distortion, transient impedance changes, and temperature change trends, providing rich data support for subsequent intelligent diagnosis. The multi-criteria fusion diagnostic model can comprehensively utilize multi-dimensional features for integrated judgment, significantly improving the accuracy of fault type identification, especially for complex faults such as high-resistance faults and intermittent arcing grounding.

[0022] Furthermore, this application possesses excellent scalability and self-evolution capabilities. By evaluating the circuit breaker's operational performance based on diagnostic results, it extends functionality from fault monitoring to equipment condition assessment. A training dataset is formed by aggregating diagnostic reports and confirmation results from multiple edge nodes in the cloud, continuously retraining and optimizing the diagnostic model. The updated model parameters are then distributed to edge nodes, creating a closed loop of edge application-cloud training-edge update. This allows the system to continuously accumulate operational experience and adapt to new fault modes, significantly improving the intelligence level of distribution network monitoring.

[0023] By deploying a multi-criteria fusion diagnostic model on edge computing nodes, real-time edge-side diagnosis is achieved, reducing fault assessment time from hours to seconds / minutes, providing core algorithmic support for rapid power restoration. By fusing multi-dimensional features of electrical and non-electrical quantities, the diagnostic accuracy is far higher than traditional single thresholds or fixed logic criteria, solving the problem of incomplete fault capture in traditional methods. By refining data into information at the edge, only lightweight diagnostic reports and key features are uploaded, significantly saving communication bandwidth and cloud storage costs, solving the problem of poor real-time diagnostics. Modular design facilitates the integration of new sensor data and new diagnostic sub-models, and the algorithm framework has good scalability, solving the problem of insufficient utilization of multi-source information. Through a closed loop of "edge application - cloud training - edge update," the system can continuously accumulate operational experience, evolve, and improve its intelligence level. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the core process of this solution; Figure 2 This is the core flowchart of edge computing in this solution; Figure 3 This is a detailed flowchart illustrating the multi-source feature extraction and fusion diagnosis process of this solution. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Terminology Explanation To clearly define the terms used in this application, the following explanations are provided: 1. Edge computing node: refers to a computing unit deployed at the power distribution site (close to the data source), possessing data acquisition, real-time processing, local storage, and intelligent analysis capabilities. In this application, the edge computing node, such as the smart cloud box, is used to perform functions such as synchronous data acquisition, multi-criteria triggered waveform recording, feature extraction, and local intelligent diagnosis.

[0027] 2. Multi-criteria fusion diagnostic model: This refers to a machine learning model deployed on edge computing nodes, used to receive multi-dimensional feature vectors and output fault diagnosis results. This model can be an algorithm trained on historical data, such as a lightweight gradient boosting tree or a pruned convolutional neural network. By fusing and analyzing waveform features, impedance features, and temperature correlation features, it achieves accurate identification of fault types.

[0028] 3. Waveform Recording: This refers to the process by which the system automatically saves the original waveform data at a high sampling rate within a time window before and after the trigger point when the trigger criterion is met. In this application, the waveform recording data includes complete waveforms of multiple cycles before and after the trigger point, which are used for subsequent feature extraction and fault analysis.

[0029] 4. Total Harmonic Distortion (THD): This is a metric that measures the degree of waveform distortion and is used to assess the harmonic content in current or voltage waveforms after a fault.

[0030] 5. Fundamental frequency: refers to the sinusoidal wave component with the same frequency as the power system's operating frequency, which is typically 50Hz or 60Hz. In this application, the fundamental frequency component is used to calculate the transient loop impedance and total harmonic distortion rate.

[0031] 6. Harmonics: refers to sinusoidal wave components whose frequencies are integer multiples of the fundamental frequency. In this application, the harmonic analysis range is from the 2nd to the 13th order, used to calculate the total harmonic distortion rate and extract waveform features.

[0032] 7. Transient loop impedance: refers to the impedance value presented by the fault loop at the instant the fault occurs. In this application, the transient loop impedance is obtained by extracting the fundamental voltage and fundamental current of a preset cycle (such as the first cycle) after the fault occurs, and calculating the ratio of voltage to current. This transient loop impedance is then compared with the normal state impedance to determine the nature of the fault.

[0033] 8. Circuit Breaker Trip Characteristic Curve: This refers to a standard curve describing the relationship between the tripping time and the current flowing through the circuit breaker. In this application, the circuit breaker's operating performance is evaluated by comparing the actual fault current curve with the standard operating characteristic curve.

[0034] 9. Cloud-Edge Collaboration: This refers to a collaborative working model between a cloud platform and edge computing nodes. In this application, it is manifested as follows: edge computing nodes are responsible for real-time data collection and diagnosis, uploading diagnostic reports to the cloud; the cloud platform aggregates data from multiple edge nodes to form a training set, retrains and optimizes the multi-criteria fusion diagnostic model, and distributes the updated model parameters to each edge node, forming a closed loop of edge application - cloud training - edge update.

[0035] 10. Current mutation rate criterion: This refers to the triggering condition for anomaly detection based on the comparison of the ratio of current change to time change with a preset threshold. In this application, data saving is triggered when the current change rate exceeds the first threshold.

[0036] 11. Temperature Change Rate Criterion: This refers to the triggering condition for anomaly detection based on the comparison of the ratio of temperature change to time change with a preset threshold. In this application, data saving is triggered when the temperature change rate exceeds the fifth threshold.

[0037] 12. JSON: JavaScript Object Notation, a lightweight data exchange format. In this application, it is used to compress and package diagnostic results, key features, and waveform data into structured data, facilitating uploading to a cloud platform via 4G / 5G networks.

[0038] To address the technical problems of traditional low-voltage distribution network fault diagnosis methods, such as incomplete fault capture, poor real-time diagnostics, insufficient utilization of multi-source information, and low level of intelligence, and to achieve revolutionary improvements in diagnostic speed, significantly enhanced diagnostic accuracy and reliability, maximized data value, strong functional scalability, and the formation of a knowledge closed loop, this paper proposes a low-voltage distribution network condition monitoring method based on full-sensing and multi-source data fusion. The method includes the following steps: Step 1: As Figure 1 As shown, data synchronous acquisition and triggering: Real-time synchronous acquisition of the timing data of the current in the three phases A, B, and C of the target line and the neutral line N. Three-phase voltage timing data And cable sheath temperature data T(t); Step 2: Multi-level anomaly detection and waveform recording: The time-series data is monitored in parallel based on multiple preset trigger criteria. When any criterion is met, high-speed cyclic waveform recording is triggered, and a preset time window before and after the trigger point is locked and saved. High sampling rate complete waveform dataset ; The multi-level triggering criteria mentioned in step 2 include: Current mutation rate criterion: ,in , This is the threshold for the rate of change of current. Criteria for exceeding the effective current limit: ,in This is the overcurrent protection threshold. Criteria for exceeding the effective voltage limit: or ; Criterion for rate of temperature change: .

[0039] Specifically: Current mutation rate threshold Overcurrent protection threshold Voltage over-limit threshold and Temperature change rate threshold It is not a fixed value, but can be preset and optimized online based on the historical operating data of the specific line, equipment parameters and expert experience, so as to adapt to the monitoring requirements of different application scenarios.

[0040] Furthermore, the above method creates diversity in criteria, simultaneously monitoring electrical quantities (current mutation rate, current RMS value, voltage RMS value) and non-electrical quantities (temperature change rate), employing a flexible combination of at least two, rather than activating all at once. This reflects adaptability to different application scenarios (cable networks, overhead lines, different load characteristics). Triggering is not for tripping (protection function), but for preserving a complete waveform dataset, providing a data foundation for subsequent multi-criteria fusion diagnosis. In traditional power system protection, temperature monitoring is usually an independent online monitoring system (such as cable temperature measurement systems), separate from electrical quantity protection devices, with no data exchange, and is not used as a criterion for triggering waveform recording. This application fuses electrical and non-electrical quantities at the triggering layer, constructing a multi-dimensional sensing grid. This requires a deep understanding of the response characteristics of different physical quantities and solving the technical challenge of synchronous acquisition of multi-source data. This application's approach of monitoring the rate of change of electrical quantities and the rate of change of temperature of non-electrical quantities in parallel and using them as the trigger condition for waveform recording breaks through the limitations of traditional devices that only consider electrical quantities and ignore thermal quantities, and only consider steady state and ignore transient states. It is a targeted solution proposed to solve the problems mentioned in the background technology, such as incomplete fault capture and insufficient utilization of multi-source information.

[0041] Step 3: As Figure 3 As shown, multi-source feature extraction: for the waveform dataset The associated steady-state data is processed to extract feature vectors that include at least the following dimensions: Waveform feature vector Based on the comparison of cycles before and after the fault, the amplitude abrupt change rate, phase offset angle, and waveform distortion rate of each phase current and voltage are extracted. Among them, the waveform distortion rate is calculated using the total harmonic distortion (THD). Specifically, for the waveform after the fault, the fundamental and main harmonic (2nd to 13th) components of each phase current and voltage are extracted and calculated according to the formula THD.

[0042] in, This is the fundamental effective value. This is the effective value of the h-th harmonic; the harmonic order ranges from 2 to 13. Impedance eigenvector Calculate the transient loop impedance during a fault. And compare the impedance under normal conditions. ; Temperature correlation characteristics : Extract the rate of temperature change before and after the trigger point and absolute temperature value T; Optionally, the impedance eigenvector in step 3 The calculation method is as follows: Select the first cycle data after the fault occurs, extract the fundamental component, and calculate the voltage of the fault loop. With current The transient impedance is obtained. ;Will The impedance ratio and impedance angle characteristics are generated by comparing the impedance with the preset normal load impedance range.

[0043] Step 4: Edge intelligence fusion diagnosis: such as Figure 2 As shown, the multi-source feature vector F=[ The input is fed into the lightweight multi-criteria fusion diagnostic model deployed on the edge computing node, and the lightweight multi-criteria fusion diagnostic model outputs a first diagnostic result; the first diagnostic result includes at least fault type classification, fault phase identification, and fault nature judgment; wherein, the fault type classification includes short-circuit transient, short-circuit short-time, overload, overvoltage, undervoltage, and single-phase grounding.

[0044] Optionally, step 6 is included after step 4: online evaluation of circuit breaker performance: if the first diagnostic result contains a short-circuit fault, then from the waveform dataset... Extracting fault current curves Circuit breaker operating time point Compare the fault current curve with the standard circuit breaker operating characteristic curve. The actual tripping time is compared with the standard time, the deviation is calculated, and the deviation is evaluated to determine whether it is within the allowable tolerance range, thus generating the circuit breaker operation performance evaluation result.

[0045] Step 5: Result Output and Reporting: Output the first diagnostic result, key feature vectors, and the waveform dataset. The compressed representation is packaged to generate a diagnostic report and uploaded to the cloud platform.

[0046] Optionally, the method further includes step 7: model adaptive iteration: the cloud platform receives diagnostic reports and corresponding final manual confirmation results from multiple edge nodes to form an incremental training dataset; the multi-criteria fusion diagnostic model is periodically retrained and optimized based on the incremental training dataset, and the updated model parameters are sent to each edge computing node.

[0047] Furthermore, each of the various triggering criteria corresponds to a preset priority. When multiple triggering criteria are met simultaneously, the time window length or sampling frequency of the saved waveform dataset is dynamically adjusted according to the priority. First, by assigning differentiated priorities to different triggering events, a preliminary classification of fault severity is achieved, enabling the system to prioritize the allocation of limited computing and storage resources to high-priority critical events. Second, the mechanism of dynamically adjusting the time window length or sampling frequency ensures that high-priority faults can obtain more complete and higher-precision waveform data for subsequent analysis, while low-priority events can adopt simplified waveform recording strategies to save storage space. Finally, this hierarchical waveform recording strategy effectively improves the resource utilization efficiency of edge computing nodes without increasing hardware costs, achieving an optimized balance between monitoring accuracy and storage overhead.

[0048] Furthermore, the waveform feature vector includes the total harmonic distortion (THD); the method further includes: identifying the fault nature as an arc fault or a metallic fault based on the value of the THD and its trend over time. First, arc faults and metallic faults have distinctly different physical characteristics. Metallic short-circuit faults exhibit smaller waveform distortion and lower THD values, while arc faults, due to the nonlinear characteristics of the arc, generate abundant harmonic components and significantly increase the THD value. The two types of faults can be directly distinguished by the THD value. Second, the trend of THD over time can further reflect the dynamic characteristics of the arc; for example, intermittent arcs exhibit periodic fluctuations in THD, while continuous stable arcs exhibit a stable high THD value. Finally, accurate identification of arc faults is of great value for preventing electrical fires and ensuring personal safety. Traditional protection devices struggle to effectively identify arc faults; this technical solution fills this technological gap.

[0049] Furthermore, the temperature-related feature vector includes a temperature-current coupling coefficient, which characterizes the correlation between temperature and current changes to distinguish between overload heating faults and equipment body contact heating faults. First, overload heating is essentially caused by increased current leading to ohmic heating of the conductor, resulting in a strong positive correlation between temperature and current changes and a high coupling coefficient. In contrast, equipment body contact heating (such as oxidation of cable joints) is caused by increased contact resistance leading to localized heating, resulting in increased temperature but minimal current change and a low coupling coefficient. Second, by introducing this innovative feature of the temperature-current coupling coefficient, accurate differentiation between these two types of thermal faults is achieved—a technical effect that traditional single temperature or current monitoring cannot achieve. Finally, accurately identifying equipment body heating faults has significant engineering value for conducting preventative maintenance and preventing equipment burn-out accidents.

[0050] Furthermore, the calculation of transient loop impedance during the fault based on the waveform dataset further includes: extracting harmonic voltage and harmonic current under a preset characteristic frequency band, calculating the harmonic transient loop impedance, and using the ratio of the fundamental transient loop impedance to the harmonic transient loop impedance as part of the impedance feature vector. First, different fault types exhibit differentiated impedance characteristics at different frequency bands. For example, arc faults show a significant increase in impedance at high frequencies, while metallic short circuits have low impedance across all frequency bands. By introducing multi-frequency impedance analysis, richer fault characteristic information is obtained than that of a single fundamental impedance. Second, the ratio of fundamental impedance to harmonic impedance eliminates the influence of common factors such as fault distance and system impedance, becoming a normalized index with good stability. Finally, this technical solution expands the application dimensions of the traditional impedance method and significantly improves the ability to identify complex faults (such as high-resistance faults and arc faults).

[0051] Furthermore, when the multi-criteria fusion diagnostic model outputs the diagnostic result, it also outputs the corresponding confidence level. The method further includes: when the confidence level is lower than a preset confidence threshold, adjusting the thresholds of the multiple triggering criteria to enter an enhanced monitoring mode. First, the introduction of confidence level allows for the quantitative expression of the reliability of the diagnostic result, providing an important reference for subsequent decision-making. Second, when the confidence level is low, the system automatically adjusts the triggering criterion thresholds (such as lowering the current mutation rate threshold and shortening the temperature change rate monitoring cycle), entering a more sensitive enhanced monitoring mode, which improves the ability to capture difficult faults. Third, this technical solution realizes the feedback control of the diagnostic result on the triggering mechanism, forming a complete monitoring-diagnosis-optimization closed loop, significantly improving the system's adaptability and intelligence level. Finally, compared with the traditional one-way process of triggering and diagnosing, this technical solution enables the system to continuously self-optimize during operation, which is especially suitable for power distribution scenarios with variable fault characteristics and complex operating conditions.

[0052] The core innovation of this application lies in the deep integration of non-electrical quantities such as temperature with electrical quantities throughout the entire process, constructing a three-dimensional perception system covering power grid status and equipment health. Specifically, at the acquisition layer, synchronous acquisition and spatiotemporal alignment of electrical quantity and temperature data are achieved, laying the foundation for subsequent correlation analysis; at the triggering layer, the temperature change rate criterion is used as an independent triggering criterion alongside current mutation rate, current effective value exceeding limit, and voltage effective value exceeding limit, enabling early warning of electrical quantity blind zone faults such as cable joint oxidation and insulation aging; at the feature layer, not only are basic features such as absolute temperature value and temperature change rate extracted, but also higher-order features such as temperature-current coupling coefficient are further extracted. By analyzing the correlation between temperature change and current change, overload heating faults can be accurately distinguished. The fault (high coupling coefficient) and the overheating fault due to poor contact with the equipment body (low coupling coefficient) are identified. At the diagnostic layer, temperature correlation features, waveform features, and impedance features are input into the multi-criteria fusion diagnostic model to assist in fault type identification and improve diagnostic confidence. Especially in scenarios with ambiguous electrical quantity features such as high-resistance grounding and arc faults, temperature features can serve as an independent source of evidence. At the evaluation layer, the degree of thermal damage to the equipment is analyzed based on temperature data to support circuit breaker performance evaluation and predictive maintenance. At the optimization layer, the temperature criterion threshold is adjusted based on the diagnostic confidence feedback to form an adaptive closed loop of the monitoring strategy.

[0053] The present application will be described in detail below through specific embodiments: Example 1: This embodiment uses the diagnosis of an A / B phase-to-A short-circuit fault in a low-voltage outgoing cable as an example to illustrate the execution process of the method of this application.

[0054] Step 1: Edge computing nodes (such as smart cloud boxes) synchronously collect data on the three-phase current, voltage, and cable temperature of A, B, and C phases at a sampling rate of 10kHz through the connected TDA-111 sensing terminal.

[0055] Step 2: In At any given moment, the current mutation rate criterion module detects... and The set threshold was exceeded within 0.5ms. This immediately triggers high-speed waveform recording. The waveform recording module saves the data. Raw waveform data for 5 cycles before and after the time point (approximately 200ms in total) .

[0056] Step 3: The feature extraction module begins to work: Calculate the first cycle after the fault relative to the cycle before the fault. and The amplitude increased by 8 times, the phase difference was about 180 degrees, and the waveform distortion was slight. and The amplitude decreases slightly, forming a waveform feature vector. ; Take the fundamental voltage and current after the fault and calculate the impedance of the A-B phase-to-phase circuit. Much smaller than the normal load impedance Furthermore, the impedance angle is close to the line impedance angle, forming an impedance characteristic vector. ; Read Temperature at time T and rate of temperature change in the first 10 minutes The temperature was found to be normal and stable, forming a temperature-related characteristic. The combination yields the feature vector. F =[ F ω , F z , F t ].

[0057] Step 4: Input the feature vector F into the edge device model. The model then... The two-phase currents show a significant increase, phase reversal, and a slight voltage drop. The low resistance characteristics in, and The temperature showed no abnormal characteristics. The probability distribution was quickly calculated, and the first diagnostic result was output: {Fault type: instantaneous short circuit; Fault phase: AB phase; Fault nature: metallic short circuit; Confidence level: 96%}.

[0058] Wherein: the lightweight multi-criteria fusion diagnostic model is a machine learning model trained with a large amount of historical data, such as a lightweight gradient boosting tree or a pruned convolutional neural network, which receives multi-source feature vectors F and outputs the probability distribution of various faults.

[0059] Step 5: The result generation module will generate the diagnostic results and impedance values. Key characteristics, as well as information such as the peak value and duration of the fault current, are packaged into a diagnostic report in JSON format, uploaded to the cloud platform via the 4G network, and trigger an alarm in the APP.

[0060] Example 2 A low-voltage distribution network status monitoring method is implemented on edge computing nodes. This method achieves real-time monitoring and fault diagnosis of the distribution network status through multi-criteria fusion diagnosis.

[0061] First, edge computing nodes synchronously collect multi-phase current time-series data, multi-phase voltage time-series data, and temperature data of the target line in real time. During the acquisition process, current and voltage sensors continuously monitor the three-phase current and voltage at a synchronous sampling frequency, while temperature sensors monitor temperature changes of key equipment. The sampling frequency is set to 10kHz to ensure that transient processes and high-frequency characteristics in the power grid can be captured.

[0062] Next, the edge computing nodes perform parallel monitoring of time-series data based on a variety of preset trigger criteria. These trigger criteria include current mutation rate criteria, current RMS value exceeding limit criteria, voltage RMS value exceeding limit criteria, and temperature change rate criteria. The current mutation rate criteria are based on comparing the ratio of current change to time change with a first threshold; data saving is triggered when the current change rate exceeds the first threshold. The current RMS value exceeding limit criteria are based on comparing the current RMS value with a second threshold; triggering this criterion occurs when the current RMS value exceeds the second threshold of the normal operating range. The voltage RMS value exceeding limit criteria are based on comparing the voltage RMS value with either a third or fourth threshold, where the third threshold is the overvoltage limit and the fourth threshold is the undervoltage limit. The temperature change rate criteria are based on comparing the temperature change rate with a fifth threshold; anomaly detection is triggered when the device temperature change rate exceeds the fifth threshold.

[0063] When any trigger criterion is met, the system immediately triggers and saves a waveform dataset containing waveform data within a time window before and after the trigger point. The time window is set to 500 milliseconds before the trigger point and 1000 milliseconds after the trigger point to ensure a complete record of the entire fault occurrence process. The waveform dataset includes three-phase current waveforms, three-phase voltage waveforms, and corresponding temperature data.

[0064] The system then processes the waveform dataset and associated steady-state data to extract waveform feature vectors, impedance feature vectors, and temperature-related feature vectors. Waveform feature vector extraction includes parameters such as peak value, RMS value, spectral characteristics, and harmonic content of current and voltage. Impedance feature vector extraction involves calculating the transient loop impedance during the fault based on the waveform dataset and comparing it with the normal-state impedance. Specifically, three cycles of data after the fault occur are selected, the fundamental component is extracted, and the voltage and current of the fault loop are calculated. The transient loop impedance is obtained through the ratio of voltage to current. Temperature-related feature vectors include features such as temperature change trends, temperature gradients, and the correlation between temperature and electrical parameters.

[0065] Finally, the waveform feature vector, impedance feature vector, and temperature-related feature vector are input into the multi-criteria fusion diagnostic model deployed on edge computing nodes to obtain diagnostic results that include fault type classification. The multi-criteria fusion diagnostic model uses deep learning algorithms to identify various fault types such as short-circuit faults, ground faults, overload faults, and equipment aging, and provides an assessment of fault severity.

[0066] After obtaining the diagnostic results, if the diagnostic results include short-circuit faults, the system further evaluates the operating performance of the associated circuit breakers based on the waveform dataset. The evaluation process includes extracting the fault current curve and the circuit breaker operating time point from the waveform dataset, comparing the fault current curve with the standard operating characteristic curve to generate performance evaluation results. By analyzing the circuit breaker's operating time, breaking capacity, and protection coordination, the system assesses whether the circuit breaker is operating normally.

[0067] To continuously optimize diagnostic accuracy, the method also includes a model optimization step. The system uses diagnostic reports and corresponding confirmation results from multiple edge computing nodes to form a training dataset, which is then used to retrain and optimize the multi-criteria fusion diagnostic model. By collecting real-world fault cases and expert confirmation results, the system continuously improves the model's diagnostic capabilities and distributes the optimized model parameters to the edge computing nodes, enabling continuous improvement and updating of the model.

[0068] This method achieves real-time monitoring and intelligent diagnosis of power distribution network status through edge computing technology. The multi-criteria fusion mechanism improves the accuracy and reliability of fault detection, while reducing dependence on the central server and improving the system's response speed and stability.

[0069] Example 3 A power distribution network monitoring system deployed on edge computing nodes is provided. The system is deployed on edge computing nodes at the power distribution site and achieves intelligent monitoring and fault diagnosis of the power distribution network through the collaborative work of multiple modules.

[0070] The system includes a data acquisition and triggering module, a feature extraction module, an intelligent diagnosis module, and a circuit breaker performance evaluation module. It also establishes a communication connection with the cloud management platform to form an edge-cloud collaborative monitoring system.

[0071] The data acquisition and triggering module is responsible for real-time synchronous acquisition of multi-phase current time-series data, multi-phase voltage time-series data, and temperature data of the target line. This module employs a high-precision data acquisition unit with a sampling frequency set to 10kHz to ensure the capture of rapidly changing signals in the distribution network. The module incorporates multiple triggering criteria, including current surge criteria, voltage dip criteria, frequency offset criteria, and temperature anomaly criteria, to monitor the acquired time-series data in parallel. When any triggering criterion meets a preset condition, the system immediately triggers the data saving mechanism, automatically saving a waveform dataset containing waveform data within a time window before and after the trigger point. The time window is set to 200ms before the trigger point and 300ms after the trigger point, totaling 500ms of waveform data, ensuring a complete record of the entire fault occurrence process.

[0072] The feature extraction module receives waveform datasets from the data acquisition and triggering module and performs in-depth processing on the waveform datasets and associated steady-state data. First, the module preprocesses the waveform data, including denoising, filtering, and normalization. Then, it extracts waveform feature vectors using signal processing techniques such as wavelet transform and Fourier transform, resulting in 64-dimensional feature parameters, including amplitude features, frequency domain features, and time-frequency features. Simultaneously, the module calculates line impedance parameters based on voltage and current data, extracting impedance feature vectors, including 12-dimensional impedance features such as positive-sequence impedance, negative-sequence impedance, and zero-sequence impedance. Furthermore, the module combines temperature sensor data to analyze the correlation between temperature and electrical parameters, extracting temperature-related feature vectors, including 8-dimensional temperature features such as temperature gradient and temperature-current correlation coefficient.

[0073] The intelligent diagnostic module is the core decision-making unit of the system, embedding a multi-criteria fusion diagnostic model. This model employs a deep neural network architecture, comprising an input layer, three hidden layers, and an output layer. It can simultaneously receive waveform feature vectors, impedance feature vectors, and temperature-related feature vectors from the feature extraction module. Through multi-layer nonlinear transformations and feature fusion, the model achieves accurate identification and classification of different types of faults. The output diagnostic results include fault type classification, covering various fault types such as short-circuit faults, grounding faults, open-circuit faults, and overload faults, and provide fault probability and confidence level assessments.

[0074] The circuit breaker performance evaluation module is specifically designed for evaluating and analyzing the operating performance of circuit breakers. When the intelligent diagnostic module's results include short-circuit faults, this module automatically initiates the evaluation program, performing a detailed analysis of the associated circuit breaker's operating performance based on the saved waveform dataset. By analyzing the changes in current waveforms before and after the circuit breaker's operation, the module calculates key performance parameters such as the circuit breaker's operating time, breaking current, and arcing time, assessing the circuit breaker's health status and remaining life, and providing a scientific basis for equipment maintenance.

[0075] The cloud management platform establishes stable communication connections with multiple edge computing nodes, using 4G / 5G wireless communication for data transmission. The platform collects monitoring data, diagnostic results, and operational status information from each edge node, building a large-scale power distribution network operation database. Based on this massive amount of data, the platform uses machine learning algorithms to continuously retrain and optimize the multi-criteria fusion diagnostic model, constantly improving its diagnostic accuracy and adaptability. Once the model optimization is complete, the platform distributes the updated model parameters to each edge computing node via the wireless network, enabling remote model upgrades and unified management.

[0076] In a preferred embodiment, the system may also include only the circuit breaker performance evaluation module without being connected to the cloud management platform, forming an independent edge monitoring unit, which is suitable for remote distribution network monitoring scenarios with limited communication conditions.

[0077] This system, through the deep integration of edge computing and cloud intelligence, enables real-time monitoring, intelligent diagnosis, and equipment performance evaluation of power distribution network faults, significantly improving the safety and reliability of power distribution network operation and providing important technical support for the construction of smart grids.

[0078] A third objective of this application is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the low-voltage distribution network status monitoring method.

[0079] A fourth objective of this application is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the low-voltage distribution network status monitoring method.

[0080] A fifth objective of this application is to provide a computer program product comprising computer instructions that instruct a computer to execute the low-voltage distribution network status monitoring method.

[0081] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0082] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0083] This application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, readable storage media, optical storage, etc.) containing computer-usable program code.

[0084] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0085] Obviously, the described embodiments are only some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort should fall within the scope of protection of this application.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation methods of this application. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this application should be covered within the protection scope of this application.

Claims

1. A method for monitoring the condition of a low-voltage distribution network, wherein the method is executed on an edge computing node deployed at the distribution site, characterized in that, include: Real-time synchronous acquisition of multi-phase current timing data, multi-phase voltage timing data, and temperature data of the target line; The time series data is monitored in parallel based on a variety of preset trigger criteria, including at least one of current mutation rate criteria and temperature change rate criteria. When any of the trigger criteria is met, a waveform dataset containing waveform data within the time window before and after the trigger point is triggered and saved. The waveform dataset and associated steady-state data are processed on the edge computing node to extract waveform feature vectors, impedance feature vectors, and temperature-related feature vectors. The waveform feature vector, impedance feature vector, and temperature-related feature vector are input into the multi-criteria fusion diagnostic model deployed on the edge computing node to obtain diagnostic results that include fault type classification.

2. The low-voltage distribution network condition monitoring method according to claim 1, characterized in that, The multiple triggering criteria include at least two of the following: current mutation rate criterion, current effective value exceeding limit criterion, voltage effective value exceeding limit criterion, and temperature change rate criterion.

3. The low-voltage distribution network condition monitoring method according to claim 1, characterized in that, The impedance feature vector includes: calculating the transient loop impedance during the fault based on the waveform dataset, and comparing the transient loop impedance with the normal state impedance.

4. The low-voltage distribution network condition monitoring method according to claim 3, characterized in that, The transient loop impedance during the fault is calculated based on the waveform dataset. Specifically, a preset cycle data after the fault occurs is selected, the fundamental component is extracted, the voltage and current of the fault loop are calculated, and the transient loop impedance is obtained. The calculation of transient loop impedance during a fault based on the waveform dataset further includes: extracting harmonic voltage and harmonic current under a preset characteristic frequency band, calculating harmonic transient loop impedance, and using the ratio of fundamental transient loop impedance to harmonic transient loop impedance as part of the impedance feature vector.

5. The low-voltage distribution network condition monitoring method according to claim 1, characterized in that, After obtaining the diagnostic results, the method further includes: if the diagnostic results include short-circuit faults, then evaluating the operating performance of the associated circuit breaker based on the waveform dataset.

6. The low-voltage distribution network condition monitoring method according to claim 5, characterized in that, Evaluating the operating performance of the associated circuit breaker based on the waveform dataset includes: extracting the fault current curve and the circuit breaker operating time point from the waveform dataset, and comparing the fault current curve with the standard operating characteristic curve to generate performance evaluation results.

7. The low-voltage distribution network condition monitoring method according to claim 1, characterized in that, The waveform feature vector includes at least one of the following: amplitude abrupt change rate, phase offset angle, and waveform distortion rate of each phase current and voltage extracted based on the comparison of cycles before and after the fault. The waveform distortion rate is the total harmonic distortion rate, and its calculation formula is as follows: in This is the fundamental effective value. This is the effective value of the h-th harmonic.

8. The low-voltage distribution network condition monitoring method according to claim 7, characterized in that, The waveform feature vector includes the total harmonic distortion rate; the method further includes: identifying the fault nature as an arc fault or a metallic fault based on the value of the total harmonic distortion rate and its trend over time. The temperature-related feature vector includes a temperature-current coupling coefficient, which is used to characterize the degree of correlation between temperature changes and current changes, in order to distinguish between overload overheating faults and overheating faults due to poor contact of the equipment body.

9. The low-voltage distribution network condition monitoring method according to claim 1, characterized in that, It also includes a model optimization step: forming a training dataset from diagnostic reports and corresponding confirmation results from multiple edge computing nodes, which is used to retrain and optimize the multi-criteria fusion diagnostic model, and then sending the optimized model parameters to the edge computing nodes.

10. A low-voltage distribution network condition monitoring system, characterized in that, Edge computing nodes deployed at power distribution sites include: The data acquisition and triggering module is used to acquire multi-phase current time-series data, multi-phase voltage time-series data and temperature data of the target line in real time and to monitor the time-series data in parallel based on a variety of preset triggering criteria. When any of the triggering criteria is met, the module triggers and saves a waveform dataset containing waveform data within the time window before and after the trigger point. The feature extraction module is used to process the waveform dataset and associated steady-state data on the edge computing node to extract waveform feature vectors, impedance feature vectors and temperature-related feature vectors. The intelligent diagnostic module, which embeds a multi-criteria fusion diagnostic model, is used to receive the waveform feature vector, impedance feature vector, and temperature-related feature vector, and output diagnostic results that include fault type classification.

Images