Fault recording intelligent analysis method and system

Abstract

The invention relates to the field of fault monitoring, in particular to a fault recording intelligent analysis method and system. According to the method, alarm signals of a monitoring system are collected in real time and automatically classified, starting signals of the fault oscillograph are extracted, and the signal triggering frequency of the same equipment is counted to generate a frequently-sent signal list. And when the triggering frequency exceeds a first threshold value of dynamic adjustment, calling a related wave recording file to carry out time domain alignment and waveform feature extraction, calculating a waveform difference quantitative index, and generating an abnormal index in combination with the triggering frequency. And if the abnormal index exceeds a dynamically adjusted second threshold value, positioning a potential defect section, generating pop-up window alarm information, and pushing the pop-up window alarm information to a monitoring terminal. The method effectively solves the problem that a conventional fault recorder cannot find the latent defect in advance, achieves the early warning of the latent defect, reduces the operation risk of a system, and improves the monitoring efficiency.

Description

Technical Field

[0001] The present invention relates to the field of fault monitoring, and in particular to a fault recording intelligent analysis method and system. Background Art

[0002] In power systems, fault recorders are crucial tools for monitoring equipment operating conditions. Their activation is typically correlated with changes in voltage, current, or switching values, providing insights into potential operational issues. However, existing technologies present numerous shortcomings. For example, patent CN107843779B proposes a method and system for classifying and analyzing power system fault recorders based on fuzzy clustering. While this method addresses the inability to automatically classify power system fault types, it still lacks real-time monitoring and intelligent analysis of fault recorder activation signals. Current fault recorders are primarily used for post-fault analysis to determine fault type, protection activation, and fault location results. They are unable to analyze recordings of latent defects before they develop into faults, enabling early detection. Furthermore, the new generation of centralized control systems is in its early stages of operation and lacks comprehensive methods for managing frequent signals. Frequent screen flooding can disrupt monitoring personnel's work, necessitating the development of folding and statistical functions for frequent signals. In addition, the existing technology lacks differentiated comparative analysis research on recording files of power grid equipment under normal and abnormal conditions, and fails to deeply explore the correspondence mechanism between latent defect types and recording waveforms. Summary of the Invention

[0003] The present invention provides a fault recorder intelligent analysis method and system, aiming to solve the problem that existing fault recorders cannot detect latent defects in advance.

[0004] To achieve the above objectives, the following technical solutions are adopted.

[0005] A fault recording intelligent analysis method comprises the following steps:

[0006] S1. Real-time collection of alarm signals from the monitoring system, automatic classification of alarm signals through a preset classification rule base, extraction of fault recorder start signals, statistics of signal triggering frequency of the same device within a preset time window, and generation of a frequent signal list

[0007] S2. When the trigger frequency of any device in the frequent signal list exceeds the dynamically adjusted first threshold, trigger the recording file comparison process to retrieve the current recording file and the historical normal recording file of the device;

[0008] S3, perform time domain alignment and waveform feature extraction on the current recording file retrieved in S2 and the historical normal recording file, calculate the waveform difference quantitative index, and generate an abnormality index based on the trigger frequency in step S1;

[0009] S4. When the abnormal index of S3 exceeds the second threshold value adjusted dynamically, the potential defect section is located according to the equipment topology data, and a pop-up alarm message is generated by associating it with the predefined defect type library and pushed to the monitoring terminal.

[0010] Optionally, the method further includes S5, training a latent defect prediction model based on the historical data of steps S1-S4, and dynamically updating the first threshold of step S2 and the second threshold of step S4 to form a closed-loop optimization mechanism.

[0011] Optionally, step S1 includes:

[0012] Build a classification rule base that includes signal type, device code, and time window parameters. The rule base supports setting differentiated statistical periods by device type.

[0013] Real-time analysis of alarm signal metadata, matching device codes and signal type tags, and classifying fault recorder activation signals into independent alarm queues;

[0014] A sliding time window algorithm is used to count the signal triggering frequency of the same device in the independent alarm queue. When the frequency exceeds the initial threshold, it is marked as a frequent signal, and a heat map list including device code, trigger time and frequency is generated.

[0015] Optionally, step S3 includes:

[0016] Preprocess the current recording file retrieved in step S2 to extract the steady-state harmonic components, transient mutation amplitude and time sequence phase characteristics of the voltage and current waveforms;

[0017] Dynamically time-align the current recording waveform with the historical normal recording waveform to eliminate the influence of nonlinear offset of the time axis;

[0018] Calculate the root mean square error, harmonic distortion and transient energy difference of the aligned waveforms to generate multi-dimensional waveform difference indicators;

[0019] The waveform difference index and the trigger frequency in step S1 are input into a weighted fusion model to output a normalized anomaly index, which is used to quantify the risk level of latent defects.

[0020] Optionally, step S4 includes:

[0021] According to the risk level of the abnormality index in step S3, the corresponding potential defect type is matched from the defect type library, wherein the defect type library stores the mapping relationship between the recording characteristics and the defects;

[0022] Locate the protection devices and upstream and downstream equipment associated with abnormal equipment based on grid topology data, and generate a defect impact range analysis diagram;

[0023] Call the disposal measures knowledge base to generate a priority list of disposal suggestions based on equipment type, defect level and real-time grid load status;

[0024] The defect location results, impact range map and disposal suggestions are encapsulated as interactive pop-up information, and a real-time waveform comparison chart is superimposed.

[0025] Optionally, step S5 includes:

[0026] Construct a historical data set, including the frequent signal trigger records in step S1, the abnormal index in step S3, and the subsequent actual fault event labels;

[0027] The latent defect prediction model is trained using a supervised learning algorithm. Input features include trigger frequency growth rate, waveform difference index change trend, and equipment operating age.

[0028] Utilizing the prediction model to output dynamically adjusted first and second thresholds, the first threshold adaptively lowers or raises according to the equipment health baseline, and the second threshold dynamically corrected according to the defect false alarm rate;

[0029] The updated threshold value is fed back to the frequent signal determination logic of step S2 and the alarm triggering condition of step S4 in real time.

[0030] A fault recording intelligent analysis system, comprising:

[0031] The signal acquisition and classification module includes a rule engine, a metadata parsing unit, and a frequency statistics unit. The signal acquisition and classification module is used to collect alarm signals from the monitoring system in real time, automatically classify the alarm signals using a preset classification rule base, extract the fault recorder start signal, count the signal trigger frequency of the same device within a preset time window, and generate a frequent signal list.

[0032] A frequent signal analysis module includes a recording file retrieval interface, a waveform alignment engine, and an anomaly index calculator. The frequent signal analysis module is configured to trigger a recording file comparison process when the trigger frequency of any device in the frequent signal list exceeds a dynamically adjusted first threshold, retrieve the current recording file and a historical normal recording file of the device, perform time domain alignment and waveform feature extraction on the retrieved current recording file and the historical normal recording file, calculate a waveform difference quantitative index, and generate an anomaly index based on the trigger frequency.

[0033] An intelligent alarm push module integrates a defect location unit, a decision tree, and a visual rendering engine. When the anomaly index exceeds a dynamically adjusted second threshold, the module locates the potential defect segment based on the device topology data, associates it with a predefined defect type library, generates a pop-up alarm message, and pushes it to the monitoring terminal.

[0034] The adaptive optimization module includes a machine learning model training platform and a threshold dynamic update interface; the adaptive optimization module is used to train a latent defect prediction model based on historical data, dynamically update the first threshold and the second threshold, and form a closed-loop optimization mechanism.

[0035] Optionally, the signal acquisition and classification module includes:

[0036] A configurable rule base allows users to set signal classification rules and statistical time windows by device type. The rules include voltage sag thresholds, current harmonic limits, and switching quantity mutation conditions.

[0037] Distributed signal processing unit, used to analyze multi-source alarm signals in parallel, extract device codes, signal trigger timestamps, and original recording file indexes;

[0038] A heat map generator is used to map the frequent signal list to the power grid geographic information system, showing the signal trigger density in different areas using color gradients.

[0039] Optionally, the frequent signal analysis module includes:

[0040] Multi-scale waveform feature extractor, used to perform wavelet transform decomposition on the recorded waveform file and extract the feature vectors of low-frequency steady-state components and high-frequency transient components;

[0041] Dynamic time warping unit, used to eliminate the time axis offset of the recorded waveform using a nonlinear alignment algorithm and calculate the waveform similarity matrix;

[0042] The anomaly index fusion device is used to perform weighted fusion of the difference index of the waveform similarity matrix and the trigger frequency of the frequent signal. The weight is dynamically adjusted according to the device type and operating environment.

[0043] Optionally, the intelligent alarm push module includes:

[0044] Defect knowledge graph, used to store the association between equipment defects and recorded waveform features, and support defect root cause analysis based on graph reasoning;

[0045] The treatment strategy optimizer is used to combine real-time grid load data, maintenance resource status, and defect impact range to generate a multi-objective optimized treatment path;

[0046] The interactive pop-up interface is used to embed waveform comparison views, defect probability cloud maps, and animated demonstrations of treatment steps, allowing monitoring personnel to directly trigger treatment work orders.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The fault recording intelligent analysis method of the present application collects the alarm signals of the monitoring system in real time and automatically classifies them, extracts the fault recorder startup signal, counts the trigger frequency of the same device signal to generate a frequent signal list, and when the trigger frequency exceeds the first threshold value of dynamic adjustment, triggers the recording file comparison process, calls the relevant recording files for time domain alignment and waveform feature extraction, calculates the waveform difference quantitative index and generates an abnormality index in combination with the trigger frequency, and then locates the potential defect section when the abnormal index exceeds the second threshold value of dynamic adjustment and generates a pop-up alarm information and pushes it to the monitoring terminal, effectively solving the problem that the previous fault recorder could not detect latent defects in advance, realizes early warning of latent defects, greatly reduces the system operation risk, and improves monitoring efficiency.

[0049] A latent defect prediction model based on historical data training has been added, and the first threshold and the second threshold are dynamically updated to form a closed-loop optimization mechanism, so that the entire analysis method can be continuously optimized and adjusted according to historical data, further improving the accuracy and reliability of latent defect prediction.

[0050] A classification rule library including signal type, device code and time window parameters is constructed, which supports setting differentiated statistical periods by device type. It also matches relevant tags through real-time analysis of alarm signal metadata, adopts sliding time window algorithm to count frequent signals, and generates heat map lists. This enhances the accuracy and detail of alarm signal classification and frequent signal statistics, and provides a more accurate data foundation for subsequent analysis and processing.

[0051] The current recording file is preprocessed to extract waveform features, dynamic time warping is used to align the waveform, the influence of nonlinear offset on the time axis is eliminated, and the multi-dimensional waveform difference index is calculated and input into the weighted fusion model to output the normalized anomaly index, which further improves the accuracy and reliability of waveform difference analysis and enables the anomaly index to more accurately quantify the risk level of latent defects.

[0052] The potential defect type is matched according to the risk level of the abnormality index, the defect impact range is located based on the power grid topology data, the disposal measure knowledge base is called to generate a priority list of disposal suggestions, and the relevant information is encapsulated as interactive pop-up information, providing monitoring personnel with more comprehensive and targeted decision-making support, which helps to quickly respond to and handle potential defects.

[0053] By constructing a historical data set, training a latent defect prediction model through a supervised learning algorithm, and dynamically adjusting the threshold using the output of the prediction model, adaptive adjustment of the threshold is achieved, further improving the intelligence and adaptability of the entire analysis method.

[0054] A fault recording intelligent analysis system includes a signal acquisition and classification module, a frequent signal analysis module, an intelligent alarm push module and an adaptive optimization module. Each module has a clear division of labor and works together to realize intelligent processing of the entire process from signal acquisition and analysis to alarm push and optimization, effectively improving the efficiency and accuracy of fault recording analysis and reducing system operation risks.

[0055] Further refinement and improvement of the signal acquisition and classification module, frequent signal analysis module and intelligent alarm push module have enhanced the system's functions and performance, enabling it to better meet the different needs of practical applications and further improving the practicality and reliability of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 The figure is a control flow diagram of an embodiment of a fault recording intelligent analysis method of the present invention.

[0057] Figure 2 This is a schematic diagram of a fault recording intelligent analysis system module of the present invention. DETAILED DESCRIPTION

[0058] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0059] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise indicated, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present invention.

[0060] Example 1

[0061] like Figure 1 As shown, a fault recording intelligent analysis method includes the following steps:

[0062] S1. Collect alarm signals from the monitoring system in real time, automatically classify the alarm signals through the preset classification rule base, extract the fault recorder startup signal, count the signal triggering frequency of the same device within the preset time window, and generate a frequent signal list.

[0063] In the next-generation monitoring system, high-precision data acquisition equipment is deployed to acquire various alarm signals generated by the monitoring system in real time. These alarm signals contain a wealth of information, such as signal type, device code, trigger timestamp, and original recording file index. The acquisition equipment typically operates at a high frequency to ensure that every generated alarm signal is captured promptly, thereby ensuring the integrity and timeliness of data collection. The acquisition frequency can range from tens to hundreds of times per second, depending on the specific scale of the power grid and the importance of the equipment.

[0064] The classification rule base is the core foundation for automatic alarm signal classification. This rule base is constructed based on key information such as signal type and device code, allowing users to set differentiated statistical periods based on different device types. In practical applications, collected alarm signals are first parsed to extract metadata, including key information such as the device code and signal type tag. This metadata is then matched against the classification rule base. Based on the matching results, fault recorder activation signals are accurately extracted from the numerous alarm signals and classified into independent alarm queues. For example, a rule can be set: If the signal type is "fault recorder activation" and the device code conforms to the power grid fault recorder device coding rules, the signal is considered a fault recorder activation signal. Simultaneously, a sliding time window algorithm is used to count the triggering frequency of the same device signal in the independent alarm queue. When the frequency exceeds an initial threshold, it is marked as a frequent signal. A heat map list is then generated, showing the device code, trigger time, and frequency. This enables accurate monitoring of fault recorder activation signals and effective statistics of frequent signals.

[0065] As a specific example, step S1 is as follows:

[0066] First, a classification rule base containing signal types, device codes, and time window parameters needs to be constructed. The design of this rule base should ensure that it can fully cover the alarm signal characteristics generated by various types of equipment in the power grid. Signal types can be subdivided into multiple categories such as voltage anomalies, current anomalies, and switching quantity changes; device codes are compiled based on the unique identification of power grid equipment to ensure that each device can be accurately identified; the setting of time window parameters needs to consider the operating characteristics and possible failure modes of different equipment. Generally, it can be set from several seconds to tens of seconds. For example, a shorter time window can be set for fast-responding equipment, while it can be appropriately extended for some relatively slow-changing equipment. In addition, the rule base supports setting differentiated statistical periods by equipment type to meet the operational monitoring needs of different types of equipment. For example, rotating equipment may require more frequent monitoring cycles, while the monitoring cycle of some stationary equipment can be relatively extended.

[0067] Next, the metadata of the alarm signals collected by the monitoring system is parsed in real time. This metadata contains rich information about the alarm signal, such as the device code, signal type tag, and signal trigger timestamp. Specialized parsing algorithms and tools are used to quickly and accurately interpret this metadata and extract key information. The extracted device code and signal type tag are then matched against the rules in the classification rule base. For example, if the device code of an alarm signal corresponds to a certain type of transformer, and the signal type is voltage sag, the signal is accurately classified into an independent alarm queue based on the settings in the rule base. This independent alarm queue is used to specifically store the fault recorder activation signal for subsequent targeted analysis and processing.

[0068] Based on the categorized independent alarm queues, a sliding time window algorithm is used to count the signal trigger frequency of the same device. This algorithm updates the statistical results in real time. As the time window slides forward, new signal data is continuously incorporated and older data outside the time range is removed, ensuring the timeliness and accuracy of the statistical results. For example, if the time window is set to 10 seconds, the number of signal triggers for a device within the current 10-second window is recounted every 1-second. When the counted frequency exceeds a pre-set initial threshold, it indicates that the device's alarm signal is abnormally frequent and is immediately marked as a frequent signal. The initial threshold is determined based on historical operating data and expert experience. Different thresholds are generally set for different types of devices and signals. For example, the threshold for critical devices can be relatively low to increase the sensitivity of the warning.

[0069] Finally, a heat map list containing device codes, trigger times, and frequencies is generated. The heat map list displays relevant information about frequent signals in an intuitive and visual manner. The device code is used to accurately identify which device has an anomaly, the trigger time helps analysts understand the temporal distribution of frequent signals, and the frequency information directly reflects the density of signal occurrence. For example, a heat map can use markers of different colors to represent the frequency of signals from different devices. The depth of the color corresponds to the frequency, and the darker the color, the higher the frequency. This allows monitoring personnel to quickly focus on devices that may have potential failure risks, thereby providing clear direction for subsequent in-depth analysis and processing.

[0070] S2. When the trigger frequency of any device in the frequent signal list exceeds the first threshold value adjusted dynamically, the recording file comparison process is triggered to retrieve the current recording file and the historical normal recording file of the device.

[0071] When the trigger frequency of any device in the frequent signal list exceeds the dynamically adjusted first threshold, the recording file comparison process is triggered. At this point, the system retrieves the device's current recording file and historical normal recording files. Current recording files are typically stored on a local storage device or cloud storage platform and can be quickly located and accessed using information such as the device code and recording timestamp. Historical normal recording files are a set of representative files pre-stored in the database that reflect the changes in electrical quantities during normal operation of the device.

[0072] Preprocessing the retrieved current waveform file and historical normal waveform files is a key step. Preprocessing includes noise removal and normalization to improve the accuracy of subsequent analysis. Noise removal can be achieved using filtering algorithms, such as wavelet transforms and low-pass filters. The appropriate filtering method should be selected based on the specific noise characteristics and signal frequency range. Normalization converts waveform data of varying amplitudes to the same dimension for easier comparison and analysis.

[0073] S3. Perform time domain alignment and waveform feature extraction on the current recording file retrieved by S2 and the historical normal recording file, calculate the waveform difference quantitative index, and generate an abnormality index based on the trigger frequency of step S1.

[0074] To accurately compare the current waveform file with a historically normal waveform file, time domain alignment is required. This waveform alignment is performed using a dynamic time warping algorithm, which effectively eliminates the effects of nonlinear time axis offsets. This alignment makes the waveforms comparable across the time dimension.

[0075] Waveform feature extraction extracts key features from the aligned waveforms. These features can reflect changes in the device's operating status. Extracted features include steady-state harmonic components, transient amplitudes, and time-series phase characteristics of the voltage and current waveforms. For example, steady-state harmonic components can be calculated using Fourier transforms, transient amplitudes can be determined by detecting waveform mutation points, and time-series phase characteristics can be analyzed based on the waveform's phase information.

[0076] Calculate waveform difference quantification indicators and generate anomaly index

[0077] After extracting waveform features, quantitative waveform difference indicators are calculated, including root mean square error (RMS), harmonic distortion, and transient energy difference. RMS error reflects the overall difference in amplitude between the two waveforms. It is calculated by squared-summing the difference in amplitude between the two waveforms at the same moment, and then taking the square root of the average. Harmonic distortion reflects the difference in harmonic components within the waveforms and can be calculated by comparing the harmonic amplitudes and phases of the two waveforms. Transient energy difference primarily measures the energy variation during transient events and can be calculated using methods such as integration.

[0078] The aforementioned waveform difference quantification indicators and the trigger frequency from step S1 are input into a weighted fusion model, which outputs a normalized anomaly index. The weighted fusion model assigns weights to each difference indicator based on its importance and relevance. For example, if transient energy differences are highly sensitive to equipment failures, they may be given a higher weight. Through weighted fusion, multiple difference indicators are combined into an anomaly index, which is used to quantify the risk level of latent defects, thereby achieving a comprehensive assessment of the equipment's operating status.

[0079] As a specific example, step S3 is as follows:

[0080] Preprocessing the current recording file retrieved in step S2 is a key starting point. The main goal of preprocessing is to purify the data and standardize it, laying the foundation for subsequent feature extraction and analysis. First of all, noise removal is an essential step, and wavelet transform filtering technology can be used. This filtering method can accurately decompose the signal and effectively remove high-frequency noise and low-frequency interference. At the same time, signal normalization processing is performed to uniformly convert waveform data of different dimensions and amplitudes into a preset standardized range. For example, both voltage and current waveform data are normalized to the range of 0-1, thereby eliminating data deviations caused by dimensional differences and providing standardized data support for subsequent feature extraction.

[0081] Next, the steady-state harmonic components, transient mutation amplitudes, and time-series phase characteristics of the voltage and current waveforms are extracted. The extraction of steady-state harmonic components can be achieved through the fast Fourier transform (FFT) algorithm, which can convert time-domain signals into frequency-domain signals, and then accurately identify the amplitude and phase of each harmonic. Usually up to the 13th harmonic can be extracted because these harmonic components can often fully reflect the electrical characteristics of the equipment under stable operation. The extraction of transient mutation amplitudes relies on detecting the mutation points of the first-order or second-order derivatives of the waveform. When a significant change in the slope of the waveform is detected, it is determined that a transient mutation has occurred, and the waveform amplitude at that moment is recorded. This is crucial for capturing sudden events such as short-circuit faults. The extraction of time-series phase characteristics requires analyzing the phase difference between the waveforms, which can be achieved by calculating the cross-correlation function between the two waveforms. The phase difference information helps to reveal the synchronous operation status of the equipment and potential phase imbalance problems.

[0082] Dynamic time warping and aligning the current recorded waveform with the historical normal recorded waveform is a key step to ensure the accuracy of waveform comparison. Since in actual operation, the waveform may be affected by various factors such as device response delay and signal transmission time difference, resulting in nonlinear offset on the time axis, a dynamic time warping algorithm is used to align the waveform. This algorithm allows elastic expansion and contraction of the time axis, and finds the optimal matching path between the two waveforms while maintaining the waveform shape characteristics, thereby eliminating the impact of time axis offset. For example, for two waveforms with similar shapes but misaligned time axes due to differences in device startup time, dynamic time warping can adjust them to the same time base, making subsequent waveform difference calculations more scientific and accurate.

[0083] The root mean square error (RMS) error, harmonic distortion, and transient energy difference of the aligned waveform are calculated to generate a multidimensional waveform difference index. The RMS error is calculated by squaring, summing, and averaging the amplitude difference between the current waveform and the historical normal waveform at the same moment, and then taking the square root. This index can comprehensively reflect the degree of deviation in the overall amplitude of the waveform. The calculation of harmonic distortion is based on the harmonic components obtained by Fourier transform. By comparing the amplitude and phase differences of each harmonic between the current waveform and the normal waveform, the harmonic distortion rate is calculated to quantify the degree of waveform distortion. The transient energy difference can be calculated using an integral method, accumulating the waveform energy during the transient process, and then comparing the transient energy difference between the current waveform and the normal waveform. This index is mainly used to measure the difference in energy changes of the waveform during the transient process. These difference indices together constitute the multidimensional waveform difference index, which comprehensively depicts the difference between the current waveform and the normal waveform from different angles.

[0084] Finally, the above-mentioned waveform difference index and the trigger frequency obtained in step S1 are input into the weighted fusion model. The core of the weighted fusion model is to assign weights according to the sensitivity of different features to the risk of latent defects. For example, after a large amount of data analysis and expert experience judgment, if the transient energy difference is more sensitive to the fault prediction of a certain equipment, it is given a larger weight. The model organically integrates the multi-dimensional difference index and the trigger frequency information through weighted calculation, and outputs a normalized anomaly index. This anomaly index presents the latent defect risk level in a standardized numerical form, which is convenient for monitoring personnel to intuitively evaluate the health status of the equipment, take corresponding maintenance and repair measures in a timely manner, effectively prevent the occurrence of faults, and ensure the stable operation of the power system.

[0085] S4. When the abnormal index of S3 exceeds the second threshold value adjusted dynamically, the potential defect section is located according to the equipment topology data, and a pop-up alarm message is generated by associating it with the predefined defect type library and pushed to the monitoring terminal.

[0086] When the anomaly index exceeds the dynamically adjusted second threshold, it indicates a potential equipment defect. The potential defect area must be located based on equipment topology data. Grid equipment topology data includes connectivity between devices, upstream and downstream equipment information, and protection device associations. By analyzing this topology data, combined with the location and electrical connections of the abnormal device, the potential impact area of ​​the defect is determined, including associated protection devices and upstream and downstream equipment. This generates a defect impact range analysis diagram.

[0087] Generate pop-up alerts based on a predefined defect type library. The library stores mappings between waveform features and defect types. By matching the waveform features currently analyzed with those in the library, the corresponding potential defect type is identified. For example, if harmonics of a specific frequency appear in the waveform and their amplitude exceeds a certain threshold, this may indicate an insulation defect in the equipment.

[0088] Finally, the defect location results, impact map, and action suggestions are packaged into an interactive pop-up window and pushed to the monitoring terminal. This pop-up window presents the monitoring personnel with an intuitive and eye-catching display, helping them quickly understand the fault situation and take appropriate measures.

[0089] As a specific example, step S4 is as follows:

[0090] When the abnormality index obtained in step S3 exceeds the pre-set dynamically adjusted second threshold, this indicates that there may be potential defects in the operating status of the current equipment. At this time, it is necessary to conduct a detailed analysis of the relevant defect information and take corresponding treatment measures. First, according to the specific risk level corresponding to the abnormality index, the potential defect type is accurately matched from the pre-built defect type library. This defect type library stores in detail the precise mapping relationship between different recording features and various types of defects. For example, for a certain specific waveform distortion feature, the library records the corresponding specific equipment failure mode, such as insulation aging, poor contact, etc. By efficiently comparing the recording features reflected by the current abnormality index with the feature patterns in the library, the potential defect type can be quickly locked.

[0091] After determining the potential defect type, the next step is to accurately locate the protection devices and upstream and downstream equipment associated with the abnormal device based on the grid topology data. Grid topology data details the electrical connections between devices and the network architecture. Through in-depth analysis of this data, the protection devices directly connected to the abnormal device, as well as other related equipment upstream and downstream of its power transmission path, can be accurately identified. Based on this, an intuitive defect impact range analysis diagram is generated. This diagram clearly illustrates the specific location and interrelationships of the abnormal device and its associated equipment in the network, allowing operations and maintenance personnel to clearly understand the potential impact area of ​​the potential defect.

[0092] At the same time, the system calls upon a specialized knowledge base of disposal measures to generate a list of disposal recommendations with a clear priority order based on the type of current equipment, the determined defect level, and the current real-time load status of the power grid. The disposal measures knowledge base has accumulated a large number of disposal plans and experience data for different types of equipment, different defect conditions, and different power grid operating conditions. Based on this information, the system comprehensively considers the importance of the equipment, the urgency of the defect, and the urgency of the power grid load, and provides operation and maintenance personnel with a series of practical disposal measures, which are arranged in order of priority. For example, for urgent defects of critical equipment under high load conditions, it is recommended to prioritize temporary load transfer and emergency repair measures; while for minor defects of secondary equipment, they can be arranged to be handled during routine maintenance.

[0093] Finally, the above-mentioned important information such as defect location results, defect impact range analysis diagram, and disposal suggestion list are effectively integrated and packaged into a rich and intuitive interactive pop-up information, which is then pushed to the monitoring terminal. In the pop-up information, a real-time waveform comparison chart is also superimposed, which shows the difference between the current abnormal waveform and the normal waveform, further assisting operation and maintenance personnel to intuitively understand the operating status and defect conditions of the equipment. At the same time, the pop-up window has an interactive function, allowing monitoring personnel to click to view detailed information, perform further operations, or confirm disposal measures, etc., thereby ensuring that operation and maintenance personnel can obtain key information in a timely and accurate manner, and quickly make reasonable decisions to effectively deal with potential equipment defects and ensure the safe and stable operation of the power grid.

[0094] Example 2

[0095] As a preferred example of the above example, the following steps are also included: S5, training a latent defect prediction model based on the historical data of steps S1-S4, dynamically updating the first threshold of step S2 and the second threshold of step S4, to form a closed-loop optimization mechanism.

[0096] Specifically, step S5 involves training a latent defect prediction model based on the historical data from steps S1 to S4. This historical data encompasses various data records generated throughout the entire process, from signal acquisition, classification, and frequent signal statistics to recording file comparison, waveform feature extraction, anomaly index calculation, and ultimately, alarm notification. For example, historical data includes key information such as the specific instance of the frequent signal trigger in step S1, the anomaly index value calculated in step S3, and the subsequent actual fault event label.

[0097] Through in-depth analysis and mining of this historical data, a latent defect prediction model is constructed and trained using appropriate machine learning algorithms. This model can learn the changing patterns and patterns of various signal characteristics, waveform differences, and anomaly indices before a fault occurs, thereby accurately predicting latent defects. During the model training process, a supervised learning algorithm can be used, taking the feature vectors in the historical data (such as the trigger frequency growth rate, the trend of waveform difference indicators, and the equipment's operating life) as input and the corresponding fault event labels as output. The model parameters are continuously adjusted to improve the model's prediction accuracy.

[0098] Based on the trained latent defect prediction model, the first threshold in step S2 and the second threshold in step S4 can be dynamically updated. Specifically, the model adaptively adjusts the first threshold based on the current health and operating trends of the equipment. For example, if the model predicts a decline in equipment health or an increase in the risk of latent defects, the first threshold is appropriately lowered, enabling the system to provide early warning and analysis of frequent signals. Conversely, if the equipment is operating well and historical data indicates a low risk of failure, the first threshold can be appropriately raised to reduce unnecessary alerts and waste of resources.

[0099] Similarly, the second threshold is dynamically adjusted based on the model's output and the false alarm rate of defects in actual applications. If the false alarm rate is high during actual operation, it may mean that the current second threshold setting is too sensitive. In this case, the second threshold can be appropriately increased based on the model's recommendations to reduce the false alarm rate. Conversely, if you want to increase the sensitivity of potential defect detection, you can appropriately lower the second threshold while maintaining a certain accuracy rate.

[0100] Through this historical data-based model training and dynamic threshold update mechanism, the entire fault recording intelligent analysis method forms a closed-loop optimization system. As grid operation data accumulates and equipment operating status changes, the system automatically adjusts its analysis and early warning strategies, maintaining efficient and accurate fault monitoring and early warning capabilities. This effectively reduces system operation risks and improves the safety and reliability of the power system.

[0101] As a specific example, step S5 is specifically performed as follows:

[0102] When it is necessary to optimize the judgment logic and alarm triggering conditions of frequent signals, a comprehensive and accurate historical data set is first constructed. The historical data set integrates multiple key information modules to ensure that all necessary data dimensions are covered in order to effectively train the latent defect prediction model. Specifically, the historical data set contains the frequent signal triggering instances recorded in step S1. These instances record in detail the time, device code, and trigger frequency of each frequent signal. At the same time, the anomaly index values ​​calculated in step S3 are included. These index values ​​reflect the degree of abnormality in the operating status of the equipment at different time points. In addition, the labels of subsequent actual fault events are also included in the data set. These labels clarify which abnormal conditions eventually evolved into actual faults, as well as the corresponding fault types and severity.

[0103] After constructing a historical data set, a supervised learning algorithm is used to train the latent defect prediction model. The supervised learning algorithm plays a central role in this process, learning patterns and associations from the historical data to enable the model to accurately predict the probability of potential equipment defects. The characteristic parameters input into the model are carefully selected to ensure the model's predictive accuracy. These include the trigger frequency growth rate, which is calculated by calculating the ratio of the difference in the trigger frequency of frequent signals within adjacent time windows to the time interval, reflecting the trend of signal trigger frequency over time; the waveform difference index change trend, which involves analyzing the waveform difference index at multiple time points to observe whether it shows an upward, downward, or fluctuating change. The direction and magnitude of the change can usually be determined by calculating the difference in the waveform difference index between adjacent time points; and the equipment's operating age. Equipment of different operating ages exhibits significant differences in failure modes and probability of occurrence, making this parameter an important reference for the model.

[0104] A fully trained latent defect prediction model can output dynamically adjusted first and second thresholds. The adjustment mechanism of the first threshold is based on the equipment health baseline, which is a set of benchmark values ​​obtained through long-term monitoring and statistical analysis of various indicators under normal operating conditions of the equipment. When the model predicts that the health of the equipment is showing a downward trend, for example, some key operating parameters of the equipment gradually deviate from the normal range, or the triggering frequency of frequent signals shows an abnormal growth trend, the system will automatically lower the first threshold, so that potential abnormalities can be captured earlier in the frequent signal judgment process, thereby triggering the subsequent recording file comparison and analysis process in advance. On the contrary, if the health of the equipment is stable and good, and historical data shows that its failure risk is low, the system will appropriately increase the first threshold to reduce unnecessary alarms and resource consumption.

[0105] The dynamic correction of the second threshold is closely centered around the false alarm rate of defects. During actual operation, the system continuously monitors the accuracy and effectiveness of alarm information and counts the number and frequency of false alarms. If the false alarm rate is high during a certain period, this may mean that the current second threshold setting is too sensitive, causing some non-fault conditions to be mistakenly judged as potential defects. At this point, the system will appropriately increase the second threshold based on the analysis results of the model and an in-depth analysis of recent false alarm cases, thereby increasing the strictness of alarm triggering and reducing the false alarm rate. On the contrary, if you want to enhance the sensitivity of detecting potential defects, for example, when the power grid is in special operating conditions or the equipment is in the aging stage, you can appropriately lower the second threshold while ensuring a certain accuracy rate.

[0106] The updated first and second thresholds are promptly applied to the frequent signal determination logic in step S2 and the alarm triggering conditions in step S4 through a real-time feedback mechanism. This feedback process is enabled by efficient communication and data update mechanisms, ensuring that the system can accurately determine and trigger operations based on the new thresholds in a timely manner. This achieves closed-loop optimization of the entire fault recording intelligent analysis method, continuously improving the system's ability to predict latent defects and the accuracy of alarms, and ensuring the stable operation of the power system and the safety and reliability of equipment.

[0107] Example 3

[0108] like Figure 2 As shown, a fault recording intelligent analysis system is described as follows:

[0109] The system consists of multiple closely integrated modules. The first is the signal acquisition and classification module, which consists of a rules engine, a metadata parsing unit, and a frequency counting unit. The rules engine accurately classifies collected alarm signals based on a preset classification rule base. The classification rule base defines the characteristics of each signal type in detail, including key information such as signal type and device code. The metadata parsing unit is responsible for extracting metadata from the alarm signal, such as the device code, signal type label, and signal trigger timestamp. Once extracted, this metadata is used by the rules engine to match it with the classification rule base, accurately filtering the fault recorder activation signal from the numerous alarm signals. The frequency counting unit uses a sliding time window algorithm to count the signal trigger frequency of the same device within a preset time window in an independent alarm queue. When the frequency exceeds an initial threshold, the unit marks the signal as a frequent signal and generates a heat map list containing the device code, trigger time, and frequency.

[0110] The system also includes a frequent signal analysis module, which consists of a recording file retrieval interface, a waveform alignment engine, and an anomaly index calculator. The recording file retrieval interface is responsible for retrieving the current recording file and historical normal recording files from the storage device. The waveform alignment engine uses a dynamic time warping algorithm to align the waveforms, eliminating the effects of time axis offset. The anomaly index calculator extracts waveform features, such as the steady-state harmonic components, transient mutation amplitude, and timing phase characteristics of the voltage and current waveforms, and calculates waveform difference quantification indicators, including root mean square error, harmonic distortion, and transient energy difference. These difference indicators are combined with the trigger frequency and a weighted fusion model is used to generate a normalized anomaly index, which is used to quantify the risk level of latent defects.

[0111] The intelligent alarm push module integrates a defect location unit, a decision tree, and a visual rendering engine. The defect location unit locates potential defect areas based on device topology data and matches potential defect types with a defect type library. The decision tree generates a prioritized list of recommended actions based on device type, defect severity, and real-time grid load status. The visual rendering engine packages the defect location results, impact area map, and recommended actions into an interactive pop-up window, overlaying a real-time waveform comparison chart, and pushes it to the monitoring terminal.

[0112] Finally, the adaptive optimization module includes a machine learning model training platform and a threshold dynamic update interface. The machine learning model training platform uses a supervised learning algorithm based on historical datasets to train a latent defect prediction model. Input features include trigger frequency growth rate, waveform difference indicator trends, and equipment operating age. The threshold dynamic update interface dynamically adjusts the first and second thresholds based on the model output and provides real-time feedback to the frequency signal analysis module and intelligent alarm push module, implementing a closed-loop optimization mechanism.

[0113] When implementing the signal acquisition and classification module of the fault recording intelligent analysis system, its composition and functions are as follows:

[0114] This module contains a configurable rule base, which is designed to allow users to flexibly set signal classification rules and statistical time windows according to different equipment types. For example, users can set corresponding signal classification rules for different types of power equipment, such as transformers and circuit breakers. These rules specifically include voltage sag thresholds, that is, when the voltage amplitude drops by more than a certain set ratio, it is judged as a voltage sag signal; current harmonic limits, that is, the upper limit of the allowable amplitude of each harmonic in the current signal is specified. If this upper limit is exceeded, it is considered an abnormal harmonic signal; and switching quantity mutation conditions, such as the judgment conditions for the state of the switching device suddenly changing from closed to open, or vice versa. Users enter these rules into the configurable rule base through the configuration interface or parameter setting file provided by the system to achieve accurate classification of signals from different devices.

[0115] The distributed signal processing units within the module are primarily responsible for the parallel parsing of alarm signals from multiple sources. In practical applications, monitoring systems receive a large number of alarm signals from a variety of sources, including geographically located substations and various types of power equipment. To efficiently process this massive volume of signals, the distributed signal processing units utilize a multi-threaded or distributed computing architecture, distributing multiple signal parsing tasks in parallel to different processing units. Each processing unit independently parses the assigned alarm signal and extracts key information, including the device code, which uniquely identifies the device issuing the signal; the signal trigger timestamp, which precisely records the time the signal occurred; and the original recording file index, which facilitates the subsequent rapid location and retrieval of the corresponding complete recording file.

[0116] The heat map generator is responsible for intuitively mapping the information in the frequent signal list to the power grid geographic information system. Specifically, it associates and matches the equipment codes recorded in the frequent signal list with the equipment location data in the power grid geographic information system. Then, based on the frequency of signal triggering, a color gradient is used for visual display on the geographic information system interface. For example, areas with high signal triggering density are represented by red, areas with medium density are represented by yellow, and areas with low density are represented by green. This visual display method allows monitoring personnel to understand the signal triggering density in different areas at a glance, so as to quickly locate areas with potential fault hazards or abnormal operation.

[0117] In the fault recording intelligent analysis system, the frequent signal analysis module is implemented as follows:

[0118] The multi-scale waveform feature extractor is a key component of this module. It uses wavelet transform technology to decompose the recorded waveform file. Wavelet transform can decompose the waveform signal into low-frequency steady-state components and high-frequency transient components of different scales. The low-frequency steady-state component reflects the basic trend and long-term changes in the signal, while the high-frequency transient component captures short-term fluctuations and sudden changes in the signal. By setting the appropriate wavelet basis function and decomposition scale, the extractor can generate feature vectors containing features at each scale. For example, when analyzing power system faults, the low-frequency component may correspond to the normal operating state of the equipment, while the high-frequency component may contain transient information generated by the fault. These feature vectors will provide multi-dimensional data support for subsequent waveform analysis.

[0119] The dynamic time warping unit is responsible for solving the problem of waveform time axis offset. In actual applications, due to various factors such as differences in device response time and signal transmission delay, waveforms from different sources may have nonlinear offsets on the time axis. This unit uses a nonlinear alignment algorithm to adjust the waveform signal on an elastic time axis. Specifically, the algorithm calculates the optimal matching path between the two waveforms so that the difference in waveform characteristics of corresponding points on the deformed waveform time axis is minimized. Through this alignment method, the analysis error caused by time offset can be eliminated, ensuring the accuracy of waveform comparison. In addition, the unit also calculates the similarity matrix of the aligned waveforms, which quantifies the degree of similarity between the waveforms and provides basic data for subsequent difference analysis.

[0120] The anomaly index fusion unit combines the difference indicators in the waveform similarity matrix with the frequency of frequent signal triggering to generate a comprehensive anomaly index. First, the difference indicators are extracted from the similarity matrix, which reflect the degree of mismatch between the waveforms. Then, these difference indicators are weightedly fused with the frequency of frequent signal triggering. The weight setting is dynamically adjusted according to the device type and operating environment. For example, for critical equipment or equipment with a complex operating environment, a higher weight may be given to the difference indicator to more sensitively reflect potential defects. The fused anomaly index is normalized so that it varies within a certain range (such as 0 to 1), thereby achieving a quantitative assessment of the risk of latent defects. Ultimately, the index provides a scientific decision-making basis for the intelligent alarm push module, enabling it to issue alarm information in a timely and accurate manner.

[0121] In the fault recording intelligent analysis system, the intelligent alarm push module is implemented as follows:

[0122] This module contains a defect knowledge graph, which is used to store the association between equipment defects and recording features, and supports defect root cause analysis based on graph reasoning. The knowledge graph associates and stores information such as equipment defect type, recording features, and equipment type in the form of nodes and edges. For example, a specific waveform distortion feature (such as a voltage sag accompanied by a specific harmonic component) is associated with the insulation defect of the equipment. When the system detects an abnormal recording feature, it uses a graph reasoning algorithm to retrieve and analyze the associated defect information from the knowledge graph to determine the root cause of the potential defect. For example, if a specific waveform feature combination is detected in the recording of a certain device, the knowledge graph can be used to infer that the device may have an insulation aging problem.

[0123] The disposal strategy optimizer in the module combines the real-time load data of the power grid, the status of maintenance resources and the scope of defect impact to generate a multi-objective optimized disposal path. Real-time load data is obtained from the power grid monitoring system and reflects the current load situation of the power grid. The status of maintenance resources includes information such as available maintenance personnel, equipment and time. The scope of defect impact is determined based on the equipment topology data and the defect type. For example, a serious defect in a key device may affect the operation of a larger area of ​​the power grid. Based on this information, a multi-objective optimization algorithm, such as a genetic algorithm or a simulated annealing algorithm, is used to comprehensively consider factors such as the urgency of restoring normal operation of the power grid, the rational use of maintenance resources, and the minimization of the impact on user power supply to generate the optimal disposal path. For example, during peak load periods of the power grid, temporary measures are given priority to ensure power supply, while maintenance resources are arranged to carry out thorough repairs during off-peak periods.

[0124] The interactive pop-up interface integrates and displays a variety of information, allowing monitoring personnel to directly trigger action work orders. The interface layout is logically divided into sections, with an embedded waveform comparison view clearly showing the difference between the current abnormal and normal waveforms. A defect probability cloud map visually displays the probability distribution of different defect types, with color depth or height indicating probability. Animated action step demonstrations dynamically demonstrate the action process. The interface also provides action buttons and options, allowing monitoring personnel to directly select and trigger the corresponding action work order within the pop-up interface. The system then generates a work order and assigns it to the appropriate maintenance personnel, ensuring rapid response and efficient action.

[0125] It is understood from common technical knowledge that the present invention may be implemented by other embodiments that do not depart from its spirit or essential features. Therefore, the embodiments disclosed above are, in all respects, merely illustrative and not exclusive. All modifications within the scope of the present invention or equivalent to the scope of the present invention are intended to be encompassed by the present invention.

Claims

1. A fault recording intelligent analysis method, characterized in that: The following steps are involved: S1. Real-time collection of alarm signals from the monitoring system, automatic classification of alarm signals through a preset classification rule base, extraction of fault recorder start signals, statistics of signal triggering frequency of the same device within a preset time window, and generation of a frequent signal list S2. When the trigger frequency of any device in the frequent signal list exceeds the dynamically adjusted first threshold, trigger the recording file comparison process to retrieve the current recording file and the historical normal recording file of the device; S3, perform time domain alignment and waveform feature extraction on the current recording file retrieved in S2 and the historical normal recording file, calculate the waveform difference quantitative index, and generate an abnormality index based on the trigger frequency in step S1; S4. When the abnormal index of S3 exceeds the second threshold value adjusted dynamically, the potential defect section is located according to the equipment topology data, and a pop-up alarm message is generated by associating it with the predefined defect type library and pushed to the monitoring terminal.

2. A fault recording intelligent analysis method according to claim 1, characterized in that: The method further includes S5, training a latent defect prediction model based on the historical data of steps S1-S4, and dynamically updating the first threshold of step S2 and the second threshold of step S4 to form a closed-loop optimization mechanism.

3. The fault recording intelligent analysis method according to claim 1, characterized in that: The step S1 comprises: Build a classification rule base that includes signal type, device code, and time window parameters. The rule base supports setting differentiated statistical periods by device type. Real-time analysis of alarm signal metadata, matching device codes and signal type tags, and classifying fault recorder activation signals into independent alarm queues; A sliding time window algorithm is used to count the signal triggering frequency of the same device in the independent alarm queue. When the frequency exceeds the initial threshold, it is marked as a frequent signal, and a heat map list including device code, trigger time and frequency is generated.

4. The fault recording intelligent analysis method according to claim 1, characterized in that: The step S3 comprises: Preprocess the current recording file retrieved in step S2 to extract the steady-state harmonic components, transient mutation amplitude and time sequence phase characteristics of the voltage and current waveforms; Dynamically time-align the current recording waveform with the historical normal recording waveform to eliminate the influence of nonlinear offset of the time axis; Calculate the root mean square error, harmonic distortion and transient energy difference of the aligned waveforms to generate multi-dimensional waveform difference indicators; The waveform difference index and the trigger frequency in step S1 are input into a weighted fusion model to output a normalized anomaly index, which is used to quantify the risk level of latent defects.

5. The fault recording intelligent analysis method according to claim 1, characterized in that: The step S4 comprises: According to the risk level of the abnormality index in step S3, the corresponding potential defect type is matched from the defect type library, wherein the defect type library stores the mapping relationship between the recording characteristics and the defects; Locate the protection devices and upstream and downstream equipment associated with abnormal equipment based on grid topology data, and generate a defect impact range analysis diagram; Call the disposal measures knowledge base to generate a priority list of disposal suggestions based on equipment type, defect level and real-time grid load status; The defect location results, impact range map and disposal suggestions are encapsulated as interactive pop-up information, and a real-time waveform comparison chart is superimposed.

6. A fault recording intelligent analysis method according to claim 2, characterized in that: The step S5 comprises: Construct a historical data set, including the frequent signal trigger records in step S1, the abnormal index in step S3, and the subsequent actual fault event labels; The latent defect prediction model is trained using a supervised learning algorithm. Input features include trigger frequency growth rate, waveform difference index change trend, and equipment operating age. Utilizing the prediction model to output dynamically adjusted first and second thresholds, the first threshold adaptively lowers or raises according to the equipment health baseline, and the second threshold dynamically corrected according to the defect false alarm rate; The updated threshold value is fed back to the frequent signal determination logic of step S2 and the alarm triggering condition of step S4 in real time.

7. A fault recording intelligent analysis system, based on a fault recording intelligent analysis method according to any one of claims 1 to 6, characterized in that: include, The signal acquisition and classification module includes a rule engine, a metadata parsing unit, and a frequency statistics unit. The signal acquisition and classification module is used to collect alarm signals from the monitoring system in real time, automatically classify the alarm signals using a preset classification rule base, extract the fault recorder start signal, count the signal trigger frequency of the same device within a preset time window, and generate a frequent signal list. A frequent signal analysis module includes a recording file retrieval interface, a waveform alignment engine, and an anomaly index calculator. The frequent signal analysis module is configured to trigger a recording file comparison process when the trigger frequency of any device in the frequent signal list exceeds a dynamically adjusted first threshold, retrieve the current recording file and a historical normal recording file of the device, perform time domain alignment and waveform feature extraction on the retrieved current recording file and the historical normal recording file, calculate a waveform difference quantitative index, and generate an anomaly index based on the trigger frequency. An intelligent alarm push module integrates a defect location unit, a decision tree, and a visual rendering engine. When the anomaly index exceeds a dynamically adjusted second threshold, the module locates the potential defect segment based on the device topology data, associates it with a predefined defect type library, generates a pop-up alarm message, and pushes it to the monitoring terminal. The adaptive optimization module includes a machine learning model training platform and a threshold dynamic update interface; the adaptive optimization module is used to train a latent defect prediction model based on historical data, dynamically update the first threshold and the second threshold, and form a closed-loop optimization mechanism.

8. The fault recording intelligent analysis system according to claim 7, characterized in that: The signal acquisition and classification module includes: A configurable rule base allows users to set signal classification rules and statistical time windows by device type. The rules include voltage sag thresholds, current harmonic limits, and switching quantity mutation conditions. Distributed signal processing unit, used to analyze multi-source alarm signals in parallel, extract device codes, signal trigger timestamps, and original recording file indexes; A heat map generator is used to map the frequent signal list to the power grid geographic information system, showing the signal trigger density in different areas using color gradients.

9. The fault recording intelligent analysis system according to claim 7, characterized in that: The frequent signal analysis module includes: Multi-scale waveform feature extractor, used to perform wavelet transform decomposition on the recorded waveform file and extract the feature vectors of low-frequency steady-state components and high-frequency transient components; Dynamic time warping unit, used to eliminate the time axis offset of the recorded waveform using a nonlinear alignment algorithm and calculate the waveform similarity matrix; The anomaly index fusion device is used to perform weighted fusion of the difference index of the waveform similarity matrix and the trigger frequency of the frequent signal. The weight is dynamically adjusted according to the device type and operating environment.

10. The fault recording intelligent analysis system according to claim 7, characterized in that: The intelligent alarm push module includes: Defect knowledge graph, used to store the association between equipment defects and recorded waveform features, and support defect root cause analysis based on graph reasoning; The treatment strategy optimizer is used to combine real-time grid load data, maintenance resource status, and defect impact range to generate a multi-objective optimized treatment path; The interactive pop-up interface is used to embed waveform comparison views, defect probability cloud maps, and animated demonstrations of treatment steps, allowing monitoring personnel to directly trigger treatment work orders.

Images