A method and system for intelligent diagnosis and prevention of malfunction of grounding current of GIS shielded wire

Abstract

This invention discloses an intelligent diagnostic method and system for grounding current of GIS shielded cables, belonging to the field of power equipment condition protection technology. This method achieves early and accurate identification and graded warning of abnormal states by real-time acquisition of grounding current and operating parameters, combined with a dynamic benchmark model and unsupervised machine learning algorithms. Its core lies in its ability to automatically generate an anti-maloperation signal and send it to the differential protection device when the abnormality is determined to originate from the external circuit of the cable shielded cable. This allows for blocking or delaying protection action in the absence of internal faults, effectively preventing false tripping. This invention solves the problems of high false alarm rate and inability to distinguish fault sources in traditional fixed threshold monitoring methods, significantly improving the accuracy of condition monitoring and the safety of main equipment operation.

Description

Technical Field

[0001] This invention belongs to the field of power equipment protection technology, and relates to a method for intelligent diagnosis and prevention of maloperation of grounding current of GIS shielded wire, and also relates to a system for intelligent diagnosis and prevention of maloperation of grounding current of GIS shielded wire. Background Technology

[0002] The grounding circuit of the cable shield between gas-insulated metal-enclosed switchgear (GIS) and the main transformer is a critical link in ensuring equipment safety and electromagnetic compatibility. Abnormal changes in its grounding current often indicate insulation degradation, partial discharge, or grounding system defects. Currently, monitoring of this circuit commonly employs current over-limit alarms based on fixed thresholds. This method has significant limitations: First, the grounding current is significantly affected by operating conditions such as unit load, system voltage, and ambient temperature; fixed thresholds cannot adapt to dynamic changes, easily leading to false alarms or missed alarms. Second, existing monitoring methods lack in-depth diagnostic capabilities; when an anomaly is detected, it cannot effectively distinguish whether the fault originates inside the transformer or in the external cable shield circuit, leading to the risk of maloperation of the generator-transformer differential protection, potentially causing unnecessary unplanned shutdowns. Furthermore, traditional methods are insensitive to the slow degradation process of insulation, making predictive maintenance difficult. Therefore, there is an urgent need for a precise diagnostic method that can intelligently identify the source of anomalies, adapt to changes in operating conditions, and coordinate with the protection system to prevent maloperation, in order to improve the reliability of condition monitoring and ensure the safe and stable operation of the main equipment.

[0003] Based on the above background, the technical problem that this application aims to solve is: how to overcome the shortcomings of existing fixed threshold monitoring methods, and provide an intelligent diagnosis and anti-maloperation method and system that can adapt to unit operating conditions, accurately identify abnormal grounding current of GIS shielding wires, and cooperate with relay protection to prevent maloperation, thereby achieving an upgrade from passive alarm to active early warning and precise protection. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for intelligent diagnosis and prevention of maloperation of GIS shielded wire grounding current, which can adapt to the unit's operating conditions, accurately identify abnormal grounding current of GIS shielded wire, and cooperate with relay protection to prevent maloperation.

[0005] The first technical solution adopted in this invention is a method for intelligent diagnosis and prevention of malfunction of grounding current of GIS shielded wire, comprising the following steps:

[0006] Step 1: Real-time acquisition of grounding current signals and associated unit operating parameters of the cable shield grounding circuit between the GIS and the main transformer, and preprocessing of the grounding current signals; Step 2: Calculation of the characteristic quantity of the current grounding current based on the preprocessed grounding current signal, comparison of the characteristic quantity with the dynamic benchmark model established based on historical health data to obtain the characteristic deviation; Step 3: Input of the characteristic quantity and characteristic deviation into the pre-trained intelligent diagnostic model, the intelligent diagnostic model outputs insulation status assessment results and early warning level;

[0007] The intelligent diagnostic model includes at least an unsupervised machine learning model for identifying transient abnormal patterns; Step 4: Issue graded warning information according to the warning level; When the warning level reaches the preset danger threshold, generate an anti-maloperation auxiliary criterion signal and send the signal to the generator-transformer differential protection device for its use in performing protection logic judgment.

[0008] Another technical solution adopted in this invention is a GIS shielded wire grounding current intelligent diagnosis and anti-maloperation system for implementing the above method, comprising:

[0009] The monitoring and acquisition module is used to collect grounding current signals and unit operating parameters in real time.

[0010] The intelligent diagnostic module, connected to the monitoring and acquisition module, is used to store the dynamic benchmark model, calculate the characteristic quantities, run the intelligent diagnostic model, and output the insulation status assessment results and early warning level.

[0011] The collaborative execution module, connected to the intelligent diagnostic module, is used to issue warnings based on the warning level and generate and send anti-maloperation auxiliary judgment signals to the generator-transformer differential protection device when the danger threshold is reached.

[0012] The data interface module is used to write the characteristic data, early warning events, and related current waveform slices output by the intelligent diagnostic module into the power plant fault recording device.

[0013] The invention is further characterized by:

[0014] Furthermore, in step 1, the associated unit operating condition parameters include one or more of the following: unit active power, system voltage, and ambient temperature; preprocessing includes filtering and calibration.

[0015] Furthermore, in step 2, the characteristic quantities include one or more of the following: the effective value of the grounding current, the zero-sequence current component, the content of a specific harmonic, the waveform statistical characteristics, and the similarity distance with the reference waveform in the dynamic reference model.

[0016] Furthermore, in step 2, the method for establishing the dynamic reference model is as follows: during the healthy operation phase of the system, the reference waveform, reference effective value, and reference harmonic content of the grounding current under different operating condition combinations are learned and stored.

[0017] Furthermore, in step 3, the intelligent diagnostic model also includes a trend prediction model, which is used to predict the changing trend of the grounding current based on the time series of the effective value, and to trigger an insulation deterioration trend warning when the predicted value exceeds the slow change warning threshold.

[0018] Furthermore, in step 3, the unsupervised machine learning model is the Isolation Forest model.

[0019] Furthermore, in step 4, the graded early warning information includes attention level, warning level, and danger level; when the early warning level is warning level or danger level, the method also includes the step of: retrieving and correlating transformer oil chromatographic data or partial discharge monitoring data to perform multi-source information fusion diagnosis.

[0020] Furthermore, in step 4, the specific steps for generating the anti-maloperation auxiliary criterion signal include: matching and comparing the abnormal characteristics of the current grounding current with the pre-stored transformer internal fault current characteristic library; if the matching fails, it is determined that the current abnormality originates from a problem outside the cable shield, and an anti-maloperation auxiliary criterion signal is generated to block differential protection or increase its action delay.

[0021] The monitoring and acquisition module includes a high-precision micro-current transformer and a signal conditioning unit installed on the grounding circuit of the shielded wires of the three-phase cables A, B, and C; the collaborative execution module and the generator-transformer differential protection device transmit signals through hard wiring or GOOSE communication messages based on the IEC61850 standard.

[0022] The beneficial effects of this invention are as follows:

[0023] 1. The precise early warning system designed in this invention eliminates operational interference through a dynamic benchmark model, establishing a "healthy baseline" for the system that varies with load and temperature. Combined with unsupervised machine learning algorithms such as isolated forests, the system can not only identify known fault modes but also keenly detect weak or unknown anomalies deviating from the baseline, achieving very early detection of insulation defects. Based on this, the system issues graded warnings according to the severity of the anomalies, and automatically correlates multi-source data such as transformer oil chromatography and partial discharge data for cross-validation and fusion analysis during medium and high-level warnings. This significantly improves the accuracy and reliability of diagnostic conclusions, upgrading the monitoring system from a simple threshold alarm to an intelligent diagnostic expert with deep cognitive capabilities.

[0024] 2. When the present invention determines that the abnormal grounding current originates from an external circuit of the cable shield (such as multiple grounding points or insulation damage) rather than an internal transformer fault, the system immediately activates the anti-maloperation collaborative logic. This involves rapidly matching real-time abnormal characteristics with a pre-stored database of internal transformer fault characteristics. If a match fails, a clear anti-maloperation auxiliary criterion signal (such as "blocking" or "increasing delay") is automatically generated. This signal is sent in real-time to the generator-transformer group differential protection device via a reliable channel (hard-wired or GOOSE message), and incorporated into its protection logic as a high-priority auxiliary criterion. This ensures that the differential protection will not trip malfunctioning in the event of an external circuit anomaly, effectively preventing unexpected power outages caused by non-core equipment issues, and significantly improving the operational safety of the main transformer and the continuity of power grid supply. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart of the method of the present invention;

[0027] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0028] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0029] The following is in conjunction with the appendix Figure 1 To be continued Figure 2 The invention will be described in detail with specific embodiments:

[0030] A method for intelligent diagnosis and prevention of malfunction of grounding current in GIS shielded wires, referenced Figure 1 This includes the following steps:

[0031] Step 1: Real-time acquisition of the grounding current signal of the cable shield grounding circuit between the GIS and the main transformer, as well as the associated unit operating parameters, and preprocessing of the grounding current signal, specifically implemented according to the following steps:

[0032] The associated unit operating condition parameters include one or more of the unit's active power, system voltage, and ambient temperature; the preprocessing includes filtering and calibration.

[0033] Step 1.1 Installation and Data Acquisition: In the grounding circuit of the cable shield between the GIS and the main transformer, install high-precision micro-current transformers on the three-phase grounding wires A, B, and C respectively, and synchronously acquire the real-time instantaneous value signals of the three-phase grounding current. ;

[0034] Step 1.2 Synchronously collect operating parameters: Simultaneously collect the unit's active power. System voltage and ambient temperature As a parameter related to operating conditions;

[0035] Step 1.3 Signal Preprocessing: The acquired raw current signal is digitally filtered and calibrated sequentially; the digital filtering uses a notch filter to filter out power frequency and its main harmonic interference; the calibration is used to convert the secondary side signal into the actual primary side current value and perform normalization processing.

[0036] Step 2: Based on the preprocessed grounding current signal, calculate the characteristic quantity of the current grounding current, and compare the characteristic quantity with the dynamic benchmark model established based on historical health data to obtain the characteristic deviation. This is implemented according to the following steps:

[0037] The characteristic quantities include: the effective value of the grounding current, the zero-sequence current component, the content of specific harmonics, waveform statistical characteristics, and the similarity distance with the reference waveform in the dynamic reference model.

[0038] Step 2.1, Real-time characteristic quantity calculation: Based on the preprocessed current signal, a set of characteristic quantities characterizing the grounding current state are periodically calculated. The characteristic quantities include the total effective value of the three-phase current. Zero-sequence current component Specific subharmonic content The time-domain statistical characteristics of the current waveform and its similarity distance to the reference waveform. or correlation coefficient .

[0039] Step 2.2, Dynamic Benchmark Model Comparison: The calculated real-time feature quantities are compared with the benchmark values ​​and normal fluctuation ranges stored in the dynamic benchmark model under the corresponding operating conditions; the dynamic benchmark model is established by learning historical data under different operating condition combinations during the system's healthy operation phase; the comparison results generate a set of quantitative feature deviation indicators.

[0040] Step 3: The feature quantity and the feature deviation are input into the pre-trained intelligent diagnostic model. The intelligent diagnostic model outputs the insulation status assessment result and the warning level, which is implemented according to the following steps:

[0041] The intelligent diagnostic model includes:

[0042] Unsupervised machine learning models are used to identify transient anomalous patterns; specifically, unsupervised machine learning models are the Isolation Forest model.

[0043] The trend prediction model is used to predict the changing trend of ground current characteristics based on time series, and to trigger an insulation degradation trend warning when the predicted value exceeds the slow change warning threshold.

[0044] Step 3.1, Model Input: Simultaneously input the real-time feature values ​​calculated in Step 2.1 and the feature deviation values ​​generated in Step 2.2 into the pre-trained intelligent diagnostic model set.

[0045] Step 3.2 Trend Prediction Analysis: The trend prediction model in the intelligent diagnostic model set analyzes the historical sequence of key characteristic quantities and predicts their future short-term change trends; if the predicted value continues to exceed the slow change warning threshold obtained from historical data statistics, an insulation degradation trend warning is triggered.

[0046] Step 3.3, Anomaly Detection and Analysis: Simultaneously, the unsupervised anomaly detection model in the intelligent diagnostic model set calculates its anomaly score using real-time feature quantities as input; if the anomaly score exceeds a set threshold, an immediate anomaly alarm is triggered.

[0047] Step 3.4, Comprehensive Status Assessment: Integrate the results of trend warnings and abnormal alarms, combine them with the current operating parameters for comprehensive analysis, and output the final insulation status assessment result and the corresponding warning level.

[0048] Step 4: Issue graded warning information according to the warning level; when the warning level reaches the preset danger threshold, generate an anti-maloperation auxiliary criterion signal and send the signal to the generator-transformer differential protection device for use in protection logic judgment. This is implemented according to the following steps:

[0049] Step 4.1: Based on the warning level output in Step 3.4, perform tiered response and diagnosis:

[0050] Step 4.1.1, Attention Level: Only display status prompts on the monitoring system's human-machine interface, without triggering external alarms.

[0051] Step 4.1.2, Warning Level: Issue an audible and visual alarm in the monitoring backend and automatically retrieve relevant online transformer oil chromatography monitoring data and partial discharge monitoring data. This includes monitoring for abnormal grounding current characteristics and characteristic gases in the oil chromatography (such as...). The content of the content and the amplitude-phase spectrum of the partial discharge were correlated and cross-validated to form a preliminary multi-source information fusion diagnostic report, which was then pushed to the operators.

[0052] Step 4.1.3, Hazard Level: Issue the highest level alarm and immediately execute the protection collaboration decision logic of Step 4.2.

[0053] Step 4.2: When the warning level reaches the danger level, the system activates the following protection coordination logic:

[0054] Step 4.2.1, Fault Feature Database Comparison: Quickly match and compare the abnormal ground current features that trigger the alarm (such as specific harmonic combinations and sudden increase waveforms) with the pre-stored "Typical Fault Current Feature Database of Transformer" (which includes simulation and measured feature maps of faults such as inter-turn short circuit, core grounding, and winding deformation).

[0055] Step 4.2.2, Criterion Generation Logic: If the comparison result is "match failed," meaning the current abnormal characteristics do not match the typical characteristics of internal transformer faults, then the current abnormality is determined to originate from a problem in the external circuit of the cable shield. The system then generates an auxiliary criterion signal with a clear logical meaning to prevent maloperation. This signal is configured to perform one or more of the following operations: request the generator-transformer differential protection device to temporarily block the relevant protection section, increase its operating current setting value, or add a settable short delay.

[0056] Step 4.2.3, Criterion Transmission and Execution: The anti-maloperation auxiliary criterion signal is transmitted reliably and in real time to the generator-transformer differential protection device via hard-contact output or a GOOSE communication message conforming to the IEC 61850 standard. After receiving this signal, the generator-transformer differential protection device incorporates it as a high-priority auxiliary criterion into its protection logic, thereby achieving accurate anti-maloperation.

[0057] Step 4.3, Full-process data archiving: The original current waveform segments, calculated characteristic quantities, early warning event records, multi-source fusion diagnostic reports, generated anti-maloperation judgment signals and transmission timing of the above process are synchronously written into the power plant's existing fault recording device through a standard data interface to form a complete "intelligent diagnostic recording file" with in-depth diagnostic information for post-event traceability and in-depth analysis.

[0058] The method for establishing the dynamic reference model is as follows: during the system's healthy operation phase, the reference waveform, reference RMS value, and reference harmonic content of the grounding current under different operating condition combinations are learned and stored. This is implemented according to the following steps:

[0059] Step A.1: Health Data Collection and Preparation

[0060] Step A.1.1: Determine the healthy operation phase: Select a stable operating period for the generator set with no alarms or fault records for a long period (e.g., 1-3 consecutive months) as the "learning period" of the model.

[0061] Step A.1.2: Synchronously collect all data: During this learning period, the following target data and operating condition label data will be collected synchronously and continuously. The target data is the raw waveform data of the grounding current of the shield wires of the three-phase cables A, B, and C (high sampling rate, such as above 4kHz); the operating condition label data is the active power of the unit at the same moment. System voltage Ambient temperature It can also be extended to humidity, load power factor, etc.

[0062] Step A.1.3, Data Preprocessing and Storage: Perform the same filtering and calibration preprocessing on the collected raw current data as on real-time diagnostics, and align it with timestamp and operating condition label data to form a structured historical health dataset.

[0063] Step A.2, Working Condition Division and Data Tagging:

[0064] Step A.2.1: Define operating condition dimensions and intervals: Divide continuous operating condition parameters into discrete intervals with engineering significance. For example: active power. Classified as "low load" "Medium load" "High load" Three temperature ranges. Classified as "low temperature" ), "normal temperature" ( ),"high temperature"( Three intervals.

[0065] Step A.2.2: Combine and label the operating conditions: Combine the above intervals to form typical operating condition combination labels, such as "high load - high temperature" and "medium load - normal temperature".

[0066] Step A.2.3, Data Classification: Each data point in the historical health dataset is automatically classified into the corresponding working condition combination label based on its actual working condition parameter values ​​at the time of collection.

[0067] Step A.3, Feature Learning and Benchmark Extraction:

[0068] Step A.3.1, Reference Waveform Extraction: Align and average all current waveforms within this subset to calculate a representative average waveform, which serves as the reference waveform for this operating condition. Simultaneously, its upper and lower fluctuation envelopes can be calculated to define the normal fluctuation range.

[0069] Step A.3.2, Calculation of reference RMS value: Statistically calculate the RMS value of the current within this subset. The distribution is analyzed, and its statistical median or mean is taken as the benchmark effective value, and its standard deviation is calculated. To define the normal fluctuation range of valid values ​​(such as the benchmark value) ).

[0070] Step A.3.3, Calculation of reference harmonic content: Perform spectrum analysis on each waveform in the subset, statistically analyze the distribution of the amplitude of each harmonic (e.g., 3rd, 5th, 7th), take the statistical median as the reference harmonic content, and calculate its fluctuation range in the same way.

[0071] Step A.4, Model Encapsulation and Storage: The reference waveform and its envelope, reference RMS value and its fluctuation range, and reference harmonic content and its fluctuation range calculated for each operating condition combination in Step A.3 above are collectively encapsulated into a structured, queryable database or data model, namely the "dynamic reference model". The dynamic reference model uses the operating condition combination label as an index and can quickly retrieve and output the corresponding complete set of reference data based on the input real-time operating condition parameters.

[0072] One of the core technologies of the intelligent diagnosis of this invention is to use the isolated forest algorithm as an unsupervised machine learning model to quickly and accurately identify potential abnormal states from real-time ground current feature data.

[0073] Isolation Forest is an anomaly detection algorithm based on ensemble learning. Its core idea is that anomalous data points, due to their scarcity and distinctiveness, are more easily isolated within a randomly partitioned feature space. Compared to supervised learning models that require large amounts of labeled data, Isolation Forest, as an unsupervised model, only needs to be trained using the learning data of a dynamic benchmark model. It eliminates the need to collect and label difficult-to-obtain fault samples, making it particularly suitable for the engineering scenario described in this invention, where "fault samples are scarce, and the goal is to discover unknown anomaly patterns." It also boasts high computational efficiency, making it suitable for online real-time monitoring.

[0074] After establishing the dynamic baseline model, the Isolation Forest model is trained using its corresponding historical health dataset. The specific steps are as follows:

[0075] Step B.1, Training Sample Construction: Extract a large number of data segments (e.g., tens of thousands to hundreds of thousands) from the historical health dataset. For each data segment, calculate its multi-dimensional feature vector compared with the synchronous dynamic benchmark model as a training sample. This feature vector typically includes the deviation of the effective value of each phase current from the benchmark value, the deviation of the main harmonic content from the benchmark value, and waveform similarity indicators (such as DTW distance).

[0076] Step B.2, Forest Construction: Set two key parameters for the isolated forest: the number of subsamples (e.g., 256) and the number of trees (e.g., 100). Repeatedly and randomly sample subsets from the entire training dataset. For each subset, recursively and randomly select a feature dimension and a split point between the maximum and minimum values ​​of that dimension to construct an "isolated tree." Recursively split until a data point is isolated or the maximum depth of the tree is reached. Repeat the above process to construct a "forest" composed of multiple isolated trees. Healthy data requires a longer, on average, path to isolation in the forest, while outliers have shorter paths.

[0077] Step B.3, Online Model Application and Anomaly Scoring: In real-time monitoring, for each newly calculated real-time multidimensional feature vector... The model is then traversed using a pre-trained Isolation Forest model to calculate its anomaly score. ;

[0078] Step B.3.1, Path length calculation: For feature vectors Record the path length it takes from the root node to the isolated leaf node in each isolated tree. .

[0079] Step B.3.2, Anomaly Score Calculation: Based on the average path length of all trees. The abnormal score is calculated using the following formula. :

[0080]

[0081] in, It is the subsampling size. It is given The average normalization factor for the path length over time. Score The closer a value is to 1, the higher the probability that the sample is abnormal; the closer a value is to 0, the more normal the sample is.

[0082] Step B.3.3, Anomaly Detection: Set an anomaly threshold. (e.g., 0.6). When the real-time feature vector abnormal scores When the model determines that an "immediate anomaly" has occurred in the current state, it triggers an anomaly alarm and outputs the result to the comprehensive diagnostic module.

[0083] To ensure the continued effectiveness of the isolated forest model during long-term operation, this invention establishes a corresponding model update mechanism. This mechanism mainly includes two parts: First, periodic validation, i.e., every quarter or every six months, the model is validated using newly generated confidence health data, and its false positive rate is calculated; if the false positive rate increases significantly, it indicates that the model's understanding of the current system's "new normal" may have deviated. Second, rolling retraining, triggered when one of the following two situations occurs: 1. The dynamic baseline model completes a major update (e.g., expanding to cover new operating conditions); 2. The aforementioned periodic validation results indicate a decline in model performance. At this time, the system will use the updated, more representative historical health dataset to fully retrain the isolated forest model, thereby synchronizing its internal decision boundary with the latest health status characteristics of the system, maintaining the accuracy and reliability of its anomaly detection.

[0084] By employing the Isolation Forest model as an unsupervised anomaly detector, this invention significantly improves the performance and practicality of the diagnostic system. This model can learn normal states solely from historical health data without relying on any predefined fault modes, thus keenly capturing and effectively identifying unknown anomalies deviating from the baseline. It exhibits high sensitivity to weak but persistent anomaly patterns, enabling early fault warnings. Simultaneously, this approach greatly lowers the engineering implementation threshold, eliminating the need for difficult-to-obtain fault samples for model training. Furthermore, the Isolation Forest algorithm boasts fast inference speed and low computational overhead, fully meeting the stringent requirements of online real-time monitoring.

[0085] The intelligent diagnostic system of this invention includes a trend prediction model, which aims to quantitatively track and provide early warning of the slow degradation process of insulation status, making up for the shortcomings of unsupervised models that may be insensitive to gradual changes, and achieving true predictive maintenance.

[0086] Model Positioning and Input Feature Selection: The trend prediction model is a time series prediction model. Its core task is to predict the short-term trajectory of grounding current based on historical data sequences of key grounding current characteristics. Typically, the "effective value of the three-phase total grounding current" is selected as the most comprehensive indicator of the grounding current level. "As the primary prediction target, it can also simultaneously predict key derived features such as the content of specific subharmonics (e.g., the 3rd harmonic) to provide a more comprehensive trend view."

[0087] Model building and training: The autoregressive integral moving average (ARIMA) model is adopted as a preferred implementation scheme for the trend prediction model.

[0088] Step C.1: Extract target features (such as...) from the historical health database and long-term monitoring data after the system was put into operation. The time series is a series of data points at equal intervals (e.g., one data point per hour). Before training, the series needs to be tested for stationarity (e.g., ADF test) and subjected to necessary differencing to eliminate long-term trends or periodicity and obtain a stationary series.

[0089] Step C.2: By analyzing the autocorrelation function (ACF) and partial autocorrelation function (PACF) plots of stationary sequences, the order of the ARIMA model is preliminarily identified. ,in: : Order of the autoregressive term. The minimum difference order required to make the sequence stationary. The order of the moving average term. Using methods such as grid search, the optimal combination of parameters that minimizes the Akaike Information Criterion (AIC) is found within a predetermined range. The maximum likelihood estimation method is used to fit the model parameters, thus completing the model training.

[0090] Step C.3, Online Prediction and Early Warning: After the model is put into online operation, the following process will be followed:

[0091] Step C.3.1, Rolling Prediction: Based on the current time... Based on the past Using a time window (e.g., data from the past 72 hours) as input, it drives a pre-trained ARIMA model to predict the future. A time window (e.g., the next 24 hours) value sequence .

[0092] Step C.3.2, Trend Analysis and Early Warning Judgment: Compare the predicted sequence with the slow-changing early warning threshold. This threshold is based on the baseline effective value of the corresponding operating condition in the dynamic baseline model. And its historical normal fluctuation range setting, for example, can be set to or Warning trigger condition: If the values ​​of multiple consecutive prediction points (e.g., more than 12 points within the next 24 hours) exceed the slow-change warning threshold, a clear insulation degradation trend is determined. Warning output: The model triggers an "insulation degradation trend warning," outputting the corresponding warning level (usually "attention" or "warning"), and the prediction curve and threshold line can be visualized.

[0093] Step C.4, Model Update and Maintenance: To ensure prediction accuracy, the following update mechanism is established: Sliding Window Retraining: Every certain period (e.g., monthly), the latest, validated healthy data is added to the training set, while the earliest data is removed, maintaining a fixed training set window length. The ARIMA model is then retrained using the updated data to capture the latest dynamic characteristics of the system. Performance Monitoring: The model's predicted values ​​are compared with the actual observed values, and metrics such as the Mean Absolute Percentage Error (MAPE) are calculated. If the error continues to increase and exceeds the allowable range, model retraining is triggered immediately.

[0094] By introducing and implementing a trend prediction model, this invention can issue early warnings several hours to several days in advance, before the actual value of the grounding current exceeds the standard, but before the development trajectory clearly points to a fault. This provides a valuable time window for arranging preventative maintenance. Furthermore, it complements the isolated forest model in the time dimension, jointly constructing a complete state assessment system covering both "short-term anomalies" and "long-term trends."

[0095] The following is a combination of appendices Figure 2 This paper describes the implementation of the "Intelligent Diagnosis and Anti-maloperation System for Grounding Current of GIS Shielded Wire" described in this invention. This system is used to implement the aforementioned method, and its core lies in the synergy between hardware modules and software functions to achieve a closed loop of monitoring, diagnosis, early warning, and protection.

[0096] Overall system composition: such as Figure 2 As shown, the system physically comprises four main functional modules: a monitoring and acquisition module, an intelligent diagnostic module, a collaborative execution module, and a data interface module. These modules are connected via industrial communication networks and signal cables, and are integrated with the power plant's existing generator-transformer differential protection devices and fault recording devices.

[0097] The monitoring and acquisition module includes high-precision micro-current transformers installed on the grounding leads of the shielded cables of phases A, B, and C to collect instantaneous grounding current signals. The secondary outputs of the transformers are connected to a nearby signal conditioning unit, which includes a multi-channel synchronous acquisition circuit capable of filtering, analog-to-digital conversion, and adding time stamps to the signals. Simultaneously, this module obtains the unit's active power, system voltage, and ambient temperature as operating parameters from the power plant monitoring system via a communication interface. The processed data is then uploaded to the intelligent diagnostic module via a communication line.

[0098] The intelligent diagnostic module is deployed in an industrial computer or a dedicated embedded device. Its software system includes a dynamic benchmark model database for storing data such as the benchmark waveform, benchmark RMS value, and benchmark harmonic content of the grounding current under different operating conditions; a feature calculation unit for real-time calculation of features such as the RMS value of the grounding current, harmonic content, and waveform similarity; a model runtime environment for loading and running pre-trained isolated forest and trend prediction models. The former performs anomaly detection on real-time features, while the latter performs trend prediction on the time series of key quantities such as the RMS value; and a diagnostic fusion unit that integrates the model output results with feature deviations, combined with operating parameters, to generate insulation status assessment results and early warning levels.

[0099] The collaborative execution module is responsible for executing early warnings and generating protection commands. Specifically, based on the early warning level output by the diagnostic module, it publishes graded information (attention, warning, danger) on the power plant monitoring backend. When the warning level or above is reached, data such as transformer oil chromatography can be retrieved for auxiliary diagnosis. Specifically, when the early warning level is "dangerous" and the abnormality is determined to be not an internal transformer fault, an auxiliary judgment criterion signal to prevent false tripping is generated. This module is equipped with a hard-wired output interface (relay) and / or a network interface supporting IEC 61850 GOOSE communication to send this signal to the generator-transformer differential protection device.

[0100] The data interface module is integrated in software form. It packages key data such as the original current waveform of the triggering event, the calculated characteristic quantity, the early warning event record and the anti-maloperation judgment signal according to a standard format (such as COMTRADE format), and writes them to the power plant's existing fault recording device through the network to form a traceable intelligent diagnostic recording file.

[0101] The present invention has been further described above with reference to specific embodiments. However, it should be understood that the specific description herein should not be construed as limiting the nature and scope of the present invention. Various modifications made to the above embodiments by those skilled in the art after reading this specification are all within the scope of protection of the present invention.

Claims

1. A method for intelligent diagnosis and prevention of malfunction of grounding current in GIS shielded wires, characterized in that, Includes the following steps: Step 1: Collect the grounding current signal of the cable shield grounding circuit between the GIS and the main transformer in real time, as well as the associated unit operating condition parameters, and preprocess the grounding current signal; Step 2: Based on the preprocessed grounding current signal, calculate the characteristic quantity of the current grounding current, and compare the characteristic quantity with the dynamic benchmark model established based on historical health data to obtain the characteristic deviation. Step 3: Input the feature quantity and the feature deviation into the pre-trained intelligent diagnostic model, and the intelligent diagnostic model outputs the insulation status assessment result and the warning level; The intelligent diagnostic model includes at least an unsupervised machine learning model for identifying transient abnormal patterns; Step 4: Issue graded early warning information according to the early warning level; When the early warning level reaches the preset danger threshold, generate an anti-maloperation auxiliary criterion signal and send the signal to the generator-transformer differential protection device for its use in performing protection logic judgment.

2. The intelligent diagnosis and anti-maloperation method for grounding current of GIS shielded wire according to claim 1, characterized in that, In step 1, the associated unit operating condition parameters include one or more of the unit active power, system voltage, and ambient temperature; the preprocessing includes filtering and calibration.

3. The intelligent diagnosis and anti-maloperation method for grounding current of GIS shielded wire according to claim 1, characterized in that, In step 2, the characteristic quantities include one or more of the following: the effective value of the grounding current, the zero-sequence current component, the content of a specific harmonic, waveform statistical characteristics, and the similarity distance with the reference waveform in the dynamic reference model.

4. The intelligent diagnosis and anti-maloperation method for grounding current of GIS shielded wire according to claim 1, characterized in that, In step 2, the method for establishing the dynamic reference model is as follows: during the healthy operation phase of the system, the reference waveform, reference effective value and reference harmonic content of the grounding current under different operating condition combinations are learned and stored.

5. The intelligent diagnosis and anti-maloperation method for grounding current of GIS shielded wire according to claim 1, characterized in that, In step 3, the intelligent diagnostic model also includes a trend prediction model, which is used to predict the changing trend of the grounding current based on the time series of the effective value of the grounding current, and to trigger an insulation deterioration trend warning when the predicted value exceeds the slow change warning threshold.

6. The intelligent diagnosis and anti-maloperation method for grounding current of GIS shielded wire according to claim 1 or 5, characterized in that, In step 3, the unsupervised machine learning model is the Isolation Forest model.

7. The intelligent diagnosis and anti-maloperation method for grounding current of GIS shielded wire according to claim 1, characterized in that, In step 4, the graded early warning information includes attention level, warning level and danger level; when the early warning level is warning level or danger level, the method further includes the step of: retrieving and correlating transformer oil chromatography data or partial discharge monitoring data to perform multi-source information fusion diagnosis.

8. The intelligent diagnosis and anti-maloperation method for grounding current of GIS shielded wire according to claim 1, characterized in that, In step 4, the specific steps for generating the anti-maloperation auxiliary criterion signal include: matching and comparing the abnormal characteristics of the current grounding current with the pre-stored transformer internal fault current characteristic library; if the matching fails, it is determined that the current abnormality originates from a problem outside the cable shield, and an anti-maloperation auxiliary criterion signal is generated to block differential protection or increase its action delay.

9. A GIS shielded wire grounding current intelligent diagnosis and anti-maloperation system implementing the method described in any one of claims 1 to 8, characterized in that, include: The monitoring and acquisition module is used to collect grounding current signals and unit operating parameters in real time. The intelligent diagnostic module, connected to the monitoring and acquisition module, is used to store the dynamic benchmark model, calculate the feature quantity, run the intelligent diagnostic model, and output the insulation status assessment result and early warning level. The collaborative execution module, connected to the intelligent diagnostic module, is used to issue warnings according to the warning level, and generate and send the anti-maloperation auxiliary criterion signal to the generator-transformer differential protection device when the danger threshold is reached. The data interface module is used to write the feature data, early warning events and related current waveform slices output by the intelligent diagnostic module into the power plant fault recording device.

10. The intelligent diagnostic and anti-maloperation system for grounding current of GIS shielded wires according to claim 9, characterized in that, The monitoring and acquisition module includes a high-precision micro-current transformer and a signal conditioning unit installed on the grounding circuit of the shielded wires of the three-phase cables A, B, and C; the collaborative execution module and the generator-transformer differential protection device transmit signals through hard wiring or GOOSE communication messages based on the IEC 61850 standard.

Images