A hierarchical early warning method and system for transformer operating state monitoring

Abstract

The application discloses a kind of transformer operating state monitoring hierarchical early warning method and system, through real-time gateway machine, service gateway machine, intelligent patrol host and communication protocol real-time collection transformer multi-source operating data, after the data collected are washed, standardized and feature extraction, input difference monitoring model for comprehensive evaluation, based on equipment account and substation three-dimensional GIS model establishes including horizontal comparison and longitudinal trend analysis difference monitoring model, transformer operating state is divided into normal, attention, abnormal, serious four grades and generates hierarchical early warning information, difference is shown on the two-dimensional or three-dimensional visualization interface of centralized control system, finally executes early warning push and closed-loop disposal.The application realizes the overall perception, accurate early warning and closed-loop management of transformer operating state, improves the early warning accuracy and operation and maintenance efficiency, and is suitable for substation centralized monitoring scene.

Description

Technical Field

[0001] This invention relates to a power system automation and intelligent monitoring technology, and more particularly to a hierarchical early warning method and system for monitoring transformer operating status based on differential analysis. Background Technology

[0002] Transformers are the core equipment of substations, and their operating status directly affects the safety of the power grid. With the in-depth advancement of the construction of "transparent substations", the centralized control system's ability to perceive the transformer status is constantly improving, but there are still the following shortcomings in differentiated monitoring and hierarchical early warning: (1) The data acquisition dimension is single, relying only on electrical quantity parameter over-limit alarms, lacking comprehensive acquisition of non-electrical quantity parameters such as oil temperature, tap position, and cooler status; (2) The early warning method is crude, based on a single parameter to set a threshold, lacking horizontal comparison between transformers of the same model and vertical trend analysis of a single transformer, which is prone to missed and false alarms; (3) The display method is single, and lists or simple graphics are difficult to intuitively present the differences in transformer status, making it impossible to quickly locate anomalies; (4) The early warning and handling are disconnected, lacking linkage with the operation and maintenance system, and the handling process cannot be tracked, making it difficult to form a closed-loop management. Therefore, there is an urgent need for a transformer operating status monitoring and hierarchical early warning method based on differentiated analysis to achieve collaborative analysis and accurate early warning of multi-source data. Summary of the Invention

[0003] Purpose of the invention: The purpose of this invention is to provide a hierarchical early warning method that enables comprehensive perception, accurate early warning and closed-loop management of transformer operating status monitoring. Another purpose of this invention is to provide a hierarchical early warning system that implements this method.

[0004] Technical solution: The present invention provides a graded early warning method for monitoring the operating status of a transformer, comprising:

[0005] (1) Real-time acquisition of multi-source operating data of transformers through real-time gateway, service gateway, intelligent inspection host and communication protocol; the multi-source operating data includes electrical quantity data, non-electrical quantity data, status quantity data and environmental data;

[0006] (2) Perform data preprocessing and feature extraction on multi-source operating data to extract transformer state feature parameters; the state feature parameters include oil temperature feature value, load feature value, tap position feature value, cooler state feature value and on-load tap changer action feature value;

[0007] (3) Based on the transformer equipment ledger and the three-dimensional GIS model of the substation, a differentiated monitoring model for transformers is established; the differentiated monitoring model includes horizontal comparative analysis and vertical trend analysis.

[0008] (4) Input the state characteristic parameters into the differentiated monitoring model for comprehensive evaluation, calculate the state evaluation score of the transformer, and classify the transformer operating status into normal state, warning state, abnormal state and severe state according to the state evaluation score;

[0009] (5) Generate graded early warning information based on the classified status levels; the graded early warning information includes early warning level and early warning type;

[0010] (6) The transformer status is displayed differently on the two-dimensional or three-dimensional visualization interface of the centralized control system. Different colors and rendering methods are used to distinguish the display according to the status level, and abnormal parameters are highlighted and located.

[0011] (7) Implement early warning information push and closed-loop handling, push the graded early warning information to the terminal, automatically generate handling work orders, track the handling progress, and verify the effect after the handling is completed, forming a closed-loop management of early warning, handling and verification.

[0012] In step (1), the electrical quantity data includes the transformer three-phase voltage, three-phase current, active power, reactive power, power factor and frequency; the non-electrical quantity data includes top oil temperature, bottom oil temperature, winding temperature and oil level; the status quantity data includes on-load tap changer position, cooler start / stop status, fan running status, oil pump running status and gas relay status; the environmental data includes ambient temperature, ambient humidity and transformer room temperature.

[0013] In step (3), the horizontal comparison analysis is used to compare and analyze the operating parameters of transformers of the same model in the same substation. Taking transformers of the same model in the same substation as the comparison object, under the condition that the load rate difference is less than the set threshold, the oil temperature, winding temperature and operating status parameters of each transformer are compared and analyzed. When the parameter deviation exceeds the set threshold, the abnormality is automatically marked and an early warning is triggered. The vertical trend analysis is used to analyze the trend of historical operating data of a single transformer. Based on the historical data of a single transformer, the periodic sliding window method is used to analyze the parameter change trend. The exponential weighted moving average control chart method, cumulative deviation auxiliary criterion and isolated forest are used to perform multi-dimensional joint anomaly detection. When a nonlinear mutation of the parameter or a continuous deviation from the historical mean exceeds the set multiple standard deviation is detected, the abnormality is automatically marked and an early warning is triggered.

[0014] In step (4), the comprehensive evaluation of the differentiated surveillance model includes the following steps:

[0015] (41) Standardized scoring of indicators: standardized scores of each dimension The range normalization method is used to calculate the score based on the degree to which the current parameter values ​​of each dimension deviate from the benchmark value. The formula is as follows:

[0016]

[0017] in, For the standardized score of the j-th dimension, This is the current real-time parameter value. This serves as the baseline value for this dimension. This represents the maximum permissible deviation for this dimension;

[0018] (42) Dynamic weight calculation: The weights of each dimension are determined by a dynamic weighting method that combines the analytic hierarchy process (AHP) and the entropy weighting method, and are corrected in combination with the ambient temperature.

[0019] (43) Overall score: The overall score is calculated using the weighted summation formula;

[0020] (44) Classification: Based on the comprehensive evaluation score, the operating status of the transformer is divided into four levels.

[0021] In step (5), the graded early warning information is divided into severe early warning, abnormal early warning and attention early warning according to the severity; the early warning types include transformer oil temperature abnormal early warning, transformer oil temperature difference early warning, on-load tap changer missliding early warning, on-load tap changer operation exceeding the limit early warning, transformer cooler abnormal early warning, transformer low load high temperature operation abnormal early warning and transformer oil temperature transmitter abnormal early warning.

[0022] In step (6), the two-dimensional visualization display includes using different colors to mark the transformer status on the main wiring diagram of the centralized control system; the three-dimensional visualization display includes dynamically rendering the transformer model in a three-dimensional scene, using green, yellow, orange and red colors to display the status level respectively, and highlighting abnormal components in the three-dimensional GIS model of the substation; the differentiated display also supports displaying the real-time values ​​and historical change trends of key transformer parameters in the form of dashboards, trend curves, bar charts and heat maps.

[0023] In step (7), the closed-loop management includes the following steps:

[0024] (71) After the graded early warning information is generated, it is pushed to the monitoring terminal in real time through the centralized control system message bus, and displayed in the alarm window and pop-up window in a graded manner, and pushed to the mobile operation and maintenance terminal simultaneously;

[0025] (72) The centralized control system automatically associates the transformer equipment ledger, historical defect records and maintenance records to generate a handling work order containing abnormal parameters, fault cause analysis and standard handling suggestions;

[0026] (73) The work order is transferred to the operation and maintenance personnel through the intelligent operation and maintenance platform. After the operation and maintenance personnel handle the work on site, they report the handling results. The centralized control system collects the transformer data after the handling to verify the handling effect.

[0027] (74) Feed back the handling results and verification data to the early warning model, and use the Bayesian update method to dynamically adjust the model parameters and early warning thresholds to achieve continuous optimization of the early warning model.

[0028] In step (74), the continuous optimization of the early warning model adopts the Bayesian update method, constructs a feedback dataset D from the disposal results and validation data, and uses maximum a posteriori estimation. Dynamically adjust key parameter vectors ;in, These are the baseline weight parameters in the AHP-entropy weight method. The thresholds for attention status, abnormal status, and severe status are specified.

[0029] The present invention discloses a graded early warning system for monitoring the operating status of a transformer, comprising:

[0030] Data acquisition module: used to acquire multi-source operating data of transformers in real time through real-time gateway, service gateway, intelligent inspection host and communication protocol;

[0031] Data processing and feature extraction module: used for cleaning, standardizing and extracting features from multi-source operational data;

[0032] Differentiated monitoring model construction module: used to establish a differentiated monitoring model based on equipment ledgers and substation 3D GIS models, including horizontal comparative analysis and vertical trend analysis;

[0033] Anomaly detection module: includes EWMA control chart detection unit, CUSUM criterion detection unit and isolated forest joint detection unit, used to perform multi-dimensional anomaly detection on state feature parameters and output the anomaly detection results to state comprehensive evaluation module;

[0034] Dynamic weighting module: includes AHP initial weight calculation unit, entropy weight method real-time correction unit and ambient temperature correction unit, used to determine the weight of each assessment dimension in the state comprehensive assessment, and output the corrected comprehensive weight to the state comprehensive assessment module;

[0035] The comprehensive condition assessment module is used to calculate the comprehensive score of the transformer and classify its level based on the anomaly detection results and dynamic weights.

[0036] The graded early warning module is used to generate three levels of early warning information based on the severity of the status level determined by the status assessment module.

[0037] Differentiated display module: Used to differentiate the transformer status on the centralized control system's visualization interface;

[0038] Closed-loop management module: includes an early warning push unit, a work order generation unit, and a Bayesian feedback self-optimization unit, used to execute early warning information push and closed-loop processing.

[0039] The connection relationships among the modules are as follows: the data acquisition module is connected to the data preprocessing and feature extraction module; the data preprocessing and feature extraction module is connected to the differentiated monitoring model construction module and the anomaly detection module, respectively; the anomaly detection module and the dynamic weighting module are connected to the state comprehensive evaluation module, respectively; the differentiated monitoring model construction module is connected to the state comprehensive evaluation module; the state comprehensive evaluation module is connected to the hierarchical early warning module and the differentiated display module, respectively; the hierarchical early warning module is connected to the closed-loop handling management module; the Bayesian feedback self-optimization unit in the closed-loop handling management module is connected to the dynamic weighting module and the state comprehensive evaluation module, respectively, forming a feedback optimization loop.

[0040] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: (1) In the longitudinal trend analysis, the EWMA control chart, CUSUM criterion and isolated forest triple joint detection are adopted to achieve full coverage detection of parameter mutation, continuous deviation and multi-parameter joint anomaly. The three have clear division of labor and complement each other; (2) Bayesian feedback optimization loop is introduced to make the warning parameters continuously self-optimize with the operating data, replacing the traditional fixed experience threshold; (3) By establishing a differentiated monitoring model that combines horizontal comparative analysis and longitudinal trend analysis, the multi-dimensional comprehensive evaluation and hierarchical warning of transformer operating status are realized. The triple joint anomaly detection method can effectively make up for the shortcomings of the single threshold method and improve the accuracy and robustness of the warning; (4) The intuitiveness and efficiency of anomaly location are improved by two-dimensional or three-dimensional differentiated display method; and the full-process traceability of transformer operation and maintenance is realized through the closed-loop management mechanism of warning-handling-verification, which effectively reduces the risk of equipment failure. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the transformer multi-source data acquisition and communication architecture of the present invention;

[0042] Figure 2 This is a schematic diagram of the transformer differential monitoring model structure of the present invention;

[0043] Figure 3 This is a schematic diagram illustrating the tiered early warning, differentiated display, and tiered push process of the present invention;

[0044] Figure 4 This is a flowchart illustrating the closed-loop management process for early warning and response in this invention.

[0045] Figure 5 This is a flowchart of the transformer condition comprehensive evaluation and algorithm optimization of the present invention. Detailed Implementation

[0046] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0047] Example 1

[0048] The specific implementation of the graded early warning method for monitoring transformer operating status provided by this invention includes:

[0049] (1) Multi-source data acquisition

[0050] like Figure 1 As shown, in this embodiment, through the centralized control system platform (Safety Zone I / II / IV), via the real-time gateway (IEC 104 protocol), service gateway (IEC 104, DL / T 860 protocol) and intelligent inspection host, multi-source operation data of the transformer is collected in real time, specifically including: (1) Electrical quantities: through protection devices, measurement and control devices, and intelligent terminals, the data is accessed through the station protocol to Safety Zone I to collect electrical parameters such as three-phase voltage, three-phase current, active power, reactive power, and power factor; (2) Non-electrical quantities: through oil temperature sensors, tap position sensors, partial discharge sensors, gas relays, etc., the data is collected to collect top oil temperature, bottom oil temperature, winding temperature, oil level, tap position, partial discharge, oil chromatography, and cooler status; (3) Status quantities: through the tap position sensor, the data is collected to collect non-electrical parameters such as three-phase voltage, three-phase current, active power, reactive power, and power factor. Position sensor, gas relay, temperature controller and cooler status feedback, collect on-load tap changer position, cooler start and stop status, fan running status, oil pump running status, gas relay status and protection signal; (4) Auxiliary sensing data: access camera video, drone / robot inspection image, infrared spectrum, three-dimensional point cloud and wireless sensing terminal data through intelligent inspection host, and collect fault waveform, security / fire protection signal, dynamic environment / cable trench monitoring data and model file at the same time; (5) Environmental data: collect ambient temperature, ambient humidity and transformer room temperature.

[0051] (2) Data preprocessing and feature extraction

[0052] The collected multi-source data is cleaned (zero values, negative values, and obvious outliers are removed; obvious outliers are values ​​that deviate from the sliding window mean by more than 3 times the standard deviation; missing data is filled with previous values; data segments that are missing for more than 2 hours are marked as invalid and not included in subsequent calculations), standardized (normalized to the [0,1] interval), and transformer status characteristic parameters are extracted, including: top oil temperature characteristic values ​​(current value, mean, rate of change, the rate of change is the average rate of change Δx / Δt in the most recent 1 hour), load characteristic values ​​(load rate, load change rate), tap characteristic values ​​(current tap, number of tap changes), cooler status characteristic values ​​(number of coolers in operation and running time), and on-load tap changer action characteristic values ​​(cumulative number of actions, recent action frequency).

[0053] (3) Establish a differentiated monitoring model

[0054] like Figure 2 As shown, a differentiated monitoring model is established based on the transformer equipment ledger and the substation 3D GIS model. The substation 3D GIS model and the equipment ledger serve as the basic data sources for the differentiated monitoring model. The differentiated monitoring model monitors the transformer operating status using the following two analysis methods:

[0055] (31) Horizontal comparative analysis: Taking the same type of transformer in the same substation as the comparison object, under the condition that the load rate difference is less than 10%, the oil temperature difference of each transformer is compared and analyzed. When the oil temperature difference between two transformers is greater than 10℃, the transformer with high oil temperature is automatically marked as abnormal and the "Transformer oil temperature difference warning" is triggered. Typical example: Two main transformers (#1 and #2) of the same type in a 220kV substation, with load rates of 62% and 58% respectively, the top oil temperature of the #1 main transformer is 65℃ and the top oil temperature of the #2 main transformer is 53℃, with a temperature difference of 12℃. The system automatically issues the "XX station transformer oil temperature difference warning" and suggests taking an oil sample for analysis of the #1 main transformer.

[0056] (32) Longitudinal trend analysis: Based on the historical data of a single transformer, the statistical characteristics (mean, standard deviation, and rate of change) of parameters such as transformer oil temperature and load rate are calculated using the periodic sliding window method (preferably 7 days with a step size of 1 hour). In the anomaly detection stage, the exponentially weighted moving average (EWMA) control chart method is introduced for online anomaly detection: Let the observed oil temperature at time t be... The EWMA statistic is defined as follows: The smoothing coefficient Preferred value: 0.2, initial value Take the average of historical data The control limits are set as follows:

[0057]

[0058] Where UCL is the upper control limit and LCL is the lower control limit. Let L be the standard deviation, and L be the control limit width (preferably 3). A change exceeding the UCL or LCL is considered a nonlinear abrupt change. A cumulative bias (CUSUM) statistic is also added as an auxiliary criterion: a statistical measure. ,in Let | be the oil temperature observation value at time i, when | |Exceeding the decision interval H ( 0.5 is preferred. H is preferred 5 When the value is 0, it is determined to be a persistent offset.

[0059] Furthermore, the Isolation Forest algorithm is used for collaborative detection of multi-parameter joint anomalies: a three-dimensional feature vector is constructed using top oil temperature, load rate, and gear shift frequency. ψ samples (preferably 256) are randomly sampled to build an isolation tree, and anomaly score s(x,ψ) is calculated. When s(x,ψ) > 0.6, it is marked as a multi-parameter joint anomaly. The longitudinal trend analysis comprehensively uses EWMA control charts, CUSUM criteria, and the Isolation Forest algorithm for multi-dimensional anomaly detection. The aforementioned EWMA control charts and CUSUM statistical criteria are used to replace or supplement traditional fixed threshold methods, improving the sensitivity and robustness of anomaly detection. When a nonlinear mutation of parameters is detected (such as an abnormal increase in oil temperature exceeding 5°C / hour in a short period of time) or a continuous deviation from the historical mean exceeding 2 standard deviations, anomalies are automatically marked and an early warning is triggered.

[0060] (4) Comprehensive status assessment and classification

[0061] like Figure 5 As shown, the extracted state feature parameters are input into the differentiated monitoring model for comprehensive evaluation, specifically including the following steps:

[0062] (41) Standardized scoring of indicators: standardized scores of each dimension The range normalization method is used to calculate the score based on the degree to which the current parameter values ​​of each dimension deviate from the benchmark value. The formula is as follows:

[0063]

[0064] in, For the standardized score of the j-th dimension, This is the current real-time parameter value. This serves as the baseline value for this dimension. This represents the maximum permissible deviation for this dimension.

[0065] (42) Dynamic weight calculation: The weights of each dimension are determined by a dynamic weighting method that combines the analytic hierarchy process (AHP) with the entropy weight method, and then corrected by the ambient temperature.

[0066] AHP initial weights: Based on a nine-level scaling method, professional technicians construct a judgment matrix by comparing five evaluation dimensions—oil temperature status, load status, on-load tap changer status, cooler status, and environmental corrections—pairwise. Calculate the normalized eigenvector corresponding to the largest eigenvalue of A as the initial weights. .

[0067] Real-time correction using entropy weighting: An evaluation matrix X = (xᵢⱼ)_(N×5) is constructed using the data of each dimension of indicators within a sliding time window N (preferably 72 hours). After standardization, the information entropy of the j-th indicator is calculated.

[0068]

[0069] in, .

[0070] The corrected weights are:

[0071]

[0072] in, The memory coefficient (preferably 0.6) balances the weight of prior knowledge with the entropy weight driven by real-time data. When the variability of a certain dimension increases, its information entropy... Reduce, entropy weight / Increase the sensitivity to abnormal dimensions to achieve adaptive enhancement.

[0073] Ambient temperature correction: Introduce an ambient temperature correction factor. / (β is preferably 0.1, The current ambient temperature. =20°C), the oil temperature weight automatically increases during high summer temperatures and decreases during low winter temperatures. The corrected overall weight satisfies =1, participates in the weighted calculation of the comprehensive evaluation score. Weighting configuration examples are as follows: Oil temperature status (base weight 0.3, automatic adjustment range 0.2~0.4), load status (base weight 0.15, adjustment range 0.1~0.2), on-load tap changer status (base weight 0.25, adjustment range 0.15~0.35), cooler status (base weight 0.2, adjustment range 0.15~0.3), environmental correction (base weight 0.1, adjustment range 0.05~0.15).

[0074] (43) Overall score: The overall score S is calculated using the weighted summation formula: ,in The comprehensive weights are dynamically adjusted using the AHP-entropy weighting method. Standardized scores are assigned to each dimension. S is a dimensionless value between 0 and 1, which is multiplied by 100 to obtain the overall percentage score.

[0075] (44) Classification: Based on the comprehensive evaluation score, the transformer operating status is divided into four levels:

[0076] Normal status (overall score ≥ 85 points): All parameters are within the normal range, indicated by green, no need to pay attention.

[0077] Note the status (70 points ≤ overall score < 85 points): Some parameters slightly deviate from the normal range, marked in yellow, requiring closer monitoring.

[0078] Abnormal status (50 points ≤ overall score < 70 points): Multiple parameters are obviously abnormal, marked in orange, and need to be dealt with in time.

[0079] Severe condition (overall score < 50 points or single parameter seriously exceeds the standard): The parameter seriously exceeds the standard or there is a major safety hazard. It is marked in red and needs to be dealt with immediately.

[0080] (5) Generate tiered early warning information

[0081] like Figure 3 As shown, graded early warning information is generated based on the status level, and is divided into the following categories according to severity:

[0082] Level I (Severe) Warning: Triggered when the transformer is in a severe condition, an alarm window will appear and a pop-up window will be automatically pushed, an accident audio message will be issued, and the message will also be pushed to the mobile terminal of the operation and maintenance personnel, requiring immediate action.

[0083] Level II (Abnormal) Warning: Triggered when the transformer is in an abnormal state, the warning window is displayed, and the warning information is pushed to the monitoring terminal, requiring timely handling.

[0084] Level III (Attention) Warning: Triggered when the transformer is in a warning state, displayed in the alarm window to prompt monitoring personnel to strengthen monitoring.

[0085] The warning types cover the following seven categories:

[0086] Transformer oil temperature anomaly warning: Based on longitudinal trend analysis, it is triggered when the oil temperature experiences a non-linear sudden change or continuously deviates from the historical threshold.

[0087] Transformer oil temperature difference warning: Based on horizontal comparative analysis, it is triggered when the oil temperature difference of transformers of the same model at the same station exceeds 10℃.

[0088] On-load tap changer mis-sliding warning: Based on the analysis of voltage ratio changes before and after tap change, it is triggered when the ratio change is lower than the mis-sliding threshold or higher than the tap-sliding threshold.

[0089] On-load tap changer operation count over-limit warning: Based on the statistical analysis of the cumulative number of operations, it is triggered when it approaches the standard threshold (such as 95% of the value specified in DL / T 574-2021) and corresponding maintenance strategy suggestions are given;

[0090] Transformer cooler abnormality early warning: Based on the correlation analysis between oil temperature trend and cooler start-up, it is triggered when the oil temperature reaches the start-up temperature but the cooler fails to start correctly, or when the cooler is operating under poor conditions.

[0091] Transformer low-load high-temperature operation abnormality warning: triggered when the transformer load rate is below 50% and the oil temperature is above 70℃;

[0092] Transformer oil temperature transmitter abnormality warning: Based on the analysis of the oil temperature curve shape, it is triggered when the oil temperature undergoes a non-linear sudden change (not caused by load change), indicating that the transmitter may be faulty.

[0093] (6) Two-dimensional or three-dimensional differentiated display

[0094] The transformer status is displayed in a differentiated manner on the centralized control system's visualization interface. In two-dimensional display, different colors (green / yellow / orange / red corresponding to four statuses) are used to identify transformer elements on the substation's main wiring diagram, supporting mouse hover to display key parameter panels. In three-dimensional display, the substation's 3D GIS model is dynamically rendered and colored in a 3D scene, using green, yellow, orange, and red according to the status level, and supporting highlighting and flashing prompts for abnormal components (such as areas with high oil temperature or abnormal gear positions). This differentiated display also supports showing the real-time values ​​and historical trends of key transformer parameters in various formats, including dashboards, trend curves, bar charts, and heat maps, achieving a "transparent" display of the transformer status.

[0095] (7) Closed-loop management of early warning and response

[0096] like Figure 4 As shown, the execution of early warning information push and closed-loop processing specifically includes:

[0097] (71) Early warning information push: The hierarchical early warning information is pushed to the monitoring terminal in real time through the centralized control system message bus and displayed hierarchically in the alarm window and pop-up window. At the same time, it is pushed to the mobile terminal of the operation and maintenance personnel through the mobile operation and maintenance platform.

[0098] (72) Automatically generate disposal work orders: The centralized control system automatically associates the transformer equipment ledger, historical defect records and maintenance records, and combines them with the current operating data to generate disposal work orders that include abnormal parameters, fault cause analysis and standard disposal suggestions.

[0099] (73) Handling Flow: Handling work orders are transferred to maintenance personnel through the intelligent operation and maintenance platform, supporting personnel assignment, progress tracking and result feedback. Monitoring personnel can view the handling progress in real time on the centralized control system.

[0100] (74) On-site handling and effect verification: After the maintenance personnel complete the on-site handling, the centralized control system automatically collects the transformer operation data after the handling and compares and analyzes it with the abnormal data before the warning to verify the handling effect. If the verification is successful, the work order is closed; otherwise, the handling process is retried.

[0101] (75) Model optimization and parameter adjustment: Feedback of the handling results and verification data to the early warning model to construct a feedback learning loop. After each closed-loop handling is completed, the verification results (including actual fault type, severity label and handling effect score η∈ [0,1]) are stored in the database according to confidence level. After accumulating M valid feedback data (M preferably 20 data), the model parameter optimization is automatically triggered once. The key parameter vector Θ=[w1~w5,T1,T2,T3] is adjusted by Bayesian update method, where w1~w5 are the baseline weight parameters in AHP-entropy weight method, and T1~T3 are the attention state threshold, abnormal state threshold and severe state threshold, respectively. Let the parameter prior distribution be p(Θ0)~N(μ0,σ0²), where μ0 is the prior mean and σ0² is the prior variance. Based on the newly arrived feedback dataset D={η1,η2,...,η m Construct the likelihood function p(D|Θ)=∏L(ηᵢ|Θ), where L(ηᵢ|Θ) is the likelihood function of a single feedback sample, and ∏ is the multiplication operator. Calculate the optimal values ​​of the parameters using maximum a posteriori estimation. After multiple iterations, the posterior distribution gradually converges, and the parameter configuration gradually approaches the global optimum, replacing the fixed empirical threshold. This enables the early warning model to continuously self-optimize and adaptively learn, thereby continuously improving the accuracy of early warnings.

[0102] This method is applied to a graded early warning system based on transformer operating status monitoring, which includes the following modules connected in sequence:

[0103] (1) Data acquisition module

[0104] This device is used to collect multi-source operating data of transformers in real time through a real-time gateway, service gateway, intelligent inspection host and communication protocols (IEC 104, DL / T860). The multi-source operating data includes electrical quantity data, non-electrical quantity data, status quantity data and environmental data, and transmits the collected data to the data preprocessing module.

[0105] (2) Data preprocessing and feature extraction module

[0106] This is used to perform data cleaning (removing zero values, negative values, and obvious outliers, and filling missing data with previous values) and standardization (normalizing to the [0,1] interval) on the collected multi-source operating data, and extract transformer state characteristic parameters. The state characteristic parameters include top oil temperature characteristic values ​​(current value, mean, and rate of change), load characteristic values ​​(load rate and load change rate), tap characteristic values ​​(current tap and number of tap changes), cooler state characteristic values ​​(number of coolers in operation and running time), and on-load tap changer action characteristic values ​​(cumulative number of actions and recent action frequency). The preprocessed characteristic parameters are then transmitted to the differentiated monitoring model construction module and the anomaly detection module, respectively.

[0107] (3) Differentiated monitoring model construction module

[0108] This is used to establish a differentiated monitoring model based on transformer equipment ledgers and substation 3D GIS models, including horizontal comparative analysis and vertical trend analysis. The horizontal comparative analysis uses transformers of the same model in the same substation as the comparison object. Under the condition that the load rate difference is less than 10%, the oil temperature and winding temperature differences of each transformer are compared one by one. When the temperature difference exceeds the threshold (greater than 10℃), it is automatically marked as abnormal. The vertical trend analysis uses the historical data of a single transformer as the benchmark and adopts the periodic sliding window method (window period of 7 days, step size of 1 hour) to calculate the statistical characteristics of parameters such as transformer oil temperature and load rate. The established differentiated monitoring model is then output to the status comprehensive evaluation module.

[0109] (4) Anomaly detection module

[0110] This module is used for multi-dimensional anomaly detection of state characteristic parameters and outputs the anomaly detection results to the state comprehensive evaluation module. The anomaly detection module includes an EWMA control chart detection unit, a CUSUM criterion detection unit, and an isolated forest joint detection unit.

[0111] (41) EWMA control chart detection unit, used to detect abrupt changes in parameters: Let the oil temperature observed at time t be... The EWMA statistic is defined as follows: ( (For smoothing coefficient), control limits are set to ( For standard deviation, To control the width limit), when A mutation exceeding the UCL or LCL is considered a nonlinear mutation.

[0112] (42) CUSUM criterion detection unit, used to detect the persistence of parameter shifts: statistics ( (where i is the oil temperature observation value at time i), when | | When the deviation exceeds the decision interval H, it is determined to be a persistent deviation.

[0113] (43) Isolated forest joint detection unit, used to detect multi-parameter joint anomalies: the top oil temperature, load rate and gear operation frequency constitute a three-dimensional feature vector, and an isolated tree is constructed by random sampling. When the anomaly score exceeds the threshold, it is marked as a multi-parameter joint anomaly.

[0114] (5) Dynamic weighting module

[0115] This module is used to determine the weights of each assessment dimension in the comprehensive state assessment and outputs the corrected comprehensive weights to the comprehensive state assessment module. The dynamic weighting module includes an AHP initial weight calculation unit, an entropy weight method real-time correction unit, and an ambient temperature correction unit.

[0116] (51) The AHP initial weight calculation unit is used to construct a judgment matrix based on the analytic hierarchy process and the nine-level scaling method for five evaluation dimensions: oil temperature state, load state, on-load tap changer state, cooler state and environmental correction. The normalized eigenvector corresponding to the largest eigenvalue of the judgment matrix is ​​calculated as the initial weight.

[0117] (52) The real-time correction unit using entropy weighting is used to construct an evaluation matrix from the indicator data of each dimension within the sliding time window, calculate the information entropy of each indicator, and the corrected weight is... ,in, This is the memory factor.

[0118] (53) Ambient temperature correction unit, used to introduce ambient temperature correction coefficient. / The weights are dynamically adjusted based on the ambient temperature.

[0119] (6) Status Comprehensive Assessment Module

[0120] To calculate the comprehensive transformer score and classify it based on anomaly detection results and dynamic weights: First, the standardized scores for each dimension are calculated using the range normalization method.

[0121]

[0122] Then calculate the overall score using the weighted summation formula. Finally, based on the comprehensive evaluation score, the transformer operating status is divided into four levels: normal status (≥85 points), warning status (70~85 points), abnormal status (50~70 points), and severe status (<50 points). The status level is then output to the graded early warning module and the differentiated display module.

[0123] (7) Tiered early warning module

[0124] Based on the status level determined by the status assessment module, three levels of early warning information are generated according to the severity: Level I (Severe) warning corresponds to a severe state, triggering a pop-up window and an accident voice message and pushing it to the operation and maintenance mobile terminal; Level II (Abnormal) warning corresponds to an abnormal state, displayed in the alarm window and pushed to the monitoring terminal; Level III (Attention) warning corresponds to an attention state, displayed in the alarm window in a hierarchical manner; the early warning information generated by the hierarchical early warning module is output to the differentiated display module and the closed-loop handling management module.

[0125] (8) Differentiated display module

[0126] This is used to differentiate the transformer status on the centralized control system's visualization interface: In terms of two-dimensional display, different colors are used to identify transformer elements on the substation's main wiring diagram (green / yellow / orange / red correspond to four statuses); in terms of three-dimensional display, the transformer is dynamically rendered and colored in the substation's three-dimensional GIS model, and it supports displaying the real-time values ​​and historical trends of key transformer parameters in various forms such as dashboards, trend curves, bar charts, and heat maps.

[0127] (9) Closed-loop disposal management module

[0128] This unit is used to perform early warning information push and closed-loop processing, including an early warning push unit, a work order generation unit, and a Bayesian feedback self-optimization unit.

[0129] (91) Early warning push unit, used to push graded early warning information to the monitoring terminal and operation and maintenance mobile terminal in real time through the centralized control system message bus.

[0130] (92) Work order generation unit, which is used to automatically associate transformer equipment ledger, historical defect records and maintenance records to generate work orders containing abnormal parameters, fault cause analysis and standard handling suggestions, and supports personnel assignment and progress tracking.

[0131] (93) Bayesian feedback self-optimization unit, used to feed back the treatment results and verification data to the early warning model, and use the Bayesian update method to adjust the key parameter vector Θ=[w1~w5,T1,T2,T3] (w1~w5 are the baseline weight parameters, T1~T3 are the thresholds of the three state levels), and use maximum a posteriori estimation. The system calculates the optimal values ​​of the parameters and feeds back the updated weights and thresholds to the dynamic weighting module and the state comprehensive evaluation module, respectively, to achieve continuous self-optimization of the early warning model.

[0132] The connections between the modules are as follows: the data acquisition module is connected to the data preprocessing and feature extraction module; the data preprocessing and feature extraction module is connected to the differentiated monitoring model construction module and the anomaly detection module, respectively; the anomaly detection module and the dynamic weighting module are connected to the state comprehensive evaluation module, respectively; the differentiated monitoring model construction module is connected to the state comprehensive evaluation module; the state comprehensive evaluation module is connected to the hierarchical early warning module and the differentiated display module, respectively; the hierarchical early warning module is connected to the closed-loop handling management module; the Bayesian feedback self-optimization unit in the closed-loop handling management module is connected to the dynamic weighting module and the state comprehensive evaluation module, forming a feedback optimization loop.

[0133] Example 2

[0134] This embodiment provides a transformer oil temperature difference early warning example: At a 500kV substation, two identical main transformers (#1 and #2) were operating under similar load conditions. The top oil temperature of transformer #1 remained consistently high, with a temperature difference of 12℃ compared to transformer #2. The system identified this anomaly through horizontal comparative analysis and issued an "XX Substation Transformer Oil Temperature Difference Early Warning (Level II)," simultaneously pushing the warning to the monitoring terminal and the maintenance mobile terminal. After reviewing historical curves and verifying their accuracy, the monitoring personnel notified maintenance personnel to inspect the site. The maintenance personnel found that some heat sink fins in the cooler of transformer #1 were blocked. After cleaning, the temperature difference returned to the normal range of 3℃. The system collected the data, verified the effectiveness of the treatment, and automatically closed the work order.

[0135] Example 3

[0136] This embodiment provides an example of an on-load tap changer misalignment warning: After an on-load tap change operation is performed on a 220kV main transformer, the system collects voltage ratio change data before and after the tap change, calculates the ratio difference, and compares it with the theoretical threshold. When the tap is changed from tap 3 to tap 4, the measured ratio change is only 40% of the theoretical value (below the misalignment judgment threshold K1). The system automatically issues a "XX transformer misalignment warning (Level II)," prohibiting remote tap change operations and simultaneously notifying maintenance personnel for inspection. On-site inspection confirms that the tap changer linkage is broken, causing the tap to not switch properly. A power outage is promptly arranged to prevent the fault from escalating.

[0137] Example 4

[0138] This embodiment provides an example of an on-load tap changer operation count early warning system: The system statistically analyzes the cumulative operation count of an on-load tap changer for a 110kV main transformer. When the cumulative operation count reaches 4750 times (95% of the 5000 times specified in the regulations for oil-immersed on-load tap changers), the system automatically issues a Level III early warning: "The on-load tap changer of the XX main transformer has accumulated 4750 operations. It is recommended to conduct tests on the moisture content and breakdown voltage of the insulating oil in the tap changer's oil chamber." Monitoring personnel notify maintenance and repair personnel, who then arrange for insulating oil testing. They discover that the moisture content of the insulating oil in the oil chamber exceeds the standard and promptly address the issue, preventing a tap changer failure.

Claims

1. A graded early warning method for monitoring the operating status of transformers, characterized in that, include: (1) Real-time collection of multi-source operation data of transformers through real-time gateway, service gateway, intelligent inspection host and communication protocol; The multi-source operating data includes electrical quantity data, non-electrical quantity data, status quantity data, and environmental data; (2) Perform data preprocessing and feature extraction on multi-source operating data to extract transformer state feature parameters; the state feature parameters include oil temperature feature value, load feature value, tap position feature value, cooler state feature value and on-load tap changer action feature value; (3) Based on the transformer equipment ledger and the three-dimensional GIS model of the substation, a differentiated monitoring model for transformers is established; the differentiated monitoring model includes horizontal comparative analysis and vertical trend analysis. (4) Input the state characteristic parameters into the differentiated monitoring model for comprehensive evaluation, calculate the state evaluation score of the transformer, and classify the transformer operating status into normal state, warning state, abnormal state and severe state according to the state evaluation score; (5) Generate graded early warning information based on the classified status levels; the graded early warning information includes early warning level and early warning type; (6) The transformer status is displayed differently on the two-dimensional or three-dimensional visualization interface of the centralized control system. Different colors and rendering methods are used to distinguish the display according to the status level, and abnormal parameters are highlighted and located. (7) Implement early warning information push and closed-loop handling, push the graded early warning information to the terminal, automatically generate handling work orders, track the handling progress, and verify the effect after the handling is completed, forming a closed-loop management of early warning, handling and verification.

2. The graded early warning method according to claim 1, characterized in that, In step (1), the electrical quantity data includes the transformer three-phase voltage, three-phase current, active power, reactive power, power factor and frequency; the non-electrical quantity data includes top oil temperature, bottom oil temperature, winding temperature and oil level; the status quantity data includes on-load tap changer position, cooler start / stop status, fan running status, oil pump running status and gas relay status; the environmental data includes ambient temperature, ambient humidity and transformer room temperature.

3. The graded early warning method according to claim 1, characterized in that, In step (3), the horizontal comparison analysis is used to compare and analyze the operating parameters of transformers of the same model in the same substation. Taking transformers of the same model in the same substation as the comparison object, under the condition that the load rate difference is less than the set threshold, the oil temperature, winding temperature and operating status parameters of each transformer are compared and analyzed. When the parameter deviation exceeds the set threshold, the abnormality is automatically marked and an early warning is triggered. The vertical trend analysis is used to analyze the trend of historical operating data of a single transformer. Based on the historical data of a single transformer, the periodic sliding window method is used to analyze the parameter change trend. The exponential weighted moving average control chart method, cumulative deviation auxiliary criterion and isolated forest are used to perform multi-dimensional joint anomaly detection. When a nonlinear mutation of the parameter or a continuous deviation from the historical mean exceeds the set multiple standard deviation is detected, the abnormality is automatically marked and an early warning is triggered.

4. The graded early warning method according to claim 1, characterized in that, In step (4), the comprehensive evaluation of the differentiated surveillance model includes the following steps: (41) Standardized scoring of indicators: standardized scores of each dimension The range normalization method is used to calculate the score based on the degree to which the current parameter values ​​of each dimension deviate from the benchmark value. The formula is as follows: in, For the standardized score of the j-th dimension, This is the current real-time parameter value. This serves as the baseline value for this dimension. This represents the maximum permissible deviation for this dimension; (42) Dynamic weight calculation: The weights of each dimension are determined by a dynamic weighting method that combines the analytic hierarchy process (AHP) and the entropy weighting method, and are corrected in combination with the ambient temperature. (43) Overall score: The overall score is calculated using the weighted summation formula; (44) Classification: Based on the comprehensive evaluation score, the operating status of the transformer is divided into four levels.

5. The graded early warning method according to claim 1, characterized in that, In step (5), the graded early warning information is divided into severe early warning, abnormal early warning and attention early warning according to the severity; the early warning types include transformer oil temperature abnormal early warning, transformer oil temperature difference early warning, on-load tap changer missliding early warning, on-load tap changer operation exceeding the limit early warning, transformer cooler abnormal early warning, transformer low load high temperature operation abnormal early warning and transformer oil temperature transmitter abnormal early warning.

6. The graded early warning method according to claim 1, characterized in that, In step (6), the two-dimensional visualization display includes using different colors to mark the transformer status on the main wiring diagram of the centralized control system; the three-dimensional visualization display includes dynamically rendering the transformer model in a three-dimensional scene, using green, yellow, orange and red colors to display the status level respectively, and highlighting abnormal components in the three-dimensional GIS model of the substation; the differentiated display also supports displaying the real-time values ​​and historical change trends of key transformer parameters in the form of dashboards, trend curves, bar charts and heat maps.

7. The graded early warning method according to claim 1, characterized in that, In step (7), the closed-loop management includes the following steps: (71) After the graded early warning information is generated, it is pushed to the monitoring terminal in real time through the centralized control system message bus, and displayed in the alarm window and pop-up window in a graded manner, and pushed to the mobile operation and maintenance terminal simultaneously; (72) The centralized control system automatically associates the transformer equipment ledger, historical defect records and maintenance records to generate a handling work order containing abnormal parameters, fault cause analysis and standard handling suggestions; (73) The work order is transferred to the operation and maintenance personnel through the intelligent operation and maintenance platform. After the operation and maintenance personnel handle the work on site, they report the handling results. The centralized control system collects the transformer data after the handling to verify the handling effect. (74) Feed back the handling results and verification data to the early warning model, and use the Bayesian update method to dynamically adjust the model parameters and early warning thresholds to achieve continuous optimization of the early warning model.

8. The graded early warning method according to claim 7, characterized in that, In step (74), the continuous optimization of the early warning model adopts the Bayesian update method, constructs a feedback dataset D from the disposal results and validation data, and uses maximum a posteriori estimation. Dynamically adjust key parameter vectors ; in, These are the baseline weight parameters in the AHP-entropy weight method. The thresholds for attention status, abnormal status, and severe status are specified.

9. A hierarchical early warning system for monitoring the operating status of transformers, characterized in that, include: Data acquisition module: used to acquire multi-source operating data of transformers in real time through real-time gateway, service gateway, intelligent inspection host and communication protocol; Data processing and feature extraction module: used for cleaning, standardizing and extracting features from multi-source operational data; Differentiated monitoring model construction module: used to establish a differentiated monitoring model based on equipment ledgers and substation 3D GIS models, including horizontal comparative analysis and vertical trend analysis; Anomaly detection module: includes EWMA control chart detection unit, CUSUM criterion detection unit and isolated forest joint detection unit, used to perform multi-dimensional anomaly detection on state feature parameters and output the anomaly detection results to state comprehensive evaluation module; Dynamic weighting module: includes AHP initial weight calculation unit, entropy weight method real-time correction unit and ambient temperature correction unit, used to determine the weight of each assessment dimension in the state comprehensive assessment, and output the corrected comprehensive weight to the state comprehensive assessment module; The comprehensive condition assessment module is used to calculate the comprehensive score of the transformer and classify its level based on the anomaly detection results and dynamic weights. The graded early warning module is used to generate three levels of early warning information based on the severity of the status level determined by the status assessment module. Differentiated display module: Used to differentiate the transformer status on the centralized control system's visualization interface; Closed-loop management module: includes an early warning push unit, a work order generation unit, and a Bayesian feedback self-optimization unit, used to execute early warning information push and closed-loop processing.

10. The graded early warning system according to claim 9, characterized in that, The connection relationship between the modules is as follows: the data acquisition module is connected to the data preprocessing and feature extraction module; The data preprocessing and feature extraction module is connected to the differentiated monitoring model construction module and the anomaly detection module, respectively; the anomaly detection module and the dynamic weighting module are connected to the state comprehensive evaluation module, respectively; the differentiated monitoring model construction module is connected to the state comprehensive evaluation module; the state comprehensive evaluation module is connected to the hierarchical early warning module and the differentiated display module, respectively; the hierarchical early warning module is connected to the closed-loop handling management module; the Bayesian feedback self-optimization unit in the closed-loop handling management module is connected to the dynamic weighting module and the state comprehensive evaluation module, respectively, forming a feedback optimization loop.

Images