Detection device and detection method for automatically identifying faults of oil-immersed transformer

Abstract

The invention discloses a detection device and a detection method for automatically identifying faults of an oil-immersed transformer, the fault detection device comprises a sensor array unit and a data processing unit, and the detection method comprises the following steps: collecting multi-source monitoring data of the oil-immersed transformer in real time; constructing multi-dimensional fused fault feature data associated with the multi-source data; processing the fault feature data subjected to fusion processing by using a random forest algorithm to obtain a feature subset and a primary early warning signal, and performing dynamic feature weighting processing on the feature subset to obtain a weighted feature vector; constructing a physical constraint model by using the weighted feature vector, and outputting a weighted deviation between the predicted temperature and the actually measured temperature of the calculation model; and when the weighted deviation exceeds a preset deviation threshold value, determining a fault early warning type and triggering fault early warning. According to the method, the source of a fault caused by abnormal oil flow rate and oil temperature can be accurately locked, misjudgment caused by experience difference is avoided, the fault diagnosis accuracy is greatly improved, and the fault diagnosis and maintenance period is shortened.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of transformer fault diagnosis, and particularly relates to an automatic identification oil-immersed transformer fault detection device and a detection method. BACKGROUND

[0002] With the continuous expansion of the power system, the stable operation of the oil-immersed transformer as the core equipment of the power grid is crucial to guarantee the reliability of power supply. In the actual operation and maintenance scene, abnormal oil flow rate (too high or too low) and oil temperature fluctuation (too high or too low) of the transformer are typical causes of faults. Abnormal oil flow rate is easy to cause uneven local heat dissipation and accelerate the aging of insulation components. Imbalanced oil temperature will accelerate the deterioration of oil quality and increase the risk of internal discharge, threatening the safety of equipment and the stability of the power grid. At present, the troubleshooting of oil-immersed transformer faults caused by abnormal oil flow rate and oil temperature mainly relies on manual inspection mode. Under the existing technology, manual troubleshooting has the following defects: on the one hand, there is a lack of standardized and systematic fault positioning logic. Operation and maintenance personnel need to rely on personal experience to check the oil circulation system (such as oil pump, radiator, oil pipeline), cooling device, temperature control component, etc. one by one, which is difficult to quickly and accurately locate the root cause of abnormal oil flow rate or oil temperature, and is easy to cause misjudgment and omission due to experience differences. On the other hand, manual inspection relies on traditional methods such as visual observation and contact temperature measurement, which cannot efficiently and accurately capture hidden blockage of oil flow and internal subtle temperature changes, resulting in a significant reduction in the accuracy and timeliness of fault diagnosis. As a result, the maintenance cycle of the transformer is significantly prolonged, and the risk of unplanned power outage is increased, which not only affects the reliable supply of electricity, but also causes high operation and maintenance costs and social and economic losses due to long-term shutdown of equipment and expansion of faults.

[0003] Therefore, the existing fault troubleshooting technology relying on manual experience cannot meet the needs of modern power grids for efficient operation and maintenance and rapid fault disposal of oil-immersed transformers. It is urgent to break through the traditional manual mode and build a more intelligent and accurate fault diagnosis system to solve the pain points of ambiguous direction and low efficiency of manual inspection and improve the operation and maintenance level of transformers and the resilience of power grids. In view of the above technical problems, a detection method for automatically identifying oil-immersed transformer faults is developed to solve the problem of low efficiency of oil-immersed transformer fault detection SUMMARY The purpose of the present application is to provide an automatic identification oil-immersed transformer fault detection device and a detection method. The present application can accurately and efficiently locate the root cause of oil-immersed transformer faults caused by abnormal oil flow rate and oil temperature, avoid misjudgment caused by experience differences, greatly improve the accuracy of fault diagnosis, shorten the fault diagnosis and maintenance cycle, reduce the risk of unplanned power outage, guarantee the reliable supply of electricity, and reduce maintenance costs. In order to achieve the above purpose, the present application adopts the following technical effects: According to one aspect of the present application, the present application discloses a kind of automatic identification oil-immersed transformer fault detection method, the fault detection method includes the following steps: Step 1, data acquisition, real-time acquisition of oil-immersed transformer multi-source monitoring data, the multi-source monitoring data include the temperature data of multiple position points in transformer oil tank, pressure data, gas concentration data and oil flow velocity data of oil pillow pipeline; Step 2, data fusion, filtering and removing the collected multi-source data, obtain multi-source fault data under different fault types, carry out space-time synchronization alignment processing to multi-source fault data, and construct multi-dimensional fusion fault feature data associated with multi-source data; Step 3, data analysis processing;Using random forest algorithm to filter and detect the fusion processed fault feature data, obtain feature subset and preliminary warning signal, then carry out dynamic feature weighting processing to feature subset, and obtain weighted feature vector; Step 4: using weighted feature vector to construct physical constraint model, calculate model predicted temperature by physical constraint model, and calculate weighted deviation between model predicted temperature and measured temperature according to the output of physical constraint model; Step 5, when the weighted deviation exceeds the preset deviation threshold, determine the fault warning type and trigger the fault warning, and then execute corresponding fault control strategy according to the fault warning type and output corresponding three-dimensional fault positioning information of fault source.

[0004] The above scheme is preferably, in step 3, the fusion processed fault feature data is filtered and detected by random forest algorithm, including the following steps: Step 31, generate K sample subsets in the fusion processed fault feature data by self-service sampling, construct random forest decision tree using sample subsets, randomly select part of features to use gini index for node splitting, Step 32, calculate the importance score of each part of features f of each sample subset Imp(f) ; ; Wherein, ; D t The training data set of the t th decision tree, Gini Indicates the gini index of training data set D t ; f The single feature to be calculated importance, K Total number of decision trees in random forest; Gini(D t |f) Indicates the featuref The weighted Gini index after splitting n Number of fault categories p i Number of fault categories in the sample subset n The probability of; Step 33: Select features whose importance scores exceed the preset feature threshold and form a feature subset. Perform dynamic weighting on the feature subset to obtain a weighted feature vector.

[0005] In a preferred embodiment of the above scheme, the fault detection method further includes statistically comparing the proportion of feature subsets exceeding a preset feature threshold, and triggering a primary warning when the decision tree determines the data to be abnormal when the proportion of feature subsets exceeds a predetermined proportion.

[0006] The preferred approach described above, which involves performing dynamic weighting on the feature subset, includes: Calculate the real-time deviation index for each feature in the feature subset. E j satisfy: x j For the first j Real-time measurement values ​​of each feature μ j For the first j The historical mean of each characteristic σ i For the first j Historical standard deviation of each feature; Calculate the dynamic adjustment coefficient τ, where the dynamic adjustment coefficient τ satisfies: ; v j Oil flow velocity at the current measurement location ，v max To achieve the maximum design oil flow rate, T avg The average value is calculated for all temperature measurement points; the weight of each feature is then calculated. ω j It satisfies the following calculation formula: N is the number of features in the feature subset.

[0007] A further preferred embodiment of the above scheme calculates the predicted temperature using a physical constraint model, which satisfies the following: = ; In the formula, ρ is the weighted density of the transformer oil. c p Transformer oil specific heat capacity ；vThe current oil flow velocity ，T The three-dimensional temperature field distribution k for heat The conduction coefficient Q loss The winding heat loss, and is calculated by the following formula: Q loss =I 2 R [1 + 0.004 (T avg - 75)]; I The winding current, resistance R The calculation formula used satisfies: R = R 0[1+ β ( T winding -20)]; R 0 is the reference resistance at 20℃; β The winding temperature coefficient T winding The winding temperature T avg The average temperature Step 41, calculate the weighted deviation of the predicted temperature and the actual measured temperature of each measurement point, and when the weighted deviation exceeds the preset temperature threshold, trigger the fault warning.

[0008] The above scheme is preferably, the weighted deviation of the predicted temperature and the actual measured temperature of each measurement point is calculated δ T Satisfies: ; Wherein, the measurement point temperature weight coefficient γ j Satisfies: ; In the formula ，m The number of temperature sensors T model,i The physical constraint model predicted temperature of the first j position; T real,j The measured temperature of the first j position, Z j The installation height of the sensor j H is the height of the oil tank.

[0009] Preferably, the fault warning types include winding overheat fault, oil flow abnormal fault and arc discharge fault.

[0010] Preferably, the corresponding three-dimensional fault positioning information includes the following steps: Acquiring the spatial positions of the sensors p The spatial coordinates of the points x p , y p , z p and calculating the signal variation degrees Δ Sp of the sensors Sp . The signal variation degrees Δ satisfy the following formula: ; T p The real-time measurement values of the sensors at the current positions p , T p , base The temperature reference values of the sensors at the current positions p , and σ p is the historical standard deviation; Adjusting the importance weights α Sp of the current measurement positions of the sensors according to the signal variation degrees Δ p of the sensors p , wherein the importance weights α satisfy the following formula: ; Based on the current measurement positions of the sensors, the three-dimensional coordinates M of of the fault sources x , y , z are calculated; the three-dimensional coordinates M of the fault sources correspond to the corresponding three-dimensional coordinates, i.e., the three-dimensional fault positioning information, and the three-dimensional coordinates M of of the fault sources satisfy the following formula: x y z ; ; wherein N in the formula is the number of the sensors or the number of the moving measurement points of the sensors.

[0011] According to another aspect of the present application, the present application provides an automatic identification oil-immersed transformer fault detection device, the fault detection device comprising a sensor array unit and a data processing unit; the sensor array unit is used for collecting multi-source data of transformer operation; the data processing unit is used for pre-processing the collected data and executing a fault control strategy according to the processing result; the data processing unit comprises a random forest processing module, a dynamic feature weighting module, a physical constraint modeling module, a fault diagnosis module and a fault execution module; the random forest processing module is used for screening and anomaly detection of the detected multi-source fault feature data, to obtain feature data and a preliminary warning signal; the dynamic feature weighting module is used for weighting processing of the feature data, to obtain a weighted feature data set, and a physical constraint model is constructed by using the weighted feature data set; the physical constraint modeling module is used for calculating the weighted deviation of the model predicted temperature and the measured temperature of each measuring point of the sensor array unit; the fault diagnosis module is used for comparing and judging whether the calculated weighted deviation exceeds a preset deviation threshold, and if so, determining a fault warning type and triggering a fault warning; the fault execution module executes a corresponding fault report according to the fault warning type.

[0012] Preferably, the fault report comprises a fault control strategy and three-dimensional fault positioning information of the calculated output fault source.

[0013] In summary, the present application has the following technical effects: (1) The automatic fault identification detection method of the present application can accurately and efficiently lock the root cause of the oil-immersed transformer fault caused by abnormal oil flow rate and oil temperature, avoid misjudgment caused by experience difference, greatly improve the fault diagnosis accuracy, break through the traditional oil-immersed transformer fault troubleshooting mode relying on artificial experience, and greatly improve the fault troubleshooting efficiency; (2) The present application constructs an intelligent and accurate fault diagnosis system, integrates multi-scale permutation entropy, gas correlation and mutual information three types of heterogeneous features, realizes multi-dimensional data collaborative analysis through the spatio-temporal correlation between multi-source data and collaborative interpolation of the data, significantly improves the fault detection accuracy, can early warn the winding overheating fault and accurately locate the fault position; reduces the feature data dimension and improves the sensitivity of fault detection analysis, shortens the maintenance cycle, reduces the risk of unplanned power outage, improves the transformer operation and maintenance level and the power grid resilience, ensures the reliable power supply, reduces the operation and maintenance cost and social and economic loss.

[0014] (3) Without manual step-by-step troubleshooting, quickly lock the fault point, shorten the fault diagnosis and maintenance cycle, reduce the risk of unplanned power failure, guarantee the reliable supply of power, improve the response speed of power grid operation and maintenance; Reduce the high operating cost caused by manual misjudgment, long-term shutdown of equipment and expansion of faults, through intelligent diagnosis to early warning potential faults, reduce equipment wear and tear and social and economic losses, optimize the investment of operation and maintenance resources. SCHEMATIC DRAWING Figure 1 is a schematic diagram of the installation and deployment structure of an automatic identification oil-immersed transformer fault detection device of the present application; Figure 2 is an architecture schematic diagram of the data processing unit of the present application; Figure 3 is a flowchart of an automatic identification oil-immersed transformer fault detection method of the present application; In the drawings, multi-parameter sensor array 1, infrared temperature sensor 2, ultrasonic flow speed meter 3, oil-immersed transformer oil tank 10, oil pipe 11 outside the oil pillow, transformer oil pillow 12, slide rail 20. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical scheme and advantages of the present application clearer, the following preferred embodiments are combined with the drawings to further illustrate the present application. However, it should be noted that many details in the specification are only to enable the reader to have a thorough understanding of one or more aspects of the present application, and the aspects of the present application can be realized even without these specific details.

[0016] In combination Figure 1 and Figure 2, according to one aspect of the present application, the present application provides a kind of automatic identification oil-immersed transformer fault detection device, the fault detection device includes sensor array unit and data processing unit 4;The sensor array unit is used to collect the multi-source data of transformer operation;The sensor array unit includes the multi-parameter sensor array 1 of being arranged in the oil tank 10 of oil-immersed transformer, the infrared temperature sensor 2 of non-contact slidable deployment installation in the outside of oil-immersed transformer oil tank 10 and the ultrasonic flow velocity meter 3 (ultrasonic sensor) of non-contact deployment on the oil pipe 11 of the outer wall of oil pillow of oil-immersed transformer oil tank 10, the infrared temperature sensor 2 is installed on the slide rail 20 outside the top end of oil-immersed transformer oil tank 1, infrared temperature sensor 2 detects the temperature of the whole transformer in the process of moving back and forth along the slide rail 20 horizontally;The data processing unit is used to pre-process the data collected and execute fault control strategy according to the processing result;The data pre-processing unit generally uses PLC controller, FPGA controller, DSP processor to process the data detected;The data processing unit 4 includes random forest processing module, dynamic feature weighting module, physical constraint modeling module, fault diagnosis module and fault execution module;The random forest processing module is used to filter and detect anomaly to the multi-source fault feature data detected, obtains feature data and preliminary warning signal;The dynamic feature weighting module is used to weight processing to feature data, obtains weighted feature data set, and constructs physical constraint model using weighted feature data set;The physical constraint modeling module is used to calculate the weighted deviation of model predicted temperature and measured temperature of each measuring point of sensor array unit;Fault diagnosis module is used to compare whether the calculated weighted deviation exceeds preset deviation threshold, if exceeds, determine fault warning type and trigger fault warning;The fault execution module executes the corresponding fault report according to the fault warning type, and the fault report includes fault control strategy and the three-dimensional fault positioning information of the corresponding fault source calculated and output.

[0017] In the present application, as Figure 1As shown, the data processing unit 4 also includes a data display module and a wireless communication module, the data display module is used to display the real-time data detected by the sensor array unit and the fault warning data after the analysis and processing of the data processing unit, the data processing unit 4 sends the real-time data and the fault data to the remote monitoring server through the wireless communication module for early warning processing; In the present application, the multi-parameter sensor array deployed in the oil tank of the oil-immersed transformer is used to detect the operating parameters of the transformer oil in the oil tank of the oil-immersed transformer in real time; In the bottom (left lower front, left lower rear, right lower front, right lower rear, respectively used to monitor the oil density ρ and the pressure P), the middle (middle front, middle rear, middle left, middle right, respectively used to monitor the temperature T and the content of H2, C2H2, CH4), and the top (left upper front, left upper rear, right upper front, right upper rear, used to monitor the temperature T and the multi-modal sensor for monitoring the temperature T, CO, CO2), a total of 12 sensor monitoring points, divided into three layers of bottom / middle / top, respectively used to measure the transformer oil density, pressure and rare gas content; The horizontal / vertical spacing of each adjacent sensor in the sensor array is ≥200mm; The non-contact slidingly deployed infrared temperature sensor outside the oil tank of the oil-immersed transformer is used to detect the overall temperature data of the transformer outer wall in real time; The speed of the infrared temperature sensor moving and scanning along the external slide rail of the oil tank is 0.2m / s-0.5m / s, and the sampling rate is 2Hz-20Hz; The non-contact ultrasonic flow rate meter deployed on the oil pipe outside the oil pillow of the oil-immersed transformer is used to monitor the flow rate data of the transformer oil flowing through the internal oil pipe of the oil pillow in real time. The non-contact ultrasonic flow rate meter is installed on the oil pipe outside the oil pillow of the oil-immersed transformer, the probe spacing of the ultrasonic flow rate meter is 50±0.1mm, one end of the ultrasonic flow rate meter is at the upstream, and the other end is at the downstream, and the non-contact fixed ultrasonic flow rate meter is used to measure the oil flow rate, so that the measurement data is more accurate; The oil flow rate in the transformer oil v s The measurement satisfies: ; In the formula, c The sound speed in the transformer oil is 1480-1520 m / s; Δt is the ultrasonic wave propagation time difference (μs); L The probe spacing is 50±0.1mm, φ The oil thermal expansion coefficient is 6.5×10⁻ 4 / ℃, and ΔT is the temperature difference between the upstream and downstream of the pipeline (℃); The data processor is used for fault diagnosis analysis of the detected transformer oil operating parameters, transformer overall temperature data and transformer oil flow rate data, and automatically adjusts the cooling system power of the oil-immersed transformer according to the fault diagnosis analysis and generates a maintenance instruction report.

[0018] Combined Figure 1 andFigure 3 As shown, according to another aspect of the present application, the present application provides an automatic identification oil-immersed transformer fault detection method, the fault detection method comprising the following steps: Step 1, data acquisition, real-time acquisition of multi-source monitoring data of the oil-immersed transformer, the multi-source monitoring data including temperature data, pressure data, gas concentration data of multiple position points inside the transformer oil tank, and oil flow velocity data of the oil pillow pipeline; the multi-source data including overall temperature data of the outer wall of the oil-immersed transformer tank, density parameters of the transformer oil, pressure parameters of the transformer oil in the tank, and rare gas content parameters contained in the transformer oil in the tank; Step 2, data fusion, screening and removing the collected multi-source data, obtaining multi-source fault data under different fault types, and performing time and space synchronization alignment processing on the multi-source fault data to construct multi-dimensional fusion fault feature data associated with the multi-source data; when performing time and space synchronization alignment processing on the multi-source fault data, taking the sampling time of the ultrasonic velocity meter as the reference t 0, aligning the temperature and gas data by linear interpolation The multi-dimensional fusion fault feature data formed after the completion of the complementation is represented as X=[ T 1, P 1, ρ 1,..., T 12 , H 2, C 2 H 2, v ]; the fault feature data is a multi-dimensional vector data composed of [temperature, pressure, density, gas content, oil speed]; Step 3, data analysis processing; using a random forest algorithm to screen and detect anomalies of the fusion-processed fault feature data, obtaining a feature subset and a preliminary warning signal, and then performing dynamic feature weighting processing on the feature subset to obtain a weighted feature vector; wherein the dynamic feature weighting adjusts the feature weights according to the oil flow velocity and the average temperature; in the present application, the normal range of the oil temperature at the top of the transformer tank is between 85-95℃, and the maximum is not more than 95℃; the normal range of the middle oil temperature is maintained at 75-80℃; the normal range of the bottom transformer oil is maintained at 60-70℃; therefore, the main factors affecting are: (1) Temperature effect: when the oil temperature rises, the oil volume expands, and the transformer oil density will slightly decrease (about 0.5%-1% per 10℃ increase). For example, the density may decrease to 0.83-0.88 g / cm3 at 95℃, but the standard value at 20℃ should be taken as the reference (the density marked by the manufacturer is usually the value at 20℃); (2) Transformer normal operating pressure: When the transformer is in normal operation, the internal pressure of the oil tank is mainly caused by the expansion of the oil volume due to the increase in oil temperature, and a small amount of gas (non-noble gas) produced by oil decomposition, usually in a slightly positive pressure state (0.01-0.03 MPa).

[0019] Upper limit of pressure: The designed pressure of the oil tank is generally ≥0.05 MPa (to prevent explosion caused by sudden pressure rise in internal failure), and the pressure during normal operation should not exceed 0.03 MPa, otherwise it may be a signal of abnormal cooling system or internal failure (such as winding short circuit).

[0020] (3) Content of dissolved gas in transformer oil: The normal dissolved gas in transformer oil is mainly oxygen and nitrogen, as well as a small amount of combustible gas produced by slight decomposition of oil and insulating materials, and the content of noble gas (such as helium, neon, argon, etc.) is extremely small (usually considered as 0). The normal range of common combustible gases is as follows (reference IEC and national standard): Hydrogen (H2): ≤1000 ppm (some standards require ≤150 ppm, depending on the voltage grade of the transformer); Acetylene (C2H2): ≤0.5 ppm (≤0.1 ppm for new equipment); Methane (CH4), ethylene (C2H4): ≤100 ppm; Carbon monoxide (CO): ≤500 ppm; carbon dioxide (CO2): ≤1000 ppm.

[0021] (4) Oil flow rate: Abnormal flow rate (too high or too low) can easily cause uneven local heat dissipation and accelerate the aging of insulation components. During normal operation of the transformer, the oil flow rate of the oil pillow outer wall pipeline should be maintained at 0.5-1.5 m / s.

[0022] The above standard data of the oil-immersed transformer in normal operation will be connected to the data preprocessing unit (PLC controller, FPGA controller, DSP processor) through the multi-modal sensor (sensor including transformer oil density, pressure, and noble gas content measurement sensor), ultrasonic velocity meter, and infrared temperature sensor installed on the transformer to realize real-time monitoring data collection and processing. The data preprocessing unit data is exported and selected by random forest algorithm to obtain the final data; through Bootstrap, n times are drawn with replacement from n samples to form K sample subsets D K(K is the number of decision trees), ensuring that each subset is both different and overlapping, increasing the robustness of the model; randomly selecting feature subsets: for each sample subset, f features are randomly selected from m′ features to construct a single decision tree; using the Gini index to split nodes, subtrees are generated according to the nodes until the termination of splitting is met, ultimately forming dozens or even hundreds of decision trees, each of which can independently judge whether the data is "abnormal / normal"; each tree performs binary classification (or regression prediction) on the input data and outputs "abnormal probability"; In this embodiment of the invention, the process of filtering and detecting anomalies in the fused fault feature data using the random forest algorithm includes the following steps: Step 31: Generate K sample subsets from the fused fault feature data through bootstrapping. Construct a random forest decision tree using these sample subsets. Randomly select some features and use the Gini index to split nodes. During node splitting, randomly select... One feature, among which s for d The square root is rounded down; for example, if d=36, then s=6; if d=32, then s=5. A K=100 sample subset is generated using Bootstrap sampling; s=6 features are randomly selected (s= =6); Step 32: Calculate the features of each part of each sample subset. f Importance score Imp(f) ; ; in, ; D t For the first t The training dataset for decision trees, D t This is a subset of the dataset obtained through Bootstrap autosampling. Gini Represents the training dataset D t The Gini index; f For a single feature whose importance is to be calculated, K This represents the total number of decision trees in the random forest. Gini(D t |f) Representation of features f Weighted Gini index after split (dataset) Dt (the impurity of the gin) n Number of fault categories p i Number of fault categories in the sample subset n The probability of; Step 33: Select features whose importance scores exceed a preset feature threshold and form a feature subset. Perform dynamic weighting on the feature subset to obtain a weighted feature vector. The dynamic weighting of the feature subset includes: First, calculate the real-time deviation index for each feature in the feature subset. E j satisfy: x j For the first j Real-time measurement values ​​of each feature μ j For the first j The historical mean of each characteristic σ i For the first j The historical standard deviation of a feature, and the deviation index, reflect the degree to which the current feature value deviates from its historical normal fluctuations. For example, the historical mean and standard deviation of the temperature feature are derived from data under normal operating conditions over the past 30 days; Secondly, calculate the dynamic adjustment coefficient τ, where the dynamic adjustment coefficient τ satisfies: ; v Oil flow velocity at the current measurement location ，v max To achieve the maximum design oil flow rate, T avg This is the average value of all temperature measurement points; T avg >85℃ is the critical point for thermal aging of the insulation material of oil-immersed transformers; 0 represents the lower limit value, ensuring that the calculation result is not less than 0; dynamic adjustment coefficient. τ As a system-level regulation parameter that comprehensively reflects the real-time operating status of the entire oil-gas-heat system, it is used to adjust the sensitivity of characteristic weight calculation; based on oil flow velocity... v and average temperature T avg Together, they determine that when the oil flow rate decreases or the temperature increases, τ Decreasing the weighting makes it more sensitive to weight allocation (abnormal features are amplified in weight). This is especially true at high flow rates. τ The value can smooth out the false deviations caused by flow fluctuations, thus breaking through the traditional fixed threshold for temperature regulation and system cooling, and realizing the nonlinear quantification of temperature effects; Finally, the weight of each feature is calculated. ω j And form a weighted eigenvector; satisfying the following calculation formula: N is the number of features in the feature subset. Then each eigenvalue is x j × ω j , for this j real-time measurement value of each feature x j The weighted feature vector set is generated as: ; In the embodiment of the present application, when the importance score exceeds the preset feature threshold, the proportion of the feature subset exceeding the preset feature threshold is counted and compared, and when the proportion of the feature subset exceeds the predetermined proportion, the decision tree determines that the data is abnormal, and a preliminary warning is triggered.

[0023] Step 4: Construct a physical constraint model using the weighted feature vector, calculate the model predicted temperature through the physical constraint model, and calculate the weighted deviation between the model predicted temperature and the measured temperature according to the output of the physical constraint model; the physical constraint model is an oil flow-temperature field coupling model, and the model predicted temperature calculated through the physical constraint model satisfies: = ; In the formula, ρ is the weighted density of transformer oil, the unit is: g / cm³; ρ is calculated by the following formula: ; c p Cp is the specific heat capacity of transformer oil (physical constant Generally, c = 1880 J / kg K p k for heat ) ； v is the current oil flow speed ，T is the three-dimensional temperature field distribution (the predicted temperature obtained by solving the physical constraint model T ); heat The conduction coefficient, Ω The conduction coefficient k is calculated by the following formula ： ; Q loss is the winding heat loss, and is calculated by the following formula: Q loss =I 2 R [1 + 0.004 (T avg - 75)]; The winding resistance R is calculated by the formula:R = R 0[1+ β ( T winding -20)]; wherein, I I is winding current, the current measured by the transformer winding, R R0 is the reference resistance at 20℃; δ ); β K is the winding temperature coefficient (the winding is generally copper, and the temperature coefficient is 0.00393 / ℃); T winding T is the winding temperature (estimated by measuring the infrared temperature sensor outside the oil-immersed transformer oil tank); T avg T is the average temperature of the measurement point, which is the average temperature value of all measurement points of the top infrared temperature sensor; Step 41, calculate the weighted deviation of the predicted temperature and the actual measured temperature of each measurement point, and trigger the fault warning when the weighted deviation exceeds the preset temperature threshold; wherein the weighted deviation of the predicted temperature and the actual measured temperature of each measurement point is calculated γ T satisfies: wherein, the measurement point temperature weight coefficient δ j satisfies: ; m N is the number of temperature sensors; T model,i is the physical constraint model predicted temperature of the mth position; j T real,j is the measured temperature of the mth position, j Z j is the installation height of the sensor j from the bottom of the transformer oil tank, and H is the height of the oil tank; ​​Step 5, when the weighted deviation exceeds the preset deviation threshold, a fault warning type is determined and a fault warning is triggered, and then a corresponding fault control strategy and three-dimensional fault positioning information are executed according to the fault warning type. The fault warning type includes winding overheating fault, oil flow abnormality fault and electric arc discharge fault, and the fault execution strategy includes: when the winding overheating fault, the cooling system power is increased to 120% of the rated power; when the oil flow abnormality fault, the standby oil circuit is started and the sound wave descaling instruction is issued; when the electric arc discharge fault, the protection tripping instruction is triggered and the differential protection is started; when the gas is abnormal, the oil chromatography online analysis is triggered; the fault control strategy also includes automatically adjusting the cooling system power of the oil-immersed transformer and generating a maintenance instruction report. In the present application, when the deviation between the actual temperature and the model calculation value δT > 8 °C T >T hT ( δT > 8 °C ) times, T hT the preset temperature threshold is triggered to warn of winding overheating fault; when ΔC ) times ， the fault type is determined according to the relevant feature combinations, for example: when the oil flow rate is abnormal, the corresponding fault type is oil road blockage, and the oil speed v <0.4m / s, the sound wave descaling is started; when the gas concentration is abnormal, the corresponding fault is electric arc discharge, > 5 ppm / 5 min, C2H2 Sp the differential protection is triggered, when the internal pressure of the oil tank fluctuates, the corresponding fault type is winding overheating, and the pressure P max>0.03MPa, the cooling power is increased.

[0024] In the present application, the corresponding three-dimensional fault positioning information includes the following steps: First, the spatial coordinates of the space positions of each sensor p point x p , y p , z p are obtained, and the signal variation degree Sp of each sensor is calculated Sp , wherein the signal variation degree satisfies: ; T p is the real-time measurement value of the sensor at the current position p , T p ,base temperature reference value of the sensor at the current position p p historical standard deviation; signal variation degree Sp reflects the degree of deviation of the current measurement value from the normal reference, and the abnormality degree of different sensors can be compared; T p , base reflects the historical reference value of the sensor under normal working conditions, and the historical data of 30 days under normal working conditions is generally used as the mean value; sigma p reflects the historical standard deviation under normal working conditions, and the fluctuation number of 30 days of historical detection within the fluctuation range under normal working conditions is generally used as the standard deviation; Secondly, combined with the signal variation degree M of adjust the importance weight p of the current sensor measurement position p satisfies; ; Finally, based on the current sensor measurement position, the fault source M of three-dimensional coordinates x , y , z ) are calculated; the fault source M corresponds to the corresponding three-dimensional coordinates, that is, the fault positioning information, and the fault source ​ three-dimensional coordinates x , y , z ) are calculated, which satisfy the following formula: Wherein, N in the formula is the number of sensors or the number of sensor moving measurement points, when the sensor moves to measure, it is the spatial position coordinates of the moving measurement point.

[0025] The spatial coordinates of the fault source (fault point) of the present application are the weighted average of all sensor positions, and the importance weight p reflects the contribution degree of each sensor position to the fault point, which takes into account the feature importance at the same time, and also compares the abnormality degree of different sensors, so that the positioning result adapts to different fault types, and the sensor position with high weight and large abnormality is more inclined, avoiding the false positioning caused by single-point measurement fluctuation.

[0026] The above only describes the preferred embodiments of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application.​

Claims

1. A method for automatically identifying faults in oil-immersed transformers, characterized in that: The fault detection method includes the following steps: Step 1, Data Acquisition: Real-time acquisition of multi-source monitoring data of the oil-immersed transformer. The multi-source monitoring data includes temperature data, pressure data, gas concentration data, and oil flow velocity data of multiple locations inside the transformer tank. Step 2, data fusion: The collected multi-source data is filtered and eliminated to obtain multi-source fault data under different fault types. The multi-source fault data is then spatiotemporally synchronized and aligned to construct multi-dimensional fused fault feature data associated with the multi-source data. Step 3, data analysis and processing; use the random forest algorithm to filter and detect anomalies in the fused fault feature data to obtain feature subsets and primary warning signals, and then perform dynamic feature weighting on the feature subsets to obtain weighted feature vectors; Step 4: Construct a physical constraint model using weighted eigenvectors, calculate the model-predicted temperature using the physical constraint model, and calculate the weighted deviation between the model-predicted temperature and the measured temperature based on the output of the physical constraint model. Step 5: When the weighted deviation exceeds the preset deviation threshold, determine the fault warning type and trigger the fault warning. Then, execute the corresponding fault control strategy and output the corresponding three-dimensional fault location information of the fault source according to the fault warning type.

2. The method for automatically identifying faults in an oil-immersed transformer according to claim 1, characterized in that: Step 3 involves using the random forest algorithm to filter and detect anomalies in the fused fault feature data, including the following steps: Step 31: Generate K sample subsets from the fused fault feature data through bootstrapping; construct a random forest decision tree using these sample subsets; and randomly select some features to perform node splitting using the Gini index. Step 32: Calculate the features of each part of each sample subset. f Importance score Imp(f) ; ； in, ; D t For the first t The training dataset for decision trees, Gini Represents the training dataset D t The Gini index; f For a single feature whose importance is to be calculated, K This represents the total number of decision trees in the random forest. Gini(D t |f) Representation of features f The weighted Gini index after splitting n Number of fault categories p i Number of fault categories in the sample subset n The probability of; Step 33: Select features whose importance scores exceed the preset feature threshold and form a feature subset. Perform dynamic weighting on the feature subset to obtain a weighted feature vector.

3. The method for automatically identifying faults in an oil-immersed transformer according to claim 2, characterized in that: The fault detection method further includes statistically comparing the proportion of feature subsets that exceed a preset feature threshold. When the proportion of feature subsets exceeds a predetermined proportion, the decision tree determines that the data is abnormal and triggers a primary warning.

4. The method for automatically identifying faults in an oil-immersed transformer according to claim 1 or 2, characterized in that: Performing dynamic weighting on a subset of features includes: Calculate the real-time deviation index for each feature in the feature subset. E j satisfy: x j For the first j Real-time measurement values ​​of each feature μ j For the first j The historical mean of each characteristic σ i For the first j Historical standard deviation of each feature; Calculate the dynamic adjustment coefficient τ, where the dynamic adjustment coefficient τ satisfies: ； v j Oil flow velocity at the current measurement location ，v max To achieve the maximum design oil flow rate, T avg This is the average value of all temperature measurement points; Calculate the weight of each feature. ω j It satisfies the following calculation formula: N is the number of features in the feature subset.

5. The method for automatically identifying faults in an oil-immersed transformer according to claim 1, characterized in that: In step 4, the predicted temperature is calculated using the physical constraint model to satisfy: = ； In the formula, ρ is the weighted density of the transformer oil. c p Transformer oil specific heat capacity ；v Current oil flow rate ，T It represents a three-dimensional temperature field distribution; k is heat Transmission coefficient; Q loss The winding heat loss is calculated using the following formula: Q loss =I 2 R [1 + 0.004 (T avg - 75)]； I For winding current, resistance R The calculation formula used satisfies: R = R 0[1+ β ( T winding -20)]; R 0 represents the reference resistance at 20°C; β This refers to the winding temperature coefficient; T winding For winding temperature; T avg Average temperature; Step 41: Calculate the weighted deviation between the predicted temperature and the actual measured temperature at each measurement point. When the weighted deviation exceeds the preset temperature threshold, trigger a fault warning.

6. The method for automatically identifying faults in an oil-immersed transformer according to claim 1, characterized in that: Calculate the weighted deviation between the predicted temperature and the actual measured temperature at each measurement point. δ T satisfy: ； Among them, the temperature weighting coefficient of the measurement point γ j satisfy: ; in the formula ，m The number of temperature sensors, T model,i For the first j The physical constraint model at each location predicts the temperature; T real,j For the first j Measured temperature at each location, Z j For sensors j The installation height is H, where H is the height of the fuel tank.

7. The method for automatically identifying faults in an oil-immersed transformer according to claim 1, characterized in that: The fault warning types include winding overheating fault, abnormal oil flow fault, and arc discharge fault.

8. The method for automatically identifying faults in an oil-immersed transformer according to claim 1, characterized in that: Executing the corresponding three-dimensional fault location information includes the following steps: Obtain the spatial location of each sensor p Spatial coordinates of a point ( x p , y p , z p And calculate the signal variation Δ of each sensor. Sp Among them, the signal variability Δ Sp satisfy: ； T p Current location p The real-time measurement values ​​of the sensor, T p , base Current location p The temperature reference value of the sensor, σ p The historical standard deviation; Combining the signal variability Δ of each sensor Sp Adjust the importance weight α of the current sensor measurement positions p Among them, the importance weight α p satisfy; ； The fault source is calculated based on the current measurement locations of each sensor. M Three-dimensional coordinates ( x , y , z Fault source M The corresponding three-dimensional coordinates constitute the three-dimensional fault location information, and the fault source. M Three-dimensional coordinates ( x , y , z The calculation satisfies the following formula: ； In the formula, N represents the number of sensors or the number of sensor moving measurement points.

9. An automatic fault detection device for oil-immersed transformers, characterized in that: A fault detection device for implementing the automatic identification and detection method for oil-immersed transformer faults as described in any one of claims 1 to 8 includes a sensor array unit and a data processing unit. The sensor array unit is used to collect multi-source data on transformer operation. The data processing unit is used to preprocess the collected data and execute a fault control strategy based on the processing results. The data processing unit includes a random forest processing module, a dynamic feature weighting module, a physical constraint modeling module, a fault diagnosis module, and a fault execution module. The random forest processing module is used to filter and detect anomalies in the detected multi-source fault feature data to obtain feature data and primary warning signals. The dynamic feature weighting module is used to weight the feature data to obtain a weighted feature dataset and to construct a physical constraint model using the weighted feature dataset. The physical constraint modeling module is used to calculate the weighted deviation between the model-predicted temperature and the measured temperature at each measurement point of the sensor array unit. The fault diagnosis module is used to compare and determine whether the calculated weighted deviation exceeds a preset deviation threshold. If it does, the fault warning type is determined and a fault warning is triggered. The fault execution module outputs the corresponding fault report based on the fault warning type.

Images