A transformer condition monitoring device and method

Abstract

The application discloses a kind of transformer condition monitoring device and method, belong to transformer condition monitoring technical field, comprising: processor;With the temperature measurement micro motion travel switch of processor connection;With the processor of environmental temperature detection sensor connection;With the power supply component of processor and environmental temperature detection sensor connection;Wherein, the temperature measurement micro motion travel switch includes: temperature measurement component and detect whether micro motion travel switch with transformer surface contact;The transformer condition monitoring device and method, with local online monitoring and remote diagnosis, without additional auxiliary equipment, greatly increase transformer operation reliability, installation convenience, and greatly reduce monitoring cost.

Description

Technical Field

[0001] This invention belongs to the field of transformer condition monitoring technology, and specifically relates to a transformer condition monitoring device and method. Background Technology

[0002] Transformer internal components (such as insulation materials) are extremely sensitive to temperature. Prolonged overheating accelerates insulation aging, leading to decreased dielectric strength and even serious faults such as breakdown and short circuits. Abnormal temperatures can be caused by various factors, including overload, cooling system malfunctions, internal partial discharge, or winding short circuits. Monitoring temperature changes can identify potential risks; for example, an abnormally high top-layer oil temperature may indicate poor heat dissipation, and excessive winding hotspot temperatures may lead to oil decomposition or even explosion. The transformer's load capacity is limited by its hotspot temperature. Precise temperature measurement allows for dynamic assessment of the transformer's short-term overload potential, improving the flexibility of grid dispatch while ensuring safety. Furthermore, the lifespan of insulation materials follows Arrhenius's Law: the aging rate doubles for every 6-8°C increase in temperature. Controlling operating temperature can significantly slow down equipment aging and reduce total lifecycle costs.

[0003] Traditional methods for transformer temperature measurement include oil surface temperature measurement, winding temperature simulation to indirectly estimate hot spot temperatures, fiber optic thermography for direct measurement of winding hot spots, infrared thermal imaging for external hot spot detection, and distributed sensor networks for multi-location synchronous monitoring. In power systems, transformers are core equipment, and real-time monitoring of their operating status is crucial for grid security. However, traditional temperature monitoring installations often require power outages, which not only affect power supply reliability but can also lead to economic losses. This is especially true for critical locations such as hospitals, data centers, and rail transit systems, where power outages can cause significant social impact or economic losses. Traditional power outage installations require coordination with users' power outage plans, potentially leading to complaints or even contract disputes. Live-line work, on the other hand, eliminates the need for power outages. Live-line work ensures uninterrupted power supply during the installation or upgrade of temperature monitoring systems, maintaining stable grid operation and directly avoiding interference with users' production and daily lives. Therefore, conducting online temperature monitoring installations on transformers without power outages—i.e., live-line work—is of significant importance.

[0004] During transformer operation, losses in the core and windings generate heat, causing the oil temperature to rise and be conducted to the tank surface. Under normal circumstances, the temperature difference (ΔT) between the tank surface and the ambient temperature should be relatively stable. An abnormal increase or irregular fluctuation in ΔT may indicate the following problems:

[0005] 1. Internal faults: such as winding short circuit, multiple grounding points in the iron core, insulation aging, etc.

[0006] 2. Cooling system malfunction: radiator blockage, fan failure, poor oil circulation;

[0007] 3. Changes in the external environment: high temperatures, poor ventilation, etc.

[0008] Transformer temperature variations are typically related to load, ambient temperature, and internal faults. Temperature differences on the transformer tank surface can reflect internal anomalies, such as winding short circuits, core faults, or cooling system problems. Furthermore, by employing large-scale model analysis, the temperature of the transformer tank surface can be compared with the ambient temperature, and transformer fault diagnosis can be performed in conjunction with factors such as the transformer's load.

[0009] The relationship between the surface temperature of the fuel tank and the fault is: ΔT = T_fuel_tank Environment; Normal range of ΔT: usually not exceeding 10~15℃; Alarm threshold: an alarm is required if ΔT exceeds 20℃ for a short period of time or if the annual growth rate exceeds 5%;

[0010] The fault type is a partial short circuit in the winding, and the typical temperature manifestation is a significant increase in the temperature of local hot spots in the oil tank, i.e., ΔT local > 30℃;

[0011] The fault type is core fault, and the typical temperature manifestation is a uniform rise in the overall oil temperature.

[0012] The fault type is cooling system failure, and the typical temperature manifestation is abnormal oil temperature gradient, that is, the top layer oil temperature is greater than the lower layer oil temperature +10℃.

[0013] The fault type is environmental factors, and the typical temperature manifestation is that the temperature changes periodically with the day and night / load, without a continuous upward trend;

[0014] When there are slight issues such as ΔT < 25℃, it is necessary to strengthen monitoring and shorten the sampling cycle of the sensor.

[0015] An alarm will be triggered immediately if there is a serious abnormality, such as ΔT > 30℃ or a sudden change.

[0016] Therefore, it is necessary to develop a new transformer condition monitoring device and method to analyze transformer faults by testing temperature differences. Summary of the Invention

[0017] The purpose of this invention is to provide a transformer condition monitoring device and method to solve the above-mentioned problems.

[0018] To achieve the above objectives, the present invention provides the following technical solution: a transformer condition monitoring device, comprising: a processor;

[0019] Temperature-sensing micro-switch connected to the processor;

[0020] An ambient temperature detection sensor connected to the processor;

[0021] A power supply component connected to the processor and the ambient temperature detection sensor;

[0022] The temperature-measuring micro-motion limit switch includes: a temperature-measuring component and a micro-motion limit switch for detecting whether it is in contact with the transformer surface.

[0023] Preferably, the power supply component includes: a main battery, a backup battery, and a charging unit with automatic switching connected to any one or both of the main battery and the backup battery;

[0024] The charging unit includes a solar power module and a thermoelectric film; the thermoelectric film includes a high-temperature surface and a low-temperature surface in contact with the transformer, and a heat sink is attached to the low-temperature surface.

[0025] Preferably, the surface of the temperature-measuring micro-motion limit switch is a high-heat-conductivity copper mold; and a heat-insulating material is provided between the ambient temperature detection sensor and the sensor body;

[0026] The temperature difference power extraction film and the temperature measurement and acquisition film are bonded together to form an integral film. The temperature measurement and acquisition film is located on one side of the transformer and is closely attached to the transformer shell to ensure accurate data collection.

[0027] The temperature difference power-generating film is closely attached to the temperature-sensing film and kept away from the transformer casing to ensure that the high-temperature side is in close contact with the temperature-sensing film, providing the largest possible temperature difference.

[0028] Preferably, the transformer condition monitoring device further includes:

[0029] A wireless module connected to the processor, the wireless module including a network communication module and a low-power wireless module.

[0030] The present invention further provides a monitoring method for a transformer condition monitoring device, comprising:

[0031] Collect the surface temperature of the transformer tank and the ambient temperature and humidity, and calibrate the surface temperature of the transformer tank based on the ambient temperature and humidity.

[0032] The status of the temperature-sensing micro switch is collected, and an alarm is triggered if the temperature-sensing micro switch is open.

[0033] Obtain current transformer load information;

[0034] After data acquisition is complete, perform low-power management of peripheral devices.

[0035] The real-time difference between the surface temperature of the oil tank and the ambient temperature, the rate of temperature change in minutes, the standard deviation of the historical N-hour temperature difference, and the hourly slope of the ambient temperature change are input into the transformer fault analysis model.

[0036] The transformer fault analysis model includes: a BP neural network unit for handling deterministic fault modes;

[0037] And the ART2 neural network unit for identifying abnormal heat distribution patterns;

[0038] After analysis, the BP neural network unit outputs intermediate results to the ART2 neural network unit, which then uses a dynamic threshold mechanism to make a judgment.

[0039] Preferably, the calculation of the standard deviation of the historical N-hour temperature difference includes: in, This indicates the real-time difference (°C) between the surface temperature of the fuel tank and the ambient temperature. The standard deviation of the temperature difference over N hours is represented by the average rate of change of the temperature difference. N represents the total number of temperature difference data points. This represents the loop variable used in the calculation, ranging from 1 to N.

[0040] Preferably, the input expression for the transformer fault analysis model is: Input=[ΔT,ΔT_rate, [, Ambient_temp_trend, Humidity, Load_current_ratio, Load_profile]; where ΔT represents the average temperature difference; _rate represents the rate of change of temperature difference (°C / min); The standard deviation of the historical N-hour temperature difference is represented; Ambient_temp_trend represents the hourly slope of the ambient temperature change; Load_profile represents the similarity of the load curve over the past 24 hours, indicating the peak electricity consumption at different times of the day; Humidity represents the ambient humidity (%RH); Load_current_ratio represents the ratio of load current to rated current.

[0041] The probability of deterministic faults in the output of a BP neural network unit includes: overload fault, radiator blockage fault, oil circulation fault, and abnormal environment fault.

[0042] Preferably, when collecting the surface temperature of the transformer tank and the ambient temperature and humidity, a temperature correction coefficient EC is set for the surface temperature of the tank to compensate for the temperature.

[0043] The expression for the temperature correction factor EC is EC = 1 + 0.02 * (Humidity - 50);

[0044] Humidity represents the ambient humidity. For every 10% increase in humidity, the temperature measurement error increases by approximately 2%. The calculation method for the corrected tank surface temperature includes: T(correction value) = T(measured value) * EC.

[0045] Preferably, the judgment method of the transformer fault analysis model includes:

[0046] Primary threshold alarm;

[0047] In-depth diagnostic phase;

[0048] Fusion diagnosis;

[0049] The primary threshold alarm includes:

[0050] By mapping the input parameters;

[0051] By curve fitting, the probability distribution of four types of deterministic faults is output: P=[Poverload, Pradiator blockage, P oil circulation failure, P environmental abnormality]; P represents the probability distribution of the fault.

[0052] If the probability of a fault is greater than a preset threshold, it is marked as a candidate fault.

[0053] The fault type is P overload, and the judgment conditions are ΔT>25℃ and Load_current_ratio>90%.

[0054] The fault type is P heat sink blockage, and the judgment condition is ΔT_rate > 3℃ / min for 10 minutes;

[0055] The fault type is P oil circulation fault, and the judgment condition is ΔT fluctuation coefficient > 0.4;

[0056] The in-depth diagnostic phase includes:

[0057] Standardize the temperature difference time series data;

[0058] A competitive subsystem is generated in the intermediate layer, and the clustering threshold is dynamically adjusted by comparing coefficients.

[0059] Match the historical fault pattern library in layer F2 and output the fuzzy fault and confidence level.

[0060] Preferably, the fusion diagnosis includes: assigning scoring coefficients to the fuel tank surface temperature, the temperature difference between the fuel tank surface and the ambient temperature, and the temperature standard deviation over a given time range; combining the fault probability output by the BP neural network unit and the fault confidence output by the ART2 neural network unit, as well as the weights of the indicators, to calculate the final score (Final_Score) for each fault; wherein the formula for the final score (Final_Score) is as follows: ;

[0061] Fault confidence level: Fault confidence level is the probability of a fault occurring, expressed as a percentage.

[0062] Fault weight: Fault weight is the probability of a fault occurring at the location where the transformer condition monitoring device is installed;

[0063] Based on the Final_Score scores, the potential faults are sorted from highest to lowest, with the highest score being the first potential fault. The second potential faults are then obtained in turn. The faults are then compared with the preset fault thresholds, and the decision results are output. If the difference between the score of the second potential fault and the Final_Score is less than 15%, a mixed fault warning is triggered.

[0064] The technical effects and advantages of this invention are as follows: This transformer condition monitoring device and method have the functions of local online monitoring and remote diagnosis, without the need for additional auxiliary equipment, which greatly increases the reliability of transformer operation, facilitates installation, and greatly reduces monitoring costs. By adding a temperature calibration coefficient EC and dynamically updating the EC coefficient, the data drift under high humidity environment is solved. The introduction of a weighting mechanism to focus on the dominant fault characteristics solves the problem of multi-fault coupling interference. The model quantization and compression meet the real-time requirements. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the transformer condition monitoring device of the present invention;

[0066] Figure 2 This is a block diagram of the transformer condition monitoring device of the present invention;

[0067] Figure 3 This is a schematic diagram of the functional modules of the transformer condition monitoring device of the present invention;

[0068] Figure 4 This is a schematic diagram of the dual-battery power supply circuit for the transformer condition monitoring device of the present invention;

[0069] Figure 5 This is a schematic diagram of the solar energy and temperature difference power extraction of the transformer condition monitoring device of the present invention;

[0070] Figure 6 This is a schematic diagram of the mobile network wireless communication of the transformer condition monitoring device of the present invention;

[0071] Figure 7 This is a schematic diagram of the low-power wireless communication of the transformer condition monitoring device of the present invention;

[0072] Figure 8 This is a schematic diagram of the overall structure of the transformer condition monitoring device of the present invention;

[0073] Figure 9 This is a schematic diagram of temperature acquisition in the transformer condition monitoring device of the present invention;

[0074] Figure 10 This is a schematic diagram of the transformer condition monitoring method of the present invention.

[0075] In the diagram: 1. Ambient temperature sensor; 2. Solar power supply unit; 3. Sensor body; 4. Temperature difference power extraction film; 5. Temperature measurement and acquisition film; 6. Temperature measurement micro-switch; 7. Buffer pad; 8. Magnet; 9. Thermally conductive adhesive; 10. Transformer. Detailed Implementation

[0076] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0077] This invention provides, for example Figure 1 , Figure 2 , Figure 3 The present invention relates to a condition monitoring device for a live-installed transformer, comprising:

[0078] 1. Ambient temperature detection sensor; 2. Solar power supply unit connected to charging chip; 3. Sensor body; 4. Temperature difference power extraction film; 5. Temperature measurement acquisition film; 6. Temperature measurement micro-switch; 7. Buffer pad; 8. Magnet; 9. Thermally conductive adhesive.

[0079] Thermoelectric film 4 and temperature acquisition film 5 are glued together to form a single film. Conductive silver paste is applied to the edge of thermoelectric film 4 using a dot coating method to connect the electrodes of the charging port of flexible PCB.

[0080] Apply thermally conductive adhesive 9 evenly to the contact surface between the thermal difference power extraction film 4 and the oil tank, covering the entire monitoring area, to ensure that the live transformer condition monitoring device is firmly adhered to the surface of the transformer 10 oil tank.

[0081] After the glue cures, the buffer pad 7 is embedded to absorb vibration energy and prevent the device from loosening and falling off.

[0082] The live-line transformer condition monitoring device adopts a dual power source design of solar power and temperature difference power, and automatically switches to ensure normal and stable power supply to the device;

[0083] The temperature measuring micro switch 6 adopts a design that integrates temperature measuring components and micro switches to ensure that the temperature measuring surface is in close contact with the surface of the transformer 10 oil tank. At the same time, it detects whether the device is firmly attached to the surface of the transformer 10 oil tank. When the transformer condition monitoring device is installed and falls off, the switch is open. When the device is in close contact with the surface of the transformer 10 oil tank, the micro switch is closed.

[0084] The hardware circuit includes a charging circuit, a power supply circuit, a wireless communication module, an emulator interface circuit, a debugging serial port circuit, a drop detection circuit, a multi-channel temperature acquisition circuit, a battery detection circuit, and a microcontroller minimum system circuit.

[0085] The battery system uses non-rechargeable backup batteries and rechargeable high / low temperature resistant batteries or rechargeable supercapacitors. Internally, a diode-based automatic battery selection mechanism is used. The rechargeable battery voltage is around 4V; when the rechargeable battery voltage drops below 3.6V, the backup 3.6V non-rechargeable battery begins supplying power. Figure 4 As shown;

[0086] The rechargeable battery uses a 5V solar power unit 2 and a temperature difference power extraction film 4. When there is sunlight or a temperature difference on the surface of the transformer 10, the VIN input voltage is approximately 5V. The charging integrated chip CN3083 tracks the maximum power of the charging power supply and simultaneously detects when the battery is fully charged and stops charging the lithium battery. The chip integrates three-stage charging protection logic and lithium battery overcharge protection. Figure 5 As shown;

[0087] The network communication module allows the device to connect to a remote server via a 4G signal. Data transmission is achieved through a wireless 4G connection, sending temperature signals, battery voltage, time information, device identification information, internal fault and data alarm information, receiving server setting commands, and reporting rate instructions. Figure 6 As shown;

[0088] LoRa module communication allows devices to optionally transmit signals via a LoRa wireless module in a self-organizing network, with a maximum distance of up to 5km. It can send temperature signals, battery voltage, time information, device identification information, internal fault and data alarm information, receive server setting commands, and report rate instructions. Figure 7 As shown;

[0089] The device software runs on a microcontroller system, using a Fudan Microelectronics 32-bit ARM processor chip. It integrates peripherals such as timers, serial ports, ADCs, PWM controllers, RTCs, and SPI. It can measure battery level, sensor temperature, communicate with 4G and LoRa modules, and perform drop detection. The processor can be emulated online via a JTAG interface for device software debugging; for example... Figure 8 As shown;

[0090] like Figure 9 As shown, the temperature-sensing micro-switch 6 uses PT100, thermistors, and temperature-sensing films as temperature-sensing components. The temperature sensor outputs an analog signal to the microcontroller, which converts the analog signal into a high-precision temperature value using a high-precision ADC. The micro-switch detection can detect whether the device is in close contact with the test surface or whether the device has been dropped.

[0091] The bottom of the live-installed transformer condition monitoring device is attached to the surface of the transformer 10 with one or more temperature difference power-collecting films. One side of the film is the high-temperature side, which is closely attached to the surface of the transformer 10 oil tank to ensure accurate temperature measurement. The other side is the low-temperature side, which is attached with a heat dissipation plate to reduce the temperature and increase the temperature difference between the two sides of the power-collecting film.

[0092] The temperature sensing component and the metal microswitch used for contact detection are integrated into one part. The PT100 is built into the inner surface of the microswitch cap, ensuring accurate detection of whether the device has detached. The temperature sensing microswitch 6 adopts a design that integrates the temperature sensing component and the microswitch, ensuring that the temperature sensing surface is in close contact with the surface of the transformer 10 tank. The microswitch detects whether the device is firmly attached to the surface of the transformer 10 tank. When the transformer condition monitoring device is installed while energized and falls off, the switch opens; when the device is in close contact with the surface of the transformer 10 tank, the microswitch closes. The surface of the temperature sensing microswitch 6 is made of high-heat-conductivity copper mold, ensuring the accuracy of temperature acquisition.

[0093] The ambient temperature detection sensor 1 uses a heat insulation pad to insulate the main body of the live transformer status monitoring device, ensuring that the ambient temperature is not affected by the main body.

[0094] The live-line transformer status monitoring device adopts a dual power source design of solar power and temperature difference power, and automatically switches to ensure normal and stable power supply to the device.

[0095] The present invention further provides, for example, Figure 10 The transformer condition monitoring method shown includes the following steps:

[0096] Step 1: Start the device and initialize it;

[0097] Step 2: Load the transformer fault analysis model based on temperature difference, and confirm the model version with the management platform. If there is an updated model or parameter debugging, the model will be upgraded. The configuration parameters of the transformer fault analysis model based on temperature difference include various alarm thresholds, data collection and reporting intervals, and temperature difference matching curves for various abnormal states.

[0098] Step 3: Monitor the equipment status and confirm that all parts are working properly. If any abnormality occurs, report it to the equipment management platform.

[0099] Step 4: Collect the surface temperature of the transformer 10 oil tank and the ambient temperature and humidity, and calibrate the surface temperature of the transformer 10 oil tank according to the temperature and humidity for subsequent temperature calculation.

[0100] Step 5: Collect the status of the micro switch. When the switch is open, an alarm will be triggered.

[0101] Step 6: Contact the platform to obtain the current load information of transformer 10, which will be used as input for the transformer fault analysis model based on temperature difference;

[0102] Step 7: After data acquisition is complete, perform low-power management of peripherals, turn off the acquisition module, and put the MCU (processor) into sleep mode.

[0103] Step 8: After the equipment collects the data, it calculates the real-time temperature difference ΔT between the surface temperature of the oil tank and the ambient temperature, the temperature change rate in minutes - temperature difference change rate (°C / min) ΔT_rate, the temperature difference over the past N hours ΔT_std, the hourly change slope of the ambient temperature Ambient_temp_trend, and sends them together with the data obtained from the network side to the transformer fault analysis model based on temperature difference.

[0104] Step 9: Based on the input parameters, start the transformer fault analysis model based on temperature difference, and use the output of the BP neural network unit as the input of the ART2 neural network unit for secondary analysis;

[0105] Step 10: Based on the threshold setting and model output, make a judgment and report the normal, early warning, alarm status and the collected values, analysis values, and fault analysis values.

[0106] Step 11: Low power management. When an anomaly occurs, increase the sampling frequency. When there is no anomaly, report according to the reporting interval issued by the platform.

[0107] The transformer fault analysis model based on temperature difference includes: a BP neural network unit for handling deterministic fault modes (such as overload and radiator blockage), and an ART2 neural network unit for identifying abnormal heat distribution patterns (such as partial discharge and winding short circuit).

[0108] Based on the input parameters, the BP neural network unit is activated, and the output of the BP neural network unit is used as the input of the ART2 neural network unit for secondary analysis.

[0109] Based on the threshold settings and model output, the system makes judgments and reports normal, early warning, and alarm statuses, as well as collected values, analyzed values, and fault analysis values.

[0110] Low power management: when an anomaly occurs, the sampling frequency is increased; when there is no anomaly, reporting is performed according to the reporting interval issued by the platform.

[0111] For transformer fault diagnosis models based on parameters such as temperature difference (ΔT = tank surface temperature − ambient temperature), temperature change rate, fluctuation characteristics, ambient temperature and humidity and their change rate, and load characteristics, the input parameters are reconstructed and optimized based on the BP-ART2 hybrid architecture to ensure a strong correlation between parameter selection and temperature difference analysis. The BP-ART2 hybrid architecture is a hybrid neural network architecture combining BP neural networks and ART2 models. This architecture utilizes the learning ability of BP neural networks and the fast learning and online recognition capabilities of ART2 models, making it particularly suitable for handling non-stationary and nonlinear stochastic processes, such as fault diagnosis in power transformers. After screening suspicious samples based on preliminary threshold alarms, the BP-ART2 hybrid architecture model is used for refined analysis of complex fault modes. Its goal is to extract the physical essence of faults from multi-dimensional features, achieving a leap from "symptom identification" to "cause localization."

[0112] Optimized mean square logarithmic error is used in regression tasks to optimize probability prediction by introducing logarithmic transformation into error calculation, thereby mitigating the impact of unbalanced target variable values ​​or extreme values.

[0113] Enhanced temperature sensitivity: A temperature gradient constraint is added to the intermediate layer to ensure priority for overload response;

[0114] Time-frequency domain fusion: The temperature difference signal after wavelet transform (energy proportion of 1 / 4 frequency band) is used as an additional feature to improve the detection capability of transient overheating (such as inter-turn short circuit);

[0115] Heat source location: By analyzing the phase difference of time series data, the abnormal heating area (such as the upper / lower part of the fuel tank) can be located.

[0116] The temperature difference characteristic parameters in the transformer fault analysis model include:

[0117] Real-time temperature difference between the surface temperature of the fuel tank and the ambient temperature (°C);

[0118] ΔT_rate: Rate of change of temperature difference (°C / min), reflecting the dynamic performance of heat dissipation;

[0119] The standard deviation of the temperature difference over the past N hours quantifies the stability of temperature fluctuations.

[0120] Environmental disturbance characteristics;

[0121] Ambient_temp_trend: The hourly slope of the ambient temperature change (°C / h);

[0122] Load_profile: Similarity of load curves over the past 24 hours, indicating peak electricity consumption at different times of the day. Load correlation characteristics;

[0123] Load_current_ratio: The ratio of load current to rated current (%).

[0124] Improvements to the BP-ART2 hybrid architecture;

[0125] The input to the BP neural network unit (deterministic fault classification) consists of 4 core parameters and 3 derived parameters.

[0126] Input=[ , , ,Ambient_temp_trend,Humidity,Load_current_ratio,Load_profile];

[0127] Output layer: Four types of deterministic faults include: overload fault, radiator blockage fault, oil circulation fault, and environmental anomaly fault;

[0128] ART2 Neural Network Unit (Nonlinear Pattern Recognition)

[0129] Input layer: Receives probability vectors output by neural network units and raw temperature difference time series data (sliding window: 10 time points, window length = 10 minutes, step size = 1 minute).

[0130] Competition subsystem (F1 layer): Adaptive normalization to handle local anomalies in temperature difference time series; Dynamic clustering threshold: Adjust the similarity threshold based on historical data (ρ=0.85±0.05);

[0131] Memory Subsystem (F2 Layer): Stores the thermal distribution patterns of typical faults; Radiator blockage: ΔT continuously increases and ΔT_rate>2℃ / min; Oil circulation fault: ΔT fluctuates greatly and oil level drops accompanied by a sudden increase in temperature difference; Classification decision: Through F2 layer weight matching, output 5 types of fuzzy faults (such as suspected local overheating of windings).

[0132] Environmental compensation characteristics;

[0133] The ambient temperature correction factor is used to correct temperature measurement errors in high humidity environments: Factor: EC = 1 + 0.02 * (Humidity - 50) (For every 10% increase in humidity, the temperature measurement error increases by approximately 2%).

[0134] Corrected oil tank surface temperature: T (corrected value) = T (measured value) * EC;

[0135] The fault diagnosis process includes:

[0136] Primary threshold alarm;

[0137] Identify deterministic failure modes that are strongly correlated with temperature differences (such as overload, radiator blockage);

[0138] Feature mapping: mapping input parameters;

[0139] By curve fitting, the probability distribution of four types of deterministic faults is output: P=[Poverload, Pradiator blockage, P oil circulation failure, P environmental abnormality]; P represents the probability distribution of a specific fault.

[0140] Threshold decision: If the probability of a certain type of fault is greater than a preset threshold, it is marked as a candidate fault.

[0141] Table 1. Correspondence between Fault Types and Judgment Criteria

[0142]

[0143] In-depth diagnostic phase;

[0144] Used for analyzing complex and / or fuzzy fault modes (such as localized overheating of the fuel tank, multiple fault coupling), it includes the following processes:

[0145] Data input: Standardize the temperature difference time series data to ensure consistent data input scale and avoid matching failures due to differences in numerical range.

[0146] Comparison: A competing subsystem is generated in the intermediate layer, and the clustering threshold is dynamically adjusted by comparing the coefficients.

[0147] Fault matching: By matching the historical fault mode library, the fault with the highest matching degree and the corresponding confidence level are found.

[0148] Fusion diagnosis;

[0149] The transformer fault analysis model based on temperature difference outputs the judgment results for each fault, and the transformer 10 status equipment attached to different parts for judgment is used to comprehensively judge the health status of the equipment by assigning different weights to different features and using a fusion diagnosis method.

[0150] Weighted voting method: Based on expert experience and historical big data, a scoring coefficient is assigned to the key indicators monitored by the system, namely temperature, temperature difference, and standard deviation. The final score for each fault is calculated by combining the BP probability, the fault confidence output by the ART2 module, and the weight of the indicators. Final_Score is used to calculate the final fault weight.

[0151] Weighting principle:

[0152] The accuracy of the model's judgments;

[0153] Expert experience assigns weight to indicators;

[0154] The weights are updated based on real-time feedback (such as sensor data stability).

[0155] Final_Score is calculated as a weighted average of the output score of the temperature difference-based transformer fault analysis model and the weights.

[0156] Fault confidence level: Fault confidence level is the probability of a specific fault occurring, expressed as a percentage.

[0157] Fault weighting: Fault weighting represents the probability of a fault occurring at the location where the transformer condition monitoring device is installed; for example, a calculation of Final_Score is: Final_Score = (62.5% × 0.4) + (45.7% × 0.3) + (30.0% × 0.3) = 47.3%;

[0158] Then, the Final_Scores are sorted, with the highest score indicating the first potential fault, the second highest score indicating the second potential fault, and so on. The faults are then compared to preset fault thresholds, and the decision result is output, including: critical fault, potential anomaly, and normal.

[0159] Select the fault type with the highest Final_Score. If the difference between the final score of the second potential fault and the Final_Score is less than 15%, then trigger a "mixed fault" warning.

[0160] At the same time, the weights are dynamically adjusted based on the equipment's operating status, including its service life, debt ratio, ambient temperature, and other factors.

[0161] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A transformer condition monitoring device, characterized in that: include: processor; A temperature-measuring micro-switch connected to the processor for detecting whether the transformer condition monitoring device is in contact with the transformer's oil tank; An ambient temperature detection sensor connected to the processor; A power supply component connected to the processor and the ambient temperature detection sensor; The end of the temperature-measuring micro-motion limit switch that contacts the oil tank of the transformer is provided with a temperature-measuring element for obtaining the oil tank temperature of the transformer. The monitoring methods of transformer condition monitoring devices include: Collect the surface temperature of the transformer oil tank and the ambient temperature and humidity, and calibrate the surface temperature of the transformer oil tank based on the ambient temperature and humidity. The status of the temperature-sensing micro switch is collected, and an alarm is triggered if the temperature-sensing micro switch is in the open state. Obtain current transformer load information; Calculate the real-time temperature difference between the oil tank surface temperature and the ambient temperature, the rate of change of temperature difference, the standard deviation of temperature difference, and the hourly slope of ambient temperature change, and input them into the transformer fault analysis model. The transformer fault analysis model includes: a BP neural network unit for handling deterministic fault modes; An ART2 neural network unit that receives the analysis results of the BP neural network unit, identifies abnormal heat distribution patterns, and makes judgments in conjunction with a dynamic threshold mechanism; After data acquisition is complete, perform low-power management of peripheral devices. The judgment method of the transformer fault analysis model includes: Primary threshold alarm; In-depth diagnostic phase; Fusion diagnosis; The primary threshold alarm includes: By mapping the input parameters; By curve fitting, the probability distribution set of four types of deterministic faults is output: P=[Poverload, Pradiator blockage, P oil circulation failure, P environmental abnormality]; P represents the probability distribution of a specific fault. If the probability of a fault is greater than a preset threshold, it is marked as a candidate fault. If ΔT > 25℃ and Load_current_ratio > 90%, the fault type is P overload; If ΔT_rate > 3℃ / min for 10 minutes, the fault type is P radiator blockage; If the ΔT fluctuation coefficient > 0.4, the fault type is P oil circulation fault; The in-depth diagnostic phase includes: Standardize the temperature difference time series data; A competitive subsystem is generated in the intermediate layer, and the clustering threshold is dynamically adjusted by comparing coefficients. Match the historical fault pattern library at layer F2 and output fuzzy faults and confidence levels; The fusion diagnosis includes: assigning scoring coefficients to the fuel tank surface temperature, the temperature difference between the fuel tank surface and the ambient temperature, and the temperature standard deviation over a given time range; combining the fault probability output by the BP neural network unit and the fault confidence output by the ART2 neural network unit, as well as the weights of the indicators, to calculate the final score (Final_Score) for each fault; the formula for the Final_Score is as follows: ; in, For fault confidence, Fault weights; Based on the Final_Score scores, the results are sorted from highest to lowest, with the highest score being the first potential fault score, and so on to obtain the second potential fault scores. The Final_Score scores are compared with the preset thresholds defined as percentages for the faults. If the fault score is higher than the preset threshold, the fault is determined to exist and the decision result is output. If the difference between the second potential fault score and the Final_Score score is less than 15%, a mixed fault warning is triggered.

2. The transformer condition monitoring device according to claim 1, characterized in that: The power supply components include: a main battery, a backup battery, and a charging unit with automatic switching connected to any one or both of the main battery and the backup battery. The charging unit includes a solar power module and a thermoelectric thin film, wherein the thermoelectric thin film is bonded to the temperature measuring element to form an integral film. The thermoelectric film includes a high-temperature surface in contact with the transformer's oil tank and a low-temperature surface with a heat sink attached.

3. The transformer condition monitoring device according to claim 1, characterized in that: The surface of the temperature measuring component is made of a thermally conductive material, which is a high-heat-conducting copper mold or aluminum material. A heat-insulating material is provided between the ambient temperature detection sensor and the sensor body.

4. The transformer condition monitoring device according to claim 1, characterized in that: The transformer condition monitoring device also includes: A wireless module connected to the processor, the wireless module including a network communication module and a low-power wireless module.

5. The transformer condition monitoring device according to claim 1, characterized in that: The calculation of the standard deviation of the temperature difference includes: in, This indicates the real-time difference between the surface temperature of the fuel tank and the ambient temperature. The standard deviation of temperature difference; the average rate of change of temperature difference. N represents the total number of temperature difference data points within the statistical period. This represents the loop variable, ranging from 1 to N.

6. The transformer condition monitoring device according to claim 1, characterized in that: The input expression for the transformer fault analysis model is: Input=[ΔT,ΔT_rate, [, Ambient_temp_trend, Humidity, Load_current_ratio, Load_profile]; where ΔT represents the average temperature difference; ΔT_rate represents the rate of change of temperature difference; The standard deviation of the temperature difference is represented by: Ambient_temp_trend represents the hourly slope of the ambient temperature change; Load_profile represents the similarity of the load profile over the past 24 hours; Humidity represents the ambient humidity; and Load_current_ratio represents the ratio of the load current to the rated current.

7. The transformer condition monitoring device according to claim 1, characterized in that: When collecting the surface temperature of the transformer tank and the ambient temperature and humidity, a temperature correction coefficient EC is set for the surface temperature of the tank to compensate for temperature. The expression for the temperature correction factor EC is EC = 1 + 0.02 × (Humidity - 50); Wherein, Humidity represents ambient humidity; the calculation method for the corrected fuel tank surface temperature includes: T(correction value) = T(measured value) * EC.

Images