Oil-immersed transformer distributed temperature measurement method based on fluorescent optical fiber sensor

Abstract

The invention relates to the technical field of power equipment state monitoring, in particular to an oil-immersed transformer distributed temperature measurement method based on a fluorescent optical fiber sensor, and the method comprises the steps: arranging the fluorescent optical fiber sensor in a key temperature rise region in a transformer in a partitioned and layered three-dimensional topological structure, and forming a distributed temperature measurement network; exciting light is injected through a pulse laser to excite a fluorescence signal; a dual-channel phase-locked amplification technology is used to collect signals, and a dual-weight adaptive attenuation model and an oil flow coupling compensation function are combined to demodulate the temperature; reconstructing a dynamic temperature field based on a three-dimensional thermodynamic inversion algorithm, and marking high-gradient hot spots; and temperature, load current and oil flow velocity data are fused to realize graded alarm. According to the invention, the limitation of traditional single-point monitoring is broken through, the strong electromagnetic interference resistance is excellent, a global temperature field can be accurately reconstructed, dynamic early warning is realized, and the operation safety and the operation and maintenance efficiency of the transformer are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment status monitoring, and in particular to a distributed temperature measurement method for an oil-immersed transformer based on a fluorescent optical fiber sensor. Background Art

[0002] Oil-immersed transformers are core equipment in power systems, and their operating status directly affects the safety and stability of the power grid. When the equipment is operating, internal windings, cores, and other components generate heat due to energy loss. If local temperatures are too high or hot spots exceed the tolerance of the insulation material, this can accelerate insulation aging, cause short circuits, and even damage the equipment. Therefore, real-time, accurate temperature monitoring of key internal areas is crucial. Current transformer temperature measurement technology has many limitations, making it difficult to meet the needs of full-area, high-precision monitoring:

[0003] Disadvantages of traditional contact temperature measurement technology:

[0004] Mainstream contact sensors, such as thermocouples and RTDs, are limited by installation conditions and can only be deployed at a few discrete points. They cannot cover critical temperature-rising areas, such as between high-voltage winding layers and on the core surface, making it difficult to detect localized micro-scale overheating. Furthermore, these sensors have weak electromagnetic interference resistance and are prone to signal distortion in the complex electromagnetic environment inside the transformer. Furthermore, their materials and lead insulation layers are affected by long-term oil immersion and high temperatures, posing aging and safety risks.

[0005] Limitations of non-contact and indirect temperature measurement technologies:

[0006] Non-contact methods such as infrared temperature measurement and oil level thermometers can only monitor the surface or oil temperature of the equipment, but cannot reflect the actual temperature of internal components such as windings and cores, which can vary significantly. Thermal simulation calculations based on load current rely on empirical parameters, making them difficult to handle unexpected operating conditions such as cooling system anomalies and local structural defects, and lacking accuracy.

[0007] Adaptability issues of existing optical fiber temperature measurement technology:

[0008] To solve the problem of electromagnetic interference, the fiber optic temperature measurement technology explored by the industry still has obvious shortcomings: limited sensitivity and accuracy, greatly affected by environmental noise, and poor performance in high temperature and high gradient scenarios; it does not fully consider the interference of factors such as internal oil flow in the transformer, further affecting the accuracy of temperature measurement; and can only provide discrete point temperature data, unable to reconstruct the internal full-domain temperature distribution, making it difficult to locate the spatial position and diffusion trend of hot spots, resulting in delayed early warning.

[0009] Therefore, to address the above problems, a distributed temperature measurement method for oil-immersed transformers based on fluorescent fiber optic sensors is proposed. Summary of the Invention

[0010] The object of the present invention is to provide a distributed temperature measurement method for an oil-immersed transformer based on a fluorescent optical fiber sensor, so as to solve the problems raised in the above background technology.

[0011] To achieve the above object, the present invention provides the following technical solutions:

[0012] A distributed temperature measurement method for oil-immersed transformers based on a fluorescent optical fiber sensor includes the following steps:

[0013] S1. Sensor Layout: Fluorescent fiber optic sensors are deployed in a partitioned and layered three-dimensional topology in key temperature-rise areas within the oil-immersed transformer, including between high-voltage and low-voltage winding layers, on the core surface, and in hot spots in the heat dissipation oil channel. The sensors consist of multiple sections of fluorescent optical fiber connected in series via electromagnetic interference-resistant fiber optic patch cords to form a distributed temperature measurement network. The length of each optical fiber section is customized based on the spatial characteristics of the monitoring area.

[0014] S2. Excitation signal injection: A pulsed laser is used to inject a pulsed excitation light with a wavelength of 380 nm to 420 nm into the fluorescent fiber sensor to excite the fluorescent substance to generate a fluorescence afterglow signal.

[0015] S3, signal acquisition and demodulation: real-time acquisition of fluorescence afterglow signals, use of dual-channel synchronous phase-locked amplification technology to separate environmental noise, and obtain temperature measurements by calculating the fluorescence decay time constant;

[0016] S4, Temperature Field Reconstruction: Combining the sensor's three-dimensional spatial coordinates with real-time temperature data, the dynamic temperature field distribution inside the transformer is reconstructed based on a thermodynamic inversion algorithm, identifying and marking hot spots where the temperature gradient exceeds 5 degrees Celsius per centimeter;

[0017] S5. Perform time-series correlation analysis on the temperature data, transformer load current, and cooling oil flow rate. When the temperature change rate exceeds the dynamic threshold and does not match the load change trend, activate the graded alarm mechanism.

[0018] As a preferred solution, the sensor deployment in step S1 includes:

[0019] The optical fiber is laid in a spiral winding pattern between winding layers. The outer side of the optical fiber is covered with an oil-resistant polyimide protective layer with a thickness of no more than 0.2 mm and fixed to the winding insulation support bar by micro ceramic clips.

[0020] A dual-redundant cross-layout method is used in the hot spots of the heat dissipation oil channel, and two sections of optical fiber are cross-fixed at the inlet and outlet of the oil channel at an angle of 30 to 60 degrees.

[0021] As a preferred solution, the signal demodulation in step S3 adopts a dual-weight adaptive attenuation model:

[0022] First, the effective decay time constant is calculated, which is the product of the time constant of the fast fluorescence decay component and its real-time signal-to-noise ratio dynamic weight, plus the product of the time constant of the slow fluorescence decay component and its complementary weight, plus the oil flow disturbance compensation.

[0023] The real-time signal-to-noise ratio dynamic weight is equal to the real-time signal-to-noise ratio value divided by the sum of the real-time signal-to-noise ratio and the environmental noise coefficient;

[0024] The oil flow disturbance compensation is obtained by integration. The integral term includes the product of the transformer oil density and the square of the velocity gradient of the oil flow in the normal direction of the optical fiber, plus the product of the temperature-dependent dynamic viscosity and the square of the oil flow curl. The integration is performed along the length of the optical fiber microelement segment, and the result is multiplied by the optical fiber fluid sensitivity coefficient.

[0025] As a preferred solution, the temperature calculation uses the oil flow coupling compensation function:

[0026] The temperature value is equal to the fluorescent material calibration coefficient a multiplied by the square of the effective decay time constant, plus the coefficient b multiplied by the effective decay time constant, plus the constant term c, and finally the oil flow correction factor d multiplied by the logarithm of the oil flow velocity;

[0027] The oil flow velocity is obtained by calculating the square root of the sum of the squares of the three-dimensional velocity components, and a small offset is added to prevent zero-speed singularities.

[0028] As a preferred solution, the three-dimensional thermodynamic inversion algorithm of step S4 includes:

[0029] Establish a coupled heat conduction and convection equation, which includes the product of the transformer oil heat capacity and the temperature-time derivative equal to the divergence of the thermal conductivity and the second-order spatial derivative of temperature, plus the winding heat source power, minus the dot product of the oil flow velocity and the temperature gradient;

[0030] The regularized least squares inversion method is adopted. The objective function is the sum of the squares of the deviations between the measured temperature values ​​of all sensors and the calculated values, plus a smoothness constraint term of the global integration of the squares of the second-order spatial derivatives of the temperature field. The temperature field distribution is solved by minimizing the objective function.

[0031] As a preferred solution, the hierarchical alarm triggering conditions of step S5 are:

[0032] The triggering condition for the first level alarm is that the temperature time change rate is greater than the sum of the load current power function term and the cooling state exponential decay term, and the oil flow growth rate lags behind the load current growth rate;

[0033] The triggering conditions for the second-level shutdown protection are that the time rate of change of the hotspot area is greater than the dynamic diffusion threshold, and the maximum temperature gradient exceeds 8 degrees Celsius per centimeter;

[0034] where the dynamic diffusion threshold is a linear function of the base diffusion rate times the load current change from the rated value, plus a heat accumulation factor.

[0035] As a preferred option, the fluorescent fiber optic sensor uses a europium ion-doped silica-zirconia composite fluorescent material, with a temperature sensitivity of 0.85 percentage points of relative change in the decay time constant per degree Celsius, and a temperature measurement accuracy better than plus or minus 0.5 degrees Celsius; and the surface of the optical fiber is coated with a nano-aluminum oxide insulating layer with a thickness controlled within the range of 50 nanometers plus or minus 5 nanometers.

[0036] As a preferred solution, the winding heat source power is dynamically updated through the electro-thermal coupling model:

[0037] The heat source power is equal to the product of the winding resistivity and the square of the current density, plus the product of the core loss coefficient and the time derivative of the magnetic flux density;

[0038] Among them, the winding resistivity changes inversely with the increase in temperature, and the resistivity decreases by 0.4 percentage for every 1 degree Celsius increase in temperature.

[0039] As a preferred solution, the heat accumulation factor is calculated using a recursive algorithm with a forgetting factor:

[0040] The current moment heat accumulation factor is equal to the previous moment value multiplied by the exponential decay factor, plus the current temperature deviation value multiplied by the heat accumulation coefficient;

[0041] The exponential decay factor is determined by the thermal decay coefficient and the sampling time interval, and the heat accumulation coefficient is related to the thermal decay coefficient.

[0042] It can be seen from the technical solutions provided by the present invention that the distributed temperature measurement method for oil-immersed transformers based on fluorescent optical fiber sensors provided by the present invention has the following beneficial effects:

[0043] 1. Distributed, full-area, precise temperature measurement breaks through the limitations of traditional single-point monitoring:

[0044] Fluorescent fiber optic sensors are deployed in a partitioned and layered three-dimensional topology, covering key temperature-rise areas such as the interlayers between high-voltage and low-voltage windings, the core surface, and the heat dissipation oil channels. This enables distributed monitoring of the entire three-dimensional space inside the transformer, resolving the problem that traditional thermocouple / infrared temperature measurement can only obtain the temperature of local discrete points.

[0045] Combining a dual-weight adaptive attenuation model with an oil flow coupling compensation function significantly improves temperature measurement accuracy. Furthermore, through oil flow disturbance compensation, the temperature calculation deviation caused by oil flow impact is effectively eliminated.

[0046] 2. High reliability in strong electromagnetic environment, adaptable to extreme working conditions of transformers:

[0047] The fluorescent fiber optic sensor is made of special composite materials and coated with a high insulation layer on the surface. Combined with anti-electromagnetic interference design, it can work stably in the strong electric field and mechanical vibration environment inside the transformer, solving the signal distortion problem caused by electromagnetic interference in traditional electronic sensors.

[0048] The sensor is oil-resistant and high-temperature-resistant, adapting to the harsh environment of long-term operation of oil-immersed transformers;

[0049] 3. 3D dynamic temperature field reconstruction to accurately locate high-gradient hotspots:

[0050] Based on the heat conduction-convection coupled finite element model and the regularized least squares inversion algorithm, the three-dimensional temperature field inside the transformer is reconstructed in real time, which can capture microscale hot spots (such as local overheating between winding layers) that are difficult to identify with traditional methods.

[0051] By monitoring the maximum temperature gradient and analyzing the diffusion rate of hot spots, potential faults can be identified in advance, significantly shortening response time compared to traditional methods that rely on manual inspections.

[0052] 4. Multi-source fusion and hierarchical warning to reduce the risk of false alarms and missed alarms:

[0053] By integrating multi-source data such as temperature, load current, and oil flow rate, a linkage judgment criterion of "temperature change rate-load correlation threshold-oil flow response" is established, effectively controlling the false alarm rate of the first-level alarm and improving the accuracy of the second-level shutdown.

[0054] The heat accumulation factor is dynamically calculated, combined with adaptive adjustment of the hotspot diffusion threshold, to accurately distinguish between "normal load temperature rise" and "abnormal fault temperature rise", avoiding false operation caused by fluctuations in a single parameter;

[0055] 5. Full life cycle cost optimization and adaptation to engineering applications:

[0056] Fluorescent fiber optic sensors have a long lifespan, matching the transformer's operating cycle and reducing operation and maintenance costs compared to traditional sensors. The distributed network design connects multiple areas in series using a single optical fiber, reducing wiring complexity and saving hardware costs.

[0057] The system supports integration with standardized protocols and edge computing gateways, can be directly connected to the power dispatching system without additional modification, and is adaptable to the existing substation operation and maintenance system, making engineering implementation easy.

[0058] In summary, the present invention provides a high-precision, highly reliable, and fully covered temperature measurement solution for oil-immersed transformers through the full-chain innovation of "distributed perception-anti-interference transmission-precise solution-global reconstruction-intelligent early warning". It can significantly improve the safety of equipment operation, reduce operation and maintenance costs, and has important engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 The figure is a flow chart of the steps of the distributed temperature measurement method for oil-immersed transformers based on fluorescent fiber optic sensors of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0061] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0062] like Figure 1 As shown, an embodiment of the present invention provides a distributed temperature measurement method for an oil-immersed transformer based on a fluorescent optical fiber sensor, comprising the following steps:

[0063] S1. Sensor Layout: Fluorescent fiber optic sensors are deployed in a partitioned and layered three-dimensional topology in key temperature-rise areas within the oil-immersed transformer, including between high-voltage and low-voltage winding layers, on the core surface, and in hot spots in the heat dissipation oil channel. The sensors consist of multiple sections of fluorescent optical fiber connected in series via electromagnetic interference-resistant fiber optic patch cords to form a distributed temperature measurement network. The length of each optical fiber section is customized based on the spatial characteristics of the monitoring area.

[0064] S2. Excitation signal injection: A pulsed laser is used to inject a pulsed excitation light with a wavelength of 380 nm to 420 nm into the fluorescent fiber sensor to excite the fluorescent substance to generate a fluorescence afterglow signal.

[0065] S3, signal acquisition and demodulation: real-time acquisition of fluorescence afterglow signals, use of dual-channel synchronous phase-locked amplification technology to separate environmental noise, and obtain temperature measurements by calculating the fluorescence decay time constant;

[0066] S4, Temperature Field Reconstruction: Combining the sensor's three-dimensional spatial coordinates with real-time temperature data, the dynamic temperature field distribution inside the transformer is reconstructed based on a thermodynamic inversion algorithm, identifying and marking hot spots where the temperature gradient exceeds 5 degrees Celsius per centimeter;

[0067] S5. Perform time-series correlation analysis on the temperature data, transformer load current, and cooling oil flow rate. When the temperature change rate exceeds the dynamic threshold and does not match the load change trend, activate the graded alarm mechanism.

[0068] In this embodiment, the sensor deployment in step S1 includes:

[0069] The optical fiber is laid in a spiral winding pattern between winding layers. The outer side of the optical fiber is covered with an oil-resistant polyimide protective layer with a thickness of no more than 0.2 mm and fixed to the winding insulation support bar by micro ceramic clips.

[0070] A dual-redundant cross-layout method is used in the hotspot area of ​​the heat dissipation oil channel. Two sections of optical fiber are fixed at the oil channel inlet and outlet at an angle of 30 to 60 degrees.

[0071] Furthermore, step S1 ensures full coverage of the key temperature-rise areas within the oil-immersed transformer by the fluorescent fiber sensor through the precise layout of the partitioned and layered three-dimensional topology, while avoiding interference with the transformer's insulation performance and oil flow field. This provides a reliable spatial coordinate reference and signal transmission channel for subsequent temperature acquisition and analysis. The following are the detailed steps:

[0072] Step S1-1: Three-dimensional coordinate mapping of key temperature rise areas:

[0073] Area Identification Basis: Based on transformer design drawings (including winding structure, core dimensions, and oil channel layout) and operating procedures (such as GB / T1094.2-2013), hotspots between high-voltage and low-voltage winding layers, on the core surface, and in the heat dissipation oil channels are identified as core monitoring areas. Hotspots are further precisely located based on thermal flow simulation results (e.g., areas with a heat flux density greater than 500W / m² under rated load, as calculated by ANSYS).

[0074] 3D scanning and coordinate modeling: An industrial-grade 3D laser scanner (accuracy ±0.05mm) is used to scan the monitored area inside the transformer to obtain the spatial dimensions of each area (such as the gap between winding layers of 0.5-2cm, the cross-sectional size of the oil channel of 3-8cm), the curvature radius (the average curvature of the winding is 15-30cm), and the spacing between adjacent components (such as the spacing between the core and the winding of 5-10cm). Based on the scanning data, a 3D coordinate system is constructed (with the center of the transformer tank as the origin, the axial direction as the Z axis, and the radial direction as the X axis), and the boundary coordinate matrix of each monitoring area is output. ,in, is the boundary coordinate matrix of the monitoring area, 、 、 Respectively The boundary points in the three-dimensional coordinate system 、 、 coordinate, is the number of regional boundary points;

[0075] Environmental parameter recording: synchronously record the electromagnetic environment of each area (such as the power frequency electric field strength 10-50kV / m near the high-voltage winding), oil flow path (obtain the flow velocity vector field in the oil channel through CFD simulation) and mechanical vibration frequency (10-100Hz), providing a basis for subsequent optical fiber selection and fixing method design;

[0076] Step S1-2: Partitioned and hierarchical three-dimensional topology design:

[0077] Zoning rules: Divide the monitoring area according to the dual dimensions of "electrical characteristics + thermal sensitivity":

[0078] High-voltage winding area (A area): including winding layers of different voltage levels such as 110kV / 220kV / 500kV, with one monitoring sub-area set up for every 2-3 winding layers;

[0079] Low-voltage winding area (Area B): To address the Joule heat concentration caused by high current on the low-voltage side, sub-areas are set up along the winding axis (each section is ≤1m long);

[0080] Core area (Area C): Focus on covering the core column, iron yoke and clamp surface, and set up sub-areas according to "column + plane";

[0081] Heat dissipation oil channel area (area D): includes the oil inlet, oil outlet and oil channel bend (area with significant eddy current effect), and has upstream and downstream sub-areas according to the direction of oil flow;

[0082] Layered logic setting: Along the radial direction (X-axis) of the transformer, the system is laid out in three layers: "inner-middle-outer": the inner layer is close to the winding conductor (5-10 mm from the conductor surface), the middle layer is located between the insulation paper layers, and the outer layer is close to the oil channel. Along the axial direction (Z-axis), a horizontal layer is set every 0.5 m in height to ensure that there is no blind spot in the vertical thermal gradient monitoring.

[0083] Topology verification: Use computer simulation to simulate the fiber optic layout path to verify whether it meets the following requirements:

[0084] Each hotspot area has at least two sections of optical fiber cross coverage;

[0085] The total length of optical fiber is less than or equal to the available wiring space inside the transformer (usually less than or equal to 50m);

[0086] The minimum spacing between adjacent fiber segments is ≥5cm (to avoid signal crosstalk);

[0087] Step S1-3: Fluorescent fiber selection and environmental resistance pretreatment:

[0088] Optical fiber core parameter selection:

[0089] Fluorescent material: Eu 3+ Doped SiO2-ZrO2 composite fiber to ensure temperature sensitivity meets ,in, is the temperature sensitivity, which represents the ratio of the relative change rate of fluorescence lifetime to the temperature change rate. is the fluorescence decay time constant, is temperature;

[0090] Fiber diameter: Bare fiber diameter 50-100μm (taking into account flexibility and mechanical strength), total diameter after coating with oil-resistant polyimide protective layer ≤200μm (to avoid affecting the winding insulation distance);

[0091] Insulation performance: The surface is coated with 50±5nm nano-Al2O3 insulation layer, and the insulation resistance is ≥10 14 Ω・cm (compliant with DL / T2348-2021 standard);

[0092] Environmental resistance pretreatment process:

[0093] Fiber optic cleaning: Ultrasonic cleaning (30 kHz, 5 min) with anhydrous ethanol was used to remove surface impurities, and the samples were dried in a 120 °C oven for 2 h after drying with nitrogen.

[0094] Protective layer curing: The polyimide protective layer is thermally cured (200°C, 1 hour) to improve its resistance to transformer oil (No. 25 mineral oil) immersion (after a 1000-hour immersion test, no swelling or cracking);

[0095] End treatment: Both ends of the optical fiber are cut flat (end surface roughness ≤ 0.5μm), and anti-electromagnetic interference optical fiber jumpers are fused (the jumpers are made of stainless steel armor, with shielding effectiveness ≥ 80dB@1MHz-1GHz);

[0096] Step S1-4: Implementation of region-specific fixation:

[0097] Winding layer fixation (A / B area):

[0098] Layout method: spiral winding, winding one circle every 10cm along the winding axis, and the winding radius is 5mm larger than the outer diameter of the winding (to avoid squeezing the insulation layer);

[0099] Fixing parts: Use micro ceramic clips (size 3×2×1mm, high temperature resistance 200°C), adhered to the winding insulation paper with high temperature resistant epoxy adhesive (shear strength ≥15MPa), and the clip spacing is ≤5cm to ensure that the optical fiber does not float and jitter;

[0100] Stress control: The winding tension is controlled at 5-10N (monitored in real time by a tensiometer) to avoid excessive stretching of the optical fiber, which may lead to increased attenuation (tests show that when the tension is greater than 15N, the signal attenuation increases by ≥0.5dB / m);

[0101] Cooling oil channel fixing (area D):

[0102] Laying method: A dual-redundant cross structure is used, with two sections of optical fiber laid along the length and width of the oil channel respectively, with a crossing angle of 30°-60° (adjusted according to the cross-sectional size of the oil channel, 45° for rectangular oil channels);

[0103] Fixed position: The entrance section of the optical fiber is 5 cm away from the oil channel entrance, and the exit section of the optical fiber is 5 cm away from the oil channel outlet. It is fixed to the inner wall of the oil channel by a polytetrafluoroethylene clamp (oil-resistant, low-friction). A 0.1 mm thick silicone rubber pad is placed between the clamp and the optical fiber to cushion the impact of oil flow.

[0104] Redundant design: Two sections of optical fiber are routed independently to ensure backup signals when a single section fails (redundancy ≥ 2);

[0105] Core surface fixed (area C):

[0106] Layout method: Lay out in a straight line along the axial direction of the core column, and lay out along the arc curve at the iron yoke, fitting the surface contour of the core (bending radius ≥ 5cm to avoid signal loss caused by excessive bending);

[0107] Fixing method: Use flexible high-temperature resistant tape (temperature resistance 150℃, adhesion ≥0.5N / cm) at intervals of 10cm, with a tape width of ≤5mm (to reduce the impact on core heat dissipation);

[0108] Step S1-5: Distributed network series connection and connectivity verification:

[0109] Customization of fiber segment length: Cut fiber segments according to the spatial characteristics of each monitoring area: winding layer segment length = winding height × π × number of turns (e.g. 1m high winding with 5 turns, segment length ≈ 15.7m); oil channel segment length = oil channel length + redundancy at both ends (0.5m each); iron core segment length = iron core unfolded length + 0.3m redundancy;

[0110] Series connection: Connect the fiber segments in each area in series through anti-electromagnetic interference fiber jumpers (length 0.5-1m). The jumpers and fiber segments are connected by fusion splicing (fusion loss ≤ 0.1dB). The fusion points are covered with stainless steel protective tubes (diameter 3mm) and fixed in non-hot spots.

[0111] Connectivity and signal testing: After the series connection is complete, the optical loss of the entire network is tested using an optical time-domain reflectometer (OTDR) (total loss ≤ 5dB) to ensure that there are no broken fibers or loose connections. 400nm pulsed light (10mW power) is injected to verify that each fiber segment can normally output a fluorescent signal (signal-to-noise ratio ≥ 30dB). The vibration of the transformer during operation is simulated (10-100Hz, amplitude 0.1mm). After 30 minutes of testing, no abnormal signal fluctuations are observed (fluctuation amplitude ≤ 1%).

[0112] In this embodiment, step S2 is used to precisely control the parameters and transmission path of the pulsed excitation light to efficiently excite the characteristic afterglow signal of the fluorescent substance in the fluorescent fiber, while suppressing electromagnetic interference and environmental noise, thereby providing a high signal-to-noise ratio original excitation source for subsequent signal demodulation. The following are the detailed steps:

[0113] Step S2-1: Pulsed laser parameter customization and performance benchmark calibration:

[0114] Core parameter matching design: for Eu 3+ The energy level structure of the doped SiO2-ZrO2 composite fluorescent material (its 4f-4f transition has a strong absorption peak at 394nm) uses an InGaN-based semiconductor pulse laser to ensure:

[0115] Wavelength coverage: 380nm–420nm (central wavelength: 400±5nm, excitation efficiency can be maintained at ≥80% when the deviation from the material absorption peak is ≤10nm);

[0116] Pulse width 5–50 ns (should be shorter than the fast decay component of fluorescence The typical value is 100ns, to avoid the overlap of the excitation light and the fluorescence signal in the time domain. After testing, a 50ns pulse can reduce the overlapping interference to ≤5%).

[0117] Single pulse energy 10–50 μJ (too low energy will result in insufficient fluorescence intensity, while too high energy will cause photobleaching - after 1 hour of continuous 50 μJ irradiation, a decrease in fluorescence intensity of ≤3% is considered acceptable);

[0118] Wavelength stability calibration: In a constant temperature environment of 25°C (±0.1°C), the output wavelength is calibrated using a high-precision spectrometer (resolution 0.05nm). The wavelength drift is controlled within ±0.5nm / °C using the laser's built-in temperature control module (total drift ≤5nm when the ambient temperature fluctuates by 10°C). After calibration, the wavelength-temperature response curve is recorded for subsequent real-time compensation.

[0119] Power consistency verification: Use photodiode array (response time ≤ 1ns) to monitor continuous The energy fluctuation of each pulse must have a coefficient of variation of ≤2% (i.e., 99.7% of the pulse energies must fall within ±3σ of the mean). If this exceeds the limit, closed-loop drive current regulation is initiated (drive current 50–200 mA, with an adjustment accuracy of 0.1 mA) to ensure that output stability complies with the standard of "excitation light power fluctuation ≤±2%" in DL / T2348-2021.

[0120] Step S2-2: Dynamic adaptive adjustment of pulse timing and energy:

[0121] Adaptive repetition rate control: based on the effective fluorescence decay time (typical value 1-5μs), set the laser repetition frequency satisfy (Reserve 100% safety margin) (Among them, is the laser repetition frequency, is the effective fluorescence decay time constant); for example:

[0122] when s, kHz, 1-10kHz is actually selected (to avoid fluorescence superposition caused by high-frequency pulses, the test shows kHz, the residual amount of pre-sequence fluorescence ;

[0123] By linking with real-time temperature field data decreases with increasing temperature), dynamically adjust (Each temperature rise , can be increased by 10%);

[0124] Energy gradient allocation strategy: Differentiate the pulse energy allocation based on the signal attenuation characteristics of different monitoring areas:

[0125] High-voltage winding area (strong electromagnetic interference, signal attenuation ≥20%): single pulse energy 30–50 μJ, ensuring that the energy reaching the end of the optical fiber is ≥10 μJ;

[0126] Oil channel area (oil flow scattering causes about 15% signal loss): energy 20–30 μJ;

[0127] Core area (signal attenuation ≤ 10%): energy 10–20 μJ (to avoid material aging caused by overexcitation);

[0128] Peak power constraint: by formula (in, is the peak power, is the single pulse energy, (Pulse width) controls the peak power in the range of 5-50W:

[0129] When W, the fluorescence signal-to-noise ratio (SNR) is less than 25dB, and the demodulation error increases to more than ±2°C;

[0130] W, the fluorescence quantum yield dropped by 215% after 1 hour of continuous irradiation, so it is necessary to monitor in real time through the power meter and adjust the closed loop. and The ratio of

[0131] Step S2-3: Integrated design of optical path to resist strong electromagnetic interference:

[0132] Low-loss optical chain construction: uses the transmission path of "laser → optical isolator → collimating lens → coupler → sensing fiber". Key component parameters:

[0133] Optical isolator: operating wavelength 380–420 nm, isolation ≥ 35 dB (prevents light reflected from the fiber end face (typical reflectivity 4%) from being fed back into the laser, reducing output fluctuation from ±5% to within ±1%);

[0134] Collimating lens: focal length 8mm, numerical aperture , collimate the laser beam into parallel light with a diameter of 1.5mm (divergence angle ≤ 0.5mrad) to reduce transmission loss;

[0135] Fiber coupler: GRIN lens type, coupling efficiency ≥85% (for 50-100μm core diameter sensing fiber, ), coupling loss ≤1.2dB;

[0136] Electromagnetic shielding system construction:

[0137] The laser and optical components are encapsulated in a brass shielding box (3mm thick), the inner wall of which is covered with carbonyl iron absorbing material (for Frequency band attenuation ≥70dB), can resist 50kV / m power frequency electric field interference inside the transformer;

[0138] Fiber optic patch cord mm stainless steel bellows, both ends are grounded through 360° electromagnetic sealing joints (grounding resistance , suppress common mode interference;

[0139] The drive circuit power supply uses an isolation transformer (isolation voltage 22kV) and a low-pass filter (cut-off frequency 1kHz) to reduce conducted interference;

[0140] Temperature drift compensation: Because the laser wavelength drifts with temperature (0.08nm / °C), the shielding box has a built-in TEC semiconductor cooler, which is monitored in real time by a Pt100 temperature sensor (accuracy ±0.1°C). The PID control temperature is stabilized at 25±0.3°C (at this time, the wavelength drift is ≤±0.024nm, ensuring a match of ≥95% with the material absorption peak);

[0141] Step S2-4: Excitation efficiency verification and closed-loop feedback optimization:

[0142] Fluorescence characteristic spectrum detection: The fluorescence spectrum at the output end of the sensing fiber is collected by a micro-spectrometer (resolution 0.5nm), which must meet the following requirements:

[0143] Characteristic peak intensity ratio: 615nm ( 5 D0→ 7 F2 transition) and 590nm ( 5 D0→ 7 F1 transition) peak intensity ratio ≥ 3:1 (when the ratio is stable, the material excitation state is normal);

[0144] Stimulation efficiency (in, for the excitation efficiency)( (When checking the coupler alignment accuracy or fiber damage);

[0145] Spatial uniformity verification: Fluorescence intensity is tested along the length of the sensing fiber (sampling every 1m), and the relative deviation is ≤±8%:

[0146] If the deviation of a certain section is greater than 10%, check whether there is excessive bending (when the bending radius is less than 5cm, mode loss will cause the local strength to drop by ≥15%).

[0147] The signal difference of the two optical fibers cross-laid in the oil channel area should be ≤5% to ensure the effectiveness of the redundant signal;

[0148] Real-time feedback adjustment: Building a closed loop of "spectrometer → MCU → laser driver":

[0149] when When the pulse energy is increased by 5 μJ / step (upper limit 50 μJ);

[0150] If the fluorescence intensity drops by ≥5% within 1 hour (a sign of photobleaching), the repetition rate will be automatically reduced by 20% and the decay rate will be recorded (a maintenance alert will be triggered if it exceeds 0.5% / h);

[0151] Step S2-5: Timing synchronization and interference source suppression:

[0152] Multi-module clock alignment: achieved through synchronous trigger signal (TTL level, rising edge ≤ 5ns):

[0153] The laser pulse emission time is , signal acquisition module delay (pulse width) and then start (such as ns, the data is collected from ns start);

[0154] The acquisition window duration is set to (Make sure to cover the full attenuation curve, e.g. s, the window length is 10 μs), which matches the integration time of the subsequent lock-in amplifier;

[0155] Background light filtering: A 380-420nm bandpass filter (bandwidth ±10nm) is inserted into the optical path to attenuate the autofluorescence (400-500nm) of the transformer oil-paper insulation material by 295%, increasing the fluorescence signal SNR to ≥35dB.

[0156] Pulse waveform optimization: Rectangular pulses are shaped into Gaussian pulses Width = set pulse width), its smooth edge can reduce electromagnetic radiation (radiation intensity is reduced by 225dB at 1MHz), avoiding interference with the transformer's internal relay protection signal;

[0157] Through the above steps, it is possible to stably output pulsed light that meets the excitation requirements of fluorescent materials in harsh environments such as strong electromagnetic fields, high temperatures, and oil immersion, ensuring the intensity and characteristic stability of the fluorescence afterglow signal and providing reliable input for subsequent temperature demodulation.

[0158] In this embodiment, the signal demodulation in step S3 adopts a dual-weight adaptive attenuation model:

[0159] First, the effective decay time constant is calculated, which is the product of the time constant of the fast fluorescence decay component and its real-time signal-to-noise ratio dynamic weight, plus the product of the time constant of the slow fluorescence decay component and its complementary weight, plus the oil flow disturbance compensation.

[0160] The real-time signal-to-noise ratio dynamic weight is equal to the real-time signal-to-noise ratio value divided by the sum of the real-time signal-to-noise ratio and the environmental noise coefficient;

[0161] The oil flow disturbance compensation is obtained by integration. The integral term includes the product of the transformer oil density and the square of the velocity gradient of the oil flow in the direction normal to the optical fiber, plus the product of the temperature-dependent dynamic viscosity and the square of the oil flow curl. The integration is performed along the length of the optical fiber microelement segment, and the result is multiplied by the optical fiber fluid sensitivity coefficient.

[0162] The temperature calculation uses the oil flow coupling compensation function:

[0163] The temperature value is equal to the fluorescent material calibration coefficient a multiplied by the square of the effective decay time constant, plus the coefficient b multiplied by the effective decay time constant, plus the constant term c, and finally the oil flow correction factor d multiplied by the logarithm of the oil flow velocity;

[0164] The oil flow velocity is calculated by taking the square root of the sum of the squares of the three-dimensional velocity components and adding a small offset to prevent zero-speed singularities.

[0165] Furthermore, step S3 uses high-precision signal acquisition and intelligent demodulation algorithms to extract the decay characteristics of the fluorescence afterglow from the strong noise background, accurately calculate the temperature-sensitive decay time constant, and provide high signal-to-noise ratio raw data for temperature field reconstruction. The following are the detailed steps:

[0166] Step S3-1: Setting up the fluorescence afterglow signal acquisition system:

[0167] Photoelectric detection module selection and configuration:

[0168] Use back-illuminated InGaAs photodetector (response band 450–850nm, similar to Eu 3+Fluorescence emission peak (615nm) matching), quantum efficiency ≥80%@615nm, dark current ≤10pA (25℃), can effectively capture weak fluorescence signals (minimum detectable power ≤-80dBm);

[0169] The front end of the detector is connected in series with a 615nm bandpass filter with a 3nm bandwidth (cut-off depth OD6) to filter out the residual 380–420nm excitation light (attenuation ≥10 6 times) and transformer oil autofluorescence (400–550 nm), increasing the purity of the fluorescence signal to ≥99%;

[0170] Using a low noise preamplifier (gain 10 4 –10 6 times, noise figure ≤1.5dB@1kHz), convert the photocurrent signal (nA level) into a voltage signal (mV level), and set the bandwidth to 10Hz–1MHz (covering the frequency range of the fast and slow components of fluorescence decay);

[0171] Timing synchronization acquisition control:

[0172] Based on the trigger signal of the pulse laser (t=0), the synchronous control signal is generated by FPGA: after the laser pulse ends (delay ns) to start the acquisition, and the acquisition window duration is set to 10 ( The typical value of the fast decay component is 100ns, so the window duration is 1us) to ensure that the entire process of fluorescence from excitation to decay to baseline is fully recorded;

[0173] The sampling rate is set to 100MS / s (sampling interval 10ns), and 1000 data points are collected for each signal (covering a duration of 10μs). The fast and slow decay component signals are collected synchronously through a dual-channel ADC (16-bit resolution, ENOB ≥ 14 bits), with a quantization error of ≤ 0.02%;

[0174] Parallel acquisition of multi-channel signals:

[0175] For the N optical fiber segments (N≤32) in the distributed temperature measurement network, a 1×32 optical switch (switching time ≤5ms, insertion loss ≤1.5dB) is used for polling acquisition. Each segment of optical fiber acquires data 200 times per second (sampling frequency 200Hz), ensuring a capture bandwidth of ≥10Hz for dynamic temperature changes (capable of identifying temperature fluctuations within 50ms).

[0176] An optical attenuator (adjustable from 0 to 30 dB) is connected in series between the optical switch and the detector. When the fluorescence intensity is too high (exceeding the linear range of the detector by 1 V), it automatically attenuates the signal to avoid signal saturation (linearity error ≤ 0.5%).

[0177] Step S3-2: Dual-channel synchronous phase-locked amplification and noise separation:

[0178] Phase-locked amplifier core parameter matching:

[0179] Reference signal: at the laser pulse frequency (1-10kHz) as a reference, a square wave reference signal with the same frequency (duty cycle 50%) is generated, which is synchronized with the periodic excitation characteristics of the fluorescence signal;

[0180] Dual-channel configuration: Channel 1 focuses on the fast decay component ns), the integration time is set to 200 ns (matching the time scale of fast decay); channel 2 focuses on the slow decay component ( ns), the integration time is set to 1 μs (to cover the slow decay process), and the phase difference between the two channels is locked to 0° (to ensure that the signals are superimposed in phase);

[0181] Suppression ratio: Through a 24th-order Butterworth low-pass filter (roll-off rate 120dB / dec), the suppression ratio of the power frequency harmonics (50Hz×200) at 10kHz is ≥120dB, which can reduce the noise power from -50dBm to below -170dBm;

[0182] Noise separation and signal-to-noise ratio improvement:

[0183] To address the electromagnetic noise (10kHz–1MHz) and thermal noise (white noise) inside the transformer, a “correlation detection + differential amplification” combination is used: the lock-in amplifier only responds to the fluorescent signal with the same frequency and phase as the reference signal, and the suppression ratio of random noise is ≥10 4 :1;

[0184] Real-time calculation of signal-to-noise ratio (SNR) (Where SNR is the real-time signal-to-noise ratio), when SNR < 20dB (such as when the electromagnetic interference in the high-voltage winding area is strong), the integration time of the lock-in amplifier is automatically increased (to a maximum of 5μs), at this time the SNR can be increased to dB(Demodulation Error ;

[0185] Adopting an adaptive filtering algorithm (RLS recursive least squares), the noise model is estimated in real time through a noise reference channel (collecting electromagnetic signals from the transformer housing), eliminating common-mode interference from the collected signal and further improving the SNR by 3-5dB.

[0186] Signal preprocessing:

[0187] Baseline correction: The baseline signal (dark current + ambient noise) without excitation is recorded every 100 acquisitions, and the formula (in, is the corrected signal, is the original signal, is the baseline signal) to eliminate baseline drift (drift amount mV / °C);

[0188] Pulse interference elimination: using The criterion identifies abnormal pulses (amplitude exceeds 3 standard deviations from the mean) by interpolating adjacent data points (error , avoid distortion of the attenuation curve caused by accidental interference;

[0189] Step S3-3: Solving the dual-weight adaptive attenuation model:

[0190] Fluorescence decay curve fitting:

[0191] Corrected fluorescence signal Fitting using a double exponential decay model: (in, 、 is the amplitude of the fast / slow decay component, 、 is the corresponding time constant, is the fitting residual);

[0192] The Levenberg-Marquardt nonlinear least squares algorithm (iterations ≤ 50, convergence accuracy 1e-6) is used to extract the data within a 1μs acquisition window. (fitting error ≤ 2ns) and (fitting error ≤ 5ns), and calculate the residual sum of squares ( is the fitting signal), when When the fitting is judged to be invalid, re-sampling is triggered;

[0193] Double-weighted effective decay time calculation:

[0194] Dynamic adjustment of weight coefficients: weights are calculated based on real-time signal-to-noise ratio (SNR) , (in, is the environmental noise coefficient, when SNR dB, , prioritize trusting the fast decay component; when SNR=10dB, , enhance the weight of the slow component);

[0195] Basic effective decay time: (Integrate the temperature-sensitive characteristics of fast and slow components to reduce the impact of single component fluctuations);

[0196] Oil flow disturbance compensation: by formula Calculate the fluid disturbance correction term, where s m / kg is the optical fiber fluid sensitivity coefficient, is the transformer oil density (895kg / m³@25℃), is the temperature-dependent dynamic viscosity ( Pa·s@ , decreases linearly with increasing temperature). is the oil flow shear rate, is the oil flow vortex, and the final effective decay time (Compensate for the decay time drift caused by oil flow shock. Tests show that the error can be reduced from ±2.5°C to ;

[0197] Step S3-4: Temperature calculation based on oil flow coupling compensation:

[0198] Temperature-decay time calibration curve construction:

[0199] The fluorescent fiber was calibrated over the entire temperature range (20-120°C) in a constant temperature oil bath (accuracy ±0.05°C), and data were collected every 5°C. , get the original data pair

[0200] Use a quadratic polynomial to fit the base curve: (in, , , is the calibration coefficient of the fluorescent material, and the goodness of fit ;

[0201] Oil flow velocity coupling compensation:

[0202] Introducing the oil flow velocity model (in, is the modulus of the oil flow velocity vector, 、 、 The oil flow velocity vector is 、 、 Directional component) (real-time acquisition by the oil channel flow meter, accuracy ±0.01m / s), the correction formula is: (in, is the oil flow correction factor. The higher the flow rate, The larger the value; To prevent the minimum value of zero-speed singularity);

[0203] Compensation effect: When the oil flow rate increases from 0.1m / s to 0.5m / s, the uncompensated temperature error reaches ±3°C, and after compensation, the error is ≤±0.6°C (verified on an oil circulation test platform of a certain electrical engineering institute);

[0204] Real-time temperature calculation and filtering:

[0205] A single temperature calculation takes ≤1ms (meeting the 200Hz sampling rate requirement). A sliding average filter (window size 10) is applied to 10 consecutive calculation results to smooth out high-frequency noise (such as transient fluctuations caused by electromagnetic pulses), while retaining a temperature rise trend of ≥0.5°C / s (to avoid over-filtering that obscures true hotspots).

[0206] Outlier elimination: When the deviation between a calculated temperature and the average of the five adjacent values ​​is greater than 3°C, it is considered an outlier (e.g., local contamination of the optical fiber), and the average of the previous three values ​​is automatically used instead (the proportion of outliers is ≤ 0.1%).

[0207] Through the above steps, a temperature measurement accuracy of ±0.8°C (20–120°C) can be achieved in complex environments with strong electromagnetic interference (50kV / m) and oil flow disturbances (flow velocity 0–1m / s), providing high-fidelity point temperature data for subsequent three-dimensional temperature field reconstruction.

[0208] In this embodiment, the three-dimensional thermodynamic inversion algorithm of step S4 includes:

[0209] Establish a coupled heat conduction and convection equation, which includes the product of the transformer oil heat capacity and the temperature-time derivative equal to the divergence of the thermal conductivity and the second-order spatial derivative of temperature, plus the winding heat source power, minus the dot product of the oil flow velocity and the temperature gradient;

[0210] The regularized least squares inversion method is used. The objective function is the sum of the squared deviations between the measured temperature values ​​of all sensors and the calculated values, plus a smoothing constraint term for the global integration of the squared second-order derivative of the temperature field space. The temperature field distribution is solved by minimizing this objective function.

[0211] The winding heat source power is dynamically updated through the electro-thermal coupling model:

[0212] The heat source power is equal to the product of the winding resistivity and the square of the current density, plus the product of the core loss coefficient and the time derivative of the magnetic flux density;

[0213] Among them, the winding resistivity changes inversely with the increase of temperature, and the resistivity decreases by 0.4% for every 1 degree Celsius increase in temperature;

[0214] Furthermore, step S4 uses the discrete temperature data and spatial coordinate information from the distributed sensors to reconstruct the complete dynamic temperature field inside the transformer using a three-dimensional thermodynamic inversion algorithm. This accurately identifies high-gradient hotspots and provides a global thermal distribution basis for equipment status assessment. The detailed steps are as follows:

[0215] Step S4-1: Temperature-coordinate data preprocessing and spatiotemporal alignment:

[0216] Sensor space coordinate mapping:

[0217] Based on the three-dimensional coordinate system constructed in step S1-1 (with the center of the tank as the origin and the axial direction as the Axis, radial Axis), the temperature measurement point space coordinate matrix of each fluorescent optical fiber (in, is the sensor space coordinate matrix, 、 、 For the The three-dimensional coordinates of the temperature measurement points, is the total number of temperature measurement points, usually Corresponding temperature data (in, is the measured temperature vector, For the The measured temperature of the point) is associated to generate a temperature dataset with coordinate labels ;

[0218] Coordinate accuracy calibration: The coordinates of three fixed reference points (such as the marking points on the inner wall of the fuel tank) are measured by a laser tracker (accuracy ±0.03mm) to calibrate the sensor coordinate matrix. Perform affine transformation (translation + rotation + scaling) to correct installation errors (ensure coordinate deviation mm);

[0219] Temperature data spatiotemporal purification:

[0220] Outlier elimination: Use the local outlier factor (LOF) algorithm to eliminate outliers at each temperature measurement point. Calculate the temperature deviation between it and the five neighboring points. If the LOF value is greater than 1.5, it is considered abnormal (such as a jump caused by local optical fiber damage) and replaced by Kriging interpolation (spatial correlation weighting). The interpolation error is ≤ 0.3°C.

[0221] Time series alignment: Due to the 5ms time difference in multi-channel polling acquisition, all temperature measurement point data are aligned to the same time axis through FPGA timestamp (accuracy 10ns) (in, For the Sampling time, , sampling interval 100ms), to ensure the time synchronization of the dynamic temperature field (time difference ;

[0222] Spatial density enhancement: For temperature points in sparse areas (such as the gap between winding layers > 2cm), radial basis function (RBF) interpolation is used to supplement virtual temperature measurement points (density ≥ 1 point / cm³) ,provides denser constraints for the inversion algorithm;

[0223] Step S4-2: Construction of three-dimensional thermodynamic coupling model:

[0224] Geometric model and mesh division:

[0225] Based on the transformer design drawings (including the 3D dimensions of the windings, core, oil channels, and oil tank), a 1:1 geometric model is constructed in finite element software (such as ANSYS). The mesh in key areas is refined: tetrahedral elements (1-3 mm side length) are used between winding layers, hexahedral elements (2-5 mm side length) are used for the oil channels, and shell elements (5 mm thickness) are used for the oil tank walls. The total number of elements is approximately 500,000 to 1,000,000 (balancing calculation accuracy and efficiency).

[0226] Mesh quality verification: Ensure that the element distortion rate is ≤15% and the aspect ratio is ≤5 (using the mesh quality assessment tool) to avoid calculation errors caused by mesh distortion (error ≤2%);

[0227] Modeling temperature dependence of material parameters:

[0228] Winding (copper): density , specific heat capacity at constant pressure (in, is the specific heat capacity of copper at constant pressure, is temperature) (20-120℃), thermal conductivity (in, is the thermal conductivity of copper);

[0229] Iron core (silicon steel sheet): , (in, is the constant pressure specific heat capacity of silicon steel sheet), (in, is the thermal conductivity coefficient of silicon steel sheet (along the rolling direction);

[0230] Transformer oil (No. 25 mineral oil): (in, is the density of transformer oil), (in, is the constant pressure specific heat capacity of transformer oil), (in, is the thermal conductivity of transformer oil);

[0231] Insulation paper: , (in, is the constant pressure specific heat capacity of the insulating paper), (in, is the thermal conductivity of the insulating paper) (isotropic);

[0232] Heat conduction-convection coupled governing equations:

[0233] Adopting heat conduction-convection coupled model: ,in, is the material density, is the specific heat capacity at constant pressure, is temperature, For time, is the thermal conductivity coefficient, is the gradient operator, is the Joule heat source of the winding and the eddy current heat source of the core, is the oil flow velocity vector;

[0234] Boundary condition settings:

[0235] Outer wall of the fuel tank: natural convection heat dissipation, convection heat transfer coefficient W / ( ℃)(wherein, is the convective heat transfer coefficient, is the temperature difference between the wall and the environment), the ambient temperature ℃(real-time acquisition);

[0236] Winding and core contact interface: thermal resistance ℃ / W (where, is the contact thermal resistance) (considering the influence of the insulation layer);

[0237] Oil channel inlet and outlet: known oil flow velocity (provided by the oil flow rate sensor) and inlet oil temperature (measured value), set as Dirichlet boundary;

[0238] Step S4-3: Implementation of three-dimensional thermodynamic inversion algorithm:

[0239] Solving the problem:

[0240] Based on the above model and control equations, the finite volume method is used to solve the transient temperature field: time step ms (matching the sensor sampling frequency), each time step is solved by iteration (convergence criterion: residual , output calculated temperature field ,in, To calculate the temperature field, is the spatial coordinate, For time;

[0241] Dynamic update of heat source items:

[0242] Joule heat source of winding ,in, is the Joule heat source of the winding, is the conductivity, is the current density, where the conductivity ,in, S / m is the conductivity of copper at 20°C, current density ,in, is the load current, is the number of parallel conductors, is the cross-sectional area of ​​a single conductor;

[0243] Core eddy current heat source ,in, is the core eddy current heat source, W s / (m T) is the iron loss coefficient, is the magnetic flux density, W·s / (m·T) is the iron loss coefficient, is the magnetic flux density, obtained from electromagnetic simulation;

[0244] Total heat source ,in, is the total heat source, updated every 5 time steps (in response to changes in load current);

[0245] Regularized least squares inverse optimization:

[0246] Objective function construction: The inversion goal is to minimize the deviation between the measured and calculated temperatures while constraining the smoothness of the temperature field: ,in, For the The measured temperature of each sensor, For the model in sensor coordinates The calculated temperature at is the smoothness constraint factor, is the Laplace operator of the temperature field (characterizing the curvature), is the volume of the computational domain;

[0247] Optimization algorithm: The conjugate gradient method is used for iterative solution. The initial temperature field is set to uniform distribution (25°C). The temperature field parameters (such as heat source intensity and material thermal conductivity correction term) are updated in each iteration until the objective function converges (iteration number The difference drops to less than 1% of the initial value);

[0248] Step S4-4: Real-time reconstruction and time evolution of dynamic temperature field:

[0249] Real-time reconstruction trigger mechanism:

[0250] The inversion calculation is started every time 10 sets of temperature data are collected (about 1 second), and GPU parallel acceleration is used (such as NVIDIA A100, the calculation time is ≤ 500ms) to ensure that the reconstruction result lags behind the actual measurement time by ≤ 1 second (to meet the needs of dynamic monitoring);

[0251] Historical data cache: saves the temperature field sequence of the last 300 seconds (3000 time steps) for analyzing temperature evolution trends (such as hot spot diffusion speed);

[0252] Temperature field spatiotemporal interpolation and visualization:

[0253] For the discrete grid temperatures obtained by inversion, Kriging spatial interpolation (variogram model: Gaussian model, search radius 5 cm) is used to generate a global continuous temperature field, with an interpolation accuracy of ≤0.5°C (compared with the verification points);

[0254] Output forms: 3D cloud map (isothermal surface interval 5°C), cross-sectional temperature distribution (axial / radial / circumferential), critical path temperature curve (such as winding center axis), real-time display through visualization platform (such as Paraview);

[0255] Step S4-5: Hotspot area identification and gradient verification:

[0256] Temperature gradient calculation:

[0257] For the reconstructed temperature field, the spatial gradient is calculated by the central difference method: (in, is the temperature gradient, 、 、 are the partial derivatives of temperature in the x, y, and z directions respectively), the gradient modulus ,in, is the temperature gradient modulus (unit: °C / cm);

[0258] Hot spot area determination and marking:

[0259] Hotspot definition: And temperature The hotspot boundary is extracted by region growing algorithm (accuracy ±2mm) and its volume is calculated. ( is the hot spot volume), the maximum temperature ( is the maximum temperature of the hot spot) and the center coordinates ;

[0260] Grading marking: Level 1 hotspots (5-8°C / cm) and level 2 hotspots (>8°C / cm) are marked with different colors (e.g., yellow / red) on the visual interface, and a hotspot characteristic parameter table (position, gradient, volume, duration) is output;

[0261] Reconstruction accuracy verification:

[0262] Use reserved verification points (sensors not involved in inversion) to evaluate accuracy: calculate the accuracy of the verification points (in, To calculate the temperature, is the measured temperature), the average error is required , maximum error ;

[0263] If the error exceeds the limit, adjust the smoothing constraint factor (±0.1) or grid density (local refinement), and re-invert until the accuracy requirements are met;

[0264] Through the above steps, high-precision reconstruction of the three-dimensional dynamic temperature field inside the transformer can be achieved (spatial resolution ±5mm, time resolution 1s), high-gradient hot spots can be accurately identified, and a direct thermal distribution basis can be provided for equipment overheating fault warning.

[0265] In this embodiment, the hierarchical alarm triggering conditions of step S5 are:

[0266] The triggering condition for the first level alarm is that the temperature time change rate is greater than the sum of the load current power function term and the cooling state exponential decay term, and the oil flow growth rate lags behind the load current growth rate;

[0267] The triggering conditions for the second-level shutdown protection are that the time rate of change of the hotspot area is greater than the dynamic diffusion threshold, and the maximum temperature gradient exceeds 8 degrees Celsius per centimeter;

[0268] Wherein, the dynamic diffusion threshold is equal to the basic diffusion rate multiplied by the linear function of the load current change ratio relative to the rated value, plus the heat accumulation factor;

[0269] The heat accumulation factor is calculated using a recursive algorithm with a forgetting factor:

[0270] The current moment heat accumulation factor is equal to the previous moment value multiplied by the exponential decay factor, plus the current temperature deviation value multiplied by the heat accumulation coefficient;

[0271] Among them, the exponential decay factor is determined by the thermal decay coefficient and the sampling time interval, and the heat accumulation coefficient is related to the thermal decay coefficient;

[0272] Furthermore, step S5 is to build a multi-source time series correlation model by integrating the temperature data from the distributed temperature measurement system with the transformer operating status parameters (load current, oil flow rate, etc.). This model implements a hierarchical alarm based on dynamic thresholds, accurately identifies abnormal temperature rise risks, and triggers corresponding response mechanisms to ensure safe equipment operation. The following are the detailed steps:

[0273] Step S5-1: Multi-source operation data collection and spatiotemporal synchronization:

[0274] Data source and interface specifications:

[0275] Temperature data: Extract key features from the three-dimensional temperature field reconstructed in step S4, including the real-time temperature of each monitoring point , hot spot area temperature , maximum temperature gradient and hotspot area (sampling frequency 1Hz, timestamp accuracy 1ms);

[0276] Load current data: collect the three-phase current on the high-voltage side through a current transformer (accuracy level 0.2) 、 、 , converted to effective value , normalized to (in, is the rated load current, , covering 150% overload conditions);

[0277] Oil flow rate data: The oil flow velocity vector model is collected through the built-in ultrasonic flow meter (accuracy ±0.01m / s) in the oil channel. , synchronously record oil flow temperature (for viscosity correction);

[0278] Data interface: Using the IEC61850 standard protocol, multi-source data is uniformly connected to the edge computing gateway, with interface delay ≤10ms and data packet loss rate ≤0.1%;

[0279] Time and space synchronization calibration:

[0280] Timestamp alignment: Based on the transformer GPS synchronized clock (accuracy ±1μs), the timestamp of temperature, current and oil flow rate data is aligned. Perform linear interpolation correction to ensure time deviation of data from different sources ;

[0281] Data cleaning: A sliding average filter (with a window size of 5 points) is used on the load current data to eliminate high-frequency noise (such as current harmonic interference). A median filter (with a window size of 3 points) is used on the oil flow rate data to eliminate pulse interference (such as instantaneous jumps caused by oil flow turbulence). The signal-to-noise ratio of the processed data is ≥30dB.

[0282] Step S5-2: Time series feature extraction and multi-source correlation modeling:

[0283] Temperature dynamic feature extraction:

[0284] Temperature change rate: calculated by central difference method (in, is the rate of change of temperature with time, for The temperature value at the moment, is the sampling interval), focusing on the temperature rise rate in the hot spot area (resolution 0.1°C / s);

[0285] Hotspot diffusion characteristics: Calculate the rate of change of the hotspot area ,in, is the rate of change of the hotspot area over time, unit: , characterizes the hotspot spreading trend;

[0286] Temperature-load correlation: Calculate the Pearson correlation coefficient between temperature and load current using a sliding window (window length 60s) ,in, is the Pearson correlation coefficient between temperature and load current, for The temperature value at the moment, for The load current value at the moment, 、 They are the average values ​​of temperature and load current in the window, respectively. Under normal working conditions, , the correlation coefficient is too low ( ) indicates potential abnormalities;

[0287] Oil flow-load linkage characteristics:

[0288] Oil flow response speed: Calculate the ratio of the oil flow rate change rate to the load current change rate ,in, is the oil flow response speed coefficient, is the rate of change of oil velocity modulus, in m / s2, is the rate of change of load current, in A / s, under normal working conditions (Oil flow changes synchronously with load), when When the cooling system is delayed,

[0289] Oil flow-temperature coupling characteristics: calculated through covariance analysis ,in, is the covariance between temperature and oil velocity, is the temperature random variable, 、 are the mathematical expectations of temperature and oil flow rate, respectively, which are negatively correlated under normal operating conditions ( ), a positive correlation indicates cooling failure;

[0290] Multi-source association model training:

[0291] Based on historical normal operation data (≥1000 hours), the temperature-load-oil flow correlation model is trained using the random forest algorithm to output the predicted value of the temperature change rate under normal conditions. ,in, is the predicted value of the temperature change rate, in °C / s, for The load current value at the moment, for Oil flow velocity model at each moment, is the time variable, and the model fit is goodness of fit ;

[0292] Calculate residuals ,in, is the predicted residual of the temperature change rate, in °C / s, and the normal residual range is determined by the 3σ criterion ,If it exceeds the range, it will be marked as a potential exception;

[0293] Step S5-3: Dynamic calculation of hierarchical alarm triggering conditions:

[0294] Level 1 alarm condition parameter calibration:

[0295] Determination of empirical coefficients: Through inversion of 100 typical fault cases (such as slight short circuit of winding, partial failure of cooler), we can obtain ,in, is the empirical coefficient of the load current's effect on the temperature change rate, in units of , (load current nonlinear coefficient), ℃ / s (cooling delay coefficient);

[0296] Oil circulation time constant (in, (time constant of the oil circulation system, representing the delay in the response of the oil flow to temperature changes): calculated based on the transformer oil circulation system design parameters (oil pump flow rate, oil channel volume), and corrected through experimental verification (oil temperature stabilization time after a sudden load change), with an error of ≤5s;

[0297] Level 1 alarm logic: triggered when both the "temperature change rate overload associated threshold" and "oil flow response hysteresis" are met, i.e.: ,in, is the normalized load current, is the rate of change of oil flow velocity modulus, is the load current change rate;

[0298] Secondary shutdown condition parameter calculation:

[0299] Hotspot diffusion threshold : Dynamically calculated based on the following formula: , where the basic diffusion rate ,in, is the basic diffusion rate of the hotspot area, in c / s, calibrated by hot spot diffusion experiment, heat accumulation factor ,in, From the initial moment to The accumulated heat at time, , ℃ is the rated hot spot temperature;

[0300] Maximum temperature gradient threshold: Combined with the insulation material tolerance limit (paper insulation will experience accelerated local aging when the gradient is >8°C / cm), set ,in, is the maximum temperature gradient modulus in the temperature field;

[0301] Secondary shutdown logic: triggered when the "hotspot area diffusion rate exceeds the threshold" and the "maximum temperature gradient exceeds the limit", that is: ,in, is the rate of change of the hotspot area over time;

[0302] Dynamic threshold adaptive adjustment:

[0303] Ambient temperature compensation: When the ambient temperature ℃, the first level alarm threshold is reduced by 10% ( ), improve sensitivity in high temperature environment;

[0304] Equipment aging correction: For transformers with an operating age of more than 15 years, the secondary shutdown threshold Increased to 0.3cm² / s (taking into account the accelerated diffusion of hot spots after insulation aging);

[0305] Step S5-4: Alarm response mechanism and linkage control:

[0306] Level 1 alarm response:

[0307] Local alarm: The control cabinet indicator light flashes (yellow), the buzzer sounds an intermittent alarm (frequency 1Hz), and the human-machine interface displays abnormal parameters (temperature change rate, load current, oil flow rate real-time curve);

[0308] Data recording: Automatically store multi-source data (temperature field, current, oil flow rate) 300s before and after the alarm, and generate event logs (including trigger time, associated parameter values, residuals, etc.) );

[0309] Decision support: Preliminary diagnostic results (such as "suspected cooler efficiency decline" and "load-temperature rise nonlinearity") are delivered, with recommendations to check the cooling system valve status and oil pump operating current.

[0310] Secondary shutdown response:

[0311] Emergency control: The transformer protection device is linked via a relay output signal (dry contact, capacity 220V / 5A), triggering the low-voltage side circuit breaker to trip (operation time ≤ 50ms), thus cutting off the load current;

[0312] Remote notification: Push shutdown signals (including fault type, hotspot location coordinates, and maximum temperature gradient value) to the operation and maintenance center through the power dispatching data network (DL / T476), and send text messages to the responsible person at the same time;

[0313] Safety interlock: After tripping, the closing circuit is locked and requires manual reset (enter password + on-site confirmation) to prevent accidental restart;

[0314] False alarm suppression and self-checking:

[0315] Level 1 alarm delayed confirmation: 5s delay starts after triggering, during which the parameters are continuously monitored. If the value falls back to the normal range, the alarm will be automatically cancelled (false alarm rate is controlled at <0.1 times / month);

[0316] Secondary post-shutdown verification: After shutdown, re-measure the temperature of the hotspot area with an infrared thermal imager and compare it with the system monitoring value (if the deviation is ≤2°C, it is confirmed to be valid; otherwise, it is marked as a false alarm and the threshold is adjusted);

[0317] Step S5-5: Iterative optimization and verification of the early warning model:

[0318] Parameter iteration driven by historical data:

[0319] Review alarm events (including valid alarms and false alarms) every quarter and optimize using gradient descent method Equal empirical coefficients: ,in, For the After iterations value, For the After iterations value, is the learning rate, is the number of historical events, For the The temperature change rate of the event, For the The load current value of the secondary event, For the The load change duration of this event should ensure that the model has a detection rate of ≥99% for real faults.

[0320] Digital twin simulation verification:

[0321] Reproduce typical fault scenarios (such as inter-turn short circuits and oil channel blockages) on the transformer digital twin platform, input simulation data (temperature, current, and oil flow rate) into the early warning model, and verify the alarm trigger time (deviation from the theoretical fault occurrence time ≤ 2s) and classification accuracy (no skipping or missing levels).

[0322] Extreme operating condition test: Simulates a combined scenario of "sudden load increase + sudden cooler failure" to verify whether the secondary shutdown is triggered before the hotspot temperature reaches 140°C (the critical value for carbonization of insulation paper) (with a safety margin of ≥10°C).

[0323] Model version management:

[0324] After each parameter optimization, a new version of the model (marked with date + number of iterations) is generated, and historical versions (at least 3) are retained to support backtracking comparison;

[0325] Before a new model is launched, it must be verified using an offline test set (including 1,000+ normal samples and 100+ faulty samples) and can only be deployed if its accuracy is ≥ 98%;

[0326] Through the above steps, accurate identification (response time ≤ 1s) and graded handling of abnormal transformer temperature rise can be achieved, effectively shortening the fault detection time (70% reduction compared with traditional methods) and reducing the occurrence rate of major accidents (verification data shows that the fault interception rate is ≥ 95%).

[0327] In this embodiment, the fluorescent fiber optic sensor uses a europium ion-doped silica-zirconia composite fluorescent material, whose temperature sensitivity is a relative change of 0.85 percentage points in the decay time constant per degree Celsius, and the temperature measurement accuracy is better than plus or minus 0.5 degrees Celsius; and the surface of the optical fiber is coated with a nano-aluminum oxide insulating layer, the thickness of which is controlled within the range of 50 nanometers plus or minus 5 nanometers.

[0328] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A distributed temperature measurement method for oil-immersed transformers based on a fluorescent fiber optic sensor is characterized by: The following steps are involved: S1. Sensor Layout: Fluorescent fiber optic sensors are deployed in a partitioned and layered three-dimensional topology in key temperature-rise areas within the oil-immersed transformer, including between high-voltage and low-voltage winding layers, on the core surface, and in hot spots in the heat dissipation oil channel. The sensors consist of multiple sections of fluorescent optical fiber connected in series via electromagnetic interference-resistant fiber optic patch cords to form a distributed temperature measurement network. The length of each optical fiber section is customized based on the spatial characteristics of the monitoring area. S2. Excitation signal injection: A pulsed laser is used to inject a pulsed excitation light with a wavelength of 380 nm to 420 nm into the fluorescent fiber sensor to excite the fluorescent substance to generate a fluorescence afterglow signal. S3, signal acquisition and demodulation: real-time acquisition of fluorescence afterglow signals, use of dual-channel synchronous phase-locked amplification technology to separate environmental noise, and obtain temperature measurements by calculating the fluorescence decay time constant; S4, Temperature Field Reconstruction: Combining the sensor's three-dimensional spatial coordinates with real-time temperature data, the dynamic temperature field distribution inside the transformer is reconstructed based on a thermodynamic inversion algorithm, identifying and marking hot spots where the temperature gradient exceeds 5 degrees Celsius per centimeter; S5. Perform time-series correlation analysis on the temperature data, transformer load current, and cooling oil flow rate. When the temperature change rate exceeds the dynamic threshold and does not match the load change trend, activate the graded alarm mechanism.

2. The distributed temperature measurement method for oil-immersed transformers based on a fluorescent optical fiber sensor according to claim 1 is characterized in that: The sensor deployment in step S1 includes: The optical fiber is laid in a spiral winding pattern between winding layers. The outer side of the optical fiber is covered with an oil-resistant polyimide protective layer with a thickness of no more than 0.2 mm and fixed to the winding insulation support bar by micro ceramic clips. A dual-redundant cross-layout method is used in the hot spots of the heat dissipation oil channel, and two sections of optical fiber are cross-fixed at the inlet and outlet of the oil channel at an angle of 30 to 60 degrees.

3. The distributed temperature measurement method for oil-immersed transformers based on a fluorescent optical fiber sensor according to claim 1 is characterized in that: The signal demodulation in step S3 adopts a dual-weight adaptive attenuation model: First, the effective decay time constant is calculated, which is the product of the time constant of the fast fluorescence decay component and its real-time signal-to-noise ratio dynamic weight, plus the product of the time constant of the slow fluorescence decay component and its complementary weight, plus the oil flow disturbance compensation. The real-time signal-to-noise ratio dynamic weight is equal to the real-time signal-to-noise ratio value divided by the sum of the real-time signal-to-noise ratio and the environmental noise coefficient; The oil flow disturbance compensation is obtained by integration. The integral term includes the product of the transformer oil density and the square of the velocity gradient of the oil flow in the normal direction of the optical fiber, plus the product of the temperature-dependent dynamic viscosity and the square of the oil flow curl. The integration is performed along the length of the optical fiber microelement segment, and the result is multiplied by the optical fiber fluid sensitivity coefficient.

4. The distributed temperature measurement method for oil-immersed transformers based on a fluorescent optical fiber sensor according to claim 3 is characterized in that: The temperature calculation uses the oil flow coupling compensation function: The temperature value is equal to the fluorescent material calibration coefficient a multiplied by the square of the effective decay time constant, plus the coefficient b multiplied by the effective decay time constant, plus the constant term c, and finally the oil flow correction factor d multiplied by the logarithm of the oil flow velocity; The oil flow velocity is obtained by calculating the square root of the sum of the squares of the three-dimensional velocity components, and a small offset is added to prevent zero-speed singularities.

5. The distributed temperature measurement method for oil-immersed transformers based on fluorescent optical fiber sensors according to claim 1 is characterized in that: The three-dimensional thermodynamic inversion algorithm of step S4 includes: Establish a coupled heat conduction and convection equation, which includes the product of the transformer oil heat capacity and the temperature-time derivative equal to the divergence of the thermal conductivity and the second-order spatial derivative of temperature, plus the winding heat source power, minus the dot product of the oil flow velocity and the temperature gradient; The regularized least squares inversion method is adopted. The objective function is the sum of the squares of the deviations between the measured temperature values ​​of all sensors and the calculated values, plus a smoothness constraint term of the global integration of the squares of the second-order spatial derivatives of the temperature field. The temperature field distribution is solved by minimizing the objective function.

6. The distributed temperature measurement method for oil-immersed transformers based on fluorescent optical fiber sensors according to claim 1 is characterized in that: The hierarchical alarm triggering conditions of step S5 are: The triggering condition for the first level alarm is that the temperature time change rate is greater than the sum of the load current power function term and the cooling state exponential decay term, and the oil flow growth rate lags behind the load current growth rate; The triggering conditions for the second-level shutdown protection are that the time rate of change of the hotspot area is greater than the dynamic diffusion threshold, and the maximum temperature gradient exceeds 8 degrees Celsius per centimeter; where the dynamic diffusion threshold is a linear function of the base diffusion rate times the load current change from the rated value, plus a heat accumulation factor.

7. The distributed temperature measurement method for oil-immersed transformers based on a fluorescent optical fiber sensor according to claim 1, characterized in that: The fluorescent fiber sensor uses a europium ion-doped silica-zirconia composite fluorescent material, with a temperature sensitivity of 0.85 percent relative change in decay time constant per degree Celsius, and a temperature measurement accuracy better than plus or minus 0.5 degrees Celsius; The surface of the optical fiber is coated with a nano-aluminum oxide insulation layer, and the thickness is controlled within the range of 50 nanometers plus or minus 5 nanometers.

8. The distributed temperature measurement method for oil-immersed transformers based on fluorescent optical fiber sensors according to claim 5 is characterized in that: The winding heat source power is dynamically updated through the electro-thermal coupling model: The heat source power is equal to the product of the winding resistivity and the square of the current density, plus the product of the core loss coefficient and the time derivative of the magnetic flux density; Among them, the winding resistivity changes inversely with the increase in temperature, and the resistivity decreases by 0.4 percentage for every 1 degree Celsius increase in temperature.

9. The distributed temperature measurement method for oil-immersed transformers based on fluorescent optical fiber sensors according to claim 6, characterized in that: The heat accumulation factor is calculated using a recursive algorithm with a forgetting factor: The current moment heat accumulation factor is equal to the previous moment value multiplied by the exponential decay factor, plus the current temperature deviation value multiplied by the heat accumulation coefficient; The exponential decay factor is determined by the thermal decay coefficient and the sampling time interval, and the heat accumulation coefficient is related to the thermal decay coefficient.

Images