Method and system for collecting and processing internal temperature of a metallized film capacitor

Abstract

The application provides a kind of collection and processing method and system of metallized film capacitor internal temperature, including the construction of metallized film capacitor internal temperature basic expression based on the principle of fiber grating light response;The construction and decomposition model of error transmission mechanism of disturbed variable under multi-field coupling;Multi-factor high-order nonlinear compensation model of adaptive kernel mapping;Improved least squares global optimization objective function by introducing regularization term;Optimal solution of overdetermined equation set and model dynamic calibration based on conjugate gradient method;By structural adaptation design and algorithm optimization combination and the construction of temperature signal processing system, realize the precise, anti-interference, real-time online detection of metallized film capacitor internal temperature through fiber grating core rod groove embedded type layout, double-channel wavelength demodulation and multi-factor high-order compensation.

Description

Technical Field

[0001] This invention relates to electrical engineering and optical engineering technologies, and more particularly to a method and system for acquiring and processing the internal temperature of a metallized thin-film capacitor. Background Technology

[0002] Metallized film capacitors are core energy storage devices in pulsed power systems, electromagnetic emission, laser drives, industrial pulse power supplies, and other equipment. They operate under conditions of high frequency, high current, rapid charging and discharging, high voltage, and strong electromagnetic interference. As the internal core insulation and support structure, the capacitor core rapidly generates heat during operation due to dielectric and metal losses. This heat accumulation leads to localized hot spots inside the capacitor, whose temperature is significantly higher than that of the outer shell surface.

[0003] Abnormally high temperatures can directly accelerate insulation aging, cause dielectric breakdown, reduce capacitance, and shorten lifespan. In severe cases, they can lead to thermal runaway and explosions, threatening the safety of the entire system. Existing temperature monitoring methods cannot meet the high-precision, interference-resistant, and long-term stable temperature measurement requirements of metallized film capacitors.

[0004] Existing temperature monitoring methods have the following main drawbacks: Thermocouples and infrared thermometry can only monitor the temperature of the outer casing and cannot obtain the actual hot spot temperature inside the capacitor. Traditional electrical sensors suffer from severe interference, low accuracy, and poor insulation in strong electromagnetic and high voltage environments. Conventional fiber Bragg grating temperature measurement does not take into account the thermal coupling error of the core rod groove embedding structure, pulse transient temperature rise, and demodulation drift, resulting in systematic biases in the measurement results. The lack of a dedicated temperature compensation model and demodulation correction algorithm for metallized film capacitors makes it difficult to achieve long-term stable and high-precision monitoring. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the prior art by providing a method and system for acquiring and processing the internal temperature of a metallized film capacitor, thereby enabling real-time, accurate, and reliable monitoring of the capacitor's internal temperature and ensuring the safe operation of the metallized film capacitor.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a method for acquiring and processing the internal temperature of a metallized thin-film capacitor, comprising the following steps: S1. Based on the principle of fiber optic grating photosensitive response, construct an expression for the internal temperature of a metallized thin-film capacitor; S2. Construction and decomposition of error propagation equations for disturbed variables under multi-field coupling; S3. Construction of an adaptive weighted multi-factor high-order nonlinear compensation model; S4. An improved least-squares global optimization objective function with the introduction of a regularization term; S5. Optimal solution of overdetermined equations based on the conjugate gradient method and dynamic calibration of an adaptive weighted multi-factor high-order nonlinear compensation model.

[0007] Furthermore, a system for acquiring and processing the internal temperature of a metallized film capacitor, used to implement the aforementioned method for acquiring and processing the internal temperature of a metallized film capacitor, further includes: a capacitor insulating core rod, the outer surface of which is covered with an internal metallized film of the capacitor, the outer surface of which is fitted with a metallized film capacitor shell, and stainless steel electrode rods connected to both ends of the capacitor insulating core rod, the stainless steel electrode rods at both ends being connected to a pulse charging and discharging power supply via wires. The capacitor insulating core rod has a groove along its axial direction. A fiber optic grating sensor is embedded in the groove. The fiber optic grating sensor is connected to a fiber optic grating demodulator via a transmission fiber. The fiber optic grating demodulator transmits the data to a host computer control system.

[0008] The beneficial effects of this invention are as follows: by using fiber grating core rod groove embedded layout, dual-channel wavelength demodulation and multi-factor high-order compensation, and by combining structural adaptation design and algorithm optimization, the internal temperature of the core rod can be accurately detected, anti-interference and real-time online. An axial groove is machined in the central area of ​​the capacitor insulating core rod. The groove size is precisely matched with the fiber optic grating sensor. This ensures both accurate matching of the fiber optic grating sensor and close contact between the fiber optic grating sensor and the heating core of the capacitor insulating core rod, without compromising the mechanical strength and insulation performance of the capacitor insulating core rod. High-temperature resistant insulating thermally conductive adhesive is used for filling and curing, so that the fiber optic grating sensor and the capacitor insulating core rod form a tight thermal coupling without gaps. This ensures that the temperature of the capacitor insulating core rod can be quickly and accurately transferred to the sensing area, and the internal hot spot temperature can be directly collected from the structure. A dual-channel fiber grating wavelength demodulation architecture was constructed. The probe light output from the demodulation system was fed into an embedded fiber grating to obtain the wavelength signals of the temperature-sensitive grating and the weakly temperature-sensitive reference grating, respectively. The temperature-sensitive grating signal was used as the detection signal, while the reference grating signal avoided the temperature-sensitive range of the core rod and shielded against cross-interference from strain and electromagnetic fields. The center wavelength signals of the two gratings were synchronously output after photoelectric conversion and peak identification, providing a basis for subsequent noise reduction and compensation. Turn on the fiber optic grating demodulator and the host computer control system, perform initial calibration of the demodulation system, calibrate basic parameters such as the grating center wavelength and temperature sensitivity coefficient at the reference temperature, establish a dual-channel signal acquisition link, and realize real-time and continuous acquisition of the core rod temperature optical signal and electrical signal conversion. The converted dual-channel wavelength signal is processed by difference to eliminate common-mode interference such as incident light intensity fluctuation, optical path scattering, and system baseline drift. Based on the fiber grating wavelength-temperature linear response principle and combined with the structural characteristics of the core rod groove, the basic calculation expression of the core rod temperature is derived and established. The fixed parameters of the system are combined to form a simplified temperature-variable correlation formula. To address the multi-source factors affecting measurement accuracy, such as ambient temperature fluctuations, loop pressure changes, circuit noise drift, and transient interference from pulse charging and discharging, the disturbed combination term is set as the variable to be optimized. A high-order polynomial compensation model with temperature, pressure, and circuit noise as independent variables is established, transforming the variable optimization problem into a surface fitting problem. The real-time collected disturbance signal is substituted into the calibrated high-order compensation model to achieve unified dynamic correction of multi-source interference, complete the accurate compensation of the variable to be optimized, and finally substitute the optimized variable into the temperature calculation formula to calculate the true temperature value at the groove of the metallized film capacitor core. The host computer simultaneously realizes temperature display, data storage and over-temperature early warning linkage. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of a system for acquiring and processing the internal temperature of a metallized thin-film capacitor. Figure 2(a) is a comparison chart of internal temperature monitoring and compensation in metallized thin-film capacitors; Figure 2(b) is a graph showing the measurement error analysis. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0011] A method for acquiring and processing the internal temperature of a metallized thin-film capacitor includes the following steps: S1. Based on the photosensitive response principle of fiber Bragg grating, construct the temperature expression for metallized thin film capacitors; S2. Construction and decomposition of error propagation equations for disturbed variables under multi-field coupling; S3. Construction of an adaptive weighted multi-factor high-order nonlinear compensation model; S4. An improved least-squares global optimization objective function with the introduction of a regularization term; S5. Optimal solution of overdetermined equations based on the conjugate gradient method and dynamic calibration of an adaptive weighted multi-factor high-order nonlinear compensation model.

[0012] S1 includes: S101. Establish the constitutive equation for dual-channel wavelength output; The dual-channel fiber optic demodulation mode is adopted. After the continuous broadband light emitted by the light source is transmitted into the fiber optic grating embedded in the groove of the core rod, two adjacent wavelength signals, namely the temperature-sensitive detection grating and the weak temperature-sensitive reference grating, are output respectively. The temperature-sensitive detection grating corresponds to the temperature-sensitive region of the core rod; the weakly temperature-sensitive reference grating avoids the temperature-sensitive region of the core rod and simultaneously shields against cross-interference from the pulse electric field and the micro-strain of the core rod; after conversion by a photodetector and extraction by a centroid peak identification algorithm, the two beams output the corresponding center wavelength signal, and the constitutive equation is: (1); (2); in, for Real-time center wavelength of the temperature-sensitive detection grating, in pm; This refers to the internal temperature of a metallized film capacitor. for Real-time center wavelength of the temperature-sensitive reference grating, in pm; , Reference temperature The initial center wavelengths of the lower temperature-sensitive detection grating and the weakly temperature-sensitive reference grating, in pm; , The temperature sensitivity coefficients of the temperature-sensitive detection grating and the weakly temperature-sensitive reference grating are respectively satisfied. Unit: pm / ℃; for The internal temperature of a metallized film capacitor at any given time, in °C; The thermal hysteresis correction term for the mandrel groove is determined by the multilayer thermal conductivity characteristics of the groove, adhesive layer, and grating. This is a non-temperature interference correction term for the reference channel; , These are the Gaussian white noise terms of the two signals, respectively, and follow the rules of... distributed; S102, Differential decoupling processing for common-mode interference suppression; By performing a difference operation on the two wavelength signals to eliminate common-mode interference, the expression for the wavelength difference is: (3); Let the common-mode noise residual term Structure and interference coupling terms Then formula (3) simplifies to: (4); The common-mode interference includes incident light intensity fluctuation, optical path scattering loss, and system baseline drift.

[0013] S103. Constructing the expression for the internal temperature of a metallized thin-film capacitor: From formula (4), the analytical expression for the internal temperature of a metallized thin-film capacitor is: (5); S104. Construct the standard temperature and variable correlation formula; Let the structure constant Baseline constant Substituting into formula (5), the standard temperature relationship is: (6); in, As a combination of constants after the structure and grating parameters are determined, it only needs to be calibrated once during the initial calibration.

[0014] S2 includes: S201, Decomposition of perturbed variables under multi-field coupling; Let the core disturbed combination variable be: (7); The temperature formula then simplifies to the core mapping relationship: (8); Among them, the internal temperature of metallized film capacitors Only with the perturbed combination variable Linear correlation, therefore the accuracy of temperature measurement depends entirely on the linear correlation. The effect of suppressing and compensating for multi-source interference.

[0015] In this process, the physical properties of each parameter in the standardized temperature expression are analyzed to distinguish between constants and dynamically disturbed variables: the core rod groove size, fiber grating model, and demodulation optical path structure are fixed, thus combining constants. The measured values ​​are constant values ​​that are invariant over time; however, the real-time wavelength difference, structure, interference coupling terms, and noise residual terms are all affected by environmental temperature fluctuations, system loop pressure changes, circuit baseline drift, transient electromagnetic interference from pulse re-discharge, and multi-source disturbances from the thermal coupling micro-strain of the mandrel, causing the measured values ​​to deviate from the true values.

[0016] S202, Establishment of the error propagation equation; Based on the operating conditions of metallized film capacitors, the disturbed combined variables measured value Decompose into true values By superimposing the multi-field coupling error terms, an error propagation equation is constructed: (9); in, This is the environmental temperature field error term, caused by fluctuations in the environmental temperature field. This is an electrical error term, caused by baseline drift in the demodulation circuit and photoelectric conversion noise; This is the pulse transient error term, caused by strong electromagnetic interference during pulse charging and discharging, and circuit pressure fluctuations. This is a thermal coupling error term, caused by the thermal conduction hysteresis and temperature gradient distribution of the multi-layered groove structure. The random measurement error term follows a zero-mean Gaussian distribution; S203. Construction of multi-field coupling error terms: Physical analysis is performed on each deterministic error term to provide a theoretical basis for subsequent compensation.

[0017] S2031, Ambient Temperature Field Error Term The non-uniform thermal expansion of the grating caused by the temperature difference between the ambient temperature and the core rod temperature is expressed as: (10); in, The equivalent thermal expansion coefficient of the optical fiber packaging structure; Effective elastic-optic coefficient of optical fiber; The influence coefficient of the rate of change of ambient temperature; for The ambient temperature at that moment; S2032, Thermal Coupling Error Term The thermal hysteresis caused by the multilayer thermal resistance of the groove, thermally conductive adhesive, and grating is addressed by constructing a one-dimensional unsteady-state heat conduction equation based on Fourier's law of heat conduction. (11); in, Density of thermally conductive adhesive; Specific heat capacity at constant pressure; The radial distance from the center of the mandrel to the grating; The volumetric heat generation rate of the core rod; The thermal coupling error term obtained by solving the thermal conduction hysteresis correction using the method of separation of variables is: (12); in, To test the sensitivity of the grating; Temperature sensing for the grating; The thermal conduction time constant; These are the coefficients of the Fourier series expansion; It is the order; This is the length of the heat conduction path in the groove; S2033, Pulse Transient Error Term The thermal shock and strong electromagnetic interference caused by the transient high current of pulse charging and discharging are expressed as follows: (13); in, for The charging and discharging current of the metallized film capacitor at all times; Joule thermal shock coefficient; The electromagnetic interference coefficient is the rate of change of current. S2034, Electrical Error Items Caused by wavelength drift and circuit temperature drift in the demodulation system, based on the FFP-TF temperature drift model, the expression is: (14); in, This refers to the operating temperature of the demodulation circuit. The circuit temperature drift fitting coefficient is denoted as .

[0018] Specifically, S3 is: S301, Construction of an adaptive kernel mapping high-order nonlinear compensation model; To address the multi-source deterministic error after decomposition, an adaptive kernel mapping high-order nonlinear compensation model is constructed, with ambient temperature, circuit temperature, pulse current, and heat conduction time constant as independent variables, and the compensation value of the disturbed variable as the output. The expression is as follows: (15); in, This is the actual measured amount of disturbance received; The predicted compensation value for multi-source errors is obtained by solving an adaptive weighted multi-factor high-order nonlinear compensation model, and the expression is: (16); in, Let each be the polynomial order of the input variable, and take... A third-order four-factor fully coupled polynomial is constructed to balance fitting accuracy and generalization ability. Let be the fourth-order polynomial fitting coefficient tensor, and be the parameters to be optimized. This is a real-time statistic of circuit noise, taking the signal standard deviation within the sliding window; This is the noise compensation coefficient; The number of support vectors; For Lagrange multipliers; Here, the radial basis function (RBF) kernel is used to capture the nonlinear mapping relationship of the error. The expression for the RBF kernel is: (17); in, for The input feature vector at time step; For the first Feature vectors of training samples; This is the kernel function width parameter; This refers to the operating temperature of the demodulation circuit. Introducing adaptive weight coefficients The dynamic weighting of the pulse charge-discharge transient process is expressed as follows: (18); in, It serves as a weighting factor for the instantaneous state, ensuring compensation accuracy under pulse transient conditions; S302, Matrix representation of adaptive kernel mapping high-order nonlinear compensation model; Let the total number of training samples be The input feature vector for each sample is: ; The corresponding measured perturbation variable matrix is The true value matrix is Then the matrix form is: (19); in, for A vector of error prediction values ​​for each sample; for The polynomial characteristic matrix; The number of terms in a third-order four-factor polynomial. ; for The polynomial coefficient tensor expansion vector; for The kernel function matrix, with elements of ; These are Lagrange multiplier vectors; This is a vector of noise statistics.

[0019] Specifically, S4 is: S401. Construct a global optimization objective function based on minimizing structural risk; To quantify the deviation between the compensated variables and the true variables, and to avoid overfitting, an improved least squares global optimization objective function with a regularization term is constructed based on the structural risk minimization criterion. This function measures the compensation accuracy and generalization ability. The expression for the global optimization objective function is as follows: (20); in, It is the Euclidean norm; A vector of true values; This is a vector of measured values; This is a penalty factor used to balance fitting accuracy and complexity; For smoothing coefficients; It is a second-order difference smoothing matrix; The trace of the matrix; The first term is the empirical risk term, which is the sum of squares of the error and the sum of squares; the second term is the L2 regularization term, which is used to suppress overfitting; and the third term is the smoothing regularization term, which is used to ensure the smoothness of the polynomial fitting.

[0020] The physical meaning of the global optimization objective function is to limit the amplitude and fluctuation of parameters while ensuring the minimum fitting error, thereby improving the generalization ability and long-term stability under pulse conditions.

[0021] S402. The optimal solution requires the KKT conditions (Karl-Kun-T conditions). Global optimization objective function To achieve the global minimum, the KKT conditions must be satisfied, meaning the partial derivatives of the global objective function with respect to all parameters to be optimized are zero. The optimal condition expression is: (twenty one); Wherein, let the error residual vector Substituting the KKT conditions, the optimal solution problem is transformed into a problem of solving a system of linear overdetermined equations; S403. Introduce constraints on the objective function: To ensure the effectiveness of compensation across the entire operating range of metallized film capacitors, inequality constraints are introduced to limit the maximum compensation error: (twenty two); in, To maximize the allowable compensation error, it is set to ±0.5 pm based on the temperature measurement accuracy requirements, corresponding to a temperature error of ±0.1℃.

[0022] Specifically, S5 is: S501, Construction of the overdetermined system of equations; Expanding the partial derivative equations in the KKT conditions and combining like terms, we construct a system of linear overdetermined equations in matrix form: (twenty three); Among them, the augmented coefficient matrix for The expression for a symmetric positive definite matrix is: (twenty four); in, They are respectively Rank, An identity matrix of order 1; The parameter vector to be solved It contains all the model parameters to be optimized; Right-hand constant term vector ; S502, Optimal solution based on the improved conjugate gradient method; S503. Hyperparameter optimization based on the improved beetle whisker particle swarm algorithm; An improved beetle whisker particle swarm optimization algorithm is used for global optimization, with the mean square error of the test set as the fitness function: (25); in, This represents the number of samples in the test set. The beetle whisker particle swarm optimization algorithm introduces adaptive inertia weights and a dynamic learning factor to ensure that the algorithm converges quickly to the globally optimal hyperparameter combination and avoids getting trapped in local optima. The expression for the adaptive inertia weights is as follows: (26); in, This represents the maximum value of the inertia weight; This represents the minimum value of the inertia weight. This represents the current iteration number; The total number of iterations; a constant. ; S504, Dynamic Calibration and Temperature Calculation; The optimal hyperparameters and the coefficients of the solved adaptive weighted multi-factor high-order nonlinear compensation model are substituted into the adaptive weighted multi-factor high-order nonlinear compensation model to complete the calibration of the adaptive weighted multi-factor high-order nonlinear compensation model. During real-time monitoring, the collected input feature vectors are substituted into the calibrated adaptive weighted multi-factor high-order nonlinear compensation model to obtain the optimized variables after real-time compensation. Finally, substituting the values ​​into the core temperature mapping formula, the true internal temperature of the metallized film capacitor is calculated; the true internal temperature of the metallized film capacitor is: (27); The calculated temperature value is uploaded to the host computer simultaneously, enabling real-time temperature display, data storage, and linkage with over-temperature warning.

[0023] Specifically, S502 is as follows: S5021. Initialization: Set the initial value for iteration. Initial residual Preprocessing matrix Solve for the initial search direction Set the iterative convergence threshold Maximum number of iterations ; S5022, No. Next iteration calculation: ; ; ; in, For the first The step size of the next iteration. For the first The residual vector of the next iteration.

[0024] S5023. Convergence Judgment: If... or Stop iterating and output the optimal parameter vector. Otherwise, continue iterating. ; ; S5034. After the iteration is completed, from the optimal parameter vector... The polynomial coefficient tensor is obtained by decomposition. Lagrange multipliers Noise compensation coefficient The parameters of the multi-factor high-order nonlinear compensation model with adaptive kernel mapping are solved.

[0025] Please see Figure 1 A system for acquiring and processing the internal temperature of a metallized film capacitor, used to implement the aforementioned method for acquiring and processing the internal temperature of a metallized film capacitor, further comprising: a capacitor insulating core rod 2, the outer surface of which is covered with an internal metallized film 3, the outer surface of which is fitted with a metallized film capacitor shell 1, and stainless steel electrode rods 4 connected to both ends of the capacitor insulating core rod 2, the stainless steel electrode rods 4 at both ends being connected to a pulse charging and discharging power supply via wires. The capacitor insulating core rod 2 has a groove 5 along its axial direction. A fiber optic grating sensor 6 is embedded in the groove. The fiber optic grating sensor 6 is connected to a fiber optic grating demodulator 9 via a transmission fiber 8. The fiber optic grating demodulator 9 transmits data to the host computer control system 11.

[0026] The groove 5 is fixed to the fiber optic grating sensor 6 by high-temperature resistant insulating thermally conductive adhesive.

[0027] The metallized film capacitor housing 1 serves to seal, protect, and support the internal structure; The capacitor insulating core rod 2 serves as the internal insulation and core support component of the capacitor and is the main area for heat generation. The metallized thin film 3 inside the capacitor serves as a core component and acts as a dielectric for energy storage. Stainless steel electrode rod 4 serves as the high-voltage input and current output electrode of the metallized thin-film capacitor, enabling the transmission of electrical energy during charging and discharging. The groove 5 is a pre-set axial groove in the capacitor insulating core rod 2, which is opened in the central area of ​​the core rod for embedding fiber optic grating sensors. The fiber optic grating sensor 6 is embedded in the middle of the groove 5 to realize the acquisition of internal temperature signals; Fiber 8 is used to transmit the light signal reflected by the grating and connect the internal sensing and the external demodulation module. Fiber Bragg grating demodulator 9 is used to acquire and analyze the reflected wavelength signal of fiber Bragg gratings; The pulse charge-discharge power supply 10 provides charge-discharge excitation for the metallized film capacitor, simulating actual working conditions. The host computer control system 11 performs data calculations, compensation optimization, temperature display, and over-temperature warning.

[0028] Example 1: To verify the measurement accuracy and anti-interference capability of a dual-channel temperature monitoring system for fiber Bragg gratings inside a metallized thin-film capacitor, this study models, simulates, and performs high-order nonlinear compensation for temperature calculation errors under multi-source interference, based on a method for processing the temperature signals acquired by the fiber Bragg grating. By comparing the measurement errors before and after compensation and verifying the effectiveness of adaptive kernel mapping and regularization optimization algorithms, this study provides theoretical support and data verification for practical applications.

[0029] The simulation platform is MATLAB R2020b or later, with no additional toolbox dependencies, using only MATLAB built-in functions to ensure that the simulation results are reproducible. Simulation parameters are shown in Table 1; Table 1 Simulation Parameters

[0030] The simulation conditions strictly simulate the actual working environment of the metallized film capacitor. Multi-source interference is modeled according to physical constitutive relations, as follows: the ambient temperature is 25±5 ℃, exhibiting sinusoidal fluctuations (frequency 0.5 Hz), simulating ambient temperature change interference; the circuit temperature is 30±2 ℃, containing 0.5 times Gaussian white noise, simulating temperature drift and random drift of the demodulation circuit; the pulse current is 1000 A peak value, 0.05 s pulse width, and 0.5 s period, simulating transient impact conditions. The types of interference cover environmental field errors, mandrel thermal conduction hysteresis errors, pulse Joule thermal / electromagnetic interference, demodulation circuit temperature drift nonlinearity errors, and Gaussian white noise (amplitude 0.2), comprehensively covering the sources of interference in actual engineering.

[0031] A dual-channel fiber grating demodulation mode is employed, with the detection grating and reference grating outputting wavelength signals respectively. The constitutive equation is: ; in, This is a thermal hysteresis correction term for the mandrel groove structure; This is a non-temperature interference correction term for the reference channel; , The two signals are Gaussian white noise; By performing difference calculation on the two wavelength signals to eliminate common-mode interference from incident light intensity fluctuations and optical path losses, we obtain: ; in, For structural and interference coupling terms; This is the common-mode noise residual term; The basic formula for mandrel temperature is: ; Define the disturbed combination variable measured value Decompose into true values By superimposing the multi-field coupling error terms, an error propagation equation is constructed: ; The environmental temperature field error term is given by the following formula: ; The thermal coupling hysteresis error term is given by the following formula: ; The transient error term is given by the following formula: ; The electrical error term is given by the following formula: ; The random measurement error term follows a zero-mean Gaussian distribution with an amplitude of 0.2.

[0032] A joint compensation scheme employing polynomial features, RBF, kernel mapping, and noise statistics is used. The formula for predicting the compensation amount is as follows: ; It is a multinomial feature matrix, constructed from input features (ambient temperature, circuit temperature, pulse current, ambient temperature change rate), with a total of 10 features; The kernel function matrix is ​​RBF, and the kernel width is... Number of support vectors ; For noise statistics, the standard deviation is calculated using a sliding window (window size 10). These are the polynomial coefficients, RBF kernel Lagrange multipliers, and noise compensation coefficients, respectively, and are the optimization parameters.

[0033] The optimized variables after compensation are: ; To avoid overfitting and improve algorithm stability, a least-squares objective function with L2 regularization and smoothing regularization is constructed: ; in, It is the L2 penalty factor; For smoothing coefficients; It is a second-order difference smoothing matrix; The objective function is transformed into an overdetermined system of equations. The solution is obtained using the preprocessed conjugate gradient method, with a convergence threshold. Maximum number of iterations This ensures both the accuracy and real-time performance of the solution.

[0034] The final formula for calculating the compensated temperature is: .

[0035] The specific operating steps are as follows: System initialization: Set core parameters such as raster parameters, sampling rate, and time series, and fix the random seed (to ensure the results are reproducible); Operating condition signal generation: Generate operating condition signals such as the actual temperature of the core rod, ambient temperature, pulse current, and demodulation circuit temperature; Multi-source interference modeling: Based on physical constitutive relations, various errors such as environmental field, thermal coupling, pulse transient, electrical, and random noise are generated; Disturbance signal synthesis: Various errors are superimposed with the true value of the disturbed variable to obtain the measured disturbed variable Xm(t); Feature and kernel matrix construction: Construct the polynomial feature matrix Φ, the RBF kernel function matrix K, and the noise statistics vector R; Construction of the overdetermined system of equations: Based on the regularized objective function, construct the overdetermined system of equations A·U=B; Optimal parameter solution: The optimal compensation parameters M, α, and θ are obtained by iterative solution using the PCG algorithm. Temperature calculation and compensation: Calculate the error compensation amount, calculate the compensated temperature Tcomp(t), and at the same time calculate the uncompensated temperature Tuncompensated(t) (for comparison). Results output and visualization: Plot temperature comparison curves and error comparison curves, calculate core performance indicators, and output simulation results.

[0036] In Figure 2(a), the actual temperature of the mandrel shows a steady and gradual upward trend (45~60 ℃), without any interference or noise, serving as a reference benchmark for temperature measurement accuracy. The uncompensated measured temperature is affected by multiple sources of interference such as pulse transient interference and ambient temperature drift, resulting in violent fluctuations and a serious deviation from the actual temperature, failing to meet the temperature measurement requirements. After high-order compensation, the temperature highly overlaps with the actual temperature, with minimal fluctuations, enabling precise tracking of the actual temperature change trend.

[0037] In Figure 2(b), the compensation error fluctuates greatly, with the maximum error approaching 10 °C. The error shows obvious spikes during pulse impact, which completely fails to meet the engineering temperature measurement accuracy requirements. After high-order compensation, the error waveform is stable, the error is significantly suppressed, the pulse spike interference is basically eliminated, and the error range is controlled within a very small range, meeting the high-precision temperature measurement requirements.

[0038] To quantitatively evaluate the performance of the compensation algorithm, the core temperature measurement indicators before and after compensation were calculated, as shown in Table 2. Table 2 Core Performance Indicator Data

[0039] The error suppression rate is calculated as (uncompensated RMSE - compensated RMSE) / uncompensated RMSE × 100%. Therefore, it can be seen that the uncompensated algorithm is affected by multi-source interference, resulting in severe temperature measurement distortion. The high-order compensation algorithm can effectively suppress various interferences such as environmental temperature drift, pulse transients, and circuit temperature drift, with an error suppression rate of 92.3%, significantly improving anti-interference capability. After compensation, the RMSE is reduced to 0.32 ℃, and the maximum absolute error is reduced to 0.85 ℃, meeting the engineering requirements for high-precision monitoring of the internal temperature of metallized film capacitors. The PCG algorithm converges within 86 iterations, with fast convergence speed and high computational efficiency, making it suitable for engineering deployment and embedded implementation. The simulation results are consistent with the theoretical expectations of the patented technical solution, verifying the accuracy of the mathematical model and algorithm logic.

[0040] The embodiments described above are merely illustrative of implementation methods of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be defined by the appended claims.

Claims

1. A method for acquiring and processing the internal temperature of a metalized film capacitor, characterized in that, Includes the following steps: S1. Based on the photosensitive response principle of fiber Bragg grating, construct an expression for the internal temperature of a metallized thin-film capacitor; S2. Construction and decomposition of error propagation equations for disturbed variables under multi-field coupling; S3. Construction of an adaptive weighted multi-factor high-order nonlinear compensation model; S4. An improved least-squares global optimization objective function with the introduction of a regularization term; S5. Optimal solution of overdetermined equations based on the conjugate gradient method and dynamic calibration of an adaptive weighted multi-factor high-order nonlinear compensation model.

2. The method for collecting and processing the internal temperature of a metalized film capacitor according to claim 1, characterized in that, S1 includes: S101. Establish the constitutive equation for dual-channel wavelength output; The dual-channel fiber optic demodulation mode is adopted. After the continuous broadband light emitted by the light source is transmitted into the fiber optic grating embedded in the groove of the core rod, two adjacent wavelength signals, namely the temperature-sensitive detection grating and the weak temperature-sensitive reference grating, are output respectively. The temperature-sensitive detection grating corresponds to the temperature-sensitive region of the core rod; the weakly temperature-sensitive reference grating avoids the temperature-sensitive region of the core rod and simultaneously shields against cross-interference from the pulse electric field and the micro-strain of the core rod; after conversion by a photodetector and extraction by a centroid peak identification algorithm, the two beams output the corresponding center wavelength signal, and the constitutive equation is: （1）； （2）； wherein, is the real-time center wavelength of the temperature-sensitive detection grating at the moment, in pm; is the internal temperature of the metalized film capacitor; for the real-time center wavelength of the weak temperature-sensitive reference grating, in pm; , respectively the reference temperature the initial center wavelength of the temperature-sensitive detection grating and the weak temperature-sensitive reference grating, respectively, in pm; , The temperature sensitivity coefficients of the temperature-sensitive detection grating and the weakly temperature-sensitive reference grating are respectively satisfied. Unit: pm / ℃; For Instantaneous temperature inside the metallized film capacitor, in °C; The heat conduction hysteresis correction term for the core rod groove is determined by the multi-layer heat conduction characteristics of the groove, the glue layer and the grating; Tref is the reference channel non-temperature disturbance correction term; , are two independent Gaussian white noise terms with distribution . S102, Differential decoupling processing for common-mode interference suppression; By performing a difference operation on the two wavelength signals to eliminate common-mode interference, the expression for the wavelength difference is: （3）； Let the common-mode noise residual term Structure, interference coupling terms Then formula (3) simplifies to: （4）； S103. Constructing the expression for the internal temperature of a metallized thin-film capacitor: From formula (4), the expression for the internal temperature of a metallized film capacitor is: （5）； S104. Construct the standard temperature and variable correlation formula; Let the structural constant , the baseline constant Substitute equation (5) into the standard temperature relationship equation, and the result is: （6）； wherein, is a combination constant determined after the structure, grating parameters, and only needs to be calibrated once at the initial calibration.

3. The method for collecting and processing the internal temperature of a metalized film capacitor according to claim 2, wherein, S2 includes: S201, Decomposition of perturbed variables under multi-field coupling; Let the core disturbed combination variable be: （7）； The temperature formula then simplifies to the core mapping relationship: （8）； wherein the internal temperature of the metallized film capacitor only with the disturbed combination variable linear correlation; S202, Establishment of the error propagation equation; Based on the operating conditions of metallized film capacitors, the disturbed combined variables measured value Decompose into true values By superimposing the multi-field coupling error terms, an error propagation equation is constructed: （9）； in, This is the error term for the ambient temperature field; is the electrical error term; is the pulse transient error term; is the thermal coupling error term; The random measurement error term is subject to a zero-mean Gaussian distribution. S203, Construction of multi-field coupling error terms; S2031, environmental temperature field error term : Caused by the non-uniform thermal expansion of the grating due to the temperature difference between the ambient temperature and the mandrel temperature, expressed as: （10）； in, The equivalent thermal expansion coefficient of the optical fiber packaging structure; Effective elastic-optic coefficient of optical fiber; The influence coefficient of the rate of change of ambient temperature; for The ambient temperature at that moment; S2032, Thermal Coupling Error Term The thermal hysteresis caused by the multilayer thermal resistance of the groove, thermally conductive adhesive, and grating is addressed by constructing a one-dimensional unsteady-state heat conduction equation based on Fourier's law of heat conduction. （11）； in, Density of thermally conductive adhesive; Specific heat capacity at constant pressure; The radial distance from the center of the mandrel to the grating; The volumetric heat generation rate of the core rod; The thermal coupling error term obtained by solving the thermal conduction hysteresis correction using the method of separation of variables is: （12）； in, To test the sensitivity of the grating; Temperature sensing for the grating; The thermal conduction time constant; These are the coefficients of the Fourier series expansion; It is the order; This is the length of the heat conduction path in the groove; S2033, pulse transient error term ; the skin effect thermal shock and strong electromagnetic interference caused by the transient large current of pulse charging and discharging, expressed as: （13）； wherein, is the charge and discharge current of the metalized film capacitor; is the joule heat shock coefficient; is the electromagnetic interference coefficient of the current change rate; S2034, Electrical Error Items Caused by wavelength drift and circuit temperature drift in the demodulation system, based on the FFP-TF temperature drift model, the expression is: （14）； wherein, is the operating temperature of the demodulation circuit; is the circuit temperature drift fitting coefficient.

4. The method for collecting and processing the internal temperature of a metalized film capacitor according to claim 3, wherein, Specifically, S3 is: S301, Construction of an adaptive kernel mapping high-order nonlinear compensation model; To address the multi-source deterministic error after decomposition, an adaptive kernel mapping high-order nonlinear compensation model is constructed, with ambient temperature, circuit temperature, pulse current, and heat conduction time constant as independent variables, and the compensation value of the disturbed variable as the output. The expression is as follows: （15）； wherein, is the measured disturbed variable; is the predicted compensation value of the multi-source error; it is solved by an adaptive weighted multi-factor high-order nonlinear compensation model, and the expression is: （16）； in, Let each be the polynomial order of the input variable, and take... A third-order four-factor fully coupled polynomial is constructed to balance fitting accuracy and generalization ability. Let be the fourth-order polynomial fitting coefficient tensor, and be the parameters to be optimized. This is a real-time statistic of circuit noise, taking the signal standard deviation within the sliding window; This is the noise compensation coefficient; The number of support vectors; For Lagrange multipliers; Here, the radial basis function (RBF) kernel is used to capture the nonlinear mapping relationship of the error. The expression for the RBF kernel is: （17）； in, for The input feature vector at time step; For the first Feature vectors of training samples; This is the kernel function width parameter; This refers to the operating temperature of the demodulation circuit. Introducing adaptive weight coefficients The dynamic weighting of the pulse charging and discharging transient process is performed, and the expression is: （18）； wherein, is a transient state weighting factor, ensuring compensation accuracy in the case of a pulse transient; S302, Matrix representation of adaptive kernel mapping high-order nonlinear compensation model; Let the total number of training samples be The input feature vector of each sample is: ； The corresponding measured disturbed variable matrix is , and the real value matrix is The matrix form is: （19）； in, for A vector of error prediction values ​​for each sample; for The polynomial characteristic matrix; The number of terms in a third-order four-factor polynomial. ; for The polynomial coefficient tensor expansion vector; for The kernel function matrix, with elements of ; These are Lagrange multiplier vectors; This is a vector of noise statistics.

5. The method for acquiring and processing the internal temperature of a metallized thin-film capacitor according to claim 4, characterized in that, Specifically, S4 is: S401. Construct a global optimization objective function based on minimizing structural risk; To quantify the deviation between the compensated variables and the true variables, and to avoid overfitting, an improved least squares global optimization objective function with a regularization term is constructed based on the structural risk minimization criterion. This function measures the compensation accuracy and generalization ability. The expression for the global optimization objective function is as follows: （20）； in, It is the Euclidean norm; A vector of true values; This is a vector of measured values; This is a penalty factor used to balance fitting accuracy and complexity; For smoothing coefficients; It is a second-order difference smoothing matrix; The trace of the matrix; S402. The optimal solution requires the KKT conditions (Karl-Kun-T conditions). Global optimization objective function To get the global minimum, the KKT conditions must be satisfied, i.e. the partial derivative of the global optimization objective function with respect to all parameters to be optimized is 0, and the optimal condition expression is: (21)； wherein let the error residual vector Substitute the KKT condition, and convert the optimal solution problem into the solution problem of linear over-determined equations. S403. Introduce constraints on the objective function: To ensure the effectiveness of compensation across the entire operating range of the metallized film capacitor, inequality constraints are introduced to limit the maximum compensation error: (22)； wherein, is the maximum allowed compensation error, set to ±0.5 pm according to the temperature measurement accuracy requirement, corresponding to a temperature error of ±0.1 °C.

6. The method for acquiring and processing the internal temperature of a metallized thin-film capacitor according to claim 5, characterized in that, Specifically, S5 is: S501, Construction of the overdetermined system of equations; Expanding the partial derivative equations in the KKT conditions and combining like terms, we construct a system of linear overdetermined equations in matrix form: （23）； wherein the augmented coefficient matrix is a symmetric positive definite matrix, expressed as: （24）； wherein respectively order, identity matrix of order the parameter vector to be solved containing all model parameters to be optimized; right end constant term vector ; S502, Optimal solution based on the improved conjugate gradient method; S503. Hyperparameter optimization based on the improved beetle whisker particle swarm algorithm; An improved beetle whisker particle swarm optimization algorithm is used for global optimization, with the mean square error of the test set as the fitness function: （25）； wherein, is the number of test set samples; The beetle whisker particle swarm optimization algorithm introduces adaptive inertia weights and a dynamic learning factor to ensure that the algorithm converges quickly to the globally optimal hyperparameter combination and avoids getting trapped in local optima. The expression for the adaptive inertia weights is as follows: （26）； in, This represents the maximum value of the inertia weight; This represents the minimum value of the inertia weight. This represents the current iteration number; Total number of iterations; constant ; S504, Dynamic Calibration and Temperature Calculation; The optimal hyperparameters and the coefficients of the solved adaptive weighted multi-factor high-order nonlinear compensation model are substituted into the adaptive weighted multi-factor high-order nonlinear compensation model to complete the calibration of the adaptive weighted multi-factor high-order nonlinear compensation model. During real-time monitoring, the collected input feature vectors are substituted into the calibrated adaptive weighted multi-factor high-order nonlinear compensation model to obtain the optimized variables after real-time compensation. Finally, substituting the values ​​into the core temperature mapping formula, the true internal temperature of the metallized film capacitor is calculated; the true internal temperature of the metallized film capacitor is: （27）； The calculated temperature value is uploaded to the host computer simultaneously, enabling real-time temperature display, data storage, and linkage with over-temperature warning.

7. The method for collecting and processing the internal temperature of a metalized film capacitor according to claim 6, wherein, Specifically, S502 is as follows: S5021. Initialization: Set the initial value for iteration. Initial residual Preprocessing matrix Solve for the initial search direction Set the iterative convergence threshold Maximum number of iterations ; S5022、th Sub-iteration computation: ； ； ； wherein is the step size for the th iteration, is the residual vector for the th iteration; S5023, convergence judgment: if or , stop iteration, output optimal parameter vector ; otherwise continue iteration: ； ； S5034. After the iteration is completed, from the optimal parameter vector... The polynomial coefficient tensor is obtained by decomposition. Lagrange multipliers Noise compensation coefficient The parameters of the multi-factor high-order nonlinear compensation model with adaptive kernel mapping are solved.

8. A system for acquisition and processing of internal temperature of a metalized film capacitor, characterized by: The method for collecting and processing the internal temperature of a metallized film capacitor as described in claims 1 to 7 further includes: a capacitor insulating core rod (2), the outer surface of which is covered with an internal metallized film (3), the outer surface of which is fitted with a metallized film capacitor shell (1), both ends of which are connected to stainless steel electrode rods (4), and both ends of which are connected to a pulse charging and discharging power supply (10) via wires; the capacitor insulating core rod (2) has a groove along its axial direction, and a fiber optic grating sensor (6) is embedded in the groove, the fiber optic grating sensor (6) is connected to a fiber optic grating demodulator (9) via a transmission fiber (8), and the fiber optic grating demodulator (9) transmits the data to a host computer control system (11).

Images