Enhanced high-voltage circuit breaker service life evaluation method

Abstract

The invention is suitable for the technical field of life evaluation, and provides an enhanced high-voltage circuit breaker life evaluation method, which comprises the following steps: constructing a dynamic evolution model of a contact material wear rate; constructing a nozzle degradation dynamic prediction model; constructing residual life probability distribution of the insulating material; according to the dynamic evolution model of the wear rate of the contact material, the dynamic prediction model of nozzle degradation and the residual life probability distribution of the insulating material, constructing a comprehensive life evaluation index; inputting the comprehensive life evaluation index into a preset layered prediction architecture for evaluation; wherein the first layer generates basic life distribution through a Bayesian network, the second layer outputs posterior life distribution correction parameters through a convolutional neural network, and the third layer optimizes posterior distribution through a KL divergence minimization algorithm to obtain a probability density function and a confidence interval of residual electrical life; according to the method, the error of life evaluation can be reduced on the basis of complex working conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of life assessment, and in particular to a life assessment method for an enhanced high-voltage circuit breaker. Background Art

[0002] High-voltage circuit breakers are core protective equipment that ensures the safe and stable operation of power systems, and their lifespan assessment is directly related to grid reliability. Traditional lifespan assessment methods simplify complex physical processes for the sake of modeling and calculation convenience, ignoring multi-field coupling effects. For example, there is a mutual feedback effect between arc erosion and nozzle thermal stress, and vibration accelerates insulation aging. However, simplified models do not account for this characteristic, resulting in deviations in degradation rate predictions. Due to the prevalence of dynamic operating conditions such as sudden changes in temperature and humidity and high-frequency operation in actual operation, fixed parameters cannot be adjusted as operating conditions change. In extreme scenarios, the mismatch between parameters and actual operating conditions is magnified, resulting in low lifespan prediction accuracy. In addition, traditional physical models are unable to timely incorporate dynamic information from operation to optimize assessments, resulting in large assessment errors and difficulty in warning of sudden failures. Summary of the Invention

[0003] The present application provides an enhanced high-voltage circuit breaker life assessment method for solving the problem of large life assessment errors.

[0004] This application provides an enhanced high-voltage circuit breaker life assessment method, comprising:

[0005] Based on the arc energy distribution, current waveform characteristics and contact pressure parameters during the circuit breaker opening and closing process, a dynamic evolution model of the contact material wear rate is constructed;

[0006] Combined with multi-physics simulation of circuit breakers to analyze the thermal-electrical coupling effect, a dynamic prediction model for nozzle degradation was constructed;

[0007] By monitoring the ambient temperature and humidity parameters, partial discharge intensity, and mechanical vibration parameters during circuit breaker operation, the probability distribution of the remaining life of the insulation material is constructed;

[0008] Constructing a comprehensive life evaluation index based on the dynamic evolution model of the contact material wear rate, the nozzle degradation dynamic prediction model and the remaining life probability distribution of the insulation material;

[0009] The comprehensive life assessment index is input into a preset hierarchical prediction framework for evaluation; wherein the first layer of the preset hierarchical prediction framework generates a basic life distribution through a Bayesian network, the second layer outputs a posterior life distribution correction parameter through a convolutional neural network, and the third layer optimizes the posterior distribution through a KL divergence minimization algorithm to obtain a probability density function and confidence interval of the remaining electrical life;

[0010] Taking the life prediction results as input, the reliability evolution surface under extreme working conditions is generated based on Monte Carlo simulation, and the model parameters of the hierarchical prediction architecture are dynamically corrected.

[0011] Furthermore, the dynamic evolution model of contact material wear rate is constructed based on the arc energy distribution, current waveform characteristics and contact pressure parameters during the circuit breaker opening and closing process, including:

[0012]

[0013] Where: W(t) is the cumulative wear, k1 and k2 are the weight coefficients of arc erosion and mechanical impact, which are calibrated by the circuit breaker type, F n (τ) is the real-time measurement value of the contact pressure, obtained by the embedded piezoelectric sensor, is the effective value of current, Δt arc (τ) is the single arcing time, H is the hardness of the contact material, λ is the temperature softening coefficient, T contact (τ) is the real-time temperature of the contact surface, T ref is the reference temperature, dI / dt| zero is the slope of the current when it passes through zero, I rated is the rated current of the circuit breaker, and m is the mechanical shock sensitivity index.

[0014] Furthermore, the above-mentioned method combines the multi-physics simulation of the circuit breaker to analyze the thermal-electrical coupling effect and constructs a dynamic prediction model for nozzle degradation, including:

[0015]

[0016] Where: ΔD(t) is the cumulative degradation of the nozzle throat diameter, η is the arc erosion rate coefficient, Q arc (τ) is the dynamic arc energy, A throat is the initial cross-sectional area of ​​the nozzle throat, β and n are the thermal stress creep damage coefficients, σ thermal (τ) is the thermal stress on the nozzle surface, σ yield is the yield strength of the nozzle material, γ and E a are the mechanical fatigue coefficient and fatigue activation energy, is the frequency of opening and closing operations per unit time, k B is the Boltzmann constant, and T(τ) is the real-time temperature of the nozzle surface.

[0017] Furthermore, the construction of the probability distribution of the remaining life of the insulation material by monitoring the ambient temperature and humidity parameters, partial discharge intensity and mechanical vibration parameters during the operation of the circuit breaker includes:

[0018] Construct a remaining life model based on the time-varying Weibull distribution:

[0019] η(t)=η0·AF(t) ―1 / θ

[0020] Where: η(t) is the scale parameter at time t, η0 is the initial scale parameter, which is determined by the accelerated electrical aging test of insulating materials, AF(t) ―1 / θ is the acceleration factor at time t, θ is the shape parameter of the Weibull distribution, reflecting the dispersion of life;

[0021]

[0022] Among them: θ(t) is the shape parameter at time t, θ0 is the initial parameter, PD peak (t) is the partial discharge pulse amplitude at time t, PD base is the partial discharge reference amplitude;

[0023]

[0024] Where: f(t) is the probability density function of the remaining life.

[0025] Furthermore, the comprehensive life evaluation index is constructed based on the dynamic evolution model of the contact material wear rate, the nozzle degradation dynamic prediction model and the probability distribution of the remaining life of the insulation material, including:

[0026] Mapping the contact wear amount, nozzle throat diameter degradation amount and insulation remaining life Weibull scale parameter to a unified health index range;

[0027] The weight coefficient is adaptively adjusted based on the operating conditions to construct a comprehensive life assessment index.

[0028] Furthermore, the construction of a comprehensive life evaluation index based on the dynamic evolution model of the contact material wear rate, the nozzle degradation dynamic prediction model and the probability distribution of the remaining life of the insulation material also includes:

[0029] Setting health status classification based on the time series of the comprehensive life assessment indicator;

[0030] When the time series of the comprehensive life evaluation index meets the first threshold range, it is determined to be in a healthy state; when the time series of the comprehensive life evaluation index meets the second threshold range, it is determined to be in a warning state and the enhanced partial discharge monitoring is activated; when the time series of the comprehensive life evaluation index meets the third threshold range, it is determined to be in a critical failure state and an emergency shutdown instruction is generated;

[0031] Through Monte Carlo simulation of contact wear, nozzle throat diameter degradation and insulation life distribution, the global remaining life probability density function is output.

[0032] Furthermore, the first layer of the preset hierarchical prediction architecture generates a basic lifespan distribution through a Bayesian network, including:

[0033] The cumulative contact wear, nozzle throat diameter degradation, and insulation remaining life scale parameters are input into a preset Bayesian network topology architecture; wherein, the root node of the preset Bayesian network topology architecture includes the ambient temperature and humidity parameters and the historical opening and closing operation frequency; the intermediate nodes of the preset Bayesian network topology architecture include the contact material wear rate, nozzle ablation rate, and insulation material aging rate; the leaf nodes of the preset Bayesian network topology architecture include the basic probability distribution of the remaining electrical life; the conditional probability is set based on the joint calibration of the historical fault database and expert experience rules;

[0034] The underlying probability distribution of the remaining electrical lifetime in the form of a Weibull distribution is output by fusing multiple node states through Bayesian inference.

[0035] Furthermore, the second layer outputs the posterior life distribution correction parameters through a convolutional neural network, including:

[0036] The continuous time series of the life assessment indicators and the health index association matrix are input into a preset convolutional neural network; wherein the preset convolutional neural network uses a one-dimensional convolution layer to extract local patterns of temporal features and a maximum pooling layer to compress redundant information; and a fully connected layer to map the spatiotemporal features to correction parameters, including an offset of the expected life value and a scaling factor of the variance;

[0037] The basic distribution parameters output by the Bayesian network are superimposed with the correction amount generated by the convolutional neural network to obtain the mean and variance of the posterior lifetime distribution, and the lower limit of the variance correction amount is constrained to prevent unphysical negative values.

[0038] Furthermore, the third layer optimizes the posterior distribution by using the KL divergence minimization algorithm to obtain the probability density function and confidence interval of the remaining electrical life, including:

[0039] Minimize the difference between the posterior distribution and the historical true life distribution using KL divergence as a measure;

[0040] The parameters of the posterior distribution are adjusted based on the gradient descent method, and the convergence condition is set as the change in KL divergence between two consecutive iterations is less than a threshold;

[0041] Generate optimized remaining electrical lifetime probability density function and confidence intervals.

[0042] Furthermore, the reliability evolution surface under extreme working conditions is generated based on Monte Carlo simulation using the life prediction results as input, and the model parameters of the hierarchical prediction architecture are dynamically corrected, including:

[0043] Preset multi-dimensional parameter ranges for short-circuit current amplitude, ambient temperature and humidity extremes, and tripping and closing frequencies, and perform Monte Carlo random sampling based on the Latin hypercube sampling method to generate a multi-dimensional extreme operating condition sample set encompassing the parameter space.

[0044] The remaining electrical life probability density function and confidence interval output by the hierarchical prediction architecture are input into the multi-physics field coupling simulation model to perform dynamic life prediction under various sample conditions, and the evolution surface is constructed using the cubic spline interpolation algorithm.

[0045] Based on the gradient distribution of probability in the evolution surface, the model parameters of the hierarchical prediction architecture are dynamically modified.

[0046] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0047] The dynamic evolution model of contact material wear rate constructed by the present invention can effectively cope with nonlinear degradation under multi-physical field coupling; the nozzle degradation dynamic prediction model improves the accuracy of nozzle degradation prediction; the probability distribution of the remaining life of the insulating material forms a comprehensive evaluation basis, covering the degradation state of the key components of the high-voltage circuit breaker; according to the dynamic evolution model of the contact material wear rate, the nozzle degradation dynamic prediction model and the probability distribution of the remaining life of the insulating material, a comprehensive life evaluation index is constructed to form a comprehensive evaluation basis, covering the degradation state of the key components of the high-voltage circuit breaker; using Bayesian network, convolutional neural network and KL divergence minimization algorithm, the probability density function and confidence interval of the remaining electrical life are generated, solving the problem of missing traditional probabilistic evaluation, providing quantitative support for accurate operation and maintenance decision-making, and clarifying the spatiotemporal distribution of failure risk; based on Monte Carlo simulation, the reliability evolution surface under extreme working conditions is generated, the parameters of the hierarchical prediction architecture are dynamically adjusted, and sudden failures are effectively warned. The present invention realizes the deep integration of probabilistic evaluation with data and mechanism, effectively copes with complex working conditions, provides reliable support for the construction of intelligent operation and maintenance system, and significantly reduces evaluation error. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The figure is a flow chart of an embodiment of an enhanced high-voltage circuit breaker life assessment method in the present invention. DETAILED DESCRIPTION

[0049] 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.

[0050] The enhanced high-voltage circuit breaker life assessment method in this embodiment is used to reduce the error in life assessment. The implementation method in this embodiment can be implemented in a system, a server, or a terminal, without specific limitation.

[0051] Example 1

[0052] See also Figure 1 An embodiment of an enhanced high-voltage circuit breaker life assessment method in the present invention includes the following steps:

[0053] S11. Construct a dynamic evolution model of contact material wear rate based on arc energy distribution, current waveform characteristics, and contact pressure parameters during circuit breaker opening and closing processes;

[0054] Arc energy distribution includes the effective current value and arcing time. The effective current value is obtained by real-time acquisition of the closing and opening current waveforms using a high-speed recording device, and the RMS value within the time window is calculated. The arcing time is determined by detecting the zero-crossing point of the current waveform and combining it with the arc voltage signal to determine the arc start and end moments. The current waveform characteristics are based on the high-speed recording data, and the slope within the 10μs window before and after the current zero crossing is calculated to reflect the arc reignition risk. The contact pressure is measured by integrating an embedded piezoelectric sensor into the contact spring mechanism, monitoring pressure fluctuations in real time and outputting a standard 4-20mA signal. The contact surface temperature is calibrated using a non-contact infrared thermal imager combined with embedded thermocouples on the contact surface to invert the temperature field distribution.

[0055] The expression of the dynamic evolution model is as follows:

[0056]

[0057] Where: W(t) is the contact wear, k1 and k2 are the weight coefficients of arc erosion and mechanical impact, which are calibrated by the circuit breaker type, F n (τ) is the real-time measurement value of the contact pressure, obtained by the embedded piezoelectric sensor, is the effective value of current, Δt arc (τ) is the single arcing time, H is the hardness of the contact material, λ is the temperature softening coefficient, T contact (τ) is the real-time temperature of the contact surface, T ref is the reference temperature, dI / dt| zero is the slope of the current when it passes through zero, I rated is the rated current of the circuit breaker, and m is the mechanical shock sensitivity index.

[0058] The model implementation process involves performing a sliding average filter on the current, pressure, and temperature signals to eliminate high-frequency noise. The arcing time is then analyzed using a wavelet transform to extract arc characteristic segments. After every 100 operations, k1 and k2 are inversely optimized based on the measured wear (laser scanning of the contact surface topography). λ and m are determined through an accelerated aging test using a 30kA short-circuit current interruption test. The integral term is numerically solved using the Runge-Kutta method (RK4), with the time step size adaptively adjusted. After each opening and closing operation, the accumulated wear W(t) is updated and output to the integrated life assessment system.

[0059] S12. Combine multi-physics simulation of circuit breakers to analyze the thermal-electrical coupling effect and construct a dynamic prediction model for nozzle degradation;

[0060] High-precision current transformers and fiber-optic voltage sensors are used to measure the opening and closing currents and arc voltages in real time to calculate the instantaneous arc energy. A fully coupled electromagnetic, thermal, and mechanical model of the nozzle is established based on the COMSOL Multiphysics platform. Current density, material thermal conductivity, and convection heat transfer coefficient are input, and the transient temperature field and thermal stress distribution are output. The number of operations per unit time is recorded using the built-in counter of the circuit breaker operating mechanism, and the data is transmitted to the evaluation system in real time. An infrared thermal imager and embedded thermocouples are used for combined monitoring, and the temperature data is cross-validated with the simulation results.

[0061] The expression of the nozzle degradation dynamic prediction model is as follows:

[0062]

[0063] Where: ΔD(t) is the nozzle throat diameter degradation, η is the arc erosion rate coefficient, Q arc (τ) is the dynamic arc energy, A throat is the initial cross-sectional area of ​​the nozzle throat, β and n are the thermal stress creep damage coefficients, σ thermal (τ) is the thermal stress on the nozzle surface, σ yield is the yield strength of the nozzle material, γ and E a are the mechanical fatigue coefficient and fatigue activation energy, is the frequency of opening and closing operations per unit time, k B is the Boltzmann constant, and T(τ) is the real-time temperature of the nozzle surface.

[0064] Model implementation process: First, multi-physics simulation is performed. The electromagnetic field is used to solve Maxwell's equations and calculate the Joule heat distribution of the arc plasma; the temperature field is used to couple the heat conduction equation with the convection boundary conditions to simulate the transient temperature rise of the nozzle; the structural field is used to calculate the thermal stress based on the thermal expansion coefficient and output σ thermal . Real-time acquisition of Q arc , σ thermal 、 The data was input into the integral model; the adaptive Simpson method was used for numerical integration, and the time step was dynamically adjusted according to the data update frequency; after every 100 operations, the actual degradation amount ΔD of the nozzle throat diameter was measured by laser scanning, and η, β, and γ were inversely optimized.

[0065] S13. Construct a probability distribution of the remaining life of the insulation material by monitoring the ambient temperature and humidity parameters, partial discharge intensity, and mechanical vibration parameters during circuit breaker operation;

[0066] The temperature parameter in the ambient temperature and humidity parameters monitors the internal temperature of the circuit breaker cabinet in real time using distributed digital temperature sensors with a data sampling frequency of 1Hz. Humidity is monitored using a capacitive humidity sensor installed on the surface of the insulation material to monitor condensation risks. Partial discharge intensity refers to the discharge amplitude and frequency. The amplitude is captured by an ultra-high frequency (UHF) sensor to capture partial discharge pulses and extract the peak voltage. The frequency is measured by counting the number of discharge pulses per unit time to identify the deterioration trend of insulation defects. Mechanical vibration parameters include vibration acceleration and main frequency. Acceleration is based on a triaxial piezoelectric accelerometer to monitor the mechanical impact energy during circuit breaker operation. The main frequency is analyzed by FFT analysis of the vibration signal to extract the resonant frequency component.

[0067] Scale parameters are updated dynamically:

[0068] η(t)=η0·AF(t) ―1 / θ

[0069] Where: η(t) is the scale parameter at time t, η0 is the initial scale parameter, which is determined by the accelerated electrical aging test of insulating materials, AF(t) ―1 / θ is the acceleration factor at time t, θ is the shape parameter of the Weibull distribution, reflecting the dispersion of life;

[0070] Calculation formula for multi-stress acceleration factor:

[0071]

[0072] Where: T amb is the ambient temperature at time t, T ref is the reference temperature, ΔT PD (t) is the temperature rise caused by partial discharge, which is inverted by infrared thermal imaging. RH(t) is the relative humidity at time t. PD peak (t) is the discharge amplitude at time t, PD base is the initial partial discharge reference amplitude, PD rate (t) is the discharge repetition rate at time t, PD rate,ref is the benchmark discharge frequency, A vib (t) is the vibration acceleration at time t, A ref is the reference value of vibration acceleration, α, β, γ, and ζ are the material sensitivity coefficients respectively.

[0073] Adaptive adjustment of shape parameters:

[0074]

[0075] Among them: θ(t) is the shape parameter at time t, θ0 is the initial parameter, PD peak (t) is the partial discharge pulse amplitude at time t, PD base is the baseline amplitude of partial discharge; when the amplitude of partial discharge increases, the shape parameter θ(t) increases, the life distribution tends to be concentrated, and the failure risk increases sharply.

[0076] Remaining life probability density function:

[0077]

[0078] Where: f(t) is the probability density function of remaining life. It quantifies the probability of instantaneous failure around a specified time t and supports risk early warning.

[0079] Model implementation process: real-time collection of T amb , RH, PD peak 、A vib , input into the acceleration factor equation; AF(t) is updated every 5 minutes to drive dynamic adjustment of η(t). Every 30 days, the material state is measured using an insulation resistance tester, and η0 and θ0 are reversely optimized. A remaining life probability density curve and 95% confidence interval are generated. When the integral value f(t) (cumulative failure probability) exceeds 80%, a high-level warning is triggered.

[0080] S14. Construct a comprehensive life assessment index based on the dynamic evolution model of contact material wear rate, the dynamic prediction model of nozzle degradation, and the probability distribution of the remaining life of the insulation material;

[0081] 1. Map contact wear, nozzle throat diameter degradation, and insulation remaining life Weibull scale parameters to a unified health index range;

[0082] Set multi-dimensional health indexes, including contact health index, nozzle health index, and insulation health index; map the cumulative contact wear W(t) to the [0,1] interval: Where W crit is the critical threshold of contact failure, which is calibrated by the circuit breaker type test; the nozzle throat diameter degradation ΔD(t) is normalized: where ΔD crit is the maximum allowable degradation of the nozzle; the remaining life scale parameter based on the Weibull distribution is calculated

[0083] 2. Adaptively adjust the weight coefficient based on operating conditions to construct a comprehensive life assessment index.

[0084] Based on the adaptive adjustment of the weight coefficient of the operating conditions, a comprehensive life evaluation index L is constructed. comprehensive (t):

[0085] L comprehensive (t)=w1·H contact (t)+w2·H nozzle (t)+w3·H insulation (t)

[0086] Among them: w1, w2 and w3 are the weight coefficients corresponding to the contact health index, nozzle health index and insulation health index respectively.

[0087] Calculate the initial weight using the entropy weight method satisfy Reflects the sensitivity of each subsystem to the overall life. op >N th , triggering weight adjustment: When I is detected peak >2I rated , the weights are forced to be reset to w1:w2:w3=0.6:0.3:0.1.

[0088] Specifically, based on the comprehensive indicator L comprehensive (t), perform the following operations:

[0089] Setting health status classification based on the time series of comprehensive life assessment indicators;

[0090] When the time series of the comprehensive life evaluation index meets the first threshold range, that is, L comprehensive ≥0.8, it is judged to be in a healthy state; when the time series of the comprehensive life evaluation index meets the second threshold range, that is, 0.5≤L comprehensive <0.8, it is judged as a warning state and the enhanced monitoring of partial discharge is activated; when the time series of the comprehensive life evaluation index meets the third threshold range, that is, L comprehensive <0.5, it is determined to be a critical failure state and an emergency shutdown command is generated.

[0091] Through Monte Carlo simulation of contact wear, nozzle throat diameter degradation and insulation life distribution, the global remaining life probability density function is output: the global remaining life probability density function is output through kernel density estimation (KDE):

[0092]

[0093] Where: i is the failure weight of each subsystem. Finally, the 95% confidence interval [μ―1.96σ,μ+1.96σ] is calculated to support risk quantification decision-making.

[0094] S15. The comprehensive life assessment index is input into a preset hierarchical prediction framework for evaluation. The first layer of the preset hierarchical prediction framework generates a basic life distribution using a Bayesian network, the second layer outputs a posterior life distribution correction parameter using a convolutional neural network, and the third layer optimizes the posterior distribution using a KL divergence minimization algorithm to obtain a probability density function and confidence interval for the remaining electrical life.

[0095] S151. The first layer of the pre-defined hierarchical prediction architecture generates the basic lifespan distribution through a Bayesian network, including:

[0096] 1. The cumulative contact wear, nozzle throat diameter degradation and insulation remaining life scale parameters are input into the preset Bayesian network topology architecture; the root node of the preset Bayesian network topology architecture includes the ambient temperature and humidity parameters and the historical opening and closing operation frequency; the intermediate nodes of the preset Bayesian network topology architecture include the contact material wear rate, nozzle ablation rate and insulation material aging rate; the leaf nodes of the preset Bayesian network topology architecture include the basic probability distribution of the remaining electrical life; the conditional probability is set based on the joint calibration of the historical fault database and expert experience rules.

[0097] Network structure design: The Bayesian network consists of three layers of nodes:

[0098] Root node: includes environmental temperature and humidity parameters collected in real time by distributed sensors and historical opening and closing operation frequencies extracted from the circuit breaker operation database;

[0099] Intermediate nodes: Correlate arc energy with contact pressure to determine contact material wear rate, thermal stress with operating frequency to determine nozzle ablation rate, and couple temperature, humidity, and partial discharge data to determine insulation material aging rate;

[0100] Leaf node: basic probability distribution of remaining electrical life, outputting life parameters in the form of Weibull distribution.

[0101] Conditional Probability Table (CPT) Calibration: This method defines causal relationships between nodes based on a historical fault database and expert experience rules. For example, in high-temperature and high-humidity environments, the conditional probability of the insulation aging rate increases by 30%. In high-frequency operation scenarios, the probability of the contact wear rate being marked as "high" reaches 85%. Contact wear, nozzle degradation, and insulation life parameters are collected in real time and input into the network simultaneously with data such as ambient temperature, humidity, and operation frequency. Each parameter is normalized to a uniform range to eliminate dimensional differences. For example, contact wear is mapped proportionally to the critical failure threshold.

[0102] 2. The basic probability distribution of the remaining electrical life in the form of Weibull distribution is output by fusing multiple node states through Bayesian inference.

[0103] By integrating real-time data with historical patterns using Bayesian theorem, the conditional probability of remaining life under given contact, nozzle, and insulation degradation conditions is calculated. For example, when the ambient temperature exceeds 50°C and the frequency of opening and closing is greater than 1000 times per month, the remaining life distribution is inferred to shift toward shorter life. When a partial discharge anomaly is detected, the probability weight of the insulation aging node is updated. Finally, the scale parameter (reflecting the average life) and shape parameter (characterizing the life dispersion) of the Weibull distribution are output. The parameter values ​​are determined by historical data statistics and maximum likelihood estimation.

[0104] S152. The second layer outputs the posterior life distribution correction parameters through the convolutional neural network, including:

[0105] 1. The continuous time series of life assessment indicators and the health index correlation matrix are input into a pre-set convolutional neural network. The pre-set convolutional neural network uses a one-dimensional convolutional layer to extract local patterns of temporal features and a maximum pooling layer to compress redundant information. The fully connected layer maps spatiotemporal features to correction parameters, including an offset for the expected lifespan and a scaling factor for the variance.

[0106] Temporal feature input: The continuous time series of the comprehensive life assessment index is input into a one-dimensional convolutional layer to extract local degradation patterns such as short-term fluctuations and cyclical trends. Spatial feature input: A correlation matrix is ​​constructed for the contact, nozzle, and insulation health index to quantify the synergistic degradation effects between components, such as contact wear accelerating nozzle ablation. The one-dimensional convolutional layer of the CNN architecture: 64 filters (size 5×1) with a ReLU activation function are used to capture key features in the time series signal (such as sudden changes in arc energy and surges in vibration frequency). The maximum pooling layer: Using a 2×1 pooling window, the feature dimension is compressed, retaining significant features (such as extended arcing time and abnormal temperature rise rate). The fully connected layer: Mapping the spatiotemporal features to correction parameters, including an offset to the expected life value and a variance scaling factor.

[0107] 2. The basic distribution parameters output by the Bayesian network are superimposed with the corrections generated by the convolutional neural network to obtain the mean and variance of the posterior lifetime distribution, and the lower limit of the variance correction is constrained to prevent unphysical negative values.

[0108] The basic distribution parameters output by the Bayesian network Superimposed with CNN correction; mean correction: μ post =μ prior +Δμ, if high-frequency vibration is detected, Δμ is negative and the expected life span is shortened; Variance correction: Constraint Δσ 2 ≥―0.5 to avoid variance.

[0109] The above steps achieve dynamic optimization of the lifetime distribution from a priori to posterior through data-driven feature mining and physical constraint parameter correction, providing high-precision input for the third-layer KL divergence optimization.

[0110] S153. The third layer optimizes the posterior distribution through the KL divergence minimization algorithm to obtain the probability density function and confidence interval of the remaining electrical life, including:

[0111] 1. Minimize the difference between the posterior distribution and the historical true lifespan distribution using KL divergence as a metric;

[0112] 2. Adjust the parameters of the posterior distribution based on the gradient descent method, and set the convergence condition to be that the change in KL divergence between two consecutive iterations is less than a threshold;

[0113] 3. Generate the optimized remaining electrical life probability density function and confidence interval.

[0114] In the third layer, the KL divergence is first used as the optimization target, and the model prediction deviation is quantified by comparing the difference between the posterior distribution and the historical true life distribution. Specifically, the true life distribution is constructed using historical failure data, and the KL divergence value is approximately calculated through Monte Carlo sampling to measure the degree of deviation between the posterior distribution (generated by the correction parameters output by the second layer) and the true distribution. Subsequently, the gradient descent method is used to iteratively optimize the parameters of the posterior distribution (such as the scale parameter and shape parameter of the Weibull distribution), and the parameter values ​​are gradually adjusted according to the gradient direction of the KL divergence to the parameters to minimize the distribution difference. During the optimization process, the convergence condition is set as the change in KL divergence between two consecutive iterations is less than a preset threshold (such as 10-410-4), to ensure that the algorithm terminates in a stable state and avoid overfitting or invalid iterations.

[0115] After optimization is complete, a probability density function for the remaining electrical life is generated based on the updated distribution parameters, clarifying the probability density of failure at different time points. Simultaneously, a 95% confidence interval is calculated based on parameter uncertainty information (such as variance), and the predicted life time range is output (e.g., the probability of remaining life being within 1000, 1500, 1000, and 1500 hours is 95%). The final results are presented through a visual interface, supporting operations and maintenance personnel in formulating preventive maintenance plans or emergency intervention strategies.

[0116] S16. Using the life prediction results as input, the reliability evolution surface under extreme working conditions is generated based on Monte Carlo simulation, and the model parameters of the hierarchical prediction architecture are dynamically modified.

[0117] 1. Preset multi-dimensional parameter ranges for short-circuit current amplitude, ambient temperature and humidity extremes, and tripping and closing frequencies, and perform Monte Carlo random sampling based on the Latin hypercube sampling method to generate a multi-dimensional extreme operating condition sample set encompassing the parameter space;

[0118] The preset short-circuit current is 150%-300% of the rated value, the ambient temperature and humidity are set to -40℃ to 100℃, the humidity is 90%-100%, and the opening and closing frequency is 1-10 times / minute. The multi-dimensional parameter range is set to generate an extreme operating condition sample set covering the entire parameter space based on Latin hypercube sampling to ensure sample uniformity and representativeness.

[0119] 2. Input the remaining electrical life probability density function and confidence interval output by the hierarchical prediction architecture into the multi-physics field coupling simulation model, perform dynamic life prediction under various sample conditions, and construct the evolution surface using the cubic spline interpolation algorithm;

[0120] The life probability distribution output by the hierarchical architecture is input into the electromagnetic-thermal-mechanical coupling simulation model, and the dynamic life decay process is simulated sample by sample. The time, working condition intensity and reliability probability are integrated through the cubic spline interpolation algorithm to construct a three-dimensional evolution surface to intuitively characterize the life degradation law under extreme scenarios.

[0121] 3. Dynamically modify the model parameters of the hierarchical prediction architecture based on the gradient distribution of probability in the evolution surface.

[0122] By analyzing the distribution characteristics of reliability gradients within the surface, such as the rapid attenuation trend in high-temperature and high-humidity areas, we targeted adjustments to the layered architecture parameters—the Bayesian network's conditional probability table, the CNN convolution kernel weights, and the regularization coefficient of the KL divergence objective function. We also introduced a closed-loop feedback mechanism to periodically compare simulation predictions with actual O&M data, triggering parameter retraining. For scenarios involving high-frequency switching of new energy vehicles and coastal condensation, we incorporated characteristic factors into the correction process to optimize the model's response accuracy to specific risks.

[0123] The above-mentioned embodiment realizes the deep integration of probabilistic evaluation, data and mechanism, effectively copes with complex working conditions, provides reliable support for the construction of intelligent operation and maintenance system, and significantly reduces evaluation errors.

[0124] Those skilled in the art will appreciate that the units of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0125] In the embodiments provided herein, it should be understood that the division of units is merely a logical functional division. In actual implementation, other division methods may be employed, such as combining multiple units into one unit, splitting a unit into multiple units, or ignoring certain features. Furthermore, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically as a separate unit, or two or more units may be integrated into a single unit. These integrated units may be implemented in either hardware or software functional units.

[0126] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nly Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc., various media that can store program code.

[0127] It can be understood that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

Claims

1. An enhanced high-voltage circuit breaker life assessment method, characterized in that: include: Based on the arc energy distribution, current waveform characteristics and contact pressure parameters during the circuit breaker opening and closing process, a dynamic evolution model of the contact material wear rate is constructed; Combined with multi-physics simulation of circuit breakers to analyze the thermal-electrical coupling effect, a dynamic prediction model for nozzle degradation was constructed; By monitoring the ambient temperature and humidity parameters, partial discharge intensity, and mechanical vibration parameters during circuit breaker operation, the probability distribution of the remaining life of the insulation material is constructed; Constructing a comprehensive life evaluation index based on the dynamic evolution model of the contact material wear rate, the nozzle degradation dynamic prediction model and the remaining life probability distribution of the insulation material; The comprehensive life assessment index is input into a preset hierarchical prediction framework for evaluation; wherein the first layer of the preset hierarchical prediction framework generates a basic life distribution through a Bayesian network, the second layer outputs a posterior life distribution correction parameter through a convolutional neural network, and the third layer optimizes the posterior distribution through a KL divergence minimization algorithm to obtain a probability density function and confidence interval of the remaining electrical life; Taking the life prediction results as input, the reliability evolution surface under extreme working conditions is generated based on Monte Carlo simulation, and the model parameters of the hierarchical prediction architecture are dynamically corrected.

2. The enhanced high-voltage circuit breaker life assessment method according to claim 1, characterized in that: The dynamic evolution model of contact material wear rate is constructed based on the arc energy distribution, current waveform characteristics and contact pressure parameters during the circuit breaker opening and closing process, including: Where: W(t) is the cumulative wear, k1 and k2 are the weight coefficients of arc erosion and mechanical impact, which are calibrated by the circuit breaker type, F n (τ) is the real-time measurement value of the contact pressure, obtained by the embedded piezoelectric sensor, is the effective value of current, Δt arc (τ) is the single arcing time, H is the hardness of the contact material, λ is the temperature softening coefficient, T contact (τ) is the real-time temperature of the contact surface, T ref is the reference temperature, dI / dt| zero is the slope of the current when it passes through zero, I rated is the rated current of the circuit breaker, and m is the mechanical shock sensitivity index.

3. The enhanced high-voltage circuit breaker life assessment method according to claim 1, characterized in that: The method combines the multi-physics simulation of the circuit breaker to analyze the thermal-electric coupling effect and builds a nozzle degradation dynamic prediction model, including: Where: ΔD(t) is the cumulative degradation of the nozzle throat diameter, η is the arc erosion rate coefficient, Q arc (τ) is the dynamic arc energy, A throat is the initial cross-sectional area of ​​the nozzle throat, β and n are the thermal stress creep damage coefficients, σ thermal (τ) is the thermal stress on the nozzle surface, σ yield is the yield strength of the nozzle material, γ and E a are the mechanical fatigue coefficient and fatigue activation energy, is the frequency of opening and closing operations per unit time, k B is the Boltzmann constant, and T(τ) is the real-time temperature of the nozzle surface.

4. The enhanced high-voltage circuit breaker life assessment method according to claim 1, characterized in that: The method of constructing a probability distribution of the remaining life of the insulation material by monitoring the ambient temperature and humidity parameters, partial discharge intensity, and mechanical vibration parameters during the operation of the circuit breaker includes: Construct a remaining life model based on the time-varying Weibull distribution: η(t)=η0·AF(t) ―1 / θ Where: η(t) is the scale parameter at time t, η0 is the initial scale parameter, which is determined by the accelerated electrical aging test of insulating materials, AF(t) ―1 / θ is the acceleration factor at time t, θ is the shape parameter of the Weibull distribution, reflecting the dispersion of life; Among them: θ(t) is the shape parameter at time t, θ0 is the initial parameter, PD peak (t) is the partial discharge pulse amplitude at time t, PD base is the partial discharge reference amplitude; Where: f(t) is the probability density function of the remaining life.

5. The enhanced high-voltage circuit breaker life assessment method according to any one of claims 1 to 4, characterized in that: The comprehensive life evaluation index is constructed based on the dynamic evolution model of the contact material wear rate, the nozzle degradation dynamic prediction model and the remaining life probability distribution of the insulation material, including: Mapping the contact wear amount, nozzle throat diameter degradation amount and insulation remaining life Weibull scale parameter to a unified health index range; The weight coefficient is adaptively adjusted based on the operating conditions to construct a comprehensive life assessment index.

6. The enhanced high-voltage circuit breaker life assessment method according to claim 5, characterized in that: The comprehensive life evaluation index is constructed based on the dynamic evolution model of the contact material wear rate, the nozzle degradation dynamic prediction model and the remaining life probability distribution of the insulation material, and further includes: Setting health status classification based on the time series of the comprehensive life assessment indicator; When the time series of the comprehensive life evaluation index meets the first threshold range, it is determined to be in a healthy state; when the time series of the comprehensive life evaluation index meets the second threshold range, it is determined to be in a warning state and the enhanced partial discharge monitoring is activated; when the time series of the comprehensive life evaluation index meets the third threshold range, it is determined to be in a critical failure state and an emergency shutdown instruction is generated; Through Monte Carlo simulation of contact wear, nozzle throat diameter degradation and insulation life distribution, the global remaining life probability density function is output.

7. The enhanced high-voltage circuit breaker life assessment method according to claim 1, characterized in that: The first layer of the preset hierarchical prediction architecture generates the basic lifespan distribution through a Bayesian network, including: The cumulative contact wear, nozzle throat diameter degradation, and insulation remaining life scale parameters are input into a preset Bayesian network topology architecture; wherein, the root node of the preset Bayesian network topology architecture includes the ambient temperature and humidity parameters and the historical opening and closing operation frequency; the intermediate nodes of the preset Bayesian network topology architecture include the contact material wear rate, nozzle ablation rate, and insulation material aging rate; the leaf nodes of the preset Bayesian network topology architecture include the basic probability distribution of the remaining electrical life; the conditional probability is set based on the joint calibration of the historical fault database and expert experience rules; The underlying probability distribution of the remaining electrical lifetime in the form of a Weibull distribution is output by fusing multiple node states through Bayesian inference.

8. The enhanced high-voltage circuit breaker life assessment method according to claim 7, characterized in that: The second layer outputs the posterior life distribution correction parameters through a convolutional neural network, including: The continuous time series of the life assessment indicators and the health index association matrix are input into a preset convolutional neural network; wherein the preset convolutional neural network uses a one-dimensional convolution layer to extract local patterns of temporal features and a maximum pooling layer to compress redundant information; and a fully connected layer to map the spatiotemporal features to correction parameters, including an offset of the expected life value and a scaling factor of the variance; The basic distribution parameters output by the Bayesian network are superimposed with the correction amount generated by the convolutional neural network to obtain the mean and variance of the posterior lifetime distribution, and the lower limit of the variance correction amount is constrained to prevent unphysical negative values.

9. The enhanced high-voltage circuit breaker life assessment method according to claim 8, characterized in that: The third layer optimizes the posterior distribution by the KL divergence minimization algorithm to obtain the probability density function and confidence interval of the remaining electrical life, including: Minimize the difference between the posterior distribution and the historical true life distribution using KL divergence as a measure; The parameters of the posterior distribution are adjusted based on the gradient descent method, and the convergence condition is set as the change in KL divergence between two consecutive iterations is less than a threshold; Generate optimized remaining electrical lifetime probability density function and confidence intervals.

10. The enhanced high-voltage circuit breaker life assessment method according to claim 1, characterized in that: The method uses the life prediction results as input, generates the reliability evolution surface under extreme working conditions based on Monte Carlo simulation, and dynamically corrects the model parameters of the hierarchical prediction architecture, including: Preset multi-dimensional parameter ranges for short-circuit current amplitude, ambient temperature and humidity extremes, and tripping and closing frequencies, and perform Monte Carlo random sampling based on the Latin hypercube sampling method to generate a multi-dimensional extreme operating condition sample set encompassing the parameter space. The remaining electrical life probability density function and confidence interval output by the hierarchical prediction architecture are input into the multi-physics field coupling simulation model to perform dynamic life prediction under various sample conditions, and the evolution surface is constructed using the cubic spline interpolation algorithm. Based on the gradient distribution of probability in the evolution surface, the model parameters of the hierarchical prediction architecture are dynamically modified.

Images