A method and system for monitoring the reliability of a photovoltaic inverter in a high-altitude special environment

Abstract

The application discloses a kind of plateau special environment under photovoltaic converter reliability monitoring method and system. Among them, the method comprises: first, the electrical quantity, temperature, environmental quantity and operating quantity are collected, and abnormal data are filtered, and dynamic time warping algorithm is combined to realize the accurate alignment of multi-frequency signal.Second, a set of plateau sensitive features is constructed, and the quality of the features is optimized through physical consistency test and three-stage feature selection. Then, the health index evolution model is established by fusing IGBT thermal fatigue equation and capacitor aging equation, and the prediction is carried out by using Bayesian physical information neural network. The high reliability confidence interval is output by Monte Carlo sampling. Finally, the remaining life is dynamically corrected based on the comprehensive environmental factors, and the graded maintenance decision is triggered according to the health index state, the remaining life and the confidence interval width. The application significantly improves the life prediction accuracy of photovoltaic converter under plateau environment, effectively reduces the operation and maintenance cost, and enhances the operation reliability of equipment.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic power generation equipment health management technology, specifically relating to a method and system for monitoring the reliability of photovoltaic converters in special high-altitude environments. Background Technology

[0002] The rapid development of photovoltaic power plants in high-altitude areas is facing a reliability management dilemma. The harsh environment at altitudes exceeding 3,000 meters constitutes a unique "triple strangulation chain": the thin air significantly weakens the efficiency of the heat dissipation system, making the converter power modules feel like they are being continuously tortured in a high-temperature oven; intense ultraviolet radiation acts like an invisible scalpel, relentlessly eroding the capacitor insulation materials and IGBT package casings day and night; and the drastic temperature difference between day and night acts like a hot and cold blacksmith, subjecting the metal weld layers to a daily quenching and tempering cycle. These coupled environmental stresses are not simply superimposed, but rather produce a synergistic amplification effect—the slow heat dissipation under low air pressure exacerbates the thermal expansion stress caused by temperature differences, while ultraviolet photodegradation reduces the toughness of materials against mechanical fatigue, leading to a precipitous decline in equipment lifespan.

[0003] However, current mainstream monitoring technologies remain trapped in the "plains thinking" paradigm. Temperature early warning systems rely on heat transfer models based on standard atmospheric conditions, completely ignoring the fatal impact of sudden pressure drops on air-cooled / liquid-cooled efficiency—similar to using a sea-level barometer to predict high-altitude weather. Electrical characteristic diagnostics mechanically apply plains indicators such as harmonic distortion rate, failing to capture the ripple anomalies caused by electrolyte boiling in low-pressure environments. Lifespan prediction models are obsessed with the data black box, treating physical mechanisms such as thermal fatigue crack propagation and electrochemical aging as noise filters, resulting in predictions that are like building a tower on sand in data-scarce high-altitude scenarios. Even more serious is the severe disconnect between operation and maintenance strategies and actual environmental conditions: regular maintenance ignores the seasonal patterns of rampant ultraviolet radiation in summer and extreme temperature differences in winter, resulting in a lack of monitoring during critical equipment degradation periods; fault repair is hampered by logistical difficulties in high-altitude areas, often leading to a vicious cycle of "minor problems becoming major ones."

[0004] This systemic failure is triggering a chain reaction of crises: power plant operators are caught between "overspending on maintenance" and "losing money due to downtime," equipment manufacturers are facing numerous quality claims due to frequent high-altitude malfunctions, and the entire industry's green energy transformation is being hampered by high-altitude reliability issues. The industry urgently needs a technological paradigm revolution—it needs monitoring methods that can see through the coupled damage mechanisms of "low pressure-radiation-temperature difference"; it needs predictive models that integrate physical essence and data intelligence; and it needs a dynamic operation and maintenance framework that transforms environmental stress into decision parameters, enabling high-altitude power plant operations to evolve from "firefighting emergency response" to "prevention-oriented management." Summary of the Invention

[0005] This invention proposes methods such as Bayesian inference, physical constraint fusion, data-driven approach, and uncertainty quantification to accurately monitor the reliability of photovoltaic converters in high-altitude environments.

[0006] The specific plan is as follows:

[0007] A method for reliability monitoring of photovoltaic converters under special high-altitude environments includes the following steps:

[0008] S1. Data Acquisition and Preprocessing: Electrical, temperature, environmental, and operational quantities of the plateau converter are collected. Outlier processing based on the 3σ criterion, elimination of power fluctuations, selection of stable periods, and time warping (DTW) algorithm are used to timestamp data at different sampling frequencies to ensure high data quality and accurate correspondence.

[0009] S2. Plateau-sensitive feature engineering: Construct a feature system including voltage ripple rate, high-frequency harmonic energy ratio, junction temperature fluctuation entropy, thermal response time, thermal shock coefficient and comprehensive stress index, and select the optimal feature subset through physical consistency and three-stage feature selection method;

[0010] S3. Physical constraint modeling: Establish IGBT thermal fatigue and capacitor aging models, combine health state evolution equations to quantitatively assess equipment health status, and introduce physical residuals to ensure that the data-driven model is consistent with physical laws.

[0011] S4, Bayesian-PINN fusion modeling: Construct a network architecture that includes an input layer, a Bayesian hidden layer and an output layer, and set a loss function that integrates data fitting terms, physical constraint terms and Bayesian regularization terms to achieve deep coupling between data and physical models, thereby improving the model's generalization ability and prediction accuracy;

[0012] S5, RUL Prediction Process: The conversion method from health indicators to remaining useful life (RUL) takes into account prediction uncertainty, introduces an environmental adaptive correction mechanism, and achieves precise maintenance decision support based on multi-indicator decision logic.

[0013] S6. Plateau Environment Adaptive Mechanism: Automatically adjusts model parameters and feature weights according to altitude and seasonal changes, enabling the monitoring system to maintain high-performance monitoring under different environmental conditions.

[0014] Furthermore, step S1, data acquisition and preprocessing, specifically includes:

[0015] (1) Data source

[0016] The collected data includes electrical quantities, temperature quantities, environmental quantities, and operational quantities of the plateau converter. Electrical quantities include DC bus voltage (V). dc ) and AC output current (I ac The sampling frequency is 10kHz, and the temperature is measured by a Hall sensor. Temperature parameters include the IGBT junction temperature (T). j ) and radiator temperature (T hsThe sampling frequency was 1Hz, measured using thermocouples. Environmental parameters included ultraviolet radiation (UV), altitude (h), and air pressure (P), with a sampling frequency of 0.1Hz, provided by a weather station. Operational parameters included output power (P). loss ) and switching frequency (f sw The sampling frequency is 1kHz, and the data comes from the SCADA system.

[0017] (2) Preprocessing process

[0018] ① Outlier Handling: Based on the 3σ criterion, if data x satisfies |x-μ|>3σ, it is identified as an outlier and removed. Here, μ represents the mean of the data, and σ represents the standard deviation of the data. For periods of rapid power change, if the absolute value of the power change rate satisfies… Then remove that data segment. Here, P loss P represents output power. rated This is the rated power.

[0019] ② Selection of stable time period: The light intensity must meet the requirement of >800W / m 2 Power fluctuations must be less than 5% (within a 10-minute window), and the time window is limited to 10:00-14:00 (true solar time).

[0020] ③ Signal Alignment: The DTW (Dynamic Time Warping) algorithm is used, with the goal of minimizing ∑d(a i ,b j ), where a i and b j These are data points from two time series, d(a) i ,b j This indicates the distance between them. And timestamps are applied to both environmental and electrical data.

[0021] Furthermore, step S2, the plateau-sensitive feature engineering, specifically includes:

[0022] (1) Feature system

[0023] ① The formula for calculating voltage ripple rate is:

[0024]

[0025] In the formula, V dc This is the DC bus voltage. RMS represents the root mean square value, and h is the altitude. This characteristic reflects the aging degree of the capacitor; the higher the ripple rate, the worse the capacitor's filtering effect, and the more severe the aging may be. The influence of altitude on the ripple rate is also taken into account.

[0026] ②The formula for calculating the high-frequency harmonic energy ratio is:

[0027]

[0028] In the formula, f sw Where is the switching frequency, and PSD(f) is the power spectral density. This characteristic reflects the degradation of IGBT switching characteristics; as IGBT performance declines, the proportion of high-frequency harmonic energy increases.

[0029] ③ The formula for calculating the junction temperature fluctuation entropy is:

[0030]

[0031] In the formula, T j p(T) is the junction temperature of the IGBT. j ) is its probability distribution, P loss0 For reference power, P loss This represents the actual output power. It measures the complexity of junction temperature fluctuations and reflects thermal fatigue damage; the more complex the junction temperature fluctuations, the more severe the thermal fatigue damage may be.

[0032] ④ The formula for calculating thermal response time is:

[0033]

[0034] In the formula, T j (t) is a function of junction temperature as a function of time, P loss (t) is the function of output power over time, and corr represents the correlation. This characteristic reflects the performance of the heat dissipation system; the longer the thermal response time, the more likely there are problems with the heat dissipation system or its performance has degraded.

[0035] ⑤ The formula for calculating the thermal shock coefficient is:

[0036]

[0037] In the formula, ΔT day For the diurnal temperature range, N cycle The thermal cycle count measures the impact of diurnal temperature stress on the equipment. The greater the temperature difference and the more cycles, the greater the thermal shock coefficient and the greater the environmental stress the equipment experiences.

[0038] ⑥ The formula for calculating the comprehensive stress index is:

[0039]

[0040] In the formula, UV represents ultraviolet radiation intensity, and h represents altitude. This index comprehensively considers factors such as ultraviolet radiation, thermal shock, and altitude, fully reflecting the overall degree of damage to equipment caused by the high-altitude environment.

[0041] (2) Feature fusion

[0042] ① Physical consistency check: verification Ensure that the capacitor's ripple rate changes with temperature in accordance with the capacitor's temperature characteristics. Verify |corr(S) T ,T a )|<0.3, to exclude the interference of ambient temperature on junction temperature fluctuation entropy.

[0043] ② Three-stage feature selection: First, features with small variance are removed by using a variance threshold. Then, physical screening is performed to retain features that conform to physical laws. Finally, the recursive feature elimination (RFE) algorithm is used to select the optimal feature subset.

[0044] ③ Final eigenvector: x = [R] r E h ,S T ,τ th ,Λ,K ts ] T .

[0045] Furthermore, step S3, physical constraint modeling, specifically includes:

[0046] (1) Core physical equations

[0047] ① IGBT thermal fatigue model:

[0048]

[0049] In the formula, N f This represents the thermal fatigue life of the IGBT, where A is a constant and ΔT is the thermal fatigue life. j For junction temperature variation, β is a material constant, and E a To activate energy, k B T is the Boltzmann constant. m The average temperature is given. This model describes the life characteristics of IGBTs under thermal stress.

[0050] ② Capacitor aging model:

[0051]

[0052] In the formula, L is the capacitor lifespan, L0 is the reference lifespan, and E a The activation energy is T, the temperature is V, the voltage is V0, the reference voltage is n, and the aging index is n. This model reflects the aging behavior of a capacitor under the combined effects of electrical and thermal stress.

[0053] (2) Evolutionary equation of health status

[0054] ① Definition of health indicator: u(t)∈[0,1], initial time u(0)=1, failure time u(t) fail ) = 0.2

[0055] The health indicator u(t) is a dimensionless parameter that quantifies the health status of equipment. It is defined within the range [0,1], facilitating a unified assessment and monitoring of equipment health status. At the initial moment, u(0) = 1 indicates that the equipment is in a brand-new, undamaged state; at the time of failure, u(t) = 1. fail The value of 0.2 is determined based on the equipment's failure threshold. When the health indicator drops to 0.2, the equipment is considered to have reached the end of its lifespan and needs to be maintained or replaced.

[0056] ② Degradation rate modeling:

[0057]

[0058] The degradation rate, or the rate of change of a health indicator over time, reflects the speed at which the health of equipment deteriorates. It is composed of the negative sum of the degradation rates of IGBTs and capacitors, indicating that the degradation of IGBTs and capacitors is the primary factor leading to a decline in the overall health of the equipment. Specifically,

[0059]

[0060] In the formula, C1 and C2 are constants related to materials and processes; E a,IGBT and E a,Cap The activation energies of the IGBT and capacitor are respectively, reflecting the energy barriers that their internal failure mechanisms need to overcome; R is the gas constant; T m T and T represent the average ambient temperatures of the IGBT and capacitor, respectively; ΔT j The change in IGBT junction temperature reflects the impact of thermal stress on IGBT degradation; β is a material constant related to the thermomechanical properties of the IGBT material; K ts It is the thermal shock coefficient, which quantifies the impact of thermal shocks such as diurnal temperature variations on IGBT degradation; V and V0 are the actual voltage and reference voltage that the capacitor withstands, respectively; n is the aging exponent, which describes the influence of voltage on the aging rate of the capacitor; I uv These are ultraviolet (UV) intensity indicators, reflecting the extent to which UV radiation affects capacitor aging. These parameters collectively determine the degradation rate of the IGBT and capacitor, thus influencing the overall health evolution of the device.

[0061] ③ Physical residuals:

[0062]

[0063] Physical residuals are an important indicator for measuring the deviation between a health state evolution model and physical laws. Based on the definition of health indicators and degradation rate modeling, ideally, the rate of change of health indicators should be equal to -(Γ). IGBT +Γ CapTherefore, physical residuals are used to assess the physical plausibility of a model by calculating the difference between the actual rate of change of health indicators predicted by the model and this ideal value. A large physical residual indicates a significant deviation between the model's predictions and physical laws, requiring model correction and optimization. During model training, minimizing physical residuals ensures consistency between the data-driven model and the physical model, improving the reliability and accuracy of predictions.

[0064] Furthermore, step S4, Bayesian-PINN fusion modeling, specifically includes:

[0065] (1) Network Architecture

[0066] The input layer receives time t and feature vector x as input, which are processed by Bayesian hidden layer 1 and Bayesian hidden layer 2. The output layer outputs the health indicator u(t). The weights w of the Bayesian hidden layers follow a normal distribution. The output health indicator u(t) is input into the physical constraint module, and the prediction results are combined with the physical constraints to serve as the loss function for model training.

[0067] (2) Loss Function

[0068]

[0069] In the formula,

[0070] ① Data fitting terms: The data fit term is used to measure the health indicator u predicted by the model. pred Compared with real health indicators u true The difference between them. By minimizing this error, the model can accurately fit the existing data. Here, N is the number of data samples. Taking the average value is to eliminate the influence of the amount of data on the loss function, making the magnitude of the loss function independent of the amount of data, which facilitates comparison and optimization between datasets of different sizes.

[0071] ②Physical constraints: The physical constraints ensure that the health state evolution predicted by the model conforms to physical laws. ‖PDE(u)‖ 2 It is the squared norm of the physical residual, reflecting the deviation between the model's predictions and the physical model. Minimizing this part of the loss encourages the model to learn healthy state evolution patterns that conform to physical laws during training. ReLU(u t+1 -u tThe physical constraint term is used to prevent non-physical increases in health indicators. Under normal circumstances, the health status of equipment should monotonically decrease or remain constant over time; an increase would violate physical principles. The Modified Linear Unit Function (ReLU) is used to penalize such unreasonable changes, ensuring that the trend of health indicator changes conforms to physical reality. Weighting coefficients β1 and β2 are used to balance the contributions of the two parts in the physical constraint term; their values ​​are adjusted according to the actual problem and model training to achieve the optimal physical constraint effect.

[0072] ③ Bayesian regularization term: The Bayesian regularization term is used to quantify the uncertainty of model parameters and prevent overfitting. The prior distribution p(w) is usually assumed to be a simple normal distribution, and its calculation formula is:

[0073]

[0074] In the formula, μ p It is the mean vector of the prior distribution, Σ p It is the covariance matrix of the prior distribution.

[0075] The posterior distribution q(w) is an update of the prior distribution, taking into account the distribution after observing the data. Its calculation formula is:

[0076]

[0077] In the formula, μ q It is the mean vector of the posterior distribution, Σ q It is the covariance matrix of the posterior distribution.

[0078] Bayesian regularization terms By measuring the difference between the posterior and prior distributions of model parameters, it is ensured that the model fully utilizes prior knowledge during training, avoiding overfitting of model parameters to training data. In equipment health status assessment, the prior distribution pre-determines a reasonable range for parameters based on physical models or engineering experience. The posterior distribution is adjusted during training based on actual data, reflecting the most likely values ​​of model parameters and their uncertainties under given data. By minimizing the Bayesian regularization term, the model can find a balance between data-driven approaches and physical constraints, ensuring that predictions not only conform to observed data but also align with the actual physical behavior of the equipment. This helps improve the model's generalization ability and reliability in few-shot learning and uncertainty quantification.

[0079] (3) Quantification of uncertainty

[0080] ① Monte Carlo sampling:

[0081]

[0082] In the formula, u(i) (t) represents the predicted health indicator value corresponding to the i-th sampling; x is the input feature vector; w (i) Let represent the model parameters for the i-th sampling; q(w) is the posterior distribution of the model parameters; M is the number of samplings, which is set to 100 here.

[0083] By sampling multiple sets of model parameters w from the posterior distribution q(w) (i) By using a Bayesian neural network (BNN) to make multiple predictions on the input feature x, multiple health indicators u are obtained. (i) (t). The physical significance of this step lies in quantifying the impact of model parameter uncertainty on the prediction results. By sampling the possible range of parameter values ​​multiple times, a basis is provided for subsequent uncertainty analysis.

[0084] ② Statistical calculation: mean Standard deviation

[0085] In the formula, σ is the mean of the predicted values ​​of health indicators. u (t) represents the standard deviation of the predicted values ​​of the health indicator.

[0086] Calculate the mean and standard deviation of multiple predictions. The mean reflects the central trend of the prediction, while the standard deviation quantifies the degree of uncertainty. A larger standard deviation indicates greater uncertainty in the model's prediction of health indicators, which may be related to uncertainty in model parameters, data noise, or limitations in the model structure. These two statistics can concisely describe the distribution characteristics of the prediction results, providing a basis for subsequent confidence interval estimation and decision support.

[0087] ③ Confidence interval:

[0088]

[0089] In the formula, CI 95 (t) represents the 95% confidence interval; 1.96 is the critical value for the normal distribution corresponding to the 95% confidence level.

[0090] A 95% confidence interval is constructed based on the mean and standard deviation, indicating that, given multiple sampling and predictions, there is a 95% probability that the true health indicator u(t) falls within this interval. The confidence interval provides a range of uncertainty for RUL predictions, helping to assess the reliability of the prediction results. A narrower confidence interval indicates a more reliable prediction, while a wider confidence interval suggests greater uncertainty, requiring more data or further model optimization.

[0091] Uncertainty quantification results are crucial for subsequent Remaining Life (RUL) prediction and decision support. On one hand, by providing confidence intervals for health indicators, the remaining useful life (RUL) of equipment can be estimated more accurately, rather than using a single point estimate, thus providing more comprehensive information for operational decisions. For example, in the decision-making logic, when the confidence interval width exceeds a certain threshold, a manual inspection is triggered to address significant uncertainty. On the other hand, uncertainty quantification helps assess model reliability. When model predictions show high uncertainty, the model can be adjusted promptly or more data can be collected to improve prediction accuracy, enhancing the robustness and credibility of the entire health status assessment system.

[0092] Furthermore, the S5RUL prediction process specifically includes:

[0093] (1) Conversion from health indicator u-value to RUL

[0094] ① Failure time calculation:

[0095] In the formula, t fail This indicates the estimated time when the equipment will reach its failure threshold; t is a time variable. σ is the mean of the predicted values ​​of the health indicators. u (t) represents the standard deviation of the predicted values ​​of the health indicators; 0.2 is the pre-set failure threshold for the health indicators.

[0096] The potential timeframe for equipment failure is determined by using the mean and standard deviation of health indicators. Taking into account the uncertainty of prediction, the earliest and latest failure times are given, reflecting the failure risk of the equipment at different confidence levels. This provides a basis for subsequent calculations of RUL (Remaining Operating Time), clarifies the remaining operating time of the equipment under different conditions, and helps in developing reasonable maintenance plans.

[0097] ②Basic RUL calculation: RUL = t fail -t current ,

[0098] In the formula, RUL represents the remaining useful life; t fail t represents the estimated failure time. current This is the current time.

[0099] The basic RUL is the expected length of time that a device will continue to operate normally under its current health and operating conditions. Confidence interval CI RUL This quantifies the uncertainty range of RUL, reflecting the reliability of the prediction results. It provides crucial information for equipment maintenance decisions, helping to determine the optimal maintenance timing, avoiding both resource waste caused by premature maintenance and equipment failure caused by delayed maintenance.

[0100] (2) Environmental adaptive correction

[0101] ① Calculation of environmental factors:

[0102] In the formula, Λ env It is a comprehensive environmental factor; UV is ultraviolet radiation intensity; ΔT day denoted by , where is the diurnal temperature range; and 'h' represents altitude. This formula comprehensively considers the impacts of UV intensity, diurnal temperature range, and altitude on equipment lifespan, quantifying the influence of different environmental factors into a single comprehensive factor. Specifically, UV radiation accelerates material aging, diurnal temperature range causes thermal fatigue, and altitude affects heat dissipation and insulation performance. It can be used to correct the baseline RUL (Range Limiting Usage) to better reflect actual equipment lifespan under real-world environmental conditions, thereby improving prediction accuracy.

[0103] Notes: ① Ultraviolet intensity: UV: Ultraviolet radiation intensity (unit: W / m²) 2 Denominator 100: Baseline value for high-altitude ultraviolet radiation, 100 W / m 2 This refers to the high UV threshold for high-altitude photovoltaic systems as defined by IEC 62446-3. High-altitude characteristics: Summer UV intensity in the Qinghai and Tibetan plateaus can reach 120-150 W / m². 2 Coefficient 0.4: The weight of ultraviolet radiation in the total environmental stress (40%), based on the contribution rate analysis of ultraviolet radiation in the aging study of photovoltaic backsheets in high-altitude areas.

[0104] ② Daily average temperature difference: ΔT day : Maximum and minimum temperature difference within 24 hours (unit: °C), denominator 30: Typical diurnal temperature range benchmark on the plateau, 30 °C is the plateau thermal cycling stress threshold defined by GB / T 36545, the average daily temperature difference on the Qinghai-Tibet Plateau is generally in the range of 20-35 °C. Coefficient 0.3: Weight of thermal shock in total stress (30%), based on the fact that IGBT module thermal fatigue tests show that temperature difference contributes 32% of the main cause of failure.

[0105] ③ Altitude item: h: Altitude (unit: meters), denominator 3000: The starting altitude benchmark for high-altitude environments; 3000m is the boundary line for high-altitude electrical equipment defined by the IEC 62804 standard. Coefficient 0.2: The weight of altitude in total stress (20%), based on the direct contribution rate of altitude of 18-22% in the failure statistics of high-altitude photovoltaic power plants.

[0106] ②RUL calibration:

[0107] In the formula, RUL 校正 RUL is the corrected remaining service life; RUL is the base remaining service life. It is the corrected RUL confidence interval; CI RUL It is the basic RUL confidence interval.

[0108] By incorporating comprehensive environmental factors, the baseline RUL (Remaining Lifespan) is corrected. Considering that high-altitude environments accelerate equipment aging, the corrected RUL more accurately reflects the remaining service life of the equipment under actual conditions. This allows the prediction results to better adapt to special environments such as high altitudes, avoiding prediction biases caused by environmental factors, and providing a more accurate basis for maintenance decisions in complex environments.

[0109] ③ Plateau expansion:

[0110] In the formula, σ u′ (t) is the corrected standard deviation of the health index; σ u (t) represents the standard deviation of the original health indicators; h represents the altitude.

[0111] In high-altitude environments, factors such as low air pressure increase the uncertainty of health indicators. This method corrects for the standard deviation by considering the impact of altitude on uncertainty, reflecting the changes in uncertainty in prediction results under high-altitude conditions. This provides more accurate uncertainty quantification results for subsequent uncertainty analysis and decision support, helping to more comprehensively assess equipment health status and remaining service life.

[0112] (3) Decision-making logic

[0113] The decision-making logic is as follows:

[0114] ① Normal state: When u > 0.6 and RUL > 1000h and σ total When the value is <0.05, the decision-making action is routine monitoring.

[0115] ② Warning status: When 0.3 < u ≤ 0.6 and 500h < RUL ≤ 1000h and σ total When the value is less than 0.15, the decision action is to prepare for maintenance.

[0116] ③ Emergency state: When u≤0.3 and RUL≤500h, the decision action is to shut down immediately.

[0117] ④ Confidence anomaly: When CI width > 0.3 × RUL, the decision action is manual inspection.

[0118] In the formula, u is the health indicator, RUL is the remaining useful life, and σ total The total standard deviation is represented by the CI width, which is the confidence interval width.

[0119] Furthermore, step S6, the plateau environment adaptation mechanism, specifically includes:

[0120] (1) Dynamic adjustment strategy

[0121] The formula for the dynamic adjustment strategy is as follows:

[0122] ① Adjustment of physical constraint weights: Where α represents the physical constraint weight, α0 represents the initial physical constraint weight, and h represents the altitude. As the altitude increases, the physical constraint weight is increased to strengthen the role of physical constraints in the model, making the model more focused on physical laws to cope with the complexity of the impact of the high-altitude environment on the equipment.

[0123] ②KL regularization weight adjustment: Where γ is the KL regularization weight, γ0 is the initial KL regularization weight, and h is the altitude. Appropriately reducing the KL regularization weight reduces the degree of restriction imposed on the model by regularization, avoids the model being too conservative due to the special characteristics of the plateau environment, and improves the model's flexibility and adaptability.

[0124] ③ Feature set correction: Where, σ T′ σ is the corrected characteristic standard deviation. T denoted as the original feature standard deviation, and h as the altitude. This compensates for the influence of factors such as low air pressure on the features, improving their representational ability in high-altitude environments.

[0125] ④ Increased degradation rate: Where Γ′ represents the enhanced degradation rate, Γ represents the original degradation rate, and h represents the altitude. Considering the accelerated degradation characteristics of equipment in high-altitude environments, the degradation rate is enhanced to make the model more accurately reflect the lifespan consumption of equipment in high-altitude environments.

[0126] (2) Seasonal strategy

[0127] Summer: UV radiation weighting 50%, heat shock weighting 30%.

[0128] Winter: UV radiation weighting 30%, heat shock weighting 50%.

[0129] Spring and autumn: UV radiation weighting 40%, heat shock weighting 40%.

[0130] A reliability monitoring system for photovoltaic converters in special high-altitude environments includes:

[0131] Data acquisition module: used to collect electrical, temperature, environmental, and operational data of the plateau converter;

[0132] Data preprocessing module: used to handle outliers, select stable time periods, and align signals in the collected data;

[0133] Feature engineering module: used to extract plateau-sensitive features and perform physical consistency checks and feature selection;

[0134] Physical constraint modeling module: used to build physical models and health state evolution equations, and calculate physical residuals;

[0135] Bayesian-PINN fusion modeling module: used to build fusion models, set loss functions, and quantify uncertainties;

[0136] RUL Prediction Module: Used to predict remaining useful life and perform environmental corrections to make maintenance decisions;

[0137] Plateau Environment Adaptive Module: Used to dynamically adjust model parameters and feature weights based on altitude and season.

[0138] Furthermore, the data acquisition module includes a Hall sensor, a thermocouple, and a weather station interface, used to collect electrical quantity, temperature quantity, and environmental quantity data, respectively. The Bayesian-PINN fusion modeling module is implemented using a deep learning framework and has variational inference capabilities to estimate the posterior distribution parameters of the Bayesian hidden layer weights.

[0139] The specific project implementation guidelines are as follows:

[0140] (1) System Deployment

[0141] After the on-site sensors collect data, it is transmitted to the edge computing node for preliminary processing, then features are extracted, and the feature data is sent to the cloud model service for RUL prediction. Finally, the prediction results are fed back to the operation and maintenance system.

[0142] (2) Parameter calibration

[0143] The parameter calibration information is as follows:

[0144] E a,IGBT The calibration range is 90-110 kJ / mol, and the calibration method is accelerated aging test. Among them, E... a,IGBT This is the activation energy of the IGBT. Activation energy is a physical quantity that represents the energy barrier that atoms need to overcome during diffusion or migration.

[0145] The β calibration range is 5.2-6.3, and the calibration method is thermal cycling test. Here, β is a material constant in the IGBT thermal fatigue model, reflecting the thermomechanical properties of the material.

[0146] The calibration range for n is 3.0-3.5, and the calibration method is a voltage stress experiment. Here, n is the aging exponent in the capacitor aging model, which describes the influence of voltage on the capacitor aging rate.

[0147] u threshold The calibration range is 0.15-0.25, and the calibration method is historical fault data analysis. Wherein, u threshold This is the failure threshold for health indicators, used to determine whether the equipment has reached a failure state.

[0148] (3) Maintenance strategy

[0149] The maintenance strategy is as follows:

[0150] When RUL > 2000h, an annual routine inspection should be performed. RUL stands for Remaining Service Life, indicating the expected time the equipment will continue to operate normally in its current condition.

[0151] When 1000h < RUL ≤ 2000h, a quarterly inspection shall be conducted.

[0152] When 500h < RUL ≤ 1000h, conduct monthly inspections and prepare spare parts.

[0153] When RUL≤500h, conduct weekly inspections and develop a shutdown plan.

[0154] The beneficial effects of this invention are as follows:

[0155] This invention significantly improves the accuracy of photovoltaic converter life prediction in high-altitude environments, effectively reduces operation and maintenance costs, and enhances equipment reliability.

[0156] Specifically,

[0157] 1. Through multi-dimensional data collection and refined preprocessing, the data quality and accuracy are significantly improved, providing a data foundation with high signal-to-noise ratio and high synchronization for status monitoring, and effectively reducing the interference of data noise on monitoring results.

[0158] 2. Construct a system containing multiple sensitive features and optimize feature quality to accurately capture key information on equipment health status in high-altitude environments, and significantly enhance the feature's ability to characterize equipment degradation processes.

[0159] 3. By integrating the IGBT thermal fatigue equation and the capacitor aging equation to establish a health index evolution model, and strictly following the physical laws to constrain the model prediction, the model's ability to explain and predict the health status of equipment is significantly improved.

[0160] 4. Construct a network architecture that integrates data fitting, physical constraints, and Bayesian regularization to achieve deep coupling between data-driven and physical models, significantly improving the model's generalization ability and prediction accuracy under conditions of small samples and high noise.

[0161] 5. By considering the uncertainty of prediction and introducing an environmental adaptive correction mechanism, the prediction results are dynamically adjusted to adapt to complex environmental changes, which significantly improves the reliability and accuracy of remaining service life prediction and provides a scientific basis for maintenance decisions.

[0162] 6. The system automatically adjusts model parameters and feature weights based on altitude and seasonal changes, significantly enhancing the monitoring system's adaptability to the special environment of the plateau and ensuring that the system can maintain high-performance monitoring results under different environmental conditions. Attached Figure Description

[0163] Figure 1 Overall method implementation flowchart.

[0164] Figure 2 Overall methodology implementation technology roadmap.

[0165] Figure 3 Data acquisition and preprocessing flowchart.

[0166] Figure 4 Flowchart of engineering process for high-altitude sensitive features.

[0167] Figure 5 Bayesian-PINN fusion modeling flowchart.

[0168] Figure 6 RUL Prediction and Decision-Making Flowchart. Detailed Implementation

[0169] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0170] As shown in the figure, this invention provides a method for monitoring the reliability of photovoltaic converters in special high-altitude environments, comprising the following steps:

[0171] 1. Data Acquisition and Preprocessing Implementation

[0172] (1) Sensor selection and deployment

[0173] Select an appropriate sensor based on the different characteristics of electrical quantities, temperature quantities, environmental quantities, and operational quantities. For DC bus voltage (V... dc ) and AC output current (I ac The acquisition of electrical quantities should employ a high-precision Hall sensor with a high sampling frequency (10kHz) response capability, accurately capturing rapid changes in electrical quantities, and possessing good linearity and stability to ensure the reliability of the acquired data.

[0174] For IGBT junction temperature (T) j ) and radiator temperature (T hs The measurement uses thermocouples, which have the characteristics of fast response time and high accuracy, and can adapt to the real-time monitoring needs of temperature changes. The sampling frequency is 1Hz, which meets the requirements of dynamic temperature tracking. At the same time, its temperature measurement range covers the normal operation of the equipment and possible high-temperature conditions, ensuring the validity of the data.

[0175] Environmental data such as ultraviolet radiation intensity (UV), altitude (h), and air pressure (P) are provided by the meteorological station. The meteorological station is equipped with professional environmental monitoring sensors, which have high-precision and high-reliability data acquisition capabilities, with a sampling frequency of 0.1Hz, and can stably acquire environmental information within a certain period of time.

[0176] Output power (P) in operation loss ) and switching frequency (f sw The data comes from the SCADA system, which is connected to the equipment's control system. Through its built-in power monitoring module and switching frequency counter, the SCADA system can accurately collect these operating parameters at a sampling frequency of 1kHz, ensuring real-time monitoring of the equipment's operating status.

[0177] (2) Data Acquisition System Setup

[0178] A distributed data acquisition system is constructed, connecting various sensors to data acquisition cards. These cards feature multi-channel, high-precision A / D conversion capabilities, enabling simultaneous acquisition of multiple electrical signals while ensuring the stability and synchronization of the sampling frequency. Furthermore, the acquisition cards are optimized for temperature signals, incorporating cold junction compensation and filtering functions to improve the accuracy of temperature data acquisition. Environmental and operational data are accessed through corresponding interface modules. All acquisition cards are connected to the main control computer via an industrial bus, enabling centralized data acquisition and transmission.

[0179] (3) Implementation of outlier handling algorithm

[0180] Write an outlier handling program based on the 3σ criterion. First, calculate the mean (μ) and standard deviation (σ) of each data set. For each data point x, determine whether it satisfies |x-μ|>3σ. If it does, it is identified as an outlier and removed.

[0181] For handling the power surge segment, the power change rate is calculated, i.e., by adjusting the output power (P) loss The power change rate is obtained by numerically differentiating the data. Then it is compared with 0.1 times the rated power (0.1P). rated The data is compared, and if the threshold is exceeded, the data segment is discarded. In the implementation process, a sliding window method can be used to calculate the power change rate. The window length can be selected according to the sampling frequency and power change characteristics of the actual data. For example, for power data with a sampling frequency of 1kHz, a sliding window of 10 sampling points can be selected to calculate the power change rate. This can smooth out the influence of noise and accurately detect power fluctuation segments.

[0182] After removing outliers, the data is interpolated to ensure the continuity of the data over time. Methods such as linear interpolation or spline interpolation can be used. The appropriate interpolation algorithm is selected according to the characteristics of the data and the extent of missing data. For example, linear interpolation can quickly fill in data gaps for occasional single outliers, while spline interpolation can provide a smoother transition for data gaps over a longer period of time, ensuring that the interpolated data can reflect the actual operating status of the equipment.

[0183] (4) Stable period selection strategy

[0184] Write a program to determine light intensity, extract light intensity information from environmental data provided by a weather station, and set a threshold of 800 W / m². 2 When the light intensity remains above the threshold for 10 consecutive minutes (based on true solar time), it is marked as the starting point of a possible stable period, and the corresponding start timestamp is recorded.

[0185] Calculate the power fluctuation rate within a 10-minute time window, for the output power (P) loss The data is statistically analyzed to calculate the ratio of its standard deviation to its mean, i.e., the power fluctuation rate. When the power fluctuation rate is less than 5%, combined with the judgment result of the light intensity, the period is determined to be a stable period, and the data of the period is extracted for subsequent feature engineering analysis. If the power fluctuation rate requirement is not met, the sliding time window is continued to make the next judgment until a stable period that meets the conditions is found.

[0186] The time range of the data for the stable period is limited to 10:00-14:00 (true solar time). This is based on the operating patterns and environmental characteristics of solar energy equipment. During this period, the equipment is usually in a relatively stable operating state and is less affected by external interference, so the collected data is more representative. By comparing with the timestamp information, the stable period data that meets the time range requirements is selected as the basic dataset for subsequent feature extraction and analysis.

[0187] (5) Application of signal alignment algorithm

[0188] Dynamic Time Warping (DTW) algorithm is used to timestamp and calibrate environmental and electrical data. First, the two sets of data are represented as two time series, for example, the environmental data series is [a1, a2, ..., a...]. n The electrical data sequence is [b1, b2, ..., b n ], where n and m are the number of sampling points for the two sets of data, and n ≠ m due to the different sampling frequencies.

[0189] Define a distance matrix d(i,j), where d(i,j) represents the distance between the i-th sampling point of the environmental data and the j-th sampling point of the electrical data. This distance can be calculated using Euclidean distance or other suitable distance metrics. For example, d(i,j) = |a i -b j Then, according to the recursive formula of the DTW algorithm, the cumulative distance matrix is ​​calculated, and the path that minimizes the cumulative distance is found. This path is the optimal time alignment path between the two sets of data.

[0190] Based on the optimal time alignment path, environmental and electrical data are interpolated or resampled to ensure they have the same timestamp sequence, thereby achieving signal alignment. The interpolation method can be selected according to the data characteristics, such as linear interpolation or polynomial interpolation, to ensure that the aligned data can accurately reflect the operating status of the equipment under different environmental conditions. This provides accurate time correspondence for subsequent feature extraction and analysis, improving the quality and reliability of feature engineering.

[0191] 2. Implementation of projects addressing the sensitive characteristics of high-altitude areas

[0192] (1) Voltage ripple rate calculation

[0193] From the preprocessed DC bus voltage data (V dc In this context, the root mean square (RMS) value is calculated according to the definition formula for the root mean square value, i.e.: Where N is the number of sampling points for voltage data. The average value of the voltage data is used to quantify the degree of voltage fluctuation by calculating the root mean square value, reflecting the filtering effect of the capacitor.

[0194] Obtain the altitude (h) information of the current device location. This information can be obtained from environmental data provided by a weather station or in real time via a GPS positioning module. Substitute this information into the formula for calculating the voltage ripple rate: The voltage ripple rate R is calculated. r This characteristic value can comprehensively reflect the aging degree of the capacitor and the impact of the high-altitude low-pressure environment on the capacitor performance, providing key information for subsequent health status assessment.

[0195] (2) Calculation of high-frequency harmonic energy ratio

[0196] Fourier transform can be performed on the current or voltage signals acquired during the IGBT switching process. The Fast Fourier Transform (FFT) algorithm can be used to convert the time-domain signal into a frequency-domain signal to obtain its power spectral density (PSD(f)).

[0197] Determine the switching frequency (f) swThe switching frequency and its harmonics range are determined based on the equipment specifications or through spectrum analysis. Typically, the switching frequency is between several hundred Hz and several kHz. For example, for a photovoltaic inverter, if its switching frequency is 20kHz, then the high-frequency range is defined as 2 to 5 times the switching frequency, i.e., 40kHz to 100kHz, and the low-frequency range is defined as 0 to the switching frequency, i.e., 0 to 20kHz.

[0198] To calculate the harmonic energy in the high-frequency and low-frequency bands, the power spectral density is integrated over the high-frequency and low-frequency bands respectively. This can be achieved using numerical integration methods, such as the trapezoidal rule or Simpson's rule, to calculate the high-frequency harmonic energy. and low-frequency harmonic energy Then follow the formula: Calculate the high-frequency harmonic energy ratio E h This feature can effectively reflect the degradation of IGBT switching characteristics, providing an important basis for assessing the health status of IGBTs.

[0199] (3) Calculation of junction temperature fluctuation entropy

[0200] IGBT junction temperature data (T) collected from temperature sensors j In the statistical junction temperature probability distribution p(T), j The junction temperature data can be divided into several intervals using histogram methods or kernel density estimation methods. The frequency of data occurrence in each interval can be counted to obtain the probability distribution function of the junction temperature.

[0201] Calculate the junction temperature fluctuation entropy using the formula: Perform the calculation, where P loss0 For reference power, P loss The actual output power is used. The reference power can be determined based on the rated power of the equipment. For example, for a device with a rated power of 10kW, P can be taken as the reference power. loss0 With a power rating of 10kW, the actual output power is obtained from the SCADA system. By calculating the junction temperature fluctuation entropy, the complexity of the junction temperature fluctuation can be quantified, reflecting the thermal fatigue damage and providing key indicators for the health status assessment of the equipment.

[0202] (4) Thermal response time calculation

[0203] Obtain the IGBT junction temperature change function over time (T) j (t) and the function of output power over time (P) loss (t)) The corresponding data sequences are collected from the temperature sensor and the SCADA system respectively, and converted into time series function form. In the data analysis software, time series analysis tools can be used to preprocess the data, such as removing noise and smoothing, to improve the data quality.

[0204] Calculate the correlation between junction temperature and output power using the Pearson correlation coefficient or other correlation indices to determine T. j (t) and P loss (t) The correlation at different time lags is used to find the time point t with the largest rate of change in correlation, i.e.: This point in time is called the thermal response time. By analyzing the thermal response time, we can assess the speed at which the heat dissipation system responds to changes in power and determine the performance status of the heat dissipation system.

[0205] Considering the influence of altitude on thermal response time, the calculated thermal response time is multiplied by the altitude correction factor e. 0.0001h The final thermal response time τ is obtained. th The altitude data is obtained from the weather station. This correction method can more accurately reflect the impact of the plateau environment on the performance of the heat dissipation system, providing more reliable data support for equipment maintenance decisions.

[0206] (5) Calculation of thermal shock coefficient

[0207] The diurnal temperature range (ΔT) is obtained from the weather station. day The weather station provides daily high and low temperatures, and the difference between these two values ​​is used to calculate the diurnal temperature range. It also records the number of thermal cycles (N). cycle The number of thermal cycles can be counted based on the equipment's operating time and the local diurnal temperature variation. For example, for equipment operating outdoors for extended periods, each diurnal cycle can be counted as one thermal cycle. Temperature change curves are recorded by a temperature recorder installed at the equipment site, and the number of thermal cycles is automatically counted. According to the formula:

[0208] Calculate the thermal shock coefficient K ts This feature comprehensively considers the impact of diurnal temperature variation and thermal cycling on the equipment, and can effectively quantify the degree of damage caused by thermal shock to the equipment, providing an important reference for equipment life prediction and maintenance decisions.

[0209] (6) Calculation of comprehensive stress index

[0210] Collect ultraviolet intensity (UV) and thermal shock coefficient (K). ts The data includes altitude (h), where ultraviolet radiation intensity is obtained from a weather station, and the thermal shock coefficient is calculated using the steps described above. Altitude data also comes from a weather station or a GPS positioning module. According to the formula: The comprehensive stress index Λ is calculated by weighting and summing three factors: ultraviolet intensity, thermal shock coefficient, and altitude. This comprehensive stress index can fully reflect the total damage to equipment caused by the plateau environment. This index provides a comprehensive environmental factor consideration for subsequent environmental adaptive correction and equipment health status assessment, which helps to improve the accuracy and reliability of prediction results.

[0211] (7) Physical consistency test and feature selection

[0212] To verify whether the change in ripple rate of the capacitor with temperature conforms to the temperature characteristics of the capacitor, i.e., to verify... Where T c To determine the capacitor case temperature, voltage ripple rate data at different temperatures are collected to establish R0. r For T c The scatter plot is generated, and linear regression analysis is performed. The sign of the regression coefficient is observed. If the regression coefficient is negative, it indicates that the ripple rate decreases with increasing temperature, which is consistent with the temperature characteristics of a capacitor. If not, the feature extraction process needs to be checked and corrected to ensure the physical rationality and reliability of the features.

[0213] Verify the degree of interference of ambient temperature on junction temperature fluctuation entropy, i.e., verify |corr(S) T ,T a )|<0.3, where T a Calculate the junction temperature fluctuation entropy S for ambient temperature. T With ambient temperature T a If the absolute value of the correlation is greater than 0.3, it indicates that the ambient temperature has a significant impact on the junction temperature fluctuation entropy. Further analysis of the reasons and corresponding corrective measures are needed. For example, consider introducing an ambient temperature compensation term during feature extraction to improve the accuracy and representativeness of the features.

[0214] A three-stage feature selection process is employed. First, the variance of each feature is calculated, and features with variances below a threshold are removed. For example, if a feature's value varies little across all samples and its variance is below the threshold, it is considered to have weak discriminative power regarding equipment health status and can be removed from the feature set. Next, the remaining features are screened based on physical laws and professional knowledge, retaining those that conform to physical laws. For instance, features unrelated to equipment aging mechanisms or contradicting the physical model are removed. Finally, the Recursive Feature Elimination (RFE) algorithm is used to rank and select features based on the predictive performance of the Bayesian-PINN model, progressively eliminating features that contribute the least to the model's prediction until the optimal feature subset is obtained. Cross-validation is then used to determine the optimal number and combination of features, ultimately yielding the selected feature vector: x = [R r E h ,S T ,τ th,Λ,K ts ] T This provides concise and effective feature inputs for subsequent modeling and prediction.

[0215] 3. Physical constraint modeling and implementation

[0216] (1) Determination of parameters of core physics equations

[0217] IGBT thermal fatigue model: The range of values ​​for each parameter was determined experimentally. The constant A was determined by referring to reliability data and thermal fatigue life test results provided by the IGBT manufacturer, combined with accelerated life testing methods, and by calibrating multiple sets of different junction temperature changes (ΔT). j ) and average temperature (T) m Under certain conditions, IGBTs underwent life testing, and their failure time (N) was recorded. f The range of constant A is determined through nonlinear regression fitting. The material constant β is calibrated by performing thermal cycling tests on IGBT samples with varying junction temperature changes, recording the number of failure cycles, and fitting the test data to obtain the value of β. Typically, β ranges from 5.2 to 6.3 and is closely related to the thermomechanical properties of the material. Activation energy E a Calculation: By analyzing the failure mechanism and physicochemical processes of IGBTs, and combining the Arrhenius equation, the activation energy E was determined based on experimental data. a The value range is 90-110 kJ / mol.

[0218] For the capacitor aging model: Similarly, the parameters are determined through experiments and data analysis. Determination of the reference life L0: Based on data provided by the capacitor manufacturer and relevant standard test methods, long-term aging tests are conducted on the capacitor. Under standard temperature (e.g., 25℃) and rated voltage conditions, the failure time of the capacitor is recorded to determine its reference life L0. Calibration of the aging index n: A voltage stress test is used to apply different multiples of the rated voltage (e.g., 1.1 times, 1.2 times, etc.) to the capacitor, and its failure time is recorded. The value of the aging index n is obtained by model fitting. Typically, the value of n ranges from 3.0 to 3.5. For example, through experimental data analysis, the value of n is 3.2. The aging index reflects the degree of influence of voltage on the aging rate of the capacitor; that is, the higher the voltage, the faster the capacitor ages, and the relationship is exponential. Capacitor activation energy E a Determination: Combining the material properties and aging mechanism of the capacitor, accelerated aging tests were conducted at different temperatures, and the failure time of the capacitor was recorded. The E value was obtained by fitting the capacitor using the Arrhenius equation. a The value of .

[0219] (2) Construction of the health state evolution equation

[0220] The health indicator u(t) is defined to have a range of values ​​[0,1]. The initial time u(0) = 1 indicates that the equipment is brand new and undamaged, and the failure time u(t) = 1 indicates that the equipment is undamaged. fail A degradation rate of 0.2 indicates that the equipment has reached the end of its lifespan and requires maintenance or replacement. This definition is based on the equipment's failure criteria and engineering experience. Establish a degradation rate model:

[0221] ① Calculate the IGBT degradation rate: The specific values ​​of each parameter are calculated based on the calibration results. For example, C1 is determined according to the IGBT manufacturing process and material properties, and E... a,IGBT Take 100 kJ / mol, R as the gas constant 8.314 J / (mol·K), T m ΔT represents the average operating temperature of the IGBT (e.g., 80℃, which needs to be converted to Kelvin 353.15K). j For junction temperature changes (e.g., 30K), β is taken as 5.8, K. ts Let Γ be the thermal shock coefficient (e.g., 0.5). Substituting this value into the formula, we can obtain Γ. IGBT The value reflects the degradation rate of the IGBT under current operating conditions.

[0222] ② Calculate the capacitance degradation rate: For example, C2 is determined according to the capacitor's manufacturing process, V is the actual voltage the capacitor withstands (e.g., 400V), V0 is the reference voltage (e.g., 380V), n is taken as 3.2, and E... a,Cap Take 80 kJ / mol, T as the capacitor's operating temperature (e.g., 60℃, converted to 333.15 K), I uv Substituting the ultraviolet intensity index (e.g., 0.8) into the formula, we obtain Γ. Cap The value quantifies the aging rate of the capacitor under electrical stress, thermal stress, and ultraviolet radiation.

[0223] Γ IGBT and Γ Cap Add them together and take the negative value to get the rate of change of the health indicators. The evolution of health indicators over time is calculated using numerical integration methods (such as the Euler method or the Runge-Kutta method) to obtain the time series of u(t). This series can intuitively reflect the dynamic changes in the health status of the equipment, providing basic data for subsequent RUL prediction.

[0224] ③ Calculate the physical residual: By comparing the rate of change of health indicators predicted by the actual model Compared with the degradation rate (Γ) calculated based on the physical model IGBT +Γ CapThe difference between the parameters is used to obtain the physical residual. If the physical residual is large, it indicates that the model's prediction results deviate from the physical laws, requiring model adjustment and optimization. For example, checking the accuracy of parameter values ​​and the rationality of the model structure. By continuously reducing the physical residual, the Bayesian-PINN model can be ensured to conform to physical constraints while being data-driven, thereby improving the credibility and reliability of the prediction results. 4. Implementation of Bayesian-PINN Fusion Modeling

[0225] (1) Network architecture setup

[0226] A Bayesian-PINN model is constructed using a deep learning framework. The input layer of the model receives time t and feature vector x (including the six selected features mentioned above: R). r E h ,S T ,τ th ,Λ env ,K ts The number of neurons in the input layer is determined by the dimension of the feature vector.

[0227] A Bayesian hidden layer is constructed, and a variational inference method is used to implement the Bayesian neural network. A probabilistic layer is used instead of a traditional fully connected layer. Bayesian hidden layer 1 and Bayesian hidden layer 2 contain 128 neurons. The ReLU activation function is chosen to increase the model's expressive power. The weights w of the Bayesian hidden layer follow a normal distribution. During model training, the posterior distribution parameters (mean μ) of the weights are estimated using variational inference methods. w and variance ).

[0228] The output layer outputs a health metric u(t). The output layer uses a linear activation function to ensure the output value is within the range [0,1]. The output health metric u(t) is then input into the physical constraint module, where it is combined with the physical constraints and used as part of the loss function for model training. The physical constraint module calculates the time derivative of the health metric. And the degradation rate (Γ) calculated based on the physical model. IGBT +Γ Cap The physical residuals are obtained and used to construct the loss function for the physical constraint terms.

[0229] (2) Loss function setting and optimization

[0230] Loss function for data fitting term: Collect tagged historical data, including actual health indicator values ​​of the device at different time points (u true Health indicators (u) predicted by the model can be obtained through actual equipment maintenance records, laboratory tests, or simulation data generation. predCompare the actual value with the mean squared error (MSE), and take the average value to obtain the result.

[0231] Physical constraint loss function: Construction of the model. Calculation of the square norm of the physical residual ||PDE(u)||. 2 ,in: Calculate ReLU(u) t+1 -u t The first term iterates through adjacent time points t and t+1 in the time series data, calculating the change in health indicators u(t+1)-u(t). If the change is positive (i.e., the health indicator is rising), it is made non-zero using the ReLU function (e.g., taking the positive value itself); otherwise, it is zero. These non-zero values ​​are summed to obtain this part of the loss. This constraint term is used to prevent non-physical increases in health indicators and ensure that the health state evolution predicted by the model conforms to physical laws. The weight coefficients β1 and β2 can be determined based on experience or through hyperparameter search. For example, initially set β1 = 0.1 and β2 = 0.01, and then adjusted according to the model training effect.

[0232] Bayesian regularization loss function: The calculation is as follows: The prior distribution p(w) is assumed to be normally distributed, and the posterior distribution q(w) is determined by the variational inference parameters of the Bayesian hidden layer. In the deep learning framework, the provided KL divergence calculation function is used to calculate the posterior distribution q(w). This regularization term is used to quantify the uncertainty of model parameters and prevent overfitting, which is especially important in small-sample learning scenarios.

[0233] Adding the three loss functions together, we get the total loss function: The model is trained by selecting an appropriate optimization algorithm (such as the Adam optimizer), and the model parameters (including the weight posterior distribution parameters of the Bayesian hidden layers and other network parameters) are adjusted to minimize the total loss function. During training, the decreasing trend of each loss and the prediction performance on the validation set, such as prediction accuracy and uncertainty quantification effect, are monitored. Hyperparameters (such as α, γ, β1, β2, etc.) are adjusted as needed to obtain the best model performance.

[0234] (3) Quantification of uncertainty

[0235] Monte Carlo sampling: sampling multiple sets of model parameters w from the posterior distribution q(w). i (i = 1, 2, ..., M, M = 100), in a deep learning framework, this can be achieved using the sampling function of a Bayesian neural network, for each set of sampling parameters w. i By using a Bayesian neural network (BNN) to perform forward propagation on the input feature vector x, the corresponding predicted health indicator value u is obtained.i (t) was sampled 100 times to obtain 100 predicted values. These predicted values ​​reflect the impact of the uncertainty of the model parameters on the prediction results. By simulating the possible range of parameter values ​​through multiple samplings, basic data is provided for subsequent uncertainty analysis.

[0236] Statistical calculation: Based on the predicted value u obtained from sampling i (t), calculate the mean: Sum of standard deviation: The mean reflects the central trend of the prediction, that is, the average predicted value of the health indicator after considering the uncertainty of the parameters. The standard deviation quantifies the degree of uncertainty of the prediction. The larger the standard deviation, the more uncertain the model's prediction of the health indicator is, which may be affected by data noise, model parameter uncertainty or model structure limitations. These two statistics can concisely describe the distribution characteristics of the prediction results and provide a basis for subsequent confidence interval estimation and decision support.

[0237] Confidence interval construction: Based on the mean and standard deviation, according to the formula: A 95% confidence interval is constructed, where 1.96 is the critical value of the normal distribution corresponding to the 95% confidence level. In practical applications, different confidence levels (such as 90% or 99%) can be selected as needed, and the critical value can be adjusted accordingly. This confidence interval provides an uncertainty range for RUL prediction, helping decision-makers assess the reliability of the prediction results. For example, when the confidence interval is narrow, the prediction results are more reliable, and a more accurate maintenance plan can be formulated accordingly; while when the confidence interval is wide, it indicates that there is greater uncertainty, which may require more data or further model optimization, or even trigger manual inspection to deal with potential equipment failure risks.

[0238] 5. Implementation of RUL Prediction Process

[0239] (1) Calculation of conversion of health indicator u-value to RUL

[0240] Failure time calculation: based on the average of health indicators and standard deviation σ u (t), calculate the estimated time for the equipment to reach the failure threshold: By iterating through the mean of health indicators in the time series, find the first one that satisfies... The time point t is the expected failure time. For example, in the predicted time series of health indicator mean values, it is found that when t = 1000 hours, Then t fail = 1000 hours. Similarly, calculate the earliest failure time considering uncertainties: And the latest expiration time: For example, if at t = 800 hours, but hours; at t = 1200 hours, but These failure times, calculated in hours, take into account the uncertainty of prediction and provide a range of possible equipment failure times, offering more comprehensive information for subsequent maintenance decisions.

[0241] Basic RUL calculation: Get the current time t current For example, assuming the current device has been running for 500 hours, then t current = 500 hours. Calculate the remaining useful life of the foundation: RUL = t fail -t current =1000-500=500 hours. This means that the equipment is expected to operate normally for another 500 hours without considering environmental factor corrections.

[0242] Calculate the confidence interval of the basic RUL: The confidence interval reflects the range of uncertainty of the underlying RUL (Remaining Usage Limit), which is that there is a 95% probability that the remaining useful life of the equipment is between 300 and 700 hours. This provides a risk assessment basis for maintenance decisions. For example, maintenance personnel can arrange maintenance plans reasonably based on this interval to ensure that maintenance operations are completed before equipment failure, while avoiding the waste of resources caused by premature maintenance.

[0243] (2) Implementation of environmental adaptive correction

[0244] Environmental factor calculation: Collection of ultraviolet radiation intensity (UV) and diurnal temperature range (ΔT) day And altitude (h) data, for example, assuming the UV intensity at the location of a device is 120 (units determined by the specific measuring instrument), the diurnal temperature range is 35℃, and the altitude is 3600 meters. Substitute into the formula: Calculations yielded the following:

[0245] This comprehensive environmental factor reflects the combined impact of ultraviolet radiation, diurnal temperature variation, and altitude on equipment lifespan under the current environment. The higher the value, the more severe the adverse impact of the environment on equipment lifespan, and the greater the need for correction of the basic RUL.

[0246] RUL Correction: Calculate the corrected remaining useful life based on the baseline RUL (e.g., 500 hours) and the comprehensive environmental factor (0.653). The actual remaining service life of the equipment is estimated to be 467.3 hours after taking into account the high-altitude environmental factors. This is because factors such as low air pressure in the high-altitude environment may have a complex impact on the aging process of the equipment.

[0247] Similarly, calculate the corrected RUL confidence interval: The corrected confidence interval also reflects the impact of environmental factors on the uncertainty of RUL, providing decision-makers with a more realistic range of equipment life prediction under actual environmental conditions, which helps to develop more reasonable and reliable maintenance strategies.

[0248] Plateau expansion: Obtain the altitude of the equipment location (e.g., 2000 meters), and calculate the corrected standard deviation of the health indicators based on the original standard deviation (assumed to be 0.05). This step takes into account the impact of factors such as low air pressure in high-altitude environments on the uncertainty of health indicator predictions. By adjusting the standard deviation, the additional uncertainty brought about by the high-altitude environment is quantified, providing a more accurate data foundation for subsequent uncertainty analysis and decision support. For example, when assessing the health status of equipment and developing maintenance plans, it is necessary to consider the factors that increase uncertainty to ensure the robustness and reliability of decisions.

[0249] (3) Implementation of decision-making logic

[0250] Write a decision logic program based on health indicators (u), remaining useful life (RUL), and total standard deviation (σ). total The current values ​​of the confidence interval width (CI width) and the confidence interval width are used to make decisions according to the following rules:

[0251] ① Normal state: When u > 0.6 and RUL > 1000h and σ total When the value is <0.05, the decision action is routine monitoring. At this time, the equipment is in good health, has a long remaining service life, and has low uncertainty. It is only necessary to carry out equipment inspection and data collection according to the normal monitoring frequency (such as monthly or quarterly) and record the equipment operating status. No additional maintenance operations are required.

[0252] ② Warning status: When 0.3 < u ≤ 0.6 and 500h < RUL ≤ 1000h and σ total When the value is less than 0.15, the decision action should be to prepare for maintenance. This indicates that the health status of the equipment has begun to decline, the remaining service life is gradually decreasing, and the uncertainty is within an acceptable range. At this time, preparations should begin for the materials and personnel required for maintenance, a maintenance plan should be formulated, and the monitoring frequency of the equipment should be increased (such as data collection and analysis every two weeks) in order to keep abreast of changes in the equipment status and ensure that maintenance work is completed before the equipment fails.

[0253] ③ Emergency State: When u≤0.3 and RUL≤500h, the decision action is to shut down immediately. This means that the health status of the equipment has seriously deteriorated and is about to face the risk of failure. The equipment must be stopped immediately for emergency maintenance or replacement to avoid more serious consequences that may be caused by equipment failure, such as equipment damage, safety accidents or power supply interruption. Before performing the shutdown operation, it should be ensured that necessary safety measures are taken and professional personnel should be arranged to handle the situation on site as soon as possible.

[0254] ④ Confidence Anomaly: When CI width > 0.3 × RUL, the decision action is manual inspection. This indicates that the uncertainty of the prediction result is too large, which may affect the reliability of the decision. At this time, maintenance personnel should be dispatched to the site to manually inspect the equipment, verify the actual operating status of the equipment, find the cause of the increased uncertainty, such as sensor failure, data transmission error, sudden environmental changes, etc., and take corresponding measures according to the site conditions, such as repairing the sensor, re-collecting data, adjusting model parameters, etc., to improve the prediction accuracy and the credibility of the decision.

[0255] In practical applications, the above decision-making logic is integrated into the equipment health status monitoring system, and prediction results (including u, RUL, σ) are obtained in real time. total The system automatically triggers corresponding decision-making actions based on indicators such as CI width, and generates maintenance work orders or alarm information, which are then sent to maintenance personnel to guide them in performing the corresponding operations. This achieves automation and intelligence in equipment maintenance decision-making, improving maintenance efficiency and equipment reliability.

[0256] 6. Implementation of Plateau Environment Adaptation Mechanism

[0257] (1) Implementation of dynamic adjustment strategy

[0258] Physical constraint weight adjustment: Based on the altitude (h) of the equipment location, according to the formula: Adjusting the physical constraint weight α, for example, setting the initial physical constraint weight α0 to 1.0, when the equipment is at an altitude of 3000 meters, we calculate α = 1.0 × (1 + 0.1 × 3000 / 3000) = 1.1. This shows that as the altitude increases, the physical constraint weight increases, strengthening the role of physical constraints in model training and prediction. This makes the Bayesian-PINN model pay more attention to physical laws to cope with the complexity of the impact of the plateau environment on the equipment, and improve the model's adaptability and prediction accuracy under plateau conditions.

[0259] KL regularization weight adjustment: Based on altitude h, use the formula: Adjusting the KL regularization weight γ, for example, if the initial KL regularization weight γ0 is 0.1, when the altitude is 3000 meters, we can calculate γ = 0.1 / (1 + 0.05 × 3000 / 3000) = 0.1 / 1.05 ≈ 0.095. Appropriately reducing the KL regularization weight reduces the degree of restriction imposed on the model by Bayesian regularization, avoids the model being too conservative due to the special characteristics of the plateau environment, improves the model's flexibility and adaptability, and enables it to better fit the equipment operation data in the plateau environment, while maintaining a certain generalization ability and preventing overfitting.

[0260] Feature set correction: based on the original feature standard deviation (σ) T And altitude (h), according to the formula: Correct the characteristic standard deviation, for example, the original characteristic standard deviation σ. T The value is 0.1. When the altitude is 2000 meters, the corrected characteristic standard deviation σ is calculated. T′ =0.1×(1+0.05×2000 / 1000)=0.1×1.1=0.11. In this way, the influence of factors such as low air pressure on the features is compensated, the feature's representation ability in the plateau environment is improved, and the features can accurately reflect the operating status of the equipment under plateau conditions, thereby providing more reliable data input for model training and prediction and improving model performance.

[0261] Enhanced degradation rate: Based on the original degradation rate (Γ) and altitude (h), using the formula: The degradation rate is enhanced. For example, if the original degradation rate is 0.01 / hour, the enhanced degradation rate Γ′ = 0.01 × (1 + 0.005 × 3000 / 1000) = 0.01 × 1.015 = 0.01015 / hour is calculated when the altitude is 3000 meters. Considering the characteristic of accelerated equipment degradation in high-altitude environments, by enhancing the degradation rate, the model can more accurately reflect the lifespan consumption of equipment in high-altitude environments, thereby improving the accuracy and reliability of RUL prediction and providing a more realistic basis for equipment maintenance decisions.

[0262] (2) Implementation of seasonal strategies

[0263] The weighting of ultraviolet radiation and thermal shock in the comprehensive stress index is adjusted according to seasonal changes in the region where the equipment is located.

[0264] Summer Strategy: The UV index is set at 50% and the thermal shock index at 30%. This is because the UV intensity is higher in summer, which has a more significant impact on the aging of equipment materials. At the same time, the diurnal temperature range is relatively small, and the impact of thermal shock is relatively weakened. By increasing the UV index, the main influencing factor of summer UV on equipment lifespan is highlighted, enabling the model to more accurately assess the health status and remaining service life of equipment in summer. This provides targeted guidance for summer equipment maintenance decisions. For example, before summer arrives, sun protection measures for equipment and material aging detection should be strengthened, and summer maintenance plans should be reasonably arranged.

[0265] Winter Strategy: The weight of ultraviolet radiation is set at 30%, and the weight of thermal shock is set at 50%. In winter, due to the large temperature difference between day and night, the fatigue damage caused by thermal shock to equipment is more severe. Since the intensity of ultraviolet radiation is relatively low, the weight of thermal shock is increased to emphasize the key impact of the temperature difference between day and night on the life of equipment in winter. The model will focus more on thermal shock factors in winter to detect equipment damage caused by thermal fatigue in a timely manner and guide winter maintenance work, such as strengthening the insulation measures of equipment and monitoring thermal circulation to prevent equipment failure caused by thermal shock.

[0266] Spring and Autumn Strategy: Both ultraviolet radiation and thermal shock are weighted at 40%. Spring and autumn environmental conditions are relatively mild, with moderate ultraviolet radiation intensity and diurnal temperature range. Their impact on equipment lifespan is relatively balanced; therefore, giving them equal weights allows the model to comprehensively consider the effects of both factors on the equipment. This ensures that equipment maintenance decisions in spring and autumn do not favor one factor, achieving a comprehensive and balanced assessment of equipment health. For example, routine equipment inspections and maintenance can be carried out in spring and autumn, and maintenance time and content can be rationally arranged based on model predictions to ensure stable equipment operation during seasonal transitions.

[0267] 7. Implementation Guidelines for the Project

[0268] (1) System deployment and implementation

[0269] Construct a field sensor network: Install various sensors (such as Hall effect sensors, thermocouples, ultraviolet sensors, etc.) on equipment such as photovoltaic inverters. Ensure that the sensor installation locations are reasonable and that they can accurately collect the required electrical, temperature, environmental, and operational data. The installation of sensors should comply with relevant standards and the requirements of equipment manufacturers. For example, temperature sensors should be installed in locations that can directly reflect the IGBT junction temperature and the heat sink temperature, avoiding external interference. At the same time, take protective measures for the sensors to ensure that they can operate stably for a long time in harsh outdoor environments.

[0270] Deploy edge computing nodes: Deploy edge computing devices, such as industrial-grade computers or edge computing servers, near the equipment site. These devices have sufficient computing power and data storage capacity to receive data collected by field sensors and perform preliminary data processing, including data cleaning, outlier detection, and feature extraction. Edge computing nodes connect to sensors via wired (e.g., industrial Ethernet) or wireless (e.g., 4G / 5G) communication to ensure real-time and reliable data transmission. Simultaneously, edge computing nodes are also responsible for sending the processed feature data to cloud-based model services for further RUL prediction analysis.

[0271] Cloud-based model service setup: Deploy the Bayesian-PINN model service on a cloud server. The cloud server has high-performance computing resources and large-capacity data storage, which can meet the needs of model training and large-scale data prediction. The model service interacts with edge computing nodes through RESTful API or other communication protocols, receives feature data and returns RUL prediction results. At the same time, the cloud server is also responsible for storing and managing historical data, model parameters and maintaining decision records, providing data support for the operation and optimization of the entire system.

[0272] Operation and Maintenance System Integration: The RUL prediction results are fed back to the operation and maintenance management system (such as the enterprise's equipment management system or dedicated photovoltaic power plant operation and maintenance software). The operation and maintenance management system automatically generates maintenance work orders, alarm information, and maintenance suggestions based on the prediction results and decision logic, and pushes them to the operation and maintenance personnel. The operation and maintenance personnel can view the health status, remaining service life, maintenance priority, and other information of the equipment through the operation and maintenance system, and formulate corresponding maintenance plans and work arrangements. At the same time, the operation and maintenance system can also record the execution status of maintenance work and the historical maintenance records of the equipment, providing data support for continuous model optimization and full life cycle management of equipment, and realizing closed-loop management of equipment operation and maintenance.

[0273] (2) Implementation of maintenance strategy

[0274] Develop appropriate maintenance strategies based on the predicted remaining useful life (RUL):

[0275] When RUL > 2000 hours, an annual routine inspection is conducted. The annual inspection includes a visual inspection of the equipment, an electrical connection inspection, a cooling system inspection, and a control system calibration, to ensure that the equipment can maintain good performance after long-term operation. At the same time, the equipment's operating data and inspection results are recorded, and the equipment's health record is updated to provide data support for subsequent maintenance decisions. The annual routine inspection is usually scheduled during the off-peak period of equipment operation or during planned power outages to minimize the impact on the power supply.

[0276] When 1000 hours < RUL ≤ 2000 hours, quarterly inspections are carried out. Based on the annual inspections, quarterly inspections add performance tests of key components and replacement of some parts (such as vulnerable parts, consumables, etc.). For example, functional tests are carried out on the drive circuit of IGBT, capacitance tests are carried out on capacitors, filters of the cooling system are replaced, etc. Through quarterly inspections, potential faults and hidden dangers of the equipment can be detected in time, measures can be taken in advance for repair or replacement, the service life of the equipment can be extended, and the reliable operation of the equipment can be ensured. Quarterly inspections can be flexibly carried out at a suitable time within a quarter according to the actual operation of the equipment and the power grid dispatching arrangements, minimizing interference to the operation of the equipment.

[0277] When 500 hours < RUL ≤ 1000 hours, monthly inspections are carried out and spare parts are prepared. Monthly inspections are more detailed and in-depth, including comprehensive inspections, cleaning, fastening, etc. of the internal components of the equipment. At the same time, according to the prediction results of the equipment health status and fault mode analysis, corresponding spare parts (such as IGBT modules, capacitors, circuit boards, etc.) are prepared to ensure that the equipment can be replaced and repaired in time when a fault occurs, reducing the downtime. Monthly inspections require maintenance personnel to have a high technical level and repair ability, be able to accurately judge the fault location and cause of the equipment, and take effective maintenance measures. The preparation of spare parts should be reasonably planned and inventory managed based on factors such as the fault probability and maintenance cycle of the equipment to balance the maintenance cost and equipment availability.

[0278] When RUL ≤ 500 hours, weekly inspections are carried out and a shutdown plan is formulated. Weekly inspections focus on the real-time operation status and key performance indicators of the equipment, increase the monitoring frequency, and capture abnormal changes of the equipment in time. At the same time, a detailed shutdown maintenance plan is formulated, including shutdown time, maintenance content, personnel arrangement, spare parts list, etc., to ensure that the shutdown maintenance work can be completed in an orderly manner before the equipment reaches the failure time. Shutdown maintenance should be preferably carried out during periods of low power grid load to reduce the impact on power supply, and relevant users and departments should be notified in advance to make power outage preparations and emergency measures. During the shutdown period, a comprehensive equipment overhaul, component replacement and system debugging are carried out to restore the good operation state of the equipment and ensure the stable operation of the power system.

[0279] The above are only the preferred embodiments of the present invention. It should be noted that for those skilled in the art, several adjustments and improvements can be made without departing from the core concept of the present invention, and these adjustments and improvements should also be regarded as the protection scope of the present invention.

Claims

1. A method for monitoring the reliability of photovoltaic converters under special high-altitude environments, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing: Electrical, temperature, environmental, and operational quantities of the plateau converter are collected. Outlier processing based on the 3σ criterion, elimination of power fluctuations, selection of stable periods, and time warping (DTW) algorithm are used to timestamp data at different sampling frequencies to ensure high data quality and accurate correspondence. S2. Plateau-sensitive feature engineering: Construct a feature system including voltage ripple rate, high-frequency harmonic energy ratio, junction temperature fluctuation entropy, thermal response time, thermal shock coefficient and comprehensive stress index, and select the optimal feature subset through physical consistency and three-stage feature selection method; S3. Physical constraint modeling: Establish IGBT thermal fatigue and capacitor aging models, combine health state evolution equations to quantitatively assess equipment health status, and introduce physical residuals to ensure that the data-driven model is consistent with physical laws. S4, Bayesian-PINN fusion modeling: Construct a network architecture that includes an input layer, a Bayesian hidden layer and an output layer, and set a loss function that integrates data fitting terms, physical constraint terms and Bayesian regularization terms to achieve deep coupling between data and physical models, thereby improving the model's generalization ability and prediction accuracy; S5, RUL Prediction Process: The conversion method from health indicators to remaining useful life (RUL) takes into account prediction uncertainty, introduces an environmental adaptive correction mechanism, and achieves precise maintenance decision support based on multi-indicator decision logic. S6. Plateau Environment Adaptive Mechanism: Automatically adjusts model parameters and feature weights according to altitude and seasonal changes, enabling the monitoring system to maintain high-performance monitoring under different environmental conditions.

2. The method for monitoring the reliability of photovoltaic converters under special high-altitude environments according to claim 1, characterized in that, In the data acquisition and preprocessing of step S1, outlier processing uses the 3σ criterion to remove abnormal data in electrical quantities and removes data from power fluctuation periods; the stable period is selected based on light intensity and power fluctuation rate, and the time window is limited to 10:00-14:00; signal alignment uses the DTW algorithm to timestamp and calibrate environmental data and electrical data.

3. The method for monitoring the reliability of photovoltaic converters under special high-altitude environments according to claim 1, characterized in that, In the plateau-sensitive characteristic engineering of step S2, the voltage ripple rate takes into account the influence of altitude on the ripple rate, reflecting the degree of capacitor aging, and the formula is: In the formula, V dc This is the DC bus voltage. It is its average value, RMS represents the root mean square value, and h is the altitude; The high-frequency harmonic energy ratio reflects the degradation of IGBT switching characteristics, and the formula is: In the formula, f sw Where f is the switching frequency, and PSD(f) is the power spectral density; Junction temperature fluctuation entropy measures the complexity of junction temperature fluctuations and is related to thermal fatigue damage. The formula is: In the formula, T j p(T) is the junction temperature of the IGBT. j ) is its probability distribution, P loss0 For reference power, P loss The actual output power; thermal response time reflects the performance of the heat dissipation system, and the formula is: In the formula, T j (t) is a function of junction temperature as a function of time, P loss (t) is the function of output power as a function of time, and corr represents the correlation; The thermal shock coefficient quantifies the impact of diurnal temperature variation stress on equipment; the formula is: In the formula, ΔT day For the diurnal temperature range, N cycle This refers to the number of thermal cycles. The comprehensive stress index fully reflects the total degree of damage to equipment caused by the high-altitude environment. The formula is: In the formula, UV represents ultraviolet radiation intensity, and h represents altitude; Physical consistency testing verifies the relationship between capacitor ripple rate and temperature, and junction temperature fluctuation entropy and ambient temperature. The three-stage feature selection algorithm uses variance thresholding, physical filtering, and recursive feature elimination to select the optimal feature subset.

4. The method for monitoring the reliability of photovoltaic converters under special high-altitude environments according to claim 1, characterized in that, In the physical constraint modeling of step S3, the IGBT thermal fatigue model describes the life characteristics of the IGBT under thermal stress, and the formula is: In the formula, N f This represents the thermal fatigue life of the IGBT, where A is a constant and ΔT is the thermal fatigue life. j For junction temperature variation, β is a material constant, and E a To activate energy, k B T is the Boltzmann constant. m Average temperature; The capacitor aging model describes the life characteristics of a capacitor under the combined effects of thermal and electrical stresses. The formula is as follows: In the formula, L is the capacitor lifespan, L0 is the reference lifespan, and E a The activation energy is T, the temperature is V, the voltage is V0, the reference voltage is n, and the aging index is n. The health status evolution equation defines the health index range as [0,1]. The degradation rate is composed of the negative sum of the degradation rates of IGBTs and capacitors. The physical residual is used to evaluate the deviation between the model and physical laws.

5. The method for monitoring the reliability of photovoltaic converters under special high-altitude environments according to claim 1, characterized in that, In the Bayesian-PINN fusion modeling of step S4, the network architecture input layer receives time t and feature vector x, the weights of the Bayesian hidden layer follow a normal distribution, and the output layer outputs health indicators. The loss function includes a data fitting term that measures the difference between the predicted and actual health indicators, a physical constraint term that ensures the evolution of health status conforms to physical laws, and a Bayesian regularization term that quantifies the uncertainty of model parameters. Specifically, the loss function is: In the formula, the data fitting term is: u pred For the health indicators predicted by the model, u true For true health indicators, N is the number of data samples; physical constraints: ‖PDE(u)‖ 2 It is the square norm of the physical residual, ReLU(u t+1 -u t The term β1 and β2 are used to prevent non-physical increases in health indicators; the Bayesian regularization term is used to prevent such increases. p(w) is the prior distribution, and q(w) is the posterior distribution; Uncertainty quantification is achieved through Monte Carlo sampling, calculating the mean and standard deviation, and constructing confidence intervals. Specifically, the formula for Monte Carlo sampling is: In the formula, u (i) (t) represents the predicted health indicator value corresponding to the i-th sampling; x is the input feature vector; w (i) Let represent the model parameters for the i-th sampling; q(w) is the posterior distribution of the model parameters; M is the number of samplings; The mean of the statistic is calculated as follows: Standard deviation is In the formula, u(t) is the mean of the predicted values ​​of the health indicator; σ u (t) represents the standard deviation of the predicted values ​​of the health indicator; The confidence interval is: In the formula, CI 95 (t) represents the 95% confidence interval; 1.96 is the critical value for the normal distribution corresponding to the 95% confidence level.

6. The method for monitoring the reliability of photovoltaic converters under special high-altitude environments according to claim 1, characterized in that, In the RUL prediction process of step S5, the health indicator u-value is converted to RUL, and the failure time, basic RUL and its confidence interval are calculated, specifically as follows: In the formula, t fail Indicates the estimated time when the equipment will reach its failure threshold; t is a time variable; σ is the mean of the predicted values ​​of the health indicators. u (t) represents the standard deviation of the predicted values ​​of the health indicator; 0.2 is the pre-set failure threshold for the health indicator, specifically: RUL = t fail -t current , In the formula, RUL represents the remaining useful life; t fail t represents the estimated failure time. current The current time; Environmental adaptive correction adjusts the baseline RUL and the standard deviation of health indicators by comprehensively adjusting for environmental factors, specifically: In the formula, RUL 校正 RUL is the corrected remaining service life; RUL is the base remaining service life. It is the corrected RUL confidence interval; CI RUL It is the basic RUL confidence interval; plateau extension: In the formula, σ u′ (t) is the corrected standard deviation of the health index; σ u (t) represents the standard deviation of the original health indicators; h represents altitude; The decision-making logic is based on health indicators, remaining useful life, total standard deviation, and confidence interval width to formulate maintenance decisions under different conditions, specifically: ① Normal state: When u > 0.6 and RUL > 1000h and σ total When <0.05, the decision-making action is routine monitoring; ② Warning status: When 0.3 < u ≤ 0.6 and 500h < RUL ≤ 1000h and σ total When the value is less than 0.15, the decision action is to prepare for maintenance. ③ Emergency situation: When u≤0.3 and RUL≤500h, the decision action is to shut down immediately; ④ Confidence anomaly: When CI width > 0.3 × RUL, the decision action is manual inspection; In the formula, u is the health indicator, RUL is the remaining useful life, and σ total The total standard deviation is represented by the CI width, which is the confidence interval width.

7. The method for monitoring the reliability of photovoltaic converters under special high-altitude environments according to claim 1, characterized in that, In the plateau environment adaptive mechanism of step S6, the dynamic adjustment strategy adjusts the physical constraint weights, KL regularization weights, feature sets, and degradation rates according to altitude. Specifically, the physical constraint weights are adjusted as follows: In the formula, α is the physical constraint weight, α0 is the initial physical constraint weight, and h is the altitude; KL regularization weight adjustment: In the formula, γ is the KL regularization weight, and γ0 is the initial KL regularization weight; Feature set correction: In the formula, σ T σ' is the corrected characteristic standard deviation, σ' is the standard deviation of the characteristic. T The original feature standard deviation; The seasonal strategy adjusts the weights of ultraviolet radiation and thermal shock in the comprehensive stress index for different seasons, specifically as follows: Summer: UV radiation weighting 50%, heat shock weighting 30%; Winter: UV weighting 30%, heat shock weighting 50%; Spring and autumn: UV radiation weighting 40%, heat shock weighting 40%.

8. A reliability monitoring system for photovoltaic converters in special high-altitude environments, characterized in that, To implement the method described in any of steps 1-7, the method includes: Data acquisition module: used to collect electrical, temperature, environmental, and operational data of the plateau converter; Data preprocessing module: used to handle outliers, select stable time periods, and align signals in the collected data; Feature engineering module: used to extract plateau-sensitive features and perform physical consistency checks and feature selection; Physical constraint modeling module: used to build physical models and health state evolution equations, and calculate physical residuals; Bayesian-PINN fusion modeling module: used to build fusion models, set loss functions, and quantify uncertainties; RUL prediction module: used to predict remaining useful life and perform environmental corrections to make maintenance decisions; Plateau Environment Adaptive Module: Used to dynamically adjust model parameters and feature weights based on altitude and season.

9. The photovoltaic converter reliability monitoring system under special high-altitude environments according to claim 8, characterized in that, The data acquisition module includes a Hall sensor, a thermocouple, and a weather station interface, which are used to collect electrical quantities, temperature quantities, and environmental quantity data, respectively.

10. The photovoltaic converter reliability monitoring system under special high-altitude environments according to claim 8, characterized in that, The Bayesian-PINN fusion modeling module is implemented using a deep learning framework and has variational inference capabilities to estimate the posterior distribution parameters of the Bayesian hidden layer weights.

Images