Method for predicting service life of electrolytic capacitor

Abstract

The invention provides a method for predicting the service life of an electrolytic capacitor, which comprises the following steps of: taking environment temperature, operation time and equivalent series resistance (ESR) which is calculated based on ripple voltage and ripple current and is subjected to temperature compensation as input data, and taking the residual service life RUL of the electrolytic capacitor as a label; according to the failure mechanism of the electrolytic capacitor, when the ESR is increased to two times of the initial value of the ESR, it is indicated that the electrolyte is obviously lost, the capacitor fails to form a failure threshold value, and based on the failure threshold value, the RUL corresponding to each time point in historical monitoring is reversely deduced in combination with operation time data; then, performing normalization processing on the multi-source heterogeneous data to eliminate dimensional differences, and constructing a standardized data set suitable for model training; training and evaluating the model to obtain a model with residual life prediction capability; according to the technical scheme, the problem that the capacitor core temperature and the online electrolytic capacitor capacitance value required by traditional electrolytic capacitor life prediction are difficult to obtain is solved, and the engineering problem of electrolytic capacitor life prediction is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power electronics and PHM technology, in particular to an electrolytic capacitor life prediction method. BACKGROUND

[0002] At present, electrolytic capacitors are widely used in high-voltage and low-voltage rectification filtering of power electronics. If electrolytic capacitors are used beyond their life, it will cause problems such as deformation and liquid leakage of the device, and even explosion, reduction or failure of product performance. According to statistics, the failure rate of aluminum electrolytic capacitors in switching power supplies under room temperature conditions accounts for 60% of the total number of failed devices, so it is particularly important to ensure that electrolytic capacitors are used within their life.

[0003] The existing life prediction technology of electrolytic capacitors includes model calculation based on physical degradation mechanism and data-driven prediction based on machine learning. In the papers "Real-time residual life prediction of electrolytic capacitors considering degradation correlation" and "Parameter identification and life prediction of aluminum electrolytic capacitors in converters", the life of the capacitor is predicted by using the capacitance degradation data and equivalent series resistance degradation data. In the patent "Life monitoring device of electrolytic capacitor and electronic equipment (authorized publication number: CN217766635 U)", the residual life of the electrolytic capacitor is predicted by real-time acquisition of the internal temperature of the electrolytic capacitor.

[0004] When predicting the life of electrolytic capacitors by physical mechanism model, it is necessary to rely on physical parameters such as internal temperature which are difficult to obtain online. The machine learning method needs to monitor the capacitance value and ESR in real time, but the online measurement of capacitance value has technical bottlenecks.

[0005] And the commonly used prediction model of capacitor life based on Arrhenius theory shows that the life of electrolytic capacitor is significantly negatively correlated with working temperature. Within the typical working temperature range, the chemical reaction rate approximately doubles for every 10℃ increase in temperature, resulting in a nearly halving of the life of the capacitor. Conversely, for every 10℃ decrease in temperature, the life approximately doubles. This 10℃ rule is a simplified application of the Arrhenius formula under certain conditions, providing a practical estimation basis for engineering design.

[0006] , wherein, is the calculated life in actual use; is the working life provided by the manufacturer under the highest working temperature; is the highest working temperature provided by the manufacturer; is the center point temperature of the capacitor during actual work. Among them, and can be found in the capacitor product specification, Data is difficult to obtain. And due to the dispersion of electrolytic capacitor manufacturing process, its life There are individual differences, and the capacitance of The temperature changes with the change of seasons, the change of environment, the existing life prediction formula Is a fixed value, which cannot be adjusted according to the actual working condition, and cannot reflect the life of electrolytic capacitor in real time according to the actual working condition. Therefore, the remaining life of individual capacitor must be accurately evaluated by monitoring parameters that can represent the life of electrolytic capacitor, such as the change of working environment temperature, the change of running time and the change of ESR. SUMMARY

[0007] Therefore, the purpose of the present application is to provide an electrolytic capacitor life prediction method, which effectively avoids the requirement of direct measurement of the internal temperature of electrolytic capacitor, overcomes the technical bottleneck that the capacitance value of capacitor is difficult to monitor online in real time, and realizes high-precision modeling and life prediction of the degradation process of electrolytic capacitor by fusing temperature and dynamically changing ESR characteristics after temperature compensation.

[0008] To achieve the above purpose, the present application adopts the following technical scheme: an electrolytic capacitor life prediction method, taking environmental temperature, running time and equivalent series resistance ESR calculated based on ripple voltage and ripple current and compensated by temperature as input data, and taking the remaining useful life RUL of electrolytic capacitor as label; according to the failure mechanism of electrolytic capacitor, when ESR increases to twice its initial value, it indicates that the electrolyte has been significantly depleted and the capacitor has failed to form a failure threshold, based on the failure threshold, combined with the running time data, the remaining useful life RUL corresponding to each time point in the history monitoring is back calculated; then, the multi-source heterogeneous data is normalized to eliminate the dimensional difference, and a standardized data set suitable for model training is constructed; the model is trained and evaluated to obtain a model with remaining life prediction capability.

[0009] In a preferred embodiment, the equivalent series resistance is calculated by monitoring the ripple voltage and the ripple current:

[0010]

[0011] Wherein, V ripple represents the ripple voltage, and I ripple represents the ripple current.

[0012] In a preferred embodiment, the temperature compensation of equivalent series resistance ESR is calculated by the following formula:

[0013]

[0014] ESR represents the compensated ESR value, which is at a standard reference temperature ​equivalent ESR value at a lower temperature, for fairly comparing the remaining useful life RUL of electrolytic capacitors at different temperatures; represents the original ESR value measured at the current actual case temperature of the capacitor; represents the normalized temperature compensation coefficient; represents the actual case temperature at the time of measurement.

[0015] In a preferred embodiment, the backstepping of the remaining useful life RUL corresponding to each time point in the historical monitoring includes: the sampling time interval of the data is 1 hour, and the failure criterion is that the equivalent series resistance ESR of the electrolytic capacitor increases to 2 times of its initial value at a standard temperature; the total duration from the start to reaching the failure threshold is t hours, based on which the original time index sequence is constructed: {1, 2, 3,..., t-1, t}, each time point corresponds to an observation sample; to construct the remaining useful life RUL label, the sequence is transformed in reverse order to generate the RUL label sequence: {t, t-1,..., 3, 2, 1}; each element in the RUL label sequence represents the remaining time unit number of the corresponding sample point from failure, i.e., the RUL value at the starting point is t, which then decreases by one unit at a time, to 0 at the failure time; the RUL label is assigned to each sample, thereby constructing a data set suitable for life prediction.

[0016] In a preferred embodiment, the model construction and training use an LSTM model; the LSTM model is composed of an input gate , a forget gate , an output gate , and a memory cell ; given an input sequence , where the input at each time step t contains the environmental temperature Temp, the running time Time, the equivalent series resistance ESR calculated indirectly through electrical measurement, and the remaining useful life RUL as the training label; the hidden state represents the output of the previous time step LSTM; the LSTM updates the memory cell state through the gating mechanism, dynamically adds or deletes information, to retain the content related to the prediction task; the input is multiplied by the candidate value , which controls the amount of new information added; is multiplied by the previous state , which determines how much historical information is removed from the memory cell; the output gate then generates the final output based on the current state .

[0017] In a preferred embodiment, the current ambient temperature, running time, ESR and temperature compensated ESR are input into the model, the model will return the remaining service life prediction value, which is predicted by the model trained by aging data, the time is compressed, which needs to be converted into normal use time in actual application, and the focus is to calculate the acceleration factor ;

[0018]

[0019] Wherein, E a represents the activation energy, the minimum energy required for the aging reaction; K B represents the Boltzmann constant, the proportional constant connecting the energy of micro-particles and the macro-temperature; T ref represents the reference temperature, the temperature as the benchmark; T use represents the working temperature, the temperature at which the capacitor actually works;

[0020] .

[0021] Compared with the prior art, the present application has the following beneficial effects:

[0022] 1. A practical method framework for indirectly determining the remaining life based on the trajectory of the predicted health index (ESR) is proposed.

[0023] 2. In order to improve the generalization ability of the model, multiple electrolytic capacitor operation data (including equivalent series resistance (ESR), ambient temperature and cumulative running time, etc.) are collected for training the remaining useful life (RUL) prediction model, so as to effectively overcome the performance dispersion problem of the capacitor caused by the difference in manufacturing process, and enhance the adaptability and prediction accuracy of the model among different capacitor individuals.

[0024] 3. The present application combines human historical experience and constructs RUL labels, and through deep learning-based time series data prediction, the electrolytic capacitor remaining useful life (RUL) prediction which is relatively easy to implement in engineering practice can be obtained. BRIEF DESCRIPTION OF DRAWINGS

[0025] Fig. 1 The electrolytic capacitor remaining useful life (RUL) prediction architecture diagram of the preferred embodiment of the present application;

[0026] Fig. 2 The capacitor ripple voltage and ripple current measurement diagram of the preferred embodiment of the present application;

[0027] Fig. 3 The high and low voltage electrolytic capacitor RUL information of the preferred embodiment of the present application. DETAILED DESCRIPTION

[0028] The application will be further described below with reference to the accompanying drawings and examples.

[0029] It should be noted that the following detailed description is illustrative only and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0030] It should be noted that the terms used herein are only intended to describe specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and furthermore, it should be understood that when the terms "comprise" and / or "include" are used in the specification, there is a presence of the features, steps, operations, devices, components and / or combinations thereof.

[0031] As shown in Figs. 1-3 The overall technical solution proposed by the application takes the ambient temperature, the running time, and the equivalent series resistance (ESR) calculated based on the ripple voltage and the ripple current and temperature compensated as input data, and takes the remaining useful life (RUL) of the electrolytic capacitor as a label. According to the failure mechanism of the electrolytic capacitor, when the ESR increases to twice its initial value, it indicates that the electrolyte has been significantly depleted and the capacitor has failed. Based on this failure threshold, combined with the running time data, the remaining useful life (RUL) corresponding to each time point in the historical monitoring can be deduced. Subsequently, the multi-source heterogeneous data is normalized to eliminate the dimensional differences, and a standardized data set suitable for model training is constructed. Finally, the model is trained and evaluated to have the remaining life prediction capability, and the prediction service can be called by the staff.

[0032] In the present application, the equivalent series resistance (ESR) after temperature compensation is used as an input feature for LSTM network training; the ESR temperature compensation calculation takes the capacitor shell temperature as the observation basis, and the complex nonlinear mapping relationship between the shell temperature and the capacitor core temperature, as well as the deep degradation coupling mechanism between them and the remaining useful life (RUL) of the electrolytic capacitor, do not need to rely on explicit physical modeling, and are all obtained by the data-driven LSTM prediction model through end-to-end training and self-learning.

[0033] The specific implementation steps are as follows:

[0034] 1. Data acquisition

[0035] The actual service life of the electrolytic capacitor is usually long, and in order to obtain comprehensive and diversified degradation data in a short period of time, an accelerated aging test method needs to be used. According to the comprehensive aging acceleration factor formula of the electrolytic capacitor:

[0036] The aging process can be significantly accelerated by applying multiple stress conditions such as temperature, voltage, and ripple current. The formula quantifies the combined acceleration effect of these stresses on the intrinsic physical and chemical processes such as electrolyte evaporation and oxide degradation. In experiments targeting switch mode power supply applications, temperature stress is mainly controlled to accelerate aging.

[0037] The data required for the program include ripple voltage, ripple current, ambient temperature, operating time, equivalent series resistance (ESR) calculated indirectly through electrical measurements, and remaining useful life (RUL) label data. Among them, the temperature and operating time are directly measurable features collected in real time by sensors, ESR is a calculated feature obtained indirectly from electrical measurements such as ripple voltage and ripple current, and RUL is determined to be invalid when ESR increases to twice the initial value, which is used as the end of life of the electrolytic capacitor. The remaining useful life (RUL) at each time is calculated based on historical data and used as a training label.

[0038] To improve the generalization ability of the model, this study collects multiple electrolytic capacitor operating data (including equivalent series resistance (ESR), ambient temperature, and cumulative operating time), which are used to train the remaining useful life (RUL) prediction model. This effectively overcomes the problem of capacitor performance dispersion caused by manufacturing process differences and enhances the adaptability and prediction accuracy of the model among different capacitor individuals.

[0039] (1) Ripple voltage and ripple current measurement

[0040] Capacitor ripple voltage and current are both weak signals, usually in the order of millivolts and milliamperes. Instrumentation amplifiers are used to collect ripple voltage, and operational amplifiers are used to collect precise resistance voltage. The ripple current of the capacitor is calculated through the voltage, current, and resistance formula, and the voltage and current signals are synchronously sampled through the MCU multi-channel ADC.

[0041] 1) Online measurement of ripple voltage

[0042] The AC ripple voltage across the capacitor is directly measured using a high-precision instrumentation amplifier. The DC bias is removed by a high-pass filter, and only the AC component is extracted. Since the ripple voltage is usually small, in the order of millivolts, a high-resolution ADC such as a 16-bit ADC is required for analog-to-digital conversion.

[0043] 2) Online measurement of ripple current

[0044] A small resistance such as a 0.1Ω precision sampling resistor is connected in series with the capacitor circuit to measure the voltage across it and calculate the current. The ripple current frequency may be high, such as several hundred kHz in a switch mode power supply, so the sampling circuit must have sufficient bandwidth.

[0045] (2) Equivalent series resistance of electrolytic capacitor

[0046] The equivalent series resistance of electrolytic capacitor can be calculated by the monitored ripple voltage and ripple current:

[0047]

[0048] This calculated ESR reflects the real ESR value of electrolytic capacitor under actual environmental temperature working stress.

[0049] (3) Temperature compensation of electrolytic capacitor equivalent series resistance ESR

[0050] In order to make the model calculate the remaining useful life RUL of electrolytic capacitor under the unified reference, the temperature compensation of ESR is needed. In this way, the model can not only see the absolute health condition, but also perceive the environmental stress intensity, so as to obtain comprehensive information. The temperature compensation of electrolytic capacitor equivalent series resistance ESR is calculated by the following formula (different methods can be used, which does not affect the prediction accuracy of electrolytic capacitor RUL):

[0051]

[0052] : Compensated ESR value, which is the equivalent ESR value at the standard reference temperature (usually 25°C), used to fairly compare the remaining useful life RUL of electrolytic capacitor at different temperatures.

[0053] : Original ESR value measured at the current actual case temperature of the capacitor.

[0054] : Normalized compensation coefficient. This is a dimensionless coefficient, which represents the temperature compensation coefficient when the capacitor temperature is .

[0055] : Actual case temperature at the time of measurement, which needs to be obtained in real time by the temperature sensor (such as NTC, thermocouple) installed near the capacitor.

[0056] : Standard reference temperature, usually fixed at 25°C, all ESR values are compensated to this temperature to eliminate the temperature effect.

[0057] The temperature compensation coefficient in it is fitted by mathematical function from the curve in the manufacturer's data manual. The ESR temperature characteristic of electrolytic capacitor can usually be fitted by piecewise function or polynomial, for example, a common quadratic polynomial fitting formula is:

[0058]

[0059] wherein is a constant coefficient obtained by curve fitting, is the measured temperature, and is substituted into the formula, and the .

[0060] (4) Constructing the residual service life characteristics of electrolytic capacitors

[0061] Industry experience shows that during the use of electrolytic capacitors, as the electrolyte volatilizes, the equivalent series resistance ESR gradually increases and the capacitance value C gradually decreases. Generally speaking, under the same temperature conditions, when the equivalent series resistance ESR increases to 2 times the initial value or the capacitance value C decreases to 80% of the initial value, it can be considered that the electrolytic capacitor has failed. The present application takes the increase of the equivalent series resistance ESR to 2 times the initial value as the failure characteristic, which is used as the basis for calculating the residual life of the electrolytic capacitor RUL, and combines the running time to back-calculate the residual life value corresponding to each time point in the historical monitoring data.

[0062] Suppose the sampling time interval of the data is 1 hour, and the increase of the equivalent series resistance ESR of the electrolytic capacitor to 2 times the initial value at the standard temperature is taken as the failure criterion. From the start to reaching the failure threshold, a total of t hours are required, and accordingly the original time index sequence {1, 2, 3,..., t-1, t} is constructed, each time point corresponding to an observation sample. In order to construct the residual service life RUL label, the sequence is transformed in reverse order to generate the RUL label sequence {t, t-1,..., 3, 2, 1}. Each element in the sequence represents the residual time unit of the corresponding sample point from failure, i.e. the RUL value at the starting point is t, which then decreases by one unit at a time, to zero at the failure time. Through this method, accurate RUL labels can be assigned to each sample, thereby constructing a data set suitable for life prediction.

[0063] 2. Data preprocessing

[0064] All feature data in the data set are sequentially subjected to maximum and minimum value normalization preprocessing:

[0065] (1)

[0066] In formula (1), x is the data to be normalized, x min and x max represent the maximum and minimum values in the data to be normalized, respectively. After the data is finally mapped between 0 and 1, it is input into the model as feature data for training.

[0067] 3. Model construction and training

[0068] An LSTM model is used to effectively process time series data and capture nonlinear degradation relationships. The LSTM model consists of an input gate Forgot the door Output gate and a memory unit Composition. Given an input sequence The input at each time step t Includes ambient temperature (Temp), runtime (Time), equivalent series resistance (ESR) indirectly calculated through electrical measurements, and remaining lifetime (RUL) as a training label. Hidden state. This represents the output of the LSTM at the previous time step. LSTM updates the memory cell state through a gating mechanism, dynamically adding or deleting information to retain content relevant to the prediction task. Inputs Multiply by candidate value Control the amount of new information; Compared to the previous state Multiplication determines how much historical information is removed from the memory unit; output gate Based on the current state Generate final output .

[0069] This solution first involves data acquisition and preprocessing. Input data includes actual monitored ripple voltage, ripple current, ambient temperature, and operating time. Based on this, the equivalent resistance (ESR) and remaining useful life (RUL) labels are calculated. All features are normalized before training to eliminate dimensional differences and improve training performance. Subsequently, the processed data is divided into training and testing sets, and a neural network model with multiple LSTM hidden layers is built, extracting temporal features layer by layer. During training, a loss function is used to calculate the error, and the Adam optimizer is used to iteratively optimize the network parameters to approximate the optimal value. Finally, the prediction results are denormalized through the output layer to generate actual RUL prediction values, thus obtaining the optimal model that meets the requirements.

[0070] Input Gate:

[0071]

[0072] Forgot the door:

[0073]

[0074] Output gate:

[0075]

[0076] Candidate values ​​for memory cells:

[0077]

[0078] Memory cell state update:

[0079]

[0080] Output:

[0081]

[0082] The prediction set data is input into the neural network for testing. The mean absolute error and root mean square error are used as evaluation indicators for the prediction results. The results are compared with the test dataset, and the loss is calculated. When the model test results are not ideal, the model parameters are adjusted until the test results are ideal. The trained network model architecture and network parameters are saved.

[0083] The mean squared error (MSE, MAE, MAPE) is used as the optimization objective for the LSTM-based remaining lifetime model. The evaluation metrics employed include MAE (mean absolute error), MSE (mean squared error), and MAPE (mean absolute percentage error). During iteration, the values ​​of MSE, MAE, and MAPE decrease, indicating a gradual improvement in model performance. The model loss curve reflects the model's fit to the dataset and the iterative loss process. When MSE, MAE, and MAPE are close to 0, it indicates that the remaining lifetime predicted by the LSTM-based model is closer to reality, and the accuracy is higher. When the model's performance is unsatisfactory, the model parameters are adjusted until the test results are ideal. The trained network model architecture and parameters are then saved.

[0084] Its definition is as follows:

[0085]

[0086]

[0087]

[0088] in and Let represent the predicted value and the true label of the i-th sample, respectively, and n represent the total number of samples in the test set.

[0089] The remaining useful life model based on LSTM is used to predict the remaining useful life at a future point in time. MAE and MSE are two commonly used evaluation metrics for regression models; the closer the value is to 0, the better the model fit. The optimal hyperparameter combination found during iterative training is fed into the remaining useful life prediction model for training, resulting in the remaining useful life prediction model, which is then stored in the cloud or locally for subsequent deployment.

[0090] 4. Model Utilization

[0091] Input the current ambient temperature, operating time, ESR, and temperature-compensated ESR into the model. The model will return a predicted remaining service life. This prediction is made using a model trained on aging data, and the time is compressed. In practical applications, it needs to be converted to the normal operating time. The key is to calculate the acceleration factor. .

[0092]

[0093] AFtemp: Temperature acceleration factor, the factor by which the aging rate increases due to temperature changes; Ea: Activation energy, the minimum energy required for the aging reaction to occur; KB: Boltzmann constant, the proportionality constant connecting the energy of microscopic particles to the macroscopic temperature; Tref: Reference temperature, the temperature used as a benchmark; Tuse: Operating temperature, the actual operating temperature of the capacitor.

[0094]

[0095] This invention relates to a switching power supply containing multiple high-voltage and low-voltage filter electrolytic capacitors. The remaining lifespan (RUL) of each electrolytic capacitor is displayed in real-time on the power supply's LCD interface, providing component-level lifespan monitoring to address existing switching power supply fault diagnosis issues. Current electrolytic capacitor fault diagnosis typically waits until a power electronic device malfunctions. During repair, engineers rely on appearance, odor, and other signs to determine if a capacitor is abnormal, isolate the faulty capacitor, manually measure its capacitance to assess aging or failure, and replace it promptly to eliminate potential problems. This process results in downtime and losses. This invention allows engineers to predict capacitor lifespan and replace them in advance, facilitating repairs and preventing unnecessary losses.

Claims

1. A method for predicting the lifespan of electrolytic capacitors, characterized in that, Using ambient temperature, operating time, and the equivalent series resistance (ESR) calculated based on ripple voltage and ripple current and compensated for temperature as input data, and the remaining useful life (RUL) of the electrolytic capacitor as the label, according to the failure mechanism of electrolytic capacitors, when the ESR increases to twice its initial value, it indicates that the electrolyte has been significantly depleted and the capacitor has failed, forming a failure threshold. Based on this failure threshold, combined with operating time data, the remaining useful life (RUL) corresponding to each time point in historical monitoring is inferred. Subsequently, the multi-source heterogeneous data is normalized to eliminate dimensional differences and construct a standardized dataset suitable for model training. The model is then trained and evaluated to obtain a model with remaining useful life prediction capabilities.

2. The method for predicting the lifespan of an electrolytic capacitor according to claim 1, characterized in that, The equivalent series resistance is calculated using the monitored ripple voltage and ripple current: Among them, V ripple I represents the ripple voltage. ripple This represents the ripple current.

3. The method for predicting the lifespan of an electrolytic capacitor according to claim 1, characterized in that, The temperature compensation for the equivalent series resistance (ESR) is calculated using the following formula: This represents the compensated ESR value, which is at the standard reference temperature. The equivalent ESR value is used to fairly compare the remaining life (RUL) of electrolytic capacitors at different temperatures. This indicates the current actual case temperature of the capacitor. The original ESR value obtained from the measurement; This represents the normalized temperature compensation coefficient; This indicates the actual shell temperature at the time of measurement.

4. The method for predicting the lifespan of an electrolytic capacitor according to claim 1, characterized in that, The method for reverse-engineering the Remaining Lifetime (RUL) at each time point in historical monitoring includes: a data sampling interval of 1 hour, with the failure criterion being the increase in the equivalent series resistance (ESR) of the electrolytic capacitor to twice its initial value at standard temperature; the duration from the start to reaching this failure threshold being t hours, thus constructing an original time index sequence: {1, 2, 3, ..., t-1, t}, with each time point corresponding to an observation sample; to construct the RUL label, this sequence is reversed to generate the RUL label sequence: {t, t-1, ..., 3, 2, 1}; each element in this RUL label sequence represents the number of remaining time units from failure for the corresponding sample point, i.e., the initial RUL value is t, which then decreases unit by unit until it reaches 0 at the failure time, assigning precise RUL labels to each sample, thereby constructing a dataset suitable for predicting the lifespan of electrolytic capacitors.

5. The method for predicting the lifespan of an electrolytic capacitor according to claim 1, characterized in that, The model is built and trained using an LSTM model; the LSTM model consists of an input gate. Forgot the door Output gate and a memory unit Composition; given an input sequence The input at each time step t Includes ambient temperature (Temp), runtime (Time), equivalent series resistance (ESR) indirectly calculated through electrical measurements, and remaining lifetime (RUL) as a training label; hidden state. This represents the output of the LSTM at the previous time step; the LSTM updates the memory cell state through a gating mechanism, dynamically adding or deleting information to retain content relevant to the prediction task; the inputs... Multiply by candidate value Control the amount of new information; Compared to the previous state Multiplication determines how much historical information is removed from the memory unit; output gate Based on the current state Generate final output .

6. The method for predicting the lifespan of an electrolytic capacitor according to claim 1, characterized in that, Input the current ambient temperature, operating time, ESR, and temperature-compensated ESR into the model. The model will return a predicted remaining service life. This prediction is made using a model trained on aging data, and the time is compressed. In practical applications, it needs to be converted to the normal operating time. The key is to calculate the acceleration factor. ; Among them, E a K represents the activation energy, the minimum energy required for an aging reaction to occur. B T represents the Boltzmann constant, a proportionality constant connecting the energy of microscopic particles to the macroscopic temperature; ref Indicates the reference temperature, the temperature used as a benchmark; T use This indicates the operating temperature, the actual operating temperature of the capacitor; 。

Images