A power device service life prediction method and system

Abstract

The application provides a power device service life prediction method and system, and relates to the technical field of power device monitoring. The method comprises the following steps: acquiring electrical data and environmental data of a power device; extracting a plurality of key failure precursor features for characterizing the service life of the power device from the electrical data; performing feature selection and combination on each key failure precursor feature to generate an optimal failure precursor feature combination; inputting the optimal failure precursor feature combination into an MCLSTM network based on grade division to output hierarchical time sequence degradation features; extracting auxiliary failure precursor features for characterizing the service life of the power device from the environmental data; performing fusion processing on the hierarchical time sequence degradation features and the auxiliary failure precursor features to obtain fusion features of the power device; and outputting a service life prediction result of the power device through a full connection network according to the fusion features.

Description

Technical Field

[0001] This invention relates to the field of power device monitoring technology, and in particular to a method and system for predicting the lifespan of power devices. Background Technology

[0002] Power devices are widely used in modern power systems, communication equipment, consumer electronics, and industrial control, playing a crucial role in providing stable power and voltage conversion. With the continuous advancement of electronic technology and the emergence of high-power-density devices, the operating environment of power devices has become more complex, facing higher operating pressures and more stringent performance requirements. Therefore, predicting the lifespan of power devices, especially under high-temperature, high-voltage, and high-frequency environments, is of significant practical importance for ensuring the reliability and stability of systems.

[0003] Currently, methods for predicting the lifespan of power devices can be mainly categorized into several types, including physical model-based prediction, statistical analysis-based prediction, and machine learning-based prediction. Physical model methods analyze the internal physical aging processes of power devices (such as thermal failure and mechanical stress), establish mathematical models, and validate them with experimental data, enabling relatively accurate simulation of the device's lifespan. Statistical analysis methods collect large amounts of power device failure data and apply statistical methods such as regression analysis and failure rate analysis for lifespan prediction. In recent years, with the development of artificial intelligence technology, machine learning and deep learning methods have been widely applied to power device lifespan prediction. These methods can automatically uncover potential patterns in data and possess strong predictive capabilities.

[0004] However, traditional lifetime prediction methods often fail to fully capture the key changes during the degradation process of power devices, and the limitations of feature selection directly affect the prediction accuracy of the model. Especially when dealing with multivariate and high-dimensional data, these methods may fail to effectively identify and mine potential influencing factors, thus affecting the accuracy of the prediction results. At the same time, traditional lifetime prediction methods usually rely on specific datasets for training, which is prone to overfitting, resulting in a significant decrease in prediction accuracy and poor generalization ability when using new power devices or operating under different conditions. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a power device lifetime prediction method that can solve the problem that traditional lifetime prediction methods often fail to fully capture the key changes in the degradation process of power devices, and the limitations of feature selection directly affect the prediction accuracy of the model. Especially when facing multivariate and high-dimensional data, these methods may fail to effectively identify and mine potential influencing factors, thus significantly impacting the accuracy of the prediction results. Furthermore, traditional lifetime prediction methods typically rely on training on specific datasets, which is prone to overfitting, leading to a significant decrease in prediction accuracy and poor generalization ability under new power devices or different operating environments.

[0006] A first aspect of this invention provides a method for predicting the lifetime of power devices, comprising:

[0007] S1: Acquire the electrical data and environmental data of the power device;

[0008] S2: Extract several key precursor features from the electrical data to characterize the lifespan of the power device;

[0009] S3: Select and combine the key failure precursor features to generate the optimal failure precursor feature combination;

[0010] S4: Input the optimal failure precursor feature combination into the hierarchical partitioning MCLSTM network to output hierarchical temporal degradation features;

[0011] S5: Extract auxiliary failure precursor features from the environmental data to characterize the lifespan of the power device;

[0012] S6: The hierarchical timing degradation features and the auxiliary failure precursor features are fused to obtain the fused features of the power device;

[0013] S7: Based on the fusion characteristics, output the predicted lifespan of the power device through a fully connected network.

[0014] A second aspect of the present invention provides a power device lifetime prediction system, comprising: a processor and a memory;

[0015] The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the power device lifetime prediction method as described in the first aspect.

[0016] A third aspect of the present invention provides a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the power device lifetime prediction method as described in the first aspect.

[0017] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0018] In this embodiment of the invention, by extracting multiple key pre-failure features and auxiliary features from electrical and environmental data, the critical changes in the power device degradation process can be comprehensively captured, avoiding the limitations of traditional methods in feature selection. Especially when facing multivariate and high-dimensional data, it can effectively identify potential influencing factors, thereby significantly improving prediction accuracy. Simultaneously, by combining a hierarchical partitioning-based MCLSTM network and a fully connected network, the comprehensive application of hierarchical temporal degradation features and fused features enhances the model's generalization ability, enabling the model to not only perform well on specific datasets but also maintain high prediction accuracy across different types of power devices and operating environments. Attached Figure Description

[0019] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0020] Figure 1 This is a flowchart illustrating a method for predicting the lifespan of power devices provided in an embodiment of the present invention;

[0021] Figure 2 This is a schematic diagram of a power device lifespan prediction system provided in an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0023] The power device lifespan prediction method provided by the present invention will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.

[0024] Reference manual attached Figure 1 The diagram shows a flowchart of a power device lifespan prediction method provided by an embodiment of the present invention.

[0025] This invention provides a method for predicting the lifespan of power devices, which may include the following steps:

[0026] S1: Acquire electrical data and environmental data of power devices.

[0027] Electrical data refers to data recorded and monitored in electrical systems that relates to the operation of power equipment or systems. This data is primarily used to describe the operating status, performance changes, and early signs of failure of electrical equipment, and is crucial for fault diagnosis, performance evaluation, and lifespan prediction.

[0028] In one possible implementation, the electrical data includes source-drain resistance, leakage current, input capacitance, output capacitance, case temperature, current waveform, voltage waveform, insulation resistance, switching frequency, and switching cycle.

[0029] Environmental data include: ambient temperature, ambient humidity, air pollution and dust, as well as vibration and mechanical stress.

[0030] It should be noted that the environmental parameters were determined by experts based on their experience, who selected the parameters that had the greatest impact on the lifespan of power devices.

[0031] S2: Extract several key failure precursor features from electrical data to characterize the lifespan of power devices.

[0032] Key failure precursor characteristics refer to features or parameters that can indicate equipment degradation, damage, or impending failure during operation. These characteristics are usually within the normal operating range, but their trends or abnormal fluctuations can provide important information for predicting equipment failure or its lifespan. By monitoring and analyzing these characteristics, early warnings can be provided, sudden failures can be avoided, thereby reducing maintenance costs, improving equipment reliability, and extending its service life.

[0033] In one possible implementation, S2 specifically includes:

[0034] S201: Set the environmental conditions and number of operating cycles for accelerated degradation of power devices in accelerated aging tests.

[0035] Examples include current, temperature, and frequency. These conditions are typically more stringent than those during normal operation to accelerate the degradation process. Current: Increasing the device's operating current simulates high-load conditions. In power systems or industrial control, equipment may operate under overload conditions for extended periods at peak loads. Increasing the current not only increases the electrothermal effect of power devices but also accelerates the aging process of internal materials, such as increased contact resistance and insulation degradation, thus affecting performance and lifespan. Temperature: Increasing the device's operating temperature simulates operation in high-temperature environments. When current passes through semiconductor components, heat accumulation can lead to degradation of silicon-based materials, resulting in performance decline or even failure. Increasing the temperature in accelerated testing reveals these thermal aging effects more quickly. Frequency: Increasing the operating frequency simulates more frequent switching operations. For example, frequent current switching can cause switching losses, EMI (electromagnetic interference), and thermal cycling problems.

[0036] S202: Record the initial state of the power device, wherein the initial state includes the initial values ​​of each electrical parameter in the electrical data.

[0037] S203: Operate the power device under the set environmental conditions and number of working cycles, and periodically measure the values ​​of various electrical parameters in each cycle to obtain multiple cycle values.

[0038] It should be noted that the parameters will show a gradual degradation trend as the test progresses.

[0039] S204: Perform statistical analysis on the initial values ​​and multiple periodic values ​​of each electrical parameter to determine the changing trend of each electrical parameter.

[0040] Electrical parameters refer to various electrical characteristics that describe the operating state of an electrical system or equipment. These parameters directly affect the performance, efficiency, and safety of the equipment. Common electrical parameters include current, voltage, power, frequency, impedance, and power factor.

[0041] In this embodiment of the invention, the changing trends of each electrical parameter are systematically recorded and analyzed during accelerated aging tests, providing a clear tracking of the degradation process. This enables the predictive model to accurately assess the degradation rate and remaining lifetime of power devices based on reliable historical data, thereby enhancing the credibility of the prediction results.

[0042] S205: Based on the various trends, determine the maximum change amplitude of each electrical parameter.

[0043] In this embodiment of the invention, the maximum change magnitude represents the most significant change in parameter degradation, highlighting the main failure trends of the device during accelerated aging. Selecting features with large change magnitudes helps the model focus on the factors that most affect device lifetime, thereby improving prediction accuracy.

[0044] S206: Sort the maximum change amplitudes in descending order and select the electrical parameters corresponding to the highest-ranking maximum change amplitudes as key failure precursor features.

[0045] In this embodiment of the invention, by statistically analyzing the changing trends of each electrical parameter and selecting the parameter with the largest change amplitude as a key precursor feature of failure, the main factors affecting the service life of power devices can be identified more accurately. This feature selection avoids interference from irrelevant or weakly correlated features, improving the predictive ability of the model.

[0046] S3: Select and combine key failure precursor features to generate the optimal failure precursor feature combination.

[0047] The optimal failure precursor feature combination refers to the process of selecting and optimizing features from a large amount of electrical data to identify the most predictive features, and then effectively combining these features to maximize the accuracy and reliability of the fault prediction model. The optimized feature combination can accurately reflect the degradation trend of equipment, promptly detect potential faults, provide early warnings to maintenance personnel, and prevent sudden failures.

[0048] In one possible implementation, S3 specifically includes:

[0049] S301: Through the initialization process of the Zebra Optimization Algorithm, an initial population of multiple failure precursor feature combinations is initially generated.

[0050] S302: Determine the fitness function used to evaluate the combination of failure precursor features:

[0051]

[0052] Where F represents the fitness function, Var(X) represents the feature variance in the failure precursor feature combination X, and Corr(X) represents the correlation between the features in the failure precursor feature combination X.

[0053]

[0054]

[0055] Where n represents the total number of sample points, Let represent the measured value of the i-th sample point in the failure precursor feature combination X, and μ represent the mean of the features.

[0056]

[0057] Among them, X i,k X represents the value of the i-th feature in the failure precursor feature combination X in the k-th sample. j,k μ represents the value of the j-th feature in the failure precursor feature combination X in the k-th sample. xi μ represents the mean of the i-th feature. xj Let m represent the mean of the j-th feature, and m represent the number of samples.

[0058] Specifically, a smaller variance in a feature indicates that the feature is relatively stable, and therefore has a lower fitness. A larger variance indicates that the feature has greater variability and may contribute more to the prediction result, resulting in a higher fitness. If the correlation between features is high, it indicates a lot of redundant information, and the fitness should be low.

[0059] S303: With the goal of maximizing the fitness function value, the combination of failure precursor features is gradually optimized using the Zebra Optimization Algorithm to determine the optimal combination of failure precursor features.

[0060] Optionally, the Zebra optimization algorithm specifically includes:

[0061] Initialize the positions of all zebras, where the position of each zebra represents a combination of pre-failure warning features.

[0062] Calculate the fitness value of each zebra based on the fitness function.

[0063] Based on the fitness value, combined with the Levy flight mechanism and adaptive weighting strategy, update the position of each zebra:

[0064]

[0065]

[0066]

[0067] in, Let w(t) represent the updated position of the u-th zebra after the general phase P in the v-th dimension, and let w(t) represent the weight coefficient at the t-th iteration. uv This represents the position of the u-th zebra in dimension v, where r represents the control factor, and PZ u Let represent the u-th zebra as the lead zebra, and I represent the coefficient indicating the direction of guidance. This represents the new position of the u-th zebra after foraging phase P1 in dimension v, x. min Denotes the lower bound of the v-th dimension variable, x maxLet represent the upper bound of the v-th dimension variable, s represent the random step size, e represent the natural constant, T represent the maximum number of iterations, q and d both represent random variables that follow a normal distribution, and λ represent the distribution exponent.

[0068] In this embodiment of the invention, Levy flight helps the zebra escape local optima in the search space, ensuring that the algorithm can find the global optimum. This is crucial for feature selection problems, as the feature space can be very complex and diverse, and local optima may lead to incomplete feature selection, thus affecting the performance of the prediction model. Simultaneously, the adaptive weight strategy can dynamically adjust the search strategy based on the performance of the current solution during iteration. This means that the zebra optimization algorithm can automatically adjust weights according to the progress of feature selection, thereby increasing attention to important features.

[0069] To prevent getting trapped in local optima, the positions of each zebra are further updated by combining the sine and cosine algorithms and the variable spiral search algorithm:

[0070]

[0071]

[0072] Where S1 represents the escape strategy and S2 represents the attack strategy. Let B represent the new position of the u-th zebra after defense phase P2 in dimension v, where B represents the helical coefficient, q1 represents the sine and cosine amplitude adjustment coefficients, q2 and q3 are both random numbers, R represents the perturbation control constant, sin represents the sine function, cos represents the cosine function, and BZ represents the helical coefficient. u This indicates the position of the vanguard zebra in an offensive strategy.

[0073] In this embodiment of the invention, by introducing escape and attack strategies, the Zebra Optimization algorithm can effectively explore a vast search space and avoid stagnation near local optima. The escape strategy increases the algorithm's exploratory nature through random perturbations, while the attack strategy helps the algorithm better approach the optimal solution, thereby improving the optimization's global search capability.

[0074] Calculate the fitness value of each zebra after the update.

[0075] If the fitness value of the updated zebra is greater than or equal to the fitness value of the current zebra, update the current zebra's position. If the fitness value of the updated zebra is less than the fitness value of the current zebra, keep the current zebra's position unchanged.

[0076] Repeat the above steps until the maximum number of iterations is reached, and output the failure precursor feature combination represented by the zebra with the highest fitness value as the optimal failure precursor combination.

[0077] In this embodiment of the invention, the Genetic Zebra Algorithm can automatically select the most predictive combination from a large number of candidate features without human intervention or assumptions. Through automated selection, human bias and omissions are reduced, enabling more efficient and comprehensive discovery of potentially important features. Simultaneously, by calculating the correlation between features, the Zebra Algorithm can effectively eliminate redundant features during the optimization process and ensure that the selected feature combinations minimize multicollinearity among features, thereby improving the accuracy and reliability of predictions.

[0078] S4: Input the optimal failure precursor feature combination into the hierarchical partitioning MCLSTM network to output hierarchical temporal degradation features.

[0079] Among them, the Multilevel Classifier Long Short-Term Memory Network (MCLSTM) is an improved LSTM (Long Short-Term Memory) network. Its characteristic is that it processes data in layers, combining information from multiple levels to better capture features at different levels when processing time-series data. MCLSTM networks are commonly used for processing complex time-series data tasks, such as time series forecasting, fault diagnosis, and equipment lifetime prediction.

[0080] Among them, Hierarchical Temporal Degradation Features refer to the process of independently processing and combining data at different levels when processing time-series data to capture long-term trends, short-term fluctuations, periodic changes, and sudden anomalies in the equipment degradation process.

[0081] In one possible implementation, the MCLSTM network includes hierarchical partitioning units and multi-cell units. S4 specifically includes:

[0082] S401: The optimal failure precursor feature combination is divided into high-level data, medium-high-level data, medium-level data, medium-low-level data, and low-level data by using the SoftMax activation function in the level division unit.

[0083] It should be noted that high-level data typically represents the most critical long-term trends for equipment health and fault diagnosis, usually including long-term degradation characteristics from the initial stage of equipment operation to the present, such as long-term performance degradation trends and cumulative changes in core parameters. Mid-to-high-level data typically represents some relatively long-term characteristics or changes, but not as significant as high-level data, such as large load variations and periodic degradation phenomena. Mid-level data typically represents the changing trends of equipment under normal operating conditions, reflecting performance fluctuations during medium-term operation, such as slight degradation during routine operation. Mid-to-low-level data typically reflects the performance of equipment under certain minor faults or instabilities, such as intermittent minor faults and slight fluctuations or instabilities. Low-level data typically represents detailed characteristics of equipment under short-term or transient operating conditions, such as voltage fluctuations and transient current changes.

[0084] In this embodiment of the invention, high-level data captures long-term degradation trends, mid-to-high-level data captures relatively long-term but insignificant changes, mid-level data represents slight degradation during normal operation, and low-level data reflects short-term transient fluctuations. By processing these different levels of data hierarchically, MCLSTM can simultaneously extract features from different time scales and different levels of information, thereby enhancing the model's sensitivity to multiple types of degradation and failure modes.

[0085] S402: Update data at different levels using multi-cell units, applying different update rules:

[0086]

[0087]

[0088]

[0089]

[0090]

[0091] Among them, c t (m1) represents the state of high-level data at time t, c t (m2) represents the state of the high-level data at time t, s1 represents the control parameters during the update of the high-level data, and f t c represents the output of the forget gate at time t. t-1 This represents the hidden state at time t-1. This represents the output of the input gate at time t. This represents element-wise multiplication. c represents the candidate state at time t. t (m3) represents the state of the rank data at time t, ct (m4) represents the state of the low-level data at time t, s2 represents the control parameters when updating the low-level data, and c t (m5) represents the state of the low-level data at time t, tanh represents the activation function, and W... c This represents the weight matrix related to the input. U represents the input data at time t. c This represents the output weight matrix, h. t-1 This represents the output at time t-1, b c This indicates the bias term.

[0092] It should be noted that each cell updates its state and adjusts its state and output based on the current input, the previous state, and the gating mechanism.

[0093] In this embodiment of the invention, by designing different control parameters and update rules for each data level, MCLSTM can assign different weights and importance to features at each level, thereby enabling the model to better capture the important information of features at each level. Simultaneously, based on the characteristics of the data at each level, the model's update strategy is dynamically adjusted, thereby enhancing the model's prediction accuracy, generalization ability, fault diagnosis ability, and ability to balance long-term and short-term information. This refined hierarchical processing approach allows the model to handle complex equipment degradation processes and capture degradation information at multiple levels, thus providing a more accurate basis for equipment lifespan prediction and fault diagnosis.

[0094] S403: Weighted combination of update data at each level in the multicellular unit yields hierarchical temporal degradation features:

[0095]

[0096] Among them, H t The stratified temporal degradation characteristics at time t are represented by w1, w2, w3, w4, and w5. w1 represents the weight represented by the state of high-level data at time t. w2 represents the weight represented by the state of high-level data at time t. w3 represents the weight represented by the state of mid-level data at time t. w4 represents the weight represented by the state of low-level data at time t. w5 represents the weight represented by the state of low-level data at time t.

[0097] S5: Extract auxiliary failure precursor features from environmental data to characterize the lifespan of power devices.

[0098] Among them, auxiliary failure precursor features refer to auxiliary data features closely related to equipment health and failure. These features do not directly reflect the main operating parameters of the equipment (such as current, voltage, temperature, etc.), but provide additional information about the equipment's operating environment, working conditions, or external factors, helping to improve the accuracy of failure prediction. These auxiliary features usually come from non-electrical data sources such as environmental monitoring, mechanical stress, and equipment status, but they are of great significance in predicting equipment performance degradation and failure.

[0099] In this embodiment of the invention, extracting auxiliary failure precursor features from environmental data can provide multi-dimensional support for equipment failure prediction, supplementing information that electrical data cannot fully reflect. This process can improve the accuracy of failure prediction, enhance the comprehensiveness of equipment health monitoring, reduce maintenance costs, improve equipment availability, and effectively identify hidden failure risks that electrical data fails to reveal.

[0100] S6: The hierarchical timing degradation features and auxiliary failure precursor features are fused to obtain the fused features of the power device.

[0101] In one possible implementation, S6 specifically includes:

[0102] S601: Perform temporal alignment and spatial normalization on the hierarchical temporal degradation features and auxiliary failure precursor features to obtain the target hierarchical temporal degradation features and target auxiliary failure precursor features:

[0103]

[0104] in, This represents the hierarchical temporal degradation characteristics of the target at time t. The LayerNorm represents the target-aided failure precursor features at time t, and H represents the layer normalization. t A represents the hierarchical temporal degradation characteristics at time t. t This represents the auxiliary failure precursor characteristics at time t.

[0105] In this embodiment of the invention, temporal alignment and spatial normalization processing can ensure the consistency of hierarchical temporal degradation features and auxiliary failure precursor features in time and space, avoiding information loss caused by different features due to different scales or inaccurate time alignment.

[0106] S602: Through a bimodal attention mechanism, the hierarchical temporal degradation features of the target and the auxiliary failure precursor features of the target are dynamically quantified, and the dynamic attention weights of the hierarchical temporal degradation features and the auxiliary failure precursor features of the target are determined.

[0107]

[0108] Where, α h The attention weights represent the temporal degradation features of the target, σ represents the sigmoid activation function, and W... h The weight matrix b represents the target temporal degradation characteristics. h The bias term α represents the temporal degradation characteristics of the target. a W represents the attention weights of the target-assisted failure precursor features. a The weight matrix b represents the target-assisted failure precursor features. a The bias term represents the characteristics of the precursory features of the target's auxiliary failure.

[0109] S603: Based on the dynamic attention weights, target hierarchical temporal degradation features, and target auxiliary failure precursor features, feature cross-enhancement is performed through a gating mechanism to obtain cross-enhanced features:

[0110]

[0111] Among them, F t This represents the cross-enhancement feature at time t. This represents element-wise multiplication. Indicates feature splicing, Gating mechanisms that represent the temporal degradation characteristics of the target. Gating mechanisms that represent the precursory characteristics of target auxiliary failure.

[0112] In this embodiment of the invention, feature cross-enhancement through a gating mechanism can enhance the interrelationships between different features, thereby fully exploring the potential connections between two types of features. This cross-enhancement helps the model capture complex interactions between features, further improving predictive capabilities.

[0113] S604: Based on the cross-enhancement features, the original feature information is preserved through residual connections, and the final normalization process is performed to obtain the fused features.

[0114] In one possible implementation, the fusion feature specifically includes:

[0115]

[0116] Among them, F fusion Denotes the fusion feature, LayerNorm denotes layer normalization, F t This represents the cross-enhancement feature at time t, and Dropout represents the regularization technique. This represents the hierarchical temporal degradation characteristics of the target at time t. This indicates the precursory features of the target auxiliary failure at time t. This represents the target temporal degradation characteristics and target auxiliary failure precursor characteristics at time t after splicing.

[0117] In this embodiment of the invention, residual connections allow the original feature information to be directly passed to subsequent layers, avoiding the "vanishing gradient" problem that may occur in deep networks, while preserving the key information of the original features. This means that even through complex feature fusion and transformation operations, the model can still retain important information from the initial data in the final output, thereby improving the stability and interpretability of the model.

[0118] S7: Based on the fusion characteristics, the power device lifetime prediction results are output through a fully connected network.

[0119] In one possible implementation, the fully connected network includes an input layer, a hidden layer, and an output layer. S7 specifically includes:

[0120] S701: Through each neuron in the input layer, the feature values ​​in the fused features are passed to the hidden layer in sequence.

[0121] S702: The neurons in the hidden layer perform a weighted summation of the feature values ​​in the input layer and then perform non-linear activation using the ReLU activation function to obtain the hidden layer features.

[0122] S703: Based on the hidden layer features, the class probability of the output power device's lifetime is predicted using the Softmax function in the output layer.

[0123]

[0124] in, Let represent the probability, and y represent the predicted category. This represents the probability of predicting a class as the g-th class given an input feature vector, where g represents the class label, g=1 represents the short-lived class, g=2 represents the medium-lived class, g=3 represents the long-lived class, and z represents the fused feature. This indicates the probability of being predicted to have a short lifespan. This indicates the probability of being predicted as a medium-life category. Let z represent the probability of being predicted as belonging to the long-life category, e denotes the exponential function, and z represent the probability of being predicted as belonging to the long-life category. g This represents the original predicted score for category g. This represents the summation of all original predicted scores after applying an exponential function, where p represents the index of the predicted category.

[0125] In this embodiment of the invention, the Softmax function transforms the network's output into predicted probabilities for each category. This provides not only the predicted category (short-term, medium-term, long-term lifetime) of the power device but also the confidence level for each category. Thus, the model not only provides the final classification result but also reflects its confidence in predicting each category. This facilitates more flexible decision-making in practical applications.

[0126] S704: Select the category with the highest category prediction probability as the power device lifetime prediction result, where the lifetime prediction result includes short-term lifetime, medium-term lifetime and long-term lifetime.

[0127] In this embodiment of the invention, selecting the category with the highest probability as the prediction result makes the model's output more concise and easier to understand. This method clearly indicates the expected lifespan of the equipment, thereby helping equipment managers or decision-makers make more reasonable judgments.

[0128] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0129] In this embodiment of the invention, by extracting multiple key pre-failure features and auxiliary features from electrical and environmental data, the critical changes in the power device degradation process can be comprehensively captured, avoiding the limitations of traditional methods in feature selection. Especially when facing multivariate and high-dimensional data, it can effectively identify potential influencing factors, thereby significantly improving prediction accuracy. Simultaneously, by combining a hierarchical partitioning-based MCLSTM network and a fully connected network, the comprehensive application of hierarchical temporal degradation features and fused features enhances the model's generalization ability, enabling the model to not only perform well on specific datasets but also maintain high prediction accuracy across different types of power devices and operating environments.

[0130] Reference manual attached Figure 2 The diagram shows a schematic of a power device lifespan prediction system provided in an embodiment of the present invention.

[0131] This invention provides a power device lifetime prediction system 20, including: a processor 201 and a memory 202;

[0132] The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the power device lifespan prediction method described above and achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0133] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0134] It should also be understood that the memory 202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0135] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0136] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0137] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0138] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0139] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0140] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0141] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0142] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0143] This invention provides a readable storage medium comprising: storing a program or instructions on the readable storage medium, wherein when the program or instructions are executed by a processor, the program or instructions implement the steps of the power device lifespan prediction method described above, and can achieve the same technical effect. To avoid repetition, this invention will not elaborate further.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the lifespan of power devices, characterized in that, include: S1: Acquire the electrical data and environmental data of the power device; S2: Extract several key precursor features from the electrical data to characterize the lifespan of the power device; S3: Select and combine the key failure precursor features to generate the optimal failure precursor feature combination; S4: Input the optimal failure precursor feature combination into the hierarchical partitioning MCLSTM network to output hierarchical temporal degradation features; S5: Extract auxiliary failure precursor features from the environmental data to characterize the lifespan of the power device; S6: The hierarchical timing degradation features and the auxiliary failure precursor features are fused to obtain the fused features of the power device; Specifically, S6 includes: S601: Perform time dimension alignment and spatial normalization processing on the hierarchical temporal degradation features and the auxiliary failure precursor features to obtain the target hierarchical temporal degradation features and the target auxiliary failure precursor features; S602: Through a bimodal attention mechanism, dynamically quantify the target hierarchical temporal degradation features and the target auxiliary failure precursor features, and determine the dynamic attention weights of the target hierarchical temporal degradation features and the target auxiliary failure precursor features; S603: Based on the dynamic attention weights, the target hierarchical temporal degradation features, and the target auxiliary failure precursor features, feature cross-enhancement is performed through a gating mechanism to obtain cross-enhanced features; S604: Based on the cross-enhancement features, the original feature information is preserved through residual connections, and the final normalization process is performed to obtain the fused features; S7: Based on the fusion characteristics, output the predicted lifespan of the power device through a fully connected network.

2. The method for predicting the lifespan of power devices according to claim 1, characterized in that, The electrical data includes source-drain resistance, leakage current, input capacitance, output capacitance, case temperature, current waveform, voltage waveform, insulation resistance, switching frequency, and switching cycle. The environmental data includes: ambient temperature, ambient humidity, air pollution and dust, as well as vibration and mechanical stress.

3. The method for predicting the lifespan of power devices according to claim 2, characterized in that, S2 specifically includes: S201: Set the environmental conditions and number of operating cycles for the accelerated degradation of the power device in the accelerated aging test; S202: Record the initial state of the power device, wherein the initial state includes the initial values ​​of each electrical parameter in the electrical data; S203: The power device is operated under the set environmental conditions and the number of working cycles, and the values ​​of each of the electrical parameters are measured periodically in each cycle to obtain multiple cycle values; S204: Perform statistical analysis on the initial values ​​of each electrical parameter and the multiple periodic values ​​respectively to determine the changing trend of each electrical parameter; S205: Based on the aforementioned trends, determine the maximum amplitude of change for each of the aforementioned electrical parameters; S206: Sort the maximum change amplitudes in descending order and select the electrical parameter corresponding to the highest-ranking maximum change amplitude as the key failure precursor feature.

4. The method for predicting the lifespan of power devices according to claim 1, characterized in that, S3 specifically includes: S301: Through the initialization process of the Zebra Optimization Algorithm, an initial population of multiple combinations of the aforementioned failure precursor features is initially generated; S302: Determine the fitness function used to evaluate the combination of failure precursor features; S303: With the goal of maximizing the function value of the fitness function, the zebra optimization algorithm is used to progressively optimize each combination of failure precursor features to determine the optimal combination of failure precursor features.

5. The method for predicting the lifespan of power devices according to claim 1, characterized in that, The MCLSTM network includes hierarchical partitioning units and multi-cell units; S4 specifically includes: S401: Using the SoftMax activation function in the classification unit, the optimal failure precursor feature combination is divided into high-level data, medium-high-level data, medium-level data, medium-low-level data, and low-level data. S402: Through the multi-cell unit, different update rules are applied to update the data at each level; S403: The updated data of each level in the multi-cell unit are weighted and combined to obtain the hierarchical temporal degradation features.

6. The method for predicting the lifespan of power devices according to claim 1, characterized in that, The fusion features are specifically: ； Among them, F fusion Denotes the fusion feature, LayerNorm denotes layer normalization, F t This represents the cross-enhancement feature at time t, and Dropout represents the regularization technique. This represents the hierarchical temporal degradation characteristics of the target at time t. This indicates the precursory features of the target auxiliary failure at time t. This represents the target temporal degradation characteristics and target auxiliary failure precursor characteristics at time t after splicing.

7. The method for predicting the lifespan of power devices according to claim 1, characterized in that, The fully connected network includes an input layer, a hidden layer, and an output layer; S7 specifically includes: S701: Through each neuron in the input layer, the feature values ​​in the fused features are sequentially passed to the hidden layer; S702: The neurons in the hidden layer perform a weighted summation of each feature value in the input layer and perform non-linear activation through the ReLU activation function to obtain the hidden layer features; S703: Based on the hidden layer features, output the category prediction probability of the power device's lifetime through the Softmax function in the output layer; S704: Select the category with the highest category prediction probability as the lifespan prediction result of the power device, wherein the lifespan prediction result includes short-term lifespan, medium-term lifespan and long-term lifespan.

8. A power device lifespan prediction system, characterized in that, include: Processor and memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the power device lifetime prediction method as described in any one of claims 1 to 7.

9. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the power device lifetime prediction method as described in any one of claims 1 to 7.

Images