Power utilization information acquisition real-time monitoring and service life prediction method for IGBT (Insulated Gate Bipolar Translator)

Abstract

The invention discloses an IGBT power utilization information acquisition real-time monitoring and service life prediction method, relates to the technical field of wireless communication and data processing, and solves the problems of poor data quality and real-time performance, long model training time and incapability of dynamically adjusting parameters in the existing real-time monitoring and service life prediction method. Comprising the following steps: step 1, carrying out data acquisition and preprocessing to obtain working information of an IGBT in real time; 2, extracting effective features; 3, constructing a life prediction model; 4, performing dynamic adjustment and optimization; 5, early warning and maintenance are carried out; step 6, constructing an intelligent analysis system; through the autoregression moving average algorithm and the transfer learning method, the model training time and the data demand are reduced; the service condition and the load of the IGBT are optimized through a self-adaptive parameter adjustment algorithm, so that the service life of the IGBT is prolonged, and the damage risk is reduced; according to the method, the data acquisition real-time performance and the prediction accuracy are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data monitoring, and more specifically to a method for real-time monitoring of power consumption information collection and service life prediction of an IGBT. Background Art

[0002] IGBT (Insulated-gate Bipolar Transistor) is a bidirectional thyristor, also known as a thyristor. Unlike traditional silicon thyristors, IGBT has bidirectional controllability and can work under reverse voltage. In a specific embodiment, IGBT is a commonly used power semiconductor device, which is widely used in industrial power and electronics. In order to ensure the reliability and safety of the power system, it is very important to conduct real-time monitoring and life prediction evaluation of IGBT. Among them, IGBT is generally used in power supply, motor control, electronic equipment, welding machine and other fields. They have the characteristics of high input impedance, low noise, low distortion, fast response, etc., so they are widely used. The structure includes an adjustable gate, a thyristor at the input stage and an insulated gate bipolar transistor at the output stage. There is also a control circuit of the thyristor between the input stage and the output stage. By controlling the control circuit, the switching control of the IGBT can be realized to achieve the required circuit function.

[0003] In specific applications, IGBT is a power switching device with high voltage, high current and high frequency characteristics. It combines the control capability of metal oxide semiconductor field effect transistor and the low on-state voltage drop characteristic of power BJT bipolar transistor. However, its monitoring capability has great challenges, such as: Data quality issues: Existing methods have certain defects in data collection and preprocessing. The raw data collected by the sensor may contain outliers and missing values, and existing methods fail to effectively remove or fill these problems, resulting in inaccurate subsequent analysis and prediction results.

[0004] Long model training time and high data requirements: The life prediction models in existing methods often require a large amount of data for training, and the training time is long. This limits the operability and flexibility of the method in practical applications to a certain extent.

[0005] Unable to dynamically adjust parameters: Existing methods rarely consider the impact of IGBT's use conditions and load on its life, and lack the ability to dynamically adjust parameters. This results in the inability to optimize according to real-time conditions in actual working environments, thus limiting the maximization of IGBT's service life.

[0006] Therefore, in order to solve the problems of poor data quality, long model training time and high data requirements, and inability to dynamically adjust parameters in existing IGBT real-time monitoring and service life prediction and evaluation methods; the present invention discloses an IGBT real-time monitoring and service life prediction and evaluation method. Summary of the invention

[0007] In view of the shortcomings of the prior art, the present invention discloses a method for real-time monitoring and service life prediction and evaluation of IGBTs. The present invention collects data from temperature sensors, current sensors, vibration sensors and voltage sensors, removes outliers, fills missing values ​​and standardizes the data, thereby improving the analyzability and predictability of the data; solves the problem of poor data quality and inability to accurately analyze and predict in the existing methods; extracts the working state characteristics of the IGBT by training the feature extraction model, including temperature change rate, power loss and current impact parameters, etc., to better reflect the working state of the IGBT, and solves the problem of inaccurate feature extraction and inability to effectively describe the state of the IGBT in the existing methods; constructs a life prediction model by an autoregressive sliding average algorithm and a transfer learning method, thereby reducing the training time and data requirements, and improving the flexibility of model application, and solving the problem of long model training time and strong dependence on a large amount of data in the existing methods; optimizes the use conditions and loads of the IGBT by an adaptive parameter adjustment algorithm and a reinforcement learning framework, and dynamically adjusts the working parameters and working state of the IGBT to extend the service life and reduce the risk of failure and damage. This solves the problem that the existing methods cannot be dynamically adjusted according to the actual working conditions.

[0008] In order to achieve the above technical effects, the present invention adopts the following technical solutions: A method for real-time monitoring of power consumption information collection and service life prediction of IGBT, wherein the method comprises: As a further technical solution of the present invention, the following steps are included: Step 1: Data collection and preprocessing; The working parameter data of the IGBT of the electric information terminal is collected in real time through temperature sensors, current sensors, vibration sensors and voltage sensors, and the analyzability and predictability of the original data are improved by removing abnormal values, filling missing values ​​and standardizing processing methods; Step 2: Feature extraction; The IGBT working state characteristics of the power information terminal are extracted by training the feature extraction model; the IGBT working state characteristics at least include temperature change rate, power loss and current impact parameters; Step 3: Construct a lifespan prediction model; The extracted data is analyzed and modeled by an autoregressive sliding average model to construct a life prediction model; the life prediction model migrates and adjusts the existing life prediction model on different devices by a transfer learning method to reduce training time and data requirements; Step 4: Real-time data stream processing; The collected real-time data stream is processed and analyzed by an edge processing model; the edge processing model updates the feature extraction model and the life prediction model in real time through a streaming data processing engine; Step 5: Dynamic adjustment and optimization; The use conditions and loads of IGBTs are optimized through adaptive parameter adjustment algorithms, and the working parameters and working states of IGBTs are dynamically adjusted through a reinforcement learning framework to extend the service life of IGBTs. Step 6: Early warning and maintenance; The IGBT is monitored in real time by a real-time monitoring and alarm system, which monitors the IGBT through a real-time data analysis engine and automatically triggers an alarm through a rule engine to prevent the occurrence of potential faults; the rule engine includes a temperature threshold unit, a current threshold unit, and a vibration threshold unit; Step 7: Build an intelligent analysis system; The intelligent analysis system uses a machine learning model to detect abnormal conditions during the operation of the IGBT in real time and provide automated fault diagnosis and solutions.

[0009] As a further technical solution of the present invention, the working mode of the transfer learning method is: S1. Select source domain and target domain; The source domain compares the similarities of different domains in data distribution, feature space and model parameters by a data comparison method to select a domain for which a life prediction model has been trained. The target domain selects an IGBT life prediction domain by observing and collecting data, features and model parameter information in the domain. S2, extract feature representation; Extract features from data in the source and target domains through convolutional neural networks; S3, establish a migration model; Establish a transfer model through transfer learning and domain adaptation methods to transfer knowledge from the source domain to the target domain; S4, transfer learning training; The model training method uses the source domain data to train the migration model. During the training process, dropout regularization is used to improve the generalization ability of the model. S5, migration and adjustment; The trained transfer model is applied to the data of the target domain and fine-tuned for the target domain through learning rate adjustment and weight freezing methods.

[0010] As a further technical solution of the present invention, the feature extraction model realizes real-time processing and analysis of the collected IGBT data through a signal processing module and a real-time analysis module. The signal processing module realizes signal enhancement, noise removal and frequency domain feature extraction of the IGBT data through filtering, spectrum analysis, correlation analysis, Fourier transform or wavelet transform operations; the real-time analysis module performs real-time analysis of the data through filters and time domain and frequency domain analysis methods.

[0011] As a further technical solution of the present invention, the autoregressive moving average model includes an autoregressive module, a moving average module and a difference module; the autoregressive module reflects the influence of past observations on current values ​​through an autoregressive equation, and the formula expression of the autoregressive equation is: (1) In formula (1), R represents the inertia weight, and m represent lag operators, S represents the number of differences; the sliding average module reflects the impact of the error of past observations on the current value through the error prediction method; the difference module performs multi-order differences on the time series data through the difference method, and converts the non-stationary series into a stationary series through the sequence conversion function. The formula expression of the sequence conversion function is: (2) In formula (2), Represents the function sequence conversion value, Represents the order of sequence transformation by the function, and d represents the sequence transformation error term; the autoregressive order, sliding average order and difference order are adjusted by the autoregressive sliding model function to achieve the fitting and prediction of time series data; the formula expression of the autoregressive sliding model function is: (3) In formula (3), H represents the autoregressive sliding model function; c represents the truncation window size, i represents the sliding average coefficient, and j represents the error term variance; the autoregressive sliding average model prediction establishes a model through historical time series data, and evaluates the remaining life of the IGBT through fitting and prediction.

[0012] As a further technical solution of the present invention, the adaptive parameter adjustment algorithm realizes the process of parameter optimization and model fitting by adjusting model parameters and optimization indicators. The adaptive parameter adjustment algorithm at least includes parameter optimization, feature selection and interface optimization. The formula of the adaptive parameter adjustment algorithm is as follows: For each particle , the particle position is expressed as =( 1, 2, ..., M), the particle speed is expressed as =( 1, 2, ..., M), where is the dimension of the problem; each particle Update according to the current position and speed, the relationship expression is: (4) (5) In formula (4) and formula (5), is the inertia weight, and They are individual learning factors and social learning factors. and is a random number between 0 and 1; It is a particle The best location ever. is the best position that the entire particle group has ever experienced; The optimal solution is obtained by iteratively updating the particle position and velocity until the convergence condition is met.

[0013] As a further technical solution of the present invention, the rule engine automatically monitors the real-time status of the IGBT through threshold setting, real-time monitoring, status judgment and alarm operation. The threshold setting sets the temperature limit threshold, current limit threshold, power factor threshold and loss index threshold through training sample data. The real-time monitoring automatically compares the data of the data acquisition system with the set threshold data through a data comparison method, and triggers an alarm through status judgment.

[0014] As a further technical solution of the present invention, the intelligent analysis system includes a data monitoring module, a data processing module, a data analysis module, a reasoning and decision module and a user interface module; the data monitoring module obtains the monitoring data of the IGBT in real time through a temperature sensor, a current sensor and a voltage sensor; the data processing module includes a storage unit and a processing unit; the storage unit manages and stores the collected data through a cloud database; the processing unit processes the data through data cleaning, data conversion and data aggregation operations to improve the quality and availability of the data; the data analysis module analyzes the collected data through statistical analysis, signal processing and pattern recognition methods to obtain patterns, trends and abnormal conditions in the data; the reasoning and decision module reasons and makes decisions on the analysis data through knowledge representation and reasoning engine; the user interface module converts the data into a visual form through a data visualization method and a chart library to display the decision results; the output end of the data monitoring module is connected to the input end of the data processing module; the output end of the data processing module is connected to the input end of the data analysis module; the output end of the data analysis module is connected to the input end of the reasoning and decision module; the output end of the reasoning and decision module is connected to the input end of the user interface module. As a further technical solution of the present invention, the cloud database includes a distributed file system and a cloud archiving module. The distributed file system stores data in a dispersed manner on multiple nodes through a data storage method to achieve scalability and redundant backup of data storage; the cloud archiving module processes, aggregates and analyzes the monitoring data of the IGBT through a mapping-reduction function.

[0015] As a further technical solution of the present invention, the edge processing model includes a data stream receiving module, a streaming data processing module, a real-time monitoring and alarm module and an edge computing module; the data stream receiving module receives real-time data streams from temperature sensors, current sensors, vibration sensors and voltage sensors through an Internet of Things sensor network; the streaming data processing engine module includes a feature extraction model updating unit and a life prediction model updating unit; the feature extraction model updating unit extracts and updates features in the real-time data stream through an online learning method; the life prediction model updating unit updates the life prediction model in real time through a time series analysis method, and the time series analysis method analyzes and models through an autoregressive sliding average model; the real-time monitoring and alarm module includes a real-time alarm unit and a real-time data analysis unit; the real-time alarm unit realizes real-time monitoring of temperature, current and voltage and triggers an alarm through a rule engine; the real-time data analysis unit performs instant analysis and processing of real-time data through parallel computing and streaming data processing methods; the edge computing module uses local computing resources through a distributed architecture and intelligent scheduling and resource management methods to process and analyze data in real time.

[0016] Positive beneficial effects: The present invention collects data from temperature sensors, current sensors, vibration sensors and voltage sensors, removes outliers, fills missing values ​​and performs standardization on the data, thereby improving the analyzability and predictability of the data; solves the problem of poor data quality and inability to accurately analyze and predict in the existing methods; extracts the working state characteristics of the IGBT, including temperature change rate, power loss and current impact parameters, etc., by training the feature extraction model, better reflects the working state of the IGBT, and solves the problem of inaccurate feature extraction and inability to effectively describe the state of the IGBT in the existing methods; constructs a life prediction model by an autoregressive sliding average algorithm and a transfer learning method, thereby reducing the training time and data requirements, and improving the flexibility of model application, and solving the problem of long model training time and strong dependence on a large amount of data in the existing methods; optimizes the use conditions and loads of the IGBT by an adaptive parameter adjustment algorithm and a reinforcement learning framework, and dynamically adjusts the working parameters and working state of the IGBT to extend the service life and reduce the risk of failure and damage. This solves the problem that the existing methods cannot be dynamically adjusted according to the actual working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work, among which: Figure 1 It is a schematic diagram of the process steps of the present invention; Figure 2 is a machine learning flow chart of the present invention; Figure 3 The prediction principle of the autoregressive moving average model ARIMA of the present invention; Figure 4 is a flow chart of the steps of the transfer learning method of the present invention; Figure 5 This is a working principle framework diagram of the intelligent analysis system of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] according to Figure 1-Figure 5 As shown: A method for real-time monitoring and service life prediction of IGBT power consumption information collection, comprising the following steps: Step 1: Data collection and preprocessing; The working parameter data of the IGBT of the electric information terminal is collected in real time through temperature sensors, current sensors, vibration sensors and voltage sensors, and the analyzability and predictability of the original data are improved by removing abnormal values, filling missing values ​​and standardizing processing methods; Step 2: Feature extraction; The working state characteristics of the IGBT of the power information terminal are trained and extracted through the feature extraction model; the working state characteristics of the IGBT include at least the temperature change rate, power loss and current impact parameters; feature extraction is a key step in the state monitoring and fault diagnosis of power electronic equipment. For the feature extraction of the working state of the insulated gate bipolar transistor (IGBT), it can be achieved by establishing a feature extraction model. The following are the general steps to achieve this function: In specific applications, sensors and analog-to-digital converters (ADCs) are used to collect key parameters of the IGBT in real time, such as current, voltage, temperature and switching frequency. The collected data needs to include data under normal working conditions and different fault conditions in order to establish a comprehensive feature library. The collected data is filtered, denoised and normalized to improve the data quality. It may also be necessary to perform feature selection on the data to remove redundant or unimportant features. Select a suitable feature extraction algorithm, such as feature extraction based on wavelet transform, time-frequency analysis, principal component analysis (PCA), etc. Use machine learning algorithms, such as support vector machines (SVM), random forests (RF) or deep learning methods, to train the feature extraction model. When training the model, you need to use annotated data sets that contain IGBT operating data in normal and faulty states. Evaluate the performance of the model through methods such as cross-validation. Adjust and optimize the model based on the evaluation results to improve the accuracy and robustness of feature extraction. Use the trained model to extract features from the real-time collected data. Use the extracted features for applications such as condition monitoring, fault diagnosis, or service life prediction. Apply the extracted features in real time, combined with prior knowledge and experience, to perform fault diagnosis and condition assessment. Take appropriate maintenance and protection measures based on the diagnosis results. Effectively extract key features that reflect the working status of the IGBT from the power consumption information, thereby achieving real-time monitoring and assessment of the health status of the IGBT.

[0020] Step 3: Construct a lifespan prediction model; The extracted data is analyzed and modeled by an autoregressive sliding average model to construct a life prediction model; the life prediction model migrates and adjusts the existing life prediction model on different devices by a transfer learning method to reduce training time and data requirements; Step 4: Real-time data stream processing; The collected real-time data stream is processed and analyzed by an edge processing model; the edge processing model updates the feature extraction model and the life prediction model in real time through a streaming data processing engine; Step 5: Dynamic adjustment and optimization; The use conditions and loads of IGBTs are optimized through adaptive parameter adjustment algorithms, and the working parameters and working states of IGBTs are dynamically adjusted through a reinforcement learning framework to extend the service life of IGBTs. Step 6: Early warning and maintenance; The IGBT is monitored in real time by a real-time monitoring and alarm system, which monitors the IGBT through a real-time data analysis engine and automatically triggers an alarm through a rule engine to prevent the occurrence of potential faults; the rule engine includes a temperature threshold unit, a current threshold unit, and a vibration threshold unit; Step 7: Build an intelligent analysis system; The intelligent analysis system uses a machine learning model to detect abnormal conditions during the operation of the IGBT in real time and provide automated fault diagnosis and solutions.

[0021] In a specific embodiment, the multimodal data acquisition system may be arranged near the IGBT workplace, or arranged in the IGBT workplace via a support frame; In step 2, data information features can be collected through the hardware structure, or the hardware data features can be input into the learning algorithm model to learn and train the input data information to improve data information analysis and application capabilities; the arithmetic operation software MATLAB can convert the acquired data information into micro data analysis through data information calculation to improve data analysis and application capabilities; In the specific embodiment of step three, the module can be set with reference to it; the autoregressive moving average model ARIMA is a commonly used time series modeling method, which can be used to model the autoregressive moving average model. The autoregressive moving average model is a commonly used model to describe time series, in which autoregression refers to the linear relationship between independent variables in the time series, and the moving average refers to the average of variables in the time series in different ways. The model usually contains three links: autoregression, sliding and averaging. AR is the autoregressive link, which represents the lag term of the independent variable; MA is the sliding link, which indicates that the value of the independent variable has changed smoothly; ARMA is the autoregressive moving average link, which combines AR and MA, taking into account both the characteristics of autoregression and the characteristics of sliding. The ARIMA model can be used to model the autoregressive moving average model to better describe the time series data. In the real-time monitoring and service life prediction and evaluation of IGBT, the autoregressive moving average model ARIMA is used to evaluate the temperature, voltage, current, motion state and other data information during the operation of IGBT. This method has not been seen in other reference documents. In this step, the residual is the difference between the predicted value and the actual observed value. Residual analysis is used to check whether the prediction model can capture the changing trend of IGBT life data. By plotting the residual graph and observing the distribution of the residuals, any systematic errors or anomalies can be found. If the residuals are approximately normal with zero mean and constant variance, the model is likely to be reliable. Error analysis is a quantitative assessment of the error between the life prediction results and the actual observed values. Different indicators, such as the root mean square error (RMSE) or the mean absolute error (MAE), can be calculated to measure the accuracy of the prediction model. Smaller error values ​​indicate that the model is closer to the actual data, while larger error values ​​may indicate that the model has a larger deviation. Survival analysis is a statistical method used to predict reliability and life. It is based on the time of event occurrence and the hazard rate before and after the event (i.e., the probability of an event occurring). In IGBT life prediction, survival analysis methods can be used to estimate the probability that the IGBT may fail at different time points. Survival function and hazard function are commonly used concepts in survival analysis, which can help to comprehensively evaluate the failure risk of IGBT at different time points.

[0022] In step five, since IGBT combines the advantages of MOSFET and ordinary bipolar transistor (BJT) in the application process of power electronics, it has the characteristics of high input impedance, low switching loss and high current carrying capacity, and its operating state is diverse. The adaptive parameter adjustment algorithm is an optimization algorithm based on swarm intelligence, which simulates the collective behavior of organisms such as bird flocks or fish schools. In PSO, the optimal solution of the problem is searched by using the position and velocity of a group of particles. Each particle represents a candidate solution to the problem, and updates the position and velocity according to its own experience and the experience of the group, and gradually optimizes the search process until the optimal solution is found. In the present invention, in the data function optimization, parameter optimization, and model parameter adjustment in machine learning in the operating state of IGBT, it has the advantages of fast convergence speed, easy implementation, and applicability to global optimization. When applying the PSO algorithm, it is necessary to determine the appropriate parameter settings and appropriate stop conditions to ensure the effectiveness and robustness of the algorithm. In the process of data information calculation, it can greatly improve the computing power and application ability of data information, especially when there is a lot of data information, the effect is particularly obvious; In the above embodiment, the working method of the life prediction model includes the following steps: Step 1: Collect and organize the monitoring data and service life data of IGBT through a multimodal data acquisition system, wherein the multimodal data acquisition system includes current data acquisition, temperature data acquisition and vibration data acquisition; wherein the multimodal data acquisition system includes a main control module based on the STM32F103RBT6 chip, a carrier communication module, an isolation module, an uninterrupted information iteration module, a multimodal interface module and a visual display module, wherein the main control module is respectively connected to the carrier communication module, the multimodal interface module, the isolation module, the uninterrupted information iteration module and the visual display module, the main control module is used to control different modules to be in a working state, the carrier communication module is used to transmit the collected data information, the isolation module is used to isolate the collected data information, the monitoring module is used to collect or obtain field data information in real time, the multimodal interface module is used to receive data information in the system, and the visual display module is used to display the monitored data information; In this step, the main control module of the STM32F103RBT6 chip contains a high-performance ARM Cortex -M3 32-bit RISC core, operating at a frequency of 72 MHz, high-speed embedded memory (Flash memory up to 128 Kbytes, SRAM up to 20 Kbytes), and a wide range of enhanced I / O and peripherals connected to two APB buses. All devices provide two 12-bit ADCs, three general-purpose 16-bit timers plus a PWM timer, as well as standard and advanced communication interfaces: up to two I2C and SPI, three USARTs, one USB and one CAN. The devices are powered by a 2.0 ~ 3.6 V power supply. They can be used in a temperature range of -40 to +85°C and an extended temperature range of -40 to +105°C. A comprehensive set of power saving modes allows the design of low-power applications. Control through this module has outstanding technical advantages.

[0023] In a specific embodiment, a carrier communication module is a module for modulating and demodulating carrier communication, which is usually used in digital communication systems. In carrier communication, signals are transmitted by adding an extra wave between the signals, which is called a carrier. The carrier signal is used to modulate the data signal onto the transmission line, thereby generating the movement of the signal. The carrier communication module usually includes one or more carrier input and output ports for modulating and demodulating the signal from input to output. In addition, the carrier communication module usually also includes an input / output comparator for detecting the difference between the input and output signals. The carrier communication module also has some other functions, such as digital signal processing, timing control, and adaptive modulation.

[0024] In a specific embodiment, the uninterrupted information iteration module is configured to cycle the detection information, for example, in a 2-hour, 4-hour, 6-hour or 8-hour cycle.

[0025] In a specific embodiment, a multimodal interface module refers to an interface that allows multiple different types of data sources (such as different sensors, different devices, or different applications) to communicate through the same interface. This interface can allow multiple data sources to share the same data and use this data for collaboration and decision-making. For example, a multimodal interface can allow an application to communicate with multiple sensors in order to obtain information from the data collected by these sensors. This interface can also provide more comprehensive information by integrating multiple sensor data into a central data set to facilitate more in-depth decision-making. Multimodal interfaces promote collaboration and information sharing between different data sources. In a specific embodiment, a compatible new multi-data communication interface can be set. In a specific embodiment, the visualization display module can be a display module for real-time display.

[0026] Step 2: Processing the detected input data, including data normalization, feature selection and feature extraction, to improve model performance; Step 3: Use the API provided by TensorFlow to input the prepared data into the model through the model optimization algorithm module for training; then optimize the model through iterative optimization parameters to achieve more accurate prediction of the real-time state and life of the IGBT; wherein the model optimization algorithm module includes an information recognition module, a feature selection module, a parameter setting module, an evaluation module and a classification module, wherein the information recognition module is used to identify the monitored data information, the feature selection module is used to obtain data information, the parameter setting module is used to set the calculated parameter data information, the evaluation module is used to evaluate and calculate the input data information, and the classification module is used to classify the evaluated data information, wherein the output end of the information recognition module is connected to the input end of the feature selection module, the output end of the feature selection module is connected to the input end of the parameter setting module, the output end of the parameter setting module is connected to the input end of the evaluation module, and the output end of the evaluation module is connected to the input end of the classification module; Step 4: Perform model evaluation, which uses root mean square error to evaluate the performance and prediction accuracy of the model.

[0027] In a specific embodiment, the machine learning method has efficient computing power and a flexible computing graph model. It can handle large-scale data sets and complex computing models, and can meet the computing power requirements in real-time monitoring and service life prediction and evaluation of IGBTs. TensorFlow provides a wealth of APIs and tools that can easily build, train and deploy machine learning models. It supports multiple programming languages ​​and can run on different platforms and devices. It has strong scalability and flexibility and is suitable for various practical application scenarios. At the same time, TensorFlow has built-in many powerful deep learning algorithms and models, including convolutional neural networks, recurrent neural networks, etc. These deep learning models have excellent performance in processing complex data and extracting advanced features. For IGBT real-time monitoring and service life prediction and evaluation, deep learning methods can better mine potential patterns and laws in data. In addition, TensorFlow supports distributed computing, which can distribute large-scale training tasks to multiple computing nodes for parallel processing, thereby improving training speed and efficiency. This is very useful for situations where large amounts of data and complex model training are required in real-time monitoring and service life prediction and evaluation of IGBTs.

[0028] In the above embodiment, the transfer learning method works as follows: S1. Select source domain and target domain; The source domain compares the similarities of different domains in data distribution, feature space and model parameters by a data comparison method to select a domain for which a life prediction model has been trained. The target domain selects an IGBT life prediction domain by observing and collecting data, features and model parameter information in the domain. S2, extract feature representation; Extract features from data in the source and target domains through convolutional neural networks; S3, establish a migration model; Establish a transfer model through transfer learning and domain adaptation methods to transfer knowledge from the source domain to the target domain; S4, transfer learning training; The model training method uses the source domain data to train the migration model. During the training process, dropout regularization is used to improve the generalization ability of the model. S5, migration and adjustment; The trained transfer model is applied to the data of the target domain and fine-tuned for the target domain through learning rate adjustment and weight freezing methods.

[0029] In a specific embodiment, in a method for real-time monitoring and service life prediction and evaluation of IGBTs, transfer learning can reduce training time and data requirements by migrating an existing life prediction model from one device (referred to as a source domain) to another device (referred to as a target domain), and improve the application flexibility of the model on the target device. In the source domain of the pre-training stage of the transfer learning method, large-scale data sets and appropriate life prediction models are used for training. The model can be trained based on domestic open source large models such as Tongyi Qianwen and ChatGLM2. In the transfer learning stage, the model parameters obtained by pre-training in the source domain are used as initial parameters, and part of the data in the source domain or data in a specific field is jointly trained with a small amount of target data in the target domain. By adjusting the model parameters, it is adapted to the characteristics of the target domain data and the needs of specific tasks.

[0030] In specific implementations, due to the small amount of data in the target domain, traditional methods require a large amount of target domain data to effectively train the model. Transfer learning uses the knowledge of the source domain to quickly adapt to the target domain with less target domain data, thereby improving data utilization efficiency. In addition, traditional methods require training models from scratch, which takes a lot of time. Transfer learning can significantly reduce training time by pre-training in the source domain, because some model parameters already have good initial values. Secondly, transfer learning enables the model to be applied to different devices without retraining the entire model. This can reduce the cost of the application while making the model more adaptable and generalizable.

[0031] In summary, through the transfer learning method, the existing life prediction model can be migrated and adjusted on different devices, making full use of the source domain knowledge, improving data efficiency and training efficiency, and enhancing the flexibility and adaptability of the model.

[0032] In the above embodiment, the feature extraction model realizes real-time processing and analysis of the collected IGBT data through a signal processing module and a real-time analysis module. The signal processing module realizes signal enhancement, noise removal and frequency domain feature extraction of the IGBT data through filtering, spectrum analysis, correlation analysis, Fourier transform or wavelet transform operations; the real-time analysis module performs real-time analysis of the data through filters and time domain and frequency domain analysis methods.

[0033] Feature extraction model is an important part of machine learning and data mining. Its purpose is to identify the most useful information from the raw data and convert it into a format that can be used by machine learning algorithms. In different application scenarios, the construction and application of feature extraction models will be different. In specific applications, first collect relevant data, which can be text, images, audio, sensor readings, etc. Clean the data, including removing noise, processing missing values, removing duplicate data, etc. Data transformation, such as normalization, standardization, logarithmic transformation, etc., to improve model performance. Then select the most useful features for the prediction task from a large number of features. Common feature selection methods include association rule mining, model-based feature selection (such as feature selection using the tree growth process), etc. Apply feature extraction algorithms to transform data and extract the intrinsic features of the data. Common feature extraction algorithms include principal component analysis (PCA), linear discriminant analysis (LDA), autoencoder (AE), convolutional neural network (CNN), etc. Then use the training data set to train the feature extraction model. Select a suitable machine learning algorithm, such as linear regression, support vector machine (SVM), neural network, etc. Evaluate the performance of the model through methods such as cross-validation. Adjust the model based on the verification results to improve the accuracy and generalization ability of the model. Deploy the trained model to the production environment. In practical applications, use the model to extract features and use them for subsequent prediction tasks such as classification, regression, clustering, etc. Feature extraction models are widely used in many fields. For example, in image recognition, they assist in recognition by extracting features such as edges, textures, and colors of images; in text mining, they assist in classification or sentiment analysis by extracting features such as keywords, phrases, and topics. When building a feature extraction model, you need to fully consider factors such as the characteristics of the data, the requirements of the task, and computing resources, and select the most appropriate algorithms and tools.

[0034] In a specific embodiment, the arithmetic operation software MATLAB provides a rich toolbox and function library, including signal processing, statistical analysis, machine learning, and optimization functions. These toolboxes and function libraries are used to process IGBT real-time monitoring data, establish prediction models and evaluation indicators, etc., to facilitate and quickly realize the development of algorithms and models; at the same time, MATLAB has intuitive visualization tools and interactive programming interfaces that can display and operate data, graphics, and results. For IGBT real-time monitoring and service life prediction and evaluation, this can easily display data trends, model outputs, and evaluation results, and more intuitively understand the performance and effects of the model. In terms of numerical computing capabilities: MATLAB is also an efficient numerical computing software with highly optimized matrix operations and numerical computing functions; for tasks such as data processing, model training, and reasoning calculations in IGBT real-time monitoring and service life prediction and evaluation, MATLAB can provide fast and accurate computing capabilities.

[0035] In the above embodiment, the autoregressive moving average model includes an autoregressive module, a moving average module and a difference module; the autoregressive module reflects the influence of past observations on current values ​​through an autoregressive equation, and the formula expression of the autoregressive equation is: (1) In formula (1), R represents the inertia weight, and m represent lag operators, S represents the number of differences; the sliding average module reflects the impact of the error of past observations on the current value through the error prediction method; the difference module performs multi-order differences on the time series data through the difference method, and converts the non-stationary series into a stationary series through the sequence conversion function. The formula expression of the sequence conversion function is: (2) In formula (2), Represents the function sequence conversion value, Represents the order of sequence transformation by the function, and d represents the sequence transformation error term; the autoregressive order, sliding average order and difference order are adjusted by the autoregressive sliding model function to achieve the fitting and prediction of time series data; the formula expression of the autoregressive sliding model function is: (3) In formula (3), H represents the autoregressive sliding model function; c represents the truncation window size, i represents the sliding average coefficient, and j represents the error term variance; the autoregressive sliding average model prediction establishes a model through historical time series data, and evaluates the remaining life of the IGBT through fitting and prediction.

[0036] In a specific embodiment, the autoregressive module, the moving average module and the difference module are three modules commonly used in time series analysis. They are combined to form ARIMA (autoregressive integrated moving average model), which is a common method for predicting time series data. In a specific application, the autoregressive module (AR) is a model used to describe the relationship between the current value and the historical value, and the current output value is expressed as a linear combination of the output values ​​of the previous periods. In a specific application, the moving average module (MA) is the accumulation of the error terms in the autoregressive model, which is used to describe the relationship between the current prediction error and the historical prediction error. The current prediction error et is expressed as a linear combination of the prediction errors of the previous periods. For example, an MA(2) model may indicate that the current prediction error et is a linear combination of the prediction errors et-1 and et-2 of the previous two periods, that is, et = θ1*et-1+ θ2*et-2. In a specific embodiment, the difference module is used to convert a non-stationary time series into a stationary time series, which can remove the trend and seasonal components in the time series and make the series stationary. The difference operation is usually to perform d differences on the series, where d is the order of the difference, which needs to be determined according to the characteristics of the series. Combining these three modules, we form the ARIMA model, which is generally in the form of ARIMA(p,d,q), where p is the number of autoregressive terms, d is the order of difference, and q is the number of sliding average terms. The ARIMA model captures the impact of historical values ​​on current values ​​through the autoregressive part, captures the impact of historical accumulation of forecast errors on current forecast errors through the sliding average part, and ensures the stationarity of time series data through the difference part.

[0037] In a specific embodiment, the autoregressive moving average model ARIMA model is a statistical model for processing time series data, which is widely used in modeling and prediction in time series data; for IGBT real-time monitoring data, the ARIMA model can capture the autocorrelation and moving average of time series data and provide accurate prediction results; in a specific embodiment, the ARIMA model has the ability to analyze and model the randomness and trend of time series; IGBT real-time monitoring data usually contains random noise and trend changes, and the ARIMA model can model and extract these features to achieve adaptive modeling and prediction of data; at the same time, the ARIMA model has flexible parameter adjustment and model recognition capabilities, and determines the best autoregressive, moving average and difference order parameter configurations, as well as the appropriate model order according to the characteristics of time series data. This ensures that the model has good fitting and prediction performance in real-time monitoring and service life prediction and evaluation; the prediction results of the ARIMA model have good interpretability and can intuitively display pattern information such as trends, periodicity and noise. This can easily understand and analyze the state and change trend of data for IGBT real-time monitoring and service life prediction and evaluation, and provide decision support; in a specific embodiment, the ARIMA model can be used as a component of IGBT real-time monitoring and service life prediction and evaluation, providing data preprocessing and feature extraction support for other models and algorithms; through the analysis and prediction of the ARIMA model, the trend information of the data can be extracted and used as the input feature of the subsequent model. And the analysis and prediction of IGBT real-time monitoring and service life prediction and evaluation are carried out through Table 1; Table 1 Test data table of IGBT real-time monitoring data

[0038] Table 1 includes data at seven time points, including three characteristics: temperature, current, and vibration. These characteristics can be used to establish the autoregressive moving average model ARIMA to analyze and predict IGBT real-time monitoring and service life prediction evaluation.

[0039] In the above embodiment, the adaptive parameter adjustment algorithm realizes the process of parameter optimization and model fitting by adjusting model parameters and optimization indicators. The adaptive parameter adjustment algorithm at least includes parameter optimization, feature selection and interface optimization. The formula of the adaptive parameter adjustment algorithm is as follows: For each particle , the particle position is expressed as =( 1, 2, ..., M), the particle speed is expressed as =( 1, 2, ..., M), where is the dimension of the problem; each particle Update according to the current position and speed, the relationship expression is: (4) (5) In formula (4) and formula (5), is the inertia weight, and They are individual learning factors and social learning factors. and is a random number between 0 and 1; It is a particle The best location ever. is the best position the entire particle group has ever experienced.

[0040] By iteratively updating the particle position and velocity until the convergence conditions are met, the optimal solution is obtained. The adaptive parameter adjustment algorithm is a technique in machine learning and statistical modeling that allows the model to automatically adjust its parameters to optimize model performance. This algorithm is often used to find the best model configuration to minimize prediction errors or maximize model fit. The adaptive parameter adjustment algorithm gives the model parameters an initial value based on experience or default settings. The model is trained using a training data set, during which the model generates predictions based on the current parameter values ​​and calculates the difference between the predictions and the actual values, i.e., the loss function value. When the parameter information is adjusted, the model parameters are adjusted according to the loss function value, with the goal of reducing the value of the loss function, which usually means improving the prediction accuracy of the model. Parameter adjustment uses various strategies in specific embodiments, such as gradient descent, Newton's method, conjugate gradient method, etc. Then a validation data set or cross-validation method is used to evaluate the performance of the adjusted model. This process can help identify overfitting or underfitting problems. During the iterative calculation process, if the evaluation results show that there is still room for improvement in model performance, return to step 2, retrain the model using new parameter values, and repeat the evaluation and adjustment process until the model performance reaches a satisfactory level or reaches a preset number of iterations. Once the model is fully trained and its performance is evaluated to your satisfaction, it can be deployed to a production environment for actual use. The key advantage of an adaptive parameter tuning algorithm is that it can automatically tune model parameters without manual intervention. This is particularly useful when dealing with complex models and large datasets, as manual parameter tuning can be time-consuming and error-prone. In addition, an adaptive algorithm is often able to find a better parameter configuration faster because it can search the parameter space in parallel and can use past experience to guide the search process.

[0041] In a further specific embodiment, the adaptive parameter adjustment algorithm optimizes the performance of the model by automatically adjusting the model parameters. First, various data information is parameter optimized, including determining the values ​​of each parameter in the model to minimize the prediction error or maximize the model fit. Parameter optimization generally involves using optimization techniques such as gradient descent, Newton's method, conjugate gradient method, etc., to adjust the parameters according to the loss function. Heuristic methods such as genetic algorithms, particle swarm optimization, simulated annealing, etc. may be used in the optimization process to guide the search for parameters. Feature selection is then performed to select the most useful features for the model from the original feature set. Specifically, a holdout method, cross validation, or a method based on feature importance scoring can be used.

[0042] In a further specific embodiment, the adaptive parameter adjustment algorithm simulates the group behavior in the bird flock to find the global optimal solution of the problem. It can avoid falling into the local optimal solution and has a good global search capability; at the same time, the particles in the adaptive parameter adjustment algorithm can search the solution space in parallel to accelerate the search process. Each particle uses its own historical best position and the best position in the group as a reference, and searches iteratively by updating the speed and position; in a specific embodiment, the adaptive parameter adjustment algorithm can adaptively adjust the search speed and direction according to the search experience of the particle itself and the collaborative information of the group. This enables the algorithm to better adapt to the characteristics and changes of the problem and improve the search efficiency; the adaptive parameter adjustment algorithm does not require the derivative or gradient information of the problem, but only the objective function of the problem. This makes the algorithm more flexible and convenient when dealing with complex problems.

[0043] The following is the data sheet of the adaptive parameter adjustment algorithm, which is used for the adaptive parameter adjustment algorithm of the IGBT real-time monitoring and service life prediction evaluation method: Table 2 Adaptive parameter adjustment algorithm data table

[0044] In this data table 2, feature 1 represents temperature features, feature 2 represents current features, feature 3 represents vibration features, and the objective function value represents an evaluation index calculated based on these features. In the adaptive parameter adjustment algorithm, by updating the speed and position of the particles, the optimization algorithm gradually searches for the optimal feature combination based on the minimization or maximization of the objective function value, which is used for real-time monitoring of IGBTs and life prediction evaluation.

[0045] In the above embodiment, the rule engine automatically monitors the real-time status of the IGBT through threshold setting, real-time monitoring, status judgment and alarm operation. The threshold setting sets the temperature limit threshold, current limit threshold, power factor threshold and loss index threshold through training sample data. The real-time monitoring automatically compares the data of the data acquisition system with the set threshold data through a data comparison method, and triggers an alarm through status judgment.

[0046] In a specific embodiment, the rule engine determines the state of the IGBT by setting a reasonable threshold. According to different monitoring indicators (such as temperature, current, vibration, etc.), the upper and lower thresholds of each indicator can be determined. When the indicator exceeds or approaches the set threshold, it can be judged that the IGBT is in an abnormal state or is about to fail. At the same time, the rule engine is real-time and can monitor the state of the IGBT in real time. By making rapid threshold judgments on real-time data, abnormal conditions of the IGBT can be discovered in time, and corresponding measures can be taken to deal with it, thereby reducing the risk of failure. In addition, the rule engine has a certain degree of flexibility and can adjust the threshold according to factors such as the IGBT model and operating environment to adapt to different working conditions and performance requirements. In implementation, the rule engine can be continuously tuned and optimized through actual monitoring data and failure experience. According to historical data analysis and system operation, the threshold settings can be continuously revised and updated to improve prediction accuracy and judgment reliability.

[0047] In the above embodiment, the intelligent analysis system includes a data monitoring module, a data processing module, a data analysis module, a reasoning and decision module and a user interface module; the data monitoring module obtains the monitoring data of the IGBT in real time through a temperature sensor, a current sensor and a voltage sensor; the data processing module includes a storage unit and a processing unit; the storage unit manages and stores the collected data through a cloud database; the processing unit processes the data through data cleaning, data conversion and data aggregation operations to improve the quality and availability of the data; the data analysis module analyzes the collected data through statistical analysis, signal processing and pattern recognition methods to obtain patterns, trends and abnormal conditions in the data; the reasoning and decision module reasons and makes decisions on the analysis data through knowledge representation and reasoning engine; the user interface module converts the data into a visual form through a data visualization method and a chart library to display the decision results; the output end of the data monitoring module is connected to the input end of the data processing module; the output end of the data processing module is connected to the input end of the data analysis module; the output end of the data analysis module is connected to the input end of the reasoning and decision module; the output end of the reasoning and decision module is connected to the input end of the user interface module.

[0048] In a specific embodiment, the intelligent analysis system monitors and collects relevant data of the IGBT device in real time through the data monitoring module, including indicators such as temperature, current, and voltage; the collected raw data is preprocessed through the data processing module, including noise removal, filtering, normalization, and other operations to improve the accuracy and reliability of subsequent data analysis. The preprocessed data is subjected to feature extraction and pattern recognition through the data analysis module to obtain the status information of the IGBT device, such as working status, health status, etc.; through the reasoning decision module: based on the results obtained by the data analysis module, life prediction and evaluation are performed to determine the current service life of the IGBT device, and corresponding decision suggestions are provided based on the prediction results, such as whether maintenance or replacement is required; the real-time monitoring data, life prediction results, and maintenance suggestions are displayed to the user through the user interface module, so that the user can understand the device status in a timely manner and make corresponding operations.

[0049] In a method for real-time monitoring and service life prediction and evaluation of IGBT, the intelligent analysis system can monitor the working status of IGBT equipment in real time, obtain data in a timely manner, and process and analyze it, so that problems can be discovered and solved as early as possible, thereby improving the reliability and stability of the equipment. In addition, through the analysis of data and the operation of models, the intelligent analysis system can predict the service life of IGBT equipment and provide corresponding maintenance suggestions to help users reasonably plan equipment maintenance and replacement plans and reduce failure risks and maintenance costs. Secondly, the intelligent analysis system can automatically make decisions and inferences based on data analysis results and set decision rules, reduce manual intervention, and improve efficiency and accuracy. Finally, through the user interface module, users can easily view the equipment status and prediction results, understand the equipment usage and take timely actions, thereby improving the convenience of operation and user experience.

[0050] In summary, the intelligent analysis system in an IGBT real-time monitoring and service life prediction and evaluation method provides users with timely equipment status monitoring, life prediction and maintenance suggestions through data monitoring, processing, analysis and decision-making, thereby improving equipment reliability, preventing failures, and reducing maintenance costs, and has the characteristics of automation and user-friendliness.

[0051] In the above embodiment, the cloud database includes a distributed file system and a cloud archiving module. The distributed file system stores data in a dispersed manner on multiple nodes through a data storage method to achieve scalability and redundant backup of data storage; the cloud archiving module processes, aggregates and analyzes the monitoring data of the IGBT through a mapping-reduction function.

[0052] In a specific embodiment, the cloud database Apache Hadoop is a distributed computing framework that can store and process data on multiple nodes; this makes it have good horizontal scalability, can handle large-scale data sets, and is suitable for high-frequency data storage and processing of IGBT real-time monitoring; Apache Hadoop adopts a fault-tolerant mechanism, which can ensure data reliability and system stability even when some nodes fail. This is very important for the data security and reliability of IGBT real-time monitoring; Apache Hadoop has high-speed data processing capabilities and can use parallel computing and distributed storage to efficiently process large amounts of data. This enables it to process and analyze IGBT monitoring data in real time and provide real-time service life prediction and evaluation; Apache Hadoop is based on the distributed file system HDFS, which can store and manage massive amounts of data. It can store large-scale data for IGBT real-time monitoring in a low-cost manner and has high availability and data backup functions.

[0053] In the above embodiment, the edge processing model includes a data stream receiving module, a streaming data processing module, a real-time monitoring and alarm module and an edge computing module; the data stream receiving module receives real-time data streams from temperature sensors, current sensors, vibration sensors and voltage sensors through the Internet of Things sensor network; the streaming data processing engine module includes a feature extraction model update unit and a life prediction model update unit; the feature extraction model update unit extracts and updates features in the real-time data stream through an online learning method; the life prediction model update unit updates the life prediction model in real time through a time series analysis method, and the time series analysis method analyzes and models through an autoregressive sliding average model; the real-time monitoring and alarm module includes a real-time alarm unit and a real-time data analysis unit; the real-time alarm unit realizes real-time monitoring of temperature, current and voltage and triggers an alarm through a rule engine; the real-time data analysis unit performs instant analysis and processing of real-time data through parallel computing and streaming data processing methods; the edge computing module uses local computing resources through a distributed architecture and intelligent scheduling and resource management methods to process and analyze data in real time.

[0054] In a specific embodiment, the working principle of the edge processing model is as follows: V1. Data collection and transmission: The power consumption information collection device is responsible for collecting the current, voltage, temperature and other power consumption information data of the IGBT in real time, and transmits it to the edge node through the network.

[0055] V2, Streaming Data Processing Engine: The streaming data processing engine runs on the edge node to receive and process the collected real-time data stream. This engine can process the data stream in real time to ensure fast data processing and response.

[0056] V3, Feature Extraction Model: Real-time data is passed to the feature extraction model through the streaming data processing engine. The model uses various feature extraction algorithms, such as time series feature extraction and frequency domain analysis, to convert raw data into representative features for subsequent life prediction.

[0057] V4. Life prediction model: The data after feature extraction is input into the life prediction model for analysis and prediction. The life prediction model may use machine learning algorithms (such as regression, classification) or deep learning algorithms (such as neural networks) to predict the life of IGBT by learning from historical data.

[0058] V5. Real-time model update: The edge processing model continuously receives the latest real-time data and updates the feature extraction model and life prediction model to ensure that the model accurately reflects the system status. As time goes by and data accumulates, the prediction accuracy of the model will continue to improve.

[0059] In a method for real-time monitoring and service life prediction of IGBT power consumption information collection, The edge processing model can monitor the power consumption information of IGBT in real time, detect abnormal conditions or faults in time, and take corresponding measures to avoid equipment damage or production accidents. Secondly, through the life prediction model, the life of IGBT can be predicted in advance to help formulate maintenance plans and scheduling, reduce equipment downtime, and reduce maintenance costs. In addition, the edge processing model continuously updates the feature extraction model and life prediction model, which can optimize the prediction accuracy, improve system performance, and ensure stable operation of the equipment. At the same time, through timely monitoring and prediction, production interruptions and maintenance costs caused by equipment failures can be avoided, thereby saving maintenance costs and improving production efficiency.

[0060] The working principle of the edge processing model is to realize the rapid processing and analysis of the real-time power consumption information of IGBT, and to realize the prediction and monitoring of the service life of IGBT through the combination of feature extraction and life prediction model. Such an approach can help to discover potential problems in advance, optimize maintenance plans, reduce maintenance costs, and improve equipment reliability and production efficiency.

[0061] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that these specific embodiments are only illustrative, and those skilled in the art may omit, replace, and change the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, merging the above method steps so as to perform substantially the same functions in substantially the same manner to achieve substantially the same results is within the scope of the present invention. Therefore, the scope of the present invention is limited only by the appended claims.

Claims

1. A method for real-time monitoring of IGBT power consumption information collection and service life prediction, characterized in that: The following steps are involved: Step 1: Data collection and preprocessing; The working parameter data of the IGBT of the electric information terminal is collected in real time through temperature sensors, current sensors, vibration sensors and voltage sensors, and the analyzability and predictability of the original data are improved by removing abnormal values, filling missing values ​​and standardizing processing methods; Step 2: Feature extraction; The IGBT working state characteristics of the power information terminal are extracted by training the feature extraction model; the IGBT working state characteristics at least include temperature change rate, power loss and current impact parameters; Step 3: Construct a life prediction model; The extracted data is analyzed and modeled by an autoregressive sliding average model to construct a life prediction model; the life prediction model migrates and adjusts the existing life prediction model on different devices by a transfer learning method to reduce training time and data requirements; Step 4: Real-time data stream processing; The collected real-time data stream is processed and analyzed by an edge processing model; the edge processing model updates the feature extraction model and the life prediction model in real time through a streaming data processing engine; Step 5: Dynamic adjustment and optimization; The use conditions and loads of IGBTs are optimized through adaptive parameter adjustment algorithms, and the working parameters and working states of IGBTs are dynamically adjusted through a reinforcement learning framework to extend the service life of IGBTs. Step 6: Early warning and maintenance; The IGBT is monitored in real time by a real-time monitoring and alarm system, which monitors the IGBT through a real-time data analysis engine and automatically triggers an alarm through a rule engine to prevent the occurrence of potential faults; the rule engine includes a temperature threshold unit, a current threshold unit, and a vibration threshold unit; Step 7: Build an intelligent analysis system; The intelligent analysis system uses a machine learning model to detect abnormal conditions during the operation of the IGBT in real time and provide automated fault diagnosis and solutions.

2. The method for real-time monitoring of power consumption information collection and service life prediction of IGBT according to claim 1 is characterized by: The transfer learning method works as follows: S1. Select source domain and target domain; The source domain compares the similarities of different domains in data distribution, feature space and model parameters by a data comparison method to select a domain for which a life prediction model has been trained. The target domain selects an IGBT life prediction domain by observing and collecting data, features and model parameter information in the domain. S2, extract feature representation; Extract features from data in the source and target domains through convolutional neural networks; S3, establish a migration model; Establish a transfer model through transfer learning and domain adaptation methods to transfer knowledge from the source domain to the target domain; S4, transfer learning training; The model training method uses the source domain data to train the migration model. During the training process, dropout regularization is used to improve the generalization ability of the model. S5, migration and adjustment; The trained transfer model is applied to the data of the target domain and fine-tuned for the target domain through learning rate adjustment and weight freezing methods.

3. The method for real-time monitoring of power consumption information collection and service life prediction of IGBT according to claim 1 is characterized by: The feature extraction model realizes real-time processing and analysis of the collected IGBT data through a signal processing module and a real-time analysis module. The signal processing module realizes signal enhancement, noise removal and frequency domain feature extraction of the IGBT data through filtering, spectrum analysis, correlation analysis, Fourier transform or wavelet transform operations; The real-time analysis module performs real-time analysis on the data through filters and time domain and frequency domain analysis methods.

4. The method for real-time monitoring of power consumption information collection and service life prediction of IGBT according to claim 1, characterized in that: The autoregressive moving average model includes an autoregressive module, a moving average module and a difference module; the autoregressive module reflects the influence of past observations on current values ​​through an autoregressive equation, and the formula expression of the autoregressive equation is: （1） In formula (1), R represents the inertia weight, and m represent lag operators, S represents the number of differences; the sliding average module reflects the impact of the error of past observations on the current value through the error prediction method; the difference module performs multi-order differences on the time series data through the difference method, and converts the non-stationary series into a stationary series through the sequence conversion function. The formula expression of the sequence conversion function is: （2） In formula (2), Represents the function sequence conversion value, Represents the order of sequence transformation by the function, and d represents the sequence transformation error term; the autoregressive order, sliding average order and difference order are adjusted by the autoregressive sliding model function to achieve the fitting and prediction of time series data; the formula expression of the autoregressive sliding model function is: （3） In formula (3), H represents the autoregressive sliding model function; c represents the truncation window size, i represents the sliding average coefficient, and j represents the error term variance; the autoregressive sliding average model prediction establishes a model through historical time series data, and evaluates the remaining life of the IGBT through fitting and prediction.

5. The method for real-time monitoring of power consumption information collection and service life prediction of IGBT according to claim 1, characterized in that: The adaptive parameter adjustment algorithm realizes parameter optimization and model fitting through parameter optimization, feature selection and interface optimization methods; for each particle , the particle position is expressed as =( 1, 2, ..., M), the particle speed is expressed as =( 1, 2, ..., M), where is the dimension of the problem; each particle Update according to the current position and speed, the relationship expression is: （4） （5） In formula (4) and formula (5), is the inertia weight, and They are individual learning factors and social learning factors. and is a random number between 0 and 1; It is a particle The best location ever. is the best position that the entire particle swarm has ever experienced; the optimal solution is obtained by iteratively updating the particle position and velocity until the convergence condition is met.

6. The method for real-time monitoring of power consumption information collection and service life prediction of IGBT according to claim 1, characterized in that: The rule engine automatically monitors the real-time status of the IGBT through threshold setting, real-time monitoring, status judgment and alarm operation. The threshold setting sets the temperature limit threshold, current limit threshold, power factor threshold and loss index threshold through training sample data. The real-time monitoring automatically compares the data of the data acquisition system with the set threshold data through a data comparison method, and triggers an alarm through status judgment.

7. The method for real-time monitoring of power consumption information collection and service life prediction of IGBT according to claim 1, characterized in that: The intelligent analysis system includes a data monitoring module, a data processing module, a data analysis module, an inference decision module and a user interface module; the data monitoring module obtains the monitoring data of the IGBT in real time through a temperature sensor, a current sensor and a voltage sensor; the data processing module includes a storage unit and a processing unit; the storage unit manages and stores the collected data through a cloud database; the processing unit processes the data through data cleaning, data conversion and data aggregation operations to improve the quality and availability of the data; the data analysis module analyzes the collected data through statistical analysis, signal processing and pattern recognition methods to obtain patterns, trends and abnormal conditions in the data; The reasoning and decision-making module performs reasoning and decision-making on the analysis data through knowledge representation and reasoning engine; The user interface module converts data into a visual form through a data visualization method and a chart library to display the decision results; the output end of the data monitoring module is connected to the input end of the data processing module; the output end of the data processing module is connected to the input end of the data analysis module; the output end of the data analysis module is connected to the input end of the reasoning decision module; the output end of the reasoning decision module is connected to the input end of the user interface module.

8. The method for real-time monitoring of power consumption information collection and service life prediction of IGBT according to claim 1, characterized in that: The cloud database includes a distributed file system and a cloud archiving module. The distributed file system stores data in a dispersed manner on multiple nodes through a data storage method to achieve scalability and redundant backup of data storage; the cloud archiving module processes, aggregates and analyzes the monitoring data of the IGBT through a mapping-reduction function.

9. The method for real-time monitoring of power consumption information collection and service life prediction of IGBT according to claim 1, characterized in that: The edge processing model includes a data stream receiving module, a streaming data processing module, a real-time monitoring and alarm module and an edge computing module; the data stream receiving module receives real-time data streams from temperature sensors, current sensors, vibration sensors and voltage sensors through the Internet of Things sensor network; the streaming data processing engine module includes a feature extraction model update unit and a life prediction model update unit; the feature extraction model update unit extracts and updates features in the real-time data stream through an online learning method; the life prediction model update unit updates the life prediction model in real time through a time series analysis method, and the time series analysis method is analyzed and modeled through an autoregressive sliding average model; the real-time monitoring and alarm module includes a real-time alarm unit and a real-time data analysis unit; The real-time alarm unit realizes real-time monitoring of temperature, current and voltage and triggers alarms through the rule engine; the real-time data analysis unit performs instant analysis and processing of real-time data through parallel computing and streaming data processing methods; The edge computing module uses local computing resources to process and analyze data in real time through a distributed architecture and intelligent scheduling and resource management methods.

Images