Abnormity monitoring method and device for IGBT power module

Abstract

The invention provides an abnormity monitoring method and device for an IGBT power module, and relates to the technical field of energy storage equipment. The method comprises the steps that multi-source operation data of an IGBT power module are acquired according to a key feature subset, and the key feature subset is a data category which is screened out by performing feature importance analysis on the operation data of the IGBT power module and has high junction temperature prediction contribution degree on the IGBT power module; based on the multi-source operation data and the junction temperature sequence within the preset duration, performing junction temperature prediction of the IGBT power module to obtain a junction temperature prediction value; and monitoring whether the IGBT power module operates abnormally or not according to the junction temperature predicted value. According to the invention, accurate early warning can be carried out on abnormal operation of the IGBT power module.

Description

Technical Field

[0001] This application relates to the field of energy storage equipment technology, and in particular to an anomaly monitoring method and device for IGBT power modules. Background Technology

[0002] Insulated-gate bipolar transistors (IGBTs) are core power devices in the DC-AC conversion stage of power conversion systems (PCS), and their operational reliability directly affects the safety, stability, and lifespan of the entire energy storage system. During operation, IGBT power modules generate significant heat due to switching and conduction losses, leading to increased junction temperature. Junction temperature fluctuations and overheating are key factors causing material aging, bond wire detachment, and even complete failure of IGBT power modules. Therefore, early warning of abnormal junction temperatures in IGBT power modules is crucial for preventing faults, extending equipment lifespan, and ensuring the stable operation of energy storage power stations.

[0003] In related technologies, triggering abnormal alarms of IGBT power modules based on fixed temperature thresholds can result in alarm delays. Summary of the Invention

[0004] This application provides an abnormal monitoring method and apparatus for IGBT power modules, which can achieve the effect of accurately predicting abnormal operation of IGBT power modules.

[0005] In a first aspect, embodiments of this application provide a method for anomaly monitoring of an IGBT power module, comprising:

[0006] Based on the key feature subset, multi-source operating data of IGBT power modules are obtained. The key feature subset is the data category that contributes highly to the prediction of junction temperature of IGBT power modules by performing feature importance analysis on the operating data of IGBT power modules.

[0007] Based on multi-source operating data and junction temperature sequence within a preset time period, the junction temperature of IGBT power modules is predicted to obtain the predicted junction temperature value.

[0008] Based on the predicted junction temperature, monitor whether there are any operational abnormalities in the IGBT power module.

[0009] In one possible implementation, the junction temperature of the IGBT power module is predicted based on multi-source operating data and a junction temperature sequence within a preset time period to obtain the predicted junction temperature value, including:

[0010] Based on gated recurrent units, real-time temporal features of junction temperature sequences are extracted;

[0011] The hidden states of the gated recurrent unit are weighted based on the temporal attention mechanism to obtain the weighted temporal features of the junction temperature sequence;

[0012] Based on multi-source operating data, real-time timing characteristics, and weighted timing characteristics, the junction temperature of the IGBT power module is predicted to obtain the predicted junction temperature value.

[0013] In one possible implementation, the junction temperature of the IGBT power module is predicted based on multi-source operating data, real-time timing characteristics, and weighted timing characteristics to obtain the predicted junction temperature value, including:

[0014] Multi-source operational data, real-time time-series features, and weighted time-series features are concatenated to obtain comprehensive time-series features;

[0015] Based on the comprehensive timing characteristics, a fully connected deep neural network model is used to perform nonlinear fitting on the junction temperature of the IGBT power module to obtain the predicted junction temperature value.

[0016] In one possible implementation, monitoring for operational abnormalities in the IGBT power module based on predicted junction temperature values ​​includes:

[0017] Determine the residual values ​​between the observed and predicted junction temperatures within a preset time period to obtain the residual sequence;

[0018] Based on the statistical characteristics of the residual sequence, monitor whether there are any operational abnormalities in the IGBT power module.

[0019] In one possible implementation, monitoring for operational anomalies in the IGBT power module is based on the statistical characteristics of the residual sequence, including:

[0020] Based on the sliding window mechanism, the mean and standard deviation of the residual sequence within the sliding window are calculated.

[0021] Based on the mean and standard deviation, the confidence interval of the residual sequence in the sliding window is determined, and the confidence level of the residual values ​​contained in the confidence interval is greater than the preset confidence threshold.

[0022] Based on the upper limit of the confidence interval corresponding to the sliding window and the preset residual threshold, monitor whether there are any abnormalities in the operation of the IGBT power module.

[0023] In one possible implementation, monitoring for operational abnormalities in the IGBT power module is performed based on the upper limit of the confidence interval corresponding to the sliding window and a preset residual threshold, including:

[0024] If the upper limit of the confidence interval corresponding to N consecutive sliding windows is greater than the residual threshold, it is determined that the IGBT power module has an operational abnormality, where N is the preset window threshold.

[0025] In one possible implementation, the subset of key features is obtained in the following manner:

[0026] Based on the random forest algorithm, feature importance analysis was performed on the operating data of IGBT power modules to obtain the first contribution of each data category in the operating data to the junction temperature prediction of IGBT power modules.

[0027] Based on the XGboost algorithm, feature importance analysis was performed on the running data to obtain the second contribution of each data category in the running data to the prediction of IGBT power module junction temperature;

[0028] The first and second contributions corresponding to the running data are weighted and merged to obtain the third contribution corresponding to the running data.

[0029] The target data category whose third contribution is greater than the preset contribution threshold in the running data is identified as the key feature subset.

[0030] In one possible implementation, the subset of key features is obtained in the following manner:

[0031] The nonlinear relationship between the operating data of the IGBT power module and the junction temperature of the IGBT power module is evaluated based on the mutual information algorithm, and the corresponding highly correlated subsets are obtained.

[0032] Based on the recursive feature elimination algorithm, redundant features in the relevant subset are removed to obtain the key feature subset.

[0033] In one possible implementation, before predicting the junction temperature of the IGBT power module based on multi-source operating data and a subset of key features of the IGBT power module, and obtaining the predicted junction temperature value, the method further includes:

[0034] Preprocessing of multi-source operational data includes data cleaning and data standardization.

[0035] Secondly, embodiments of this application provide an anomaly monitoring device for an IGBT power module, comprising:

[0036] The acquisition module is used to acquire multi-source operating data of IGBT power modules based on key feature subsets. The key feature subsets are data categories that contribute highly to the prediction of junction temperature of IGBT power modules by performing feature importance analysis on the operating data of IGBT power modules.

[0037] The prediction module is used to predict the junction temperature of the IGBT power module based on multi-source operating data and the junction temperature sequence within a preset time period, and obtain the predicted junction temperature value.

[0038] The monitoring module is used to monitor whether there are any abnormalities in the operation of the IGBT power module based on the predicted junction temperature.

[0039] Thirdly, embodiments of this application provide an anomaly monitoring device for an IGBT power module, comprising: a memory and a processor;

[0040] The memory stores instructions that the computer executes;

[0041] The processor executes computer execution instructions stored in memory, causing the processor to perform the methods described in the various possible implementations of the first aspect above.

[0042] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the methods described in the various possible implementations of the first aspect above.

[0043] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed, implements the methods described in the various possible implementations of the first aspect above.

[0044] The anomaly monitoring method and apparatus for IGBT power modules provided in this application obtain multi-source operating data of the IGBT power module based on a subset of key features. The subset of key features involves analyzing the feature importance of the IGBT power module's operating data to identify data categories that contribute significantly to the prediction of the IGBT power module's junction temperature. Based on the multi-source operating data and the junction temperature sequence within a preset time period, the junction temperature of the IGBT power module is predicted, obtaining a predicted junction temperature value. This effectively improves the accuracy of the predicted junction temperature value. By monitoring whether the IGBT power module exhibits any operational anomalies based on the predicted junction temperature value, early warning of IGBT power module anomalies can be achieved, ensuring the normal operation of the IGBT power module. Attached Figure Description

[0045] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0046] Figure 1 A flowchart illustrating the anomaly monitoring method for an IGBT power module provided in an embodiment of this application;

[0047] Figure 2 This is a diagram illustrating the architecture of an IGBT power module anomaly monitoring system provided in an embodiment of this application.

[0048] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0049] Figure 4A schematic diagram of the structure of the abnormal monitoring device for the IGBT power module provided in the embodiments of this application;

[0050] Figure 5 This is a schematic diagram of the structure of the abnormal monitoring device for the IGBT power module provided in the embodiment of this application.

[0051] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0052] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0053] First, let me explain the terms used in this application:

[0054] Insulated Gate Bipolar Transistor (IGBT): A fully controllable voltage-driven power semiconductor device that combines the high input impedance of a Metal-Oxide-Semiconductor Field-Effect Transistor (MOSFET) with the low on-state voltage drop of a Giant Transistor (GTR). It is the core switching device for energy conversion in power electronic devices such as energy storage converters.

[0055] Power Conversion System (PCS): A device that connects a battery system to the power grid (or load) to achieve bidirectional conversion of electrical energy. It can convert between direct current and alternating current and control the charging and discharging process of electrical energy.

[0056] Gated Recurrent Unit (GRU): An improved recurrent neural network architecture that, by introducing update and reset gate mechanisms, can more effectively capture long-term dependencies in time series and alleviate the gradient vanishing problem of traditional recurrent neural networks (RNNs).

[0057] Deep Neural Network (DNN): A type of feedforward neural network containing multiple hidden layers, capable of learning multi-level abstract representations of data to achieve complex nonlinear mappings and function approximations.

[0058] Random Forest Algorithm: The Random Forest algorithm is a classification and regression algorithm based on the idea of ​​ensemble learning. Its core is to construct multiple independent decision trees by random sampling. At the same time, when splitting at each decision tree node, some features are randomly selected from all features for optimal splitting. Finally, the output results of all decision trees are merged by voting or averaging to obtain the final prediction conclusion.

[0059] XGBoost algorithm: Extreme Gradient Boosting Algorithm. An efficient and scalable machine learning algorithm belonging to the gradient boosting decision tree family. It performs well in various prediction tasks by integrating multiple weak learners (decision trees) and optimizing their construction process.

[0060] Junction temperature: refers to the actual operating temperature of the semiconductor junction inside the IGBT chip. It is the most critical parameter leading to IGBT performance degradation and failure.

[0061] Temporal attention mechanism: a mechanism for sequence models that can automatically learn and assign different weights to each time step in the input sequence, enabling the model to "pay attention" to historical information that is more important to the current prediction task.

[0062] Prediction residual: refers to the difference between the actual observed value and the model's predicted value in a prediction model. In this invention, it is the difference between the actual junction temperature and the predicted junction temperature. Changes in the statistical characteristics of the residual sequence are a key indicator for identifying anomalies.

[0063] Sliding window: A data processing technique used to process streaming or time-series data. It defines a fixed-length window that slides forward as new data arrives, performing calculations only on the data within the window at a time.

[0064] To address the issue of delayed alarms for IGBT power modules in related technologies, this application acquires multi-source operating data of IGBT power modules in an energy storage converter within a preset time period. Based on this multi-source operating data and historical junction temperature sequences, a hybrid prediction model combining time-series attention mechanism and GRU-DNN is used to accurately predict the junction temperature of the IGBT power modules. Furthermore, based on the residual sequence between the predicted and observed junction temperatures, and by combining sliding window statistical analysis of the statistical characteristics of the residual sequence, early identification and warning of abnormal states of the IGBT modules are achieved. This effectively improves the accuracy and timeliness of IGBT power module fault warnings, ensuring the safe and stable operation of the energy storage system.

[0065] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0066] Figure 1 This is a flowchart illustrating the anomaly monitoring method for an IGBT power module provided in an embodiment of this application. Figure 1 As shown, the method includes:

[0067] S101. Based on the key feature subset, obtain multi-source operating data of the IGBT power module. The key feature subset is the data category that contributes highly to the prediction of junction temperature of the IGBT power module by performing feature importance analysis on the operating data of the IGBT power module.

[0068] Specifically, multi-source operating data includes line voltage, total active power, and timing data of the A / B / C phase three-phase voltage, current, and temperature of the IGBT power module.

[0069] Optionally, multi-source operational data can be collected at a preset data collection frequency. For example, the data collection frequency can be once every 1 minute or once every 5 minutes. The multi-source operational data can be the real-time value at the time of collection or the average value within the collection interval.

[0070] Understandably, the correlation between the junction temperature of an IGBT power module and the data in its operating data varies; some data have a significant impact on the junction temperature, while others have a relatively weak impact. By performing feature importance analysis on the operating data of the IGBT power module, and identifying multi-source operating data that are strongly correlated with the IGBT power module for use in junction temperature prediction, the accuracy of junction temperature prediction can be improved.

[0071] S102. Based on multi-source operating data and junction temperature sequence within a preset time period, predict the junction temperature of the IGBT power module and obtain the predicted junction temperature value.

[0072] In one implementation, a gated loop unit is used to extract real-time and weighted time-series features from the junction temperature sequence. Based on the real-time and weighted time-series features of the junction temperature sequence and multi-source operating data, a junction temperature prediction model is used to predict the junction temperature of the IGBT power module.

[0073] An alternative implementation involves using a temporal convolutional network to extract long-term dependency features of the junction temperature sequence, then introducing a self-attention mechanism to weight key time steps to obtain temporal features of the junction temperature sequence, and then predicting the junction temperature based on the temporal features of the junction temperature sequence and multi-source operating data.

[0074] In one embodiment, the junction temperature prediction model can be a fully connected deep neural network model, which can perform nonlinear fitting on the junction temperature of the IGBT power module based on real-time timing features, weighted timing features, and multi-source operating data to obtain the junction temperature prediction value.

[0075] Another alternative implementation uses the Transformer architecture as the backbone of time-series modeling, utilizes its multi-head attention mechanism to capture global dependencies in the junction temperature sequence, and then merges it with multi-source running data before outputting the junction temperature prediction value through a fully connected layer.

[0076] S103. Based on the predicted junction temperature, monitor whether there are any abnormalities in the operation of the IGBT power module.

[0077] In one implementation, the junction temperature prediction model is modeled and trained based on relevant operating data of the IGBT power module under normal thermal characteristics. In other words, the junction temperature prediction model outputs the theoretical junction temperature value of the IGBT power module under healthy conditions. The junction temperature prediction value can be used as a reference value. If there is a large deviation between the observed actual junction temperature value and the junction temperature prediction value, it indicates that the thermal characteristics of the IGBT power module are abnormal and there is an operational abnormality.

[0078] In another implementation, the junction temperature prediction model is modeled and trained based on the historical operating data of the IGBT power module. The predicted junction temperature value reflects the value that the junction temperature may reach at the next moment. If the predicted junction temperature value exceeds the preset junction temperature threshold, it indicates that there is a risk of excessive junction temperature at the future moment, thus determining that the IGBT power module is operating abnormally and realizing early warning of IGBT power module operating abnormality.

[0079] The anomaly monitoring method for IGBT power modules provided in this application acquires multi-source operating data of the IGBT power module based on a key feature subset. The key feature subset is a data category that contributes significantly to the prediction of the junction temperature of the IGBT power module by performing feature importance analysis on the operating data of the IGBT power module. This can effectively improve the accuracy of subsequent junction temperature prediction. Based on the predicted junction temperature value, the method monitors whether there are any operational anomalies in the IGBT power module, enabling early warning of IGBT power module anomalies and ensuring the normal operation of the IGBT power module.

[0080] In one possible implementation, the junction temperature of the IGBT power module is predicted based on multi-source operating data and a junction temperature sequence within a preset time period to obtain the predicted junction temperature value, including:

[0081] Based on the gated recurrent unit, the real-time temporal features of the junction temperature sequence are extracted; the hidden state of the gated recurrent unit is weighted based on the temporal attention mechanism to obtain the weighted temporal features of the junction temperature sequence.

[0082] Based on multi-source operating data, real-time timing characteristics, and weighted timing characteristics, the junction temperature of the IGBT power module is predicted to obtain the predicted junction temperature value.

[0083] Specifically, the gated loop unit relies on the dynamic adjustment mechanism of its reset and update gates. For historical junction temperature data within a preset time period, it first filters key information from the current junction temperature data and historical hidden states through the reset gate, and controls the retention of historical states and the integration of current information through the update gate. It outputs the hidden state at the corresponding moment as a real-time time series feature, accurately capturing the instantaneous change pattern and temporal correlation of junction temperature data in the continuous time dimension. At the same time, combined with the importance difference of data at different moments in the junction temperature sequence, the hidden states output by the gated loop unit at each moment are weighted and aggregated through preset weight allocation rules. Finally, a weighted time series feature that integrates key time series information and importance differences is obtained, which not only preserves the dynamic change characteristics of the junction temperature sequence, but also highlights the contribution of core moment data to feature representation.

[0084] By combining multi-source operating data collected from IGBT power modules under actual operating conditions, real-time time-series features extracted from junction temperature sequences through gated cyclic units, and weighted time-series features that integrate the differences in the importance of data at critical moments, the three types of features are standardized, preprocessed, and fused before being input into the junction temperature prediction model. Through deep learning and dynamic fitting of multi-dimensional features, the junction temperature prediction model can predict the junction temperature of the IGBT power module within a preset time period in the future, and finally output accurate and reliable junction temperature prediction values, providing data support for the safe operation monitoring of IGBT power modules.

[0085] The anomaly monitoring method for IGBT power modules provided in this application accurately captures the instantaneous fluctuation pattern and temporal correlation of junction temperature in the continuous time dimension through a gated loop unit, and highlights the characterization contribution of junction temperature data near the current moment and during the stage of sudden change in operating conditions, providing reliable and accurate data input for the junction temperature prediction model, thereby improving the accuracy and efficiency of junction temperature prediction.

[0086] In one possible implementation, the junction temperature of the IGBT power module is predicted based on multi-source operating data, real-time timing characteristics, and weighted timing characteristics to obtain the predicted junction temperature value, including:

[0087] Multi-source operating data, real-time time-series features, and weighted time-series features are concatenated to obtain comprehensive time-series features. Based on the comprehensive time-series features, a fully connected deep neural network model is used to perform nonlinear fitting on the junction temperature of the IGBT power module to obtain the predicted junction temperature value.

[0088] Specifically, based on the comprehensive time-series features that integrate multi-source operational data, real-time time-series features, and weighted time-series features, the features are first standardized and preprocessed to eliminate differences in dimensions and data distribution biases. Then, the processed feature vectors are input into a parameter-optimized fully connected deep neural network model. This model relies on the stacked structure of multiple hidden layers and the complex connections between neurons. It introduces nonlinear transformations through activation functions and iteratively optimizes network parameters using gradient descent algorithms to minimize prediction errors. It deeply mines the complex nonlinear mapping relationship between comprehensive time-series features and the junction temperature of IGBT power modules, achieving accurate fitting of the junction temperature change pattern and finally outputting stable and reliable predicted values ​​for IGBT power module junction temperatures.

[0089] The anomaly monitoring method for IGBT power modules provided in this application embodiment is based on a fully connected deep neural network model. It performs nonlinear fitting on the junction temperature of the IGBT power module to obtain the predicted junction temperature value, thereby achieving accurate fitting of the junction temperature change law and providing key data support for monitoring the operating status of the IGBT power module.

[0090] In one possible implementation, monitoring for operational abnormalities in the IGBT power module based on predicted junction temperature values ​​includes:

[0091] Determine the residual value between the observed junction temperature and the predicted junction temperature within a preset time period to obtain the residual sequence; based on the statistical characteristics of the residual sequence, monitor whether there are any operational abnormalities in the IGBT power module.

[0092] Specifically, starting from the current moment, the process traces back a preset time period to obtain the junction temperature observations within that preset time period. The difference between the observed junction temperature values ​​and the corresponding predicted junction temperature values ​​is then determined as the residual value. For example, if there are 100 time points within the preset time period, corresponding to 100 observed junction temperature values ​​and 100 predicted junction temperature values, subtracting the observed and predicted values ​​yields a residual sequence containing 100 residual values.

[0093] When the junction temperature prediction model is trained using operating data from a healthy IGBT power module, the predicted junction temperature output by the model is the theoretical junction temperature when the IGBT's thermal characteristics are normal. If the observed junction temperature deviates significantly from the predicted value, it indicates an abnormality in the IGBT power module's thermal characteristics. To eliminate errors caused by occasional inaccurate junction temperature observations, reliable residual information is analyzed from the residual sequence based on its statistical characteristics. For example, a 95% confidence interval for the residual value is determined. This allows for a more accurate assessment of whether the IGBT power module is experiencing operational abnormalities.

[0094] In one implementation, a variational autoencoder (VAE) is used to model the residual sequence in a healthy state. By calculating the difference between the real-time residual and the reconstructed residual, such as KL divergence or reconstruction error, it is determined whether there is an operational abnormality in the IGBT power module.

[0095] Another alternative implementation is unsupervised anomaly detection based on Isolation Forest or One-Class SVM. This method applies an unsupervised anomaly detection algorithm to the residual sequence to identify anomalies that deviate significantly from the distribution of healthy states, and is suitable for scenarios where fault samples are lacking.

[0096] The abnormal monitoring method for IGBT power modules provided in this application uses the predicted junction temperature as a reference value, determines the residual value between the observed junction temperature value and the predicted junction temperature value within a preset time period, obtains the residual sequence, and monitors whether there is an operational abnormality in the IGBT power module based on the statistical characteristics of the residual sequence. This enables accurate analysis of the junction temperature change trend of the IGBT power module and accurate and early warning of operational abnormalities in the IGBT power module.

[0097] In one possible implementation, monitoring for operational anomalies in the IGBT power module is based on the statistical characteristics of the residual sequence, including:

[0098] Based on the sliding window mechanism, the mean and standard deviation of the residual sequence within the sliding window are calculated.

[0099] Based on the mean and standard deviation, the confidence interval of the residual sequence in the sliding window is determined, and the confidence level of the residual values ​​contained in the confidence interval is greater than the preset confidence threshold.

[0100] Based on the upper limit of the confidence interval corresponding to the sliding window and the preset residual threshold, monitor whether there are any abnormalities in the operation of the IGBT power module.

[0101] For example, the residual sequence contains 100 residual values, and the sliding window length is 30. The first sliding window contains the 1st to 30th residual values ​​of the residual sequence. The mean and standard deviation of these 30 residual values ​​are calculated, and the confidence intervals corresponding to these 30 residual values ​​are determined based on the mean and standard deviation. For example, the confidence level of the confidence interval is 95%. Thus, the confidence interval corresponding to the first sliding window is obtained. Then the sliding window slides, and the second sliding window contains the 2nd to 31st residual values ​​of the residual sequence. Similarly, the confidence interval corresponding to the second sliding window is calculated, and so on, to obtain the confidence intervals corresponding to multiple sliding windows.

[0102] If the upper limit of the confidence interval corresponding to the sliding window is greater than the preset residual threshold, it indicates that there is a large deviation between the junction temperature observation value and the junction temperature prediction value, and the confidence level is high. It can be determined that a relatively credible IGBT power module malfunction has occurred at this time, and an early warning should be issued in a timely manner.

[0103] The anomaly monitoring method for IGBT power modules provided in this application adopts a sliding window mechanism to analyze the statistical characteristics of the residual sequence. By comparing the upper limit of the confidence interval corresponding to the sliding window with the residual threshold, it is determined whether there is an operational anomaly in the IGBT power module. This method can detect anomalies in a timely manner and reduce misjudgments caused by occasional jump data.

[0104] In one possible implementation, monitoring for operational abnormalities in the IGBT power module is performed based on the upper limit of the confidence interval corresponding to the sliding window and a preset residual threshold, including:

[0105] If the upper limit of the confidence interval corresponding to N consecutive sliding windows is greater than the residual threshold, it is determined that the IGBT power module has an operational abnormality, where N is the preset window threshold.

[0106] Specifically, based on the above embodiments, in order to further enhance the reliability of anomaly detection, when the upper limit of the confidence interval corresponding to N consecutive sliding windows is greater than the residual threshold, it is determined that the IGBT power module has an operational anomaly. For example, N is 3.

[0107] The IGBT power module anomaly monitoring method provided in this application triggers an alarm for IGBT power module malfunction when the upper limit of the confidence interval corresponding to multiple consecutive sliding windows is greater than the residual threshold, thus ensuring the reliability of anomaly monitoring and reducing false judgments.

[0108] In one possible implementation, the subset of key features is obtained in the following manner:

[0109] Based on the random forest algorithm, feature importance analysis was performed on the operating data of IGBT power modules to obtain the first contribution of each data category in the operating data to the junction temperature prediction of IGBT power modules.

[0110] Based on the XGboost algorithm, feature importance analysis was performed on the running data to obtain the second contribution of each data category in the running data to the prediction of IGBT power module junction temperature;

[0111] The first and second contributions corresponding to the running data are weighted and merged to obtain the third contribution corresponding to the running data.

[0112] The target data category whose third contribution is greater than the preset contribution threshold in the running data is identified as the key feature subset.

[0113] Specifically, complete operating data of the IGBT power module under different operating conditions is first collected. For example, this includes line voltage, total active power, and three-phase voltage, current, and temperature time series data of the A / B / C phases of the IGBT module. After cleaning, outlier removal, and standardization preprocessing, the operating data is input into the optimized random forest algorithm and XGboost algorithm.

[0114] Based on the powerful feature evaluation capability of the random forest algorithm, multiple independent decision trees are constructed and the prediction results are integrated. During the training process, the corresponding feature importance index is automatically calculated according to the degree of influence of each data category on the junction temperature prediction result. Finally, the feature importance score of each category in the running data is obtained by ranking the index, thereby clarifying the first contribution of each data category to the junction temperature prediction of IGBT power modules.

[0115] The XGboost algorithm iteratively constructs multiple gradient boosting decision trees, optimizing model performance by guiding the gradient descent direction of the loss function. During training, it automatically calculates the corresponding feature importance index based on the influence weight of each data category on the junction temperature prediction result. Finally, by quantifying and ranking the index, it obtains the feature importance score corresponding to each data category in the running data, thereby clarifying the second contribution of each data category to the junction temperature prediction of the IGBT power module.

[0116] To avoid the risk of inaccurate contribution assessment from a single algorithm, the first and second contribution scores are weighted to obtain a third contribution score that combines the results of both algorithms. After obtaining the third contribution score for each data category, the data categories are sorted according to the third contribution score, and the categories with a third contribution score greater than a preset contribution score threshold are identified as the key feature subset.

[0117] The IGBT power module anomaly monitoring method provided in this application calculates the contribution of each data category to junction temperature prediction based on the random forest algorithm and the XGboost algorithm, respectively, and merges the results of the two to obtain a more accurate third contribution. Based on the third contribution, the data categories that have a greater impact on junction temperature prediction are determined, and a key feature subset is obtained. Junction temperature prediction is performed based on the key feature subset, which can effectively improve the accuracy of junction temperature prediction.

[0118] In one possible implementation, the subset of key features is obtained in the following manner:

[0119] The nonlinear relationship between the operating data of the IGBT power module and the junction temperature of the IGBT power module is evaluated based on the mutual information algorithm, and the corresponding highly correlated subsets are obtained.

[0120] Based on the recursive feature elimination algorithm, redundant features in the relevant subset are removed to obtain the key feature subset.

[0121] Specifically, based on the powerful nonlinear relationship measurement capability of the mutual information algorithm, the nonlinear dependency between the two is accurately assessed by quantifying the degree of information sharing between different operating data categories and junction temperature. The operating data categories with mutual information values ​​higher than a preset threshold are selected to form a relevant subset that is highly correlated with junction temperature, effectively preserving the core information that has a significant impact on junction temperature changes.

[0122] Subsequently, based on the iterative screening mechanism of the Recursive Feature Elimination (RFE) algorithm, and relying on the optimized base model, the relevant subset is used as the initial feature set input to the model. The model is trained multiple times, feature weights are calculated, and redundant features with the lowest weights are removed. This process is repeated until the preset number of features or model performance standards are met. Finally, redundant information and duplicate features in the relevant subset are eliminated to obtain a concise key feature subset with strong representational capabilities.

[0123] The IGBT power module anomaly monitoring method provided in this application first selects a highly correlated subset based on the mutual information algorithm, and then removes redundant features from the correlated subset based on the recursive feature elimination algorithm to obtain a key feature subset, providing high-quality input for the subsequent construction of the junction temperature prediction model, and helping to improve the prediction accuracy and operating efficiency of the junction temperature prediction model.

[0124] In one possible implementation, before predicting the junction temperature of the IGBT power module based on multi-source operating data and a subset of key features of the IGBT power module, and obtaining the predicted junction temperature value, the method further includes:

[0125] Preprocessing of multi-source operational data includes data cleaning and data standardization.

[0126] Specifically, data cleaning refers to removing obvious outliers and filling in missing values, such as using forward and backward imputation or interpolation to generate intermediate values. Data standardization refers to normalizing multi-source operational data to a unified dimension; for example, normalizing multi-source operational data to the [0,1] interval.

[0127] In one implementation, principal component analysis or factor analysis is performed on multi-source operational data to map high-dimensional features to a low-dimensional space, retaining the main variation information, thereby achieving feature compression and denoising, and reducing data complexity.

[0128] In one implementation, different sampling frequencies (such as high-frequency current sampling and low-frequency temperature sampling) are used for different sensor data, and multi-rate data fusion is achieved through interpolation or alignment methods to improve data utilization.

[0129] In one implementation, when data is scarce, Generative Adversarial Networks (GANs) are used to generate multi-source running data that conforms to the real distribution, thereby expanding the training samples and improving the robustness of the model.

[0130] The IGBT power module anomaly monitoring method provided in this application preprocesses multi-source operating data to ensure data reliability and improve the accuracy of junction temperature prediction.

[0131] Figure 2 This is a diagram illustrating the architecture of an IGBT power module anomaly monitoring system provided in an embodiment of this application. Figure 2 As shown, the data acquisition module collects multi-source operating data from the IGBT module, including both regular and module-specific data, with an acquisition frequency of 1 minute or 5 minutes. The feature selection module then filters key feature variables. First, the data is preprocessed, then the importance of features is calculated using a random forest algorithm, ranked, and weighted for fusion. Finally, a subset of features contributing significantly to IGBT junction temperature prediction is selected. The temperature prediction module builds an Attention-GRU-DNN hybrid prediction model, using the historical junction temperature sequence X1 and key feature variable X2 as input to obtain the current IGBT temperature prediction value. The anomaly warning module provides anomaly warnings based on the prediction residual sequence. It calculates the residual sequence between the actual junction temperature value and the model prediction value, calculates the statistical characteristics and confidence interval of the window residual, and compares the residual mean confidence interval with a dynamic threshold. When an anomaly is detected, an warning is triggered.

[0132] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 3 As shown, the core components of the electronic device include:

[0133] Processor: Executes the computer program stored in the memory to implement the IGBT power module anomaly monitoring method in the above embodiment.

[0134] Memory: Stores the computer program and historical and real-time operating data of the IGBT module.

[0135] I / O interface: Connects to external devices (such as sensors and monitoring systems), and is responsible for the input of raw data and the output of early warning signals.

[0136] Network adapter: Used to connect devices to the cloud platform to enable data uploading and remote monitoring.

[0137] Figure 4 This is a schematic diagram of the structure of the abnormal monitoring device for the IGBT power module provided in the embodiments of this application, as shown below. Figure 4As shown, the anomaly monitoring device 40 for the IGBT power module provided in this embodiment includes:

[0138] The acquisition module 401 is used to acquire multi-source operating data of IGBT power modules based on key feature subsets. The key feature subsets are data categories that contribute highly to the prediction of junction temperature of IGBT power modules by performing feature importance analysis on the operating data of IGBT power modules.

[0139] The prediction module 402 is used to predict the junction temperature of the IGBT power module based on multi-source operating data and the junction temperature sequence within a preset time period, and obtain the predicted junction temperature value.

[0140] The monitoring module 403 is used to monitor whether there are any abnormalities in the operation of the IGBT power module based on the predicted junction temperature value.

[0141] In one possible implementation, the prediction module 402 is specifically used for:

[0142] Based on gated recurrent units, real-time temporal features of junction temperature sequences are extracted;

[0143] The hidden states of the gated recurrent unit are weighted based on the temporal attention mechanism to obtain the weighted temporal features of the junction temperature sequence;

[0144] Based on multi-source operating data, real-time timing characteristics, and weighted timing characteristics, the junction temperature of the IGBT power module is predicted to obtain the predicted junction temperature value.

[0145] In one possible implementation, the prediction module 402 is specifically used for:

[0146] Multi-source operational data, real-time time-series features, and weighted time-series features are concatenated to obtain comprehensive time-series features;

[0147] Based on the comprehensive timing characteristics, a fully connected deep neural network model is used to perform nonlinear fitting on the junction temperature of the IGBT power module to obtain the predicted junction temperature value.

[0148] In one possible implementation, the monitoring module 403 is specifically used for:

[0149] Determine the residual values ​​between the observed and predicted junction temperatures within a preset time period to obtain the residual sequence;

[0150] Based on the statistical characteristics of the residual sequence, monitor whether there are any operational abnormalities in the IGBT power module.

[0151] In one possible implementation, the monitoring module 403 is specifically used for:

[0152] Based on the sliding window mechanism, the mean and standard deviation of the residual sequence within the sliding window are calculated.

[0153] Based on the mean and standard deviation, the confidence interval of the residual sequence in the sliding window is determined, and the confidence level of the residual values ​​contained in the confidence interval is greater than the preset confidence threshold.

[0154] Based on the upper limit of the confidence interval corresponding to the sliding window and the preset residual threshold, monitor whether there are any abnormalities in the operation of the IGBT power module.

[0155] In one possible implementation, the monitoring module 403 is specifically used for:

[0156] If the upper limit of the confidence interval corresponding to N consecutive sliding windows is greater than the residual threshold, it is determined that the IGBT power module has an operational abnormality, where N is the preset window threshold.

[0157] In one possible implementation, the acquisition module 401 is further configured to:

[0158] Based on the random forest algorithm, feature importance analysis was performed on the operating data of IGBT power modules to obtain the first contribution of each data category in the operating data to the junction temperature prediction of IGBT power modules.

[0159] Based on the XGboost algorithm, feature importance analysis was performed on the running data to obtain the second contribution of each data category in the running data to the prediction of IGBT power module junction temperature;

[0160] The first and second contributions corresponding to the running data are weighted and merged to obtain the third contribution corresponding to the running data.

[0161] The target data category whose third contribution is greater than the preset contribution threshold in the running data is identified as the key feature subset.

[0162] In one possible implementation, the acquisition module 401 is further configured to:

[0163] The nonlinear relationship between the operating data of the IGBT power module and the junction temperature of the IGBT power module is evaluated based on the mutual information algorithm, and the corresponding highly correlated subsets are obtained.

[0164] Based on the recursive feature elimination algorithm, redundant features in the relevant subset are removed to obtain the key feature subset.

[0165] In one possible implementation, the acquisition module 401 is further configured to:

[0166] Preprocessing of multi-source operational data includes data cleaning and data standardization.

[0167] The abnormal monitoring device for the IGBT power module provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0168] Figure 5 This is a schematic diagram of the structure of an anomaly monitoring device for an IGBT power module provided in an embodiment of this application. Figure 5 As shown, the IGBT power module anomaly monitoring device 50 provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the IGBT power module anomaly monitoring device 50 further includes a communication interface 503. The processor 501, memory 502, and communication interface 503 are connected via a communication bus 504.

[0169] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0170] The specific implementation process of processor 501 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0171] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0172] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0173] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0174] This application also provides a computer program product, including a computer program that, when executed, implements the above-described method.

[0175] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed, implement the above-described method.

[0176] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0177] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0178] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0179] 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.

[0180] 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.

[0181] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part 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 of 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.

[0182] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0183] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. An abnormality monitoring method of an IGBT power module, characterized by, The method comprises the following steps: According to the key feature subset, the multi-source operating data of the IGBT power module is obtained, wherein the key feature subset is obtained by performing feature importance analysis on the operating data of the IGBT power module, and the data category with high contribution degree to the junction temperature prediction of the IGBT power module is screened out; Based on the multi-source operating data and the junction temperature sequence within the preset time length, the junction temperature prediction of the IGBT power module is performed to obtain the junction temperature prediction value; According to the junction temperature prediction value, it is monitored whether the IGBT power module has operating abnormity.

2. The abnormality monitoring method of an IGBT power module according to claim 1, characterized by, The method comprises the following steps: Based on the multi-source operating data and the junction temperature sequence within the preset time length, the junction temperature prediction of the IGBT power module is performed to obtain the junction temperature prediction value, which comprises the following steps: Based on the gating cycle unit, the real-time time sequence feature of the junction temperature sequence is extracted; Based on the time sequence attention mechanism, the hidden state of the gating cycle unit is weighted to obtain the weighted time sequence feature of the junction temperature sequence; 3. The abnormality monitoring method of an IGBT power module according to claim 2, characterized by, According to the multi-source operating data, the real-time time sequence feature and the weighted time sequence feature, the junction temperature prediction of the IGBT power module is performed to obtain the junction temperature prediction value. The method comprises the following steps: The multi-source operating data, the real-time time sequence feature and the weighted time sequence feature are spliced to obtain the comprehensive time sequence feature; 4. The abnormality monitoring method of an IGBT power module according to any one of claims 1 to 3, characterized by, According to the comprehensive time sequence feature, the junction temperature of the IGBT power module is nonlinearly fitted based on the deep neural network model with full connection to obtain the junction temperature prediction value. The method comprises the following steps: The method comprises the following steps:

5. The abnormality monitoring method of an IGBT power module according to claim 4, characterized by, The residual sequence is obtained by determining the residual value of the observed value of the junction temperature within the preset time length compared with the junction temperature prediction value; According to the statistical characteristics of the residual sequence, it is monitored whether the IGBT power module has operating abnormity. The method comprises the following steps: Based on the sliding window mechanism, the residual mean and standard deviation of the residual sequence within the sliding window are calculated; 6. The abnormality monitoring method of an IGBT power module according to claim 5, characterized by, According to the mean and standard deviation, the confidence interval corresponding to the sliding window of the residual sequence is determined, and the confidence degree of the residual value contained in the confidence interval is greater than the preset confidence threshold; According to the upper limit of the confidence interval corresponding to the sliding window and the preset residual threshold, it is monitored whether the IGBT power module has operating abnormity.

7. The abnormality monitoring method of an IGBT power module according to any one of claims 1 to 3, characterized by, The method comprises the following steps: If the upper limit of the confidence interval corresponding to the continuous N sliding windows is greater than the residual threshold, it is determined that the IGBT power module has operating abnormity, wherein N is a preset window threshold. The key feature subset is obtained by the following method: The first contribution degree of each data category in the operating data of the IGBT power module to the junction temperature prediction of the IGBT power module is obtained by performing feature importance analysis on the operating data of the IGBT power module based on the random forest algorithm; perform feature importance analysis on the operation data based on an XGboost algorithm to obtain a second contribution degree of each data category in the operation data to IGBT power module junction temperature prediction; weighting and fusing the first contribution degree and the second contribution degree corresponding to the operation data to obtain a third contribution degree corresponding to the operation data; determining, as the key feature subset, a target data category in the operation data corresponding to the third contribution degree greater than a preset contribution degree threshold.

8. The abnormality monitoring method of an IGBT power module according to any one of claims 1 to 3, characterized by, The key feature subset is obtained by: based on the mutual information algorithm, evaluating the nonlinear relationship between the operation data of the IGBT power module and the junction temperature of the IGBT power module to obtain a relevant subset corresponding to high correlation; based on the recursive feature elimination algorithm, eliminating redundant features in the relevant subset to obtain the key feature subset.

9. The abnormality monitoring method of an IGBT power module according to any one of claims 1 to 4, characterized by, Before performing the junction temperature prediction of the IGBT power module based on the multi-source operation data and the junction temperature sequence within the preset time length to obtain the junction temperature prediction value, the method further includes: preprocessing the multi-source operation data, wherein the preprocessing includes data cleaning and data standardization.

10. An abnormality monitoring device of an IGBT power module, characterized by, It includes: an acquisition module configured to acquire multi-source operation data of an IGBT power module according to a key feature subset, wherein the key feature subset is a data category with high contribution degree to IGBT power module junction temperature prediction, which is screened out by performing feature importance analysis on operation data of the IGBT power module; a prediction module configured to perform junction temperature prediction of the IGBT power module based on the multi-source operation data and a junction temperature sequence within a preset time length to obtain a junction temperature prediction value; a monitoring module configured to monitor whether the IGBT power module has an operation abnormality according to the junction temperature prediction value.

Images