Integrated fault prediction comprehensive power supply test method and system

Abstract

The application discloses a kind of integrated fault prediction comprehensive power supply test method and system, and it is related to power supply test technical field.A kind of integrated fault prediction comprehensive power supply test system, including have: test parameter acquisition module, reference state update module, data fusion module, fault prediction module and retest record module.The application is by setting first prediction channel and second prediction channel, and is combined with initial abnormality judgment result flexible selection model path, while guaranteeing the prediction efficiency, has improved prediction accuracy, adapts to the operation demand under different complexity scene;By setting flow analysis mechanism and dynamic flow transfer limit, and based on the weighting operation of maximum influence data item, the intelligent verification and self-adaptive adjustment of model result consistency are realized, and the stability and robustness of fault judgment process are enhanced.

Description

Technical Field

[0001] This invention relates to the field of power supply testing technology, and in particular to a comprehensive power supply testing method and system that integrates fault prediction. Background Technology

[0002] As a critical infrastructure in modern industrial and information systems, the operating status of power supply equipment directly affects the stability and security of the system. With the rapid development of intelligent manufacturing, data centers, and communication equipment, power supply systems face higher operating loads and stability requirements. However, traditional power supply testing and fault diagnosis methods mostly rely on periodic inspections and rule-based anomaly identification, making it difficult to promptly detect potential fault trends or capture multi-factor anomalies in complex scenarios.

[0003] Existing power supply testing methods mainly focus on the measurement of electrical parameters, such as real-time monitoring of voltage, current, and frequency. Some systems have introduced auxiliary means such as infrared thermal imaging for image recognition after a fault occurs. However, these methods generally suffer from the following technical shortcomings: limited data dimensions, lack of in-depth fusion analysis, lack of effective predictive capabilities, weak model generalization ability, poor adaptability to individual device differences, fragmented knowledge accumulation, and lack of closed-loop learning mechanisms.

[0004] Therefore, there is an urgent need for a power supply testing system and method that can integrate multimodal data, has adaptive and fault prediction capabilities, and supports knowledge accumulation and dynamic updates, in order to improve the ability to identify and respond to complex faults and realize the transformation from passive response to proactive prediction. Summary of the Invention

[0005] A comprehensive power supply testing method integrating fault prediction, comprising:

[0006] During power supply testing, acquire multimodal data of the power supply equipment, including electrical parameter information, operating environment information, thermal imaging information, and operation log information;

[0007] Based on historical multimodal data, a reference state transition diagram is constructed for the power supply equipment, including the fault evolution path from "normal state" to "abnormal trend" and then to "fault state", and a fault knowledge graph corresponding to the relationship of "fault type - parameter performance - handling suggestion" for each standard fault case;

[0008] The stable fluctuation period of each data in the electrical parameter information is calculated, and the weighted average of the stable fluctuation periods of each data is taken as the fusion period. All multimodal data within the fusion period are obtained and fused to obtain the periodic fusion feature. The multidimensional health status profile of the power supply equipment is obtained through feature engineering. The multidimensional health status profile is initially judged for anomalies, and the results include normal and abnormal.

[0009] The periodic fusion features are input into the fault prediction model for fault prediction. The fault prediction model selects a prediction channel based on the result of the initial anomaly judgment. It contains two independent prediction channels. The first prediction channel is based on a lightweight model for fast analysis, and the second prediction channel is based on a convolutional neural network for in-depth analysis. The results of the prediction channels are processed and analyzed within the fault prediction model to obtain the initial test results.

[0010] As a preferred embodiment of the present invention, a comprehensive power supply testing method integrating fault prediction further includes:

[0011] Based on the initial test results, a fault knowledge graph is matched, and the retest weight is obtained based on the corresponding parameter performance. The power supply equipment is tested again, and the retest weight is used to weight the periodic fusion features to obtain the retest periodic fusion features. Fault prediction is performed through the fault prediction model, and the output correction result or final result is determined based on whether the number of flow analysis exceeds the flow limit.

[0012] If the output is a corrected result, the corrected result is used to match the fault knowledge graph and obtain the retest weight. The test is then performed again until the final result is obtained, or the number of retests exceeds the set threshold.

[0013] The retest weights are incorporated into the personalized device test samples after each weighting of the periodic fusion features; the personalized device test samples are used for weighting during the data fusion process.

[0014] By performing cluster analysis on personalized equipment test samples and constructing personalized equipment features in the equipment twin digital model based on the cluster analysis results, the personalized equipment features are used to correct the fault evolution path.

[0015] As a preferred embodiment of the present invention, the fault prediction model uses a first prediction channel for analysis when the initial anomaly judgment result is normal; and uses a second prediction channel for analysis when the initial anomaly judgment result is abnormal.

[0016] If the analysis result of the prediction channel is consistent with the initial anomaly judgment result, the initial test result is output; otherwise, a flow analysis is performed. The flow analysis specifically involves switching the prediction channel for analysis. If the analysis result of the prediction channel after the flow analysis is inconsistent with the analysis result of the prediction channel before the flow analysis, the flow analysis continues until the result is consistent or the number of flow analyses exceeds the flow limit, and the initial test result is output.

[0017] As a preferred technical solution of the present invention, before the flow analysis is confirmed, the fault prediction model weights the input periodic fusion feature according to the maximum impact data item corresponding to the result of the prediction channel, that is, adds a set weight value to the corresponding part of the periodic fusion feature for the maximum impact data item.

[0018] As a preferred technical solution of the present invention, the circulation limit is a dynamic value, which is determined by using an initial setting value combined with the number of retests as independent variables, and is dynamically adjusted by setting a mathematical function.

[0019] As a preferred embodiment of the present invention, the data fusion includes:

[0020] During the fusion cycle, features are extracted from electrical parameter information, operating environment information, thermal imaging information, and operation log information to form a structured feature set;

[0021] Each feature is assigned an initial fusion weight by a feature importance assessment method, and the structured feature set is fused to generate a high-dimensional fusion vector in a unified format. The t-SNE method is then used for dimensionality reduction to obtain the final periodic fusion feature.

[0022] The periodic fusion features are matched with the device features in the personalized device test samples based on similarity. The fusion weights are dynamically adjusted according to the similarity threshold to achieve adaptive feature reconstruction for different device states.

[0023] As a preferred embodiment of the present invention, the initial anomaly determination includes:

[0024] A time series feature trajectory map is constructed based on multimodal data, and a sliding window mechanism is used to fit the trends of electrical parameter information and operating environment information to extract statistical features;

[0025] The above statistical features are analyzed by using an anomaly trend detection model. The anomaly trend detection model is an unsupervised anomaly detection model built based on the isolated forest algorithm, which is used to identify data segments that have not yet triggered the hard fault threshold but have potential anomaly trends.

[0026] The system judges and outputs the initial anomaly judgment result based on the results of the hard fault threshold and the abnormal trend detection model. The judgment method is that if only two results are normal, the initial anomaly judgment result is output as normal; otherwise, it is an anomaly.

[0027] A comprehensive power supply test system integrating fault prediction, comprising:

[0028] Test parameter acquisition module: Acquires multimodal data of the power supply device during power supply testing;

[0029] Reference State Update Module: Used to acquire and update the reference state transition diagram corresponding to the power supply equipment;

[0030] Data fusion module: performs data fusion on all multimodal data within the fusion cycle;

[0031] Fault prediction module: Used to calculate the periodic fusion data according to the fault prediction model, and obtain the initial test results, correction results and final test results of fault prediction;

[0032] Retest Record Module: Used to acquire and record the retest weight, personalized equipment test samples, and personalized equipment characteristics during the retesting process of power supply equipment.

[0033] The present invention has the following advantages:

[0034] This invention constructs a state transition diagram and a fault knowledge graph, and combines periodic fusion features and multi-dimensional health status profile analysis to achieve dynamic modeling of the operating trend of power equipment and accurate characterization of fault evolution paths, significantly improving the ability to identify abnormal trends.

[0035] This invention improves prediction accuracy while ensuring prediction efficiency by setting up a first prediction channel and a second prediction channel, and flexibly selecting the model path based on the initial anomaly judgment results, thus adapting to the operational needs of different complexity scenarios. By setting up a flow analysis mechanism and dynamic flow limits, as well as weighting operations based on the data item with the greatest impact, it realizes intelligent verification and adaptive adjustment of the consistency of model results, thereby enhancing the stability and robustness of the fault judgment process.

[0036] This invention constructs a digital twin model of equipment by introducing a clustering analysis mechanism of personalized equipment test samples and equipment features, thereby achieving adaptive fault judgment based on individual equipment differences and improving the model's generalization ability and applicability. Through a similarity matching and feature reconstruction mechanism between periodically fused features and personalized equipment samples, it supports dynamic optimization of fusion weights for specific operating states, further improving the pertinence and accuracy of feature expression.

[0037] This invention introduces the isolated forest algorithm into the initial anomaly judgment to perform discriminative analysis on trend statistical features, which can realize early warning of abnormal trends before the traditional fault threshold is reached, and has stronger foresight and initiative. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only schematic diagrams of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of the integrated fault prediction power supply test system used in an embodiment of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0041] Example 1: A comprehensive power supply testing method integrating fault prediction, comprising the following steps:

[0042] Step S1: Acquire multimodal data of the power supply equipment during the power supply test, including electrical parameter information, operating environment information, thermal imaging information and operation log information;

[0043] The electrical parameter information includes at least voltage, current, active power, reactive power, power factor, and harmonics;

[0044] The operating environment information includes external environmental indicators such as temperature, humidity, altitude, wind speed, and dust concentration, and is obtained by using a multi-functional environmental sensor installed in the test environment for real-time sampling.

[0045] The thermal imaging information is obtained by capturing the thermal distribution image of the target device at certain time intervals (e.g., 5 to 60 seconds) using a thermal imaging camera, and converting the thermal image into temperature matrix data using an infrared temperature calibration model;

[0046] The operation log information includes manual operation records, test program execution logs, and system exception logs, and supports data synchronization and capture of the operation and maintenance system through standard interface protocols (such as MODBUS, IEC61850, etc.).

[0047] All types of data are collected synchronously and aligned using timestamps to ensure consistency of data across different modalities over time.

[0048] To enhance data integrity, the system also includes a data quality assessment module, which performs integrity checks, outlier removal, and missing value imputation on the collected data. The missing value imputation methods include K-nearest neighbor interpolation, time series autoregressive interpolation, and multimodal information collaborative imputation.

[0049] Step S2: Construct a reference state transition diagram for the power supply equipment based on historical multimodal data, including the fault evolution path from "normal state" to "abnormal trend" and then to "fault state", and a fault knowledge graph of the relationship between "fault type - parameter performance - handling suggestions" for each standard fault case;

[0050] The reference state transition diagram is constructed based on cluster analysis and state labeling. It uses segmented time windows to divide historical data into blocks and calculates the characteristic trend change amount for each data block, including mean change rate, fluctuation index, anomaly frequency, etc., to calibrate the equipment operating status.

[0051] The transition paths between states are modeled using a state transition probability matrix. The states are classified into at least five categories: normal, sub-healthy, abnormal trend, critical fault, and failure. The transition paths can be fitted and optimized using Markov chain modeling.

[0052] The fault knowledge graph is a triplet graph structure, containing entity nodes (fault type, parameter behavior, and handling suggestions) and relation edges ("trigger", "corresponding", and "suggestion"), and supports storage in a graph database (Neo4j).

[0053] The fault knowledge graph has an automatic update mechanism, including:

[0054] After the fault prediction model outputs the final result, if the result is marked as a misjudged or uncovered new fault type after manual review or subsequent actual operation and maintenance verification, the current period fusion feature, prediction label, actual fault type and handling measures will be written as a new sample into the queue to be updated.

[0055] Semantic extraction and parameter standardization are performed on the data in the update queue to automatically generate a triplet of "fault type - parameter behavior - handling suggestion";

[0056] A graph embedding entity matching and fusion algorithm is adopted to determine the similarity relationship between the newly added triplet and the entity nodes in the original graph, and to decide whether to add a new node, merge existing nodes or update edge weights based on a set threshold.

[0057] After the knowledge graph structure is updated, the update information is recorded through the graph version management module, and the re-scoring and path weight adjustment of the prediction model's associated paths are triggered to ensure the consistency between the graph and the model's reasoning logic.

[0058] The update process is set to be triggered periodically (every 24 hours) or executed after manual confirmation (such as in conjunction with an expert review mechanism) to ensure the accuracy and controllability of the map update.

[0059] Step S3: Calculate the stable fluctuation period of each data in the electrical parameter information, take the weighted average of the stable fluctuation periods of each data as the fusion period, obtain all multimodal data within the fusion period and perform data fusion to obtain the periodic fusion feature, and obtain the multidimensional health status profile of the power supply equipment through feature engineering processing; perform initial anomaly judgment on the multidimensional health status profile, and the results include normal and abnormal.

[0060] The calculation of the stable fluctuation period is based on signal analysis methods. Sliding window Fourier transform (STFT) and autocorrelation function analysis are performed on each electrical parameter information to extract its dominant frequency or periodic pattern. The stable period of each signal is identified by evaluating the energy concentration and peak spacing of the periodic fluctuation.

[0061] The weighted average is calculated by taking into account the weight of each electrical parameter on the fault sensitivity, which is determined based on the rate of change of different parameters before the fault occurs in historical data and the importance ranking in the prediction model.

[0062] The data fusion includes:

[0063] During the fusion cycle, features are extracted from electrical parameter information, operating environment information, thermal imaging information, and operation log information to form a structured feature set;

[0064] The electrical parameter information features include statistical features (mean, variance, maximum value, kurtosis, etc.) and frequency domain features (harmonic content, spectral energy, etc.); the operating environment information features include mean, volatility, and statistics of abrupt events; the thermal imaging information features obtain the temperature distribution of hot spots in the equipment through image segmentation, and extract regional temperature difference, maximum / minimum temperature, temperature rise gradient, etc.; the operation log information uses natural language processing techniques (such as TF-IDF, BERT encoding) to extract keyword vectors;

[0065] Each feature is assigned an initial fusion weight by a feature importance assessment method (XGBoost backtracking weight analysis), and the structured feature set is fused to generate a high-dimensional fusion vector in a unified format. The t-SNE method is then used for dimensionality reduction to obtain the final periodic fusion feature.

[0066] The periodic fusion features are matched with the device features in the personalized device test samples for similarity. Cosine similarity or Euclidean distance is used as the similarity metric. The fusion weight is dynamically adjusted according to the similarity threshold to achieve adaptive feature reconstruction for different device states.

[0067] The initial anomaly detection includes:

[0068] A time series feature trajectory map is constructed based on multimodal data, and a sliding window mechanism is used to fit the trend of electrical parameter information and operating environment information to extract statistical features (including at least trend slope, fluctuation amplitude, and coefficient of variation).

[0069] Trend fitting methods include linear least squares, polynomial fitting, or smoothing prediction based on Kalman filtering;

[0070] The above statistical features are analyzed by using an anomaly trend detection model. The anomaly trend detection model is an unsupervised anomaly detection model built based on the isolated forest algorithm, which is used to identify data segments that have not yet triggered the hard fault threshold but have potential anomaly trends.

[0071] The initial anomaly judgment result is determined and output based on the results of the hard fault threshold and the abnormal trend detection model. The judgment method is that if only two results are normal, the initial anomaly judgment result is output as normal; otherwise, it is abnormal.

[0072] The hard threshold is set based on the power supply device's factory specifications, historical safety limits, and expert experience, and can be customized according to device type.

[0073] Step S4: Input the periodic fusion features into the fault prediction model for fault prediction. The fault prediction model selects a prediction channel based on the initial anomaly judgment result. It contains two independent prediction channels. The first prediction channel performs fast analysis based on a lightweight model, and the second prediction channel performs in-depth analysis based on a convolutional neural network. The results of the prediction channels are processed and analyzed within the fault prediction model to obtain the initial test result.

[0074] The first prediction channel is a lightweight model combination based on ensemble learning algorithms, including random forest, gradient boosting tree (GBDT) and support vector machine (SVM), which performs fast classification processing on the input periodic fusion features;

[0075] The second prediction channel is a deep analysis model based on convolutional neural networks (CNN). The input layer receives periodic fusion features, and after one-dimensional or two-dimensional convolutional layers extract local temporal features, it is connected to a fully connected layer to output the probability of the fault category.

[0076] CNN models can be configured with multiple channels to process data from different modalities, and the importance of each modal feature can be dynamically adjusted through the attention mechanism module;

[0077] The fault prediction model uses the first prediction channel for analysis when the initial anomaly judgment result is normal; and uses the second prediction channel for analysis when the initial anomaly judgment result is abnormal.

[0078] If the analysis result of the prediction channel is consistent with the initial anomaly judgment result, the initial test result is output; otherwise, a flow analysis is performed. The flow analysis specifically involves switching the prediction channel for analysis. If the analysis result of the prediction channel after the flow analysis is inconsistent with the analysis result of the prediction channel before the flow analysis, the flow analysis continues until the result is consistent or the number of flow analyses exceeds the flow limit, and the initial test result is output.

[0079] Before the flow analysis is confirmed, the fault prediction model weights the input periodic fusion feature based on the data item with the greatest impact corresponding to the prediction channel result. That is, the maximum impact data item is given a set weight value in the corresponding part of the periodic fusion feature.

[0080] The data item with the greatest impact is determined by ranking the importance of features in the model, such as by extracting it through Grad-CAM heatmap analysis in CNN;

[0081] The circulation limit is a dynamic value, which is determined by using an initial set value combined with the number of retests as independent variables, and is dynamically adjusted by setting a mathematical function; the function can be an exponential decay function or a linear weighted function, used to balance test efficiency and prediction accuracy.

[0082] The fault prediction model has a dynamic migration mechanism, including:

[0083] When the statistical characteristics of the cluster to which the personalized device test sample belongs shift significantly, or when the historical prediction accuracy is continuously lower than the preset threshold within a set period, the model migration evaluation mechanism is triggered.

[0084] The model migration evaluation mechanism calculates the distribution distance between the source domain model and the target domain data based on the current device category, historical sample distribution, and the latest fusion features, including at least the maximum mean difference (MMD) and kernel mapping similarity.

[0085] When the distribution distance exceeds the set migration trigger threshold, the model migration module is activated, and a migration learning strategy based on domain adaptation technology is used to update the existing model parameters. The strategy includes at least freezing and fine-tuning some weight layers in the convolutional neural network, or using a domain discriminator for adversarial training optimization.

[0086] The migrated and updated model replaces the original model in subsequent fault prediction analysis. The migration process and its effects are recorded in the model evolution log, and update prompts are periodically sent to the administrator.

[0087] Step S5: Match the fault knowledge graph based on the initial test results, obtain the retest weight based on the corresponding parameter performance, test the power supply equipment again, use the retest weight to weight the periodic fusion features to obtain the retest periodic fusion features, and perform fault prediction through the fault prediction model. Then, determine the output correction result or final result based on whether the number of flow analysis exceeds the flow limit.

[0088] If the output result is a corrected result, the corrected result is used to match the fault knowledge graph, and the retest weight is obtained. The test is then performed again until the final result is obtained, or the number of retests exceeds a set threshold. The retest weight is incorporated into the personalized device test sample after each weighting of the periodic fusion features. The personalized device test sample is used for weighting during the data fusion process.

[0089] The retest weight is calculated based on the degree of matching between the initial test results and the historical fault types and parameter performance in the knowledge graph; the matching degree is quantified by the similarity calculation formula (cosine similarity) to obtain the weighting coefficient of each fault type;

[0090] The generation of retest weights also takes into account the frequency and severity of historical failure cases. If a certain type of failure occurs frequently and has a significant impact, then the weight of that type will be higher.

[0091] The weighting method for the periodic fusion features of the retest weights adopts either linear weighting or adaptive weighting. The linear weighting method involves linearly combining the weighting coefficients of each feature dimension, while the adaptive weighting method uses a machine learning algorithm (weighted least squares regression) to learn the optimal weighting coefficients based on historical data.

[0092] The weighted periodic fusion features are re-input into the fault prediction model for fault prediction, and the model output will be corrected and judged based on the number of flow analyses.

[0093] If the number of iterations exceeds the iteration limit and the corrected result differs significantly from the initial result, the corrected result will be output; otherwise, the final result will be output, and all iteration processes and analysis details will be recorded.

[0094] Step S6: Perform cluster analysis on the personalized equipment test samples, and construct personalized equipment features in the equipment twin digital model based on the cluster analysis results. The personalized equipment features are used to correct the fault evolution path.

[0095] The clustering analysis employs an unsupervised learning algorithm. First, the multidimensional features of the equipment test samples are standardized. Then, the high-dimensional data is reduced using the t-SNE method to reveal the underlying structure in the data.

[0096] Then, the device samples are clustered based on the hierarchical clustering algorithm, and similar device samples are grouped into the same class; the clustering results are evaluated by indicators such as silhouette coefficient and Davies-Bouldin index to ensure the effectiveness of the clustering effect.

[0097] Each cluster represents a type of equipment with similar operating characteristics and fault features. The personalized equipment features are the cluster center points or the most representative equipment features of that type of equipment.

[0098] In the equipment twin digital model, the personalized equipment characteristics of each piece of equipment are modeled using digital twin technology, and combined with historical operating data, failure modes and equipment characteristics, a virtual image of the equipment is formed;

[0099] The construction of personalized equipment features in the digital twin model includes not only electrical parameters at the hardware level, but also dynamic factors such as environment, temperature, and load. Through real-time monitoring and analysis of these features, the digital model can be updated in real time when the equipment changes, and potential failure trends of the equipment can be identified in advance.

[0100] By updating personalized equipment characteristics in real time, the model can dynamically adjust its inferences on fault evolution paths, especially when equipment status changes significantly (such as drastic fluctuations in the working environment or changes in operating load), and automatically correct the judgment criteria of the fault prediction model.

[0101] The revised fault evolution path will be continuously optimized and added to the fault knowledge graph by comparing it with actual historical fault data, ensuring that the model and the graph are updated in sync, and further improving the accuracy of prediction.

[0102] Example 2: A comprehensive power supply test system integrating fault prediction, see [link to example]. Figure 1 As shown, it includes the following modules:

[0103] Test parameter acquisition module: Acquires multimodal data of the power supply device during power supply testing;

[0104] Reference State Update Module: Used to acquire and update the reference state transition diagram corresponding to the power supply device;

[0105] Data fusion module: performs data fusion on all multimodal data within the fusion cycle;

[0106] Fault prediction module: Used to calculate the periodic fusion data according to the fault prediction model, and obtain the initial test results, correction results and final test results of fault prediction;

[0107] Retest Record Module: Used to acquire and record the retest weight, personalized equipment test samples, and personalized equipment characteristics during the retesting process of power supply equipment.

[0108] Example 3, power equipment fault diagnosis and prediction based on integrated power supply test bench, includes:

[0109] The integrated power supply test bench adopts a method that combines multimodal data acquisition with intelligent prediction models. By analyzing the input and output characteristics, electrical parameters, environmental information, thermal imaging data and operation logs of the power supply equipment, it realizes intelligent prediction of the health status of the power supply equipment and fault diagnosis.

[0110] 1. Multimodal data acquisition:

[0111] During power supply testing, the integrated power supply test bench is equipped with various sensors and data acquisition modules to collect electrical parameters, environmental parameters, thermal imaging information, and operation log information in real time.

[0112] 2. Data fusion and periodic feature calculation:

[0113] By fusing the periodic features of multimodal data, the stable fluctuation periods of electrical parameters, environmental changes, and equipment operation are calculated to obtain fused periodic features. Feature engineering methods are then used to extract and weightedly fuse the data, generating high-dimensional fused features to ensure that the multidimensional information of the data is comprehensively considered.

[0114] 3. Preliminary anomaly assessment and intelligent prediction:

[0115] After initial data processing, an unsupervised anomaly detection model based on the Isolation Forest algorithm is used to determine the health status of the power supply equipment. If the equipment status exhibits an abnormal trend, the next step of in-depth analysis is performed.

[0116] If the initial assessment is normal, a rapid evaluation can be conducted using a lightweight predictive model.

[0117] If an anomaly is initially identified, a convolutional neural network is used to further analyze the type and cause of the anomaly.

[0118] 4. Fault Knowledge Graph and Dynamic Updates:

[0119] By combining the fault knowledge graph, after the fault prediction model outputs its results, the prediction results, actual fault types, and handling measures are automatically added to the queue to be updated. Through the graph update algorithm, fault triples are automatically extracted and graph embedding analysis is performed to improve and update the fault evolution path and diagnostic rules of the equipment.

[0120] 5. Equipment health status profiling and continuous optimization:

[0121] Based on personalized equipment test samples, cluster analysis is used to categorize the equipment into different types, constructing personalized equipment characteristics. Digital twin technology is used to update the health status of the equipment in real time, and the fault prediction path is continuously optimized based on changes in equipment status.

[0122] 6. Fault prediction and intelligent adjustment:

[0123] During the operation of the integrated power supply test bench, the equipment fault prediction model intelligently predicts potential equipment faults based on real-time collected feature data and updated fault knowledge graphs. When the prediction result matches the initial anomaly assessment, a fault warning is output; if the results are inconsistent, a flow analysis mechanism switches the prediction channel to further ensure prediction accuracy.

[0124] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A comprehensive power supply testing method integrating fault prediction, characterized in that, include: During power supply testing, acquire multimodal data of the power supply equipment, including electrical parameter information, operating environment information, thermal imaging information, and operation log information; Based on historical multimodal data, a reference state transition diagram is constructed for the power supply equipment, including the fault evolution path from "normal state" to "abnormal trend" and then to "fault state", and a fault knowledge graph corresponding to the relationship of "fault type - parameter performance - handling suggestions" for each standard fault case; The stable fluctuation period of each data in the electrical parameter information is calculated, and the weighted average of the stable fluctuation period of each data is taken as the fusion period. All multimodal data within the fusion period are obtained and data fusion is performed to obtain the periodic fusion feature. Then, a multidimensional health status profile of the power supply equipment is obtained through feature engineering. Initial anomaly assessment is performed on the multidimensional health status profile, with results including normal and abnormal; The periodic fusion features are input into the fault prediction model for fault prediction. The fault prediction model selects a prediction channel based on the result of the initial anomaly judgment. It contains two independent prediction channels. The first prediction channel is based on a lightweight model for fast analysis, and the second prediction channel is based on a convolutional neural network for in-depth analysis. The results of the prediction channels are processed and analyzed within the fault prediction model to obtain the initial test results. Based on the initial test results, a fault knowledge graph is matched, and the retest weight is obtained based on the corresponding parameter performance. The power supply equipment is tested again, and the retest weight is used to weight the periodic fusion features to obtain the retest periodic fusion features. Fault prediction is performed through the fault prediction model, and the output correction result or final result is determined based on whether the number of flow analysis exceeds the flow limit. If the output is a corrected result, the corrected result is used to match the fault knowledge graph and obtain the retest weight. The test is then performed again until the final result is obtained, or the number of retests exceeds the set threshold. The retest weights are incorporated into the personalized device test samples after each weighting of the periodic fusion features; the personalized device test samples are used for weighting during the data fusion process. By performing cluster analysis on personalized equipment test samples and constructing personalized equipment features in the equipment twin digital model based on the cluster analysis results, the personalized equipment features are used to correct the fault evolution path.

2. The integrated power supply testing method with fault prediction according to claim 1, characterized in that, When the initial anomaly judgment result is normal, the fault prediction model uses the first prediction channel for analysis. When the initial anomaly assessment result is anomaly, the second prediction channel is used for analysis; If the analysis result of the prediction channel is consistent with the initial anomaly judgment result, the initial test result is output; otherwise, a flow analysis is performed. The flow analysis specifically involves switching the prediction channel for analysis. If the analysis result of the prediction channel after the flow analysis is inconsistent with the analysis result of the prediction channel before the flow analysis, the flow analysis continues until the result is consistent or the number of flow analyses exceeds the flow limit, and the initial test result is output.

3. The integrated power supply testing method with fault prediction according to claim 2, characterized in that, Before the flow analysis is confirmed, the fault prediction model weights the input periodic fusion feature based on the data item with the greatest impact corresponding to the prediction channel result. That is, the maximum impact data item is given a set weight value in the corresponding part of the periodic fusion feature.

4. The integrated power supply testing method with fault prediction according to claim 2, characterized in that, The circulation limit is a dynamic value, which is determined by using an initial setting value combined with the number of retests as independent variables, and is dynamically adjusted by setting a mathematical function.

5. The integrated power supply testing method with fault prediction according to claim 1, characterized in that, The data fusion includes: During the fusion cycle, features are extracted from electrical parameter information, operating environment information, thermal imaging information, and operation log information to form a structured feature set; Each feature is assigned an initial fusion weight by a feature importance assessment method, and the structured feature set is fused to generate a high-dimensional fusion vector in a unified format. The t-SNE method is then used for dimensionality reduction to obtain the final periodic fusion feature. The periodic fusion features are matched with the device features in the personalized device test samples based on similarity. The fusion weights are dynamically adjusted according to the similarity threshold to achieve adaptive feature reconstruction for different device states.

6. The integrated power supply testing method with fault prediction according to claim 1, characterized in that, The initial anomaly detection includes: A time series feature trajectory map is constructed based on multimodal data, and a sliding window mechanism is used to fit the trends of electrical parameter information and operating environment information to extract statistical features. The above statistical features are analyzed by using an anomaly trend detection model. The anomaly trend detection model is an unsupervised anomaly detection model built based on the isolated forest algorithm, which is used to identify data segments that have not yet triggered the hard fault threshold but have potential anomaly trends. The system judges and outputs the initial anomaly judgment result based on the results of the hard fault threshold and the abnormal trend detection model. The judgment method is that if only two results are normal, the initial anomaly judgment result is output as normal; otherwise, it is an anomaly.

7. A comprehensive power supply testing system integrating fault prediction, characterized in that, The system applies the method of any one of claims 1 to 6, including: Test parameter acquisition module: Acquires multimodal data of the power supply device during power supply testing; Reference State Update Module: Used to acquire and update the reference state transition diagram corresponding to the power supply equipment; Data fusion module: Performs data fusion on all multimodal data within the fusion period; Fault prediction module: Used to calculate the periodic fusion data according to the fault prediction model, and obtain the initial test results, correction results and final test results of fault prediction; Retest Record Module: Used to acquire and record the retest weight, personalized equipment test samples, and personalized equipment characteristics during the retesting process of power supply equipment.

Images