Electric power product operation fault detection system and method based on big data

Abstract

The invention discloses an electric power product operation fault detection system based on big data, and relates to the field of electric power fault detection systems, and the system comprises a data collection module which collects the multi-parameter operation data of various electric power products, including electrical and equipment states and environmental parameters, can dynamically adjust the collection frequency, and caches and encrypts the transmission data; the data preprocessing module is used for carrying out abnormal value correction, noise removal, normalization and integrity check; the feature extraction module is used for extracting key and time-frequency domain features by using multiple algorithms, selecting sensitive features according to a fault mode and identifying newly added features; and the fault detection model construction module constructs a model by combining machine learning and deep learning algorithms with ensemble learning and the like, and can update training. Through multi-module cooperation, features are effectively extracted to construct the model, accurate fault diagnosis and prediction are realized, visual display and knowledge support are provided, the intelligent level and efficiency of power product fault detection are improved, and stable operation of a power system is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of power fault detection systems, and in particular to a power product operation fault detection system based on big data. Background Art

[0002] With the rapid development of the power industry, power products are widely used in various links such as power generation, transmission, transformation, distribution and consumption. The stability and reliability of their operation are crucial to ensure the safe and efficient operation of the power system. However, power products are affected by various factors in a complex operating environment and are prone to various faults.

[0003] Traditional methods of fault detection for power products rely mainly on manual inspections and regular maintenance, which have many limitations. Manual inspections are inefficient and make it difficult to achieve real-time, comprehensive monitoring of many power products. The experience and skill levels of inspectors vary, which may lead to missed faults or misjudgments. Moreover, regular maintenance cycles are often fixed and cannot respond to dynamic changes in the operating status of power products in a timely manner. Faults may occur without being detected between maintenance intervals.

[0004] In recent years, with the rapid development of information technology and sensor technology, massive amounts of data have been generated during the operation of power products. However, existing fault detection technologies face challenges in processing these big data. On the one hand, the data collection methods are not intelligent enough, some sensors have limited accuracy and poor stability, and the accuracy and integrity of the collected data are difficult to guarantee. On the other hand, there is a lack of effective processing and analysis methods for the collected big data. During data preprocessing, operations such as noise removal, outlier processing, and data normalization are not accurate enough, which affects the subsequent analysis results. Feature extraction methods are not efficient enough, and it is difficult to mine key features from massive data, resulting in low accuracy and reliability of fault detection models.

[0005] In terms of fault detection model construction, traditional machine learning algorithms have limitations when dealing with complex nonlinear relationships. Although deep learning algorithms have strong learning capabilities, their application in the power field is not mature enough, and the model training time is long and the hardware requirements are high. In addition, the existing detection system lacks systematic knowledge base management, and the fault mode library, expert experience rule library, and historical fault case library are not updated in a timely manner, which cannot provide comprehensive and accurate support for fault diagnosis and prediction. At the same time, the visualization display is not intuitive enough and the interactivity is poor, which is not conducive to the operation and maintenance personnel to quickly understand and process fault information. In summary, there is an urgent need for a power product operation fault detection system and method based on big data to improve the accuracy, timeliness, and intelligence level of fault detection. Summary of the invention

[0006] The present invention proposes a power product operation fault detection system and method based on big data to solve the problems mentioned in the above-mentioned prior art.

[0007] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a power product operation fault detection system based on big data, comprising: The data acquisition module collects the operation data of power products in real time through various types of sensors and monitoring equipment. The power products include power generation equipment, power transmission equipment, power transformation equipment, power distribution equipment and power consumption equipment. The operation data covers electrical parameters such as voltage, current, resistance, power factor, frequency, equipment status parameters such as temperature, humidity, vibration amplitude, pressure, speed, insulation resistance, and environmental parameters such as ambient temperature, ambient humidity, and atmospheric pressure. The collected data has an accurate timestamp and has a data cache function. It temporarily stores data when the network is abnormal and automatically uploads it after the network is restored. The data acquisition module can also dynamically adjust the collection frequency according to the type of power product and the operating status, collect key parameters at high frequency, and collect stable parameters at low frequency. The data preprocessing module is connected to the data acquisition module and performs preprocessing operations on the collected data, including: using an outlier detection method based on statistical principles to identify and correct obviously erroneous data points, such as data that exceeds a certain multiple of the mean standard deviation is considered abnormal; using a combination of sliding average filtering and Kalman filtering to remove noise interference in the data; mapping data of different magnitudes to a specific interval through data normalization to facilitate subsequent processing; establishing data rationality inspection rules based on the operating rules of power products and expert experience, checking the integrity and logical consistency of the data, and filling in missing data; The feature extraction module uses principal component analysis (PCA) and linear discriminant analysis (LDA) and other dimensionality reduction algorithms to extract features from preprocessed data and obtain key feature vectors. It also uses wavelet transform and empirical mode decomposition (EMD) time-frequency analysis methods to extract the time-frequency domain features of the data, including frequency distribution and energy distribution. It selects features that are sensitive to faults based on the power product fault pattern library, such as features related to dissolved gas content in oil for transformer faults and zero-sequence current and zero-sequence voltage features for transmission line faults. It can also automatically identify new fault features, analyze the differences between new data and historical fault data through machine learning algorithms, and discover potential important features. The fault detection model construction module builds a fault detection model based on machine learning algorithms and deep learning algorithms. The specific steps are as follows: the extracted features are divided into training sets and test sets, of which the training set accounts for 70%-85%; the training set is trained using support vector machines (SVM), decision trees, random forest machine learning algorithms, and convolutional neural networks (CNN), recurrent neural networks (RNN) and their variants such as long short-term memory networks (LSTM) and gated recurrent units (GRU) deep learning algorithms. During the training process, the cross-validation method is used to evaluate the model performance and adjust the model hyperparameters; the attention mechanism is introduced to focus the model on key features; the results of multiple base models are integrated through ensemble learning methods such as voting and weighted average methods to improve the accuracy and generalization ability of the model; The fault diagnosis and prediction module receives the feature vector after feature extraction from real-time data, inputs it into the fault detection model for fault diagnosis and prediction, and outputs the fault type, fault location, fault severity, and remaining service life information; when a fault is detected, the case-based reasoning (CBR) method is used to find similar cases in the historical fault case library to provide reference for maintenance; based on the hidden Markov model (HMM), the operating status of the power product for a period of time in the future is predicted to provide early warning of potential faults; The visualization module displays the collected data, preprocessing results, feature extraction results, fault diagnosis and prediction results in the form of intuitive charts, including real-time data curves, fault distribution heat maps, and prediction trend line charts; it provides an interactive operation interface, where users can view detailed data by clicking and zooming; it supports multi-dimensional data comparison and display, which facilitates the analysis of the impact of different factors on the operation of power products; The knowledge base management module has built-in power product failure mode library, expert experience rule library and historical failure case library. The failure mode library records the characteristic manifestations of various failures, the expert experience rule library stores the failure judgment rules summarized by experts, and the historical failure case library contains detailed failure case information; establishes a knowledge update mechanism to update the knowledge base in time when new failure types are detected or new expert experience is obtained; and has knowledge query and retrieval functions to facilitate users to quickly obtain relevant knowledge; The communication module is responsible for data communication between various modules of the system and with external devices. It supports multiple communication methods such as wired communication such as Ethernet, RS-485 and wireless communication such as Wi-Fi, 4G / 5G, and Bluetooth. The communication module has data encryption function and uses AES and RSA encryption algorithms to ensure data transmission security. It has communication protocol conversion function and can be compatible with communication protocols of different devices. It also has communication status monitoring and automatic reconnection mechanism to ensure the stability of data transmission.

[0008] Furthermore, the sensors and monitoring equipment in the data acquisition module have an automatic calibration function, and the measurement accuracy is regularly calibrated according to a standard signal source; the collected data is compressed before transmission, and a lossless compression algorithm such as LZ77 and LZ78 is used to reduce the amount of data transmission without losing data accuracy.

[0009] Furthermore, when normalizing the data, the data preprocessing module selects an appropriate normalization method according to different types of power products and parameter characteristics, such as using maximum and minimum normalization for voltage and current parameters, and using mean normalization for parameters whose power factor is meaningful within a specific range; in the data integrity check, interpolation and mean filling methods are used to complete missing data based on the power product operation cycle and data correlation relationship.

[0010] Furthermore, when performing principal component analysis (PCA), the feature extraction module determines the principal components by calculating the eigenvalues ​​and contribution rates, and retains the principal components with a cumulative contribution rate of more than 85%; when using wavelet transform to extract time-frequency domain features, a suitable wavelet basis function is selected according to the signal characteristics, such as selecting the Daubechies wavelet basis for mutation signals and the Haar wavelet basis for stationary signals.

[0011] Furthermore, when using the deep learning algorithm, the fault detection model construction module adopts a transfer learning method and uses the pre-trained model to perform fine-tuning on the power field data set to speed up the model training speed; when evaluating the model, in addition to the accuracy and recall rate indicators, the F1-score and root mean square error RMSE are also used to comprehensively evaluate the model performance.

[0012] Furthermore, a method for detecting power product operation faults based on big data comprises the following steps: A data collection step, collecting power product operation data through the data collection module described in claim 1; Data preprocessing step, using a data preprocessing module to preprocess the collected data; Feature extraction step, using feature extraction module to extract features from preprocessed data; A model building step, building a fault detection model with the help of a fault detection model building module; Fault diagnosis and prediction step: using the fault diagnosis and prediction module to perform fault diagnosis and prediction based on the constructed model; Visualization display step: display relevant data and results through the visualization display module; Knowledge base management step, relying on the knowledge base management module to manage the knowledge base; The communication management step uses the communication module to manage data communication.

[0013] Furthermore, in the data collection step, the collection frequency is dynamically adjusted according to the importance and operating status of the power products, and the collection frequency is increased to 2-5 times the original for power products in a fault warning state or at a critical node.

[0014] Furthermore, in the data preprocessing step, a density-based clustering algorithm such as DBSCAN is used to detect outliers, a small number of isolated outliers are corrected, and a large number of clustered outliers are alarmed to indicate possible fault hazards; after data normalization, the normalized data is checked for rationality again to prevent new anomalies from being introduced during the normalization process.

[0015] Furthermore, in the feature extraction step, correlation analysis is first performed on the high-dimensional data to remove redundant features with high correlation, and then dimensionality reduction is performed; when extracting time-frequency domain features, the Hilbert-Huang transform (HHT) is combined to further improve the time-frequency resolution.

[0016] Furthermore, in the model building step, the fault detection model is updated and trained regularly. When the operating status of the power product changes significantly or a new fault type appears, the model update is triggered in time. In the model training process, an adaptive learning rate strategy is adopted to dynamically adjust the learning rate according to the model convergence situation to improve the training efficiency.

[0017] Compared with the prior art, the present invention has the following beneficial effects: The present invention realizes comprehensive and real-time collection of operating data of various power products through the data acquisition module. The design of various types of sensors and dynamic adjustment of acquisition frequency ensures the accuracy and effectiveness of the data, providing a rich source of information for subsequent fault detection. The data preprocessing module is used to effectively remove noise and outliers, and perform data normalization and integrity checks, which improves data quality and makes subsequent analysis more reliable. The feature extraction module uses a variety of advanced algorithms to extract key features and time-frequency domain features, and can automatically identify new fault features, enhance the ability to capture fault features, and help improve the accuracy of fault detection.

[0018] The fault detection model construction module builds models based on a variety of machine learning and deep learning algorithms, and improves model performance through integrated learning and attention mechanisms. It can more accurately diagnose the type, location and severity of faults and predict the remaining service life. The fault diagnosis and prediction module combines case-based reasoning and hidden Markov model methods, which can not only quickly locate faults, but also warn of potential faults in advance, providing strong support for power system operation and maintenance. The visualization display module displays various data and results in intuitive charts, which is convenient for operation and maintenance personnel to quickly grasp the operating status of power products and improve work efficiency. The knowledge base management module ensures the timely update and effective use of the knowledge base, and provides a rich knowledge reserve for fault detection. The communication module ensures the security, stability and efficiency of data transmission, and ensures the normal operation of the entire system. The present invention improves the overall intelligence level and efficiency of power product operation fault detection, reduces the risk of faults, and ensures the stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic block diagram of a power product operation fault detection system based on big data proposed by the present invention; Figure 2 This is a schematic block diagram of the power product operation fault detection method based on big data proposed in the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise" and "counterclockwise" indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0022] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. In addition, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, and it can be the internal connection of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0023] Reference Figure 1-2 :A power product operation fault detection system based on big data, comprising: The data acquisition module collects the operation data of power products in real time through various types of sensors and monitoring equipment. The power products include power generation equipment, power transmission equipment, power transformation equipment, power distribution equipment and power consumption equipment. The operation data covers electrical parameters such as voltage, current, resistance, power factor, frequency, equipment status parameters such as temperature, humidity, vibration amplitude, pressure, speed, insulation resistance, and environmental parameters such as ambient temperature, ambient humidity, and atmospheric pressure. The collected data has an accurate timestamp and has a data cache function. It temporarily stores data when the network is abnormal and automatically uploads it after the network is restored. The data acquisition module can also dynamically adjust the collection frequency according to the type of power product and the operating status, collect key parameters at high frequency, and collect stable parameters at low frequency. The data preprocessing module is connected to the data acquisition module and performs preprocessing operations on the collected data, including: using an outlier detection method based on statistical principles to identify and correct obviously erroneous data points, such as data that exceeds a certain multiple of the mean standard deviation is considered abnormal; using a combination of sliding average filtering and Kalman filtering to remove noise interference in the data; mapping data of different magnitudes to a specific interval through data normalization to facilitate subsequent processing; establishing data rationality inspection rules based on the operating rules of power products and expert experience, checking the integrity and logical consistency of the data, and filling in missing data; The feature extraction module uses principal component analysis (PCA) and linear discriminant analysis (LDA) and other dimensionality reduction algorithms to extract features from preprocessed data and obtain key feature vectors. It also uses wavelet transform and empirical mode decomposition (EMD) time-frequency analysis methods to extract the time-frequency domain features of the data, including frequency distribution and energy distribution. It selects features that are sensitive to faults based on the power product fault pattern library, such as features related to dissolved gas content in oil for transformer faults and zero-sequence current and zero-sequence voltage features for transmission line faults. It can also automatically identify new fault features, analyze the differences between new data and historical fault data through machine learning algorithms, and discover potential important features. The fault detection model construction module builds a fault detection model based on machine learning algorithms and deep learning algorithms. The specific steps are as follows: the extracted features are divided into training sets and test sets, of which the training set accounts for 70%-85%; the training set is trained using support vector machines (SVM), decision trees, random forest machine learning algorithms, and convolutional neural networks (CNN), recurrent neural networks (RNN) and their variants such as long short-term memory networks (LSTM) and gated recurrent units (GRU) deep learning algorithms. During the training process, the cross-validation method is used to evaluate the model performance and adjust the model hyperparameters; the attention mechanism is introduced to focus the model on key features; the results of multiple base models are integrated through ensemble learning methods such as voting and weighted average methods to improve the accuracy and generalization ability of the model; The fault diagnosis and prediction module receives the feature vector after feature extraction from real-time data, inputs it into the fault detection model for fault diagnosis and prediction, and outputs the fault type, fault location, fault severity, and remaining service life information; when a fault is detected, the case-based reasoning (CBR) method is used to find similar cases in the historical fault case library to provide reference for maintenance; based on the hidden Markov model (HMM), the operating status of the power product for a period of time in the future is predicted to provide early warning of potential faults; The visualization module displays the collected data, preprocessing results, feature extraction results, fault diagnosis and prediction results in the form of intuitive charts, including real-time data curves, fault distribution heat maps, and prediction trend line charts; it provides an interactive operation interface, where users can view detailed data by clicking and zooming; it supports multi-dimensional data comparison and display, which facilitates the analysis of the impact of different factors on the operation of power products; The knowledge base management module has built-in power product failure mode library, expert experience rule library and historical failure case library. The failure mode library records the characteristic manifestations of various failures, the expert experience rule library stores the failure judgment rules summarized by experts, and the historical failure case library contains detailed failure case information; establishes a knowledge update mechanism to update the knowledge base in time when new failure types are detected or new expert experience is obtained; and has knowledge query and retrieval functions to facilitate users to quickly obtain relevant knowledge; The communication module is responsible for data communication between various modules of the system and with external devices. It supports multiple communication methods such as wired communication such as Ethernet, RS-485 and wireless communication such as Wi-Fi, 4G / 5G, and Bluetooth. The communication module has data encryption function and uses AES and RSA encryption algorithms to ensure data transmission security. It has communication protocol conversion function and can be compatible with communication protocols of different devices. It also has communication status monitoring and automatic reconnection mechanism to ensure the stability of data transmission.

[0024] In the present invention, the sensors and monitoring equipment in the data acquisition module have an automatic calibration function, and the measurement accuracy is calibrated regularly according to a standard signal source; the collected data is compressed before transmission, and a lossless compression algorithm such as LZ77 and LZ78 is used to reduce the amount of data transmission without losing data accuracy.

[0025] In the present invention, the data acquisition module plays a key role in data acquisition. The internal sensors and monitoring equipment have a highly intelligent automatic calibration function. Every certain period, such as once a week or once a month (the specific period can be set according to the equipment operation stability and accuracy requirements), the equipment will automatically connect to the pre-set high-precision standard signal source, compare the collected standard signal data with its own measurement value, and accurately adjust the internal measurement parameters according to the comparison results to ensure that the measurement accuracy is always maintained at a high level, effectively reducing the measurement error drift caused by the long-term operation of the equipment.

[0026] At the same time, in order to improve the efficiency of data transmission, advanced lossless compression algorithms are used to process the collected data before data transmission. Taking the LZ77 algorithm as an example, it uses repeated character strings in the data to build a dictionary and represents the repeated parts in the original data with dictionary indexes, thereby greatly reducing the amount of data; the LZ78 algorithm builds a dynamic dictionary, matches the input data string with the character strings in the dictionary, and encodes and outputs the unmatched parts, which also achieves efficient compression. These lossless compression algorithms completely retain the original information of the data while reducing the amount of data transmission, ensuring that the data accuracy is not affected in any way, providing a solid data foundation for subsequent data analysis, and ensuring the accuracy of fault detection.

[0027] In the data preprocessing module of the present invention, the data normalization link is highly targeted and adaptable. For different types of power products, such as generators in power generation equipment, high-voltage lines in power transmission equipment, transformers in substation equipment, distribution cabinets in distribution equipment, and motors in power consumption equipment, as well as various parameter characteristics, there are normalization strategies adapted thereto. Taking voltage and current parameters as an example, since their numerical range varies greatly and there is no fixed interval restriction, the maximum and minimum normalization method is adopted, that is, by subtracting the minimum value from the actual value and dividing it by the difference between the maximum value and the minimum value, it is mapped to the [0,1] interval, so that data of different magnitudes can be compared and analyzed on the same scale. For parameters such as power factor that have clear meanings only within a specific interval (such as [-1,1]), mean normalization is adopted, first calculating its mean, and then subtracting the mean from each data point and dividing it by the standard deviation, so that the data is normalized around the mean, ensuring that the data is within a reasonable interval and can highlight the relative relationship between the data.

[0028] In terms of data integrity check, the characteristics of the power product operation cycle and the inherent correlation between data are fully considered. When missing data is detected, if the missing data is in the stable operation stage of the power product and the surrounding data fluctuates slightly, the mean filling method is used to supplement the missing value with the average value of the surrounding data to maintain the continuity and stability of the data; if the data missing occurs in the stage of changing the operation status of the power product, such as the startup or shutdown process, at this time the data change has a certain regularity, then the interpolation method is used to perform interpolation calculations based on the changing trends of the previous and next data points to accurately complete the missing data, laying a solid foundation for the subsequent accurate analysis of the operation status of the power product.

[0029] In the fault detection model building module of the present invention, when using deep learning algorithms, transfer learning methods play an important role. First, select models that have been pre-trained on large-scale general data sets, and these pre-trained models have learned rich feature representations. Then, apply it to the power field data set and fine-tune it for the specific task of power product operation fault detection. For example, using a convolutional neural network model trained on massive image data and migrating it to the power equipment image fault detection scenario, only the last few layers of the model need to be adjusted to adapt it to the feature classification requirements of the power field, thereby effectively reducing training time and computing resource consumption, speeding up model training, and enabling it to adapt to power product fault detection tasks more quickly.

[0030] In the model evaluation stage, in addition to the commonly used accuracy and recall indicators, F1-score and root mean square error (RMSE) are introduced in order to measure the model performance more comprehensively and accurately. F1-score takes into account the accuracy and recall rate, and can more balancedly evaluate the performance of the model in fault detection of different categories. The root mean square error (RMSE) measures the degree of deviation between the predicted value and the true value from the perspective of numerical difference, which is of great significance for evaluating the accuracy of the model in predicting the severity of power product faults and other continuous value outputs. Through these multi-dimensional evaluation indicators, the model can be optimized more effectively and the reliability of fault detection can be improved.

[0031] In the present invention, a method for detecting power product operation faults based on big data comprises the following steps: A data collection step, collecting power product operation data through the data collection module described in claim 1; Data preprocessing step, using a data preprocessing module to preprocess the collected data; Feature extraction step, using feature extraction module to extract features from preprocessed data; A model building step, building a fault detection model with the help of a fault detection model building module; Fault diagnosis and prediction step: using the fault diagnosis and prediction module to perform fault diagnosis and prediction based on the constructed model; Visualization display step: display relevant data and results through the visualization display module; Knowledge base management step, relying on the knowledge base management module to manage the knowledge base; The communication management step uses the communication module to manage data communication.

[0032] In the data collection step of the present invention, a set of intelligent dynamic adjustment mechanism for collection frequency is constructed in view of the diversity and operation complexity of power products. First, the importance of power products is graded according to their criticality in the power system. For example, the core units in power generation equipment and the hub substations in the transmission network are classified as high importance levels, while some auxiliary power equipment are at a relatively low level. At the same time, the operation status of power products is monitored in real time, and the collected multi-dimensional data such as voltage, current, temperature, etc. are analyzed in real time to determine whether they are in a fault warning state, such as detecting that the temperature of a certain device continues to rise close to the critical value, or the current fluctuations are abnormally frequent. For high-importance power products that are in a fault warning state or located at key nodes (such as main transmission lines during peak load periods of the power grid), the collection frequency is rapidly increased from the conventional once per hour to once every 12 minutes or even higher, that is, increased to 2-5 times the original. Through this dynamic adjustment, it is ensured that the detailed data of key power products at critical operating moments can be captured in time, providing more abundant and timely data support for subsequent accurate fault diagnosis and prediction.

[0033] In the data preprocessing step of the present invention, the outlier detection and processing link is particularly critical. The density-based clustering algorithm DBSCAN is adopted, and its principle is to divide the density-connected data points into a cluster by calculating the density between the data points. For those data points whose density is lower than the set threshold and exist in isolation, they are determined to be possible outliers. For a small number of isolated outliers, the system will adopt correction methods such as mean substitution and linear interpolation according to the distribution law of the normal data around them to make them consistent with the overall data trend. When a large number of data points are found to form an abnormal clustering area, this may indicate a serious hidden danger of failure. The system will immediately send an alarm signal to notify the operation and maintenance personnel to conduct a detailed inspection.

[0034] After completing the data normalization operation, in order to ensure data quality, the normalized data will be checked for rationality again. Because during the normalization process, some originally reasonable data may become abnormal due to data transformation. For example, check whether the normalized data is still within a reasonable numerical range, whether the logical relationship between the data remains consistent, etc. Through this additional check step, new abnormal situations introduced by the normalization operation are effectively avoided, providing a reliable data basis for subsequent feature extraction and fault detection.

[0035] In the feature extraction step of the present invention, a careful design is made for high-dimensional data processing. First, correlation analysis techniques are used, such as calculating the Pearson correlation coefficient between features. Features whose absolute value of the correlation coefficient is greater than a set threshold (such as 0.8) are considered to be redundant features with high correlation. In this way, features that contribute little to fault detection and have repeated information are accurately identified and removed, effectively reducing the data dimension, and reducing the amount of calculation and storage space requirements.

[0036] When performing dimensionality reduction processing, methods such as principal component analysis (PCA) can be used. PCA calculates the eigenvalues ​​and eigenvectors of the data covariance matrix, selects eigenvectors with larger contribution rates to form a new feature space, and projects the original high-dimensional data into the low-dimensional space, achieving the dual purposes of data compression and feature extraction.

[0037] When extracting time-frequency domain features, the Hilbert-Huang transform (HHT) is innovatively introduced. HHT first performs empirical mode decomposition (EMD) on the signal, decomposing the complex signal into multiple intrinsic mode functions (IMFs). Each IMF represents the local characteristics of different frequency bands. Then, each IMF is Hilbert transformed to obtain the instantaneous frequency and instantaneous amplitude, thereby constructing a time-frequency spectrum. This method can adaptively process nonlinear and non-stationary signals. Compared with traditional time-frequency analysis methods, such as short-time Fourier transform (STFT), it can more accurately characterize the time-varying characteristics of the signal, further improve the time-frequency resolution, and provide strong support for accurately capturing the operating fault characteristics of power products.

[0038] In the model building step of the present invention, a complete update mechanism is constructed to ensure the accuracy and timeliness of the fault detection model. In terms of regular update training, a weekly or monthly update cycle (which can be adjusted according to actual conditions) is set as an update cycle. In each cycle, the model is retrained using the newly collected power product operation data. This helps the model adapt to the gradual factors in the long-term operation of power products, such as performance changes caused by equipment aging.

[0039] When there are major changes in the operating status of power products, such as grid structure adjustment, access to new power equipment, or new types of faults, the system will immediately trigger a model update. These key events can be discovered in a timely manner by real-time monitoring of changes in key indicators of operating data, such as mutation patterns of voltage and current, abnormal distribution of equipment temperature, etc. Once triggered, relevant new data is quickly collected and used to perform targeted optimization training on the model.

[0040] During the model training process, an adaptive learning rate strategy is adopted. For example, a moderate learning rate (such as 0.001) is initially set. During the training process, after a certain number of iterations (such as 100 times), the rate of change of the loss function of the model on the validation set is calculated. If the loss function has a clear downward trend, the learning rate is appropriately reduced (such as multiplied by 0.9) to adjust the model parameters more finely; if the loss function tends to be stable or rising, the learning rate is increased (such as multiplied by 1.1) so that the model can jump out of the local optimal solution, thereby accelerating the training convergence speed, improving the training efficiency, and ensuring that the model always maintains good performance.

[0041] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can replace or change the technical solution and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A power product operation fault detection system based on big data, characterized in that: include: The data acquisition module collects the operation data of power products in real time through various types of sensors and monitoring equipment. The collected data is time-stamped and has a data cache function, which temporarily stores the data when the network is abnormal and automatically uploads it after the network is restored. The data acquisition module dynamically adjusts the acquisition frequency according to the type of power product and operating status, with high-frequency acquisition of key parameters and low-frequency acquisition of stable parameters; The data preprocessing module is connected to the data acquisition module and performs preprocessing operations on the collected data, including: using an outlier detection method based on statistical principles to identify and correct erroneous data points, and treating data that exceeds the mean standard deviation as abnormal; using a combination of sliding average filtering and Kalman filtering to remove noise interference in the data; mapping data of different magnitudes to specific intervals through data normalization; establishing data rationality inspection rules based on the operating rules of power products and expert experience, checking the integrity and logical consistency of the data, and filling in missing data; The feature extraction module uses principal component analysis (PCA) and linear discriminant analysis (LDA) and other dimensionality reduction algorithms to extract features from preprocessed data and obtain key feature vectors. It also uses wavelet transform and empirical mode decomposition (EMD) time-frequency analysis methods to extract the time-frequency domain features of the data, including frequency distribution and energy distribution. It selects features that are sensitive to faults based on the power product fault pattern library, including features related to dissolved gas content in oil for transformer faults and zero-sequence current and zero-sequence voltage features for transmission line faults. It automatically identifies new fault features, analyzes the differences between new data and historical fault data through machine learning algorithms, and discovers potential important features. The fault detection model construction module builds a fault detection model based on machine learning algorithms and deep learning algorithms. The specific steps are as follows: the extracted features are divided into training sets and test sets, of which the training set accounts for 70%-85%; the training set is trained using support vector machines (SVM), decision trees, random forest machine learning algorithms, and convolutional neural networks (CNN), recurrent neural networks (RNN) and their variants of long short-term memory networks (LSTM), and gated recurrent units (GRU) deep learning algorithms. During the training process, the cross-validation method is used to evaluate the model performance and adjust the model hyperparameters; the attention mechanism is introduced to focus the model on key features; the results of multiple base models are integrated through ensemble learning methods such as voting and weighted average methods; The fault diagnosis and prediction module receives the feature vector after feature extraction of real-time data, inputs it into the fault detection model for fault diagnosis and prediction, and outputs the fault type, fault location, fault severity and remaining service life information; when a fault is detected, the case-based reasoning (CBR) method is used to find similar cases in the historical fault case library to provide reference for maintenance; the operating status of power products in the future is predicted based on the hidden Markov model (HMM); Visual display module, which displays the collected data, preprocessing results, feature extraction results, fault diagnosis and prediction results in the form of intuitive charts; The knowledge base management module has built-in power product failure mode library, expert experience rule library and historical failure case library. The failure mode library records the characteristic manifestations of various failures, the expert experience rule library stores the failure judgment rules summarized by experts, and the historical failure case library contains detailed failure case information; establish a knowledge update mechanism to update the knowledge base when a new failure type is detected or new expert experience is obtained; The communication module is responsible for data communication between various modules of the system and with external devices. It supports Ethernet or RS-485 wired communication and Wi-Fi or 4G or 5G or Bluetooth wireless communication methods. The communication module has data encryption function and uses AES and RSA encryption algorithms to ensure data transmission security. It has communication protocol conversion function and is compatible with communication protocols of different devices. It also has communication status monitoring and automatic reconnection mechanism.

2. The power product operation fault detection system based on big data according to claim 1 is characterized in that: The sensors and monitoring equipment in the data acquisition module have an automatic calibration function, and the measurement accuracy is calibrated according to the standard signal source; the collected data is compressed before transmission, and lossless compression algorithms such as LZ77 and LZ78 are used to reduce the data transmission volume without losing data accuracy.

3. The power product operation fault detection system based on big data according to claim 1 is characterized in that: When normalizing data, the data preprocessing module selects an appropriate normalization method according to different types of power products and parameter characteristics, including maximum and minimum normalization for voltage and current parameters, and mean normalization for parameters with meaningful power factor within a specific range; in the data integrity check, interpolation and mean filling methods are used to complete missing data based on the power product operation cycle and data correlation relationship.

4. The power product operation fault detection system based on big data according to claim 1 is characterized in that: When performing principal component analysis (PCA), the feature extraction module determines the principal components by calculating the eigenvalues ​​and contribution rates, and retains the principal components with cumulative contribution rates of more than 85%. When using wavelet transform to extract time-frequency domain features, a suitable wavelet basis function is selected according to signal characteristics, including selecting the Daubechies wavelet basis for mutation signals and the Haar wavelet basis for stationary signals.

5. The power product operation fault detection system based on big data according to claim 1 is characterized in that: When using the deep learning algorithm, the fault detection model building module adopts the transfer learning method and uses the pre-trained model to perform fine-tuning on the power field data set to speed up the model training speed; when evaluating the model, in addition to the accuracy and recall rate indicators, the F1-score and root mean square error RMSE are also used to comprehensively evaluate the model performance.

6. A method for using the big data-based power product operation fault detection system according to any one of claims 1 to 5, characterized in that: The following steps are involved: Data collection step, collecting power product operation data through a data collection module; Data preprocessing step, using a data preprocessing module to preprocess the collected data; Feature extraction step, using feature extraction module to extract features from preprocessed data; A model building step, building a fault detection model with the help of a fault detection model building module; Fault diagnosis and prediction step: using the fault diagnosis and prediction module to perform fault diagnosis and prediction based on the constructed model; Visualization display step: display relevant data and results through the visualization display module; Knowledge base management step, relying on the knowledge base management module to manage the knowledge base; The communication management step uses the communication module to manage data communication.

7. The method for detecting power product operation faults based on big data according to claim 6 is characterized in that: In the data collection step, the collection frequency is dynamically adjusted according to the importance and operating status of the power products. For power products in a fault warning state or at key nodes, the collection frequency is increased to 2-5 times the original.

8. According to the big data-based power product operation fault detection method described in claim 6, in the data preprocessing step, a density-based clustering algorithm such as DBSCAN is used to detect outliers, a small number of isolated outliers are corrected, and a large number of clustered outliers are alarmed to indicate possible fault hazards; after data normalization, the normalized data is checked for rationality again to prevent new anomalies from being introduced during the normalization process.

9. According to the big data-based power product operation fault detection method of claim 6, in the feature extraction step, correlation analysis is first performed on the high-dimensional data to remove redundant features with high correlation, and then dimensionality reduction processing is performed; when extracting time-frequency domain features, the Hilbert-Huang transform HHT is combined to further improve the time-frequency resolution.

10. According to the big data-based power product operation fault detection method of claim 6, in the model construction step, the fault detection model is updated and trained regularly, and when the operating status of the power product changes significantly or a new fault type appears, the model update is triggered in time; in the model training process, an adaptive learning rate strategy is adopted to dynamically adjust the learning rate according to the model convergence situation to improve the training efficiency.

Images