A signal fault detection method for a single-phase rectifier

Abstract

The present application relates to the field of signal processing technology, and provides a signal fault detection method for a single-phase rectifier. The method comprises: reading the vital sign data of the target single-phase rectifier; building a multidimensional constraint space based on the data, and retrieving the rectifier fault record; mining and analyzing the fault record, clustering and generating a fault waveform decision tree; building a signal fault identification channel based on the decision tree, with a built-in similarity comparator; collecting the output signal waveform, transferring it to the identification channel for comparison, and outputting the fault type. The present application solves the technical problem that the traditional method is difficult to locate the fault, and frequent misjudgments and missed detections occur due to low detection efficiency and insufficient accuracy when facing complex electrical characteristics, rectification characteristics, and scene characteristics, thereby extending the maintenance time and increasing the waste of maintenance resources. The method improves the efficiency and accuracy of signal fault detection, quickly locates the fault point, reduces misjudgments and missed detections, saves maintenance time and maintenance resources, and achieves the effect of ensuring the stable operation of the equipment.

Description

Technical Field

[0001] The present application relates to the field of electronic engineering technology, specifically to the field of signal processing technology, and in particular to a signal fault detection method for a single-phase rectifier. Background Art

[0002] With the rapid development and widespread application of power electronics technology, single-phase rectifiers play an increasingly important role in power systems. They not only involve efficient energy conversion, but are also directly related to the stable operation of multiple key areas such as motor drive and communication power supply. However, during the operation of single-phase rectifiers, they are affected by multiple factors such as electrical characteristics, rectification characteristics, and changing scene characteristics, and their signal fault detection has become a complex and critical technical issue. Traditional single-phase rectifier signal fault detection methods often seem to be powerless when faced with these complex characteristics. Due to low detection efficiency and insufficient accuracy, it is difficult for traditional methods to accurately and quickly locate the fault point, and misjudgment and missed detection often occur. This not only prolongs the maintenance time and increases the waste of maintenance resources, but also may pose a potential threat to the stable operation of the power system. Summary of the invention

[0003] The present application provides a signal fault detection method for a single-phase rectifier, aiming to solve the technical problem that when faced with complex electrical characteristics, rectification characteristics and scenario characteristics, traditional methods have low detection efficiency and insufficient accuracy, resulting in difficulty in fault location, frequent misjudgments and missed detections, and thus prolonged maintenance time and increased waste of maintenance resources.

[0004] In view of the above problems, the present application provides a signal fault detection method for a single-phase rectifier.

[0005] The present application provides a signal fault detection method for a single-phase rectifier, the method comprising: reading vital sign data of a target single-phase rectifier, wherein the vital sign data comprises electrical characteristics, rectification characteristics and scenario characteristics; constructing a multidimensional constraint space according to the electrical characteristics, rectification characteristics and scenario characteristics, and retrieving rectification fault records with the multidimensional constraint space as a restriction condition; performing mining and analysis of the rectification fault records, and clustering and generating a fault waveform decision tree; constructing a signal fault identification channel based on the fault waveform decision tree, wherein a similarity comparator is embedded in the signal fault identification channel; collecting an output signal waveform, and transferring the output signal waveform to the signal fault identification channel for similarity comparison, and outputting a fault identification type based on the comparison result.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The above-mentioned signal fault detection method for a single-phase rectifier reads the physical data of the single-phase rectifier, which covers multiple aspects such as electrical characteristics, rectification characteristics, and scene characteristics. Subsequently, a multidimensional constraint space is constructed using these data, and the rectification fault record is quickly retrieved through this space. After obtaining the fault record, an in-depth analysis is performed, and the common characteristics of the fault waveform are identified by mining the fault data, and then a fault waveform decision tree is generated by clustering. Based on this decision tree, a signal fault identification channel is established. This channel is embedded with a similarity comparator. When the output signal waveform of the single-phase rectifier is collected, it will be input into this identification channel for comparison. By comparison, it is determined whether the signal waveform is similar to the known fault waveform, so as to output the corresponding fault identification type. This method realizes efficient and accurate detection of single-phase rectifier signal faults through the construction of a multidimensional constraint space, mining and analysis of fault waveforms, and establishment of an intelligent identification channel, greatly improving the accuracy and efficiency of fault location, helping to reduce the waste of maintenance time and resources, and ensuring the stable operation of equipment.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 It is a schematic flow chart of a signal fault detection method for a single-phase rectifier in one embodiment;

[0011] Figure 2 A schematic diagram of obtaining rectification fault records of a signal fault detection method for a single-phase rectifier in an embodiment. DETAILED DESCRIPTION

[0012] The embodiment of the present application provides a signal fault detection method for a single-phase rectifier to solve the technical problem that when faced with complex electrical characteristics, rectification characteristics and scenario characteristics, traditional methods have low detection efficiency and insufficient accuracy, resulting in difficulty in fault location, frequent misjudgments and missed detections, and thus prolonged maintenance time and increased waste of maintenance resources.

[0013] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0014] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0015] Embodiment 1

[0016] like Figure 1 , Figure 2 As shown, the present application provides a signal fault detection method for a single-phase rectifier, the method comprising:

[0017] Reading vital sign data of a target single-phase rectifier, wherein the vital sign data includes electrical characteristics, rectification characteristics, and scene characteristics;

[0018] With the rapid development of science and technology and the continuous expansion of industrial scale, single-phase rectifiers have been widely used in many fields such as electricity, communication, and transportation. As a key device for power conversion and control, the stable operation of single-phase rectifiers is crucial to the reliability and safety of the entire system.

[0019] In an embodiment of the present application, during the signal fault detection process of a single-phase rectifier, the system terminal first collects detailed vital sign data about the rectifier. These data are targeted, and they cover the electrical characteristics, rectification characteristics, and specific scene characteristics of the rectifier when it is in operation. Electrical characteristics involve changes in electrical parameters such as current and voltage, which directly reflect the working state of the rectifier. Rectification characteristics are more related to the efficiency and stability of the rectifier in converting electrical energy, such as rectification efficiency, waveform distortion, etc. The scene characteristics refer to the environment and working conditions of the rectifier, such as temperature, load changes, etc. Through the comprehensive collection and in-depth analysis of these vital sign data, the operating status of the single-phase rectifier can be more accurately understood, so as to promptly discover and solve potential signal fault problems. This not only helps to ensure the stable operation of the rectifier, but also improves the reliability and safety of the entire system.

[0020] Building a multidimensional constraint space according to the electrical characteristics, rectification characteristics and scenario characteristics, and retrieving rectification fault records with the multidimensional constraint space as a restriction condition;

[0021] In one embodiment, during the signal fault detection process of the single-phase rectifier, the system terminal constructs a multidimensional constraint space based on the collected electrical characteristics, rectification characteristics and scenario characteristic data. This space is a customized search framework, which uses this multidimensional constraint space as a constraint condition to help the system terminal accurately retrieve the rectification fault records that match these characteristics. By building a multidimensional constraint space, fault cases similar to the current rectifier operating conditions can be located more quickly and accurately. These fault records not only contain the specific conditions for the occurrence of the fault, but when a new fault occurs, these historical records can be referred to to quickly find the root cause of the problem and take appropriate measures to repair it. In general, the construction of a multidimensional constraint space provides an efficient and accurate fault retrieval tool for the system terminal, which makes the signal fault detection of the single-phase rectifier more intelligent and efficient.

[0022] Furthermore, the present application provides a method for retrieving and acquiring rectification fault records using the multidimensional constraint space as a restriction condition, and the method further includes:

[0023] The electrical characteristics include voltage characteristics and current characteristics, the rectification characteristics include rectification type, rectification efficiency and filtering characteristics, and the scenario characteristics include source equipment, electromagnetic interference characteristics and load characteristics;

[0024] Building an eight-dimensional constraint space based on the voltage characteristics, current characteristics, rectification type, rectification efficiency, filtering characteristics, source equipment, electromagnetic interference characteristics, and load characteristics;

[0025] Preferably, in the signal fault detection of the single-phase rectifier, the system terminal comprehensively considers multiple key characteristics to build an accurate fault analysis framework. These characteristics include voltage and current characteristics in electrical characteristics, rectification type, rectification efficiency and filtering characteristics in rectification characteristics, and source equipment, electromagnetic interference and load characteristics in scenario characteristics.

[0026] After obtaining these characteristics, the system terminal conducts a detailed analysis of each feature to understand its physical meaning, value range, and impact on rectifier performance, and determine the importance and sensitivity of each feature in fault detection. Subsequently, appropriate constraints are set for each dimension based on engineering experience and expertise. These conditions include value ranges (such as upper and lower limits of voltage and current), type classification (such as rectification type), efficiency threshold, filtering effect requirements, etc. The setting of conditions must ensure that they can cover various operating scenarios and failure modes that the rectifier may encounter. After that, the voltage characteristics, current characteristics, rectification type, rectification efficiency, filtering characteristics, source equipment, electromagnetic interference characteristics, and load characteristics are used as eight dimensions to construct an eight-dimensional space model. In this space model, each dimension represents a feature, and the points in the space represent the rectifier state under a specific feature combination. Then, the set constraints are input into the eight-dimensional space model, the points in the model are constrained, and the points that do not meet the conditions are eliminated, thereby constructing a limited and accurate eight-dimensional constraint space. Through this space, the system terminal can more accurately locate the feature combination that matches the specific fault mode, so as to quickly retrieve the relevant fault records. This multi-dimensional analysis method not only improves the accuracy of fault detection, but also enables the system terminal to more comprehensively understand the performance of the single-phase rectifier under different operating conditions. This provides strong support for subsequent fault analysis and repair work, and helps to achieve stable operation and fault prevention of single-phase rectifiers.

[0027] Based on the electric power big data, data retrieval is performed with the eight-dimensional constraint space as a restriction condition to obtain the rectification fault record.

[0028] Preferably, the system terminal uses the previously constructed eight-dimensional constraint space as a constraint condition based on the power big data to perform data retrieval. The system terminal quickly obtains a large amount of diversified power operation data through the power big data platform. Then, these data are input into the eight-dimensional constraint space. The eight-dimensional constraint space performs a preliminary screening of the input data according to the internally formulated constraint conditions, that is, the associated scale evaluation function, and then uses the preset associated scale threshold to extract the results of the preliminary screening to construct a rectifier fault record. These fault records contain rich fault information and data, which can help to more deeply understand the causes, characteristics and laws of the faults. In this way, the fault problems of the rectifier can be located and solved more efficiently and accurately, and the stability and reliability of the power system can be improved.

[0029] Furthermore, the present application provides a method for obtaining the rectification fault record, and the method further includes:

[0030] Configure a predetermined correlation scale threshold;

[0031] Evaluating the search data based on the correlation scale evaluation function, and extracting the search data that meets the predetermined correlation scale threshold to construct the rectification fault record;

[0032] Optionally, in the process of obtaining rectifier fault records, the system terminal first configures a predetermined correlation scale threshold. This threshold is set based on historical experience and actual needs, and is used to measure the degree of correlation between the retrieved data and the rectifier fault. Subsequently, the system terminal evaluates the retrieved data using a pre-built correlation scale evaluation function. This function calculates the correlation with the rectifier fault based on multiple dimensions and features of the data. Through this function, the correlation of the data can be quantified, and data that is highly correlated with the fault can be screened out. Finally, the retrieved data that meets the predetermined correlation scale threshold is extracted, and these data are constructed into rectifier fault records. These records are an important basis for further analyzing and solving rectifier faults. By configuring the threshold and using the evaluation function, data related to the rectifier fault can be obtained more accurately, improving the accuracy and efficiency of fault detection.

[0033] The association scale evaluation function is:

[0034] ;

[0035] Wherein, C represents the correlation scale between the retrieval data and the eight-dimensional constraint space, For the The weight coefficient of the attribute is set based on the influence ratio. The first of the eight dimensional features of the retrieved data Attribute feature value, is the first Attribute characteristic value.

[0036] Optional, pre-built correlation scale evaluation functions are as follows:

[0037] ;

[0038] Among them, C in the function is the result value used to characterize the correlation scale between the retrieval data and the eight-dimensional constraint space. It represents the similarity or correlation between the two. Indicates The weight coefficient of the attribute. The weight coefficient is set according to the importance of each attribute in the evaluation, and is determined based on the influence ratio. Different attributes may contribute differently to the association scale, so this difference is reflected through the weight coefficient. Represents the first of the eight dimensional features of the retrieved data The characteristic values ​​of the attributes. These characteristic values ​​are extracted from the retrieved data and represent the specific characteristics of the data. is the first Attribute characteristic value. The constraint space is set according to known conditions or rules, and the characteristic value represents the attribute value associated with a specific condition or rule. When calculating the association scale, the system terminal calculates the attribute value associated with each attribute. , calculate the eigenvalues ​​of the retrieved data and the eigenvalues ​​of the eight-dimensional constraint space The difference between the two. This difference reflects the degree of deviation of the retrieved data from the constraint space on this attribute. Then, find the minimum and maximum differences among all attributes. This step is to determine the range of the difference so that it can be normalized later. After that, according to each attribute The weight coefficient , and weight the corresponding differences. The weight coefficient reflects the importance of different attributes in the association scale evaluation, so the weighted processing can more accurately reflect the degree of association between the retrieval data and the constraint space. Then, the system terminal sums the weighted differences of all attributes to obtain the final association scale C. This value combines the contribution of all attributes to the association scale and can fully reflect the degree of association between the retrieval data and the eight-dimensional constraint space. Through this process, the system terminal uses the association scale evaluation function to quantify the degree of association between the retrieval data and the constraint space, thereby screening out data that is highly relevant to specific conditions or rules, providing strong support for subsequent fault detection and analysis.

[0039] Performing mining analysis on the rectifier fault records and clustering to generate a fault waveform decision tree;

[0040] In one embodiment, when performing mining analysis of rectifier fault records, the system terminal first performs in-depth data mining on the obtained fault records based on the predetermined fault type, extracts the key features and patterns in the fault records, and constructs several fault data sets. These features and patterns help to understand the causes, processes and impacts of the faults. Subsequently, cluster analysis is performed on several fault data sets in turn, and similar fault records are classified into one category according to the similarities and differences between the data. In this way, the system terminal can divide a large number of fault records into several representative fault categories, which is convenient for subsequent analysis and processing. On the basis of clustering, the system terminal further constructs a fault waveform decision tree. A decision tree is a classification model with a tree structure, which can classify layer by layer according to the characteristics and attribute values ​​of the data, and finally obtain the classification results of the fault waveform. By constructing a decision tree, complex fault waveform data can be simplified into a set of rules that are easy to understand and explain, thereby facilitating the diagnosis of faults. In summary, by mining and analyzing rectifier fault records, and using clustering algorithms and decision tree technology, a large amount of fault data is converted into valuable fault waveform classification information. This information not only helps to deeply understand the nature and laws of faults, but also provides an important reference for subsequent fault prevention and processing.

[0041] Furthermore, the present application provides a method for performing mining analysis of the rectifier fault records and clustering to generate a fault waveform decision tree, and the method also includes:

[0042] Performing data mining on the rectifier fault records based on predetermined fault types to obtain a plurality of fault data sets;

[0043] Preferably, the system terminal performs in-depth data mining on the rectifier fault records based on the preset fault types. Among them, the preset fault types are obtained by clustering based on the K-means algorithm, including VD1 fault, VD2 fault, filter fault and resistor fault, etc. Specifically, the system terminal has an in-depth understanding of each fault type in the predetermined fault type, and grasps its characteristics, causes and manifestation patterns. Subsequently, according to the characteristics of the predetermined fault type, relevant feature data is extracted from the rectifier fault record. The extracted features are then screened to remove features that are not highly correlated with the fault type, and key features are retained. Afterwards, the rectifier fault records are traversed and analyzed, and the feature data of each fault record is compared with the predetermined fault type to determine its category. Then, according to the comparison results, the fault records are classified into corresponding fault data sets to ensure that each fault data set only contains records related to its corresponding fault type. Finally, the constructed fault data set is verified to ensure the accuracy and completeness of the data. If there is no obvious abnormality in the verification result, the obtained several fault data sets are output for subsequent analysis and application. These data sets can not only help system terminals more clearly understand the characteristics and laws of each fault type, but also provide strong data support for subsequent fault analysis and processing. In this way, fault problems in rectifiers can be located and solved more accurately, improving the stability and reliability of the power system.

[0044] Performing cluster analysis on the plurality of fault data sets in sequence, and performing waveform contour enhancement on the cluster analysis results to obtain a plurality of standard waveform contour images, wherein the standard waveform contour images correspond to the fault types one by one;

[0045] Preferably, the system terminal performs cluster analysis on each fault data set. The purpose of this step is to cluster similar fault waveform images together, thereby revealing the main waveform characteristics of each fault type. The result of the cluster analysis is a series of high-frequency fault waveform images, which represent the most common waveform patterns of each fault type. Specifically, the system terminal applies the K-means clustering algorithm to each fault data set to determine the appropriate K value, that is, the number of clusters to be divided, and this K value is determined based on the professional knowledge and historical experience of the fault type. Subsequently, the centroids of the K clusters are initialized, each data point is iteratively assigned to the nearest cluster, and the centroid of each cluster is recalculated until the convergence condition is reached. After that, each fault data set will be divided into K clusters, each cluster representing a specific fault mode or manifestation. In each cluster obtained by clustering, the frequency of occurrence of the waveform image in the cluster is counted. The waveform image with the highest frequency of occurrence is selected as the representative of the cluster, that is, the high-frequency fault waveform image, which can represent the typical characteristics of the fault mode represented by the cluster. Further, the waveform contour of the selected high-frequency fault waveform image is enhanced. The system terminal first applies the Canny edge detection algorithm to extract the edge information of the fault waveform image. The Canny algorithm is a multi-stage algorithm, including noise suppression, calculation of gradient strength and direction, non-maximum suppression and double threshold processing. According to the characteristics and requirements of the fault waveform image, the parameters of the Canny algorithm, such as the standard deviation of the Gaussian filter, high and low thresholds, etc., are adjusted to optimize the effect of edge detection. Subsequently, the system terminal obtains the image processed by the Canny algorithm, which can highlight the edge contour of the waveform. After that, the image after Canny edge detection is filtered. During the filtering process, the system terminal adjusts the parameters of the filter according to the actual situation to further smooth the image and reduce noise. Then, according to the grayscale distribution of the image and the strength of the edge information, a suitable contrast enhancement method is selected, and the relevant parameters are adjusted to achieve the best effect. For example, histogram equalization, adaptive contrast enhancement technology, etc. Finally, the enhanced waveform contour image is evaluated to observe the accuracy of edge detection and the clarity of the contour. After the waveform contour enhancement process, the representative waveform image of each cluster is used as the standard waveform contour image of the fault type and output to obtain several standard waveform contour images. These standard waveform contour images correspond to the fault types one by one and can be used for subsequent fault identification and diagnosis. In general, by clustering the fault data set and performing waveform contour enhancement, the standard waveform contour images of each fault type were successfully extracted, providing strong support for subsequent fault identification and diagnosis.

[0046] Based on the decision tree principle, the fault waveform decision tree is constructed with the fault type as the main node and the standard waveform contour image as the subsidiary node.

[0047] Preferably, the system terminal constructs a decision tree for fault waveform analysis based on the decision tree principle. In this decision tree, the fault type is set as the main node, that is, the root node of the tree, which represents the main target to be identified and classified. Each specific fault type corresponds to one or more branches, which are divided according to different fault characteristics. The standard waveform contour image is used as a subsidiary node, that is, the leaf node of the decision tree. These leaf nodes store the waveform contour image information corresponding to the main node (fault type), which provides intuitive and specific fault manifestations, and helps to quickly understand and identify different fault types. Specifically, in the process of building this decision tree, the system terminal first determines all possible fault types based on the fault data set, and uses each fault type as a main node (root node or intermediate node) of the decision tree. Subsequently, the standard waveform contour image is analyzed to extract features that can distinguish different fault types. These features include the amplitude, frequency, phase, shape, etc. of the waveform. After that, starting from the root node, the data set is divided into different subsets according to the extracted features. For each subset, the best feature is selected as the division criterion to generate a new child node. The above process is performed recursively until the stopping condition is met, that is, all samples in the current node belong to the same category. When the recursive process reaches the stopping condition, the current node is set as a leaf node. The standard waveform contour image of the fault type corresponding to the current node is used as the auxiliary information of the leaf node. Then, the constructed decision tree is optimized, such as by pruning to avoid overfitting. Once the decision tree is built, the system terminal can use it to build a signal fault identification channel to facilitate the identification and analysis of new fault waveform images. By comparing and matching the new waveform image with the standard waveform contour image in the decision tree, the fault type corresponding to the waveform image can be quickly determined, thereby providing strong support for fault diagnosis and processing.

[0048] Furthermore, the present application provides setting a predetermined fault type, and the method further includes:

[0049] Performing cluster analysis on the rectifier fault records based on a K-means algorithm to obtain multiple rectifier fault data sets;

[0050] performing fault type identification and frequency statistics on the multiple rectifier fault data sets in sequence to generate a fault type sequence;

[0051] The fault type that meets the preset ranking in the fault type sequence is extracted and set as the predetermined fault type.

[0052] Optionally, the system terminal uses a similar method to the above-mentioned method of obtaining high-frequency fault waveform images, based on the K-means algorithm, to perform cluster analysis on the rectifier fault records, cluster similar fault data together, and form multiple rectifier fault data sets. Subsequently, the fault type is identified and the frequency is counted for each data set in turn, so that a fault type sequence can be generated, which reflects the frequency of occurrence of different fault types in the data set. After that, the fault types that meet the preset ranking in this sequence are extracted. These types are considered to be predetermined fault types, that is, the fault types that are focused on and processed, such as VD1 fault, VD2 fault, filter fault, and resistor fault. Through such a process, rectifier faults can be effectively classified and identified, providing strong support for subsequent processing and analysis.

[0053] Based on the fault waveform decision tree, a signal fault identification channel is constructed, wherein the signal fault identification channel is embedded with a similarity comparator;

[0054] In one embodiment, the system terminal uses the previously constructed fault waveform decision tree as the core foundation and uses a convolutional neural network to build a signal fault identification channel, in which the fault waveform decision tree is embedded in the signal fault identification channel. This channel is specifically used to identify the type of fault in the signal. Inside the channel, there is also an embedded similarity comparator. The function of this comparator is to compare the input fault waveform image with the standard waveform contour image stored in the decision tree for similarity. Through comparison, it is possible to quickly determine which standard waveform the input waveform is most similar to, thereby determining the corresponding fault type. This method not only improves the accuracy of fault identification, but also greatly speeds up the identification speed, making signal fault identification more efficient and reliable.

[0055] Furthermore, the present application provides a method for building a signal fault identification channel based on the fault waveform decision tree, and the method also includes:

[0056] Building a signal fault identification channel based on a convolutional neural network, and embedding the fault waveform decision tree in the signal fault identification channel;

[0057] The signal fault identification channel includes an input layer, a similarity calculation layer, a comparison and determination layer and an output layer, wherein the similarity calculation layer is embedded with a similarity comparator, and the comparison and determination layer includes a predetermined similarity constraint and a predetermined judgment logic;

[0058] Preferably, the system terminal constructs a signal fault identification channel based on a convolutional neural network, and embeds the previously created fault waveform decision tree into it. This identification channel has multiple levels, including an input layer, a similarity calculation layer, a comparison judgment layer, and an output layer. Specifically, the system terminal designs the architecture of the convolutional neural network, including an input layer, a convolution layer, an output layer, etc., and determines the parameters of each layer, such as the number and size of convolution kernels, step size, pooling method, etc. ReLU is then used as an activation function. The ReLU function can keep the gradient unchanged when the input is positive, which helps to accelerate the training process of the neural network. Subsequently, the cross entropy loss function is selected as a criterion for measuring the difference between the model output and the true label. The cross entropy loss function is applicable to multi-classification problems and helps the convolutional neural network to better learn and classify fault signals. Afterwards, the system terminal uses the historical fault signal data set to train the convolutional neural network, and uses the back propagation algorithm and the gradient descent optimizer to update the weights and biases of the network. Then iterate the training until the set number of training rounds is completed. After completing the training of the convolutional neural network, the system terminal introduces a similarity calculation layer after the output layer of the convolutional neural network. In the similarity calculation layer, a similarity comparator is embedded to calculate the similarity between the input signal and the standard waveform profile in the fault waveform decision tree. The similarity comparator determines the similarity between the input signal and the standard waveform by comparing the waveform features. Then, after the similarity calculation layer, a comparison decision layer is set. In the comparison decision layer, a predetermined similarity constraint and a predetermined judgment logic are defined. The predetermined similarity constraint is used to determine the maximum similarity threshold between the input signal and the standard waveform profile, and the predetermined judgment logic determines the fault type to which the input signal that does not meet the predetermined similarity constraint belongs based on the similarity result. Finally, the system terminal integrates the convolutional neural network, the similarity calculation layer, and the comparison decision layer into a complete signal fault identification channel, and determines the input and output formats of the channel to ensure that the fault signal to be identified can be received and the corresponding fault type can be output. In this way, the system terminal not only utilizes the powerful feature extraction and classification capabilities of the convolutional neural network, but also combines the prior knowledge of the fault waveform decision tree, making signal fault identification more accurate and efficient.

[0059] The predetermined judgment logic is that if there is a similarity calculation result that satisfies the predetermined similarity constraint, the fault type corresponding to the maximum similarity is output; if there is no similarity calculation result that satisfies the predetermined similarity constraint, the expert judgment channel is activated to perform fault type judgment, the expert judgment result is output, and the signal fault identification channel is incrementally learned based on the input image and the expert judgment result.

[0060] Preferably, comparing the predetermined judgment logic in the judgment layer is a complex process, and the fault type of the corresponding situation is obtained by comparing the similarity calculation results with the predetermined similarity constraints. Specifically, if any of the similarity calculation results matches the predetermined similarity constraints, the system terminal selects the result with the highest similarity and uses the corresponding fault type as the final output. However, if no result meets the predetermined similarity constraints, this means that the system terminal has encountered a complex situation that is difficult to automatically identify. At this time, the expert judgment channel will be activated to determine the fault type, and the expert's judgment result will be used as the output. At the same time, in order to continuously improve the accuracy and adaptability of the signal fault identification channel, the system terminal annotates the input image and uses the fault type determined by the expert as a label. Subsequently, the annotated input image and its fault type label are organized into an incremental learning data set. Afterwards, the incremental learning data set is used to fine-tune the convolutional neural network in the signal fault identification channel. During the fine-tuning process, the parameters of some network layers can be fixed, and only some layers are trained to accelerate the learning process and maintain the stability of the model. Then, according to the size and complexity of the incremental learning data set, a suitable learning rate and number of iterations are selected for training. After fine-tuning is completed, the convolutional neural network is verified using the validation data set to evaluate the effect of incremental learning. By comparing the performance indicators of the model on the validation data set, such as accuracy and recall, it is determined whether incremental learning has effectively improved the performance of the convolutional neural network. If incremental learning significantly improves the performance of the model, and new fault types frequently appear in the incremental learning data set, the fault waveform decision tree is updated, and the new fault type and its corresponding waveform profile are added to the decision tree to enrich the coverage of the decision tree. In this way, the signal fault identification channel can not only ensure the rapid and accurate identification of the fault type in most cases, but also make judgments with the help of experts when encountering complex situations, and continuously improve its own performance through learning.

[0061] Furthermore, the present application provides a method for generating the similarity comparer, the method further comprising:

[0062] Building a feature extraction unit based on predetermined waveform comparison indicators, wherein the predetermined waveform comparison indicators include peak value, valley value, period, amplitude, and phase;

[0063] A similarity comparison unit is built based on the predetermined waveform comparison index and the Euclidean decision distance, and the similarity comparator is generated in combination with the feature extraction unit.

[0064] Optionally, the feature extraction unit is constructed based on a predetermined waveform comparison index. The predetermined waveform comparison index includes peak value, valley value, period, amplitude, phase, etc. Specifically, the peak value refers to the highest point in the signal waveform, while the valley value is the lowest point. In order to construct the feature extraction unit, the system terminal first designs a peak and valley value extraction algorithm. This algorithm traverses the entire waveform of the signal, finds all local maximum values ​​(peak values) and local minimum values ​​(valley values), and extracts these values, which can reflect the extreme value of the signal and help understand the range and dynamic characteristics of the signal. Subsequently, a period extraction algorithm is designed. This algorithm estimates the period of the signal by calculating the time difference between consecutive peaks or valley values. This period is the time interval for the signal waveform to recur. In order to improve accuracy, the period extraction algorithm also calculates the average value of multiple periods to smooth the result and avoid accidental errors. After that, an amplitude extraction algorithm is designed. This algorithm calculates the amplitude by measuring the difference between the peak value and the valley value. The amplitude describes the amplitude of the signal and reflects the energy intensity of the signal. Then, a phase extraction algorithm is designed. This decomposes the entire waveform into components of different frequencies through Fourier transform, and then extracts the phase information of each component. Phase information is very important for signal synthesis, demodulation, and analysis of relative delay of signals. It describes the relative position relationship between the various frequency components of the signal. Finally, the system terminal integrates the peak, valley, period, amplitude, and phase extraction algorithms designed above to construct a unified feature extraction unit. The main task of this feature extraction unit is to extract key waveform features from the input signal for subsequent comparison with the standard waveform.

[0065] When building the feature extraction unit, the system terminal also built a similarity comparison unit at the same time. Specifically, the Euclidean distance is a method for measuring the similarity between two data points, which is based on the Euclidean distance calculation in the feature space. In the similarity comparison unit, the system terminal defines the calculation formula of the Euclidean distance, which is used to compare the similarity between the feature vectors of two signal waveforms. Subsequently, an algorithm is designed that can receive the feature vectors of two signal waveforms as input and calculate the distance between the two feature vectors using the Euclidean distance formula. Then, based on the calculated distance value, the similarity between the two signal waveforms is judged. The smaller the distance, the higher the similarity. After that, a similarity threshold is set according to the needs of the actual application. When the Euclidean distance of the two signal waveforms is less than this threshold, they are considered similar; otherwise, they are considered dissimilar. Then, the above-constructed algorithm and the calculation formula of the Euclidean distance are integrated into an independent similarity comparison unit. Finally, the system terminal connects the feature extraction unit and the similarity comparison unit to construct a complete similarity comparer. The similarity comparator extracts feature vectors from the input signal waveform through the feature extraction unit, and then passes the extracted feature vectors to the similarity comparison unit, which calculates the Euclidean decision distance and outputs the similarity judgment result. In summary, the feature extraction unit is responsible for extracting key waveform features from the input signal, and the similarity comparison unit uses these features and the Euclidean decision distance to construct a similarity comparator, thereby realizing the comparison and judgment of the similarity between the input signal and the standard waveform.

[0066] The output signal waveform is collected, and the output signal waveform is transferred to the signal fault identification channel for similarity comparison, and the fault identification type is output based on the comparison result.

[0067] In one embodiment, after the system terminal collects the output signal waveform, it transfers this waveform to the signal fault identification channel through the data transmission interface. The signal fault identification channel receives this data and stores it in an internal buffer for subsequent processing and analysis. Subsequently, the signal fault identification channel performs a similarity comparison on the output signal waveform that flows to the internal part. During the comparison process, the pre-set waveform comparison index and Euclidean judgment distance are used to determine the similarity between the output signal waveform and the known fault waveform. Based on the comparison results, the corresponding fault identification type is output to help users quickly locate and solve the fault problem. In summary, by comparing the output signal waveform with the waveform in the fault waveform library, the fault type can be accurately identified, improving the efficiency and accuracy of fault handling.

[0068] Furthermore, the present application provides that before transferring the output signal waveform to the signal fault identification channel for similarity comparison, the method further includes:

[0069] Retrieve a sample input waveform set and a sample output waveform set based on the eight-dimensional constraint space;

[0070] Performing deviation analysis on the sample input waveform set according to the standard input waveform to obtain an input waveform deviation set;

[0071] Performing deviation analysis on the sample output waveform set according to the standard output waveform to obtain an output waveform deviation set;

[0072] Optionally, the system terminal retrieves a sample input waveform set and a sample output waveform set based on the eight-dimensional constraint space. These two sets contain input and output waveform samples under specific constraint conditions, respectively. Subsequently, the sample input waveform set is subjected to deviation analysis using the standard input waveform as a reference. This process is to identify and quantify the degree of deviation of the sample input waveform from the standard by comparing the difference between the standard input waveform and the sample input waveform. Through this analysis, an input waveform deviation set is obtained, which records the deviation information between the sample input waveform and the standard input waveform. At the same time, the system terminal also obtains the output waveform deviation set using the same method. These deviation information helps to understand the variation and potential problems of the sample input waveform and the sample output waveform, and provides an important basis for subsequent analysis and processing.

[0073] Training a waveform deviation analysis model based on the input waveform deviation set and the output waveform deviation set to obtain a converged waveform deviation analysis model that meets expected constraints;

[0074] Optionally, the system terminal trains a waveform deviation analysis model based on the input waveform deviation set and the output waveform deviation set. This model gradually adjusts its own parameters and structure by learning the relationship between the input and output waveform deviations to optimize the prediction and analysis capabilities of the waveform deviation. Specifically, the system terminal selects a convolutional neural network as the waveform deviation analysis model because the convolutional neural network can effectively extract local features and spatial relationships when processing waveform data. The input layer of the model is then designed to match the dimension of the input waveform deviation set, the output layer is designed to match the dimension of the output waveform deviation set, and the number of hidden layers and the number of nodes in each layer are determined according to the complexity of the problem and the characteristics of the data. Subsequently, the weights and biases of the model are assigned using a random initialization method, and appropriate training parameters such as a learning rate, batch size, and number of iterations are set. Afterwards, the input waveform deviation set is divided into a training set and a validation set, and the training set is used for model training, and the validation set is used to monitor the performance of the model. In each training iteration, a batch of data is randomly extracted from the training set as input, and the predicted output of the model is calculated by forward propagation. Calculate the loss function value between the predicted output and the corresponding output waveform, and calculate the gradient of the loss function to the model parameters through the back-propagation algorithm. Then, use the optimization algorithm to update the parameters of the model according to the gradient to minimize the loss function value. Repeat the above steps until the preset number of iterations is reached. When the training is completed, the system terminal uses the validation set to evaluate the performance of the model and calculates the accuracy, recall and other indicators of the model. And adjust the parameters and structure of the model according to the validation results, such as increasing the number of hidden layers, changing the number of nodes, adjusting the learning rate, etc. When the performance of the model on the validation set is stable and meets the expected constraints, the system terminal determines that the model has converged and outputs the trained waveform deviation analysis model. This model is used for subsequent tasks such as waveform deviation identification and compensation to improve the accuracy and efficiency of waveform processing and analysis.

[0075] The input waveform deviation is collected and transferred to the converged waveform deviation analysis model to obtain the output waveform deviation, and the output signal waveform is compensated based on the output waveform deviation.

[0076] Optionally, the system terminal collects input waveform deviation data and uses the same method to transfer the input waveform deviation data to the trained convergent waveform deviation analysis model. This model will analyze and calculate based on the input waveform deviation and output the corresponding output waveform deviation result. Once the system terminal obtains the output waveform deviation, it compensates the output signal waveform according to the deviation information. The purpose of compensation is to adjust the output signal waveform to make it closer to the expected waveform, thereby eliminating or reducing the impact caused by the deviation. Through this series of steps, the output signal waveform can be optimized and improved, and the accuracy and reliability of signal processing can be improved.

[0077] In summary, the embodiments of the present application have at least the following technical effects:

[0078] The embodiment of the present application reads the physical data of the target single-phase rectifier, including electrical characteristics, rectification characteristics and scenario characteristics, and builds a multidimensional constraint space based on these data to retrieve and obtain rectification fault records. Subsequently, these fault records are mined and analyzed, and a fault waveform decision tree is generated by clustering. After that, a signal fault identification channel is built based on the fault waveform decision tree, and a similarity matcher is embedded in the channel. Then the output signal waveform is collected and transferred to the signal fault identification channel for similarity comparison, and the fault identification type is output according to the comparison result. In addition, in order to obtain the rectification fault record more accurately, the associated scale evaluation function is used to evaluate the retrieved data, and the data that meets the predetermined associated scale threshold is extracted to construct the fault record. At the same time, in order to construct the fault waveform decision tree, data mining and cluster analysis are performed on the fault record based on the predetermined fault type to obtain a standard waveform contour image, and a decision tree is constructed in combination with the decision tree principle. When building the signal fault identification channel, a convolutional neural network is used, and a fault waveform decision tree is embedded. At the same time, a feature extraction unit and a similarity comparison unit are also built to generate a similarity matcher to improve the comparison accuracy. If the comparison result does not meet the predetermined similarity constraint, the expert judgment channel is activated to determine the fault type, and the signal fault identification channel is incrementally learned. These technical effects jointly solve the technical problems that traditional methods face complex electrical characteristics, rectification characteristics, and scene characteristics, which lead to difficulties in fault location, frequent misjudgments and missed detections, and thus prolong maintenance time and increase the waste of maintenance resources due to low detection efficiency and insufficient accuracy. They improve the efficiency and accuracy of signal fault detection, quickly locate fault points, reduce misjudgments and missed detections, save maintenance time and maintenance resources, and achieve the effect of ensuring stable operation of equipment.

[0079] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. Other embodiments are within the scope of the attached claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0080] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0081] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A signal fault detection method for a single-phase rectifier, characterized in that: The method further comprises: Reading vital sign data of a target single-phase rectifier, wherein the vital sign data includes electrical characteristics, rectification characteristics, and scene characteristics; Building a multidimensional constraint space according to the electrical characteristics, rectification characteristics and scenario characteristics, and retrieving rectification fault records with the multidimensional constraint space as a restriction condition; Performing mining analysis on the rectifier fault records and clustering to generate a fault waveform decision tree; Based on the fault waveform decision tree, a signal fault identification channel is constructed, wherein the signal fault identification channel is embedded with a similarity comparator; Collecting the output signal waveform, and transferring the output signal waveform to the signal fault identification channel for similarity comparison, and outputting the fault identification type based on the comparison result; The performing of mining and analysis of the rectification fault records and clustering to generate a fault waveform decision tree includes: Performing data mining on the rectifier fault records based on predetermined fault types to obtain a plurality of fault data sets; Performing cluster analysis on the plurality of fault data sets in sequence, and performing waveform contour enhancement on the cluster analysis results to obtain a plurality of standard waveform contour images, wherein the standard waveform contour images correspond to the fault types one by one; Based on the decision tree principle, the fault waveform decision tree is constructed with the fault type as the main node and the standard waveform contour image as the subsidiary node; Wherein, the signal fault identification channel is established based on the fault waveform decision tree, including: Building a signal fault identification channel based on a convolutional neural network, and embedding the fault waveform decision tree in the signal fault identification channel; The signal fault identification channel includes an input layer, a similarity calculation layer, a comparison and determination layer and an output layer, wherein the similarity calculation layer is embedded with a similarity comparator, and the comparison and determination layer includes a predetermined similarity constraint and a predetermined judgment logic; The predetermined judgment logic is that if there is a similarity calculation result that satisfies the predetermined similarity constraint, the fault type corresponding to the maximum similarity is output; if there is no similarity calculation result that satisfies the predetermined similarity constraint, the expert judgment channel is activated to perform fault type judgment, the expert judgment result is output, and the signal fault identification channel is incrementally learned based on the input image and the expert judgment result.

2. The method according to claim 1, characterized in that Taking the multi-dimensional constraint space as a constraint condition, searching and obtaining rectification fault records includes: The electrical characteristics include voltage characteristics and current characteristics, the rectification characteristics include rectification type, rectification efficiency and filtering characteristics, and the scenario characteristics include source equipment, electromagnetic interference characteristics and load characteristics; Building an eight-dimensional constraint space based on the voltage characteristics, current characteristics, rectification type, rectification efficiency, filtering characteristics, source equipment, electromagnetic interference characteristics, and load characteristics; Based on the electric power big data, data retrieval is performed with the eight-dimensional constraint space as a restriction condition to obtain the rectification fault record.

3. The method according to claim 2, characterized in that Obtaining the rectification fault record includes: Configure a predetermined correlation scale threshold; Evaluating the search data based on the correlation scale evaluation function, and extracting the search data that meets the predetermined correlation scale threshold to construct the rectification fault record; The association scale evaluation function is: ； Wherein, C represents the correlation scale between the retrieval data and the eight-dimensional constraint space, For the The weight coefficient of the attribute is set based on the influence ratio. The first of the eight dimensional features of the retrieved data Attribute feature value, is the first Attribute characteristic value.

4. The method according to claim 1, characterized in that: The method further comprises: Performing cluster analysis on the rectifier fault records based on a K-means algorithm to obtain multiple rectifier fault data sets; performing fault type identification and frequency statistics on the multiple rectifier fault data sets in sequence to generate a fault type sequence; The fault type that meets the preset ranking in the fault type sequence is extracted and set as the predetermined fault type.

5. The method according to claim 1, characterized in that The method further comprises: Building a feature extraction unit based on predetermined waveform comparison indicators, wherein the predetermined waveform comparison indicators include peak value, valley value, period, amplitude, and phase; A similarity comparison unit is built based on the predetermined waveform comparison index and the Euclidean decision distance, and the similarity comparator is generated in combination with the feature extraction unit.

6. The method according to claim 2, characterized in that The output signal waveform is transferred to the signal fault identification channel for similarity comparison, and the process also includes: Retrieve a sample input waveform set and a sample output waveform set based on the eight-dimensional constraint space; Performing deviation analysis on the sample input waveform set according to the standard input waveform to obtain an input waveform deviation set; Performing deviation analysis on the sample output waveform set according to the standard output waveform to obtain an output waveform deviation set; Training a waveform deviation analysis model based on the input waveform deviation set and the output waveform deviation set to obtain a converged waveform deviation analysis model that meets expected constraints; The input waveform deviation is collected and transferred to the converged waveform deviation analysis model to obtain the output waveform deviation, and the output signal waveform is compensated based on the output waveform deviation.

Images