Power plant electric circuit fault diagnosis data processing method and system based on big data

Abstract

The application relates to the technical field of power plant electrical equipment fault diagnosis, and discloses a power plant electrical circuit fault diagnosis data processing method and system based on big data, which collects time sequence monitoring data, image data and text data of a power plant electrical circuit, carries out standardization processing, and constructs a standardized multi-modal data set; an adaptive multi-modal noise suppression algorithm is used to suppress noise in the time sequence monitoring data, the image data and the text data; shallow features and deep features of the denoised time sequence monitoring data, the denoised image data and the denoised text data are respectively extracted, and are mapped to a unified semantic space for weighted fusion to obtain a multi-modal fusion feature vector; the multi-modal fusion feature vector is input into a Transform-CNN hybrid model, a diagnosis result is output, and early warning information is sent; and the application realizes early warning and accurate intelligent diagnosis of power plant electrical circuit faults.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for electrical equipment in power plants, and specifically to a data processing method and system for fault diagnosis of electrical circuits in power plants based on big data. Background Technology

[0002] Electrical circuits in power plants are the core support for power plant production and operation, and their operating status directly determines the stability and safety of power supply. As power plant units develop towards larger capacity and higher parameters, the structure of electrical circuits becomes increasingly complex. During operation, they are susceptible to various factors such as load fluctuations, environmental interference, and equipment aging, leading to various faults such as poor contact, insulation damage, localized overheating, and abnormal parameters. If these faults are not diagnosed and addressed promptly and accurately, they can easily cause equipment damage and shutdown accidents, resulting in huge economic losses and safety hazards. Currently, the data processing mode for fault diagnosis in power plant electrical circuits is singular, mostly focusing only on single... The analysis of one type of monitoring data fails to integrate multi-source heterogeneous multimodal data, resulting in missing fault feature information and an inability to fully reflect the operating status of electrical circuits. The data utilization rate is low, as the collected raw data contains a large amount of environmental noise and equipment interference signals. Existing preprocessing methods mostly use fixed threshold denoising or simple filtering, which cannot adaptively match the noise characteristics under different operating conditions, causing effective fault features to be masked by noise and significantly reducing the accuracy of subsequent diagnosis. It is difficult to mine the deep correlation features between multimodal data, resulting in low accuracy in identifying complex faults and early weak faults, with a high false positive and false negative rate. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing a data processing method and system for fault diagnosis of power plant electrical circuits based on big data.

[0004] The first aspect of this invention provides a data processing method for fault diagnosis of electrical circuits in power plants based on big data, the method comprising the following steps: Collect time-series monitoring data, image data, and text data of the power plant's electrical circuits, standardize the collected data, and construct a standardized multimodal dataset; For standardized multimodal datasets, an adaptive multimodal noise suppression algorithm is adopted to suppress noise in time-series monitoring data, image data, and text data; Shallow and deep features of the denoised time-series monitoring data, denoised image data, and denoised text data are extracted respectively, and mapped to a unified semantic space for weighted fusion to obtain a multimodal fusion feature vector. The multimodal fusion feature vector is input into the Transformer-CNN hybrid model to diagnose fault type, fault level and fault location, output the diagnosis results, and issue warning information based on the diagnosis results and preset fault thresholds; Specifically, inputting the multimodal fusion feature vector into the Transformer-CNN hybrid model includes: inputting the multimodal fusion feature vector into the Transformer-CNN hybrid model, capturing and enhancing the different modal correlation features in the multimodal fusion feature vector through the multi-head attention mechanism of the Transformer module, mining the deep correlation information related to the fault in the feature vector, and outputting the attention-enhanced feature vector; inputting the feature vector output by the Transformer module into the CNN module, extracting local features from the feature vector through the convolutional layer of the CNN module, and then filtering and compressing the extracted local features through the pooling layer to retain the fault feature information, and outputting the filtered fault feature vector.

[0005] Optionally, in a first implementation of the first aspect of the present invention, the step of employing an adaptive multimodal noise suppression algorithm to suppress noise in time-series monitoring data, image data, and text data for a standardized multimodal dataset includes: An adaptive wavelet packet threshold denoising algorithm combined with particle swarm optimization is applied to the time-series monitoring data in the standardized multimodal dataset to obtain the denoised time-series monitoring data. Image data from a standardized multimodal dataset is input into a lightweight MobileNetV3 model to extract edge features from the images, mark the edge regions corresponding to device defects, and then dynamically adjust the spatial variance and grayscale variance of the filter based on the edge regions to obtain the denoised image data. Calculate the importance weight of each word vector in the text data of the standardized multimodal dataset related to fault diagnosis. Based on the weight, select the keyword vectors related to fault diagnosis, identify outliers in the keyword vectors, and replace the identified outliers with the mean of the word vectors corresponding to the historical text data of the same period to obtain the denoised text data.

[0006] Optionally, in a second implementation of the first aspect of the present invention, the step of performing an adaptive wavelet packet threshold denoising algorithm combined with particle swarm optimization on the time-series monitoring data in the standardized multimodal dataset to obtain denoised time-series monitoring data includes: The db4 wavelet basis was selected as the basis function for wavelet packet decomposition. Wavelet packet decomposition was performed on the standardized time-series monitoring data to obtain wavelet packet coefficients in different frequency bands. Initialize the parameters of the particle swarm, where the position of each particle corresponds to a combination of wavelet packet threshold and decomposition level; Calculate the fitness value of each particle, update the individual optimal position of each particle and the global optimal position of the entire particle swarm, and adjust the velocity and position of each particle so that the particles gradually move closer to the global optimal position. When the number of iterations of the particle swarm reaches the preset number, the iteration stops, and the parameters corresponding to the global optimal position are the optimal wavelet packet threshold and the number of decomposition layers. By employing the optimal wavelet packet threshold and decomposition level, the wavelet packet coefficients are thresholded, and the wavelet packet coefficients after thresholding are decomposed inversely to obtain the denoised time-series monitoring data.

[0007] Optionally, in a third implementation of the first aspect of the present invention, the step of extracting shallow and deep features from the denoised time-series monitoring data, the denoised image data, and the denoised text data, respectively, and mapping them to a unified semantic space for weighted fusion to obtain a multimodal fusion feature vector, includes: The denoised time-series monitoring data is analyzed segment by segment to extract the effective value, peak factor, kurtosis and waveform factor of the data, and integrated to form a shallow feature set of the time-series monitoring data; A CNN-LSTM hybrid neural network is used to extract deep features from the denoised time-series monitoring data, and the deep feature vector of the time-series monitoring data is output. The denoised image data is processed to grayscale, and then the contrast, entropy, and energy of the image are calculated and integrated to form a shallow feature set of the image data. The denoised image data is input into a ResNet-50 neural network with SE-Net attention mechanism. The spatial features of the image are extracted step by step through convolutional layers and pooling layers. The attention mechanism is used to enhance the ability to extract features of the defective areas of the device and output the deep feature vector of the image data. The word vectors of the denoised text are filtered to extract the word vectors corresponding to keywords related to fault diagnosis of electrical circuits in power plants, and integrated to form a shallow feature set of the text data. The denoised text word vectors are input into the BERT-BiLSTM hybrid model. The semantic features of the text are extracted through the encoding layer to capture the contextual associations related to faults in the text. Then, the temporal correlation of the text features is enhanced through the BiLSTM network to output the deep feature vector of the text data. The shallow and deep features of the extracted time-series monitoring data, image data, and text data are concatenated to form a time-series comprehensive feature vector, an image comprehensive feature vector, and a text comprehensive feature vector. The temporal integrated feature vector, image integrated feature vector, and text integrated feature vector are mapped to a unified semantic space and then weighted and fused to obtain a multimodal fused feature vector.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, the step of using a CNN-LSTM hybrid neural network to extract deep features from the denoised time-series monitoring data and outputting a deep feature vector of the time-series monitoring data includes: The denoised time-series monitoring data is divided into continuous sample segments of fixed length. The sample segments are then input into a CNN network, and the convolutional layers are used to extract features from the sample segments to capture the local features of the data. The local features output by the CNN network are input into the LSTM network, and the long-term dependencies of the time series data are captured through the hidden layers, outputting the deep feature vector of the time series monitoring data.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, the step of performing fault type, fault level, and fault location diagnosis, outputting diagnosis results, and issuing warning information based on the diagnosis results and a preset fault threshold includes: The fault feature vector output by the CNN module is fed into the fully connected layer. The fully connected layer integrates and maps the fault feature vector, and outputs the feature mapping results corresponding to the fault type, fault level and fault location respectively. The feature mapping results output by the fully connected layer are classified and judged, and then integrated to obtain the fault diagnosis results. The output fault diagnosis results are compared with the preset fault warning thresholds to determine whether the current fault diagnosis results meet the warning conditions. If the warning conditions are met, the corresponding warning information is generated according to the fault level and fault type.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, the step of classifying and judging the feature mapping results output by the fully connected layer and integrating them to obtain the fault diagnosis result includes: The feature mapping data corresponding to the fault type is input into the softmax classifier. The softmax classifier performs probability allocation to obtain the probability value corresponding to each fault type. The fault type with the highest probability value is used to determine the fault type result. Obtain the feature thresholds corresponding to different preset fault levels, compare the feature mapping data corresponding to the fault level with each level threshold one by one, and determine the corresponding fault level based on the comparison results. Calculate the attention weights of different location features in the feature mapping data corresponding to the fault location, filter out the location features with the highest weights, and determine the power plant electrical circuit node corresponding to the location features with the highest weights as the fault location. By integrating the identified fault type, fault level, and fault location, a fault diagnosis result is obtained.

[0011] A second aspect of the present invention provides a big data-based power plant electrical circuit fault diagnosis data processing system, the system comprising: The data processing module is used to collect time-series monitoring data, image data, and text data of the power plant's electrical circuits, and to standardize the collected data to construct a standardized multimodal dataset. The noise reduction module is used to suppress noise in time-series monitoring data, image data, and text data by employing an adaptive multimodal noise suppression algorithm for standardized multimodal datasets. The weighted fusion module is used to extract shallow and deep features from the denoised time-series monitoring data, denoised image data, and denoised text data respectively, and map them to a unified semantic space for weighted fusion to obtain a multimodal fusion feature vector. The fault diagnosis module is used to input multimodal fusion feature vectors into the Transformer-CNN hybrid model to diagnose fault type, fault level and fault location, output diagnosis results, and issue early warning information based on the diagnosis results and preset fault thresholds. Specifically, inputting the multimodal fusion feature vector into the Transformer-CNN hybrid model includes: inputting the multimodal fusion feature vector into the Transformer-CNN hybrid model, capturing and enhancing the different modal correlation features in the multimodal fusion feature vector through the multi-head attention mechanism of the Transformer module, mining the deep correlation information related to the fault in the feature vector, and outputting the attention-enhanced feature vector; inputting the feature vector output by the Transformer module into the CNN module, extracting local features from the feature vector through the convolutional layer of the CNN module, and then filtering and compressing the extracted local features through the pooling layer to retain the fault feature information, and outputting the filtered fault feature vector.

[0012] A third aspect of the present invention provides a big data-based power plant electrical circuit fault diagnosis data processing device, the big data-based power plant electrical circuit fault diagnosis data processing device comprising a memory and at least one processor, the memory storing instructions; the at least one processor calling the instructions in the memory to cause the big data-based power plant electrical circuit fault diagnosis data processing device to perform the various steps of the big data-based power plant electrical circuit fault diagnosis data processing method as described in any of the preceding claims.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the big data-based power plant electrical circuit fault diagnosis data processing method as described in any of the preceding claims.

[0014] The technical solution provided by this invention involves collecting time-series monitoring data, image data, and text data of power plant electrical circuits, standardizing the collected data to construct a standardized multimodal dataset, and employing an adaptive multimodal noise suppression algorithm to suppress noise in the time-series monitoring data, image data, and text data. Shallow and deep features are extracted from the denoised time-series monitoring data, denoised image data, and denoised text data, respectively, and mapped to a unified semantic space for weighted fusion to obtain a multimodal fusion feature vector. This multimodal fusion feature vector is then input into a Transformer-CNN hybrid model for fault type, fault level, and fault location diagnosis, outputting diagnostic results, and issuing early warning information based on the diagnostic results and preset fault thresholds. This invention achieves early warning and accurate intelligent diagnosis of power plant electrical circuit faults, improving the safety and maintenance efficiency of power plant electrical equipment. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 A flowchart of a big data-based power plant electrical circuit fault diagnosis data processing method provided in an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a big data-based power plant electrical circuit fault diagnosis data processing system provided in an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a big data-based power plant electrical circuit fault diagnosis data processing device provided in an embodiment of the present invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The flowchart of the power plant electrical circuit fault diagnosis data processing method based on big data provided in this embodiment of the invention includes the following steps: Step 101: Collect time-series monitoring data, image data, and text data of the power plant's electrical circuits; standardize the collected data and construct a standardized multimodal dataset. In this embodiment, deploying high-precision sensors at key nodes of the power plant's electrical circuits is fundamental for acquiring time-series monitoring data. Examples include current sensors, voltage sensors, and temperature sensors, which operate in real-time at a fixed sampling frequency, such as 100 times per second, accurately capturing key data such as current, voltage, and temperature. After data acquisition, a Z-Score normalization method is used to eliminate the influence of different dimensions and magnitudes. For each time-series data point... x Through formula

[0019] Calculate its standardized value, where μ It is the mean of the time series data. σ It is the standard deviation, which ultimately transforms the data into standard normal distribution data with a mean of 0 and a standard deviation of 1.

[0020] Image data acquisition relies on high-definition cameras installed at key locations on the power plant's electrical equipment. These cameras capture images of the equipment at preset time intervals, such as every 5 minutes, or under trigger conditions, such as changes in equipment status, ensuring comprehensive information about the equipment in different states. For standardization, the images are first uniformly adjusted to a fixed size, such as 224×224 pixels, to adapt to the input requirements of subsequent models. Then, normalization is performed, linearly mapping pixel values ​​from the [0,255] range to the [0,1] range, thereby enhancing the stability and comparability of the image data.

[0021] The text data originates from power plant monitoring system logs, equipment operation records, and operator reports. After collection, a series of preprocessing operations are required, including stop word and punctuation removal, and stemming, to clean up the text content. Subsequently, using a pre-trained word vector model Word2Vec, each word is converted into a fixed-dimensional word vector, such as 300-dimensional, to construct a word vector representation of the text data, achieving standardized processing of the text data. Finally, the various standardized data are integrated to successfully construct a standardized multimodal dataset.

[0022] Step 102: For the standardized multimodal dataset, an adaptive multimodal noise suppression algorithm is used to suppress noise in time-series monitoring data, image data, and text data; In this embodiment, an adaptive wavelet packet threshold denoising algorithm combined with particle swarm optimization is applied to the time-series monitoring data in the standardized multimodal dataset to obtain denoised time-series monitoring data. The image data in the standardized multimodal dataset is input into a lightweight MobileNetV3 model to extract edge features from the images, mark the edge regions corresponding to equipment defects, and then dynamically adjust the spatial variance and gray-level variance of the filter according to the edge regions to obtain denoised image data. The importance weight of each word vector in the text data of the standardized multimodal dataset related to fault diagnosis is calculated, and keyword vectors related to fault diagnosis are selected according to the weights. Outliers in the keyword vectors are identified, and the mean of the word vectors corresponding to the historical text data of the same period is used to replace the identified outliers to obtain denoised text data.

[0023] In this embodiment, the db4 wavelet basis is selected as the basis function for wavelet packet decomposition. Wavelet packet decomposition is performed on the standardized time-series monitoring data to obtain wavelet packet coefficients in different frequency bands. The parameters of the particle swarm are initialized, where the position of each particle corresponds to a combination of wavelet packet threshold and decomposition level. The fitness value of each particle is calculated, and the individual optimal position of each particle and the global optimal position of the entire particle swarm are updated. At the same time, the velocity and position of each particle are adjusted so that the particles gradually move closer to the global optimal position. When the number of iterations of the particle swarm reaches the preset number, the iteration is stopped, and the parameters corresponding to the global optimal position are the optimal wavelet packet threshold and decomposition level. The optimal wavelet packet threshold and decomposition level are used to perform thresholding on the wavelet packet coefficients. Wavelet packet inverse decomposition is then performed on the thresholded wavelet packet coefficients to obtain the denoised time-series monitoring data.

[0024] Step 103: Extract the shallow and deep features of the denoised time-series monitoring data, denoised image data, and denoised text data respectively, and map them to a unified semantic space for weighted fusion to obtain a multimodal fusion feature vector. In this embodiment, the denoised time-series monitoring data is analyzed segment by segment to extract the effective values, peak factor, kurtosis, and waveform factor, which are then integrated to form a shallow feature set of the time-series monitoring data. A CNN-LSTM hybrid neural network is used to extract the deep features of the denoised time-series monitoring data, outputting a deep feature vector. The denoised image data is then processed in grayscale, and the contrast, entropy, and energy of the image are calculated, integrating them to form a shallow feature set of the image data. The denoised image data is then input into a ResNet-50 neural network with an added SE-Net attention mechanism. Spatial features of the image are extracted step by step through convolutional and pooling layers. The attention mechanism enhances the ability to extract features of equipment defect areas, outputting a deep feature vector of the image data. Finally, the denoised text word vectors are processed... The process involves filtering and extracting word vectors corresponding to keywords related to fault diagnosis in power plant electrical circuits, integrating them to form a shallow feature set of text data. The denoised text word vectors are then input into a BERT-BiLSTM hybrid model. The semantic features of the text are extracted through the encoding layer to capture fault-related contextual relationships within the text. The BiLSTM network is then used to enhance the temporal correlation of text features, outputting a deep feature vector of the text data. The extracted shallow and deep features from the time-series monitoring data, image data, and text data are concatenated to form a time-series comprehensive feature vector, an image comprehensive feature vector, and a text comprehensive feature vector. These three feature vectors are then mapped to a unified semantic space and weighted and fused to obtain a multimodal fusion feature vector.

[0025] In this embodiment, the denoised time-series monitoring data is divided into continuous sample segments of a fixed length. The sample segments are input into a CNN network, and feature extraction is performed on the sample segments through convolutional layers to capture the local features of the data. The local features output by the CNN network are input into an LSTM network, and the long-term dependencies of the time-series data are captured through hidden layers to output the deep feature vector of the time-series monitoring data.

[0026] Step 104: Input the multimodal fusion feature vector into the Transformer-CNN hybrid model to diagnose fault type, fault level, and fault location, output the diagnosis results, and issue early warning information based on the diagnosis results and preset fault thresholds. Specifically, inputting the multimodal fusion feature vector into the Transformer-CNN hybrid model includes: inputting the multimodal fusion feature vector into the Transformer-CNN hybrid model, capturing and strengthening the different modal correlation features in the multimodal fusion feature vector through the multi-head attention mechanism of the Transformer module, mining the deep correlation information related to the fault in the feature vector, and outputting the attention-enhanced feature vector; inputting the feature vector output by the Transformer module into the CNN module, extracting local features from the feature vector through the convolutional layer of the CNN module, and then filtering and compressing the extracted local features through the pooling layer to retain fault feature information and outputting the filtered fault feature vector.

[0027] In this embodiment, the multimodal fusion feature vector is input into the Transformer-CNN hybrid model. Through the multi-head attention mechanism of the Transformer module, the related features of different modalities in the multimodal fusion feature vector are captured and enhanced, mining deep correlation information related to the fault in the feature vector, and outputting the attention-enhanced feature vector. The feature vector output by the Transformer module is input into the CNN module. The convolutional layer of the CNN module extracts local features from the feature vector, and then the pooling layer filters and compresses the extracted local features, retaining fault feature information, and outputting the filtered fault feature vector. The fault feature vector output by the CNN module is input into the fully connected layer. The fully connected layer integrates and maps the fault feature vector, outputting feature mapping results corresponding to fault type, fault level, and fault location. The feature mapping results output by the fully connected layer are classified and judged, and integrated to obtain the fault diagnosis result. The output fault diagnosis result is compared with a preset fault warning threshold to determine whether the current fault diagnosis result meets the warning conditions. If the warning conditions are met, corresponding warning information is generated according to the fault level and fault type.

[0028] After the multimodal fusion feature vector is input into the Transformer module, the multi-head attention mechanism begins to function. The core of the multi-head attention mechanism is to split the input feature vector into multiple heads, with each head performing attention calculations independently, and finally concatenating the results.

[0029] The feature vector output from the Transformer module is input into the CNN module. First, it enters the convolutional layers. This embodiment uses three convolutional layers, each with 64 3×3 kernels, a stride of 1, and "same" padding to maintain the feature map size. The convolutional layers perform convolution operations with local regions of the input feature vector through the kernels, extracting local features. Next is the pooling layer, using max pooling with a 2×2 window and a stride of 2. The pooling layer filters and compresses the local features extracted by the convolutional layers, reducing feature dimensionality and computational cost while retaining important fault feature information. After processing by the three convolutional and three pooling layers, the filtered fault feature vector is output. The fault feature vector output from the CNN module is then fed into the fully connected layer. The fully connected layer contains three hidden layers with 256, 128, and 64 neurons respectively. The fully connected layer integrates and maps the fault feature vector through the fully connected relationships between neurons. During the training of the fully connected layers, the ReLU activation function is used to introduce non-linearity and enhance the model's expressive power. Simultaneously, Dropout technology with a dropout rate of 0.5 is employed to prevent overfitting. After processing by the fully connected layers, feature mapping results corresponding to fault type, fault level, and fault location are output.

[0030] In this embodiment, the feature mapping data corresponding to the fault type is input into a softmax classifier. The softmax classifier performs probability allocation to obtain the probability value corresponding to each fault type. The fault type with the highest probability value is determined as the fault type result. Preset feature thresholds corresponding to different fault levels are obtained. The feature mapping data corresponding to each fault level is compared with each level threshold one by one. The corresponding fault level is determined based on the comparison results. The attention weights of different location features in the feature mapping data corresponding to the fault location are calculated. The location feature with the highest weight is selected. The power plant electrical circuit node corresponding to the location feature with the highest weight is determined as the fault location. The determined fault type, fault level, and fault location are integrated to obtain the fault diagnosis result.

[0031] Please see Figure 2 A schematic diagram of the structure of a power plant electrical circuit fault diagnosis data processing system based on big data provided in this embodiment of the invention. The system includes: The data processing module is used to collect time-series monitoring data, image data, and text data of the power plant's electrical circuits, and to standardize the collected data to construct a standardized multimodal dataset. The noise reduction module is used to suppress noise in time-series monitoring data, image data, and text data by employing an adaptive multimodal noise suppression algorithm for standardized multimodal datasets. The weighted fusion module is used to extract shallow and deep features from the denoised time-series monitoring data, denoised image data, and denoised text data respectively, and map them to a unified semantic space for weighted fusion to obtain a multimodal fusion feature vector. The fault diagnosis module is used to input multimodal fusion feature vectors into the Transformer-CNN hybrid model to diagnose fault type, fault level and fault location, output diagnosis results, and issue early warning information based on the diagnosis results and preset fault thresholds. Specifically, the process of inputting the multimodal fusion feature vector into the Transformer-CNN hybrid model includes: inputting the multimodal fusion feature vector into the Transformer-CNN hybrid model, capturing and enhancing the different modal correlation features in the multimodal fusion feature vector through the multi-head attention mechanism of the Transformer module, mining the deep correlation information related to faults in the feature vector, and outputting the attention-enhanced feature vector; inputting the feature vector output by the Transformer module into the CNN module, extracting local features from the feature vector through the convolutional layers of the CNN module, and then filtering and compressing the extracted local features through the pooling layer to retain fault feature information and outputting the filtered fault feature vector.

[0032] Figure 3 This is a schematic diagram of a power plant electrical circuit fault diagnosis data processing device 300 based on big data, provided in an embodiment of the present invention. This device 300 can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the power plant electrical circuit fault diagnosis data processing device 300 based on big data. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the power plant electrical circuit fault diagnosis data processing device 300 based on big data to implement the method provided in the above embodiment.

[0033] The big data-based power plant electrical circuit fault diagnosis data processing equipment 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The structure of the power plant electrical circuit fault diagnosis data processing equipment based on big data shown does not constitute a limitation on the computer equipment provided by the present invention. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0034] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the various steps of the big data-based power plant electrical circuit fault diagnosis data processing method provided in the above embodiments.

[0035] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described equipment or apparatus / unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0036] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0037] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A data processing method for fault diagnosis of electrical circuits in power plants based on big data, characterized in that, The method includes the following steps: Collect time-series monitoring data, image data, and text data of the power plant's electrical circuits, standardize the collected data, and construct a standardized multimodal dataset; For standardized multimodal datasets, an adaptive multimodal noise suppression algorithm is adopted to suppress noise in time-series monitoring data, image data, and text data; Shallow and deep features of the denoised time-series monitoring data, denoised image data, and denoised text data are extracted respectively, and mapped to a unified semantic space for weighted fusion to obtain a multimodal fusion feature vector. The multimodal fusion feature vector is input into the Transformer-CNN hybrid model to diagnose fault type, fault level and fault location, output the diagnosis results, and issue warning information based on the diagnosis results and preset fault thresholds; Specifically, the step of inputting the multimodal fusion feature vector into the Transformer-CNN hybrid model includes: inputting the multimodal fusion feature vector into the Transformer-CNN hybrid model, capturing and enhancing the different modal correlation features in the multimodal fusion feature vector through the multi-head attention mechanism of the Transformer module, mining the deep correlation information related to faults in the feature vector, and outputting the attention-enhanced feature vector; inputting the feature vector output by the Transformer module into the CNN module, extracting local features from the feature vector through the convolutional layer of the CNN module, and then filtering and compressing the extracted local features through the pooling layer to retain fault feature information and outputting the filtered fault feature vector. Specifically, the process involves extracting shallow and deep features from the denoised time-series monitoring data, denoised image data, and denoised text data, respectively, and mapping them to a unified semantic space for weighted fusion to obtain a multimodal fusion feature vector, including: The denoised time-series monitoring data is analyzed segment by segment to extract the effective value, peak factor, kurtosis and waveform factor of the data, and integrated to form a shallow feature set of the time-series monitoring data; A CNN-LSTM hybrid neural network is used to extract deep features from the denoised time-series monitoring data, and the deep feature vector of the time-series monitoring data is output. The denoised image data is processed to grayscale, and then the contrast, entropy, and energy of the image are calculated and integrated to form a shallow feature set of the image data. The denoised image data is input into a ResNet-50 neural network with SE-Net attention mechanism. The spatial features of the image are extracted step by step through convolutional layers and pooling layers. The attention mechanism is used to enhance the ability to extract features of the defective areas of the device and output the deep feature vector of the image data. The word vectors of the denoised text are filtered to extract the word vectors corresponding to keywords related to fault diagnosis of electrical circuits in power plants, and integrated to form a shallow feature set of the text data. The denoised text word vectors are input into the BERT-BiLSTM hybrid model. The semantic features of the text are extracted through the encoding layer to capture the contextual associations related to faults in the text. Then, the temporal correlation of the text features is enhanced through the BiLSTM network to output the deep feature vector of the text data. The shallow and deep features of the extracted time-series monitoring data, image data, and text data are concatenated to form a time-series comprehensive feature vector, an image comprehensive feature vector, and a text comprehensive feature vector. The temporal integrated feature vector, image integrated feature vector, and text integrated feature vector are mapped to a unified semantic space and then weighted and fused to obtain a multimodal fused feature vector.

2. The data processing method for fault diagnosis of power plant electrical circuits based on big data as described in claim 1, characterized in that, The aforementioned adaptive multimodal noise suppression algorithm, applied to standardized multimodal datasets, suppresses noise in time-series monitoring data, image data, and text data, including: An adaptive wavelet packet threshold denoising algorithm combined with particle swarm optimization is applied to the time-series monitoring data in the standardized multimodal dataset to obtain the denoised time-series monitoring data. Image data from a standardized multimodal dataset is input into a lightweight MobileNetV3 model to extract edge features from the images, mark the edge regions corresponding to device defects, and then dynamically adjust the spatial variance and grayscale variance of the filter based on the edge regions to obtain the denoised image data. Calculate the importance weight of each word vector in the text data of the standardized multimodal dataset related to fault diagnosis. Based on the weight, select the keyword vectors related to fault diagnosis, identify outliers in the keyword vectors, and replace the identified outliers with the mean of the word vectors corresponding to the historical text data of the same period to obtain the denoised text data.

3. The data processing method for fault diagnosis of power plant electrical circuits based on big data as described in claim 2, characterized in that, The step involves performing an adaptive wavelet packet threshold denoising algorithm combined with particle swarm optimization on the time-series monitoring data in the standardized multimodal dataset to obtain denoised time-series monitoring data, including: The db4 wavelet basis was selected as the basis function for wavelet packet decomposition. Wavelet packet decomposition was performed on the standardized time-series monitoring data to obtain wavelet packet coefficients in different frequency bands. Initialize the parameters of the particle swarm, where the position of each particle corresponds to a combination of wavelet packet threshold and decomposition level; Calculate the fitness value of each particle, update the individual optimal position of each particle and the global optimal position of the entire particle swarm, and adjust the velocity and position of each particle so that the particles gradually move closer to the global optimal position. When the number of iterations of the particle swarm reaches the preset number, the iteration stops, and the parameters corresponding to the global optimal position are the optimal wavelet packet threshold and the number of decomposition layers. By employing the optimal wavelet packet threshold and decomposition level, the wavelet packet coefficients are thresholded, and the wavelet packet coefficients after thresholding are decomposed inversely to obtain the denoised time-series monitoring data.

4. The data processing method for fault diagnosis of power plant electrical circuits based on big data as described in claim 1, characterized in that, The method employs a CNN-LSTM hybrid neural network to extract deep features from the denoised time-series monitoring data, outputting a deep feature vector of the time-series monitoring data, including: The denoised time-series monitoring data is divided into continuous sample segments of fixed length. The sample segments are then input into a CNN network, and the convolutional layers are used to extract features from the sample segments to capture the local features of the data. The local features output by the CNN network are input into the LSTM network, and the long-term dependencies of the time series data are captured through the hidden layers, outputting the deep feature vector of the time series monitoring data.

5. The data processing method for fault diagnosis of power plant electrical circuits based on big data as described in claim 1, characterized in that, The process of diagnosing fault type, fault level, and fault location, outputting diagnostic results, and issuing early warning information based on the diagnostic results and preset fault thresholds includes: The fault feature vector output by the CNN module is fed into the fully connected layer. The fully connected layer integrates and maps the fault feature vector, and outputs the feature mapping results corresponding to the fault type, fault level and fault location respectively. The feature mapping results output by the fully connected layer are classified and judged, and then integrated to obtain the fault diagnosis results. The output fault diagnosis results are compared with the preset fault warning thresholds to determine whether the current fault diagnosis results meet the warning conditions. If the warning conditions are met, the corresponding warning information is generated according to the fault level and fault type.

6. The data processing method for fault diagnosis of power plant electrical circuits based on big data as described in claim 5, characterized in that, The classification and judgment of the feature mapping results output by the fully connected layer, and the integration to obtain the fault diagnosis results, include: The feature mapping data corresponding to the fault type is input into the softmax classifier. The softmax classifier performs probability allocation to obtain the probability value corresponding to each fault type. The fault type with the highest probability value is used to determine the fault type result. Obtain the feature thresholds corresponding to different preset fault levels, compare the feature mapping data corresponding to the fault level with each level threshold one by one, and determine the corresponding fault level based on the comparison results. Calculate the attention weights of different location features in the feature mapping data corresponding to the fault location, filter out the location features with the highest weights, and determine the power plant electrical circuit node corresponding to the location features with the highest weights as the fault location. By integrating the identified fault type, fault level, and fault location, a fault diagnosis result is obtained.

7. A data processing system for fault diagnosis of electrical circuits in power plants based on big data, characterized in that, The system includes: The data processing module is used to collect time-series monitoring data, image data, and text data of the power plant's electrical circuits, and to standardize the collected data to construct a standardized multimodal dataset. The noise reduction module is used to suppress noise in time-series monitoring data, image data, and text data by employing an adaptive multimodal noise suppression algorithm for standardized multimodal datasets. The weighted fusion module extracts shallow and deep features from denoised time-series monitoring data, denoised image data, and denoised text data, respectively, and maps them to a unified semantic space for weighted fusion to obtain a multimodal fusion feature vector. It analyzes the denoised time-series monitoring data segment by segment, extracting effective values, peak factors, kurtosis, and waveform factors, and integrating them to form a shallow feature set. A CNN-LSTM hybrid neural network is used to extract deep features from the denoised time-series monitoring data, outputting a deep feature vector. The denoised image data undergoes grayscale processing, and its contrast, entropy, and energy are calculated, integrating them to form a shallow feature set. The denoised image data is then input into a ResNet-50 neural network with an added SE-Net attention mechanism. Through convolutional and pooling layers, spatial features of the image are gradually extracted, and the attention mechanism enhances the processing of spatial features. The system extracts features from defective areas of equipment and outputs deep feature vectors from image data. It filters denoised text word vectors, extracting word vectors corresponding to keywords related to fault diagnosis in power plant electrical circuits, and integrates them to form a shallow feature set of text data. The denoised text word vectors are then input into a BERT-BiLSTM hybrid model. The semantic features of the text are extracted through the encoding layer, capturing fault-related contextual relationships within the text. The BiLSTM network then enhances the temporal correlation of text features, outputting deep feature vectors from the text data. The extracted shallow and deep features from the time-series monitoring data, image data, and text data are concatenated to form a time-series comprehensive feature vector, an image comprehensive feature vector, and a text comprehensive feature vector. These three feature vectors are then mapped to a unified semantic space and weighted and fused to obtain a multimodal fusion feature vector. The fault diagnosis module is used to input multimodal fusion feature vectors into the Transformer-CNN hybrid model to diagnose fault type, fault level, and fault location, output diagnosis results, and issue early warning information based on the diagnosis results and preset fault thresholds. Specifically, inputting the multimodal fusion feature vectors into the Transformer-CNN hybrid model includes: inputting the multimodal fusion feature vectors into the Transformer-CNN hybrid model, capturing and enhancing the different modal correlation features in the multimodal fusion feature vectors through the multi-head attention mechanism of the Transformer module, mining deep correlation information related to the fault in the feature vectors, and outputting the attention-enhanced feature vectors; inputting the feature vectors output by the Transformer module into the CNN module, extracting local features from the feature vectors through the convolutional layers of the CNN module, and then filtering and compressing the extracted local features through pooling layers to retain fault feature information, outputting the filtered fault feature vectors.

8. A data processing device for fault diagnosis of electrical circuits in power plants based on big data, characterized in that, The big data-based power plant electrical circuit fault diagnosis data processing device includes a memory and at least one processor. The memory stores instructions. The at least one processor invokes the instructions in the memory to cause the big data-based power plant electrical circuit fault diagnosis data processing device to perform each step of the big data-based power plant electrical circuit fault diagnosis data processing method as described in any one of claims 1-6.

9. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement each step of the big data-based power plant electrical circuit fault diagnosis data processing method as described in any one of claims 1-6.

Images