Power industry data classification and grading method and system based on deep learning
By combining the deep learning method of CNN and CAA attention mechanism, the difficult problem of data classification and grading in power enterprises was solved, the rapid and accurate classification and grading of power industry data was achieved, and the data management and decision-making capabilities were improved.
Patent Information
- Application Number
- CN202510810200.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-17
- Publication Date
- 2025-10-03
AI Technical Summary
Power companies are faced with a huge amount of data with complex and diverse types. Existing technologies make it difficult to quickly and accurately classify and grade power industry data, which affects data management and decision-making capabilities.
A deep learning-based method is adopted, combined with convolutional neural network (CNN) and CAA attention mechanism. Through data preprocessing, feature extraction, classification and grading, the classification and grading network is optimized to achieve fast and accurate classification and grading of power industry data.
It improves the accuracy and efficiency of data processing in the power industry, enhances data management and decision-making capabilities, can effectively process power grid operation data, equipment status monitoring data and user electricity consumption data, and improves the accuracy and practicality of classification and grading.
Smart Images
Figure CN120744604A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data classification technology, and in particular to a method and system for data classification and grading in the power industry based on deep learning. Background Art
[0002] In the power industry, data classification and grading are crucial measures to ensure data security and promote the rational development and utilization of data. With the rapid development of the digital economy, power companies, as key data holders and processors, face the challenge of massive amounts of data, complex and diverse data types, and enormous value potential. Power companies urgently need to establish unified and scientific data classification and grading methods to more efficiently process the industry's complex data, enabling rapid and accurate data classification and grading, and improving the industry's data management and decision-making capabilities. Summary of the Invention
[0003] To address the problems in the existing technology, the present invention provides a method and system for data classification and grading in the power industry based on deep learning. The specific technical solution is as follows:
[0004] A method for classifying and grading data in the power industry based on deep learning, comprising the following steps:
[0005] Step S1, collecting power industry data and preprocessing it to obtain preprocessed power industry data;
[0006] Step S2: constructing a CNN model, inputting the pre-processed power industry data into the CNN model and extracting a power industry data feature map;
[0007] Step S3, inputting the power industry data characteristic graph into the CAA module to obtain an enhanced power industry data characteristic graph;
[0008] Step S4: constructing a classification and grading network based on deep learning, and using the power industry data feature map and the enhanced power industry data feature map to train the classification and grading network to obtain a trained classification and grading network;
[0009] Step S5, optimizing the trained classification and grading network to obtain an optimized classification and grading network;
[0010] Step S6: Collect the power industry data to be inspected, pre-process it, and input it into the optimized classification and grading network to obtain the fault type and health status level corresponding to the data.
[0011] Preferably, the preprocessing in step S1 includes data cleaning, data standardization, normalization processing, feature encoding, and data enhancement processing.
[0012] Preferably, the step S2 constructs a CNN model, and extracts the power industry data feature graph after inputting the pre-processed power industry data into the CNN model, specifically comprising the following steps:
[0013] Step S21, construct a CNN model consisting of 3 convolutional layers, 2 pooling layers and 1 fully connected layer;
[0014] Step S21, extracting spatial features of the power industry data through convolution operations, and using convolution layers to capture local patterns and edge information of the power industry data;
[0015] In step S23, features at different levels are integrated through structures such as skip connections or residual connections to obtain a data feature map of the power industry.
[0016] Preferably, the step S3 of inputting the electric power industry data characteristic graph into the CAA module to obtain the enhanced electric power industry data characteristic graph specifically includes the following steps:
[0017] Step S31, performing global average pooling on the power industry data feature map to extract global context information;
[0018] Step S32, adjusting the number of channels through convolution to generate correlations between different spatial positions in the power industry data feature map; Step S33, applying depthwise separable convolution to the convolved power industry data feature map;
[0019] In step S34, attention weights are generated through the Sigmoid activation function to selectively enhance or suppress different areas of the power industry data feature map to obtain an enhanced power industry data feature map.
[0020] Preferably, the step S4 constructs a classification and grading network based on deep learning, and trains the classification and grading network using the power industry data feature map and the enhanced power industry data feature map, and obtains the trained classification and grading network, specifically including the following steps:
[0021] Step S41, constructing a classification and grading network including two fully connected layers and one output layer, including a classifier and a classifier, for classifying and grading power industry data;
[0022] Step S42: fusing the electric power industry data feature graph and the enhanced electric power industry data feature graph to obtain a fused electric power industry data feature graph;
[0023] In step S43, the fused electric power industry data feature map is input into the classifier and classifier to classify and classify the electric power industry data, and the cross entropy loss is selected as the loss function to train the classification and grading network to obtain a trained classification and grading network.
[0024] Preferably, in step S5, the trained classification and grading network is optimized to obtain the optimized classification and grading network, which specifically includes the following steps:
[0025] Step S51, using the Bayesian optimization algorithm to optimize the hyperparameters of the classification and grading network to find the best model parameter combination;
[0026] Step S52: Verify the classification and grading results obtained by training the classification and grading network on the training set on the validation set, calculate the evaluation index of the classification and grading network, and output the optimized classification and grading network when the evaluation index reaches a preset value.
[0027] A data classification and grading system for the power industry based on deep learning, applying the method described above, includes:
[0028] The data preprocessing module is used to collect and preprocess the power industry data to obtain the preprocessed power industry data; the CNN model construction module is used to construct the CNN model, input the preprocessed power industry data into the CNN model, and then extract the power industry data feature map;
[0029] A CAA module, used for inputting the power industry data feature map into the CAA module to obtain an enhanced power industry data feature map;
[0030] The classification and grading network training module is used to build a classification and grading network based on deep learning. The classification and grading network is trained using the power industry data feature map and the enhanced power industry data feature map to obtain a trained classification and grading network.
[0031] The classification and grading network optimization module is used to optimize the trained classification and grading network to obtain the optimized classification and grading network;
[0032] The classification and grading module is used to collect the power industry data to be inspected, pre-process it and input it into the optimized classification and grading network to obtain the fault type and health status level corresponding to the data.
[0033] A computer-readable storage medium includes a stored program, wherein when the program is run, the device where the computer-readable storage medium is located is controlled to execute the method for classifying and grading power industry data based on deep learning.
[0034] A processor is used to run a program, wherein when the program is running, the method for classifying and grading data in the power industry based on deep learning is executed.
[0035] Compared with the prior art, the present invention has the following beneficial effects:
[0036] The present invention improves the accuracy and efficiency of data processing in the power industry by combining convolutional neural networks (CNN) and CAA attention mechanisms. The algorithm includes the following steps: data preprocessing, cleaning, standardizing and normalizing the raw data of the power industry to enhance the robustness of the model; feature extraction: utilizing the powerful feature extraction capability of CNN to automatically extract key features from the preprocessed data. CNN captures the spatial hierarchical features of data through multiple layers of convolutional layers; a CAA module is introduced on the basis of CNN, which extracts global context information through a global average pooling layer and gradually captures larger context information through depthwise separable convolution, and finally uses a sigmoid activation function to generate attention weights for feature maps to selectively enhance or suppress features in different regions; classification and grading: combining CAA-enhanced feature maps, using deep learning technology to classify and grade data to identify different levels of data importance and urgency; further optimizing and adjusting the classification and grading results to ensure the accuracy and practicality of the results. The present invention can effectively process complex data in the power industry, including grid operation data, equipment status monitoring data, and user electricity consumption data, and achieve fast and accurate classification and grading of data. By introducing the CAA attention mechanism, the algorithm not only improves the efficiency of data processing, but also enhances the data management and decision-making capabilities of the power industry. It has important practical application value and broad market prospects. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly describes the drawings required for the specific embodiments or the description of the prior art. Similar elements or parts are generally identified by similar reference numerals throughout the drawings. Elements or parts in the drawings are not necessarily drawn to scale.
[0038] Figure 1 Flow chart of the method of the present invention.
[0039] Figure 2 This is a schematic diagram of the CAA module of the present invention. DETAILED DESCRIPTION
[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.
[0041] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.
[0042] It should also be understood that the terms used in the present specification are only for the purpose of describing particular embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.
[0043] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.
[0044] Example 1:
[0045] like Figure 1 As shown, this embodiment provides a method for classifying and grading data in the power industry based on deep learning, including the following steps:
[0046] Step S1: Collect and pre-process the power industry data to obtain pre-processed power industry data. The pre-processing includes data cleaning, data standardization, normalization, feature encoding, and data enhancement. The details are as follows:
[0047] (1) Data cleaning: After collecting data from the power industry, remove invalid, erroneous, or redundant data. For example, delete outliers and fill in missing values to ensure data integrity and accuracy. For grid operation data, remove records of abnormal fluctuations in parameters such as voltage and current. For equipment status monitoring data, remove erroneous readings caused by sensor failures. For user electricity consumption data, fill in the missing electricity consumption periods caused by data collection failures.
[0048] (2) Data standardization. Convert data with different dimensions and units to unified dimensions and units. For example, convert voltage from kilovolts (kV) to volts (V) and current from amperes (A) to milliamperes (mA) so that the model can better process and compare data from different sources.
[0049] (3) Normalization. Scale the data to a specific range (0 to 1) to reduce the numerical differences between different features and improve the stability and convergence speed of model training. The normalization method can be the minimum-maximum normalization method or the Z-score normalization method.
[0050] (4) Feature encoding. Perform one-hot encoding or label encoding on categorical variables to make them processable by the model. For example, for user types (such as residential, commercial, industrial, etc.) in user electricity consumption data, use one-hot encoding to convert them into multiple binary features.
[0051] (5) Data augmentation. Increase data diversity through rotation, scaling, cropping, and other methods to improve the generalization ability of the model. For example, for the time series of power grid operation data, new training samples can be generated by randomly cropping and splicing data from different time periods.
[0052] Step S2: Build a CNN model, input the pre-processed power industry data into the CNN model, and extract the power industry data feature map. This specifically includes the following steps:
[0053] Step S21: Build a CNN model and design a multi-layer convolutional neural network structure, including convolutional layers, activation layers, pooling layers, etc. Specifically, build a CNN model containing 3 convolutional layers, 2 pooling layers, and 1 fully connected layer to extract features of power industry data.
[0054] Among them, the structure and sequence of the entire CNN model can be expressed as:
[0055] Input data → first convolutional layer → ReLU → first pooling layer → second convolutional layer → ReLU → second pooling layer → third convolutional layer → ReLU → flatten → fully connected layer → output feature vector.
[0056] This structural design makes full use of the feature extraction capability of the convolutional layer and the dimensionality reduction capability of the pooling layer, and finally integrates the extracted features into a feature vector that can be used for classification and grading tasks through the fully connected layer.
[0057] In the constructed convolutional neural network (CNN) model, preprocessed power industry data is first input into the first convolutional layer. This layer convolves the input data with a 3x3 convolution kernel to extract local features and enhances the model's nonlinear expression capabilities through the ReLU activation function. The output is the feature map after the first convolution. Subsequently, the feature map after the first convolution is input into the first pooling layer, which uses maximum pooling (or average pooling) to downsample the feature map to reduce its dimensionality while retaining important information. The output is the feature map after the first pooling.
[0058] The first layer of pooled feature maps is used as input to the second convolutional layer, where a 3x3 convolution kernel is used again for convolution to further extract deeper local features. A nonlinear transformation is then performed using the ReLU activation function to obtain the second layer of convolution feature maps. The second layer of convolution feature maps is then input to the second pooling layer for maximum pooling (or average pooling) downsampling, and the second layer of pooled feature maps is output.
[0059] The feature map after the second layer of pooling is input to the third convolutional layer, which uses a 3x3 convolution kernel to perform convolution operations to extract higher-level features. Nonlinear processing is then performed using the ReLU activation function, outputting the feature map after the third convolution. The feature map after the third convolution is input to the fully connected layer, which first flattens the feature map into a one-dimensional vector. The fully connected layer then further processes and integrates the features, ultimately outputting a feature vector for subsequent classification and grading tasks.
[0060] The formula for the convolution operation is as follows:
[0061]
[0062] Among them, (f*g)(t) represents the result of the convolution operation, which is also a function that depends on the variable t. Simply put, the function g "slides" along the function f and calculates the area of the overlapping area (or volume in the case of multidimensionality) at each position. The sum of these areas is the result of the convolution.
[0063] In step S21, convolution operations are used to extract spatial features of the power industry data. Convolutional layers are used to capture local patterns and edge information in the power industry data. In the first convolutional layer, a 3x3 convolution kernel is used to convolve the data, extracting voltage fluctuation patterns in the grid operation data and characteristics of equipment operating status changes in the equipment status monitoring data. Reinforced linear unit (ReLU) activation functions are used to enhance the model's expressiveness. Max pooling or average pooling is also used to reduce feature dimensionality and improve the model's abstraction capabilities.
[0064] The formula for the ReLU activation function is:
[0065] ReLU(x)=max(0,x);
[0066] For any input x: if x is positive or zero, the ReLU function directly returns x; if x is negative, the ReLU function returns 0.
[0067] In step S23, features at different levels are integrated through structures such as skip connections or residual connections to obtain a feature map of the power industry data, thereby enhancing the feature transfer capability of the model.
[0068] In the present invention, a skip connection is chosen for implementation. By selecting feature maps at different levels (such as the feature map F1 of the first convolutional layer and the feature map F3 of the third convolutional layer), and after aligning their sizes (such as through upsampling or downsampling operations), they are added or concatenated element by element, thereby fusing feature information at different levels. This allows the model to capture both local details and global structural information of the data, enhancing the model's ability to comprehensively capture local details and global structure.
[0069] Step S3: Input the power industry data feature graph into the CAA module to obtain an enhanced power industry data feature graph. Figure 2 As shown, the specific steps include:
[0070] In step S31, global average pooling is performed on the power industry data feature map to extract global context information. The feature map output by the CNN model is processed through the global average pooling layer to obtain a global feature vector, which contains the average feature information of the entire data. Global average pooling is a special type of average pooling, except that it does not divide the feature map into several rectangular areas. Instead, it averages all elements in the entire feature map and outputs it to the next layer. It is defined as follows:
[0071]
[0072] where x i Represents the input i-th feature map, which can be a multidimensional array containing data for multiple channels. n represents the total number of elements in the feature map. In the case of a two-dimensional feature map, n is equal to the width of the feature map multiplied by its height multiplied by the number of channels.
[0073] Step S32: Adjust the number of channels through convolution to generate correlations between different spatial locations in the power industry data feature map. Specifically, a 1x1 convolution is applied to the feature map after global average pooling, adjusting its number of channels from 64 to 32, and generating correlations between different spatial locations in the feature map.
[0074] In step S33, depthwise separable convolution is applied to the convolved power industry data feature map. Specifically, depthwise separable convolution is applied to the 1x1 convolved feature map, gradually expanding the receptive field from 3x3 to 5x5, ultimately capturing a wider range of contextual information.
[0075] In step S34, attention weights are generated by the Sigmoid activation function to selectively enhance or suppress different regions of the power industry data feature map to obtain an enhanced power industry data feature map. Specifically, the feature map after the depthwise separable convolution is subjected to the Sigmoid activation function to generate attention weights. The weight values are between 0 and 1. Regions with higher weight values indicate that the features are more important, and the model will strengthen its attention to the features of this region. The formula of the Sigmoid activation function is:
[0076]
[0077] For any input x, the output value of the Sigmoid function is compressed to the range (0,1).
[0078] In this invention, the CAA module serves as a key link between feature extraction and classification and grading. After CNN feature extraction, the CAA module further optimizes the feature representation. Through this series of operations, it selectively enhances or suppresses features in different regions, allowing the subsequent classification and grading network to operate based on more discriminative features. Compared to traditional methods that rely solely on CNN feature extraction, the introduction of the CAA module significantly improves the efficiency and accuracy of data processing, enhancing data management and decision-making capabilities in the power industry.
[0079] Step S4: construct a classification and grading network based on deep learning, and train the classification and grading network using the power industry data feature map and the enhanced power industry data feature map to obtain a trained classification and grading network. Specifically, the following steps are included:
[0080] Step S41: Construct a classification and grading network, designing a deep learning network structure for classification and grading. Specifically, a classification and grading network consisting of two fully connected layers and one output layer is constructed to classify and grade power industry data. Step S42: Feature fusion is performed on the power industry data feature map output by the CNN model and the enhanced power industry data feature map output by the CAA module to obtain a fused power industry data feature map, allowing the model to more accurately identify the data category and grade.
[0081] The classification and grading network consists of a classifier and a classifier, which are based on different parts of the same network and are used to map the fused feature maps of power industry data to specific classification and grading results. The classifier is typically a fully connected layer structure that maps feature maps to category probability distributions and is suitable for classification tasks; the classifier, on the other hand, is used for regression tasks and outputs specific numerical grades. Although the two differ in the design of the last few layers, the entire network is trained and optimized as a whole. By sharing the universality of the feature extraction module and leveraging the specificity of the classifier and classifier, it can efficiently handle the classification and grading tasks of power industry data.
[0082] In step S43, the fused power industry data feature graph is input into a classifier and a grader to classify and grade the power industry data. A cross-entropy loss is selected as the loss function for training the classification and grading network, resulting in a trained classification and grading network. Specifically, a softmax classifier is designed to classify fault types in the power grid operation data, and a regression grader is designed to classify the health status of the equipment status monitoring data, identifying the importance and urgency of the data.
[0083] Step S5, optimizing the trained classification and grading network to obtain an optimized classification and grading network. Specifically, the following steps are included:
[0084] Step S51: Use the Bayesian optimization algorithm to optimize the hyperparameters of the classification and grading network to find the best model parameter combination. Optimizing the classification and grading results and further adjusting the classification and grading results of the model through the optimization algorithm can improve the accuracy and practicality of the results.
[0085] In step S52, the classification and grading results obtained by training the classification and grading network on the training set are verified on the validation set to ensure the accuracy and practicality of the results. Evaluation metrics for the classification and grading network are calculated, and when the evaluation metrics reach preset values, the optimized classification and grading network is output. Evaluation metrics include precision, recall, and F1 score, which are used to evaluate model performance.
[0086] The model is evaluated using metrics such as accuracy, recall, and F1 score, and iteratively optimized based on the evaluation results to improve its performance. For example, if the model's accuracy on the validation set is low, the reasons for misclassification and grading can be analyzed, and the model's structure or parameters can be adjusted, with iterative optimization performed until the model's performance reaches the desired level. Step S6: Collect the power industry data to be inspected, preprocess it, and then input it into the optimized classification and grading network to obtain the corresponding fault type and health status level.
[0087] In step S6, the optimized classification and grading network is used to process the power industry data to be tested. After preprocessing, this data is input into the network, which then outputs the corresponding fault type and health status level. Fault type and health status level are key results of power industry data classification and grading. They are closely linked to the input data and reflect various aspects of the power system's operating status. Input data includes grid operation data (such as voltage, current, power, and frequency), equipment status monitoring data (such as equipment temperature, vibration, and insulation resistance), environmental data (such as temperature, humidity, and wind speed), and user electricity usage data (such as load and duration). After preprocessing, this data is input into the classification and grading network. Based on the data's characteristics and patterns, the network identifies the corresponding fault type and health status level. For example, abnormal fluctuations in voltage and current may be identified as a short circuit fault, while persistent increases in equipment temperature and abnormal vibration may indicate a moderately abnormal equipment state. Abnormal load fluctuations in user electricity usage data may be identified as an overload fault on the user side. In this way, the classification and grading network can convert complex power industry data into specific fault types and health status levels, providing decision support for the operation and maintenance of the power system.
[0088] In the classification task, specifically for fault type classification in power grid operation data, the model achieved an accuracy of 95.2%, a recall rate of 94.8%, and an F1 score of 95.0%. When identifying short-circuit faults, the model accurately distinguished between phase-to-phase short circuits and ground short circuits, with accuracy rates of 96.5% and 94.3%, respectively. In equipment status monitoring data, the model achieved an accuracy of 93.7%, a recall rate of 93.2%, and an F1 score of 93.5%, effectively identifying insulation faults, overload faults, and other types.
[0089] In the classification task, that is, for the classification of equipment health status levels, the model achieved an accuracy of 92.1%, a recall rate of 91.8%, and an F1 score of 92.0%. For example, when judging whether the equipment is in a moderately abnormal state, the model achieved an accuracy of 93.4%, a recall rate of 92.9%, and an F1 score of 93.2%. In user electricity usage data, the model achieved an accuracy of 90.5%, a recall rate of 90.2%, and an F1 score of 90.3% for the classification of user electricity usage status, and was able to effectively identify abnormal electricity usage. When comprehensively considering the performance of the classification and classification tasks, the overall accuracy of the model reached 91.5%, the recall rate was 91.2%, and the F1 score was 91.4%. These results show that the method of the present invention can not only accurately identify the fault type, but also effectively evaluate the health status level of equipment and users, providing strong support for the operation and maintenance of the power system.
[0090] The present invention greatly improves the efficiency of data processing in the power industry and reduces manual intervention and time costs through the automated feature extraction and classification and grading of deep learning models. The present invention combines the deep learning model with the CAA attention mechanism and applies it to the task of data classification and grading in the power industry, giving full play to the advantages of both, and can more accurately capture the key features and contextual information in the power industry data, thereby improving the accuracy of data classification and grading. After optimization of data preprocessing and post-processing, the algorithm that combines the CNN model with the CAA attention mechanism has strong robustness to noise and outliers in the power industry data, and can stably output reliable classification and grading results. The algorithm can effectively process complex data in the power industry, including power grid operation data, equipment status monitoring data, and user electricity consumption data, and has a wide range of practical application scenarios and market prospects.
[0091] Example 2:
[0092] Based on the same inventive concept as Example 1, this embodiment provides a data classification and grading system for the power industry based on deep learning, applying the method described above, including:
[0093] The data preprocessing module is used to collect and preprocess the power industry data to obtain the preprocessed power industry data; the CNN model construction module is used to construct the CNN model, input the preprocessed power industry data into the CNN model, and then extract the power industry data feature map;
[0094] A CAA module, used for inputting the power industry data feature map into the CAA module to obtain an enhanced power industry data feature map;
[0095] The classification and grading network training module is used to build a classification and grading network based on deep learning. The classification and grading network is trained using the power industry data feature map and the enhanced power industry data feature map to obtain a trained classification and grading network.
[0096] The classification and grading network optimization module is used to optimize the trained classification and grading network to obtain the optimized classification and grading network;
[0097] The classification and grading module is used to collect the power industry data to be inspected, pre-process it and input it into the optimized classification and grading network to obtain the fault type and health status level corresponding to the data.
[0098] Example 3:
[0099] Based on the same inventive concept as Example 1, this embodiment provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the method for classifying and grading power industry data based on deep learning.
[0100] Example 4:
[0101] Based on the same inventive concept as Example 1, this embodiment provides a processor, which is used to run a program, wherein when the program is running, the method for classifying and grading power industry data based on deep learning is executed.
[0102] Those skilled in the art will appreciate that the modules of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0103] In the embodiments provided by the present invention, it should be understood that the division of modules is merely a logical function division, and there may be other division methods in actual implementation, for example, multiple modules can be combined into one module, one module can be split into multiple modules, or some features can be ignored, etc.
[0104] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing module, or each module may exist physically separately, or two or more modules may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or software functional modules.
[0105] If the integrated module is implemented in the form of a software functional module 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, 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. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nly Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc., various media that can store program code.
[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.
Claims
1. A data classification and grading method for the power industry based on deep learning, characterized by: The following steps are involved: Step S1, collecting power industry data and preprocessing it to obtain preprocessed power industry data; Step S2: constructing a CNN model, inputting the pre-processed power industry data into the CNN model and extracting a power industry data feature map; Step S3, inputting the power industry data characteristic graph into the CAA module to obtain an enhanced power industry data characteristic graph; Step S4: constructing a classification and grading network based on deep learning, and using the power industry data feature map and the enhanced power industry data feature map to train the classification and grading network to obtain a trained classification and grading network; Step S5, optimizing the trained classification and grading network to obtain an optimized classification and grading network; Step S6: Collect the power industry data to be inspected, pre-process it, and input it into the optimized classification and grading network to obtain the fault type and health status level corresponding to the data.
2. A method for classifying and grading power industry data based on deep learning according to claim 1, characterized in that: The preprocessing in step S1 includes data cleaning, data standardization, normalization processing, feature encoding, and data enhancement processing.
3. The method for classifying and grading power industry data based on deep learning according to claim 1 is characterized in that: The step S2 constructs a CNN model, and inputs the pre-processed power industry data into the CNN model to extract the power industry data feature map, which specifically includes the following steps: Step S21, construct a CNN model consisting of 3 convolutional layers, 2 pooling layers and 1 fully connected layer; Step S21, extracting spatial features of the power industry data through convolution operations, and using convolution layers to capture local patterns and edge information of the power industry data; In step S23, features at different levels are integrated through structures such as skip connections or residual connections to obtain a data feature map of the power industry.
4. The method for classifying and grading power industry data based on deep learning according to claim 1 is characterized in that: The step S3 of inputting the power industry data characteristic graph into the CAA module to obtain the enhanced power industry data characteristic graph specifically includes the following steps: Step S31, performing global average pooling on the power industry data feature map to extract global context information; Step S32, adjusting the number of channels by convolution to generate correlations between different spatial positions in the power industry data feature map; Step S33, applying depthwise separable convolution to the convolved power industry data feature map; In step S34, attention weights are generated through the Sigmoid activation function to selectively enhance or suppress different areas of the power industry data feature map to obtain an enhanced power industry data feature map.
5. The method for classifying and grading power industry data based on deep learning according to claim 1 is characterized in that: The step S4 constructs a classification and grading network based on deep learning, and trains the classification and grading network using the power industry data feature map and the enhanced power industry data feature map to obtain the trained classification and grading network, specifically including the following steps: Step S41, constructing a classification and grading network including two fully connected layers and one output layer, including a classifier and a classifier, for classifying and grading power industry data; Step S42: fusing the electric power industry data feature graph and the enhanced electric power industry data feature graph to obtain a fused electric power industry data feature graph; In step S43, the fused electric power industry data feature map is input into the classifier and classifier to classify and classify the electric power industry data, and the cross entropy loss is selected as the loss function to train the classification and grading network to obtain a trained classification and grading network.
6. The method for classifying and grading power industry data based on deep learning according to claim 1 is characterized in that: In step S5, the trained classification and grading network is optimized to obtain the optimized classification and grading network, which specifically includes the following steps: Step S51, using the Bayesian optimization algorithm to optimize the hyperparameters of the classification and grading network to find the best model parameter combination; Step S52: Verify the classification and grading results obtained by training the classification and grading network on the training set on the validation set, calculate the evaluation index of the classification and grading network, and output the optimized classification and grading network when the evaluation index reaches a preset value.
7. A data classification and grading system for the power industry based on deep learning, characterized by: The method applied to any one of claims 1 to 6, comprising: The data preprocessing module is used to collect and preprocess the power industry data to obtain the preprocessed power industry data; CNN model construction module, used to build a CNN model, input the pre-processed power industry data into the CNN model and extract the power industry data feature map; A CAA module, used for inputting the power industry data feature map into the CAA module to obtain an enhanced power industry data feature map; The classification and grading network training module is used to build a classification and grading network based on deep learning. The classification and grading network is trained using the power industry data feature map and the enhanced power industry data feature map to obtain a trained classification and grading network. The classification and grading network optimization module is used to optimize the trained classification and grading network to obtain the optimized classification and grading network; The classification and grading module is used to collect the power industry data to be inspected, pre-process it and input it into the optimized classification and grading network to obtain the fault type and health status level corresponding to the data.
8. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute a method for classifying and grading power industry data based on deep learning as described in any one of claims 1 to 6.
9. A processor, characterized in that: The processor is used to run a program, wherein when the program is run, it executes a method for classifying and grading power industry data based on deep learning as described in any one of claims 1 to 6.