Load identification method, system and device based on Radon transformation and deep learning, and storage medium
By combining Radon transform and deep learning, the problems of noise resistance and computational complexity in load identification are solved, achieving efficient and accurate load identification.
Patent Information
- Application Number
- CN202511799558.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-02
- Publication Date
- 2026-02-10
AI Technical Summary
Existing AI-based load recognition methods lack noise resistance and have high computational complexity, making it difficult to meet real-time recognition requirements.
A method combining Radon transform and deep learning is adopted. Load measurement data is converted into images through Markov transfer fields. After denoising using wavelet transform, image features are extracted using Radon transform, and a convolutional neural network is constructed for load identification.
It improves the accuracy and robustness of load recognition, reduces computational complexity, enhances the model's noise resistance and computational efficiency, and is suitable for recognition tasks in different scenarios.
Smart Images
Figure CN121502227A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of load identification, and in particular to a load identification method, system, device and storage medium based on Radon transformation and deep learning. BACKGROUND
[0002] In recent years, deep learning technology has achieved many results in the field of power systems, especially in load identification. Deep learning can build multiple hidden layers, and through layer-by-layer training, it can extract different features of load measurement data in each layer. In the high layer of the network, the shallow features are transformed into higher-dimensional abstract features, showing the deep characteristics of the data, thereby completing the load identification.
[0003] The current load identification results based on artificial intelligence are as follows. Document 1 (Wang Yi, Yi Huan, Li Songcong, etc. Non-intrusive load identification method based on VMD-LSTM [J]. Electronic Technology Application, 2023, 49(02): 127-132.) proposes a non-intrusive load identification method based on LSTM. The LSTM network is used for load identification. Document 2 (Dong Tian. Research on power load pattern identification and prediction method based on deep learning [D]. Jilin University, 2023.) describes in detail the use of deep learning methods to learn the characteristics of power load data, thereby improving the accuracy of load identification. Document 3 (Li Zheng, Wang Ze, Feng Wei, etc. Non-intrusive electrical appliance load identification method based on CNN and K-means clustering [J]. Journal of Hebei University of Science and Technology, 2022, 43(04): 365-373.) proposes a non-intrusive household appliance identification method based on the combination of convolutional neural network and K-means clustering. The K-means algorithm is used to improve the convolutional neural network, increase the difference between similar characteristic load features, and improve the accuracy of load identification. Document 4 (Yang Rui, Zou Xiaosong, Xiong Wei, etc. Non-intrusive load identification method based on VMD and PSO-SVM [J / OL]. Electrical Measurement and Instrumentation: 1-9 [2023-12-30].) proposes the advantages of VMD in signal processing, and proposes a load identification algorithm based on VMD-FastICA and VMD-Entropy-PSOSVM. This method constructs a multi-dimensional feature matrix input to establish a particle swarm optimization support vector machine (PSO-SVM) to classify and identify the load.
[0004] However, the existing artificial intelligence load identification method still has the following technical deficiencies: (1) poor noise resistance: load measurement data is easily affected by power grid fluctuations, external electromagnetic interference and other factors during acquisition, resulting in a large amount of noise in the signal, and the existing identification model based on deep learning usually lacks effective denoising preprocessing link, resulting in unstable identification results. (2) High computational complexity: the traditional deep learning model has complex network structure and long training time when processing large-scale load data, and the computational efficiency is low, which is difficult to meet the real-time identification demand. SUMMARY
[0005] The purpose of the present application is to provide a load identification method, system, device and storage medium based on Radon transform and deep learning, to improve the accuracy and robustness of load identification.
[0006] Technical scheme: a load identification method based on Radon transform and deep learning, comprising:
[0007] Data preprocessing: intercepting load measurement data, and converting the load measurement data into load images by using Markov transition field;
[0008] Image transformation: denoising the load image containing noise by using wavelet transform; and extracting image features from the denoised load image by using Radon transform;
[0009] Load identification: dividing the image features into a training set and a test set, and performing normalization processing on the training set and the test set respectively; constructing a convolutional neural network, inputting the normalized training set into the convolutional neural network to train the model, obtaining a load identification model, testing the load identification model by using the normalized test set, and identifying the load by using the tested load identification model.
[0010] Further, the intercepting load measurement data and converting the load measurement data into load images by using Markov transition field comprises:
[0011] Intercepting load measurement data, performing MTF transformation on the load measurement data according to a given image size, and obtaining load images;
[0012] Combining the load image data with the load labels of the original load measurement data to form training data.
[0013] Further, the intercepting load measurement data and converting the load measurement data into load images by using Markov transition field comprises:
[0014] Extracting the column corresponding to the voltage value , dividing the time series into A quantile box, labeled , and the same amount of data in each quantile box;
[0015] Change each data in the time series to the serial number of its corresponding quantile box;
[0016] Construct a transition matrix :
[0017] ;
[0018] Wherein, represents the frequency of the quantile box transferred to the quantile box ;
[0019] Based on the transition matrix Construct a Markov transition field :
[0020] .
[0021] Further, the wavelet transform is used for denoising the load image containing noise, comprising:
[0022] Using wavelet basis function to carry out multi-scale wavelet decomposition on the load image containing noise, to obtain the low-frequency part and multiple high-frequency detail parts of the image, to obtain the wavelet coefficients on each scale;
[0023] Compress or denoise the wavelet coefficients on each scale;
[0024] Inverse wavelet transform the processed wavelet coefficients to obtain the reconstructed image after denoising.
[0025] Further, the load image after denoising is extracted by Radon transform, comprising:
[0026] Using Radon transform to project the load image after denoising along each direction under different projection angles, selecting the projection of a specific angle range to form a projection subset, and normalizing the projection subset to form a feature vector group and extracting the final feature.
[0027] Further, the expression of Radon transform is as follows:
[0028] ;
[0029] Wherein, is the projection angle, is the distance from the origin to the straight line.
[0030] Further, the convolutional neural network comprises an input layer, two convolutional layers, an activation function layer, a pooling layer, a full connection layer and an output layer connected in sequence; the input layer receives image data and pre-processes; the convolutional layer extracts local features using a plurality of convolution kernels; the activation function layer introduces a non-linear factor to enhance the model expression ability; the pooling layer reduces the feature map size by down-sampling operation and retains important information; the full connection layer maps the feature map output by the pooling layer to the output category; a suitable output layer structure is selected, and a Softmax function is selected for the classification task to generate the probability distribution of the category.
[0031] Based on the same inventive concept, the load identification system based on Radon transform and deep learning comprises:
[0032] A data preprocessing module is configured to intercept load measurement data and convert the load measurement data into load images by using a Markov transition field.
[0033] An image transformation module is configured to denoise the load images containing noise by using wavelet transform and extract image features from the denoised load images by using Radon transform.
[0034] A load identification module is configured to divide the image features into a training set and a test set, normalize the training set and the test set respectively, construct a convolutional neural network, input the normalized training set into the convolutional neural network to train the model, obtain a load identification model, test the load identification model by using the normalized test set, and identify the load by using the tested load identification model.
[0035] Based on the same inventive concept, the load identification device based on Radon transform and deep learning comprises a processor and a memory, the memory stores computer instructions, and the processor is configured to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the device implements the steps of the load identification method based on Radon transform and deep learning.
[0036] Based on the same inventive concept, the computer readable storage medium stores a computer program, and when the program is executed by the processor, the steps of the load identification method based on Radon transform and deep learning are implemented.
[0037] Beneficial effects: Compared with the prior art, the significant technical effects of the present application are as follows: (1) Anti-noise performance is enhanced: the present application effectively suppresses the fluctuation problem caused by random noise and environmental interference in load measurement data by adopting wavelet transform for multi-scale denoising processing of the load image, solving the technical problem of decreased recognition accuracy of the traditional load identification method under noise conditions. Through this technical means, the system can still maintain the stability of feature extraction and the accuracy of model identification in a complex power load environment, significantly improving the anti-noise performance and robustness. (2) The calculation efficiency is improved: the present application introduces Radon transform to extract features and direction information from the denoised load image in the feature extraction process, reducing the redundant feature dimension, thereby reducing the computational load of the subsequent convolutional neural network. This technical means effectively solves the technical problems of high computational complexity and long training time of traditional load identification algorithms when processing large-scale load data, achieving higher calculation efficiency and lower calculation cost. (3) Generalization ability is strengthened: due to the combination of wavelet transform and Radon transform, the load identification model of the present application has stronger generalization ability and is suitable for identification tasks in different scenarios. BRIEF DESCRIPTION OF DRAWINGS
[0038] Figure 1 is a flowchart of a load identification method based on Radon transform and deep learning disclosed by an embodiment of the present application;
[0039] Figure 2 is a flowchart of a wavelet transform denoising method disclosed by an embodiment of the present application;
[0040] Figure 3 is a flowchart of a Radon transform algorithm disclosed by an embodiment of the present application;
[0041] Figure 4 is a feature extraction and identification flowchart of a CNN convolutional neural network disclosed by an embodiment of the present application;
[0042] Figure 5 is a structural diagram of a CNN convolutional neural network disclosed by an embodiment of the present application;
[0043] Figure 6 is an algorithm loss function result graph disclosed by an embodiment of the present application;
[0044] Figure 7 is a structural diagram of a load identification system based on Radon transform and deep learning disclosed by an embodiment of the present application;
[0045] Figure 8 is a structural diagram of a load identification device based on Radon transform and deep learning disclosed by an embodiment of the present application. DETAILED DESCRIPTION
[0046] The present application will be described in detail below with reference to the drawings and specific embodiments. Those skilled in the art will understand that the purposes and advantages achieved by the present application are not limited to the above specific beneficial effects described, and the above and other purposes achieved by the present application will be more clearly understood according to the following detailed description.
[0047] Those of ordinary skill in the art will understand that the exemplary components, systems and methods described in connection with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. The specific implementation depends on the specific application and design constraints imposed on the overall system. Those of ordinary skill in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.
[0048] Reference to "embodiments" in the present application means that the specific features, structures or characteristics described in connection with the embodiments can be included in at least one embodiment of the present application. The appearance of this phrase at various places in the specification does not necessarily refer to the same embodiment, nor is it independent or alternative to other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.
[0049] Embodiment 1
[0050] Please refer to Figure 1 , Figure 1 is a flowchart of a load identification method based on Radon transform and deep learning disclosed in an embodiment of the present application. Among them, Figure 1 The load identification method described is applied in the power system, such as for load identification, etc., and the present application is not limited to the embodiments. As Figure 1 shown, the load identification method based on Radon transform and deep learning can include the following operations:
[0051] S1, data preprocessing: intercepting load measurement data, and converting the load measurement data into load images using Markov transition field (MTF).
[0052] Dataset Transformation and Processing: The dataset used in this solution comes from the WHITED global household and industrial instantaneous energy dataset. Since this dataset was recorded using a sound card, the storage format is FLAC (a common lossless audio file format). The dataset contains 1100 different records from 110 different devices, which can be divided into 47 different types across 6 different regions. First, the FLAC files need to be converted using the `soundfile.read()` function to read the data and sampling rate. Then, the voltage and current of the first 22,000 samples are extracted, and the current and voltage values are calculated based on the parameters provided in the dataset. Finally, the file is saved as a one-dimensional data CSV file.
[0053] To convert one-dimensional data into image data, MTF (Markov Transition Field) transformation is used. First, a CSV file is read, then the columns corresponding to voltage values are extracted and converted into NumPy arrays. The first 22,000 elements of the array are extracted. A MarkovTransitionField object is created, with the image size set to 24. The extracted array is then subjected to MTF transformation, and the resulting image data is saved. These image data are then used to construct the image dataset.
[0054] A time-series image coding method based on Markov transition matrices. This method treats the temporal progression of a time series as a Markov process, meaning that given the current state, its future evolution does not depend on its past evolution. Based on this, a Markov transition matrix is constructed, which is then extended to a Markov transition field to achieve image coding.
[0055] The specific implementation process of step S1 is as follows:
[0056] S1.1 Image construction based on Markov Transition Field (MTF): Extract load measurement data, perform MTF transformation on the load measurement data according to the given image size, and obtain the load image.
[0057] In step S1.1, firstly, load measurement data is extracted to form the object data of the MTF (Mean Transformer). Based on the given image size, an MTF transformation is performed to obtain the load image data. For the column X corresponding to the extracted voltage value... The image encoding steps for the image (t=1,2,...,T) are as follows:
[0058] S1.1.1 Extract the column corresponding to the voltage value time series become One quantile bin, labeled as And the amount of data in each bin is the same;
[0059] S1.1.2 Change each data point in the time series to its corresponding quantile bin number;
[0060] S1.1.3 Constructing the transition matrix :
[0061] ;
[0062] in, Indicates the partition box Transfer to the partition box The frequency;
[0063] S1.1.4, Based on the transition matrix Constructing Markov transition fields :
[0064] .
[0065] S1.2 Training Data Construction: Combine the load image data obtained in step S1.1 with the load labels of the original load measurement data to form training data.
[0066] S2. Image Transformation: Denoising the noisy load image is performed using wavelet transform; Radon transform is then used to extract image features from the denoised load image.
[0067] The specific implementation process of step S2 is as follows:
[0068] S2.1 Noise Reduction Processing: such as Figure 2 As shown, wavelet transform is used to denoise the noisy load image.
[0069] In step S2.1, firstly, image preprocessing is performed using the image dataset obtained from the Markov transfer field method to convert the original image into a grayscale image, thereby reducing computational load and eliminating the influence of color on features. A grayscale conversion function can be used to convert a color image to a grayscale image. The color overlay image generated by the Markov transfer field method is converted to a grayscale image. The conversion formula for the pixel values after converting the pixel values of a color image to grayscale is as follows:
[0070] .
[0071] Then, a suitable wavelet basis function is selected, such as the Haar wavelet, DB wavelet, or SMIC wavelet. This invention selects the Haar wavelet. Specifically:
[0072] S2.1.1. Use wavelet basis functions to perform multi-scale wavelet decomposition on the noisy load image to obtain the low-frequency part and multiple high-frequency detail parts of the image, and obtain the wavelet coefficients at each scale.
[0073] S2.1.2. Wavelet coefficients at various scales can be compressed or denoised using methods such as thresholding, soft thresholding, and hard thresholding.
[0074] S2.1.3. Perform inverse wavelet transform on the processed wavelet coefficients to obtain the denoised reconstructed image. Perform necessary post-processing on the reconstructed image, such as adjusting contrast and equalization.
[0075] S2.2 Projection Transformation: such as Figure 3 As shown, Radon transform is used to extract image features from the denoised load image. The specific operation is as follows:
[0076] First, the Radon transform is used to analyze the denoised load image at different projection angles. Projecting along the lower edge in various directions, selecting projections within a specific angle range to form a projection subset, combining the projection subsets into a feature vector group, normalizing it, and extracting the final features.
[0077] This projection transformation process can map the spatial features of the original load image into a directional feature distribution, thereby providing a richer and more stable feature representation for subsequent feature learning and recognition.
[0078] The Radon transform is primarily used for projective transformations of images, converting them from physical coordinate space to Radon parameter space. It can be used to reconstruct the original image by measuring its projection at different angles. This transform is widely used in computed tomography (CT) scans to obtain a dimensionality-reduced representation of the image, and is a mathematical transformation method widely applied in image processing and medical imaging.
[0079] In this embodiment, the formula for the Radon transform can be described in the following form:
[0080] .
[0081] in, For the projection angle, To represent the distance from the origin to the line (also known as radial distance), this formula is a completely new functional correspondence, representing the given independent variable. and There will be a corresponding function value. The physical meaning of this value is brightness (or superposition, attenuated value, matching degree, etc.).
[0082] S3. Load Recognition: Divide the image features into training and test sets, and normalize the training and test sets respectively; construct a convolutional neural network, input the normalized training set into the convolutional neural network to train the model, obtain the load recognition model, test the load recognition model using the normalized test set, and use the tested load recognition model to perform load recognition.
[0083] The specific implementation process of step S3 is as follows:
[0084] S3.1, Dataset splitting: The training dataset and labels are shuffled using the same seed, and then the dataset is split according to the proportion of the labels, with 80% used as the training set and 20% as the test set.
[0085] S3.2 Normalization: Call the sklearn library to perform data normalization on the training and test sets respectively.
[0086] S3.3 Constructing a Convolutional Neural Network
[0087] A CNN (Convolutional Neural Network) model is constructed, and its learning ability is improved through iterative training. The CNN model structure consists of multiple layers, such as... Figure 4 As shown, in this embodiment, the convolutional neural network includes an input layer, two convolutional layers, an activation function layer, a pooling layer, a fully connected layer (the fully connected layer is the same as that in a conventional neural network), and an output layer connected in sequence.
[0088] S3.4 Model Training: Input the normalized training set into the convolutional neural network to train the model and obtain the load identification model. Test the load identification model using the normalized test set and use the tested load identification model for load identification.
[0089] like Figure 5 As shown, firstly, the input layer receives image data and performs necessary preprocessing; then, the convolutional layer uses multiple convolutional kernels to extract local features; next, a non-linear factor is introduced into the activation function layer to enhance the model's expressive power; subsequently, the pooling layer reduces the feature map size while retaining important information through downsampling; then, the fully connected layer maps the feature map output by the pooling layer to the output category; finally, a suitable output layer structure is selected, and the Softmax function is used for the classification task to generate the probability distribution of the categories. Throughout the process, gradients are calculated and network parameters are updated using the backpropagation algorithm, and multiple iterations of training are used to improve model performance and generalization ability.
[0090] Model Training and Classification Learning: The constructed convolutional neural network (CNN) is trained on an image dataset. A suitable dataset is selected, and necessary preprocessing is performed, such as image normalization and resizing. Based on the specific task and application scenario, a suitable CNN structure is chosen, and relevant parameters and hyperparameters are set, such as kernel size, number of filters, and pooling method. The CNN model is trained using the training set, calculating the loss function and updating network parameters using backpropagation in each iteration. Through multiple iterations, the model's performance and generalization ability are continuously optimized. The trained model is validated using a validation set to check for overfitting or underfitting. Optimization can be achieved by adjusting parameters and adding regularization. The validated model is tested using a test set to obtain performance and accuracy metrics in practical applications. The trained CNN model is then deployed to practical applications for tasks such as image classification and object detection.
[0091] The training and classification learning process of CNN models requires a lot of experiments and optimization to improve the model's performance and generalization ability. At the same time, it is necessary to build the network and set the parameters reasonably according to the specific task and application scenario.
[0092] Model Evaluation and Matrix Plotting: After training a CNN model, the training process and performance can be analyzed by plotting matrices such as the loss function. The specific steps are as follows:
[0093] Record the loss function value, accuracy, and other metrics for each round of training and save them in matrix form. Use visualization tools (such as matplotlib, TensorBoard, etc.) to convert these matrix data into visual images. Plotting the loss function curve allows observation of the model's loss function changes on the training and validation sets, and whether overfitting or underfitting occurs. Plotting the accuracy curve allows observation of the model's classification accuracy changes on the training and validation sets, and whether fluctuations or declines occur. Analyze the loss function and accuracy curves, and adjust and optimize the model according to actual needs, such as adjusting the learning rate and increasing regularization. Plotting loss function and other matrix graphs helps us better understand the model's training process and performance, thereby improving the model's performance and generalization ability. The loss function results of the algorithm in this invention are shown in the figure below. Figure 6 As shown, the accuracy of the training set and the accuracy of the validation set reach about 90%, while the loss of the training set and the loss of the validation set decrease significantly, less than 0.2, which meets the recognition requirements, indicating that the present invention has good recognition accuracy in this embodiment.
[0094] This invention uses wavelet transform and Radon transform for image preprocessing to extract local and global features; simultaneously, it utilizes a CNN convolutional network model for feature extraction and classification. This invention improves the accuracy and robustness of load identification. It has the advantages of simple implementation and high identification accuracy, and can be used in the fields of load regulation and identification.
[0095] This invention improves upon existing technologies in areas such as load data image denoising, feature extraction, and the complexity of load identification models by introducing multi-level feature extraction and denoising techniques, including Markov transfer field, wavelet transform, and Radon transform.
[0096] Example 2
[0097] Please see Figure 7 , Figure 7 This is a schematic diagram of a load identification system based on Radon transform and deep learning disclosed in an embodiment of the present invention. This system can achieve load identification and specifically includes:
[0098] The data preprocessing module is used to extract load measurement data and convert it into a load image using a Markov transfer field.
[0099] The image transformation module is used to denoise noisy load images using wavelet transform; and to extract image features from the denoised load images using Radon transform.
[0100] The load recognition module is used to divide image features into training and test sets, and normalize the training and test sets respectively; construct a convolutional neural network, input the normalized training set into the convolutional neural network to train the model, obtain the load recognition model, test the load recognition model using the normalized test set, and use the tested load recognition model to perform load recognition.
[0101] In one optional implementation, the load identification method based on Radon transform and deep learning includes: a) converting the obtained load data into a load image using a Markov transfer field; b) denoising the image using wavelet transform and projecting the load image using Radon transform to obtain an output image; c) performing classification learning using a convolutional neural network to obtain a load identification classification model.
[0102] Example 3
[0103] Please see Figure 8 , Figure 8 This is a schematic diagram of a load identification device based on Radon transform and deep learning, as disclosed in an embodiment of the present invention. Figure 8The described device can be applied to power systems, such as for load identification, and the embodiments of the present invention are not limited thereto.
[0104] like Figure 8 As shown, the device may include a processor and a memory, wherein the memory stores computer instructions, and the processor executes the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device performs the steps of the method described in the above embodiments and achieves the same technical effect as the above method. The memory may include a computer system readable medium in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The device may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the memory may be used to read and write non-removable, non-volatile magnetic media (commonly referred to as a "hard disk drive"). A program / utility having a set (at least one) of program modules may be stored in, for example, memory. Such program modules include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. The program modules typically perform the functions and / or methods described in the embodiments of the present invention.
[0105] The processor executes various functional applications and data processing by running programs stored in memory, such as the method provided in Embodiment 1 of the present invention.
[0106] Example 4
[0107] Embodiment 4 of the present invention also provides a computer-readable storage medium storing a computer program thereon. When the program is executed by a processor, it implements the steps of the method described in the above embodiments and achieves the same technical effect as the above method.
[0108] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
[0109] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wireline, optical fiber, RF, etc., or any suitable combination thereof.
[0110] Computer program code for performing the operations of this invention can be written in one or more programming languages or a combination thereof. Programming languages include object-oriented programming languages such as Java, Smalltalk, and C++, as well as conventional procedural programming languages such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0111] Of course, the computer-executable instructions provided in the embodiments of the present invention are not limited to the above-described method operations, but can also perform related operations in the methods provided in any embodiment of the present invention.
[0112] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A load identification method based on Radon transform and deep learning, characterized in that, include: Data preprocessing: Extract load measurement data and use Markov transfer field to convert the load measurement data into a load image; Image transformation: Wavelet transform is used to denoise the noisy load image; Radon transform is used to extract image features from the denoised load image; Load identification: Image features are divided into training and test sets, and the training and test sets are normalized respectively; a convolutional neural network is constructed, and the normalized training set is input into the convolutional neural network to train the model to obtain the load identification model. The load identification model is tested using the normalized test set, and the tested load identification model is used for load identification.
2. The load identification method based on Radon transform and deep learning according to claim 1, characterized in that, The process of extracting load measurement data and converting it into a load image using a Markov transfer field includes: Extract load measurement data, perform MTF transformation on the load measurement data according to the given image size, and obtain the load image; The load image data is combined with the load labels of the original load measurement data to form the training data.
3. The load identification method based on Radon transform and deep learning according to claim 2, characterized in that, The process of extracting load measurement data and performing MTF transformation on the load measurement data according to a given image size to obtain a load image includes: Extract the column corresponding to the voltage value time series Divided into One quantile bin, labeled as And the amount of data in each bin is the same; Change each data point in the time series to its corresponding quantile bin number; Construct the transition matrix : ； in, Indicates the partition box Transfer to the partition box The frequency; Based on the transition matrix Constructing Markov transition fields : 。 4. The load identification method based on Radon transform and deep learning according to claim 1, characterized in that, The denoising process for the noisy load image using wavelet transform includes: Wavelet basis functions are used to perform multi-scale wavelet decomposition on noisy load images to obtain the low-frequency part and multiple high-frequency detail parts of the image, and to obtain the wavelet coefficients at each scale. Compress or denoise the wavelet coefficients at each scale; The processed wavelet coefficients are subjected to inverse wavelet transform to obtain the denoised reconstructed image.
5. The load identification method based on Radon transform and deep learning according to claim 1, characterized in that, The step of extracting image features from the denoised load image using Radon transform includes: The Radon transform is used to project the denoised load image along various directions at different projection angles. Projections within a specific angle range are selected to form a projection subset. The projection subsets are then combined into a feature vector group, normalized, and the final features are extracted.
6. The load identification method based on Radon transform and deep learning according to claim 5, characterized in that, The expression for the Radon transform is as follows: ； in, For the projection angle, This represents the distance from the origin to the line.
7. The load identification method based on Radon transform and deep learning according to claim 1, characterized in that: The convolutional neural network includes an input layer, two convolutional layers, an activation function layer, a pooling layer, a fully connected layer, and an output layer connected in sequence. The input layer receives and preprocesses image data; the convolutional layer uses multiple convolutional kernels to extract local features; nonlinear factors are introduced into the activation function layer to enhance the model's expressive power; the pooling layer reduces the feature map size and retains important information through downsampling; the fully connected layer maps the feature map output by the pooling layer to the output category; a suitable output layer structure is selected, and the Softmax function is used for the classification task to generate the probability distribution of the categories.
8. A load identification system based on Radon transform and deep learning, characterized in that, include: The data preprocessing module is used to extract load measurement data and convert it into a load image using a Markov transfer field. The image transformation module is used to denoise noisy load images using wavelet transform; and to extract image features from the denoised load images using Radon transform. The load recognition module is used to divide image features into training and test sets, and normalize the training and test sets respectively; construct a convolutional neural network, input the normalized training set into the convolutional neural network to train the model, obtain the load recognition model, test the load recognition model using the normalized test set, and use the tested load recognition model to perform load recognition.
9. A load identification device based on Radon transform and deep learning, characterized in that, The device includes a processor and a memory, the memory storing computer instructions, the processor executing the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the device implements the steps of the load identification method based on Radon transform and deep learning as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the load identification method based on Radon transform and deep learning as described in any one of claims 1 to 7.