A method for suppressing impulse noise in power line carrier communication system

Abstract

The present invention relates to the field of power line data communication technology, and in particular to a method for suppressing pulse noise in a power line carrier communication system. The method is applied to a noise suppression system, and the method includes: establishing digital communication with a power line carrier communication system, and obtaining a communication data set within a preset time period based on the digital communication; the communication data set includes data points collected at a preset collection frequency within the preset time period; each data point includes multiple features; based on the communication data set, a communication feature set is established; a pre-set deep learning network model is optimized through the communication feature set to obtain an optimized deep learning network model; the current data point is read, and the current data point is input into the optimized deep learning network model to obtain predicted configuration parameters; and the power line carrier signal in the current data point is subjected to noise suppression processing according to the predicted configuration parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of power line data communication, and in particular to a method for suppressing pulse noise in a power line carrier communication system. Background Art

[0002] Power line communication (PLC) is a communication technology that transmits data over power lines and is widely used in smart grids, home automation, and industrial control. However, PLC environments are subject to various types of noise interference, particularly impulse noise, which significantly impacts communication quality. The random and transient nature of this noise makes it difficult to fully and effectively suppress it using traditional filtering and averaging methods, resulting in reduced communication efficiency and increased data transmission errors.

[0003] Existing noise suppression methods typically rely on fixed rules or simple statistical properties, making them difficult to adapt to complex and changing communication environments. With the development of artificial intelligence and deep learning technologies, combining machine learning methods with communication noise suppression has become a new research direction, enabling dynamic analysis of the communication environment and targeted optimization. However, efficiently extracting communication features and building reliable noise suppression mechanisms in complex power line environments remains a key challenge in current technological development. Summary of the Invention

[0004] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method for suppressing impulse noise in a power line carrier communication system.

[0005] In order to achieve the above objectives, the main technical solutions adopted by the present invention include:

[0006] An embodiment of the present invention provides a method for suppressing impulse noise in a power line carrier communication system. The method is applied to a noise suppression system and includes:

[0007] establishing digital communication with a power line carrier communication system, and acquiring a communication data set within a preset time period based on the digital communication;

[0008] The communication data set includes data points collected at a preset collection frequency within a preset time period; each data point includes multiple features;

[0009] establishing a communication feature set based on the communication data set;

[0010] Optimizing a preset deep learning network model using the communication feature set to obtain an optimized deep learning network model;

[0011] The deep learning network model includes a deep learning network and an additional attention layer embedded in the deep learning network for optimizing the deep learning network during the training process of the deep learning network model;

[0012] Read the current data point and input it into the optimized deep learning network model to obtain the predicted configuration parameters;

[0013] The configuration parameters include precoding parameters, adaptive spectrum and subcarrier mapping;

[0014] Noise suppression processing is performed on the power line carrier signal in the current data point according to the predicted configuration parameters.

[0015] Preferably, establishing a communication feature set based on the communication data set specifically includes:

[0016] Preprocessing the data points in the communication data set to obtain a preprocessed communication data set;

[0017] Based on the features in the preprocessed communication dataset, obtain the corresponding covariance matrix;

[0018] Performing eigenvalue decomposition on the covariance matrix to obtain a plurality of data groups, wherein each data group includes an eigenvalue and its corresponding eigenvector;

[0019] Selecting eigenvectors corresponding to N eigenvalues ​​from multiple data groups according to a pre-set strategy, and projecting each data point in the preprocessed communication data set onto each of the selected eigenvectors based on the selected eigenvectors to obtain a dimensionality-reduced data representation corresponding to the data point;

[0020] The dimensionality-reduced data representations corresponding to all data points in the preprocessed communication data set are used to form a communication feature set.

[0021] Preferably, selecting eigenvectors corresponding to N eigenvalues ​​in a plurality of data groups according to a pre-set strategy specifically includes:

[0022] For each data set’s eigenvalue and corresponding eigenvector, use formula (1) to obtain the target value of the data set;

[0023] Wherein, formula (1) is:

[0024] ;

[0025] Q is the target value of the data set;

[0026] is the eigenvalue in the data set;

[0027] is the eigenvalue in the data set The variance of the corresponding eigenvector;

[0028] M is the sum of the eigenvalues ​​in all data sets;

[0029] is the preset weight parameter;

[0030] Based on the target value of each data group, the eigenvectors corresponding to the eigenvalues ​​in the top N data groups with the largest target values ​​are selected.

[0031] Preferably, preprocessing the communication data set to obtain a preprocessed communication data set specifically includes:

[0032] Screening out data points whose time series attenuation coefficients meet a preset threshold value from the communication data set to form an initial communication data set;

[0033] Filtering the initial communication data set according to a pre-set filtering strategy to obtain a pre-processed communication data set;

[0034] Among them, the time series attenuation coefficient of the data point is calculated using formula (2);

[0035] The formula (2) is:

[0036] ;

[0037] in, is a preset first decay rate;

[0038] is a preset first weight coefficient;

[0039] is a preset second decay rate;

[0040] is a preset second weight coefficient;

[0041] is a preset third weight coefficient;

[0042] t is the time from the acquisition moment of the data point corresponding to the time series attenuation coefficient to the current moment;

[0043] It is the time series attenuation coefficient corresponding to the data point when the time between the acquisition moment and the current moment is t.

[0044] Preferably, the filtering of the initial communication data set according to a preset filtering strategy to obtain a preprocessed communication data set specifically includes:

[0045] Screening out data points that meet similar conditions in the initial communication data set, and marking the screened data points that meet similar conditions with screening identifiers;

[0046] Using a stepped window, trend adaptation analysis is performed on the data points with screening marks in the initial communication data set, so that each data point with screening marks is given a corresponding fitness score;

[0047] Filtering data points with a filter identifier in the initial communication data set to obtain a first data point set;

[0048] The first data point set consists of data points with a screening identifier whose fitness scores are greater than a preset fitness value;

[0049] Obtaining a comprehensive score corresponding to each data point in the first data point set according to the fitness score and a preset credibility score corresponding to the data point;

[0050] The data points in the first data point set whose comprehensive scores are greater than a preset threshold and the preset configuration parameters corresponding to the data points are combined to form a preprocessed communication data set.

[0051] Preferably, a step window is used to perform trend adaptation analysis on the data points with screening identifiers in the initial communication data set, so that each data point with screening identifiers obtains a corresponding fitness score, specifically including:

[0052] Using stepped windows, the initial communication data set is divided into multiple consecutive or overlapping time windows, each of which contains a preset number of data points;

[0053] In each step window, a predetermined fitting algorithm is used to perform trend fitting on the designated features of the data points with the screening mark, and a fitting trend line corresponding to the designated features is generated. The predicted value of the designated feature of each data point with the screening mark is calculated based on the fitting trend line, and the fitness score of the data point is further calculated based on the predicted value of the designated feature.

[0054] The specified characteristics include: signal strength, noise level, signal-to-noise ratio;

[0055] Wherein, the predetermined fitting algorithm is a linear fitting algorithm.

[0056] Preferably, the step of calculating the fitness score of a data point based on the predicted value of a specified feature specifically includes:

[0057] Based on the predicted value of the specified feature, the fitness score of the data point is calculated using formula (3);

[0058] The formula (3) is:

[0059] ;

[0060] ;

[0061] For the first i The fitness score of each data point;

[0062] For the first i The true value of the signal strength in the data point;

[0063] For the first i The predicted value of the signal strength in the data point;

[0064] For the first i The true value of the noise level in the data points;

[0065] For the first i The predicted value of the noise level in the data points;

[0066] For the first i The true value of the signal-to-noise ratio in the data points;

[0067] For the first i The predicted value of the signal-to-noise ratio in the data points;

[0068] is the preset maximum deviation value of signal strength;

[0069] is the preset maximum deviation value of the noise level;

[0070] is the preset maximum deviation value of the signal-to-noise ratio;

[0071] is the time position weight, which indicates the importance of the i-th data point with the screening mark in the step window;

[0072] is the collection time corresponding to the i-th data point with the screening mark;

[0073] is the center time point of the step window;

[0074] is a preset parameter value.

[0075] Preferably, obtaining a comprehensive score corresponding to each data point in the first data point set according to the fitness score and a preset credibility score corresponding to the data point specifically includes:

[0076] According to the fitness score and the preset credibility score corresponding to each data point in the first data point set, the comprehensive score corresponding to the data point is obtained using formula (4);

[0077] The formula (4) is:

[0078] ;

[0079] is the comprehensive score corresponding to the i-th data point with the screening mark;

[0080] is the average of the fitness scores corresponding to all data points with the screening flag;

[0081] is the standard deviation of the fitness scores corresponding to all data points with the screening mark;

[0082] is the pre-set credibility score corresponding to the i-th data point with the screening flag;

[0083] is the average of the credibility scores corresponding to all data points with the screening mark;

[0084] is the standard deviation of the confidence scores corresponding to all data points with the screening mark;

[0085] is the pre-set fitness weight coefficient;

[0086] is a pre-set credibility weight coefficient.

[0087] Preferably,

[0088] Among them, the deep learning network model is obtained by embedding an additional attention layer between the feature extraction layer and the fully connected layer in the convolutional neural network;

[0089] The attention addition layer adopts any one of the self-attention mechanism, the soft attention mechanism and the hard attention mechanism.

[0090] The present invention provides the following beneficial effects: A method for suppressing impulse noise in a power line carrier communication system utilizes a deep learning network model combined with a communication feature set to analyze environmental characteristics and optimize noise suppression parameter configuration, thereby specifically suppressing impulse noise in power line communications. Compared to existing technologies, this method significantly improves suppression effectiveness. Furthermore, dynamic acquisition of configuration parameters (such as precoding parameters, adaptive spectrum, and subcarrier mapping) enables the communication system to adapt to complex and changing power line environments, effectively reducing bit error rates and improving communication stability. Furthermore, inputting current data points and predicting configuration parameters based on the optimized model ensures that the communication system can rapidly respond to environmental changes and maintain high-quality data transmission. The noise environment in power line carrier communications is complex and variable. The predictive capabilities of deep learning models can accurately address diverse noise characteristics, making the system more suitable for practical application scenarios. The deep learning network model, embedded with an additional attention layer, enhances its focus on key features, enabling adaptive optimization and adapting to noise characteristics in diverse environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 This is a flow chart of a method for suppressing impulse noise in a power line carrier communication system according to the present invention;

[0092] Figure 2 Schematic diagram of the structure of an exemplary electronic device of the present invention.

[0093] Reference numerals

[0094] 300: bus; 301: receiver; 302: processor; 303: transmitter; 304: memory; 305: bus interface. DETAILED DESCRIPTION

[0095] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation methods in conjunction with the accompanying drawings.

[0096] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. Instead, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0097] Example 1

[0098] For example, in a residential smart meter network, a noise suppression system monitors communication interference on the line to ensure accurate transmission of energy metering data, preventing billing errors or data loss caused by pulse noise interference. Alternatively, in an industrial plant, a noise suppression system monitors the high-frequency pulse interference generated when large motors start or stop, ensuring stable communication between production equipment.

[0099] See also Figure 1 This embodiment provides a method for suppressing impulse noise in a power line carrier communication system. The method is applied to a noise suppression system and includes:

[0100] establishing digital communication with a power line carrier communication system, and acquiring a communication data set within a preset time period based on the digital communication;

[0101] In this embodiment, the noise suppression system exchanges data with the power line carrier communication system via a digital communication interface and collects communication data at a preset interval (e.g., hourly or daily). This enables the noise suppression system to systematically collect and analyze communication data within a time period, ensuring a comprehensive understanding of the communication process. This ensures continuous monitoring and collection of communication data from the power line carrier communication system, providing sufficient data support for subsequent noise suppression.

[0102] The communication data set includes data points collected at a preset collection frequency within a preset time period; each data point includes multiple features; such as: signal strength, noise level, signal-to-noise ratio, and communication error rate.

[0103] Based on the communication data set, a communication feature set is established; through the extraction of the feature set, the dimension of the data is compressed, and the main information in the communication process can be better reflected, thereby enhancing the accuracy of the noise suppression process.

[0104] Optimizing a preset deep learning network model using the communication feature set to obtain an optimized deep learning network model;

[0105] In which, the deep learning network model includes a deep learning network and an additional attention layer embedded in the deep learning network for optimizing the deep learning network during the training process of the deep learning network model; in this embodiment, the additional attention layer enhances the network's responsiveness to important communication features and can better identify important information related to noise suppression.

[0106] The system reads the current data point and feeds it into an optimized deep learning network model to obtain predicted configuration parameters. This system responds and predicts appropriate configuration parameters in real time, enabling noise suppression to quickly adapt to the current communication state and improving the system's dynamic adjustment capabilities. Configuration parameters determine key communication settings, including precoding, spectrum selection, and subcarrier mapping, which directly impact communication stability and anti-interference capabilities.

[0107] The configuration parameters include precoding parameters, adaptive spectrum and subcarrier mapping;

[0108] The power line carrier signal at the current data point is subjected to noise suppression processing according to the predicted configuration parameters. This embodiment adjusts signal transmission according to the predicted configuration parameters and performs noise suppression to reduce interference and optimize the power line carrier signal quality.

[0109] In this embodiment, in a smart grid, smart meters in homes and businesses transmit energy consumption data to the power company via power line carrier communication (PLC). Due to the complex power line environment, pulse noise (such as the inrush current from high-power equipment startup and power outage recovery) may interfere, resulting in unstable data transmission. A noise suppression system predicts and adjusts communication configuration parameters in real time to ensure accurate transmission of meter data. This reduces electricity metering errors and user disputes caused by communication errors. Alternatively, in industrial parks or factories, PLC is widely used for signal transmission between sensors, actuators, and controllers. However, industrial equipment such as motors, welding machines, and inverters can generate strong pulse noise that interferes with communication signals. A deep learning network model optimized by the noise suppression system can quickly adapt to complex industrial environments, achieving highly reliable communication and ensuring synchronization and real-time monitoring between production line equipment. Furthermore, in rural power grid transformation projects, remote control devices (such as voltage regulators or distribution switches) can be remotely operated via PLC. However, since rural power grid lines are long and old, they are easily affected by power equipment switching or external interference. Through the dynamic adjustment capabilities of the noise suppression system, communication parameters can be optimized in real time, the anti-interference performance of long-distance power line communication can be improved, and the accuracy and timeliness of remote control operations can be guaranteed.

[0110] Specifically, based on the communication data set, a communication feature set is established, which specifically includes:

[0111] Preprocessing the data points in the communication data set to obtain a preprocessed communication data set;

[0112] Based on the features in the preprocessed communication dataset, obtain the corresponding covariance matrix;

[0113] Performing eigenvalue decomposition on the covariance matrix to obtain a plurality of data groups, wherein each data group includes an eigenvalue and its corresponding eigenvector;

[0114] Selecting eigenvectors corresponding to N eigenvalues ​​from multiple data groups according to a pre-set strategy, and projecting each data point in the preprocessed communication data set onto each of the selected eigenvectors based on the selected eigenvectors to obtain a dimensionality-reduced data representation corresponding to the data point;

[0115] The dimensionality-reduced data representations corresponding to all data points in the preprocessed communication data set are used to form a communication feature set.

[0116] For example, assume a communication dataset for a power line carrier communication system includes the following characteristics: signal strength, noise level, and communication error rate. A dataset containing 100 data points is collected. The data points in the communication dataset are preprocessed to obtain a preprocessed communication dataset; assuming the preprocessed communication dataset contains 80 data points.

[0117] Then, the corresponding covariance matrix is ​​obtained for the features in the preprocessed communication dataset. Assume that the covariance matrix Cov of the three features in the preprocessed communication dataset is:

[0118] ;

[0119] Then, perform eigenvalue decomposition on the covariance matrix to obtain the following three eigenvalues ​​and their corresponding eigenvectors:

[0120] Eigenvalues: λ1=2.0; λ2=1.5; λ3=0.6;

[0121] Eigenvectors: v1=[0.8, 0.4, 0.4]; v2=[0.3, 0.9, 0.3]; v3=[0.1, 0.1, 0.9];

[0122] According to a pre-set strategy (e.g., selecting two eigenvalues), the corresponding eigenvectors v1 and v2 are selected. Each data point is then projected onto the eigenvectors v1 and v2, and the reduced-dimensional data representation is calculated. Finally, the reduced-dimensional data representations corresponding to all data points in the preprocessed communication dataset are combined to form a communication feature set.

[0123] In this embodiment, by decomposing and projecting eigenvalues, high-dimensional data is compressed into a low-dimensional feature set, reducing computational complexity. Feature vector selection focuses on the main direction of change, which helps to more efficiently identify noise characteristics and improve communication quality. By extracting and reducing the dimension of key features, the feature set can more effectively cope with complex noise environments and provide more accurate data support for subsequent noise suppression strategies.

[0124] Specifically, the eigenvectors corresponding to the N eigenvalues ​​are selected from the multiple data groups according to a pre-set strategy, including:

[0125] For each data set’s eigenvalue and corresponding eigenvector, use formula (1) to obtain the target value of the data set;

[0126] Wherein, formula (1) is:

[0127] ;

[0128] Q is the target value of the data set; is the eigenvalue in the data set; is the eigenvalue in the data set The variance of the corresponding eigenvector; M is the sum of the eigenvalues ​​in all data groups; is the preset weight parameter;

[0129] Based on the target value of each data group, the eigenvectors corresponding to the eigenvalues ​​in the top N data groups with the largest target values ​​are selected.

[0130] Formula (1) in this embodiment is used to optimize the target value Q by analyzing the eigenvalues ​​and eigenvectors according to a pre-set strategy, providing more accurate input for tasks such as noise suppression. It helps to identify noise fluctuations and reduce interference with the signal by optimizing the processing method. In this embodiment, the data dimension is greatly reduced through dimensionality reduction, thereby reducing the computational complexity. Assuming that the original data point has 10 features, only 3 features may be needed for processing after dimensionality reduction. In this way, the subsequent calculation speed will be greatly accelerated. Selecting the eigenvector with the largest eigenvalue can retain the core information in the data point. This helps to remove irrelevant data noise, such as unimportant background noise or signal fluctuations in communication, thereby improving the accuracy of subsequent algorithms (such as deep learning models). The dataset after dimensionality reduction is more concise, reduces noise interference, and helps to more accurately predict and suppress noise, thereby improving communication quality.

[0131] Preprocessing the communication data set to obtain a preprocessed communication data set; specifically comprising:

[0132] Screening out data points whose time series attenuation coefficients meet a preset threshold value from the communication data set to form an initial communication data set;

[0133] Filtering the initial communication data set according to a pre-set filtering strategy to obtain a pre-processed communication data set;

[0134] Among them, the time series attenuation coefficient of the data point is calculated using formula (2);

[0135] The formula (2) is:

[0136] ;

[0137] in, is a preset first decay rate; is a preset first weight coefficient; is a preset second decay rate; is a preset second weight coefficient; is the third pre-set weight coefficient; t is the time between the acquisition moment of the data point corresponding to the time series attenuation coefficient and the current moment; It is the time series attenuation coefficient corresponding to the data point when the time between the acquisition moment and the current moment is t.

[0138] Specifically, the calculation formula for the time series decay coefficient is mainly used to perform time series screening of data points. Formula (2) adjusts the time series characteristics of the data points through the decay rate and weight coefficient, performing weighted decay based on time (t) and the moment of data collection, thereby affecting the weight and importance of the data points. Specifically, the time series decay coefficient of the data point gradually decays over time, which can effectively screen out data points that are relatively "close" to the current moment and have higher reference value, thereby improving the quality of the data set during the preprocessing process, reducing unnecessary noise effects, and optimizing the efficiency and accuracy of data processing.

[0139] The initial communication data set is screened according to a pre-set screening strategy to obtain a pre-processed communication data set, specifically including:

[0140] Screening out data points that meet similar conditions in the initial communication data set, and marking the screened data points that meet similar conditions with screening identifiers;

[0141] By marking and filtering the identifiers, the range of target data points in the initial communication data set is clarified, the interference of irrelevant data is reduced, and the analysis efficiency is improved.

[0142] A stepped window is used to perform trend adaptation analysis on the data points with filtering marks in the initial communication data set, so that each data point with filtering marks obtains a corresponding fitness score; a stepped window method is introduced, and the trend changes of data points are evaluated in combination with the fitness score, so that the filtered data is more in line with actual needs and reflects the dynamic characteristics of the data.

[0143] Filtering data points with a filter identifier in the initial communication data set to obtain a first data point set;

[0144] The first data point set consists of data points with a screening identifier whose fitness scores are greater than a preset fitness value;

[0145] According to the fitness score and the pre-set credibility score corresponding to each data point in the first data point set, the comprehensive score corresponding to the data point is obtained; the comprehensive score is calculated by the fitness score and the credibility score, and the data points are further filtered to ensure that the remaining data is of higher quality and more representative.

[0146] The data points in the first data point set whose comprehensive scores are greater than a pre-set threshold, along with the preset configuration parameters corresponding to each data point, form a pre-processed communication data set. The corresponding configuration parameters are retained based on the data points to facilitate subsequent operations and analysis, forming a more complete pre-processed data set.

[0147] This embodiment combines time series analysis and trend adaptability assessment to select data points that meet specific similarity criteria and exhibit distinct trend characteristics. This dual screening mechanism of fitness and comprehensive score further enhances the representativeness and accuracy of data points while reducing the interference of noisy data. It is suitable for multiple scenarios in communication dataset processing, such as anomaly detection, trend prediction, and pattern recognition.

[0148] In the practical application of this embodiment, a step window is used to perform trend adaptation analysis on the data points with the screening identifier in the initial communication data set, so that each data point with the screening identifier obtains a corresponding fitness score, specifically including:

[0149] Using stepped windows, the initial communication data set is divided into multiple consecutive or overlapping time windows, each of which contains a preset number of data points;

[0150] In this embodiment, stepped window partitioning breaks down the initial communication data set into multiple small time windows, facilitating analysis of local trends within each window. This segmented analysis can better capture short-term fluctuations or local changes that are not easily obscured by overall trends.

[0151] Within each stepped window, a predetermined fitting algorithm is used to perform trend fitting on the specified features of the data points marked with the filter, generating a fitted trend line corresponding to the specified features. The fitted trend line then calculates the predicted value of the specified feature for each data point marked with the filter, and further calculates the fitness score of the data point based on the predicted value of the specified feature. Specifically, by screening the identified data points, the analysis is focused on data points with specific characteristics or importance, rather than conducting a comprehensive analysis of the entire dataset. This selective analysis method can significantly improve efficiency and accuracy, quantifying the importance or degree of deviation of each data point through the fitness score.

[0152] The specified characteristics include: signal strength, noise level, signal-to-noise ratio;

[0153] The predetermined fitting algorithm is a linear fitting algorithm. The predetermined fitting algorithm in this embodiment may also be a nonlinear fitting algorithm, which is not specifically limited.

[0154] Preferably, the step of calculating the fitness score of a data point based on the predicted value of a specified feature specifically includes:

[0155] Based on the predicted value of the specified feature, the fitness score of the data point is calculated using formula (3);

[0156] The formula (3) is:

[0157] ;

[0158] ;

[0159] For the first i The fitness score of each data point; For the first i The true value of the signal strength in the data point; For the first i The predicted value of the signal strength in the data point; For the first i The true value of the noise level in the data points; For the first i The predicted value of the noise level in the data points; For the first i The true value of the signal-to-noise ratio in the data points; For the first i The predicted value of the signal-to-noise ratio in the data points; is the preset maximum deviation value of signal strength; is the preset maximum deviation value of the noise level; is the preset maximum deviation value of the signal-to-noise ratio; is the time position weight, which indicates the importance of the i-th data point with the screening mark in the step window; is the collection time corresponding to the i-th data point with the screening mark; is the center time point of the step window; is a pre-set parameter value. In this embodiment, formula (3) introduces multiple features (signal strength, noise level, signal quality) to comprehensively calculate the fitness score of the data point. This method is more comprehensive than a single feature and can more accurately reflect the overall quality or importance of the data point. At the same time, formula (3) The weighted method introduces a time decay mechanism, which takes into account the time attribute of the data point. The further away the data point, the smaller its weight, highlighting temporal relevance. This helps give more weight to the latest data in trend analysis, thereby improving timeliness and accuracy. Formula (3) achieves a quantitative evaluation of data point quality by calculating the deviation between the characteristic value and the predicted value and combining the normalized range of the deviation (such as R1, R2, and R3). This evaluation method is both specific and detailed.

[0160] The step of obtaining a comprehensive score corresponding to each data point in the first data point set according to the fitness score and the preset credibility score corresponding to the data point specifically includes:

[0161] According to the fitness score and the preset credibility score corresponding to each data point in the first data point set, the comprehensive score corresponding to the data point is obtained using formula (4);

[0162] The formula (4) is:

[0163] ;

[0164] For the first i The comprehensive score corresponding to each data point; is the average of the fitness scores corresponding to all data points with the screening flag; is the standard deviation of the fitness scores corresponding to all data points with the screening mark; For the first i The pre-set confidence score corresponding to each data point; is the average of the credibility scores corresponding to all data points with the screening mark; is the standard deviation of the confidence scores corresponding to all data points with the screening mark; is the pre-set fitness weight coefficient; The present embodiment integrates the fitness score and the credibility score through a weighted combination, which not only considers the trend and deviation of the data points, but also introduces the credibility evaluation. This multi-factor fusion method surpasses the traditional single evaluation model.

[0165] Among them, the deep learning network model is obtained by embedding an additional attention layer between the feature extraction layer and the fully connected layer in the convolutional neural network;

[0166] The attention addition layer adopts any one of the self-attention mechanism, the soft attention mechanism and the hard attention mechanism.

[0167] In this embodiment, the first few layers of a convolutional neural network (CNN) extract low-level to high-level features, such as edges, textures, and patterns, from the input data through convolution and pooling operations. These features are passed to the fully connected layer for further processing. The fully connected layer maps high-dimensional features to the output space of the classification or regression task and is a key step in completing the task. However, the fully connected layers of traditional CNNs generally have difficulty processing long-range dependencies between features or global contextual information. In this embodiment, an additional attention layer is embedded between the feature extraction layer and the fully connected layer to optimize feature selection and information flow, thereby improving model performance. Embedding the additional attention layer between the feature extraction layer and the fully connected layer of the CNN not only optimizes feature selection and transmission, but also effectively solves the problems of the traditional CNN's fixed receptive field and insufficient global information capture by the fully connected layer.

[0168] In addition, this embodiment also provides an electronic device, comprising: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the power line carrier communication system pulse noise suppression method described in embodiment 1.

[0169] Attachment Figure 2 This is a schematic diagram of the structure of an exemplary electronic device of this application. Figure 2 In the figure, the bus architecture is represented by bus 300, which can include any number of interconnected buses and bridges. Bus 300 connects various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are all well known in the art and therefore will not be described further herein. Bus interface 305 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 can be the same component, namely a transceiver, which provides a unit for communicating with various other devices over a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 can be used to store data used by processor 302 when performing operations.

[0170] Example 2

[0171] This embodiment provides a method for suppressing pulse noise in a power line carrier communication system, addressing the existing technical problem of nonlinear noise introduced into the power line carrier communication system by the high peak-to-average ratio characteristics of OFDM (Orthogonal Frequency Division Multiplexing) modulation, resulting in reduced communication signal quality. By configuring an additional attention layer to optimize the deep learning network, reading real-time communication data, configuring precoding parameters, adaptive spectrum, and subcarrier mapping, and performing noise suppression processing, the technical effect of improving the signal quality of the power line carrier communication system is achieved. This embodiment provides a method for suppressing pulse noise in a power line carrier communication system, including:

[0172] Step 1: Establish digital communication with the power line carrier communication system, and obtain a communication data set within a preset time period based on the digital communication.

[0173] Specifically, based on specific communication protocols and standards, ensure that data can be efficiently transmitted between the two systems. A digital communication connection is established between the power line carrier communication system and the noise suppression system. A time period is set, such as several hours, a day, a week, or even longer, depending on the frequency requirements of the data analysis and application. Once digital communication is established, data points of communication data are collected within the preset time period. This information, including signal strength, noise level, and communication error rate, is crucial for analyzing and optimizing the communication process. Preprocessing operations such as denoising, verification, and formatting are performed on the collected data points, and the processed data points are stored as a communication dataset.

[0174] Step 2: Extract communication features based on the communication data set and establish a communication feature set.

[0175] Specifically, the communication data set is analyzed and the covariance matrix is ​​calculated to reflect the correlation between the various features in the communication data set. The covariance matrix is ​​subjected to eigenvalue decomposition to obtain a set of eigenvalues ​​and corresponding eigenvectors. The eigenvalues ​​represent the variance of the data in the direction of these eigenvectors, while the eigenvectors represent the main directions of change in the data. Based on the size of the eigenvalues, the eigenvectors are sequentially screened, and the eigenvectors corresponding to the largest eigenvalues ​​are usually selected because they represent the main directions of change in the data. Based on the screened eigenvectors, the features of the communication data set are extracted, and the original data is projected onto these eigenvectors to obtain a reduced-dimensional data representation. The extracted features are organized into a feature set, namely the communication feature set. Extracting key feature information from the original communication data and establishing a communication feature set helps the system better understand the communication environment and improve the stability and reliability of communication.

[0176] Step 3: construct a deep learning network through big data, configure an additional attention layer through the communication feature set, and optimize the deep learning network through the additional attention layer.

[0177] Specifically, big data is used for data collection to establish a basic dataset. The collected data is then screened for similarity, and similarity screening markers are established to remove potentially inaccurate or irrelevant data points. The preprocessed data is then randomly clustered, and the dataset is divided into training, validation, and test datasets. A deep learning network is trained, consisting of multiple layers, such as convolutional layers and fully connected layers, to identify and process complex patterns in the data. An attention mechanism is selected as an additional layer, such as self-attention, soft attention, or hard attention. Self-attention allows the model to focus on different parts of the input sequence, while soft attention allows the model to learn a weight distribution, and hard attention directly selects a specific part of the input sequence. The structure of the attention layer, depending on the selected attention mechanism, typically consists of a query, key, and value sequence, followed by an attention mechanism module. The attention layer is integrated into the deep learning network, typically by adding it after a specific layer. The attention layer is used to optimize the deep learning network, known as model pruning, using various techniques to identify and remove unimportant parts, reducing model complexity. Model pruning is a technique that reduces model complexity by removing unimportant connections or neurons from a network. This reduces the number of model parameters, thereby lowering computational costs and increasing model speed. During pruning, various techniques are used to identify and remove unimportant components. For example, neuron importance can be assessed based on model weights, activation values, and gradients. After removing less important neurons or connections, the model must be retrained to adapt to the new structure, which may involve adjusting network parameters and refining the training algorithm. Model pruning can reduce the amount of computation required, thereby improving local accuracy and responsiveness. Due to the reduced number of parameters, the pruned model can process new data more quickly while maintaining high prediction accuracy. The attention mechanism is a technique that improves a model's ability to focus on and respond to important features. It automatically learns the important parts of the input data and adjusts the network's output accordingly. Additional attention layers can be applied to sequence data, image data, or text data to enhance the model's ability to identify key components of the data. By configuring additional attention layers, the network can process data more efficiently, especially when dealing with long-range dependencies. By configuring additional attention layers in the deep learning network, the model's ability to identify key parts of the data can be improved, helping the model automatically learn the important parts of the data, and significantly improving the local accuracy and response speed of the power line carrier communication system.

[0178] Step 4: Read real-time communication data, and configure precoding parameters, adaptive spectrum, and subcarrier mapping based on the real-time communication data according to the optimized deep learning network.

[0179] Specifically, sensors and network monitoring equipment are used to collect real-time communication data, including signal strength, noise levels, and communication error rates. Models trained and optimized using big data and deep learning networks are used to process real-time communication data. The optimized deep learning network can identify and understand key features in the communication data and configure precoding parameters, adaptive spectrum, and subcarrier mapping accordingly. Precoding parameters refer to specific settings used in the precoding process, such as the coding algorithm, modulation method, and signal power allocation. These settings can improve the signal's resistance to interference and reduce transmission losses. Adaptive spectrum refers to a technology that dynamically selects and adjusts the communication frequency based on current spectrum conditions during communication. It can automatically identify and avoid interfering frequency bands and select the optimal communication frequency, helping to adapt to complex power line environments. Subcarrier mapping is a key step in orthogonal frequency division multiplexing technology. It determines how data is allocated to different subcarriers, improving the communication system's spectral efficiency and interference resistance. In power line carrier communication systems, subcarrier mapping maps the data stream onto a series of subcarriers that are orthogonal in the frequency domain. The performance of power line carrier communication systems can be significantly improved through the reasonable configuration and optimization of precoding parameters, adaptive spectrum and subcarrier mapping.

[0180] Step 5: Perform noise suppression processing on the power line carrier signal in the real-time communication data according to the configured precoding parameters, adaptive spectrum and subcarrier mapping.

[0181] Specifically, based on the output of the previously optimized deep learning network, precoding parameters, including the coding algorithm, modulation method, and signal power allocation, are configured. This helps improve the signal's resistance to interference, reduce transmission losses, and thus enhance communication quality. Based on the configured parameters, the communication frequency is selected and adjusted, and adaptive spectrum and subcarrier mapping are performed. Adaptive spectrum technology automatically identifies and avoids interfering frequency bands, selecting the optimal communication frequency. Subcarrier mapping maps the data stream onto a series of subcarriers that are mutually orthogonal in the frequency domain, helping to improve spectral efficiency and interference resistance. Based on the configured precoding parameters, adaptive spectrum, and subcarrier mapping, noise suppression is performed on the power line carrier signal in real-time communication data. Various signal processing techniques, such as filter design, noise estimation and suppression, and signal reconstruction, are used to reduce the impact of noise and interference on the signal. By suppressing the noise of the power line carrier signal, the system can adapt to changes in the communication environment in real time, improving communication stability.

[0182] Furthermore, step one in this embodiment includes: obtaining a data iteration cycle of the power line carrier communication system; establishing a timing attenuation coefficient based on the data iteration cycle; performing data identification of the communication data set within a preset time period using the timing attenuation coefficient, and completing acquisition of the communication data set based on the data identification result.

[0183] Specifically, an appropriate data iteration cycle is determined based on the characteristics of the power line carrier communication system, such as signal stability and noise frequency. The data iteration cycle refers to the time interval between data collection and updates, which determines the frequency of data updates and the scope of application of the time decay coefficient. During this cycle, the system collects relevant power line carrier communication data, such as signal strength, noise level, and communication error rate. Based on the determined data iteration cycle, a time decay coefficient is established to represent the degree to which the importance or influence of data decreases over time. Over time, the influence of older data gradually decreases, while the influence of newer data increases. The time decay coefficient is used to identify communication data sets within a preset time period. Each data point is labeled or classified based on the data timestamp and time decay coefficient to reflect its relative importance and influence. Based on the data identification results, the required data is filtered and extracted from the data set, including selecting data within a specific time period and excluding outdated or unimportant data, to complete the communication data set acquisition. The time decay coefficient is a coefficient that represents the change in data importance over time. By establishing the time decay coefficient, the system can more accurately assess the impact of historical data on the current communication status. By determining the data iteration cycle, establishing the time series attenuation coefficient, and performing data identification and acquisition, the system can more effectively utilize historical data and improve the accuracy of data processing.

[0184] In the specific application of this embodiment, step two includes: calculating and obtaining the covariance matrix of the communication data set; performing eigenvalue decomposition on the covariance matrix to establish eigenvalues ​​and mapping eigenvectors; sequentially screening the mapping eigenvectors based on the eigenvalues, and completing communication feature extraction based on the sequential screening results.

[0185] Specifically, the acquired communication data set is cleaned to remove invalid and erroneous data. The data is normalized or standardized to ensure the accuracy of the calculation results. For the processed data, the covariance matrix is ​​calculated. The covariance matrix is ​​a symmetric matrix used to represent the correlation between the features in the data set, where each element represents the covariance between two features in the data set. The covariance matrix is ​​subjected to eigenvalue decomposition, which is a mathematical method used to decompose a matrix into a set of eigenvalues ​​and corresponding eigenvectors. The eigenvalues ​​represent the variance of the data in the direction of these eigenvectors, while the eigenvectors represent the main directions of change of the data. Mapping the eigenvectors refers to the new representation obtained by projecting the original data onto the eigenvectors, which can be regarded as converting the data from the original feature space to a new feature space, which is defined by the eigenvectors.

[0186] The eigenvalues ​​are sorted, typically from largest to smallest. The size of the eigenvalue reflects the variance of the data in the direction of the corresponding eigenvector, so larger eigenvalues ​​indicate greater data variation in that direction. Based on the eigenvalue sorting results, the eigenvectors corresponding to the first few largest eigenvalues ​​are selected. These represent the primary direction of variation in the data and are therefore the most representative features of the data. Based on the selected eigenvectors, the features of the communication dataset are extracted, and the original data is projected onto these eigenvectors to obtain a reduced-dimensional data representation. By calculating the covariance matrix, eigenvalue decomposition, eigenvector screening, and communication feature extraction, data complexity is reduced and the performance of the communication system is improved.

[0187] Furthermore, step three of this application includes: collecting data through big data to establish a basic data set; performing data preprocessing on the basic data set to establish data preprocessing results; performing random data clustering on the data preprocessing results, and establishing a training data set, a verification data set, and a test data set based on the data random clustering results; completing the construction of a deep learning network based on the training data set, the verification data set, and the test data set.

[0188] Specifically, large amounts of data are collected from various sources. This data can be structured or unstructured, depending on the application scenario. For power line carrier communication systems, this includes communication signal data, noise data, and channel state information. Using sensors, web crawlers, and database access, data from various sources and formats is collected, including real-time monitoring of the communication system, historical records, and user data. The collected data is organized and cleaned to remove irrelevant or erroneous data points. This data is then integrated into a basic dataset suitable for machine learning. This basic dataset should contain sufficient information to facilitate subsequent analysis and modeling. Data similarity screening is performed on the basic dataset to identify data points with similar characteristics or patterns and establish similarity screening markers. The dataset is partitioned into multiple fixed-size windows, each containing a certain number of data points. Within-window trend fitness analysis is performed on the data with similarity screening markers using a stepped window. Based on the results of the within-window trend fitness analysis, a fitness ranking is generated for each data point within the window. Based on the fitness ranking, data with similarity screening markers are eliminated, eliminating data points with low fitness ranking scores. Data preprocessing is completed based on the data elimination results.

[0189] Random clustering is performed on the data preprocessing results, grouping data points into different clusters so that data points within the same cluster are similar, while data points in different clusters are dissimilar. Random data clustering is an unsupervised learning technique that initializes the clustering process by randomly selecting initial cluster centers. It does not require prior knowledge of the data's category labels and can automatically discover natural groupings within the data. Typically, it involves selecting initial cluster centers, assigning data points to the nearest cluster center, and updating the cluster centers. This step is repeated until a termination condition is met, such as when the change in cluster center is less than a certain threshold. In power line carrier communication systems, random data clustering can be used to group communication data into different clusters based on its characteristics (such as signal strength, noise level, and communication error rate). Each cluster represents a different communication mode or combination of features, which helps to understand and analyze complex patterns in communication data.

[0190] Based on the clustering results, the dataset is split into three different subsets: a training dataset, a validation dataset, and a test dataset. The training dataset is used to train the model, that is, to use the model to learn the characteristics and patterns of the data; the validation dataset is used to evaluate the performance of the model during the training process and help adjust the model parameters; and the test dataset is used to ultimately evaluate the model's generalization ability, that is, the model's performance on unseen data.

[0191] Based on the characteristics of the power line carrier communication system and analysis requirements, an appropriate deep learning model architecture, such as a convolutional neural network, recurrent neural network, or generative adversarial network, is selected. First, the network is trained using a training dataset, continuously attempting to learn the data's features and patterns and predict or classify new data points. During training, parameters are continuously adjusted to minimize prediction error. Next, a validation dataset is used to evaluate the model's performance to prevent overfitting, where the model performs well only on the training data but poorly on new or unseen data. By comparing the predictions from the validation dataset with the actual results, the network's parameters are adjusted to optimize model performance. Finally, a test dataset is used to evaluate the model's generalization ability. The test dataset contains unseen data points, allowing for a better assessment of the model's performance on new data. Leveraging big data, a trained and optimized deep learning network is constructed for signal processing and analysis in the power line carrier communication system. This network effectively identifies and processes complex patterns in communication data based on characteristics such as signal strength, noise level, and communication error rate, improving communication stability.

[0192] Furthermore, the present application also includes the following steps: performing data similarity screening on the basic data set and establishing a similar screening identifier; configuring a step window, performing in-window trend adaptation analysis on the data with similar screening identifiers in the basic data set through the step window, and generating a fitness ranking; eliminating data with similar screening identifiers according to the fitness ranking, and completing data preprocessing based on the data elimination results.

[0193] Specifically, similarity screening is performed on the underlying dataset to identify similar patterns or features within the dataset, such as data points with similar signal strengths, noise levels, or communication error rates. Similarity screening typically involves the use of various data processing and analysis techniques, such as cluster analysis and feature matching. Based on the results of similarity screening, labels are established for the filtered similar data points, indicating whether they are similar data. For example, a basic dataset containing a large amount of communication data can be used to identify data points with similar signal strengths and noise levels through data similarity screening, and these data points can be labeled, such as "high noise level group" or "low signal strength group."

[0194] A stepped window is a sliding window that slides across a dataset to analyze data trends within the window. The window size and shape can be configured based on the characteristics of the data and the analysis requirements. The dataset is divided into multiple fixed-size windows, each containing a certain number of data points. Trend adaptation analysis is performed on the data within each window to identify patterns and trends in the dataset over time, including fluctuations in signal strength, periodic changes in noise levels, and trends in communication error rates. By analyzing how data changes over time, long-term and short-term trends, as well as periodic or random variations, can be revealed. Trend adaptation analysis typically involves the use of various statistical and machine learning techniques, such as time series analysis and autoregressive models.

[0195] Based on the results of the trend adaptation analysis within the window, a fitness ranking is generated for each data point within the window. This ranking reflects the changing trends and patterns of the data within the window, helping the system identify and understand key features in the data. For example, data points are sorted based on the changing trends of signal strength to identify important communication patterns. Based on similarity screening indicators, data points with similar screening indicators are eliminated. Elimination is based on various reasons, such as data redundancy, data unreliability, or low data value. For example, data points with minimal changes in signal strength or small fluctuations in noise levels are eliminated.

[0196] The purpose of eliminating these similar data points is to optimize the dataset by removing potentially inaccurate or irrelevant data points. This helps reduce the size and complexity of the dataset and improve the efficiency of subsequent analysis. Based on the results of data elimination, data preprocessing is completed to obtain an optimized and filtered dataset containing more critical and useful data points. This ensures that the preprocessed dataset contains diverse and representative data while reducing redundancy and noise. By establishing a similarity screening representation and performing trend adaptation analysis on the base dataset, data points that do not match the trend within the window or have high similarity are eliminated, effectively preprocessing the base dataset, improving data quality, and preparing the data for deep learning model training.

[0197] Furthermore, the present application also includes the following steps: obtaining the data source of the basic data set, performing a trustworthy evaluation of the data based on the data source, and establishing a trustworthy evaluation ranking; performing a weighted analysis based on the trustworthy evaluation ranking and the fitness ranking, and eliminating data identified by similar screening based on the weighted analysis results.

[0198] Specifically, when collecting basic data, the source of each data point is usually recorded, including sensor data, network monitoring data, user behavior data, etc. Based on the data source, factors such as the data collection method, data consistency, and data update frequency are analyzed to conduct a credibility evaluation of the data, assess the data quality and reliability, and identify possible errors or deviations. If a data point comes from a data source that is considered more reliable, then its credibility will be higher. For example, sensor data is considered more reliable because it comes directly from the communication system, followed by network monitoring data, and user behavior data has lower reliability due to its subjectivity. Data points are sorted according to their credibility, with data points with higher credibility ranked first. For example, sensor data points are given a higher credibility score, while user behavior data points are given a lower score.

[0199] Weighted analysis combines trustworthiness ranking and fitness ranking to assign a weighted value to each data point based on its reliability and its trend adaptability within the window. Weighted analysis typically involves using various statistical and machine learning techniques, such as weighted averaging and weighted ranking, to comprehensively consider multiple aspects of the data. For example, sensor data points can be given higher weights because they are more reliable, while also considering the importance of signal strength trends. Similarity screening is performed on datasets to identify data points with similar characteristics or patterns, such as signal strength, noise levels, or communication error rates. Based on the results of the weighted analysis, data identified with similarity screening is eliminated, prioritizing those with low weighted analysis scores. For example, data points with low weighted analysis scores that are similar to other data points are eliminated to optimize the dataset. The goal of data elimination is to optimize the dataset by removing potentially inaccurate or irrelevant data points, thereby improving the efficiency and accuracy of subsequent analysis and model building. Weighted analysis can identify and eliminate inaccurate or irrelevant data points, thereby improving dataset quality.

[0200] Furthermore, the present application also includes the following steps: performing real-time processing monitoring on the power line carrier signal noise suppression processing to establish a monitoring and evaluation data set; performing performance monitoring and evaluation using the monitoring and evaluation data set to generate compensation data; and performing optimized deep learning network compensation processing using the compensation data.

[0201] Specifically, a real-time processing monitoring system is implemented, using various sensors, network monitoring devices, or software tools to continuously monitor the effectiveness of the power line carrier signal noise suppression processing. Data such as signal quality, noise level, and communication error rate are collected in real time to promptly identify signal processing issues, such as poor noise suppression and increased communication error rates. Based on the real-time monitoring results, a monitoring and evaluation dataset is established, consisting of processed signal data, raw signal data, noise data, and relevant performance indicators such as signal quality and communication error rate. Performance testing and evaluation are conducted using the monitoring and evaluation dataset to assess the effectiveness of the current noise suppression processing and compare it against pre-set performance standards. This includes calculating signal quality indices (such as signal-to-noise ratio), evaluating noise suppression efficiency, and analyzing communication error rates. Based on the test results, compensation data is generated to adjust noise suppression parameters, such as precoding parameters, adaptive spectrum, and subcarrier mapping. The purpose of this compensation data is to correct deficiencies in the noise suppression processing, improve signal quality, and reduce communication error rates. Next, the generated compensation data is applied to the optimized deep learning network, including adjusting network weights, changing algorithm parameters, and optimizing the model structure. The deep learning network is adjusted and optimized based on the compensation data, and the network is retrained to adapt to the new compensation parameters. Real-time monitoring of the effectiveness of noise suppression processing, generation of compensation data based on the monitoring results, and application of this data to adjust the deep learning network help the system adapt to the ever-changing communication environment.

[0202] In summary, the method for suppressing pulse noise in a power line carrier communication system provided by the present application establishes digital communication with the power line carrier communication system, obtains a communication data set within a preset time period based on the digital communication; extracts communication features based on the communication data set to establish a communication feature set; constructs a deep learning network through big data, configures an additional attention layer through the communication feature set, and optimizes the deep learning network through the additional attention layer; reads real-time communication data, configures precoding parameters, adaptive spectrum, and subcarrier mapping based on the real-time communication data according to the optimized deep learning network; and performs power line carrier signal noise suppression processing in the real-time communication data according to the configured precoding parameters, adaptive spectrum, and subcarrier mapping. In other words, by configuring the additional attention layer to optimize the deep learning network, reading real-time communication data, configuring precoding parameters, adaptive spectrum, and subcarrier mapping, and performing noise suppression processing, the technical effect of improving the signal quality of the power line carrier communication system is achieved.

[0203] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0204] In the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they may refer to fixed connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; and internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0205] In the present invention, unless otherwise expressly specified or limited, when a first feature is "above" or "below" a second feature, it may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium. Furthermore, when a first feature is "above," "above," or "above" a second feature, it may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is at a higher level than the second feature. When a first feature is "below," "below," or "below" a second feature, it may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is at a lower level than the second feature.

[0206] In the description of this specification, the terms "one embodiment", "some embodiments", "embodiments", "examples", "specific examples" or "some examples" refer to the specific features, structures, materials or characteristics described in conjunction with the embodiment or example and included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and features of different embodiments or examples, unless they are mutually inconsistent.

[0207] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may alter, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for suppressing impulse noise in a power line carrier communication system, characterized in that: The method is applied to a noise suppression system, and the method comprises: establishing digital communication with a power line carrier communication system, and acquiring a communication data set within a preset time period based on the digital communication; The communication data set includes data points collected at a preset collection frequency within a preset time period; each data point includes multiple features; establishing a communication feature set based on the communication data set; Optimizing a preset deep learning network model using the communication feature set to obtain an optimized deep learning network model; The deep learning network model includes a deep learning network and an additional attention layer embedded in the deep learning network for optimizing the deep learning network during the training process of the deep learning network model; Read the current data point and input it into the optimized deep learning network model to obtain the predicted configuration parameters; The configuration parameters include precoding parameters, adaptive spectrum and subcarrier mapping; performing noise suppression processing on the power line carrier signal in the current data point according to the predicted configuration parameters; The establishing of the communication feature set includes: Preprocessing the data points in the communication data set to obtain a preprocessed communication data set; Based on the features in the preprocessed communication dataset, obtain the corresponding covariance matrix; Performing eigenvalue decomposition on the covariance matrix to obtain a plurality of data groups, wherein each data group includes an eigenvalue and its corresponding eigenvector; Selecting eigenvectors corresponding to N eigenvalues ​​in a plurality of data groups according to a pre-set strategy, and projecting each data point in the preprocessed communication data set onto each selected eigenvector to obtain a dimensionality-reduced data representation corresponding to the data point; The dimensionality-reduced data representations corresponding to all data points in the preprocessed communication data set are used to form a communication feature set; The preprocessed communication data set includes: Screening out data points whose time series attenuation coefficients meet a preset threshold value from the communication data set to form an initial communication data set; Screening out data points that meet similar conditions in the initial communication data set and marking them with screening identifiers; Using a stepped window, trend adaptation analysis is performed on the data points with screening marks in the initial communication data set, so that each data point with screening marks is given a corresponding fitness score; Filtering data points with a filter identifier in the initial communication data set to form a first data point set, wherein the data points have a fitness score greater than a preset fitness value; Obtaining a comprehensive score corresponding to each data point in the first data point set according to the fitness score and a preset credibility score corresponding to the data point; The data points in the first data point set whose comprehensive scores are greater than a predetermined threshold and the preset configuration parameters corresponding to the data points are combined into a pre-processed communication data set; The fitness score is calculated using the following formula: ； ； Among them, the fitness score of the i-th data point with the screening mark and its corresponding collection time are and ; The true value and predicted value of the signal intensity in the i-th data point with the screening mark are and ; The true value and predicted value of the noise level in the i-th data point with the screening mark are and ; The true value and predicted value of the signal-to-noise ratio in the i-th data point with the screening mark are and ; is the preset maximum deviation value of signal strength; is the preset maximum deviation value of the noise level; is the preset maximum deviation value of the signal-to-noise ratio; is the time position weight, which indicates the importance of the i-th data point with the screening mark in the step window; is the center time point of the step window; is a preset parameter value.

2. The method for suppressing impulse noise in a power line carrier communication system according to claim 1, wherein: in, The deep learning network model is obtained by embedding an additional attention layer between the feature extraction layer and the fully connected layer in the convolutional neural network; The attention addition layer adopts any one of the self-attention mechanism, the soft attention mechanism and the hard attention mechanism.

Images