A carrier noise identification method and system including a high proportion of power electronics operation

Abstract

The application discloses a carrier noise identification method and system containing a high proportion of power electronic operation, acquires power line carrier test noise data, extracts features of the power line carrier test noise data by using a self-organizing mapping neural network, obtains multidimensional feature data, calculates multidimensional noise feature approximation residuals by using the multidimensional feature data, determines the number of active neurons of the self-organizing mapping neural network according to the multidimensional noise feature approximation residuals, clusters and matches the multidimensional feature data and a multidimensional noise feature library according to the number of active neurons, and obtains a noise identification result, wherein the multidimensional noise feature library is constructed by extracting features of power line carrier test noise sample data by using a super-natural spiral evolution method, and the method improves the power line noise identification speed by dynamically improving parameters in the self-organizing mapping neural network by using multidimensional noise feature approximation residual result calculation.

Description

Technical Field

[0001] This invention relates to the field of power grid noise identification technology, and in particular to a carrier noise identification method and system that incorporates a high proportion of power electronic operation. Background Technology

[0002] In response to the national "dual-carbon" strategic goal, the construction of new power systems is accelerating, placing higher demands on their communication capabilities. Power line carrier communication (PLC) offers advantages such as wide coverage and low cost, playing a crucial role in supporting the construction of power communication networks. However, new power systems exhibit "dual-high" characteristics: a high proportion of power electronic equipment connected to the grid creates a harsh electromagnetic environment for PLC communication, making the noise environment of PLC channels more complex and exacerbating signal attenuation. This severely restricts the improvement of reliability and transmission speed, significantly reducing communication quality.

[0003] Power line noise identification (PLN) technology can effectively identify the complex noise characteristics of power lines, assess the on-site noise environment, and thus achieve accurate noise reduction, improving the reliability and transmission speed of power line communication and enhancing communication quality. However, traditional PSN technology neglects the extraction of power line carrier noise data features from multiple dimensions of power electronic operating environments, including time, frequency, and energy domains. It cannot select the most representative noise features to construct a noise identification dictionary, resulting in poor accuracy in PSN identification. Furthermore, the noise identification network structure used is complex and converges slowly, making it difficult to achieve rapid network convergence and rapid noise identification, thus slowing down the PSN identification speed. Summary of the Invention

[0004] To address the aforementioned technical problems, embodiments of the present invention provide a carrier noise identification method and system incorporating a high proportion of power electronic operation. By utilizing multidimensional noise features to approximate residual results and dynamically improving the parameters in a self-organizing mapping neural network, the speed of power line noise identification is increased.

[0005] A first aspect of the present invention provides a carrier noise identification method for a system in which a high proportion of power electronic operation is incorporated, the method comprising:

[0006] Collect power line carrier test noise data;

[0007] Power line carrier test noise data is input into a pre-constructed self-organizing map neural network (SOMN) to extract features from the power line carrier test noise data, obtaining multi-dimensional feature data. Multi-dimensional noise feature approximation residuals are then calculated using these multi-dimensional feature data. The number of active neurons in the SOMN is then determined based on these multi-dimensional noise feature approximation residuals. Clustering matching of the multi-dimensional feature data and the multi-dimensional noise feature library is then performed based on the number of active neurons to obtain the noise identification result. The multi-dimensional noise feature library is constructed by extracting features from the power line carrier test noise sample data using a supernatural spiral evolution method.

[0008] In this embodiment, power line carrier test noise data is collected, and features are extracted from the power line carrier test noise data using a self-organizing map neural network to obtain multi-dimensional feature data. Multi-dimensional noise feature approximation residuals are calculated using these multi-dimensional feature data, and the number of active neurons in the self-organizing map neural network is determined based on these residuals. Clustering matching is then performed on the multi-dimensional feature data and the multi-dimensional noise feature library based on the number of active neurons to obtain the noise identification result. The multi-dimensional noise feature library is constructed by extracting features from the power line carrier test noise sample data using a supernatural spiral evolution method. This method improves the speed of power line noise identification by dynamically improving the parameters in the self-organizing map neural network using the multi-dimensional noise feature approximation residual results.

[0009] In one possible implementation of the first aspect, the multidimensional noise feature library is constructed by extracting features from sample power line carrier test noise data using a supernatural spiral evolution method, specifically as follows:

[0010] Noise sample data of power line carrier test is collected, and feature extraction is performed on the power line carrier test noise sample data to obtain noise features, which include original noise features and noise-related features.

[0011] The noise features are screened to obtain typical noise features, and a multidimensional noise feature library is constructed based on these typical noise features.

[0012] In one possible implementation of the first aspect, noise features are screened to obtain typical noise features, and a multidimensional noise feature library is constructed based on these typical noise features, specifically as follows:

[0013] Initialize the location of each noise feature;

[0014] The contribution of each noise feature is iteratively calculated. The position of each noise feature is updated according to its contribution, and the contribution of each noise feature is updated. This process is repeated until the number of iterations is reached, and the final contribution of each noise feature is obtained.

[0015] The noise features are ranked according to their final contribution, and the noise feature with the largest contribution is selected as the typical noise feature. A multidimensional noise feature library is then constructed based on the typical noise feature.

[0016] In one possible implementation of the first aspect, the positions of each noise feature are updated according to its contribution, resulting in the updated positions of each noise feature, specifically as follows:

[0017] The noise features whose contribution is greater than a preset value within a preset radius are aggregated to obtain an adjacency set.

[0018] Calculate the movement probability of each noise feature to noise features in its adjacent set. The formula for calculating the movement probability is as follows:

[0019]

[0020] Among them, prob nk (t) represents the probability, c k (t) represents the contribution of noise feature k, c n (t) represents the contribution of noise feature n;

[0021] The target moving noise features are determined based on the movement probability, and the Euclidean distance between each noise feature and the target moving noise feature corresponding to each noise feature is calculated.

[0022] The updated positions of each noise feature are obtained by multiplying the spiral shape parameter and the Euclidean distance. The formula for calculating the updated positions is as follows:

[0023]

[0024] In the formula, n represents the noise characteristic, k represents the target movement noise characteristic, and x n (t) and x k (t) represents the position of the noise feature and the target movement noise feature at iteration t, respectively. n (t+1) and x k (t+1) represents the position of the noise feature and the target movement noise feature at iteration t+1, respectively. |x k (t)-x n (t) represents the distance between the noise feature and the other target noise features, where ||xk (t)-x n (t)|| is the norm of the distance between the noise feature and the other target noise features, σ is the preset logarithmic spiral shape parameter, v is a random number between [-1,1], and t is the iteration number of the current iteration.

[0025] In this embodiment, noise features with a contribution greater than a preset value within a preset radius are aggregated to obtain an adjacency set. The movement probability of each noise feature to a noise feature in the adjacency set is then calculated. Based on the movement probability, the target moving noise feature of each noise feature is determined. The Euclidean distance between each noise feature and its corresponding target moving noise feature is calculated. The updated position of each noise feature is obtained by multiplying the spiral shape parameter and the Euclidean distance. This method compares the contribution of each noise feature with the contribution of other noise features within the detection radius using a supernatural spiral evolution method. The spiral evolution method enables each supernatural individual to perform an efficient global search in a spiral manner, selects the most representative noise feature, constructs a multidimensional noise library for power line carriers, improves the adaptability of noise feature recognition to the multidimensional noise feature library for power lines, and achieves accurate identification of power line noise categories.

[0026] In one possible implementation of the first aspect, the contribution of each noise feature is updated to obtain the updated contribution, specifically as follows:

[0027] The importance function of each noise feature is transformed into the contribution value of the typical noise feature, resulting in the updated contribution value. The transformation process is as follows:

[0028]

[0029] In the formula, c n (t) represents the contribution of typical noise characteristics, D(x) n (t) represents the importance function of the noise feature, λ represents the enhancement coefficient of the typical noise feature, η represents the attenuation degree of the contribution of the typical noise feature, and t represents the number of iterations in the tth iteration.

[0030] In one possible implementation of the first aspect, a self-organizing map neural network extracts features from power line carrier test noise data to obtain multidimensional feature data. The multidimensional noise feature approximation residual is then calculated using this multidimensional feature data. Finally, the number of active neurons in the self-organizing map neural network is determined based on the multidimensional noise feature approximation residual. Specifically:

[0031] Feature extraction is performed on power line carrier noise data to obtain multidimensional features. These multidimensional features are then compared with typical noise features in a multidimensional noise feature library to obtain multidimensional noise feature approximation residuals. The formula for calculating the multidimensional noise feature approximation residuals is as follows:

[0032]

[0033] In the formula, I represents the total number of input noise feature data, e represents the approximation residual of the i-th input noise feature data, and z represents the total number of input noise feature data. i For the i-th input noise data, the multidimensional feature value is... This is the estimated multidimensional feature value of the i-th input noise data in the multidimensional noise feature library;

[0034] The residual approximated by the multidimensional noise feature is compared with the preset multidimensional noise feature approximation residual to obtain the residual difference. The number of active neurons is then calculated based on the residual difference. The formula for calculating the number of active neurons is as follows:

[0035]

[0036] In the formula, e represents the residual approximating the multidimensional noise feature, e min This indicates that the preset multidimensional noise features approximate the residual.

[0037] In one possible implementation of the first aspect, clustering matching is performed on the multidimensional feature data and the multidimensional noise feature library based on the number of active neurons to obtain the noise recognition result, specifically as follows:

[0038] The multidimensional features and weight vectors are normalized to obtain normalized multidimensional features and weight vectors. The inner product of the normalized multidimensional features and weight vectors is then calculated, with the formula for the inner product being:

[0039]

[0040] in, This represents the normalized multidimensional features. d represents the normalized weight vector. o (p) denotes the inner product;

[0041] The neurons in the self-organizing map neural network are sorted according to their inner product, and the sorting results are obtained. Neurons with an inner product greater than a preset value are selected as active neurons from the sorting results.

[0042] Active neighborhoods are determined based on active neurons, and the weights within the active neighborhoods are adjusted based on the deviation between the normalized multidimensional features and the normalized weight vector to obtain the adjusted weights. The feature data are then clustered using the adjusted weights to obtain the clustering results.

[0043] Calculate the correlation between the clustering results and typical noise features, and select the clustering result with the highest correlation as the noise identification result.

[0044] In one possible implementation of the first aspect, the constructed self-organizing map neural network includes an input layer and an output layer. The input layer corresponds to a high-dimensional input vector, and the output layer consists of a series of ordered nodes organized on a two-dimensional grid. The input nodes and the output nodes are connected by a weight vector.

[0045] A second aspect of the present invention provides a carrier noise identification system incorporating a high proportion of power electronic operation, the system comprising:

[0046] The acquisition module is used to acquire power line carrier test noise data;

[0047] The feature extraction module is used to input power line carrier test noise data into a pre-constructed self-organizing map neural network (SOMN). The SOMN then extracts features from the power line carrier test noise data to obtain multidimensional feature data. Multidimensional noise feature approximation residuals are calculated using these multidimensional feature data. The number of active neurons in the SOMN is then determined based on these residuals. Clustering matching of the multidimensional feature data and the multidimensional noise feature library is performed based on the number of active neurons to obtain the noise identification result. The multidimensional noise feature library is constructed by extracting features from the power line carrier test noise sample data using a supernatural spiral evolution method.

[0048] In one possible implementation of the second aspect, the multidimensional noise feature library is constructed by extracting features from sample power line carrier test noise data using a supernatural spiral evolution method, specifically as follows:

[0049] Noise sample data of power line carrier test is collected, and feature extraction is performed on the power line carrier test noise sample data to obtain noise features, which include original noise features and noise-related features.

[0050] The noise features are screened to obtain typical noise features, and a multidimensional noise feature library is constructed based on these typical noise features. Attached Figure Description

[0051] Figure 1 : A flowchart illustrating an embodiment of the carrier noise identification method with a high proportion of power electronic operation provided by the present invention;

[0052] Figure 2 : A schematic diagram of the identification process of an embodiment of the carrier noise identification method with a high proportion of power electronic operation provided by the present invention;

[0053] Figure 3 : A schematic diagram of the system structure of another embodiment of the carrier noise identification method with a high proportion of power electronic operation provided by the present invention;

[0054] Figure 4 : A schematic diagram of the specific system structure of another embodiment of the carrier noise identification method with a high proportion of power electronic operation provided by the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Example 1

[0057] Please refer to Figure 1 This is a flowchart illustrating an embodiment of a carrier noise identification method with a high proportion of power electronic operation provided by the present invention, including steps S11 to S12, each step of which is as follows:

[0058] S11. Collect power line carrier test noise data.

[0059] In this embodiment, based on the real-time monitoring of carrier noise involving a high proportion of power electronics operation, a large amount of noise with distinct characteristics is extracted during a specific time period.

[0060] S12. Input the power line carrier test noise data into the constructed self-organizing map neural network so that the self-organizing map neural network can extract features from the power line carrier test noise data to obtain multi-dimensional feature data. Calculate the multi-dimensional noise feature approximation residual using the multi-dimensional feature data, and then determine the number of active neurons in the self-organizing map neural network based on the multi-dimensional noise feature approximation residual. Perform cluster matching on the multi-dimensional feature data and the multi-dimensional noise feature library based on the number of active neurons to obtain the noise identification result. The multi-dimensional noise feature library is constructed by extracting features from the power line carrier test noise sample data using the supernatural spiral evolution method.

[0061] In a preferred embodiment, the multidimensional noise feature library is constructed by extracting features from sample power line carrier test noise data using a supernatural spiral evolution method, specifically as follows:

[0062] Noise sample data of power line carrier test is collected, and feature extraction is performed on the power line carrier test noise sample data to obtain noise features, which include original noise features and noise-related features.

[0063] The noise features are screened to obtain typical noise features, and a multidimensional noise feature library is constructed based on these typical noise features.

[0064] In a preferred embodiment, noise features are screened to obtain typical noise features, and a multidimensional noise feature library is constructed based on these typical noise features, specifically as follows:

[0065] Initialize the location of each noise feature;

[0066] The contribution of each noise feature is iteratively calculated. The position of each noise feature is updated according to its contribution, and the contribution of each noise feature is updated. This process is repeated until the number of iterations is reached, and the final contribution of each noise feature is obtained.

[0067] The noise features are ranked according to their final contribution, and the noise feature with the largest contribution is selected as the typical noise feature. A multidimensional noise feature library is then constructed based on the typical noise feature.

[0068] In a preferred embodiment, the positions of each noise feature are updated according to its contribution, resulting in the updated positions of each noise feature, specifically as follows:

[0069] The noise features whose contribution is greater than a preset value within a preset radius are aggregated to obtain an adjacency set.

[0070] Calculate the movement probability of each noise feature to noise features in its adjacent set. The formula for calculating the movement probability is as follows:

[0071]

[0072] Among them, prob nk (t) represents the probability, c k (t) represents the contribution of noise feature k, c n (t) represents the contribution of noise feature n;

[0073] The target moving noise features are determined based on the movement probability, and the Euclidean distance between each noise feature and the target moving noise feature corresponding to each noise feature is calculated.

[0074] The updated positions of each noise feature are obtained by multiplying the spiral shape parameter and the Euclidean distance. The formula for calculating the updated positions is as follows:

[0075]

[0076] In the formula, n represents the noise characteristic, k represents the target movement noise characteristic, and x n (t) and x k (t) represents the position of the noise feature and the target movement noise feature at iteration t, respectively.n (t+1) and x k (t+1) represents the position of the noise feature and the target movement noise feature at iteration t+1, respectively. |x k (t)-x n (t) represents the distance between the noise feature and the other target noise features, where ||x k (t)-x n (t)|| is the norm of the distance between the noise feature and the other target noise features, σ is the preset logarithmic spiral shape parameter, v is a random number between [-1,1], and t is the iteration number of the current iteration.

[0077] In a preferred embodiment, the contribution of each noise feature is updated to obtain the updated contribution, specifically as follows:

[0078] The importance function of each noise feature is transformed into the contribution value of the typical noise feature, resulting in the updated contribution value. The transformation process is as follows:

[0079]

[0080] In the formula, c n (t) represents the contribution of typical noise characteristics, D(x) n (t) represents the importance function of the noise feature, λ represents the enhancement coefficient of the typical noise feature, η represents the attenuation degree of the contribution of the typical noise feature, and t represents the number of iterations in the tth iteration.

[0081] In this embodiment, as Figure 2 As shown, the carrier noise situation with a high proportion of power electronics operation is monitored in real time, and a large amount of noise with obvious characteristics in a specific time period is extracted as test noise, providing data for the subsequent construction of a multi-dimensional noise feature library.

[0082] The multidimensional noise feature library contains raw test noise data, raw noise features, noise correlation features, and noise types. Raw noise features characterize the inherent properties of the noise signal itself, including: sample mean square error (MSE) describing the dispersion of the noise data's average value; sample skewness (SK) describing the direction and degree of skewness in the noise data distribution; and approximate entropy describing the regularity of the noise. Noise correlation features characterize the correlation between different noise signals, including: standard deviation of the correlation between different noise signals; Pearson correlation coefficient reflecting the direction of linear relationship and correlation between signals from different channels; and approximate entropy of the sum / difference of different noise signals. This multidimensional noise feature library provides noise type query and identification capabilities for the subsequent training process of multivariate dynamic active improved self-organizing map neural networks, enabling rapid noise identification.

[0083] The supernatural spiral evolution method is used to select the aforementioned original noise features and noise-related features. First, based on the firefly algorithm, the supernatural spiral evolution method moves towards a better position by comparing the contribution of each noise feature with the contribution of other noise features within the detection radius, eventually converging to one or more extreme points to select typical noise features. Second, the spiral evolution method allows each supernatural individual to perform a global search in a spiral manner, enabling each individual to search more space, enhancing the global search capability, and thus improving search efficiency. Based on the above typical noise feature selection method based on the supernatural spiral evolution method, the required typical noise features can be obtained, laying a solid foundation for the rapid identification of power line communication noise. The specific steps are as follows:

[0084] Step 1: Initialization of the Supernatural Individual Group. The supernatural individual group contains all the aforementioned noise feature data, with each supernatural individual representing a noise feature data point. The positions of each supernatural individual are randomly initialized as X = {x1, x2, ..., x...}. n ,...,x N}, where x n Let N be the position of the nth supernatural individual, and N be the number of supernatural individuals. Each supernatural individual can be considered as the independent variable of the noise feature importance function. Each supernatural individual is assigned a different noise feature importance based on its position. As the iteration progresses, all supernatural individuals will converge to the extreme point of the objective function. The following parameters are defined: initial noise typical feature contribution c0, noise typical feature contribution attenuation η, noise typical feature enhancement coefficient λ, supernatural individual group size o, and preset maximum number of iterations S for supernatural spiral evolution. max Preset detection radius r d Candidate radius r w The candidate domain update rate variable α, and the threshold number of adjacent supernatural individuals z. thr It should be noted that the more typical the noise feature, the greater its contribution to the noise's typicality.

[0085] Step 2: Noise Feature Location Update. First, define the set of supernatural spiral evolution iterations T = {1,2,...,t,...,T}, and select the radius of the noise feature n for the t-th iteration. All contents that conform to c n (t)<c k The noise features required by (t) are aggregated into an adjacency set H. n (t). The candidate radius determines the size of the candidate domain, and each supernatural individual will only seek individuals within the candidate domain whose contribution to the noise typical features is greater than its own; at the same time, the candidate radius cannot be greater than the preset maximum detection radius r. dBased on this, the probability probability (prob) of the noise feature n moving to other noise features k in the adjacent region during the t-th iteration is calculated. nk (t), represented as:

[0086]

[0087] In the formula, prob nk (t) represents the probability of moving, c k (t) represents the contribution of noise feature k, c n (t) represents the contribution of noise feature n;

[0088] Finally, based on the obtained movement probabilities, a roulette wheel method is used to determine the target individual k; that is, the higher the movement probability, the more likely it is to become the target individual. After the target individual k is determined, the noise feature positions are updated by multiplying the Euclidean distance between the two supernatural individuals by the spiral shape parameter, as shown in the following formula:

[0089]

[0090] In the formula, n represents the current supernatural individual, k represents the target supernatural individual, and x n (t) and x k (t) represents the positions of noise features n and k in iteration t, respectively, while x n (t+1) and x k (t+1) represents the positions of noise features n and k in iteration t+1, respectively, |x k (t)-x n (t) represents the distance between the current noise feature and the remaining target noise features, ||x k (t)-x n (t)|| represents the norm of the distance between the current noise feature and the other target noise features, σ is the preset logarithmic spiral shape parameter, v is a random number between [-1,1], and t is the iteration number of the current iteration.

[0091] During successive iterations, the step size of the position update decreases as the number of iterations increases, which solves the problem of oscillation of supernatural individuals near the extreme point. At the same time, by adding a logarithmic spiral morphological parameter, the algorithm can make each supernatural individual no longer adopt a linear movement strategy, but gradually spiral to converge to the global extreme point.

[0092] Step 3: Update the contribution of typical noise features. The noise feature importance function D(x) of the typical noise feature n in the t-th iteration is then updated. n (t) is converted into the contribution of typical noise characteristics c. n (t), represented as:

[0093]

[0094] In the formula, c n (t) represents the contribution of typical noise characteristics, D(x) n (t) represents the importance function of the noise feature, λ represents the enhancement coefficient of the typical noise feature, η represents the attenuation degree of the contribution of the typical noise feature, and t represents the number of iterations in the tth iteration.

[0095] Here, the contribution of typical noise features consists of two parts: the importance function of noise features and the attenuation of the contribution of typical noise features in history. This fully simulates the phototaxis of fireflies in nature and the absorption and attenuation effect of the natural environment on the light intensity emitted by fireflies.

[0096] Repeat the above steps to perform updates and iterations until the maximum number of iterations S is reached. max At this point, the typical noise features are sorted from largest to smallest according to their contribution. The N typical noise features with the largest contribution are selected as the input to the multivariate dynamic active improved self-organizing map neural network. At the same time, the selected N typical noise features are added to the multidimensional noise feature library.

[0097] In a preferred embodiment, the self-organizing map neural network extracts features from power line carrier test noise data to obtain multidimensional feature data. The multidimensional feature data is then used to calculate a multidimensional noise feature approximation residual. Finally, the number of active neurons in the self-organizing map neural network is determined based on the multidimensional noise feature approximation residual. Specifically:

[0098] Feature extraction is performed on power line carrier noise data to obtain multidimensional features. These multidimensional features are then compared with typical noise features in a multidimensional noise feature library to obtain multidimensional noise feature approximation residuals. The formula for calculating the multidimensional noise feature approximation residuals is as follows:

[0099]

[0100] In the formula, I represents the total number of input noise feature data, e represents the approximation residual of the i-th input noise feature data, and z represents the total number of input noise feature data. i For the i-th input noise data, the multidimensional feature value is... This is the estimated multidimensional feature value of the i-th input noise data in the multidimensional noise feature library;

[0101] The residual approximated by the multidimensional noise feature is compared with the preset multidimensional noise feature approximation residual to obtain the residual difference. The number of active neurons is then calculated based on the residual difference. The formula for calculating the number of active neurons is as follows:

[0102]

[0103] In the formula, e represents the residual approximating the multidimensional noise feature, emin This indicates that the preset multidimensional noise features approximate the residual.

[0104] In a preferred embodiment, clustering matching is performed on the multidimensional feature data and the multidimensional noise feature library based on the number of active neurons to obtain the noise recognition result, specifically as follows:

[0105] The multidimensional features and weight vectors are normalized to obtain normalized multidimensional features and weight vectors. The inner product of the normalized multidimensional features and weight vectors is then calculated, with the formula for the inner product being:

[0106]

[0107] in, This represents the normalized multidimensional features. d represents the normalized weight vector. o (p) denotes the inner product;

[0108] The neurons in the self-organizing map neural network are sorted according to their inner product, and the sorting results are obtained. Neurons with an inner product greater than a preset value are selected as active neurons from the sorting results.

[0109] Active neighborhoods are determined based on active neurons, and the weights within the active neighborhoods are adjusted based on the deviation between the normalized multidimensional features and the normalized weight vector to obtain the adjusted weights. The feature data are then clustered using the adjusted weights to obtain the clustering results.

[0110] Calculate the correlation between the clustering results and typical noise features, and select the clustering result with the highest correlation as the noise identification result.

[0111] In a preferred embodiment, the constructed self-organizing map neural network includes an input layer and an output layer. The input layer corresponds to a high-dimensional input vector, and the output layer consists of a series of ordered nodes organized on a two-dimensional grid. The input nodes and the output nodes are connected by a weight vector.

[0112] In this embodiment, a self-organizing map neural network improved by a multivariate dynamic active method is used to quickly and automatically cluster power line noise feature data. The network's mapping learning rules and competitive transfer function are dynamically improved based on the residual results approximating the multidimensional noise features. This allows the number of active neurons selected each time to be dynamically adjusted according to the actual situation, achieving rapid network identification and rapid convergence of noise clustering. The specific steps are as follows:

[0113] (1) Construction of a multivariate dynamic active improved self-organizing map neural network model. The noise feature data set is defined as Z = {z1, z2, ..., z...}. i ,...,z I}, where z i Given an N-dimensional vector, a multivariate dynamically active improved self-organizing map neural network suitable for fast classification and identification of power line carrier noise feature data includes an input layer and an output layer. The input layer corresponds to a high-dimensional input vector, and the output layer consists of a series of ordered nodes organized on a 2D grid. Input and output nodes are connected by weight vectors. The number of neurons in the input layer is I, and the number of neurons in the output layer is O. The maximum number of iterations for training the multivariate dynamically active improved self-organizing map neural network is defined as P. max The weights of each node in the output layer of the neural network in the p-th iteration are W. o (p), where the connection weight between input layer node i and output layer node o is w. io (p).

[0114] (2) Initialization. Initialize the connection weights of the input and output layers, assigning them a small connection weight. Define the initial connection weight as W. o (0), establish the initial active neighborhood The initial learning rate is preset to η(0).

[0115] (3) Calculation of Multidimensional Noise Feature Approximation Residual. Multidimensional features of the carrier noise data to be input into the neural network are extracted and compared with existing multidimensional noise features in the multidimensional noise feature library. The multidimensional noise feature approximation residual *e* is calculated, which is expressed as:

[0116]

[0117] In the formula, I represents the total number of input noise feature data, e represents the approximation residual of the i-th input noise feature data, and z represents the total number of input noise feature data. i For the i-th input noise data, the multidimensional feature value is... is the estimated value of the multidimensional feature corresponding to the i-th input noise data in the multidimensional noise feature library.

[0118] (4) Calculation of the number of active neurons based on the multidimensional noise feature approximation residual. The multidimensional noise feature approximation residual e is compared with a preset multidimensional noise feature approximation residual threshold e. min The number of active neurons, Θ, is calculated by comparing the residuals. A larger residual indicates more characteristics of the input noise, requiring more active neurons for clustering and representation. The formula for calculating the number of active neurons, Θ, is as follows:

[0119]

[0120] In the formula, e represents the residual approximating the multidimensional noise feature, e min This indicates that the preset multidimensional noise features approximate the residual.

[0121] (5) Selection of active neurons. The carrier noise feature data and weight vector to be input into the neural network are normalized to obtain normalized carrier noise data. With weight vector And calculate noise data With weight vector The inner product of the input carrier noise vector and the weight vector, when both have a magnitude of 1, results in a larger inner product, a smaller Euclidean distance between the two points, and thus a greater similarity. (Carrier noise data) With weight vector The inner product d o (p) The calculation formula is:

[0122]

[0123] in, This represents the normalized multidimensional features. d represents the normalized weight vector. o (p) denotes the inner product;

[0124] After obtaining carrier noise data With weight vector The inner product d o (p) After that, the output layer neurons of the multivariate dynamic active improved self-organizing map neural network are sorted according to the inner product. Based on the sorting result, the output layer neurons with the largest inner product Θ are selected as active neurons, where Θ is a preset fixed value.

[0125] (6) Using Θ active neurons as the central nodes, determine the active neighborhood, which is the adjustment domain of the neuron weights, and gradually reduce its active neighborhood as the iteration increases.

[0126] (7) Based on the active neighborhood, based on carrier noise data With weight vector The gradient of the deviation adjusts the weights of each node in the neighborhood. Simultaneously, the larger the residual, the lower the learning rate, thereby improving the training accuracy of the multivariate dynamic active self-organizing map neural network and avoiding overfitting. Furthermore, since the difference between nodes in the inactive neighborhood and the center node is too large, to reduce the computational cost of the weight update process and improve training speed, the weights of nodes in the inactive neighborhood remain unchanged. The weight update formula is expressed as:

[0127]

[0128] In the formula, Represents the weight vector. W represents carrier noise data. o (p+1) represents the updated weight vector.

[0129] (8) When the preset maximum number of iterations P is reached max The clustering results of the noise feature data were then obtained.

[0130] (9) Rapid Noise Identification and Classification. The correlation between the clustering results of noise feature data and different power line noise distribution types is calculated, and the result with the highest correlation is selected as the final noise identification and classification result. Because this invention dynamically improves the network's mapping learning rules and competitive transmission functions based on the residual results approximating multidimensional noise features, the number of active neurons selected each time can be dynamically adjusted according to the actual situation. Multiple clustering results can be obtained simultaneously, effectively improving the clustering speed of the multivariate dynamically active improved self-organizing mapping neural network. Furthermore, different power line noise features are already present in the carrier multidimensional noise feature library, allowing for rapid noise identification by consulting the carrier multidimensional noise feature library.

[0131] This invention calculates the number of active neurons in an improved self-organizing map neural network (SAMR) based on the residual results approximated by multidimensional noise features, and performs multivariate dynamic active SAMR output layer neuron ranking based on the inner product. Based on the ranking results, it selects active neurons in the SAMR output layer, thereby enabling rapid automatic clustering of power line noise feature data. Simultaneously, during SAMR training, the learning rate and weight parameters of the network are dynamically improved using multidimensional noise feature approximation residual results, thereby increasing the recognition speed and noise convergence speed of the power line noise identification network and achieving rapid identification of power line noise categories.

[0132] Example 2

[0133] Accordingly, see Figure 3 , Figure 3 This invention provides a carrier noise identification system incorporating a high proportion of power electronic operation, as shown in the figure. The carrier noise identification system incorporating a high proportion of power electronic operation includes:

[0134] Acquisition module 301 is used to acquire power line carrier test noise data;

[0135] The feature extraction module 302 is used to input power line carrier test noise data into the constructed self-organizing map neural network, so that the self-organizing map neural network can extract features from the power line carrier test noise data to obtain multi-dimensional feature data. The multi-dimensional noise feature approximation residual is calculated using the multi-dimensional feature data, and the number of active neurons in the self-organizing map neural network is determined based on the multi-dimensional noise feature approximation residual. The multi-dimensional feature data and the multi-dimensional noise feature library are clustered and matched based on the number of active neurons to obtain the noise recognition result. The multi-dimensional noise feature library is constructed by extracting features from the power line carrier test noise sample data using the supernatural spiral evolution method.

[0136] In a preferred embodiment, the multidimensional noise feature library is constructed by extracting features from sample power line carrier test noise data using a supernatural spiral evolution method, specifically as follows:

[0137] Noise sample data of power line carrier test is collected, and feature extraction is performed on the power line carrier test noise sample data to obtain noise features, which include original noise features and noise-related features.

[0138] The noise features are screened to obtain typical noise features, and a multidimensional noise feature library is constructed based on these typical noise features.

[0139] As an example of this embodiment, such as Figure 4 As shown, the carrier noise rapid identification system with a high proportion of power electronic operation proposed in this invention includes: a communication module, a power supply module, a storage module, a multi-dimensional data noise feature extraction module, a multi-dimensional feature noise database construction module, and a power line communication noise rapid identification module, which are described in detail below:

[0140] Communication module: The communication module is responsible for communicating with various power terminals via power line carrier. It receives the carrier signals transmitted by the terminals and extracts a large amount of noise with obvious characteristics as test noise. Alternatively, it receives the carrier noise data identification and resolution results generated by the power line communication noise rapid identification module and sends them back to the power terminal for further noise reduction processing.

[0141] Power supply module: The power supply module is responsible for providing a stable voltage level to each module in the carrier noise rapid identification device, which contains a high proportion of power electronic operation, and ensuring the normal operation of the system circuit.

[0142] Storage Module: The data storage unit mainly consists of flash memory and random access memory (RAM). The RAM serves as the memory module for the carrier noise rapid identification device, providing the foundation for the high-speed operation of noise feature extraction and noise rapid identification programs. The flash memory, as the long-term storage module for device data, provides sufficient storage space for the acquired noise signal data.

[0143] Multidimensional data noise feature extraction module: The multidimensional data noise feature extraction module is responsible for extracting power line carrier noise data features from different dimensions such as time domain, frequency domain, and energy domain under the power electronic operating environment.

[0144] Multidimensional feature noise library construction module: Based on the supernatural spiral evolution method, the most representative noise features are selected and the extracted typical noise features are stored for subsequent rapid identification of power line communication noise based on multivariate dynamic active improved self-organizing map neural network.

[0145] Power line communication noise rapid identification module: The power line communication noise rapid identification module uses a multivariate dynamic active improved self-organizing map neural network to cluster and match the received noise signal with the noise feature information in the carrier multidimensional noise feature library. Finally, it performs classification judgment according to the preset noise feature-noise type correlation index to achieve rapid identification of carrier noise type.

[0146] The aforementioned carrier noise identification system incorporating a high proportion of power electronic operation can implement one of the carrier noise identification methods incorporating a high proportion of power electronic operation described in the above-described method embodiments. The options in the above method embodiments are also applicable to this embodiment and will not be detailed here. The remaining content of this application's embodiments can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0147] In summary, implementing the embodiments of the present invention has the following beneficial effects:

[0148] Power line carrier test noise data is collected, and features are extracted from the power line carrier test noise data using a self-organizing map neural network to obtain multidimensional feature data. Multidimensional noise feature approximation residuals are calculated using these multidimensional feature data, and the number of active neurons in the self-organizing map neural network is determined based on these residuals. Clustering matching is then performed on the multidimensional feature data and the multidimensional noise feature library based on the number of active neurons to obtain the noise identification result. The multidimensional noise feature library is constructed by extracting features from power line carrier test noise sample data using a supernatural spiral evolution method. This method improves the speed of power line noise identification by dynamically improving the parameters in the self-organizing map neural network using the multidimensional noise feature approximation residual results.

[0149] 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 scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A carrier noise identification method incorporating a high proportion of power electronic operation, characterized in that, include: Collect power line carrier test noise data; The power line carrier test noise data is input into the constructed self-organizing map neural network, so that the self-organizing map neural network can extract features from the power line carrier test noise data to obtain multi-dimensional feature data. The multi-dimensional noise feature approximation residual is calculated using the multi-dimensional feature data. The multi-dimensional noise feature approximation residual is compared with the preset multi-dimensional noise feature approximation residual to obtain the residual difference value. The number of active neurons is obtained based on the residual difference value. Clustering matching is performed on the multidimensional feature data and the multidimensional noise feature library based on the number of active neurons to obtain noise identification results. The multidimensional noise feature library includes multiple typical noise features, which are constructed by extracting features from power line carrier test noise sample data using a supernatural spiral evolution method. The clustering matching process for the noise identification results is as follows: The multidimensional feature data and weight vector are normalized to obtain normalized multidimensional features and normalized weight vectors, and the inner product of the normalized multidimensional feature data and the normalized weight vectors is calculated. The neurons in the self-organizing map neural network are sorted according to the inner product to obtain a sorting result. Among the sorting results, neurons with an inner product greater than a preset value are selected as active neurons. Active neighborhoods are determined based on the active neurons, and the weights within the active neighborhoods are adjusted based on the gradient of the deviation between the normalized multidimensional features and the normalized weight vector to obtain the adjusted weights. The multidimensional feature data are then clustered using the adjusted weights to obtain the clustering results. Calculate the correlation between the clustering results and the typical noise features, and select the clustering result with the highest correlation as the noise identification result.

2. The carrier noise identification method incorporating a high proportion of power electronic operation as described in claim 1, characterized in that, The multidimensional noise feature library was constructed by extracting features from sample power line carrier test noise data using a supernatural spiral evolution method. Specifically: Collect power line carrier test noise sample data, and extract features from the power line carrier test noise sample data to obtain noise features, wherein the noise features include original noise features and noise-related features; The noise features are filtered to obtain typical noise features, and a multidimensional noise feature library is constructed based on the typical noise features.

3. The carrier noise identification method incorporating a high proportion of power electronic operation as described in claim 2, characterized in that, The process of filtering the noise features to obtain typical noise features, and constructing a multidimensional noise feature library based on these typical noise features, specifically involves: The positions of each of the noise features are initialized; The contribution of each noise feature is iteratively calculated, and the position of each noise feature is updated according to the contribution of each noise feature to obtain the updated position of each noise feature. The contribution of each noise feature is also updated to obtain the updated contribution. This step is repeated until the number of iterations is reached and the iteration stops to obtain the final contribution of each noise feature. The noise features are sorted according to their final contribution, and the noise feature with the largest contribution is selected as the typical noise feature. The multidimensional noise feature library is then constructed based on the typical noise feature.

4. The carrier noise identification method incorporating a high proportion of power electronic operation as described in claim 3, characterized in that, The step of updating the position of each noise feature based on its contribution to obtain the updated position of each noise feature is specifically as follows: The noise features whose contribution is greater than a preset value within a preset radius are aggregated to obtain an adjacency set. Calculate the movement probability of each noise feature to the noise feature in the adjacent set, wherein the formula for calculating the movement probability is: Among them, prob nk (t) represents the probability, c k (t) represents the contribution of the noise feature k, c n (t) represents the contribution of the noise feature n; Based on the movement probability, the target moving noise feature of each noise feature is determined, and the Euclidean distance between each noise feature and the target moving noise feature corresponding to each noise feature is calculated. The updated position of each noise feature is obtained by multiplying the spiral shape parameter and the Euclidean distance, wherein the calculation formula for the updated position is as follows: In the formula, n represents the noise characteristic, k represents the target movement noise characteristic, and x n (t) and x k (t) represents the position of the noise feature and the target movement noise feature at iteration t, respectively. n (t+1) and x k (t+1) represents the position of the noise feature and the target movement noise feature at iteration t+1, respectively, |x k (t)-x n (t) represents the distance from the noise feature to the other target noise features, ||x k (t)-x n (t)|| is the norm of the distance between the noise feature and the other target noise features, σ is a preset logarithmic spiral shape parameter, v is a random number between [-1, 1], t is the iteration number of the current iteration, and S max This represents the maximum number of iterations.

5. The carrier noise identification method incorporating a high proportion of power electronic operation as described in claim 3, characterized in that, The contribution of each noise feature is then updated to obtain the updated contribution, specifically as follows: The importance function of each noise feature is transformed into a noise typical feature contribution value to obtain the updated contribution value. The transformation process is as follows: In the formula, c n (t) represents the contribution of typical noise characteristics, D(x) n (t) represents the importance function of the noise feature, λ represents the enhancement coefficient of the typical noise feature, η represents the attenuation degree of the contribution of the typical noise feature, and t represents the number of iterations in the tth iteration.

6. The carrier noise identification method incorporating a high proportion of power electronic operation as described in claim 1, characterized in that, The self-organizing map neural network extracts features from the power line carrier test noise data to obtain multi-dimensional feature data. The multi-dimensional noise feature approximation residual is then calculated using this multi-dimensional feature data. Specifically: Feature extraction is performed on the power line carrier test noise data to obtain multidimensional features. These multidimensional features are then compared with typical noise features in the multidimensional noise feature library to obtain multidimensional noise feature approximation residuals. The calculation formula for these multidimensional noise feature approximation residuals is as follows: In the formula, I represents the total number of input noise feature data, e represents the approximation residual of the i-th input noise feature data, and z represents the total number of input noise feature data. i For the i-th input noise data, the multidimensional feature value is... This is the estimated multidimensional feature value of the i-th input noise data in the multidimensional noise feature library; The formula for calculating the number of active neurons is as follows: In the formula, e represents the residual approximating the multidimensional noise feature, e min This indicates that the preset multidimensional noise features approximate the residual.

7. The carrier noise identification method incorporating a high proportion of power electronic operation as described in claim 1, characterized in that, The formula for calculating the inner product is: in, This represents the normalized multidimensional features. d represents the normalized weight vector. o (p) represents the inner product.

8. The carrier noise identification method incorporating a high proportion of power electronic operation as described in claim 1, characterized in that, The constructed self-organizing map neural network includes an input layer and an output layer. The input layer corresponds to a high-dimensional input vector, and the output layer consists of a series of ordered nodes organized on a two-dimensional grid. The input nodes and the output nodes are connected by a weight vector.

9. A carrier noise identification system incorporating a high proportion of power electronic operation, characterized in that, include: The acquisition module is used to acquire power line carrier test noise data; The feature extraction module is used to input the power line carrier test noise data into the constructed self-organizing map neural network, so that the self-organizing map neural network can extract features from the power line carrier test noise data to obtain multi-dimensional feature data. The multi-dimensional noise feature approximation residual is calculated using the multi-dimensional feature data. The multi-dimensional noise feature approximation residual is compared with the preset multi-dimensional noise feature approximation residual to obtain the residual difference value. The number of active neurons is obtained based on the residual difference value. Clustering matching is performed on the multidimensional feature data and the multidimensional noise feature library based on the number of active neurons to obtain noise identification results. The multidimensional noise feature library includes multiple typical noise features and is constructed by extracting features from power line carrier test noise sample data using a supernatural spiral evolution method. The clustering matching process for the noise identification results is as follows: The multidimensional feature data and weight vector are normalized to obtain normalized multidimensional features and normalized weight vectors, and the inner product of the normalized multidimensional feature data and the weight vector is calculated. The neurons in the self-organizing map neural network are sorted according to the inner product to obtain a sorting result. Among the sorting results, neurons with an inner product greater than a preset value are selected as active neurons. Active neighborhoods are determined based on the active neurons, and the weights within the active neighborhoods are adjusted based on the gradient of the deviation between the normalized multidimensional features and the normalized weight vector to obtain the adjusted weights. The multidimensional feature data are then clustered using the adjusted weights to obtain the clustering results. Calculate the correlation between the clustering results and the typical noise features, and select the clustering result with the highest correlation as the noise identification result.

10. The carrier noise identification system incorporating a high proportion of power electronic operation as described in claim 9, characterized in that, The multidimensional noise feature library was constructed by extracting features from sample power line carrier test noise data using a supernatural spiral evolution method. Specifically: Collect power line carrier test noise sample data, and extract features from the power line carrier test noise sample data to obtain noise features, wherein the noise features include original noise features and noise-related features; The noise features are filtered to obtain typical noise features, and a multidimensional noise feature library is constructed based on the typical noise features.

Images