Noise modeling model training method, device, electronic device and storage medium

Abstract

This application discloses a method, apparatus, electronic device, and storage medium for training a noise modeling model. The method includes: structuring the original data samples to obtain sample data features corresponding to the original data samples; decoding the sample data features based on each first parameter vector and an initial model to obtain predicted noise characteristics corresponding to each first parameter vector; determining the competition degree corresponding to each first parameter vector; determining multiple second parameter vectors based on the competition degree corresponding to each first parameter vector; transforming the multiple second parameter vectors to obtain multiple third parameter vectors; decoding the sample data features based on each third parameter vector and the initial model to obtain predicted noise characteristics corresponding to each third parameter vector; determining a first target parameter vector from the multiple third parameter vectors; and training the initial model based on the first target parameter vector to obtain a noise modeling model.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to a noise modeling training method, apparatus, electronic device, and storage medium. Background Technology

[0002] Low-voltage power line carrier communication (LPC) technology refers to a technology that uses low-voltage power distribution networks as a transmission medium to achieve data and information exchange. This technology isolates power frequency current and high-frequency signals, ensuring that signals and power frequency currents occupy different frequency bands during transmission, thus guaranteeing orderly and efficient transmission for both. However, existing power transmission lines were not originally designed for data transmission but only for energy transmission, and they also connect various complex electrical devices, resulting in highly complex noise when used as communication channels, including various narrowband, broadband, and pulse noise. Noise increases the bit error rate and reduces the signal-to-noise ratio, thus affecting communication quality. Currently, existing technologies use backpropagation neural networks (BPNNs) to model power line channel noise. However, due to the inherent limitations of BPNNs, they are prone to getting trapped in local optima and lack global search capabilities, thus reducing the accuracy of noise modeling. This makes it impossible to filter noise in the power line channel based on the noise characteristics obtained from the noise modeling, affecting communication quality. Summary of the Invention

[0003] This application provides a noise modeling training method, apparatus, electronic device, and storage medium. By improving the accuracy of noise modeling, noise in power line channels is filtered based on the noise characteristics obtained from the noise modeling, thereby improving communication quality.

[0004] In a first aspect, embodiments of this application provide a method for training a noise modeling model, the method comprising:

[0005] Data is collected from the power line channel to construct training samples, which include original data samples and noise characteristic samples.

[0006] The original data samples are structured to obtain the sample data features corresponding to the original data samples;

[0007] Based on each first parameter vector and the initial model, the features of the sample data are decoded to obtain the predicted noise characteristics corresponding to each first parameter vector, wherein each first parameter vector is randomly generated;

[0008] Based on the predicted noise characteristics corresponding to each first parameter vector and the noise characteristic samples, the degree of competition corresponding to each first parameter vector is determined.

[0009] Based on the competitiveness corresponding to each first parameter vector, multiple second parameter vectors are determined from the multiple first parameter vectors, wherein the multiple second parameter vectors are the first parameter vectors whose competitiveness is greater than a first preset threshold among the multiple first parameter vectors;

[0010] The plurality of second parameter vectors are transformed to obtain a plurality of third parameter vectors;

[0011] Based on each third parameter vector and the initial model, the features of the sample data are decoded to obtain the prediction noise characteristics corresponding to each third parameter vector;

[0012] Based on the predicted noise characteristics corresponding to each third parameter vector and the noise characteristic samples, a first target parameter vector is determined from multiple third parameter vectors;

[0013] The initial model is trained based on the first target parameter vector to obtain a noise modeling model.

[0014] Secondly, embodiments of this application provide a noise modeling method, which includes:

[0015] Data is collected from the power line channel to obtain initial data;

[0016] The initial data is structured to obtain initial data features;

[0017] The initial features are input into the noise modeling model to obtain the target noise characteristics corresponding to the initial data, wherein the noise modeling model is trained by the method of the first aspect.

[0018] Thirdly, embodiments of this application provide a noise modeling training device, which includes: a transceiver unit and a processing unit;

[0019] The transceiver unit is used to collect data on the power line channel and construct training samples, wherein the training samples include original data samples and noise characteristic samples.

[0020] The processing unit is used to structure the original data sample to obtain the sample data features corresponding to the original data sample.

[0021] The processing unit is used to decode the features of the sample data based on each first parameter vector and the initial model to obtain the prediction noise characteristics corresponding to each first parameter vector, wherein each first parameter vector is randomly generated;

[0022] The processing unit is used to determine the degree of competition for each first parameter vector based on the predicted noise characteristics corresponding to each first parameter vector and the noise characteristic samples.

[0023] The processing unit is configured to determine multiple second parameter vectors from multiple first parameter vectors based on the competition degree corresponding to each first parameter vector, wherein the multiple second parameter vectors are first parameter vectors whose competition degree is greater than a first preset threshold among the multiple first parameter vectors;

[0024] The processing unit is used to transform the plurality of second parameter vectors to obtain a plurality of third parameter vectors;

[0025] The processing unit is used to decode the features of the sample data based on each third parameter vector and the initial model to obtain the prediction noise characteristics corresponding to each third parameter vector.

[0026] The processing unit is configured to determine a first target parameter vector from multiple third parameter vectors based on the predicted noise characteristics corresponding to each third parameter vector and the noise characteristic samples.

[0027] The processing unit is used to train the initial model based on the first target parameter vector to obtain a noise modeling model.

[0028] Fourthly, embodiments of this application provide a noise modeling apparatus, which includes: a transceiver unit and a processing unit.

[0029] The transceiver unit is used to collect data on the power line channel to obtain initial data;

[0030] The processing unit is used to structure the initial data to obtain initial data features;

[0031] The processing unit is used to input the initial features into the noise modeling model to obtain the target noise characteristics corresponding to the initial data, wherein the noise modeling model is trained by the method of the first aspect.

[0032] Fifthly, embodiments of this application provide an electronic device, including: a processor and a memory, the processor being connected to the memory, the memory being used to store a computer program, and the processor being used to execute the computer program stored in the memory, so that the electronic device performs the methods as described in the first and second aspects.

[0033] Sixthly, embodiments of this application provide a computer-readable storage medium storing a computer program that causes a computer to perform the methods as described in the first and second aspects.

[0034] In a seventh aspect, embodiments of this application provide a computer program product, the computer program product including a non-transitory computer-readable storage medium storing a computer program, and a computer operable to perform the methods as described in the first and second aspects.

[0035] Implementing the embodiments of this application has the following beneficial effects:

[0036] As can be seen, in this embodiment, training samples are constructed by collecting data on the power line channel, including original data samples and noise characteristic samples. The original data samples are structured to obtain sample data features corresponding to the original data samples. The sample data features are decoded based on each first parameter vector and the initial model to obtain the predicted noise characteristics corresponding to each first parameter vector, wherein each first parameter vector is randomly generated. The competition degree corresponding to each first parameter vector is determined based on the predicted noise characteristics and noise characteristic samples corresponding to each first parameter vector. Based on the competition degree corresponding to each first parameter vector, multiple second parameter vectors are determined from multiple first parameter vectors, wherein the multiple second parameter vectors are first parameter vectors with a competition degree greater than a first preset threshold among multiple first parameter vectors. The multiple second parameter vectors are transformed to obtain multiple third parameter vectors. The sample data features are decoded based on each third parameter vector and the initial model to obtain the predicted noise characteristics corresponding to each third parameter vector. Based on the predicted noise characteristics and noise characteristic samples corresponding to each third parameter vector, a first target parameter vector is determined from multiple third parameter vectors. The initial model is trained based on the first target parameter vector to obtain a noise modeling model. First, the sample data features are decoded using randomly generated first parameter vectors and an initial model to obtain the predicted noise characteristics corresponding to each first parameter vector. Then, based on the competitiveness of each first parameter vector, multiple second parameter vectors with competitiveness greater than a first preset threshold are determined from the multiple first parameter vectors. In this application, instead of directly using the first parameter vector with the highest competitiveness as the first target parameter vector to prevent the obtained first target parameter vector from being a local optimum, the multiple second parameter vectors are further transformed to expand the search range and obtain multiple third parameter vectors. Then, the first target parameter vector is determined from the multiple third parameter vectors, making the obtained first target parameter vector a global optimum. The initial model is then trained based on the global optimum to obtain a noise modeling model. This not only improves the accuracy of noise modeling and filters noise in the power line channel based on the noise characteristics obtained from noise modeling, thereby improving communication quality, but also improves the convergence speed of the noise modeling model. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 A flowchart illustrating a noise modeling training method provided in an embodiment of this application;

[0039] Figure 2 A schematic diagram of a noise acquisition system provided in an embodiment of this application;

[0040] Figure 3 An internal circuit diagram of a noise acquisition system provided in an embodiment of this application;

[0041] Figure 4 A schematic diagram illustrating the convergence speed of a prior art BPNN provided in this application embodiment;

[0042] Figure 5 A schematic diagram illustrating the convergence speed of an optimized BPNN provided in an embodiment of this application;

[0043] Figure 6 A schematic diagram of a prior art BPNN error is provided for an embodiment of this application;

[0044] Figure 7 A schematic diagram of an optimized BPNN error provided in an embodiment of this application;

[0045] Figure 8 A flowchart illustrating another noise modeling training method provided in this application embodiment;

[0046] Figure 9 A flowchart illustrating a noise modeling method provided in an embodiment of this application;

[0047] Figure 10 This is a schematic diagram illustrating the application of a noise modeling method provided in an embodiment of this application;

[0048] Figure 11 A functional unit block diagram of a noise modeling training device provided in this application embodiment;

[0049] Figure 12 A functional unit block diagram of a noise modeling device provided in an embodiment of this application;

[0050] Figure 13 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0051] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0052] The terms "first," "second," "third," and "fourth," etc., used in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0053] In this document, the term "embodiment" means that a particular feature, result, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0054] To facilitate understanding of the technical solution of this application, the relevant technical terms involved in this application will be explained first.

[0055] Power line communication (PLC) channels are a technology that uses existing AC or DC power lines as a transmission medium for high-speed data transmission.

[0056] The main characteristics of noise in power line channels can be summarized as follows:

[0057] (1) Randomness. Noise sources on power lines are very widespread, and many of these noise sources are unstable, including noise during rain and thunder, and the constant switching on and off of various electrical equipment, all of which generate a large amount of random noise.

[0058] (2) Periodicity and continuity. In addition to random noise, there is also some slowly changing periodic noise. The amplitude of this periodic noise can reach 10V, and it usually comes from low-power electrical appliances on the wires. At the same time, due to the wide variety of equipment, a large number of different harmonics are generated and superimposed together, becoming continuous noise on the communication channel;

[0059] (3) Variability. Depending on the equipment connected in different regions and at different times, as well as the structure of the power distribution network, the amplitude and frequency of the noise will vary.

[0060] Secondly, the five types of power line noise are: colored background noise, narrowband noise, periodic impulse noise synchronous with the power frequency, periodic impulse noise asynchronous with the power frequency, and asynchronous non-periodic impulse noise.

[0061] Due to the different interference capabilities and continuously changing characteristics of the aforementioned five types of noise, they can be classified into two major categories. Based on their interference capabilities, the first three types can be unified as background noise. Since the other two types not only have high noise intensity but are mostly continuous, sudden, and impulse noise, researchers classify them as impulse noise.

[0062] (1) Colored background noise is a type of combined noise with a wide spectrum that can occupy the entire channel bandwidth. Its power spectral density is relatively small compared to other noises and it slows down over time.

[0063] (2) Narrowband noise is noise composed of sinusoidal amplitude modulation signals, and its power spectrum is relatively high within a limited frequency range;

[0064] (3) The main reason for the generation of pulse noise asynchronous with the power frequency is the periodic opening and closing of high-output power supply equipment, and its periodic frequency is generally 50-200kHz.

[0065] (4) The characteristics of pulse noise synchronized with the power frequency are high power, long working duration, working frequency of 50Hz or an integer multiple thereof, and periodicity of 10ms-20ms.

[0066] (5) Random noise is different from other types of noise. It refers to a group of impulse interferences with relatively concentrated energy, short duration, and wide frequency spectrum.

[0067] See Figure 1 , Figure 1 This is a flowchart illustrating a noise modeling training method provided in an embodiment of this application. The method includes, but is not limited to, steps 101-109:

[0068] 101: Collect data on the power line channel to construct training samples.

[0069] For example, the training samples include original data samples and noise characteristic samples. During the model training process, multiple data are collected to form a training set and a test set. In this application, the training sample is just any sample in the training set. This application uses the training sample to train the initial model as an example to illustrate a noise modeling training method. When there are multiple training samples, the training method with only one training sample is referred to.

[0070] This application uses the collection of training samples as an example for illustration. The original data samples are the collected measured data. Specifically, through methods such as... Figure 2 The noise acquisition system shown measures noise by connecting to the low-voltage power line on the external power distribution box of the laboratory using a coupling circuit and a high-speed acquisition card. The acquired raw data samples are then sent to a computer. The coupling circuit couples the noise to the high-speed acquisition card, which acquires the raw data samples before transmitting them to the computer for further storage and analysis. It should be noted that the raw data samples include both the data originally intended to be transmitted by the power line and the noise data generated during transmission. The noise characteristic samples represent the characteristic behavior of the noise in the raw data samples, that is, the performance and variation of noise in the power line channel under different times, frequencies, and conditions, specifically including amplitude distribution, power spectral density, and periodicity.

[0071] Furthermore, the internal circuitry of the noise acquisition system, such as Figure 3 The system includes: a low-voltage power line, a power frequency transformer, couplers, a high-speed data acquisition card, and a computer. Each coupler has one end connected to channel A and channel B of the power line, and the other end connected to the neutral wire. Similarly, each power frequency transformer has one end connected to channel A and channel B of the power line, and the other end connected to the live wire. High-frequency interference and noise in the power line network are isolated by the couplers and power frequency transformers. The voltage range is adjusted to -5VDC to +5VDC, and the noise frequency range is adjusted to 1MHz to 2.5MHz, providing a clean power supply to the data acquisition card and reducing the impact of high-frequency noise from the power line on the card. The coupling circuit uses an inductive coupling coil based on the electromagnetic induction principle of a transformer, meaning that one integrated circuit transmits electrical energy or signals to another integrated circuit. In the acquisition of raw data samples, the coupling circuit mainly serves to: prevent permanent damage to the acquisition card; increase the strong electrical signal to suppress the interference of power frequency signals on noise data testing and ensure the safety of the acquisition equipment; use an inductive coupler to couple with high-frequency signals to obtain accurate power line noise data; and reduce the attenuation of high-frequency signals to ensure that noise signals within the frequency range required for the experiment are coupled into the acquisition system to acquire raw data samples of the PLC channel.

[0072] 102: Structure the original data samples to obtain the sample data features corresponding to the original data samples.

[0073] For example, the original data samples are structured to obtain the corresponding sample data features. This includes: cleaning the original data samples to remove or correct outliers, missing values, and duplicate values ​​in the dataset to ensure data quality. For missing values, interpolation or imputation methods can be used; for outliers, thresholds can be set to identify and remove them. Then, the cleaned original data samples are normalized, scaling the data to a specific range, typically [0,1] or [-1,1]. This helps accelerate the learning speed of the neural network because data with small numerical ranges is easier to process. Finally, the dimensions of the data are adjusted according to the structural requirements of the BPNN to obtain the sample data features.

[0074] 103: Based on each first parameter vector and the initial model, the features of the sample data are decoded to obtain the prediction noise characteristics corresponding to each first parameter vector.

[0075] For example, the sample data features are decoded based on each first parameter vector and the initial model to obtain the predicted noise characteristics corresponding to each first parameter vector, wherein each first parameter vector is randomly generated.

[0076] It should be noted that the initial model refers to a BPNN, and the number of layers in a BPNN can be set to 1 for both the input and output layers. The number of hidden layers has a significant impact on the neural network; fewer hidden layers lead to slower convergence and weaker generalization ability, while more hidden layers increase training time. Furthermore, the number of neurons in the input and output layers can be determined by the dimensionality of the data. In this paper, the input data can be 7-dimensional, while the output data can be 1-dimensional, meaning the number of neurons in the input layer can be 7, and the number of neurons in the output layer can be 1. Further, the training function for the initial model is the Levenberg-Marquardt algorithm from the Trainlm function, which has a relatively fast convergence speed.

[0077] Specifically, the first parameter vector is a set of BPNN weights and thresholds. Before model training, multiple first parameter vectors are randomly generated and applied to the initial model. Based on each first parameter vector, the features of the sample data are decoded, and the initial model outputs the predicted noise characteristics corresponding to each first parameter vector. The predicted noise characteristics and the noise characteristic samples have the same dimension. The predicted noise characteristics represent the characteristics and behavior of the noise predicted by the initial model.

[0078] 104: Based on the predicted noise characteristics and noise characteristic samples corresponding to each first parameter vector, determine the degree of competition corresponding to each first parameter vector.

[0079] For example, determining the competitiveness of each first parameter vector based on the predicted noise feature and noise feature samples corresponding to each first parameter vector includes: obtaining the matching degree between the predicted noise feature corresponding to each first parameter vector and the features of the sample data; obtaining the second matching degree between the predicted noise features corresponding to each first parameter vector; obtaining the first similarity between the predicted noise feature corresponding to each first parameter vector and the predicted noise feature corresponding to other first parameter vectors based on the second matching degree and a second preset threshold; determining the frequency of the predicted noise feature corresponding to each first parameter vector among the predicted noise features corresponding to multiple first parameter vectors based on the first similarity; and determining the competitiveness of each first parameter vector based on the matching degree and frequency between the predicted noise feature corresponding to each first parameter vector and the features of the sample data.

[0080] Optionally, the matching degree between the predicted noise characteristic corresponding to each first parameter vector and the sample data feature sum is obtained. The matching degree can be represented by the Euclidean distance between the predicted noise characteristic corresponding to each first parameter vector and the sample data feature sum. The matching degree between the predicted noise characteristic corresponding to each first parameter vector and the sample data feature sum can be determined by formula (1):

[0081]

[0082] Among them, M(a i ,c m ) represents a i and c m The degree of matching, a i Let a represent the prediction noise characteristic corresponding to the i-th first parameter vector among multiple prediction noise characteristics corresponding to first parameter vectors. i,k c represents the k-th dimension of the predicted noise characteristic corresponding to the i-th first parameter vector. m c represents the characteristics of the sample data. m,k Let k represent the k-th dimension of the sample data features, and L be the total dimension of the predicted noise characteristics and the sample data features.

[0083] The first matching degree is obtained between the predicted noise characteristics and the sample data features corresponding to each first parameter vector. This represents the first matching degree between the predicted noise characteristics obtained by applying each first parameter vector and the sample features. The larger the first matching degree, the smaller the error of the predicted noise characteristics obtained by decoding the first parameter vector in the initial model. Therefore, the first matching degree can be used to measure the error between the predicted noise characteristics obtained by applying each first parameter vector to the initial model and the sample data features.

[0084] Optionally, a second matching degree is obtained between the predicted noise characteristics corresponding to each first parameter vector. Specifically, the second matching degree between the predicted noise characteristics corresponding to each first parameter vector can be determined by formula (2).

[0085]

[0086] Among them, M(a i ,b j ) represents a i and b j The degree of matching, a i Let a represent the prediction noise characteristic corresponding to the i-th first parameter vector among multiple prediction noise characteristics corresponding to first parameter vectors. i,k Let b represent the k-th dimension of the predicted noise characteristic corresponding to the i-th first parameter vector. j This represents the prediction noise characteristic corresponding to the j-th first parameter vector among multiple prediction noise characteristics corresponding to first parameter vectors, where i and j are not equal, and b j,k Let L represent the k-th dimension of the predicted noise characteristic corresponding to the j-th first parameter vector, and L be the total dimension of the predicted noise characteristic.

[0087] Optionally, based on the second matching degree and the second preset threshold, a first similarity is obtained between the predicted noise characteristics corresponding to each first parameter vector and the predicted noise characteristics corresponding to other first parameter vectors. The first similarity between the predicted noise characteristics corresponding to each first parameter vector and the predicted noise characteristics corresponding to other first parameter vectors can be determined by formula (3):

[0088]

[0089] Among them, S(a i ,b j ) represents a i and b j Similarity, a i Let b represent the prediction noise characteristic corresponding to the i-th first parameter vector among multiple prediction noise characteristics corresponding to first parameter vectors. j Let x represent the predicted noise characteristic corresponding to the j-th first parameter vector among multiple predicted noise characteristics corresponding to first parameter vectors, where i and j are not equal, and x represents the second preset threshold.

[0090] Optionally, based on the first similarity, the frequency of the predicted noise feature corresponding to each first parameter vector among the predicted noise features corresponding to multiple first parameter vectors is determined. Specifically, this can be determined by formula (4):

[0091]

[0092] Among them, ai Let P(a) represent the prediction noise characteristic corresponding to the i-th first parameter vector among multiple prediction noise characteristics corresponding to first parameter vectors. i S(a) represents the frequency of the predicted noise feature corresponding to the i-th first parameter vector among the predicted noise features corresponding to multiple first parameter vectors, where N is the number of predicted noise features corresponding to multiple first parameter vectors. i ,b j ) represents a i and b j similarity, b j This represents the prediction noise characteristic corresponding to the j-th first parameter vector among multiple prediction noise characteristics corresponding to first parameter vectors.

[0093] It should be noted that determining the frequency of the predicted noise characteristic corresponding to each first parameter vector among the predicted noise characteristics corresponding to multiple first parameter vectors represents the uniqueness of the predicted noise characteristic corresponding to each first parameter vector. A lower frequency indicates a more unique predicted noise characteristic, which can increase the diversity of predicted noise characteristics across multiple first parameter vectors. Conversely, a higher frequency indicates that the predicted noise characteristic appears more frequently, which may not increase the diversity of predicted noise characteristics across multiple first parameter vectors. Instead, it may cause the predicted noise characteristics across multiple first parameter vectors to gradually become more homogeneous, potentially leading the model to get trapped in local optima and thus affecting the model's global search capability.

[0094] Optionally, based on the first matching degree and frequency, the competition degree corresponding to each first parameter vector is determined, specifically, it can be determined by formula (5):

[0095] V(a i )=y*M(a i ,c m )-z*P(a i ) Formula (5)

[0096] Among them, a i V(a) represents the prediction noise characteristic corresponding to the i-th first parameter vector among multiple prediction noise characteristics corresponding to first parameter vectors. i ) represents the competitiveness corresponding to the i-th first parameter vector, y is the third preset threshold, z is the fourth preset threshold, P(a i M(a) represents the frequency of the predicted noise characteristic corresponding to the i-th first parameter vector among the predicted noise characteristics corresponding to multiple first parameter vectors. i ,c m ) represents a i and c m The degree of matching, c mThis represents the characteristics of the sample data.

[0097] The competitiveness of each first parameter vector represents the degree to which each first parameter vector optimizes the performance of the initial model. The higher the competitiveness of the first parameter vector, the stronger the performance of the initial model using that first parameter vector.

[0098] As can be seen, in this application, the competitiveness of each first parameter vector is determined based on the first matching degree and frequency. The higher the first matching degree and the lower the frequency of the first parameter vector, the higher the competitiveness, and the stronger the performance of the initial model using that first parameter vector. A higher first matching degree indicates that the error of the predicted noise characteristics obtained by decoding the first parameter vector in the initial model is smaller, and a lower frequency indicates that the predicted noise characteristics corresponding to the first parameter vector are more unique. The predicted noise characteristics corresponding to the first parameter vector can increase the diversity of predicted noise characteristics corresponding to multiple first parameter vectors, thereby improving the global search capability of the model. Therefore, determining the competitiveness of each first parameter vector based on the first matching degree and frequency can identify the first parameter vector with high first matching degree and low frequency, so that the initial model can be trained based on the first parameter vector. This not only improves the accuracy of the model but also enhances its global search capability, enabling the model to find the global optimal solution and ensuring the global convergence capability of the model.

[0099] 105: Based on the competitiveness corresponding to each first parameter vector, determine multiple second parameter vectors from multiple first parameter vectors.

[0100] For example, based on the competitiveness corresponding to each first parameter vector, multiple second parameter vectors are determined from multiple first parameter vectors, wherein the multiple second parameter vectors are the first parameter vectors whose competitiveness is greater than a first preset threshold among the multiple first parameter vectors. Specifically, the competitiveness corresponding to each first parameter vector is sorted in descending order, and the median value is used as the first preset threshold. The first parameter vectors whose competitiveness is greater than the first preset threshold among the multiple first parameter vectors are used as multiple second parameter vectors. It should be noted that the first parameter vectors corresponding to the first N / 2 competitiveness values ​​after sorting can also be directly used as multiple second parameter vectors.

[0101] As can be seen, in this embodiment, based on the competitiveness corresponding to each first parameter vector, multiple second parameter vectors are determined from multiple first parameter vectors. The multiple second parameter vectors are first parameter vectors whose competitiveness is greater than a first preset threshold among the multiple first parameter vectors. By filtering the first parameter vectors with high first matching degree and low frequency of occurrence through the magnitude of the competitiveness corresponding to each first parameter vector, the first parameter vectors with high matching degree and low frequency of occurrence are selected, so that the first parameter vectors with high frequency are suppressed and the first parameter vectors with high matching degree and low frequency are retained. This ensures the diversity of the prediction noise characteristics corresponding to the first parameter vectors. On the basis of improving the accuracy of the model, it can also improve the global search capability of the model, enabling the model to find the global optimal solution and ensuring the global convergence capability of the model.

[0102] 106: Transform multiple second parameter vectors to obtain multiple third parameter vectors.

[0103] For example, multiple second parameter vectors are transformed to obtain multiple third parameter vectors, including: copying the multiple second parameter vectors to obtain multiple fourth parameter vectors; cross-merging the multiple fourth parameter vectors to obtain multiple fifth parameter vectors; and mutating the multiple fifth parameter vectors to obtain multiple third parameter vectors. First, the multiple second parameter vectors are copied to obtain multiple fourth parameter vectors. The selected better solution (i.e., multiple second parameter vectors) is copied to generate multiple identical copies; the number of copies can be fixed. By copying the better solution, the resulting multiple fourth parameter vectors can maintain the diversity of parameter vectors while increasing the propagation speed of high-quality parameter vectors, thus accelerating convergence.

[0104] Optionally, multiple fourth-parameter vectors can be cross-fused to obtain multiple fifth-parameter vectors. One or more parameter vectors are then randomly selected from these fourth-parameter vectors for pairing. Following a certain cross-cross probability and strategy (such as single-point cross-cross, multi-point cross-cross, uniform cross-cross, etc.), partial data from the paired parameter vectors are exchanged to generate new parameter vectors, i.e., multiple fifth-parameter vectors. The cross-cross operation can combine the advantages of two parameter vectors, producing new, potentially better solutions. Furthermore, cross-crossing can explore new regions in the solution space, increasing the chance of finding the global optimum.

[0105] Optionally, multiple fifth parameter vectors can be mutated to obtain multiple third parameter vectors. Fifth parameter vectors are selected for mutation according to a certain mutation probability. The values ​​at one or more positions of the selected fifth parameter vectors are randomly changed. Mutation methods can include random inversion, random addition or subtraction of a constant, etc. Mutation operations can introduce new values, increase the diversity of the fifth parameter vectors, prevent the model from getting trapped in local optima, and help the model perform a refined search within a local range. It can also help the model escape local optima; when the model is trapped in a local optimum, appropriate mutation operations can help it escape that local region and continue searching for a better solution.

[0106] As can be seen, in this embodiment, multiple second parameter vectors are copied to obtain multiple fourth parameter vectors; multiple fourth parameter vectors are cross-fused to obtain multiple fifth parameter vectors; and multiple fifth parameter vectors are mutated to obtain multiple third parameter vectors. The copying, cross-fertilizing, and mutating operations on multiple second parameter vectors yield multiple third parameter vectors. First, by copying the better solution (i.e., multiple second parameter vectors), the resulting multiple fourth parameter vectors maintain parameter vector diversity while increasing the propagation speed of high-quality parameter vectors. Second, cross-fertilizing explores new regions in the solution space, increasing the chance of finding the global optimum. Third, mutation introduces new values, increasing the diversity of the fifth parameter vectors, preventing the model from getting trapped in local optima, and helping the model to perform fine-grained searches within local ranges. It can also escape local optima; when the model gets trapped in a local optimum, appropriate mutation operations can help it escape that local region and continue searching for a better solution until the optimal solution or a near-optimal solution is found.

[0107] 107: Based on each third parameter vector and the initial model, the features of the sample data are decoded to obtain the prediction noise characteristics corresponding to each third parameter vector.

[0108] It should be noted that the sample data features are decoded based on each third parameter vector and the initial model to obtain the predicted noise characteristics corresponding to each third parameter vector. Each third parameter vector is applied to the initial model, and the sample data features are decoded based on each third parameter vector, resulting in the predicted noise characteristics corresponding to each third parameter vector output by the initial model. The predicted noise characteristics and the noise characteristic samples have the same dimension. The predicted noise characteristics represent the characteristics and behavior of the noise predicted by the initial model.

[0109] 108: Based on the predicted noise characteristics and noise characteristic samples corresponding to each third parameter vector, determine the first target parameter vector from multiple third parameter vectors.

[0110] For example, determining the first target parameter vector from multiple third parameter vectors based on the predicted noise characteristics and noise characteristic samples corresponding to each third parameter vector includes: determining the competition degree corresponding to each third parameter vector based on the predicted noise characteristics and noise characteristic samples corresponding to each third parameter vector; and taking the third parameter vector with the highest competition degree among the multiple third parameter vectors as the first target parameter vector. It should be noted that the method for determining the competition degree corresponding to each third parameter vector based on the predicted noise characteristics and noise characteristic samples corresponding to each third parameter vector is similar to the method for determining the competition degree corresponding to each first parameter vector based on the predicted noise characteristics and noise characteristic samples corresponding to each first parameter vector described in step 104, and will not be repeated here.

[0111] It should be noted that before using the third parameter vector with the highest competitiveness among the multiple third parameter vectors as the first target parameter vector, the method further includes: updating the multiple third parameter vectors, sorting them in descending order based on the competitiveness corresponding to each third parameter vector, retaining the first N / 2 third parameter vectors, removing the last N / 2 third parameter vectors, and randomly generating N / 2 sixth parameter vectors, and using the first N / 2 third parameter vectors and N / 2 sixth parameter vectors as multiple first parameter vectors again. This process can be understood as updating the parameter vectors, which is done once every time the model iterates.

[0112] Furthermore, after updating multiple third parameter vectors, it is determined whether the iteration process meets preset conditions. Specifically, the preset conditions can be whether the number of iterations is greater than a fifth preset threshold, or whether the error is less than a sixth preset threshold. In this application, the example of whether the error is less than the sixth preset threshold is used for explanation. The third parameter vector with the highest competitiveness among multiple third parameter vectors is obtained, and the third matching degree between the predicted noise characteristics and the sample data features corresponding to the third parameter vector with the highest competitiveness is obtained. If the third matching degree is greater than the sixth preset threshold, the corresponding third parameter vector with the highest competitiveness is used as the first target parameter vector.

[0113] Furthermore, if the third matching degree is less than the sixth preset threshold, the sample data features are decoded again based on the new multiple first parameter vectors, and the process of steps 103-108 is repeated until the first target parameter vector is determined.

[0114] As can be seen, in this embodiment, the sample data features are decoded based on each first parameter vector and the initial model to obtain the predicted noise characteristics corresponding to each first parameter vector, wherein each first parameter vector is randomly generated; the competition degree corresponding to each first parameter vector is determined based on the predicted noise characteristics and noise characteristic samples corresponding to each first parameter vector; multiple second parameter vectors are determined from multiple first parameter vectors based on the competition degree corresponding to each first parameter vector, wherein the multiple second parameter vectors are first parameter vectors whose competition degree is greater than a first preset threshold among multiple first parameter vectors; multiple second parameter vectors are transformed to obtain multiple third parameter vectors; the sample data features are decoded based on each third parameter vector and the initial model to obtain the predicted noise characteristics corresponding to each third parameter vector; and a first target parameter vector is determined from multiple third parameter vectors based on the predicted noise characteristics and noise characteristic samples corresponding to each third parameter vector. As can be seen, this application starts by searching for the first objective parameter vector (optimal solution) from multiple randomly generated first parameter vectors (initial solutions). This determination method is not limited to specific problems, nor does it emphasize the quality of algorithm parameter settings and initial solutions. Utilizing its heuristic intelligent search mechanism, even if it starts with a poor initial solution, it can eventually search for the global optimal solution of the problem. It is not highly dependent on the problem and the initial solution, and has strong adaptability and robustness.

[0115] 109: The initial model is trained based on the first objective parameter vector to obtain the noise modeling model.

[0116] For example, training an initial model based on a first target parameter vector to obtain a noise modeling model includes: decoding the features of sample data based on the first target parameter vector to obtain the predicted noise characteristics corresponding to the first target parameter vector; obtaining the loss between the predicted noise characteristics corresponding to the first target parameter vector and the noise characteristic samples; and training the model based on the loss to obtain the noise modeling model.

[0117] It should be noted that after obtaining the first target parameter vector through steps 101-108, the BPNN needs to be trained using the first target parameter vector. Specifically, the sample data features are decoded based on the first target parameter vector to obtain the predicted noise characteristics corresponding to the first target parameter vector. The first target parameter vector is applied to the initial model, and the sample data is input into the initial model. The initial model decodes the sample data features to obtain the predicted noise characteristics corresponding to the first target parameter vector. The loss between the predicted noise characteristics corresponding to the first target parameter vector and the noise characteristic samples is obtained. The loss can be the mean squared error, cross-entropy loss, etc., between the predicted noise characteristics corresponding to the first target parameter vector and the noise characteristic samples, which is not limited in this application. If the loss meets the preset termination condition, the model converges, and a trained noise modeling model is obtained. If the loss does not meet the preset termination condition, the first target parameters are adjusted using the loss to obtain new first target parameters until the model converges, and a trained noise modeling model is obtained. In practical applications, the losses corresponding to multiple training samples can be weighted to obtain a weighted loss, and the first target parameter of the initial model can be adjusted using the weighted loss; alternatively, the first target parameter of the initial model can be adjusted one by one using the loss of each training sample. This application does not limit this.

[0118] As can be seen, in this embodiment, training samples are constructed by collecting data on the power line channel, including original data samples and noise characteristic samples. The original data samples are structured to obtain sample data features corresponding to the original data samples. The sample data features are decoded based on each first parameter vector and the initial model to obtain the predicted noise characteristics corresponding to each first parameter vector, wherein each first parameter vector is randomly generated. The competition degree corresponding to each first parameter vector is determined based on the predicted noise characteristics and noise characteristic samples corresponding to each first parameter vector. Based on the competition degree corresponding to each first parameter vector, multiple second parameter vectors are determined from multiple first parameter vectors, wherein the multiple second parameter vectors are first parameter vectors with a competition degree greater than a first preset threshold among multiple first parameter vectors. The multiple second parameter vectors are transformed to obtain multiple third parameter vectors. The sample data features are decoded based on each third parameter vector and the initial model to obtain the predicted noise characteristics corresponding to each third parameter vector. Based on the predicted noise characteristics and noise characteristic samples corresponding to each third parameter vector, a first target parameter vector is determined from multiple third parameter vectors. The initial model is trained based on the first target parameter vector to obtain a noise modeling model. First, the sample data features are decoded using randomly generated first parameter vectors and an initial model to obtain the predicted noise characteristics corresponding to each first parameter vector. Then, based on the competitiveness of each first parameter vector, multiple second parameter vectors with competitiveness greater than a first preset threshold are determined from the multiple first parameter vectors. In this application, instead of directly using the first parameter vector with the highest competitiveness as the first target parameter vector to prevent the obtained first target parameter vector from being a local optimum, the multiple second parameter vectors are further transformed to expand the search range and obtain multiple third parameter vectors. Then, the first target parameter vector is determined from the multiple third parameter vectors, making the obtained first target parameter vector a global optimum. The initial model is then trained based on the global optimum to obtain a noise modeling model. This not only improves the accuracy of noise modeling and filters noise in the power line channel based on the noise characteristics obtained from noise modeling, thereby improving communication quality, but also improves the convergence speed of the noise modeling model.

[0119] In one embodiment of this application, to verify the effectiveness of the training method, a simulation operation was also performed, specifically including: constructing a training set; building a simulation experimental platform using MATLAB language; and using the built-in BPNN toolbox, continuously optimizing the initial weight threshold (i.e., parameter vector) of the BPNN using the algorithm process described in steps 101-107 to obtain a first target parameter vector; and using the first target parameter vector as input to the initial model BPNN for subsequent power line channel noise modeling. The BPNN performs nonlinear fitting based on different activation functions. Without an activation function, its learning form is linear, and it can only handle some simple two-dimensional functions. Therefore, an activation function is selected to improve the fitting effect of the BPNN on nonlinear problems. After multiple simulations, the activation function was determined to be the tan-sigmoid function.

[0120] Then, the initial model (i.e., BPNN) is trained using the method described in step 108 to obtain a trained noise modeling model.

[0121] like Figure 4 and Figure 5 As shown, compared to noise modeling using existing BPNN techniques, noise modeling using the optimized BPNN described in this application can significantly improve the model's iterative convergence speed. Figure 4 For and Figure 5 The horizontal axis represents the number of iterations, and the vertical axis represents the error. Figure 4 To model noise using existing BPNN technology, the iterative convergence speed and error variation of the model are analyzed. Figure 5 The changes in the model's iterative convergence speed and error are clearly shown in the figure for noise modeling using the optimized BPNN in this application (i.e., optimized using the method described in steps 101-109). It can be clearly seen that the optimized BPNN can significantly improve the model's iterative convergence speed, reduce the error, and improve the accuracy.

[0122] In addition, such as Figure 6 and Figure 7 As shown, the horizontal axis represents the test sample number, and the vertical axis represents the specific value. Figure 4 To model noise using existing BPNN technology, the expected value and error value are obtained; Figure 5 The expected value and error value obtained by noise modeling using the optimized BPNN in this application (i.e., optimized by the method described in steps 101-109) can be clearly seen from the figure. The difference between the expected value and error value obtained by noise modeling using the optimized BPNN is significantly smaller than the expected value and error value obtained by noise modeling using the BPNN of the prior art.

[0123] Furthermore, as shown in Table 1, two different noise modeling methods are compared: the existing BPNN and the BPNN optimized in this application (represented as the optimized BPNN in Table 1). The test set size for both methods is 500 samples. Table 1 shows that the average prediction errors of the existing BPNN and the optimized BPNN are 15% and 3%, respectively; the mean square errors of the prediction data are 7.60E-04 and 9.10E-05, respectively; and the correlation coefficients between the existing BPNN and the optimized BPNN and the actual data are 73% and 97%, respectively. Therefore, it can be seen that the optimized BPNN has the characteristics of small error and high accuracy.

[0124] Table 1

[0125]

[0126] In one embodiment of this application, another method for training a noise modeling model is also provided, see below. Figure 8 , Figure 8 This application provides a flowchart illustrating another noise modeling training method, which includes, but is not limited to, steps 801-813:

[0127] 801: Input raw data sample.

[0128] 802: Randomly generate multiple first parameter vectors.

[0129] 803: Determine the first matching degree and frequency corresponding to each first parameter vector.

[0130] 804: Determine the competition degree corresponding to each first parameter vector based on the first matching degree and frequency corresponding to each first parameter vector.

[0131] 805: Based on the competitiveness corresponding to each first-order parameter vector, multiple second-order parameter vectors are obtained.

[0132] Specifically, the first parameter vectors corresponding to the first N / 2 degrees of competition are excited, and the first parameter vectors corresponding to the last N / 2 degrees of competition are suppressed, resulting in multiple second parameter vectors.

[0133] 806: Transform multiple second parameter vectors to obtain multiple third parameter vectors.

[0134] 807: Update multiple third parameter vectors.

[0135] 808: Does the iteration process meet the preset conditions?

[0136] 809: If so, use the third parameter vector with the highest corresponding competitiveness as the first target parameter vector.

[0137] If not, repeat steps 803-809 until the first target parameter vector is determined.

[0138] 810: Initial model.

[0139] Specifically, the initial model decodes the features of the sample data based on the first target parameter vector to obtain the prediction noise characteristics corresponding to the first target parameter vector.

[0140] 811: Obtain the loss between the predicted noise characteristics and the noise characteristic samples corresponding to the first target parameter vector.

[0141] 812: Determine whether the loss meets the preset conditions.

[0142] 813: If so, the model converges, and a well-trained noise modeling model is obtained.

[0143] If not, adjust the initial model based on the loss.

[0144] The methods in steps 801-813 have been described in detail in steps 101-109, and will not be repeated here.

[0145] It should be noted that the model undergoes an immune memory process with each iteration.

[0146] See Figure 9 , Figure 9 This is a flowchart illustrating a noise modeling method provided in an embodiment of this application. The method includes, but is not limited to, steps 901-903:

[0147] 901: Collect data from the power line channel to obtain initial data.

[0148] like Figure 10 As shown, the data is collected using the method described in step 101, which will not be elaborated upon here.

[0149] 902: Structure the initial data to obtain the initial data features.

[0150] like Figure 10 The data is structured in a manner similar to the structuring method described in step 102, and will not be repeated here.

[0151] 903: Input the initial features into the noise modeling model to obtain the target noise characteristics corresponding to the initial data.

[0152] The noise modeling model is trained using the methods described in steps 101-109.

[0153] Following 903, the method further includes: performing spectral analysis on the target noise characteristics to obtain the frequency components of the noise data in the initial data; and filtering the initial data based on the frequency components to obtain the actual transmitted data in the initial data.

[0154] Specifically, such as Figure 10 The diagram illustrates how a noise model is used to obtain the target noise characteristics. A spectral analysis of these characteristics is then performed to identify the frequency components of the noise data in the initial dataset. Specifically, a Fast Fourier Transform or other spectral analysis methods are used to perform a spectral analysis of the target noise characteristics to identify the main noise frequency components.

[0155] Furthermore, based on the frequency components, the initial data is filtered to obtain the actual transmitted data from the initial data. Specifically, based on the results of the spectrum analysis (i.e., the frequency components), a band-stop filter is designed with its notch frequency aligned with the main noise frequency. For example... Figure 10 As shown, the designed filter is applied to the target noise characteristics to filter out noise components of a fixed frequency, and the initial data is filtered to obtain the actual transmitted data in the initial data.

[0156] As can be seen, in this embodiment, data is collected from the power line channel to obtain initial data; the initial data is structured to obtain initial data features; the initial features are input into a noise modeling model to obtain target noise characteristics corresponding to the initial data, wherein the noise modeling model is trained using a noise modeling model training method; spectral analysis is performed on the target noise characteristics to obtain the frequency components of the noise data in the initial data; based on the frequency components, the initial data is filtered to obtain the actual transmitted data in the initial data. Extracting the target noise characteristics from the initial data using a trained noise model can improve the accuracy and efficiency of noise feature extraction. Then, spectral analysis and filtering based on the target noise characteristics can effectively remove noise from the initial data, filtering noise in the power line channel and thus improving communication quality.

[0157] See Figure 11 , Figure 11 This is a functional unit block diagram of a noise modeling training device provided in an embodiment of this application. The noise modeling training device 1100 includes: a transceiver unit 1101 and a processing unit 1102;

[0158] The transceiver unit 1101 is used to collect data on the power line channel and construct training samples, wherein the training samples include original data samples and noise characteristic samples.

[0159] Processing unit 1102 is used to structure the original data sample to obtain the sample data features corresponding to the original data sample;

[0160] This is used to decode the features of the sample data based on each first parameter vector and the initial model, to obtain the predicted noise characteristics corresponding to each first parameter vector, wherein each first parameter vector is randomly generated;

[0161] Used to determine the degree of competition for each first parameter vector based on the predicted noise characteristics corresponding to each first parameter vector and the noise characteristic samples;

[0162] This is used to determine multiple second parameter vectors from multiple first parameter vectors based on the degree of competition corresponding to each first parameter vector, wherein the multiple second parameter vectors are first parameter vectors whose degree of competition is greater than a first preset threshold among the multiple first parameter vectors;

[0163] This is used to transform the plurality of second parameter vectors to obtain a plurality of third parameter vectors;

[0164] It is used to decode the features of sample data based on each third parameter vector and the initial model, and obtain the prediction noise characteristics corresponding to each third parameter vector;

[0165] Used to determine a first target parameter vector from multiple third parameter vectors based on the predicted noise characteristics corresponding to each third parameter vector and the noise characteristic samples;

[0166] This is used to train the initial model based on the first target parameter vector to obtain a noise modeling model.

[0167] In one embodiment of this application, in determining the competitiveness of each first parameter vector based on the predicted noise characteristics and the noise characteristic samples, the processing unit 1102 is specifically used for:

[0168] Obtain the first matching degree between the predicted noise characteristics corresponding to each first parameter vector and the features of the sample data;

[0169] Obtain the second matching degree between the predicted noise characteristics corresponding to each first parameter vector;

[0170] Based on the second matching degree and the second preset threshold, the first similarity between the predicted noise characteristics corresponding to each first parameter vector and the predicted noise characteristics corresponding to other first parameter vectors is obtained.

[0171] Based on the first similarity, determine the frequency of the predicted noise feature corresponding to each first parameter vector among the predicted noise features corresponding to multiple first parameter vectors;

[0172] Based on the first matching degree and frequency, the competition degree corresponding to each first parameter vector is determined.

[0173] In one embodiment of this application, in the process of transforming the plurality of second parameter vectors to obtain a plurality of third parameter vectors, the processing unit 1102 is specifically used for:

[0174] The plurality of second parameter vectors are copied to obtain a plurality of fourth parameter vectors;

[0175] The multiple fourth parameter vectors are cross-fused to obtain multiple fifth parameter vectors;

[0176] The plurality of fifth parameter vectors are mutated to obtain a plurality of third parameter vectors.

[0177] In one embodiment of this application, in determining a first target parameter vector from multiple third parameter vectors based on the predicted noise characteristics corresponding to each third parameter vector and the noise characteristic samples, the processing unit 1102 is specifically used for:

[0178] Based on the predicted noise characteristics corresponding to each third parameter vector and the noise characteristic samples, the degree of competition corresponding to each third parameter vector is determined.

[0179] The third parameter vector with the highest competitiveness among multiple third parameter vectors is taken as the first target parameter vector.

[0180] In one embodiment of this application, in the process of training the initial model based on the first target parameter vector to obtain a noise modeling model, the processing unit 1102 is specifically used for:

[0181] Based on the first target parameter vector, the features of the sample data are decoded to obtain the prediction noise characteristics corresponding to the first target parameter vector;

[0182] Obtain the loss between the predicted noise characteristics corresponding to the first target parameter vector and the noise characteristic samples;

[0183] The noise modeling model is obtained by training the model based on the loss.

[0184] See Figure 12 , Figure 12 This is a functional unit block diagram of a noise modeling device provided in an embodiment of this application. The noise modeling device 1200 includes: a transceiver unit 1201 and a processing unit 1202;

[0185] Transceiver unit 1201 is used to collect data on the power line channel to obtain initial data;

[0186] Processing unit 1202 is used to structure the initial data to obtain initial data features;

[0187] The processing unit 1202 is used to input the initial features into the noise modeling model to obtain the target noise characteristics corresponding to the initial data, wherein the noise modeling model is trained by a noise modeling model training device.

[0188] In one embodiment of this application, the processing unit 1202 is specifically used for: performing spectral analysis on the target noise characteristics to obtain the frequency components of the noise data in the initial data; and filtering the initial data based on the frequency components to obtain the actual transmission data in the initial data.

[0189] See Figure 13 , Figure 13 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 13 As shown, the electronic device 1300 includes a transceiver 1301, a processor 1302, and a memory 1303. They are connected to each other via a bus 1304. The memory 1303 is used to store computer programs and data, and can transfer the data stored in the memory 1303 to the processor 1302.

[0190] Optionally, the processor 1302 is used to read the computer program in the memory 1303 and perform the following operations:

[0191] The transceiver 1301 collects data on the power line channel and constructs training samples, wherein the training samples include original data samples and noise characteristic samples;

[0192] The original data samples are structured to obtain the sample data features corresponding to the original data samples;

[0193] Based on each first parameter vector and the initial model, the features of the sample data are decoded to obtain the predicted noise characteristics corresponding to each first parameter vector, wherein each first parameter vector is randomly generated;

[0194] Based on the predicted noise characteristics corresponding to each first parameter vector and the noise characteristic samples, the degree of competition corresponding to each first parameter vector is determined.

[0195] Based on the competitiveness corresponding to each first parameter vector, multiple second parameter vectors are determined from the multiple first parameter vectors, wherein the multiple second parameter vectors are the first parameter vectors whose competitiveness is greater than a first preset threshold among the multiple first parameter vectors;

[0196] The plurality of second parameter vectors are transformed to obtain a plurality of third parameter vectors;

[0197] Based on each third parameter vector and the initial model, the features of the sample data are decoded to obtain the prediction noise characteristics corresponding to each third parameter vector;

[0198] Based on the predicted noise characteristics corresponding to each third parameter vector and the noise characteristic samples, a first target parameter vector is determined from multiple third parameter vectors;

[0199] The initial model is trained based on the first target parameter vector to obtain a noise modeling model.

[0200] Specifically, the transceiver 1301 described above can be... Figure 11 The transceiver unit 1101 of the noise modeling training device 1100 in the embodiment, the processor 1302 can be... Figure 11 The processing unit 1102 of the noise modeling training device 1100 in the embodiment.

[0201] Specifically, the transceiver 1301 described above can be... Figure 11 The transceiver unit 1101 of the noise modeling training device 1100 in the embodiment, the processor 1302 can be... Figure 11 The processing unit 1102 of the noise modeling training device 1100 in the embodiment. Therefore, the specific function of the processor 1302 can be referred to the specific function of the processing unit 1102, and the specific function of the transceiver 1301 can be referred to the specific function of the transceiver unit 1101.

[0202] Optionally, the processor 1302 is used to read the computer program in the memory 1303 and perform the following operations:

[0203] The transceiver 1301 is controlled to collect data on the power line channel to obtain initial data;

[0204] The initial data is structured to obtain initial data features;

[0205] The initial features are input into the noise modeling model to obtain the target noise characteristics corresponding to the initial data, wherein the noise modeling model is trained by a noise modeling model training device.

[0206] Specifically, the transceiver 1301 described above can be... Figure 12 The transceiver unit 1201 of the noise modeling apparatus 1200 in the embodiment, the processor 1302 can be Figure 12 The processing unit 1202 of the noise modeling apparatus 1200 in the embodiment.

[0207] Specifically, the transceiver 1301 described above can be... Figure 12 The transceiver unit 1201 of the noise modeling apparatus 1200 in the embodiment, the processor 1302 can be Figure 12The noise modeling apparatus 1200 of the embodiment is a processing unit 1202. Therefore, the specific function of the processor 1302 can be referred to the specific function of the processing unit 1202, and the specific function of the transceiver 1301 can be referred to the specific function of the transceiver unit 1201.

[0208] It should be understood that the electronic devices mentioned in this application may include smartphones (such as Android phones, iOS phones, Windows Phones, etc.), tablet computers, PDAs, laptops, mobile internet devices (MIDs), or wearable devices. The above-mentioned electronic devices are merely examples and not exhaustive, and include, but are not limited to, the electronic devices described above. In practical applications, the above-mentioned electronic devices may also include: intelligent in-vehicle terminals, computer equipment, etc.

[0209] This application also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement some or all of the steps of any of the noise modeling training methods described in the above method embodiments.

[0210] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps of any of the noise modeling training methods described in the above method embodiments.

[0211] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0212] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0213] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0214] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0215] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software program module.

[0216] If the integrated unit is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0217] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0218] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for training a noise modeling system, characterized in that, The method includes: Data is collected from the power line channel to construct training samples, which include original data samples and noise characteristic samples. The original data samples are structured to obtain the sample data features corresponding to the original data samples; Based on each first parameter vector and the initial model, the features of the sample data are decoded to obtain the predicted noise characteristics corresponding to each first parameter vector, wherein each first parameter vector is randomly generated; Based on the predicted noise characteristics corresponding to each first parameter vector and the noise characteristic samples, the degree of competition corresponding to each first parameter vector is determined. Based on the competitiveness corresponding to each first parameter vector, multiple second parameter vectors are determined from the multiple first parameter vectors, wherein the multiple second parameter vectors are the first parameter vectors whose competitiveness is greater than a first preset threshold among the multiple first parameter vectors; The plurality of second parameter vectors are transformed to obtain a plurality of third parameter vectors; Based on each third parameter vector and the initial model, the features of the sample data are decoded to obtain the prediction noise characteristics corresponding to each third parameter vector; Based on the predicted noise characteristics corresponding to each third parameter vector and the noise characteristic samples, a first target parameter vector is determined from multiple third parameter vectors; The initial model is trained based on the first target parameter vector to obtain a noise modeling model.

2. The method according to claim 1, characterized in that, The step of determining the competitiveness of each first parameter vector based on the predicted noise characteristics and the noise characteristic samples corresponding to each first parameter vector includes: Obtain the first matching degree between the predicted noise characteristics corresponding to each first parameter vector and the features of the sample data; Obtain the second matching degree between the predicted noise characteristics corresponding to each first parameter vector; Based on the second matching degree and the second preset threshold, the first similarity between the predicted noise characteristics corresponding to each first parameter vector and the predicted noise characteristics corresponding to other first parameter vectors is obtained. Based on the first similarity, determine the frequency of the predicted noise feature corresponding to each first parameter vector among the predicted noise features corresponding to multiple first parameter vectors; Based on the first matching degree and frequency, the competition degree corresponding to each first parameter vector is determined.

3. The method according to claim 1, characterized in that, The transformation process performed on the plurality of second parameter vectors to obtain a plurality of third parameter vectors includes: The plurality of second parameter vectors are copied to obtain a plurality of fourth parameter vectors; The multiple fourth parameter vectors are cross-fused to obtain multiple fifth parameter vectors; The plurality of fifth parameter vectors are mutated to obtain a plurality of third parameter vectors.

4. The method according to claim 3, characterized in that, The step of determining the first target parameter vector from multiple third parameter vectors based on the predicted noise characteristics corresponding to each third parameter vector and the noise characteristic samples includes: Based on the predicted noise characteristics corresponding to each third parameter vector and the noise characteristic samples, the degree of competition corresponding to each third parameter vector is determined. The third parameter vector with the highest competitiveness among multiple third parameter vectors is taken as the first target parameter vector.

5. The method according to claim 2, characterized in that, The step of training the initial model based on the first target parameter vector to obtain a noise modeling model includes: Based on the first target parameter vector, the features of the sample data are decoded to obtain the prediction noise characteristics corresponding to the first target parameter vector; Obtain the loss between the predicted noise characteristics corresponding to the first target parameter vector and the noise characteristic samples; The noise modeling model is obtained by training the model based on the loss.

6. A noise modeling method, characterized in that, The method includes: Data is collected from the power line channel to obtain initial data; The initial data is structured to obtain initial data features; The initial data features are input into the noise modeling model to obtain the target noise characteristics corresponding to the initial data, wherein the noise modeling model is trained by any one of the methods in claims 1-5.

7. The method according to claim 6, characterized in that, The method further includes: Spectral analysis of the target noise characteristics is performed to obtain the frequency components of the noise data in the initial data; Based on the frequency components, the initial data is filtered to obtain the actual transmitted data from the initial data.

8. A noise modeling training device, characterized in that, The device includes: Transceiver unit and processing unit; The transceiver unit is used to collect data on the power line channel and construct training samples, wherein the training samples include original data samples and noise characteristic samples. The processing unit is used to structure the original data sample to obtain the sample data features corresponding to the original data sample. The processing unit is used to decode the features of the sample data based on each first parameter vector and the initial model to obtain the prediction noise characteristics corresponding to each first parameter vector, wherein each first parameter vector is randomly generated; The processing unit is used to determine the degree of competition for each first parameter vector based on the predicted noise characteristics corresponding to each first parameter vector and the noise characteristic samples. The processing unit is configured to determine multiple second parameter vectors from multiple first parameter vectors based on the competition degree corresponding to each first parameter vector, wherein the multiple second parameter vectors are first parameter vectors whose competition degree is greater than a first preset threshold among the multiple first parameter vectors; The processing unit is used to transform the plurality of second parameter vectors to obtain a plurality of third parameter vectors; The processing unit is used to decode the features of the sample data based on each third parameter vector and the initial model to obtain the prediction noise characteristics corresponding to each third parameter vector. The processing unit is configured to determine a first target parameter vector from multiple third parameter vectors based on the predicted noise characteristics corresponding to each third parameter vector and the noise characteristic samples. The processing unit is used to train the initial model based on the first target parameter vector to obtain a noise modeling model.

9. An electronic device, characterized in that, include: A processor and a memory, the processor being connected to the memory, the memory being used to store a computer program, and the processor being used to execute the computer program stored in the memory to cause the electronic device to perform the method as claimed in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a computer program that is executed by a processor to implement the method as claimed in any one of claims 1-7.

Images