Noise detection method and device, computer equipment and readable storage medium

Abstract

The invention relates to a noise detection method and device, computer equipment and a computer readable storage medium. The method comprises the following steps: under the condition that a first confidence coefficient is smaller than a confidence coefficient threshold value, performing simulation correction on transient voltage and transient current through an initial digital twinborn model to obtain a first error value; the first confidence corresponds to the first predicted noise intensity and is output by the initial noise detection model based on the first feature vector; under the condition that the first error value is greater than an error threshold value, correcting at least one of the initial noise detection model, the first feature vector and the initial digital twinborn model according to the first error value to obtain a first correction result; obtaining the target predicted noise intensity of the target node according to the first correction result; and when the target prediction noise intensity is greater than or equal to the alarm threshold, sending alarm information. By adopting the method, the accuracy of noise detection can be improved.

Description

Technical Field

[0001] This application relates to the field of power grid technology, and in particular to a noise detection method, apparatus, and computer equipment. Background Technology

[0002] With the large-scale deployment of smart grids, smart meters, and distributed energy, power line communication technology for distribution networks has become an important means of communication in distribution networks due to its lack of the need for additional communication carriers, low cost, and ease of deployment.

[0003] In traditional technologies, noise detection of power line communication signals in distribution networks mainly relies on time-domain filters to suppress fixed-frequency band interference, frequency-domain transformations (such as FFT or wavelets) plus threshold discrimination, and quantitative assessment based on statistics such as kurtosis. In recent years, some studies have introduced machine learning or deep learning algorithms to classify and estimate the noise characteristics of power line communication signals in distribution networks, but most of these studies remain at the stage of offline training and single-node applications.

[0004] However, due to the complex environment of power distribution lines, the diverse and dynamic noise interference reduces the accuracy of power line communication signal noise detection in power distribution networks. Summary of the Invention

[0005] Therefore, it is necessary to provide a noise detection method, apparatus, computer equipment, and computer-readable storage medium that can improve accuracy in addressing the aforementioned technical problems.

[0006] In a first aspect, this application provides a noise detection method, the method comprising:

[0007] The first feature vector of the target node in the distribution network is obtained and input into the initial noise detection model to obtain the first predicted noise intensity and the first confidence level corresponding to the first predicted noise intensity; wherein, the first feature vector is determined based on transient voltage, transient current and topological features;

[0008] If the first confidence level is less than the confidence level threshold, the transient voltage and the transient current are simulated and corrected using an initial digital twin model to obtain a first error value; wherein, the initial digital twin model is determined based on the measured line parameters of each node in the distribution network and is a digital twin model into which the first predicted noise intensity is injected.

[0009] If the first error value is greater than the error threshold, at least one of the initial noise detection model, the first feature vector, and the initial digital twin model is corrected based on the first error value to obtain a first correction result;

[0010] Based on the first correction result, the target predicted noise intensity of the target node is obtained;

[0011] An alarm message is sent if the target predicted noise intensity at the target node is greater than or equal to the alarm threshold.

[0012] In one embodiment, the first correction result includes at least one of the corrected initial noise detection model, the corrected first feature vector, and the corrected initial digital twin model; the first error value includes multi-domain feature error; the multi-domain feature error includes time-domain feature error, frequency-domain feature error, and topological feature error; the step of correcting at least one of the initial noise detection model, the first feature vector, and the initial digital twin model according to the first error value to obtain the first correction result includes:

[0013] If the time-domain feature error is greater than the time-domain threshold, the hyperparameters and learning rate in the initial noise detection model are adjusted to obtain the corrected initial noise detection model; wherein, the time-domain feature error is determined based on the root mean square of the simulated voltage in the time-domain simulation feature vector and the root mean square of the transient voltage in the time-domain measured feature vector, or the root mean square of the simulated current in the time-domain simulation feature vector and the root mean square of the transient current in the time-domain measured feature vector.

[0014] When the frequency domain feature error is greater than the frequency domain threshold, the first feature vector after correction is obtained by performing target series wavelet packet decomposition on the transient voltage of the target node; wherein, the frequency domain feature error is determined based on the simulated subband energy in the frequency domain simulated feature vector and the measured subband energy in the frequency domain measured feature vector.

[0015] If the topology feature error is greater than the topology threshold, the branch reactance that is greater than the topology threshold is corrected in the initial digital twin model to obtain a corrected initial digital twin model; wherein, the topology feature error is determined based on the sensitivity of the branch reactance.

[0016] In one embodiment, obtaining the target predicted noise intensity of the target node based on the first correction result includes:

[0017] If the second confidence level is greater than or equal to the confidence level threshold, then at least one of the following is executed: using the corrected initial noise detection model as the target noise detection model and using the corrected first feature vector as the second feature vector; wherein, the second confidence level is the output of the corrected initial noise detection model based on the first feature vector or the corrected first feature vector, or the output of the initial noise detection model based on the corrected first feature vector;

[0018] Obtain the second predicted noise intensity corresponding to the second confidence level, and use it as the target predicted noise intensity of the target node.

[0019] In one embodiment, obtaining the target predicted noise intensity of the target node based on the first correction result includes:

[0020] If the first confidence level is less than the confidence threshold, the transient voltage and the transient current are simulated and corrected using the corrected initial digital twin model to obtain a second error value;

[0021] If the second error value is less than or equal to the error threshold, the first predicted noise intensity is taken as the target predicted noise intensity of the target node;

[0022] The method further includes: using the corrected initial digital twin model as the target digital twin model; wherein the target digital twin model is used to update the measured line parameters of each node in the distribution network.

[0023] In one embodiment, obtaining the target predicted noise intensity of the target node based on the first correction result includes:

[0024] If the second confidence level is less than the confidence level threshold, the transient voltage and the transient current are simulated and corrected using the corrected initial digital twin model to obtain a second error value; wherein, the corrected initial digital twin model is injected with a second predicted noise intensity; the second predicted noise intensity corresponds to the second confidence level; the second confidence level is the output of the corrected initial noise detection model based on the first feature vector or the corrected first feature vector, or the output of the initial noise detection model based on the corrected first feature vector;

[0025] If the second error value is less than or equal to the error threshold, the second predicted noise intensity is taken as the target predicted noise intensity of the target node;

[0026] The method further includes: using the corrected initial digital twin model as the target digital twin model; wherein the target digital twin model is used to update the measured line parameters of each node in the distribution network.

[0027] In one embodiment, the method further includes:

[0028] If the second error value is greater than the error threshold, at least one of the corrected initial noise detection model, the corrected first feature vector, and the corrected initial digital twin model is corrected again based on the second error value to obtain a second correction result;

[0029] Based on the second correction result, the target predicted noise intensity of the target node is obtained.

[0030] In one embodiment, the step of simulating and correcting the transient voltage and transient current using an initial digital twin model to obtain a first error value includes:

[0031] The transient voltage and transient current are simulated using the initial digital twin model to obtain the simulated voltage and simulated current.

[0032] Based on the simulated current and the simulated voltage, the time-domain simulation feature vector and the frequency-domain simulation feature vector are obtained;

[0033] Based on the time-domain simulation feature vector, the frequency-domain simulation feature vector, and the topology simulation features of the target node, the simulation feature vector and the simulation noise intensity are determined.

[0034] The first error value is determined based on the simulated feature vector, the first feature vector, the simulated noise intensity, and the first predicted noise intensity.

[0035] Secondly, this application also provides a noise detection device, the device comprising:

[0036] The acquisition module is used to input the first feature vector of the target node in the distribution network into the initial noise detection model, and to obtain the first predicted noise intensity and the first confidence level corresponding to the first predicted noise intensity of the initial noise detection model; wherein, the first feature vector is determined based on transient voltage, transient current and topological features;

[0037] The first error determination module is used to perform simulation correction on the transient voltage and the transient current through an initial digital twin model when the first confidence level is less than the confidence level threshold, so as to obtain a first error value; wherein, the initial digital twin model is determined based on the measured line parameters of each node in the distribution network and is a digital twin model of the first predicted noise intensity.

[0038] The correction module is used to correct at least one of the initial noise detection model, the first feature vector, and the initial digital twin model based on the first error value when the first error value is greater than the error threshold, so as to obtain a first correction result.

[0039] The result determination module is used to obtain the target predicted noise intensity of the target node based on the first correction result;

[0040] The alarm module is used to send alarm information when the target predicted noise intensity of the target node is greater than or equal to the alarm threshold.

[0041] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0042] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0043] The aforementioned noise detection method, apparatus, computer equipment, and computer-readable storage medium, firstly, when the first confidence level is less than a confidence threshold, wherein the initial noise detection model outputs a first predicted noise intensity and a corresponding first confidence level based on a first feature vector; an initial digital twin model corresponding one-to-one with the actual line parameters is introduced, and closed-loop simulation correction is performed in conjunction with the first confidence level to obtain a first error value for simulation verification; secondly, when the first error value is greater than the error threshold, at least one of the initial noise detection model, the first feature vector, and the initial digital twin model is corrected according to the first error value to obtain a first correction result, and the physical parameters of the first feature, the initial noise detection model, and the initial digital twin model are dynamically adjusted in the process to maintain high robustness and accuracy to environmental changes; finally, based on the dynamically adjusted first correction result, the target predicted noise intensity of the target node is obtained, and when the target predicted noise intensity is greater than or equal to an alarm threshold, an alarm message is sent, thereby improving the accuracy of target predicted noise intensity detection and ensuring the accuracy of alarm messages based on target predicted noise intensity. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating a noise detection method in one embodiment;

[0046] Figure 2 This is a flowchart illustrating the first correction result determined based on a first error value in one embodiment.

[0047] Figure 3 This is a flowchart illustrating the process of obtaining the target prediction noise intensity of a target node based on the first correction result in one embodiment.

[0048] Figure 4This is a flowchart illustrating the process of obtaining the target predicted noise intensity of a target node based on the first correction result in another embodiment.

[0049] Figure 5 This is a flowchart illustrating the process of obtaining the target prediction noise intensity of a target node based on the first correction result in yet another embodiment.

[0050] Figure 6 This is a schematic diagram illustrating the process of obtaining a first error value based on an initial digital twin model in one embodiment.

[0051] Figure 7 This is a flowchart illustrating the noise detection method in another embodiment;

[0052] Figure 8 This is a structural block diagram of a noise detection device in one embodiment;

[0053] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0055] In one embodiment, such as Figure 1 As shown, a noise detection method is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and to a system including both a terminal and a server, and can be implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps S102 to S110.

[0056] Step S102: Input the first feature vector of the target node in the distribution network into the initial noise detection model to obtain the first predicted noise intensity and the first confidence level corresponding to the first predicted noise intensity output by the initial noise detection model.

[0057] Wherein, the first eigenvector is denoted as It is determined based on transient voltage, transient current and topological characteristics.

[0058] Optionally, the first feature vector is generated by: determining a time-domain measured feature vector based on the transient voltage and transient current of the target node; determining a frequency-domain measured feature vector based on the transient voltage of the target node; determining a topological measured feature vector based on the topological features of the target node; and determining the first feature vector based on the time-domain measured feature vector, the frequency-domain measured feature vector, and the topological measured feature vector.

[0059] Furthermore, based on the transient voltage and transient current of the target node, the time-domain measured feature vector is determined, including: determining the root mean square of the transient voltage based on the transient voltage of the target node; determining the root mean square of the transient current based on the transient current of the target node; and determining the time-domain measured feature vector based on the root mean square of the transient voltage and the root mean square of the transient current.

[0060] Furthermore, the acquisition equipment uses a sampling rate Transient voltage of a target node in a power distribution network With transient current conduct Continuous sampling within the time window allows the acquisition device to capture the transient voltage of the target node. With transient current Send to the terminal. The terminal then processes the data based on the transient voltage of the target node. The root mean square value is calculated as shown in formula (1), and the root mean square value of the transient voltage is obtained. .

[0061] Formula (1)

[0062] The terminal uses the transient current of the target node as a reference. The root mean square value is calculated as shown in formula (2), and the root mean square value of the transient current is obtained. .

[0063] Formula (2)

[0064] The terminal will convert the root mean square of the transient voltage. and root mean square of transient current By concatenating the vectors, we obtain the time-domain measured feature vectors. .

[0065] Furthermore, based on the transient voltage of the target node, the frequency domain measured feature vector is determined, including: performing a first target-level wavelet packet decomposition on the transient voltage of the target node to obtain the measured sub-band coefficients; determining the energy of each measured sub-band based on the measured sub-band coefficients; and determining the frequency domain measured feature vector based on the energy of each measured sub-band.

[0066] Furthermore, the terminal responds to transient voltages Perform the first objective-level wavelet packet decomposition on the transient voltage. Perform wavelet packet decomposition to the th Level, obtain the measured subband coefficient, denoted as The terminal uses formula (3) based on the measured sub-band coefficients. Calculate the energy of each measured subband. The frequency domain measured eigenvectors are obtained. .

[0067] Formula (3)

[0068] Furthermore, the terminal reads the degree of the target node from the distribution network topology. Maximum degree in distribution network topology Wherein, degree refers to the number of lines directly connected to the target node; the terminal uses formula (4) to determine the degree of the target node. and maximum degree Calculate normalized degree Normalized degree As topological feature vectors .

[0069] Formula (4)

[0070] Optionally, the terminal will use the time-domain measured feature vector 2. Measured eigenvectors in the frequency domain and topological eigenvectors Normalization is performed separately to obtain the normalized time-domain measured eigenvectors. Normalized frequency domain measured eigenvectors and normalized topological feature vectors The terminal will normalize the time-domain measured feature vector. Normalized frequency domain measured eigenvectors and normalized topological feature vectors By concatenating the components, we obtain the first feature vector. .

[0071] By integrating the characteristics of the time domain, electromagnetic spectrum domain, and distribution network topology domain in multiple dimensions, the system can comprehensively capture power frequency interference and broadband transient noise, significantly improving the accuracy of power line communication signal noise detection in the distribution network.

[0072] Optionally, the terminal will use the first feature vector of the target node in the distribution network to... Input the pre-trained initial noise detection model and obtain the first predicted noise intensity and the first predicted noise intensity of the initial noise detection model. The corresponding first confidence level The initial noise detection model can be a graph convolutional network model, denoted as... First confidence level This reflects the reliability of the initial noise detection model for the detection results of the first feature vector.

[0073] Furthermore, the terminal will acquire the first feature vector of the target node in the distribution network. Before inputting the pre-trained initial noise detection model, the process also includes plotting several key measurement points (such as smart meters and branch nodes) in the distribution network as nodes in a graph, with a total number of nodes. Each node The features are the sample feature vectors ,in .in, Let represent the d-dimensional real space. Construct the adjacency matrix of the undirected graph based on the actual wiring relationships of the distribution network. As shown in formula (5). Furthermore, self-loops are introduced during training and inference. And calculate the degree matrix. .in, Adjacency matrix A refers to an N×N identity matrix, where N is the number of nodes in the graph. In graph neural networks (such as GCN), the adjacency matrix A represents the connections between nodes. However, without self-loops, where each node is not connected to itself, its own features are ignored during information aggregation. Therefore, adding self-loops ensures that each node considers its own features when aggregating neighbor information at each layer, ensuring that each layer includes the node's own representation, which is beneficial for feature preservation and gradient propagation.

[0074] Formula (5)

[0075] Taking a two-layer GCN as an example, define the layer The hidden features are ,in, In a graph convolutional network (GCN), the first... The dimension of the hidden representation (features) of the layer, assuming the hidden dimension of the first layer GCN output is set to... =64, that is, the number of nodes is 64. The layer output feature vector length is 64; d represents the input feature dimension of each graph node. Initial layer As shown in formula (6), the propagation formula for each layer is shown in formula (7).

[0076] Formula (6)

[0077] Formula (7)

[0078] In the formula, It is a non-linear activation function (such as ReLU). For the first Layer learnable weights, let , The dimension represents the final graph node. h1 represents the feature dimension of layer 0 (initial layer) of the graph convolutional network, and h2 represents the feature dimension of layer 2 of the graph convolutional network. Indicates a self-loop; Representation degree matrix.

[0079] The terminal hides its representation from all nodes. Perform global average pooling to obtain the global graph feature vector. As shown in formula (8).

[0080] Formula (8)

[0081] In the formula, N is the total number of nodes. Hidden representation of identifier nodes, Let h represent the h-dimensional real space.

[0082] Furthermore, a fully connected regression head is used to output the predicted noise intensity of the sample feature vector. The specific details are shown in formula (9).

[0083] Formula (9)

[0084] In the formula, Indicates the predicted noise intensity. This represents a set of fully connected layer weight parameters used for noise intensity estimation (regression task). This is the noise intensity bias term, where T represents the transpose of a vector or matrix, used to... Transpose it into a row vector, and it can be used with the global graph feature vector. Perform the dot product.

[0085] Furthermore, a fully connected layer with a sigmoid function is used to output the confidence score. As shown in formula (10).

[0086] Formula (10)

[0087] In the formula, The unnormalized score representing the confidence level is calculated as shown in formula (11).

[0088] , Formula (11)

[0089] In the formula, This represents a set of fully connected layer weight parameters used for confidence. It is the confidence bias term. This represents the feature vector of the global graph.

[0090] Terminal to sample dataset Training is performed using a multi-task loss function, and the loss function is... As shown in formula (12). Where, This represents the true noise intensity and the true confidence label.

[0091] Formula (12)

[0092] In the formula, For batch size, Hyperparameters for balancing regression and classification losses; This represents the difference between the predicted noise intensity of the i-th sample and the actual noise intensity of the i-th sample; The true confidence label represents the i-th sample; This represents the prediction confidence level for the i-th sample.

[0093] Step S104: If the first confidence level is less than the confidence level threshold, the transient voltage and transient current are simulated and corrected using the initial digital twin model to obtain the first error value.

[0094] The initial digital twin model is determined based on the measured line parameters of each node in the distribution network, including conductor impedance. Branch reactor and load impedance The terminal is connected to the impedance of the wire. Branch reactor and load impedance Reconstruct the equivalent physical network model, which is the initial digital twin model.

[0095] Furthermore, conductor impedance The determination methods include: determining the conductor impedance based on the conductor resistance per unit length, conductor reactance, and line length. The terminal is determined based on formula (13) using the conductor resistance per unit length. Conductor reactance and line length Determine the impedance of the conductor .

[0096] Formula (13)

[0097] In the formula, j is the imaginary unit.

[0098] For each branch, its load impedance is collected. and branch length The branch reactance is calculated using formula (14). .

[0099] Formula (14)

[0100] Furthermore, by utilizing SCADA systems or power distribution network GIS data, a node-branch adjacency matrix can be constructed. In digital twin software (such as MATLAB / Simulink, DIgSILENT, PowerFactory, etc.), based on Construct a multi-node network, and combine the above , and Enter the parameters as network element parameters and configure the simulation step size. (e.g., 0.1ms), simulation duration and acquisition window length To maintain consistency, the original digital twin model is constructed. A first predictive noise intensity is then injected into the original digital twin model to obtain the initial digital twin model.

[0101] Optionally, if the first confidence level is greater than or equal to the confidence level threshold, denoted as , that is, the first confidence level... If the noise intensity is greater than or equal to the confidence threshold, it is preliminarily considered that the noise intensity output by the initial noise detection model, i.e., the graph convolutional network model, is... If it is sufficiently reliable, the first predicted noise intensity will be used as the target predicted noise intensity of the target node.

[0102] Optionally, if the first confidence level is less than the confidence level threshold, denoted as , that is, the first confidence level... If the error value is less than or equal to the confidence threshold, the initial noise detection model is considered unreliable. In this case, an initial digital twin model needs to be introduced to simulate and correct the transient voltage and transient current of the target node to obtain the first error value.

[0103] It is important to note the simulation time step. Aligned with the actual sampling time, if there is a discrepancy between the two clocks, it can be corrected by interpolation or timestamp mapping.

[0104] Step S106: If the first error value is greater than the error threshold, at least one of the initial noise detection model, the first feature vector, and the initial digital twin model is corrected according to the first error value to obtain the first correction result.

[0105] Optionally, if the first error value is greater than the error threshold, the terminal corrects at least one of the initial noise detection model, the first feature vector, and the initial digital twin model based on the parameters in the first error value to obtain a first correction result. For example, the first feature vector is corrected; another example is the initial noise detection model and the first feature vector are corrected; yet another example is that the initial noise detection model, the first feature vector, and the initial digital twin model are all corrected.

[0106] Step S108: Based on the first correction result, obtain the target predicted noise intensity of the target node.

[0107] Optionally, the terminal corrects the first feature vector based on the first correction result, for example, and then inputs the corrected first feature vector back into the initial noise detection model. The resulting second noise detection intensity is the target predicted noise intensity. Alternatively, if the initial noise detection model and the first feature vector are corrected, the corrected first feature vector is input back into the corrected initial noise detection model. The resulting second noise detection intensity is the target predicted noise intensity.

[0108] Step S110: If the target prediction noise intensity of the target node is greater than or equal to the alarm threshold, send an alarm message.

[0109] Optionally, if the target predicts noise intensity ≥ alarm threshold If a noise alarm is triggered, the terminal records key information, which may include the trigger time, the triggered alarm threshold, location information, and the first error value. The terminal then sends an alarm message.

[0110] In the above noise detection method, firstly, when the first confidence level is less than a confidence threshold, the initial noise detection model outputs a first predicted noise intensity and a corresponding first confidence level based on a first feature vector; an initial digital twin model corresponding one-to-one with the actual line parameters is introduced, and closed-loop simulation correction is performed in combination with the first confidence level to obtain a first error value for simulation verification; secondly, when the first error value is greater than the error threshold, at least one of the initial noise detection model, the first feature vector, and the initial digital twin model is corrected according to the first error value to obtain a first correction result. The physical parameters of the first feature vector, the initial noise detection model, and the initial digital twin model are dynamically adjusted in the process to maintain high robustness and accuracy to environmental changes; finally, based on the dynamically adjusted first correction result, the target predicted noise intensity of the target node is obtained, and an alarm message is sent when the target predicted noise intensity is greater than or equal to the alarm threshold. This improves the accuracy of target predicted noise intensity detection and also ensures the accuracy of alarm messages based on target predicted noise intensity.

[0111] In one exemplary embodiment, such as Figure 2 As shown, the first correction result includes at least one of the corrected initial noise detection model, the corrected first feature vector, and the corrected initial digital twin model; the first error value includes multi-domain feature error; the multi-domain feature error includes time-domain feature error, frequency-domain feature error, and topological feature error; the first correction result is obtained by correcting at least one of the initial noise detection model, the first feature vector, and the initial digital twin model according to the first error value, including steps S202 to S206. Wherein:

[0112] Step S202: If the temporal feature error is greater than the temporal threshold, adjust the hyperparameters and learning rate in the initial noise detection model to obtain the corrected initial noise detection model.

[0113] Among them, the time-domain characteristic error is determined based on the root mean square of the simulated voltage in the time-domain simulation characteristic vector and the root mean square of the transient voltage in the time-domain measured characteristic vector, or the root mean square of the simulated current in the time-domain simulation characteristic vector and the root mean square of the transient current in the time-domain measured characteristic vector.

[0114] Optionally, the terminal performs simulation correction on the transient voltage and transient current using the initial digital twin model to obtain the simulated voltage and simulated current. The terminal calculates the root mean square (RMS) of the simulated voltage to obtain the RMS of the simulated voltage; the terminal calculates the RMS of the simulated current to obtain the RMS of the simulated current; and the time-domain simulation feature vector is obtained by concatenating the RMS of the simulated current and the RMS of the simulated voltage.

[0115] Furthermore, the terminal is based on the root mean square of the simulated voltage in the time-domain simulation feature vector. and the root mean square of transient voltage in the time-domain measured eigenvector The time-domain characteristic error is determined. Furthermore, the terminal uses the root mean square of the transient voltage... With simulated root mean square voltage Take the absolute value of the difference and divide by the root mean square of the transient voltage. The time-domain characteristic error is obtained, that is... / When the temporal feature error exceeds a temporal threshold, such as 50%, the terminal adjusts the hyperparameters in the initial noise detection model. The learning rate and the corrected initial noise detection model are obtained. Hyperparameters are used to balance the regression and classification losses.

[0116] Furthermore, the terminal can also base its simulation on the root mean square of the simulated current in the time-domain simulation feature vector. and the root mean square of transient current in the time-domain measured eigenvector The time-domain characteristic error is determined. Furthermore, the terminal uses the root mean square of the transient current... With simulated root mean square current Take the absolute value of the difference and divide by the root mean square of the transient current. The time-domain characteristic error is obtained, that is... When the temporal feature error exceeds a temporal threshold, such as 50%, the terminal adjusts the hyperparameters in the initial noise detection model. The learning rate and the corrected initial noise detection model are obtained. Hyperparameters are used to balance the regression and classification losses.

[0117] Step S204: If the frequency domain feature error is greater than the frequency domain threshold, target series wavelet packet decomposition is performed on the transient voltage of the target node to obtain the corrected first feature vector.

[0118] Among them, the frequency domain feature error is based on the simulated subband energy in the frequency domain simulation feature vector. Measured subband energy in frequency domain measured eigenvectors Sure.

[0119] Optionally, the terminal performs wavelet packet decomposition on the simulated voltage down to the [missing information]. The simulation sub-band coefficients are obtained at the first stage; the terminal calculates the simulation sub-band coefficients to obtain the energy of each simulation sub-band; and the frequency domain simulation feature vector is obtained based on the energy of each simulation sub-band.

[0120] Optionally, the measured subband energy With simulated subband energy The frequency domain characteristic error is determined by taking the absolute value of the difference and finding its maximum value. Assuming a frequency domain threshold, for example, set to 20%, the frequency domain characteristic error is determined when it exceeds this threshold. ≥20% indicates that the wavelet subband division is not fine enough. The terminal performs target-level wavelet packet decomposition on the transient voltage of the target node to obtain the corrected first feature vector. The target level is greater than the first target technique L, and can be L+1. Further, the terminal performs wavelet packet decomposition on the simulated voltage down to the [missing value]. The system first calculates the new simulation sub-band coefficients; then it calculates the new simulation sub-band coefficients to obtain the energy of each new simulation sub-band; based on the energy of each new simulation sub-band, it obtains a new frequency domain simulation feature vector, and replaces the frequency domain simulation feature vector with the new frequency domain simulation feature vector to obtain the corrected first feature vector.

[0121] Step S206: If the topology feature error is greater than the topology threshold, correct the branch reactance that is greater than the topology threshold in the initial digital twin model to obtain the corrected initial digital twin model.

[0122] The topology characteristic error is determined based on the sensitivity of branch reactance. This involves: while keeping other parameters constant, applying a small perturbation (e.g., 1% of its rated value) to the target parameter, rerunning the simulation, and calculating the change in the error function (e.g., the Euclidean distance between the simulated and measured eigenvectors) before and after the perturbation. Based on this, the local sensitivity of the parameter is estimated. Furthermore, all parameters are categorized, and the average sensitivity of each category is calculated, then normalized to obtain the contribution ratio of each category of parameters to the overall error. If the sensitivity ratio of the branch impedance parameter exceeds a preset threshold (e.g., 60%), it can be determined that the error is mainly concentrated in the branch impedance modeling.

[0123] The first error value includes the multi-domain feature error, denoted as... .

[0124] Optionally, if the topology feature error is greater than the topology threshold, the branch reactance greater than the topology threshold is corrected in the initial digital twin model by formula (15) to obtain the corrected initial digital twin model.

[0125] Formula (15)

[0126] In the formula, Represents the new branch reactor, Represents the impedance of the conductor; It can be determined by line search; Represents partial derivatives; This represents multi-domain feature error.

[0127] In this embodiment, the initial results of the initial noise detection model are simulated and verified using an initial digital twin model, eliminating the risks of a purely data-driven model and the problem of the purely data-driven model being detached from physical constraints, thereby improving the accuracy of noise detection.

[0128] In one exemplary embodiment, such as Figure 3 As shown, based on the first correction result, the target predicted noise intensity of the target node is obtained, including steps S302 to S304. Wherein:

[0129] Step S302: If the second confidence level is greater than or equal to the confidence level threshold, then at least one of the following is executed: using the corrected initial noise detection model as the target noise detection model and using the corrected first feature vector as the second feature vector.

[0130] The second confidence level is the output of the corrected initial noise detection model based on the first feature vector or the corrected first feature vector, or the output of the initial noise detection model based on the corrected first feature vector.

[0131] Optionally, based on the temporal feature error in the first error value, if the temporal feature error is greater than a temporal threshold, the hyperparameters and learning rate in the initial noise detection model are adjusted to obtain a corrected initial noise detection model. At this time, the terminal inputs the first feature vector into the corrected initial noise detection model to obtain the corresponding second predicted noise intensity and the second confidence level corresponding to the second predicted noise intensity. If the second confidence level is greater than or equal to the confidence threshold, it indicates that the corrected initial noise detection model is reliable, and the corrected initial noise detection model is then used as the target noise detection model.

[0132] Optionally, based on the frequency domain feature error in the first error value, if the frequency domain feature error is greater than the frequency domain threshold, a target-level wavelet packet decomposition is performed on the transient voltage of the target node to obtain the corrected first feature vector. At this time, the terminal inputs the corrected first feature vector into the initial noise detection model to obtain the corresponding second predicted noise intensity and the second confidence level corresponding to the second predicted noise intensity. If the second confidence level is greater than or equal to the confidence threshold, the corrected first feature vector is used as the second feature vector.

[0133] Optionally, if the first error value includes both time-domain feature errors greater than a time-domain threshold and frequency-domain feature errors greater than a frequency-domain threshold, the terminal performs target-level wavelet packet decomposition on the transient voltage of the target node to obtain the corrected first feature vector. The terminal adjusts the hyperparameters and learning rate in the initial noise detection model to obtain the corrected initial noise detection model. At this time, the terminal inputs the corrected first feature vector into the corrected initial noise detection model to obtain the corresponding second predicted noise intensity and the second confidence level corresponding to the second predicted noise intensity. If the second confidence level is greater than or equal to the confidence threshold, the corrected first feature vector is used as the second feature vector.

[0134] Step S304: Obtain the second predicted noise intensity corresponding to the second confidence level, and use it as the target predicted noise intensity of the target node.

[0135] Optionally, the terminal obtains the second predicted noise intensity corresponding to the second confidence level, and uses it as the target predicted noise intensity of the target node.

[0136] In this embodiment, by performing at least one of correcting the first feature vector and correcting the initial noise detection model, the digital twin simulation and the initial noise detection model are synergistically optimized, thereby improving the accuracy of noise detection.

[0137] In one exemplary embodiment, such as Figure 4 As shown, based on the first correction result, the target predicted noise intensity of the target node is obtained, including steps S402 to S404. Wherein:

[0138] In step S402, if the first confidence level is less than the confidence level threshold, the transient voltage and transient current are simulated and corrected using the corrected initial digital twin model to obtain the second error value.

[0139] Optionally, if the first confidence level is less than the confidence threshold, indicating that the initial noise detection model is unreliable, the terminal corrects the branch reactance exceeding the topological threshold in the initial digital twin model based on the first error value, obtaining a corrected initial digital twin model. The terminal then uses the corrected initial digital twin model to simulate and correct the transient voltage and transient current, obtaining a second error value.

[0140] Step S404: If the second error value is less than or equal to the error threshold, the first predicted noise intensity is taken as the target predicted noise intensity of the target node.

[0141] The second error value includes multi-domain feature error. and noise intensity error .

[0142] Optionally, if the second error value is less than or equal to the error threshold, the terminal uses the first predicted noise intensity as the target predicted noise intensity of the target node.

[0143] Optionally, if the second error value is greater than the error threshold, that is, in the case of multi-domain feature error... Greater than the multi-domain error threshold and noise intensity error If at least one of the noise intensity error thresholds is greater than the second error value, the terminal recalibrates at least one of the corrected initial noise detection model, the corrected first feature vector, and the corrected initial digital twin model based on the second error value to obtain the second calibration result; based on the second calibration result, the target predicted noise intensity of the target node is obtained.

[0144] The method also includes step S406, which uses the corrected initial digital twin model as the target digital twin model.

[0145] The target digital twin model is used to update the measured line parameters of each node in the distribution network.

[0146] Optionally, the terminal uses the corrected initial digital twin model as the target digital twin model. The target digital twin model is used to update the measured line parameters of each node in the distribution network to minimize the error between the digital twin model and the measured line parameters.

[0147] In this embodiment, by correcting only the digital twin model, the minimum error between the digital twin model and the measured line parameters can be reduced, thereby improving the accuracy of the predicted noise intensity output by the digital twin model collaborative noise reduction detection model.

[0148] In one exemplary embodiment, such as Figure 5 As shown, based on the first correction result, the target predicted noise intensity of the target node is obtained, including steps S502 to S504. Wherein:

[0149] In step S502, if the second confidence level is less than the confidence level threshold, the transient voltage and transient current are simulated and corrected using the corrected initial digital twin model to obtain the second error value.

[0150] The corrected initial digital twin model is injected with a second predicted noise intensity; the second predicted noise intensity corresponds to a second confidence level; the second confidence level is the output of the corrected initial noise detection model based on the first feature vector or the corrected first feature vector, or the output of the initial noise detection model based on the corrected first feature vector.

[0151] Optionally, if the terminal fails to satisfy the second confidence level being greater than or equal to the confidence threshold after correcting at least one of the initial noise detection model and the first feature vector, then the terminal needs to superimpose the verification from the simulation of the corrected initial digital twin model; the transient voltage and transient current are then simulated and corrected using the corrected initial digital twin model to obtain the second error value. Further judgment is then made based on the second error value.

[0152] Step S504: If the second error value is less than or equal to the error threshold, the second predicted noise intensity is taken as the target predicted noise intensity of the target node.

[0153] The second error value includes multi-domain feature error and noise intensity error.

[0154] Optionally, if the second error value is less than or equal to the error threshold, the second predicted noise intensity is taken as the target predicted noise intensity of the target node.

[0155] Optionally, if the second error value is greater than the error threshold, that is, in the case of multi-domain feature error... Greater than the multi-domain error threshold and noise intensity error If at least one of the noise intensity error thresholds is greater than the target noise intensity of the target node, the second predicted noise intensity is used as the target predicted noise intensity. The terminal recalibrates at least one of the calibrated initial noise detection model, the calibrated first feature vector, and the calibrated initial digital twin model based on the second error value to obtain the second calibration result; based on the second calibration result, the target predicted noise intensity of the target node is obtained.

[0156] The method also includes step S506, which uses the corrected initial digital twin model as the target digital twin model.

[0157] The target digital twin model is used to update the measured line parameters of each node in the distribution network.

[0158] In this embodiment, the accuracy of noise detection is improved by synergistic optimization of digital twin simulation and initial noise detection model through correction of at least one of the first feature vector and correction of the initial noise detection model, as well as by superimposing the correction of the digital twin model.

[0159] In an exemplary embodiment, the method further includes: if the second error value is greater than the error threshold, recalibrating at least one of the calibrated initial noise detection model, the calibrated first feature vector, and the calibrated initial digital twin model according to the second error value to obtain a second calibration result; and obtaining the target predicted noise intensity of the target node according to the second calibration result.

[0160] Optionally, whether the terminal only corrects the initial digital twin model or corrects at least one of the first feature vector and the initial noise detection model, as well as the superimposed correction of the digital twin model, there is a second error value that is greater than the error threshold. In this case, it is necessary to correct at least one of the corrected initial noise detection model, the corrected first feature vector, and the corrected initial digital twin model again based on the second error value to obtain the second correction result; based on the second correction result, the target predicted noise intensity of the target node is obtained.

[0161] It is important to note that a maximum number of calibrations needs to be set. (For example, 5 times). If the result is still not greater than the confidence threshold or less than the error threshold, an anomaly can be recorded and reported for manual intervention.

[0162] In this embodiment, under the condition that the initial noise detection model, the first feature vector, and the initial digital twin model are calibrated multiple times, the digital twin simulation and the initial noise detection model can be synergistically optimized, thereby improving the accuracy of noise detection.

[0163] In one exemplary embodiment, such as Figure 6 As shown, the transient voltage and transient current are simulated and corrected using an initial digital twin model to obtain the first error value, including steps S602 to S608. Wherein:

[0164] Step S602: Simulate the transient voltage and transient current using the initial digital twin model to obtain the simulated voltage and simulated current.

[0165] Optionally, the terminal performs simulation processing on transient voltage and transient current using an initial digital twin model to obtain simulated voltage and simulated current.

[0166] Step S604: Based on the simulated current and simulated voltage, obtain the time-domain simulation feature vector and the frequency-domain simulation feature vector.

[0167] Optionally, the terminal performs root mean square (RMS) calculation on the simulated voltage to obtain the RMS of the simulated voltage; the terminal performs RMS calculation on the simulated current to obtain the RMS of the simulated current; and the time-domain simulation feature vector is obtained by concatenating the RMS of the simulated current and the RMS of the simulated voltage.

[0168] Optionally, the terminal performs wavelet packet decomposition on the simulated voltage down to the [missing information]. The simulation sub-band coefficients are obtained at the first stage; the terminal calculates the simulation sub-band coefficients to obtain the energy of each simulation sub-band; and the frequency domain simulation feature vector is obtained based on the energy of each simulation sub-band.

[0169] Step S606: Based on the time-domain simulation feature vector, the frequency-domain simulation feature vector, and the topology simulation features of the target node, determine the simulation feature vector and the simulation noise intensity.

[0170] Optionally, the terminal concatenates the time-domain simulation feature vector, the frequency-domain simulation feature vector, and the topology simulation feature of the target node together to determine the simulation feature vector; then, the terminal inputs the simulation feature vector into the initial noise detection model to obtain the simulation noise intensity.

[0171] Step S608: Determine the first error value based on the simulation feature vector, the first feature vector, the simulation noise intensity, and the first predicted noise intensity.

[0172] Optionally, the terminal uses the simulation feature vector according to formula (16). and the first eigenvector Determine the multi-domain feature error .

[0173] Formula (16)

[0174] The terminal calculates the simulated noise intensity based on formula (17). and predict noise intensity Determine the noise intensity error .

[0175] Formula (17)

[0176] Terminal based on multi-domain feature error and noise intensity error Determine the first error value.

[0177] Optionally, in The multi-domain feature error is greater than the multi-domain error threshold and the noise intensity error. If at least one of the noise intensity error thresholds is greater than the first error value, at least one of the initial noise detection model, the first feature vector, and the initial digital twin model is corrected according to the first error value to obtain the first correction result.

[0178] In this embodiment, secondary decision-making is performed using multi-domain feature errors and intensity errors, which can eliminate the risks of pure data-driven models and the problem of pure data-driven models deviating from physical constraints, thereby improving the accuracy of noise detection.

[0179] In one embodiment, such as Figure 7 As shown, transient voltage, transient current, and topological features are sampled. The time-domain measured feature vector and the frequency-domain measured feature vector are determined based on the transient voltage and transient current. The time-domain measured feature vector, the frequency-domain measured feature vector, and the topological feature vector are concatenated to obtain the first feature vector. The first feature vector is input into the initial noise detection model to obtain the first predicted noise intensity and the first confidence level corresponding to the first predicted noise intensity output by the initial noise detection model.

[0180] Case 1: If the first confidence level is greater than or equal to the confidence threshold, the first predicted noise intensity is taken as the target noise intensity.

[0181] Scenario 2: When the first confidence level is less than the confidence threshold, the transient voltage and transient current are simulated and corrected using the initial digital twin model to obtain the first error value. The first error value includes multi-domain characteristic error and noise intensity error. The multi-domain characteristic error includes time-domain characteristic error, frequency-domain characteristic error, and topological characteristic error. Sub-scenario 1: Satisfies multi-domain characteristic error. Greater than the multi-domain error threshold and noise intensity error If at least one of the noise intensity error thresholds is greater than the frequency domain feature error, and the target transient voltage is decomposed into a target-level wavelet packet based on the transient voltage of the target node, the corrected first feature vector is obtained. If the time domain feature error is greater than the time domain threshold, the hyperparameters and learning rate in the initial noise detection model are adjusted to obtain the corrected initial noise detection model. The corrected first feature vector is input into the corrected initial noise detection model to obtain the second confidence level and the corresponding second predicted noise intensity. If the second confidence level is less than the confidence threshold, the transient voltage and transient current are simulated and corrected using the corrected initial digital twin model to obtain the second error value. If the second error value is greater than the error threshold, the corrected initial digital twin model is corrected again based on the second error value to obtain the second corrected initial digital twin model. The multi-domain feature error and noise intensity error between the corrected first feature vector and the simulation results are calculated again using the second corrected initial digital twin model. Less than or equal to the multi-domain error threshold and noise intensity error If the noise intensity error threshold is less than or equal to the noise intensity error threshold, the second predicted noise intensity is used as the target predicted noise intensity of the target node. Here, the corrected initial noise model is used as the target noise model, the corrected first initial feature vector is used as the second feature vector, and the second predicted noise intensity is the output of the target noise model based on the second feature vector.

[0182] Sub-case 2: In multi-domain feature errors Less than or equal to the multi-domain error threshold and noise intensity error If the noise intensity error threshold is less than or equal to the first predicted noise intensity, the first predicted noise intensity is taken as the target noise intensity.

[0183] Target Predicted Noise Intensity ≥ alarm threshold If a noise alarm is triggered, the terminal records key information, which may include the trigger time, the triggered alarm threshold, location information, and the first error value. The terminal then sends an alarm message.

[0184] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0185] Based on the same inventive concept, this application also provides a noise detection device for implementing the noise detection method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more noise detection device embodiments provided below can be found in the limitations of the noise detection method described above, and will not be repeated here.

[0186] In one exemplary embodiment, such as Figure 8 As shown, a noise detection device is provided, including: an acquisition module 801, a first error determination module 802, a correction module 803, a result determination module 804, and an alarm module 805, wherein:

[0187] The acquisition module 801 is used to input the first feature vector of the target node in the distribution network into the initial noise detection model to obtain the first predicted noise intensity and the first confidence level corresponding to the first predicted noise intensity of the initial noise detection model; wherein, the first feature vector is determined based on transient voltage, transient current and topological features.

[0188] The first error determination module 802 is used to perform simulation correction on transient voltage and transient current through an initial digital twin model when the first confidence level is less than the confidence level threshold, so as to obtain a first error value; wherein, the initial digital twin model is determined based on the measured line parameters of each node in the distribution network and a digital twin model with the first predicted noise intensity is injected.

[0189] The correction module 803 is used to correct at least one of the initial noise detection model, the first feature vector, and the initial digital twin model based on the first error value when the first error value is greater than the error threshold, so as to obtain a first correction result.

[0190] The result determination module 804 is used to obtain the target predicted noise intensity of the target node based on the first correction result.

[0191] The alarm module 805 is used to send alarm information when the target prediction noise intensity of the target node is greater than or equal to the alarm threshold.

[0192] In an exemplary embodiment, the first correction result includes at least one of the corrected initial noise detection model, the corrected first feature vector, and the corrected initial digital twin model; the first error value includes multi-domain feature error; the multi-domain feature error includes time-domain feature error, frequency-domain feature error, and topological feature error; the correction module 803 is used to adjust the hyperparameters and learning rate in the initial noise detection model to obtain the corrected initial noise detection model when the time-domain feature error is greater than the time-domain threshold; wherein, the time-domain feature error is determined based on the root mean square of the simulated voltage in the time-domain simulated feature vector and the root mean square of the transient voltage in the time-domain measured feature vector, or the root mean square of the simulated current in the time-domain simulated feature vector and the root mean square of the transient current in the time-domain measured feature vector.

[0193] The correction module 803 is also used to obtain the corrected first feature vector by performing target series wavelet packet decomposition on the transient voltage of the target node when the frequency domain feature error is greater than the frequency domain threshold; wherein, the frequency domain feature error is determined based on the simulated subband energy in the frequency domain simulated feature vector and the measured subband energy in the frequency domain measured feature vector.

[0194] The correction module 803 is also used to correct the branch reactance that is greater than the topology threshold in the initial digital twin model when the topology feature error is greater than the topology threshold, so as to obtain the corrected initial digital twin model; wherein the topology feature error is determined based on the sensitivity of the branch reactance.

[0195] In an exemplary embodiment, the result determination module 804 is further configured to, if the second confidence level is greater than or equal to the confidence level threshold, perform at least one of using the corrected initial noise detection model as the target noise detection model and using the corrected first feature vector as the second feature vector; wherein, the second confidence level is the output of the corrected initial noise detection model based on the first feature vector or the corrected first feature vector, or the output of the initial noise detection model based on the corrected first feature vector; and obtain the second predicted noise intensity corresponding to the second confidence level as the target predicted noise intensity of the target node.

[0196] In an exemplary embodiment, the result determination module 804 is further configured to, when the first confidence level is less than the confidence level threshold, perform simulation correction on the transient voltage and transient current through the corrected initial digital twin model to obtain a second error value; and when the second error value is less than or equal to the error threshold, use the first predicted noise intensity as the target predicted noise intensity of the target node.

[0197] The noise detection device also includes an update module, which is used to use the corrected initial digital twin model as the target digital twin model; wherein, the target digital twin model is used to update the measured line parameters of each node in the distribution network.

[0198] In an exemplary embodiment, the result determination module 804 is further configured to, when the second confidence level is less than the confidence level threshold, perform simulation correction on the transient voltage and transient current using the corrected initial digital twin model to obtain a second error value; wherein, the corrected initial digital twin model is injected with a second predicted noise intensity; the second predicted noise intensity corresponds to the second confidence level; the second confidence level is the output of the corrected initial noise detection model based on the first feature vector or the corrected first feature vector, or the output of the initial noise detection model based on the corrected first feature vector; when the second error value is less than or equal to the error threshold, the second predicted noise intensity is used as the target predicted noise intensity of the target node.

[0199] The noise detection device also includes an update module, which is used to use the corrected initial digital twin model as the target digital twin model; wherein, the target digital twin model is used to update the measured line parameters of each node in the distribution network.

[0200] In an exemplary embodiment, the noise detection device further includes a loop execution module, configured to, when the second error value is greater than the error threshold, recalibrate at least one of the calibrated initial noise detection model, the calibrated first feature vector, and the calibrated initial digital twin model according to the second error value to obtain a second calibration result; and obtain the target predicted noise intensity of the target node according to the second calibration result.

[0201] In an exemplary embodiment, the first error determination module 802 is further configured to perform simulation processing on transient voltage and transient current through an initial digital twin model to obtain simulated voltage and simulated current; based on the simulated current and simulated voltage, obtain time-domain simulation feature vector and frequency-domain simulation feature vector; based on the time-domain simulation feature vector, frequency-domain simulation feature vector and topology simulation features of the target node, determine the simulation feature vector and simulation noise intensity; and based on the simulation feature vector, the first feature vector, the simulation noise intensity and the first predicted noise intensity, determine the first error value.

[0202] Each module in the aforementioned noise detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0203] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores power line communication data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements a noise detection method.

[0204] Those skilled in the art will understand that Figure 9The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0205] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0206] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0207] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0208] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0209] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0210] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A noise detection method, characterized in that, The method includes: The first feature vector of the target node in the distribution network is obtained and input into the initial noise detection model to obtain the first predicted noise intensity and the first confidence level corresponding to the first predicted noise intensity. The first feature vector is determined based on transient voltage, transient current and topological features. If the first confidence level is less than the confidence level threshold, the transient voltage and the transient current are simulated and corrected using an initial digital twin model to obtain a first error value; wherein, the initial digital twin model is determined based on the measured line parameters of each node in the distribution network and is a digital twin model into which the first predicted noise intensity is injected. If the first error value is greater than the error threshold, at least one of the initial noise detection model, the first feature vector, and the initial digital twin model is corrected based on the first error value to obtain a first correction result; Based on the first correction result, the target predicted noise intensity of the target node is obtained; An alarm message is sent if the target predicted noise intensity at the target node is greater than or equal to the alarm threshold.

2. The method according to claim 1, characterized in that, The first correction result includes at least one of the corrected initial noise detection model, the corrected first feature vector, and the corrected initial digital twin model; the first error value includes multi-domain feature error; the multi-domain feature error includes time-domain feature error, frequency-domain feature error, and topological feature error; the step of correcting at least one of the initial noise detection model, the first feature vector, and the initial digital twin model according to the first error value to obtain the first correction result includes: If the time-domain feature error is greater than the time-domain threshold, the hyperparameters and learning rate in the initial noise detection model are adjusted to obtain the corrected initial noise detection model; wherein, the time-domain feature error is determined based on the root mean square of the simulated voltage in the time-domain simulation feature vector and the root mean square of the transient voltage in the time-domain measured feature vector, or the root mean square of the simulated current in the time-domain simulation feature vector and the root mean square of the transient current in the time-domain measured feature vector. When the frequency domain feature error is greater than the frequency domain threshold, the first feature vector after correction is obtained by performing target series wavelet packet decomposition on the transient voltage of the target node; wherein, the frequency domain feature error is determined based on the simulated subband energy in the frequency domain simulated feature vector and the measured subband energy in the frequency domain measured feature vector. If the topology feature error is greater than the topology threshold, the branch reactance that is greater than the topology threshold is corrected in the initial digital twin model to obtain a corrected initial digital twin model; wherein, the topology feature error is determined based on the sensitivity of the branch reactance.

3. The method according to claim 2, characterized in that, The step of obtaining the target predicted noise intensity of the target node based on the first correction result includes: If the second confidence level is greater than or equal to the confidence level threshold, then at least one of the following is executed: using the corrected initial noise detection model as the target noise detection model and using the corrected first feature vector as the second feature vector; wherein, the second confidence level is the output of the corrected initial noise detection model based on the first feature vector or the corrected first feature vector, or the output of the initial noise detection model based on the corrected first feature vector; Obtain the second predicted noise intensity corresponding to the second confidence level, and use it as the target predicted noise intensity of the target node.

4. The method according to claim 2, characterized in that, The step of obtaining the target predicted noise intensity of the target node based on the first correction result includes: If the first confidence level is less than the confidence threshold, the transient voltage and the transient current are simulated and corrected using the corrected initial digital twin model to obtain a second error value; If the second error value is less than or equal to the error threshold, the first predicted noise intensity is taken as the target predicted noise intensity of the target node; The method further includes: using the corrected initial digital twin model as the target digital twin model; wherein the target digital twin model is used to update the measured line parameters of each node in the distribution network.

5. The method according to claim 2, characterized in that, The step of obtaining the target predicted noise intensity of the target node based on the first correction result includes: If the second confidence level is less than the confidence level threshold, the transient voltage and the transient current are simulated and corrected using the corrected initial digital twin model to obtain a second error value; wherein, the corrected initial digital twin model is injected with a second predicted noise intensity; the second predicted noise intensity corresponds to the second confidence level; the second confidence level is the output of the corrected initial noise detection model based on the first feature vector or the corrected first feature vector, or the output of the initial noise detection model based on the corrected first feature vector; If the second error value is less than or equal to the error threshold, the second predicted noise intensity is taken as the target predicted noise intensity of the target node; The method further includes: using the corrected initial digital twin model as the target digital twin model; wherein the target digital twin model is used to update the measured line parameters of each node in the distribution network.

6. The method according to any one of claims 4 or 5, characterized in that, The method further includes: If the second error value is greater than the error threshold, at least one of the corrected initial noise detection model, the corrected first feature vector, and the corrected initial digital twin model is corrected again based on the second error value to obtain a second correction result; Based on the second correction result, the target predicted noise intensity of the target node is obtained.

7. The method according to claim 1, characterized in that, The step of simulating and correcting the transient voltage and transient current using an initial digital twin model to obtain a first error value includes: The transient voltage and transient current are simulated using the initial digital twin model to obtain the simulated voltage and simulated current. Based on the simulated current and the simulated voltage, the time-domain simulation feature vector and the frequency-domain simulation feature vector are obtained; Based on the time-domain simulation feature vector, the frequency-domain simulation feature vector, and the topology simulation features of the target node, the simulation feature vector and the simulation noise intensity are determined. The first error value is determined based on the simulated feature vector, the first feature vector, the simulated noise intensity, and the first predicted noise intensity.

8. A noise detection device, characterized in that, The device includes: The acquisition module is used to input the first feature vector of the target node in the distribution network into the initial noise detection model, and to obtain the first predicted noise intensity and the first confidence level corresponding to the first predicted noise intensity of the initial noise detection model; wherein, the first feature vector is determined based on transient voltage, transient current and topological features; The first error determination module is used to perform simulation correction on the transient voltage and the transient current through an initial digital twin model when the first confidence level is less than the confidence level threshold, so as to obtain a first error value; wherein, the initial digital twin model is determined based on the measured line parameters of each node in the distribution network and is a digital twin model of the first predicted noise intensity. The correction module is used to correct at least one of the initial noise detection model, the first feature vector, and the initial digital twin model based on the first error value when the first error value is greater than the error threshold, so as to obtain a first correction result. The result determination module is used to obtain the target predicted noise intensity of the target node based on the first correction result; The alarm module is used to send alarm information when the target predicted noise intensity of the target node is greater than or equal to the alarm threshold.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images