Abnormal energy measurement data identification method and system based on time-frequency transformation

Abstract

The invention relates to an abnormal energy measurement data identification method and system based on time-frequency transformation, and the method comprises the following steps: S1, describing a randomness feature through analyzing the certainty features of four frequency spectrum aspects: frequency spectrum center frequency, frequency spectrum bandwidth, frequency spectrum energy and frequency spectrum phase, and four statistical features: mean value, variance, inclination and kurtosis; establishing an energy measurement data feature set; s2, taking the extracted features as input features, and constructing a training data set; s3, using an SVM algorithm to train an anomaly study and judgment matching model based on the training data set; s4, predicting a to-be-tested data set by using the trained anomaly study and judgment matching model to obtain abnormal data points; and S5, calculating an abnormal score of each abnormal data point by using a mahalanobis distance to represent an abnormal degree of the data point relative to normal data, and marking the abnormal data point on the time-frequency graph. According to the method, the accuracy and robustness of abnormal data detection can be improved, visual reminding is carried out, and the detection efficiency of workers is improved.

Description

Technical Field

[0001] The present invention relates to the field of data monitoring, and in particular to a method and system for identifying abnormal energy metering data based on time-frequency transformation. Background Art

[0002] The existing means of monitoring energy metering data anomalies are single, the troubleshooting technology is not sophisticated, and the problem is not handled in a timely manner. The process of determining metering data anomalies involves many factors, the data is complicated, and it consumes a lot of manpower and material resources, and there is a risk of inaccurate judgment.

[0003] Under the existing technology, the anomaly detection methods for multi-dimensional time series data such as energy user data are generally divided into statistical methods and deep learning methods. Statistical methods mainly include mean method, forward method, and hot card filling method. Although statistical methods are relatively simple, when the abnormal part of the data accounts for a large proportion of the overall data, the accuracy of statistics will be greatly reduced. Deep learning methods usually require a large amount of labeled data to train the model, and may have insufficient generalization ability when the amount of data is small or the data distribution is uneven. Summary of the invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide a method for identifying abnormal energy metering data based on time-frequency transformation, which can improve the accuracy and robustness of abnormal data detection.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for identifying abnormal energy metering data based on time-frequency transformation includes the following steps:

[0007] S1: By analyzing the deterministic characteristics of four spectrum aspects, namely, spectrum center frequency, spectrum bandwidth, spectrum energy and spectrum phase, and the four statistical characteristics of mean, variance, slope and kurtosis, the random characteristics are described to establish the energy metering data feature set;

[0008] S2: Use the extracted features as input features, mark normal data and abnormal data, and build a training data set;

[0009] S3 uses the SVM algorithm to train the abnormality analysis and matching model based on the training data set, and separates normal data from abnormal data by finding the optimal hyperplane;

[0010] S4: Use the trained anomaly analysis and matching model to predict the test data set to obtain abnormal data points;

[0011] S5: Use the Mahalanobis distance to calculate the anomaly score of each abnormal data point, indicating the degree of abnormality of the data point relative to normal data, and mark the abnormal data point on the time-frequency graph, and select the color, shape and size of the mark according to the size of the anomaly score.

[0012] Furthermore, the step S1 is specifically as follows: obtaining energy metering data of the power system and performing preprocessing; using Fourier transform to transform the preprocessed energy metering data of the power system, converting the time domain data into time-frequency domain data, and extracting features from the time-frequency domain data, including four deterministic features of the spectrum, namely, spectrum center frequency, spectrum bandwidth, spectrum energy, and spectrum phase, and calculating four statistical features, namely, mean, variance, slope, and kurtosis.

[0013] Furthermore, the preprocessing includes data cleaning, denoising and data format conversion.

[0014] Furthermore, the Fourier transform is used to transform the pre-processed power system energy metering data, convert the time domain data into time-frequency domain data, and extract features from the time-frequency domain data, specifically:

[0015] The pre-processed energy metering data is converted from the time domain to the frequency domain using discrete Fourier transform to obtain spectrum information;

[0016] Perform discrete Fourier transform on the signal x[n] of length N to obtain the spectrum X[k]:

[0017]

[0018] Among them, k is the frequency index and t is the time index;

[0019] Calculate the energy feature E based on the spectrum information, which represents the total energy of the signal;

[0020]

[0021] Analyze spectrum information and extract spectrum features, including spectrum peak f peak , spectrum center f centroid

[0022] f peak = arg max k |X[k]|

[0023]

[0024] Draw a time-frequency diagram based on the spectrum information to obtain time-frequency characteristics, including instantaneous energy and instantaneous frequency: E instan [t]=|X[t]| 2 ;

[0025]

[0026] Here, θ[t] is the phase information.

[0027] Furthermore, the S3 is specifically:

[0028] The goal of SVM is to find an optimal hyperplane that maximizes the distance between normal data points and the hyperplane, while keeping abnormal data points away from the hyperplane. The optimization problem of SVM is expressed as:

[0029]

[0030] subject to you i (w T x i +b)≥1-ξ i , for i = 1, 2, ..., n

[0031] Where w is the normal vector of the hyperplane, b is the bias term, ξ is the slack variable, and C is the penalty parameter;

[0032] Select the kernel function as Gaussian kernel function:

[0033] K(x,x′)=exp(-γ||xx′|| 2 )

[0034] Among them, x and x′ are eigenvectors, and γ is the parameter of the Gaussian kernel function;

[0035] The SVM model is trained based on the training set, and the Bayesian optimization method is used to find the optimal hyperparameter combination in the parameter space to obtain the anomaly judgment matching model.

[0036] Furthermore, the Bayesian optimization method is used to find the optimal hyperparameter combination in the parameter space, as follows:

[0037] A Gaussian process model is established to estimate the performance of the parameter combination. The Gaussian process model is expressed as:

[0038] f(θ)~GP(m(θ),k(θ,θ′));

[0039] Where f(θ) is the performance of the parameter combination θ, m(θ) is the mean function, k(θ,θ′) is the covariance function, and GP() represents the Gaussian process;

[0040] In Bayesian optimization, a Gaussian process model is used for posterior prediction, giving the next parameter combination θ next Predicted values ​​and confidence intervals for :

[0041]

[0042] in, is the predicted value; σ(θ next ) is the standard deviation, z is the confidence interval coefficient;

[0043] The next parameter combination is selected by maximizing its predicted value and uncertainty, using the following moment function:

[0044]

[0045] An abnormal energy metering data identification system based on time-frequency transformation includes a data processing module, an abnormal analysis and matching module and a visualization module;

[0046] The data processing module describes the random characteristics by analyzing the deterministic characteristics of four spectrum aspects, namely, spectrum center frequency, spectrum bandwidth, spectrum energy and spectrum phase, and the four statistical characteristics of mean, variance, slope and kurtosis, and establishes the energy metering data feature set; and uses the extracted features as input features to mark normal data and abnormal data, and construct a training data set;

[0047] The abnormal analysis and matching module is provided with a trained abnormal analysis and matching model, which predicts the test data set to obtain abnormal data points;

[0048] The visualization module uses the Mahalanobis distance to calculate the anomaly score of each abnormal data point to indicate the degree of abnormality of the data point relative to normal data, and marks the abnormal data point on the time-frequency graph, and selects the color, shape and size of the mark according to the size of the anomaly score.

[0049] Furthermore, the data processing module obtains the energy metering data of the power system and performs preprocessing; the preprocessed energy metering data of the power system is transformed by Fourier transform, the time domain data is converted into time-frequency domain data, and features are extracted from the time-frequency domain data, including four deterministic features of the spectrum, namely, the spectrum center frequency, spectrum bandwidth, spectrum energy and spectrum phase, and four statistical features, namely, mean, variance, slope and kurtosis, are calculated; the preprocessing includes data cleaning, denoising and data format conversion;

[0050] Fourier transform is used to transform the pre-processed power system energy metering data, convert the time domain data into time-frequency domain data, and extract features from the time-frequency domain data, specifically:

[0051] The pre-processed energy metering data is converted from the time domain to the frequency domain using discrete Fourier transform to obtain spectrum information;

[0052] Perform discrete Fourier transform on the signal x[n] of length N to obtain the spectrum X[k]:

[0053]

[0054] Among them, k is the frequency index and t is the time index;

[0055] Calculate the energy feature E based on the spectrum information, which represents the total energy of the signal;

[0056]

[0057] Analyze spectrum information and extract spectrum features, including spectrum peak f peak , spectrum center f centroid

[0058] f peak = arg max k |X[k]|

[0059]

[0060] Draw a time-frequency diagram based on the spectrum information to obtain time-frequency characteristics, including instantaneous energy and instantaneous frequency: E instan [t]=|X[t]| 2 ;

[0061]

[0062] Here, θ[t] is the phase information.

[0063] Furthermore, the abnormal analysis and matching model is constructed based on SVM, as follows:

[0064] The goal of SVM is to find an optimal hyperplane that maximizes the distance between normal data points and the hyperplane, while keeping abnormal data points away from the hyperplane. The optimization problem of SVM is expressed as:

[0065]

[0066] subject to you i (w T x i +b)≥1-ξ i , for i = 1, 2, ..., n

[0067] Where w is the normal vector of the hyperplane, b is the bias term, ξ is the slack variable, and C is the penalty parameter;

[0068] Select the kernel function as Gaussian kernel function:

[0069] K(x,x′)=exp(-γ||xx′|| 2 )

[0070] Among them, x and x′ are eigenvectors, and γ is the parameter of the Gaussian kernel function;

[0071] The SVM model is trained based on the training set, and the Bayesian optimization method is used to find the optimal hyperparameter combination in the parameter space to obtain the anomaly judgment matching model.

[0072] Furthermore, the Bayesian optimization method is used to find the optimal hyperparameter combination in the parameter space, as follows:

[0073] A Gaussian process model is established to estimate the performance of the parameter combination. The Gaussian process model is expressed as:

[0074] f(θ)~GP(m(θ),k(θ,θ′));

[0075] Where f(θ) is the performance of the parameter combination θ, m(θ) is the mean function, k(θ,θ′) is the covariance function, and GP() represents the Gaussian process;

[0076] In Bayesian optimization, a Gaussian process model is used for posterior prediction, giving the next parameter combination θ next Predicted values ​​and confidence intervals for :

[0077]

[0078] in, is the predicted value; σ(θ next ) is the standard deviation, z is the confidence interval coefficient;

[0079] The next parameter combination is selected by maximizing its predicted value and uncertainty, using the following moment function:

[0080]

[0081] The present invention has the following beneficial effects:

[0082] 1. The present invention combines the time-frequency transformation and support vector machine methods to more comprehensively analyze the time-frequency characteristics of energy metering data, and uses the support vector machine model to accurately identify abnormal data, which can improve the accuracy and robustness of abnormal data detection and provide more effective support for energy management and optimization;

[0083] 2. The present invention efficiently searches for the best hyperparameter combination in the parameter space through Bayesian optimization, thereby optimizing the performance of the support vector machine model. It can effectively utilize previous evaluation results, dynamically adjust the selection of parameter combinations, and improve the accuracy and reliability of the model in identifying abnormal energy metering data. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1is a flow chart of the method of the present invention;

[0085] Figure 2 Schematic diagram of feature extraction process in one embodiment of the present invention. DETAILED DESCRIPTION

[0086] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0087] refer to Figure 1 , a method for identifying abnormal energy metering data based on time-frequency transformation, comprising the following steps:

[0088] S1: By analyzing the deterministic characteristics of four spectrum aspects, namely, spectrum center frequency, spectrum bandwidth, spectrum energy and spectrum phase, and the four statistical characteristics of mean, variance, slope and kurtosis, the random characteristics are described to establish the energy metering data feature set;

[0089] S2: Use the extracted features as input features, mark normal data and abnormal data, and build a training data set;

[0090] S3 uses the SVM algorithm to train the abnormality analysis and matching model based on the training data set, and separates normal data from abnormal data by finding the optimal hyperplane;

[0091] S4: Use the trained anomaly analysis and matching model to predict the test data set to obtain abnormal data points;

[0092] S5: Use the Mahalanobis distance to calculate the anomaly score of each abnormal data point, indicating the degree of abnormality of the data point relative to normal data, and mark the abnormal data point on the time-frequency graph, and select the color, shape and size of the mark according to the size of the anomaly score.

[0093] In this embodiment, step S1 is specifically as follows: obtaining energy metering data of the power system and performing preprocessing; using Fourier transform to transform the preprocessed energy metering data of the power system, converting the time domain data into time-frequency domain data, and extracting features from the time-frequency domain data, including four deterministic features of the spectrum, namely, spectrum center frequency, spectrum bandwidth, spectrum energy, and spectrum phase, and calculating four statistical features, namely, mean, variance, slope, and kurtosis.

[0094] In this embodiment, the preprocessing includes data cleaning, denoising and data format conversion.

[0095] In this embodiment, Fourier transform is used to transform the pre-processed power system energy metering data, convert the time domain data into time-frequency domain data, and extract features from the time-frequency domain data, specifically:

[0096] The pre-processed energy metering data is converted from the time domain to the frequency domain using discrete Fourier transform to obtain spectrum information;

[0097] Perform discrete Fourier transform on the signal x[n] of length N to obtain the spectrum X[k]:

[0098]

[0099] Among them, k is the frequency index and t is the time index;

[0100] Calculate the energy feature E based on the spectrum information, which represents the total energy of the signal;

[0101]

[0102] Analyze spectrum information and extract spectrum features, including spectrum peak f peak , spectrum center f centroid

[0103] f peak = argmax k |X[k]|

[0104]

[0105] Draw a time-frequency diagram based on the spectrum information to obtain time-frequency characteristics, including instantaneous energy and instantaneous frequency: E instan [t]=|X[t]| 2 ;

[0106]

[0107] Here, θ[t] is the phase information.

[0108] In this embodiment, S3 is specifically:

[0109] The goal of SVM is to find an optimal hyperplane that maximizes the distance between normal data points and the hyperplane, while keeping abnormal data points away from the hyperplane. The optimization problem of SVM is expressed as:

[0110]

[0111] subject to you i (w T x i +b)≥1-ξ i , for i = 1, 2, ..., n

[0112] Where w is the normal vector of the hyperplane, b is the bias term, ξ is the slack variable, and C is the penalty parameter;

[0113] Select the kernel function as Gaussian kernel function:

[0114] K(x,x′)=exp(-γ||xx′|| 2 )

[0115] Among them, x and x′ are eigenvectors, and γ is the parameter of the Gaussian kernel function;

[0116] The SVM model is trained based on the training set, and the Bayesian optimization method is used to find the optimal hyperparameter combination in the parameter space to obtain the anomaly judgment matching model.

[0117] In this embodiment, the Bayesian optimization method is used to find the optimal hyperparameter combination in the parameter space, as follows:

[0118] A Gaussian process model is established to estimate the performance of the parameter combination. The Gaussian process model is expressed as:

[0119] f(θ)~GP(m(θ),k(θ,θ′));

[0120] Where f(θ) is the performance of the parameter combination θ, m(θ) is the mean function, k(θ,θ′) is the covariance function, and GP() represents the Gaussian process;

[0121] In Bayesian optimization, a Gaussian process model is used for posterior prediction, giving the next parameter combination θ next Predicted values ​​and confidence intervals for :

[0122]

[0123] in, is the predicted value; σ(θ next ) is the standard deviation, z is the confidence interval coefficient;

[0124] The next parameter combination is selected by maximizing its predicted value and uncertainty, using the following moment function:

[0125]

[0126] This embodiment also provides an abnormal energy metering data identification system based on time-frequency transformation, including a data processing module, an abnormal analysis and matching module, and a visualization module;

[0127] The data processing module describes the random characteristics by analyzing the deterministic characteristics of four spectrum aspects, namely, spectrum center frequency, spectrum bandwidth, spectrum energy and spectrum phase, and the four statistical characteristics of mean, variance, slope and kurtosis, and establishes the energy metering data feature set; and uses the extracted features as input features to mark normal data and abnormal data, and construct a training data set;

[0128] The abnormal analysis and matching module is provided with a trained abnormal analysis and matching model, which predicts the test data set to obtain abnormal data points;

[0129] The visualization module uses the Mahalanobis distance to calculate the anomaly score of each abnormal data point to indicate the degree of abnormality of the data point relative to normal data, and marks the abnormal data point on the time-frequency graph, and selects the color, shape and size of the mark according to the size of the anomaly score.

[0130] In this embodiment, preferably, the data processing module obtains the energy metering data of the power system and performs preprocessing; the preprocessed energy metering data of the power system is transformed by Fourier transform, the time domain data is converted into time-frequency domain data, and features are extracted from the time-frequency domain data, including four spectrum deterministic features of spectrum center frequency, spectrum bandwidth, spectrum energy and spectrum phase, and four statistical features of mean, variance, slope and kurtosis are calculated; the preprocessing includes data cleaning, denoising and data format conversion;

[0131] Fourier transform is used to transform the pre-processed power system energy metering data, convert the time domain data into time-frequency domain data, and extract features from the time-frequency domain data, specifically:

[0132] The pre-processed energy metering data is converted from the time domain to the frequency domain using discrete Fourier transform to obtain spectrum information;

[0133] Perform discrete Fourier transform on the signal x[n] of length N to obtain the spectrum X[k]:

[0134]

[0135] Among them, k is the frequency index and t is the time index;

[0136] Calculate the energy feature E based on the spectrum information, which represents the total energy of the signal;

[0137]

[0138] Analyze spectrum information and extract spectrum features, including spectrum peak f peak , spectrum center f centroid

[0139] f peak = arg max k |X[k]|

[0140]

[0141] Draw a time-frequency diagram based on the spectrum information to obtain time-frequency characteristics, including instantaneous energy and instantaneous frequency: E instan [t]=|X[t]|2 ;

[0142]

[0143] Here, θ[t] is the phase information.

[0144] In this embodiment, preferably, the abnormality analysis and matching model is constructed based on SVM, which is as follows:

[0145] The goal of SVM is to find an optimal hyperplane that maximizes the distance between normal data points and the hyperplane, while keeping abnormal data points away from the hyperplane. The optimization problem of SVM is expressed as:

[0146]

[0147] subject to you i (w T x i +b)≥1-ξ i , for i = 1, 2, ..., n

[0148] Where w is the normal vector of the hyperplane, b is the bias term, ξ is the slack variable, and C is the penalty parameter;

[0149] Select the kernel function as Gaussian kernel function:

[0150] K(x,x′)=exp(-γ||xx′|| 2 )

[0151] Among them, x and x′ are eigenvectors, and γ is the parameter of the Gaussian kernel function;

[0152] The SVM model is trained based on the training set, and the Bayesian optimization method is used to find the optimal hyperparameter combination in the parameter space to obtain the anomaly judgment matching model.

[0153] In this embodiment, preferably, a Bayesian optimization method is used to find the optimal hyperparameter combination in the parameter space, as follows:

[0154] A Gaussian process model is established to estimate the performance of the parameter combination. The Gaussian process model is expressed as:

[0155] f(θ)~GP(m(θ),k(θ,θ′));

[0156] Where f(θ) is the performance of the parameter combination θ, m(θ) is the mean function, k(θ,θ′) is the covariance function, and GP() represents the Gaussian process;

[0157] In Bayesian optimization, a Gaussian process model is used for posterior prediction, giving the next parameter combination θ nextPredicted values ​​and confidence intervals for :

[0158]

[0159] in, is the predicted value; σ(θ next ) is the standard deviation, z is the confidence interval coefficient;

[0160] The next parameter combination is selected by maximizing its predicted value and uncertainty, using the following moment function:

[0161]

[0162] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0163] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0164] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0165] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0166] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A method for identifying abnormal energy metering data based on time-frequency transformation, characterized in that: The following steps are involved: S1: By analyzing the deterministic characteristics of four spectrum aspects, namely, spectrum center frequency, spectrum bandwidth, spectrum energy and spectrum phase, and the four statistical characteristics of mean, variance, slope and kurtosis, the random characteristics are described to establish the energy metering data feature set; S2: Use the extracted features as input features, mark normal data and abnormal data, and build a training data set; S3 uses the SVM algorithm to train the abnormality analysis and matching model based on the training data set, and separates normal data from abnormal data by finding the optimal hyperplane; S4: Use the trained anomaly analysis and matching model to predict the test data set to obtain abnormal data points; S5: Use the Mahalanobis distance to calculate the anomaly score of each abnormal data point, indicating the degree of abnormality of the data point relative to normal data, and mark the abnormal data point on the time-frequency graph, and select the color, shape and size of the mark according to the size of the anomaly score.

2. According to the method for identifying abnormal energy metering data based on time-frequency transformation in claim 1, it is characterized in that: The step S1 is specifically as follows: obtaining energy metering data of the power system and performing preprocessing; using Fourier transform to transform the preprocessed energy metering data of the power system, converting the time domain data into time-frequency domain data, and extracting features from the time-frequency domain data, including four deterministic features of the spectrum, namely, spectrum center frequency, spectrum bandwidth, spectrum energy and spectrum phase, and calculating four statistical features, namely, mean, variance, slope and kurtosis.

3. According to the method for identifying abnormal energy metering data based on time-frequency transformation in claim 1, it is characterized in that: The preprocessing includes data cleaning, denoising and data format conversion.

4. The method for identifying abnormal energy metering data based on time-frequency transformation according to claim 2 is characterized in that: The Fourier transform used is used to transform the pre-processed power system energy metering data, convert the time domain data into time-frequency domain data, and extract features from the time-frequency domain data, specifically: The pre-processed energy metering data is converted from the time domain to the frequency domain using discrete Fourier transform to obtain spectrum information; Perform discrete Fourier transform on the signal x[n] of length N to obtain the spectrum X[k]: Among them, k is the frequency index and t is the time index; Calculate the energy feature E based on the spectrum information, which represents the total energy of the signal; Analyze spectrum information and extract spectrum features, including spectrum peak f peak , spectrum center f centroid f peak =arg max k |X[k]| Draw a time-frequency diagram based on the spectrum information to obtain time-frequency characteristics, including instantaneous energy and instantaneous frequency: E ins tan [t]=|X[t]| 2 ; Here, θ[t] is the phase information.

5. The method for identifying abnormal energy metering data based on time-frequency transformation according to claim 1 is characterized in that: The S3 is specifically: The goal of SVM is to find an optimal hyperplane that maximizes the distance between normal data points and the hyperplane, while keeping abnormal data points away from the hyperplane. The optimization problem of SVM is expressed as: subject to y i (w T x i +b≥1-ξ i ,for i=1,2,...,n Where w is the normal vector of the hyperplane, b is the bias term, ξ is the slack variable, and C is the penalty parameter; Select the kernel function as Gaussian kernel function: K(x,x′)=exp(-γ||xx′|| 2 )in, x and x′ are eigenvectors, and γ is the parameter of the Gaussian kernel function; The SVM model is trained based on the training set, and the Bayesian optimization method is used to find the optimal hyperparameter combination in the parameter space to obtain the anomaly judgment matching model.

6. The method for identifying abnormal energy metering data based on time-frequency transformation according to claim 5 is characterized in that: The Bayesian optimization method is used to find the optimal hyperparameter combination in the parameter space, as follows: A Gaussian process model is established to estimate the performance of the parameter combination. The Gaussian process model is expressed as: f(θ)~GP(m(θ),k(θ,θ′)); Where f(θ) is the performance of the parameter combination θ, m(θ) is the mean function, k(θ,θ′) is the covariance function, and GP() represents the Gaussian process; In Bayesian optimization, a Gaussian process model is used for posterior prediction, giving the next parameter combination θ next Predicted values ​​and confidence intervals for : in, is the predicted value; σ(θ next ) is the standard deviation, z is the confidence interval coefficient; The next parameter combination is selected by maximizing its predicted value and uncertainty, using the following moment function:

7. An abnormal energy metering data identification system based on time-frequency transformation, characterized in that: It includes data processing module, abnormal analysis and matching module and visualization module; The data processing module describes the random characteristics by analyzing the deterministic characteristics of four spectrum aspects, namely, spectrum center frequency, spectrum bandwidth, spectrum energy and spectrum phase, and the four statistical characteristics of mean, variance, slope and kurtosis, and establishes the energy metering data feature set; and uses the extracted features as input features to mark normal data and abnormal data, and construct a training data set; The abnormal analysis and matching module is provided with a trained abnormal analysis and matching model, which predicts the test data set to obtain abnormal data points; The visualization module uses the Mahalanobis distance to calculate the anomaly score of each abnormal data point to indicate the degree of abnormality of the data point relative to normal data, and marks the abnormal data point on the time-frequency graph, and selects the color, shape and size of the mark according to the size of the anomaly score.

8. The abnormal energy metering data identification system based on time-frequency transformation according to claim 7 is characterized in that: The data processing module acquires the energy metering data of the power system and performs preprocessing; uses Fourier transform to transform the preprocessed energy metering data of the power system, converts the time domain data into time-frequency domain data, and extracts features from the time-frequency domain data, including four deterministic features of the spectrum, namely, spectrum center frequency, spectrum bandwidth, spectrum energy and spectrum phase, and calculates four statistical features, namely, mean, variance, slope and kurtosis; Preprocessing includes data cleaning, denoising and data format conversion; Fourier transform is used to transform the pre-processed power system energy metering data, convert the time domain data into time-frequency domain data, and extract features from the time-frequency domain data, specifically: The pre-processed energy metering data is converted from the time domain to the frequency domain using discrete Fourier transform to obtain spectrum information; Perform discrete Fourier transform on the signal x[n] of length N to obtain the spectrum X[k]: Among them, k is the frequency index and t is the time index; Calculate the energy feature E based on the spectrum information, which represents the total energy of the signal; Analyze spectrum information and extract spectrum features, including spectrum peak f peak , spectrum center f centroid f peak =arg max k |X[k]| Draw a time-frequency diagram based on the spectrum information to obtain time-frequency characteristics, including instantaneous energy and instantaneous frequency: E ins tan [t]=|X[t]| 2 ; Here, θ[t] is the phase information.

9. The abnormal energy metering data identification system based on time-frequency transformation according to claim 7 is characterized in that: The abnormal analysis and matching model is constructed based on SVM, as follows: The goal of SVM is to find an optimal hyperplane that maximizes the distance between normal data points and the hyperplane, while keeping abnormal data points away from the hyperplane. The optimization problem of SVM is expressed as: subject to y i (w T x i +b)≥1-ξ i ,for i=1,2,...,n Where w is the normal vector of the hyperplane, b is the bias term, ξ is the slack variable, and C is the penalty parameter; Select the kernel function as Gaussian kernel function: K(x,x′)=exp(-γ||x-x′|| 2 ) Among them, x and x′ are eigenvectors, and γ is the parameter of the Gaussian kernel function; The SVM model is trained based on the training set, and the Bayesian optimization method is used to find the optimal hyperparameter combination in the parameter space to obtain the anomaly judgment matching model.

10. The method for identifying abnormal energy metering data based on time-frequency transformation according to claim 9 is characterized in that: The Bayesian optimization method is used to find the optimal hyperparameter combination in the parameter space, as follows: A Gaussian process model is established to estimate the performance of the parameter combination. The Gaussian process model is expressed as: f(θ)~GP(m(θ),k(θ,θ′)); Where f(θ) is the performance of the parameter combination θ, m(θ) is the mean function, k(θ,θ′) is the covariance function, and GP() represents the Gaussian process; In Bayesian optimization, a Gaussian process model is used for posterior prediction, giving the next parameter combination θ next Predicted values ​​and confidence intervals for : in, is the predicted value; σ(θ next ) is the standard deviation, z is the confidence interval coefficient; The next parameter combination is selected by maximizing its predicted value and uncertainty, using the following moment function:

Images