Electric quantity abnormity intelligent monitoring method and system considering event waves

Abstract

The invention relates to an electric quantity abnormity intelligent monitoring method considering event waves. The method comprises the following steps: S1, collecting event wave data through a sensor and an intelligent device; s2, using a time-frequency analysis algorithm to extract time sequence features and frequency spectrum features of the electric quantity demand related event waves; s3, constructing a training set according to the extracted time sequence features and frequency spectrum features, and training an electric quantity demand motivation analysis model constructed based on machine learning and deep learning; s4, constructing an electric quantity abnormity monitoring model, calibrating a basic rule through an event wave amplitude value and frequency domain energy, learning normal waveform characteristics through an auto-encoder by the abnormity detection model, and realizing intelligent electric quantity abnormity monitoring in combination with motivation analysis; and S5, constructing a hierarchical visualization model, and graphically displaying the electric quantity change reason, the event behavior and the abnormal early warning information. According to the invention, real-time monitoring and early warning of electric quantity abnormity are realized, and challenges of transient, steady and complex non-stationary events in an electric power system can be effectively handled.

Description

Technical Field

[0001] The present invention relates to the field of electric quantity monitoring, and in particular to an intelligent method and system for monitoring electric quantity abnormality taking event waves into consideration. Background Art

[0002] In modern power systems, with the increasing complexity of loads and energy demand, real-time monitoring and anomaly detection of power consumption have become one of the core technologies in power grid operation and management. Traditional power monitoring systems mainly rely on rule threshold methods to monitor power data, such as implementing simple over-limit alarms by presetting upper and lower limits of voltage, current or power. However, with the rapid development of industrial automation, smart grids and distributed energy systems, the complexity of the operating environment of power systems has increased significantly, and traditional methods have gradually exposed many limitations.

[0003] The operation of the power grid not only includes steady-state power fluctuations (such as normal load changes of equipment), but is also affected by transient events (such as load switching, short-circuit faults) and complex non-stationary events (such as irregular power fluctuations and harmonic interference). The nonlinearity and dynamics of these complex event waves make it difficult for traditional static monitoring methods to thoroughly analyze and distinguish them. Moreover, event wave data often appear as signals with strong time-varying characteristics, that is, non-stationary signals, which cannot be fully processed using fixed frequency domain analysis methods. For example, during load switching, the energy of the signal may be distributed in the high frequency band, while in steady-state fluctuations, the signal is mainly concentrated in the fundamental frequency. In addition, different event waves may have obvious differences in the distribution characteristics of the time and frequency domains, which puts higher requirements on signal processing.

[0004] Traditional methods that rely on rule-based threshold calibration can usually only detect obvious anomalies (such as a single voltage limit or power amplitude exceeding the limit), and cannot cope with more complex abnormal patterns. For example, nonlinear loads and harmonic pollution may require more in-depth analysis rather than relying on a single parameter. Summary of the invention

[0005] In order to solve the above problems, the purpose of the present invention is to provide an intelligent monitoring method and system for power anomalies taking event waves into consideration, so as to realize real-time monitoring and early warning of power anomalies, and effectively cope with the challenges of transient, steady-state and complex non-stationary events in the power system.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] An intelligent monitoring method for power anomaly considering event waves comprises the following steps:

[0008] S1: Collect event wave data through sensors and smart devices;

[0009] S2: Use the time-frequency analysis algorithm to extract the time series characteristics and spectrum characteristics of the event wave related to the power demand, as follows:

[0010] The sliding window size is set according to the event characteristics. Based on the sliding window segmentation signal, the event wave x(t) is segmented into n window signals:

[0011] x n (t)=x(t),t∈[(n-1)Ts,(n-1)Ts+Tw];

[0012] Where Tw is the length of the time window; Ts is the sliding window size;

[0013] The event wave is filtered to reduce noise. High-frequency noise is removed by low-pass filtering and band-pass filtering is used to extract f 1 -f 2 The frequency components of

[0014]

[0015] Where H(f) is the filter transfer function;

[0016] The input signal after filtering is:

[0017]

[0018] in, represents Fourier transform;

[0019] Normalize the signal to a standard range:

[0020]

[0021] Based on the time-frequency analysis of the processed event wave, different time-frequency analysis methods are used for different events:

[0022] Extract time domain, frequency domain and time-frequency joint features, and construct feature vector matrix; generate visual two-dimensional time-frequency graph based on time-frequency distribution

[0023] S3: Based on the extracted time series features and spectrum features, a training set is constructed to train a power demand factor analysis model based on machine learning and deep learning;

[0024] S4: Build an abnormal power monitoring model, calibrate the basic rules through event amplitude and frequency domain energy, and the abnormal detection model learns the normal waveform characteristics through the autoencoder, combined with the cause analysis, to realize intelligent abnormal power monitoring;

[0025] S5: Build a hierarchical visualization model to graphically display the causes of power changes, event behaviors, and abnormal warning information.

[0026] Furthermore, event wave collection is derived from power system operation and is collected through hierarchical equipment:

[0027] On the distribution side, smart meters are used to collect the voltage, current, and power of low-voltage / medium-voltage distribution network users;

[0028] On the transmission side, a synchronous phasor measurement unit is used, located at the transmission network node, to collect the amplitude, phase, and frequency of the three-phase voltage and current, and to supplement the transient event parameters;

[0029] And through the high-frequency transient recorder device, the short-term and violent fluctuations of arc extinguishing, reconstruction and resonance are captured.

[0030] Furthermore, different time-frequency analysis methods are used for different events, as follows:

[0031] For steady-state events, using short-time Fourier transform, the time domain value of the event wave x(t) is represented by a time variable τ, then

[0032]

[0033] Where x(τ) is the time signal; w(t-τ) is the Gaussian window; e -j2πfτ is the Fourier kernel;

[0034] For transient events, use wavelet transform to map the signal into time-frequency space and analyze local time and frequency changes:

[0035]

[0036] Among them, a is the scale factor; b is the translation factor; ψ * is the mother wavelet function;

[0037] For complex non-stationary events, use the Hilbert-Huang transform; first perform empirical mode decomposition (EMD) on the signal and decompose it into several IMFs:

[0038]

[0039] For each IMF i Then perform Hilbert transform to extract the instantaneous frequency:

[0040]

[0041] Furthermore, a training set is constructed based on the extracted time series features and spectrum features, and a power demand factor analysis model based on machine learning and deep learning is trained, as follows:

[0042] Generate data labels and numerical labels for each event wave, and generate classification labels Y according to the type of event wave cls , describing the motivation type:

[0043]

[0044] Generate a numerical label Y for the amplitude prediction of power fluctuation reg , the label |ΔP| represents the magnitude of the power demand or the maximum deviation of the voltage and current waveforms:

[0045] Y reg =ΔP=max(x t )-min(x t ), t∈window range;

[0046] The final label matrix is ​​defined as:

[0047]

[0048] By extracting time series features, frequency domain features, and time-frequency features, a feature vector matrix is ​​constructed, and a training data set is constructed by combining the label matrix and the feature vector matrix;

[0049] Establish an analysis model of power demand drivers, including classification model and regression model;

[0050] Based on the training data set, train the power demand factor analysis model;

[0051] Real-time analysis and prediction of electricity demand drivers are achieved based on the trained electricity demand driver analysis model.

[0052] Furthermore, the feature vector matrix is ​​constructed by extracting the time series features, frequency domain features, and time-frequency features, as follows:

[0053] Extract time domain features and extract statistical information from the original time series to capture signal fluctuation behavior, including mean μ, standard deviation σ, root mean square value RMS and peak value PeakValue:

[0054]

[0055] Peak Value = max(x(t));

[0056] Extract frequency domain features, including the main frequency f max , total harmonic distortion THD and total spectrum energy E:

[0057]

[0058] E = ∫|X(f)| 2 df;

[0059] Where X(f) is the spectrum of the signal, H n is the amplitude of the nth order harmonic;

[0060] The local characteristics of the signal are extracted by the joint time-frequency method to obtain the time-frequency concentration:

[0061]

[0062] The final eigenvector matrix structure:

[0063]

[0064] Among them, X is N samples of the event wave, and m time series, frequency domain and time-frequency features are extracted from each sample.

[0065] Furthermore, a power demand factor analysis model is established, including a classification model and a regression model, as follows:

[0066] Logistic regression is used to build a classification model for analyzing the types of power demand drivers:

[0067]

[0068] Among them, β k is the parameter corresponding to the kth class:

[0069] Ridge regression is used to build a regression model to predict the fluctuation amplitude of electricity demand Y reg :

[0070]

[0071] Among them, λ is the regularization parameter and β is the model parameter;

[0072] For the classification model, the cross entropy loss function is used:

[0073]

[0074] Where II is the indicator function, y i is the true category of the i-th sample; is the probability of being predicted as event k;

[0075] For the regression model, the loss function uses the mean square error:

[0076]

[0077] The gradient descent SGD optimization algorithm is used to minimize the loss function during training.

[0078] Furthermore, S4 is specifically:

[0079] Fixed rules are constructed based on amplitude and frequency domain energy. The fixed rules include abnormal amplitude rules and frequency domain energy abnormal rules. Specifically:

[0080] Abnormal amplitude rule: max(xt )-min(x t )>ΔP max

[0081] High frequency energy ratio:

[0082] Among them, ΔP max The upper limit of the fluctuation amplitude set by experience; f c is the high frequency critical value frequency; is the threshold of the high-frequency energy ratio; if any of the above rules is met, it is preliminarily marked as a potential abnormal wave;

[0083] The autoencoder network is used to learn the distribution characteristics of normal power waves and capture potential anomalies.

[0084] Combined with the power demand driver analysis model, specific anomaly categories and category-related amplitudes are explained.

[0085] Furthermore, the autoencoder includes an encoder and a decoder:

[0086] The encoder projects the input waveform X into a low-dimensional latent space Z:

[0087] Z=f enc (X) = σ(W env X+b enc );

[0088] Decoder, reconstructs input data

[0089]

[0090] Among them, σ is the activation function, W enc ,b enc are the weight and bias of the encoder respectively; W dec ,b dec are the weight and bias of the decoder respectively;

[0091] During the training process, normal waveform data is used as training data, and the loss function is to minimize the reconstruction error:

[0092]

[0093] Using the Adam optimizer, update the parameter W enc ,b enc ,W dec ,b dec .

[0094] Furthermore, the hierarchical visualization model includes the macro level, event level, and anomaly and alarm level, as follows:

[0095] At the macro level, the global trend of long-term real-time fluctuations in electricity consumption is displayed, and steady-state / unsteady-state distribution diagrams are displayed;

[0096] Event level: displays the waveform and frequency domain characteristics of a single event, showing the classification of the cause and the amplitude change;

[0097] The abnormality and alarm levels highlight the abnormal time interval, its cause, and magnitude, and generate alarm statistics.

[0098] A system for intelligently monitoring power anomalies taking event waves into consideration, comprising a processor, a memory and a computer program stored in the memory. When the processor executes the computer program, the system specifically performs the steps in the method for intelligently monitoring power anomalies taking event waves into consideration as described above.

[0099] The present invention has the following beneficial effects:

[0100] 1. The present invention combines time-frequency analysis with intelligent learning technology, from multi-dimensional feature extraction to intelligent dynamic modeling, and can effectively cope with the challenges of transient, steady-state and complex non-stationary events in the power system. It not only improves the ability to capture anomalies, but also improves the interpretability of power changes through visualization and cause analysis, providing strong support for the safe operation and intelligent management of smart grids;

[0101] 2. The present invention realizes the real-time classification and prediction of the factors of power demand through data label generation. The classification task identifies the demand type, and the regression task predicts the fluctuation amplitude. The model can effectively assist power dispatching and event analysis;

[0102] 3. The present invention can intuitively reflect the causes of electricity changes, event behaviors and abnormal warning information through intelligent time waveform, cause analysis and classification and comprehensive visualization of frequency domain characteristics, thereby improving the initiative and response efficiency of power operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] Figure 1 is a flow chart of the method of the present invention;

[0104] Figure 2 Schematic diagram of a power demand driver analysis model in one embodiment of the present invention. DETAILED DESCRIPTION

[0105] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0106] refer to Figure 1 In this embodiment, a method for intelligently monitoring power anomalies considering event waves is provided, comprising the following steps:

[0107] S1: Collect event wave data, including voltage, current, and power, through sensors and smart devices;

[0108] S2: Use the time-frequency analysis algorithm to extract the time series characteristics and spectrum characteristics of the event wave related to the power demand, as follows:

[0109] The sliding window size is set according to the event characteristics. Based on the sliding window segmentation signal, the event wave x(t) is segmented into n window signals:

[0110] x n (t)=x(t),t∈[(n-1)Ts,(n-1)Ts+Tw];

[0111] Where Tw is the length of the time window; Ts is the sliding window size;

[0112] The length of each time window Tw is set to be determined by the event characteristics:

[0113] Steady-state events: Tw = 1s, analyzing load fluctuations of industrial equipment;

[0114] Transient events: Tw = 50ms-200ms, capturing fast-changing load events;

[0115] Complex non-stationary events: Tw>1s, adjusted according to the dynamic characteristics of the signal;

[0116] The step size of the sliding window is 50%-75% of the window length (eg, Ts=0.5Tw) to ensure the continuity of data in adjacent windows.

[0117] The event wave is filtered and denoised. The high-frequency noise is removed by low-pass filtering, and the fundamental wave 50Hz and low-order harmonic components are retained. The f 1 -f 2 Frequency components, such as analyzing the dynamic characteristics of 50Hz and its harmonic components;

[0118]

[0119] Where H(f) is the filter transfer function;

[0120] The input signal after filtering is:

[0121]

[0122] in, represents Fourier transform;

[0123] Normalize the signal to a standard range:

[0124]

[0125] Based on the time-frequency analysis of the processed event wave, different time-frequency analysis methods are used for different events:

[0126] Extract time domain, frequency domain and time-frequency joint features, and construct feature vector matrix; generate visual two-dimensional time-frequency graph based on time-frequency distribution

[0127] S3: Based on the extracted time series features and spectrum features, a training set is constructed to train a power demand factor analysis model based on machine learning and deep learning;

[0128] S4: Build an abnormal power monitoring model, calibrate the basic rules through event amplitude and frequency domain energy, and the abnormal detection model learns the normal waveform characteristics through the autoencoder, combined with the cause analysis, to realize intelligent abnormal power monitoring;

[0129] S5: Build a hierarchical visualization model to graphically display the causes of power changes, event behaviors, and abnormal warning information.

[0130] In this embodiment, event wave collection is derived from power system operation and is collected through hierarchical equipment:

[0131] On the distribution side, smart meters are used to collect the voltage, current, and power of low-voltage / medium-voltage distribution network users;

[0132] On the transmission side, a synchronous phasor measurement unit is used, located at the transmission network node, to collect the amplitude, phase, and frequency of the three-phase voltage and current, and to supplement the transient event parameters;

[0133] And through the high-frequency transient recorder device, the short-term and violent fluctuations of arc extinguishing, reconstruction and resonance are captured.

[0134] In this embodiment, different time-frequency analysis methods are used for different events, as follows:

[0135] For steady-state events, short-time Fourier transform is used (such as load fluctuations in industrial equipment). The time domain value of the event wave x(t) is represented by a time variable τ, then

[0136]

[0137] Where x(τ) is the time signal; w(t-τ) is the Gaussian window; e -j2πfτ is the Fourier kernel;

[0138] For transient events (such as large-scale equipment switching), wavelet transform is used to map the signal into time-frequency space and analyze local time and frequency changes:

[0139]

[0140] Among them, a is the scale factor; b is the translation factor; ψ* is the mother wavelet function;

[0141] For complex non-stationary events, use the Hilbert-Huang transform (such as tripping waveform); first perform empirical mode decomposition (EMD) on the signal and decompose it into several IMFs:

[0142]

[0143] For each IMF i Then perform Hilbert transform to extract the instantaneous frequency:

[0144]

[0145] In this embodiment, a training set is constructed based on the extracted time series features and spectrum features, and an electricity demand factor analysis model based on machine learning and deep learning is trained, as follows:

[0146] Generate data labels and numerical labels for each event wave, and generate classification labels Y according to the type of event wave cls , describing the motivation type:

[0147]

[0148] Generate a numerical label Y for the amplitude prediction of power fluctuation reg， The label |ΔP| represents the magnitude of the power demand or the maximum deviation of the voltage and current waveforms:

[0149] Y reg =ΔP=max(x t )-min(x t ), t∈window range;

[0150] The final label matrix is ​​defined as:

[0151]

[0152] By extracting time series features, frequency domain features, and time-frequency features, a feature vector matrix is ​​constructed, and a training data set is constructed by combining the label matrix and the feature vector matrix;

[0153] Establish an analysis model of power demand drivers, including classification model and regression model;

[0154] Based on the training data set, train the power demand factor analysis model;

[0155] Real-time analysis and prediction of electricity demand drivers are achieved based on the trained electricity demand driver analysis model.

[0156] In this embodiment, a feature vector matrix is ​​constructed by extracting time series features, frequency domain features, and time-frequency features, as follows:

[0157] Extract time domain features and extract statistical information from the original time series to capture signal fluctuation behavior, including mean μ, standard deviation σ, root mean square value RMS and peak value PeakValue:

[0158]

[0159]

[0160] Peak Value = max(x(t));

[0161] Extract frequency domain features, including the main frequency f max , total harmonic distortion THD and total spectrum energy E:

[0162]

[0163] E = ∫|X(f)| 2 df;

[0164] Where X(f) is the spectrum of the signal, H n is the amplitude of the nth order harmonic;

[0165] The local characteristics of the signal are extracted by the joint time-frequency method to obtain the time-frequency concentration:

[0166]

[0167] The final eigenvector matrix structure:

[0168]

[0169] Where X is N samples of the event wave, and each sample extracts m time series, frequency domain and time-frequency features. In this embodiment, an electricity demand motivation analysis model is established, including a classification model and a regression model, as follows:

[0170] Logistic regression is used to build a classification model for analyzing the types of power demand drivers:

[0171]

[0172] Among them, β k is the parameter corresponding to the kth class:

[0173] Ridge regression is used to build a regression model to predict the fluctuation amplitude of electricity demand Y reg :

[0174]

[0175] Among them, λ is the regularization parameter and β is the model parameter;

[0176] For the classification model, the cross entropy loss function is used:

[0177]

[0178] Where II is the indicator function, y i is the true category of the i-th sample; is the probability of being predicted as event k;

[0179] For the regression model, the loss function uses the mean square error:

[0180]

[0181] The gradient descent SGD optimization algorithm is used to minimize the loss function during training.

[0182] In this embodiment, S4 is specifically:

[0183] Fixed rules are constructed based on amplitude and frequency domain energy. The fixed rules include abnormal amplitude rules and frequency domain energy abnormal rules. Specifically:

[0184] Abnormal amplitude rule: max(x t )-min(x t )>ΔP max

[0185] High frequency energy ratio:

[0186] Among them, ΔP max The upper limit of the fluctuation amplitude set by experience; f c is the high frequency critical value frequency; is the threshold of the high-frequency energy ratio; if any of the above rules is met, it is preliminarily marked as a potential abnormal wave;

[0187] The autoencoder network is used to learn the distribution characteristics of normal power waves and capture potential anomalies.

[0188] Combined with the power demand driver analysis model, specific anomaly categories and category-related amplitudes are explained.

[0189] In this embodiment, the autoencoder includes an encoder and a decoder:

[0190] The encoder projects the input waveform X into a low-dimensional latent space Z:

[0191] Z=f enc (X) = σ(W enc X+b enc );

[0192] Decoder, reconstructs input data

[0193]

[0194] Among them, σ is the activation function, W enc ,b enc are the weight and bias of the encoder respectively; W dec ,b dec are the weight and bias of the decoder respectively;

[0195] During the training process, normal waveform data is used as training data, and the loss function is to minimize the reconstruction error:

[0196]

[0197] Using the Adam optimizer, update the parameter W enc ,b enc ,W dec ,b dec .

[0198] In this embodiment, the hierarchical visualization model includes a macro level, an event level, and an anomaly and alarm level, as follows:

[0199] At the macro level, the global trend of long-term real-time fluctuations in electricity (daily, weekly or monthly) is displayed, and steady-state / unsteady-state distribution diagrams are displayed;

[0200] Event level: displays the waveform and frequency domain characteristics of a single event, showing the classification of the cause and the amplitude change;

[0201] The abnormality and alarm levels highlight the abnormal time interval, its cause, and magnitude, and generate alarm statistics.

[0202] 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.

[0203] 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.

[0204] 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.

[0205] 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 The steps for the functions specified in one or more boxes.

[0206] 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. An intelligent monitoring method for power anomaly considering event waves, characterized in that: The following steps are involved: S1: Collect event wave data through sensors and smart devices; S2: Use the time-frequency analysis algorithm to extract the time series characteristics and spectrum characteristics of the event wave related to the power demand, as follows: The sliding window size is set according to the event characteristics. Based on the sliding window segmentation signal, the event wave x(t) is segmented into n window signals: x n (t)=x(t),t∈[(n-1)Ts,(n-1)Ts+Tw]; Where Tw is the length of the time window; Ts is the sliding window size; The event wave is filtered and denoised. High-frequency noise is removed through low-pass filtering, and the frequency components of f1-f2 are extracted through band-pass filtering. Where H(f) is the filter transfer function; The input signal after filtering is: in, represents Fourier transform; Normalize the signal to a standard range: Based on the time-frequency analysis of the processed event wave, different time-frequency analysis methods are used for different events: Extract time domain, frequency domain and time-frequency joint features, and construct feature vector matrix; generate visual two-dimensional time-frequency graph based on time-frequency distribution S3: Based on the extracted time series features and spectrum features, a training set is constructed to train a power demand factor analysis model based on machine learning and deep learning; S4: Build an abnormal power monitoring model, calibrate the basic rules through event amplitude and frequency domain energy, and the abnormal detection model learns the normal waveform characteristics through the autoencoder, combined with the cause analysis, to realize intelligent abnormal power monitoring; S5: Build a hierarchical visualization model to graphically display the causes of power changes, event behaviors, and abnormal warning information.

2. The method for intelligently monitoring power anomalies considering event waves according to claim 1, characterized in that: The event wave collection is derived from the operation of the power system and is collected through layered equipment: On the distribution side, smart meters are used to collect the voltage, current, and power of low-voltage / medium-voltage distribution network users; On the transmission side, a synchronous phasor measurement unit is used, located at the transmission network node, to collect the amplitude, phase, and frequency of the three-phase voltage and current, and to supplement the transient event parameters; And through the high-frequency transient recorder device, the short-term and violent fluctuations of arc extinguishing, reconstruction and resonance are captured.

3. The method for intelligently monitoring power anomalies considering event waves according to claim 1, characterized in that: Different time-frequency analysis methods are used for different events, as follows: For steady-state events, using short-time Fourier transform, the time domain value of the event wave x(t) is represented by a time variable τ, then Where x(τ) is the time signal; w(t-τ) is the Gaussian window; e -j2πfτ is the Fourier kernel; For transient events, use wavelet transform to map the signal into time-frequency space and analyze local time and frequency changes: Among them, a is the scale factor; b is the translation factor; ψ * is the mother wavelet function; For complex non-stationary events, use the Hilbert-Huang transform; first perform empirical mode decomposition (EMD) on the signal and decompose it into several IMFs: For each IMF i Then perform Hilbert transform to extract the instantaneous frequency:

4. The method for intelligently monitoring power anomalies considering event waves according to claim 1, characterized in that: According to the extracted time series features and spectrum features, a training set is constructed to train the power demand factor analysis model based on machine learning and deep learning, as follows: Generate data labels and numerical labels for each event wave, and generate classification labels Y according to the type of event wave cls , describing the motivation type: Generate a numerical label Y for the amplitude prediction of power fluctuation reg , the label |ΔP| represents the magnitude of the power demand or the maximum deviation of the voltage and current waveforms: Y reg =ΔP=max(x t )-min(x t ), t∈window range; The final label matrix is ​​defined as: By extracting time series features, frequency domain features, and time-frequency features, a feature vector matrix is ​​constructed, and a training data set is constructed by combining the label matrix and the feature vector matrix; Establish an analysis model of power demand drivers, including classification model and regression model; Based on the training data set, train the power demand factor analysis model; Real-time analysis and prediction of electricity demand drivers are achieved based on the trained electricity demand driver analysis model.

5. The method for intelligently monitoring power anomalies considering event waves according to claim 5 is characterized in that: The feature vector matrix is ​​constructed by extracting the time series features, frequency domain features, and time-frequency features, as follows: Extract time domain features and extract statistical information from the original time series to capture signal fluctuation behavior, including mean μ, standard deviation σ, root mean square value RMS and peak value PeakValue: Peak Value = max(x(t)); Extract frequency domain features, including the main frequency f max , total harmonic distortion THD and total spectrum energy E: E=∫|X(f)| 2 df; Where X(f) is the spectrum of the signal, H n is the amplitude of the nth order harmonic; The local characteristics of the signal are extracted by the joint time-frequency method to obtain the time-frequency concentration: The final eigenvector matrix structure: Among them, X is N samples of the event wave, and m time series, frequency domain and time-frequency features are extracted from each sample.

6. The method for intelligently monitoring power anomalies considering event waves according to claim 5, characterized in that: The power demand factor analysis model is established, including a classification model and a regression model, which is specifically as follows: a classification model is constructed using logistic regression to analyze the power demand factor type: Among them, β k is the parameter corresponding to the kth class: Ridge regression is used to build a regression model to predict the fluctuation amplitude of electricity demand Y reg : Among them, λ is the regularization parameter and β is the model parameter; For the classification model, the cross entropy loss function is used: Where II is the indicator function, y i is the true category of the i-th sample; is the probability of being predicted as event k; For the regression model, the loss function uses the mean square error: The gradient descent SGD optimization algorithm is used to minimize the loss function during training.

7. The method for intelligently monitoring power anomalies considering event waves according to claim 1, characterized in that: The S4 is specifically: Fixed rules are constructed based on amplitude and frequency domain energy. The fixed rules include abnormal amplitude rules and frequency domain energy abnormal rules. Specifically: Abnormal amplitude rule: max(x t )-min(x t )>ΔP max High frequency energy ratio: Among them, ΔP max The upper limit of the fluctuation amplitude set by experience; f c is the high frequency critical value frequency; is the threshold of the high-frequency energy ratio; if any of the above rules is met, it is preliminarily marked as a potential abnormal wave; The autoencoder network is used to learn the distribution characteristics of normal power waves and capture potential anomalies. Combined with the power demand driver analysis model, specific anomaly categories and category-related amplitudes are explained.

8. The method for intelligently monitoring power anomalies considering event waves according to claim 8, characterized in that: The autoencoder includes an encoder and a decoder: The encoder projects the input waveform X into a low-dimensional latent space Z: Z=f enc (X)=σ(W enc X+b enc ); Decoder, reconstructs input data Among them, σ is the activation function, W enc , b enc are the weight and bias of the encoder respectively; W dec , b dec are the weight and bias of the decoder respectively; During the training process, normal waveform data is used as training data, and the loss function is to minimize the reconstruction error: Using the Adam optimizer, update the parameter W enc ,b enc ,W dec ,b dec .

9. The method for intelligently monitoring power anomalies considering event waves according to claim 1, characterized in that: The hierarchical visualization model includes a macro level, an event level, and an anomaly and alarm level, as follows: At the macro level, the global trend of long-term real-time fluctuations in electricity consumption is displayed, and steady-state / unsteady-state distribution diagrams are displayed; Event level: displays the waveform and frequency domain characteristics of a single event, showing the classification of the cause and the amplitude change; The abnormality and alarm levels highlight the abnormal time interval, its cause, and magnitude, and generate alarm statistics.

10. An intelligent monitoring system for power anomaly considering event waves, characterized in that: It includes a processor, a memory and a computer program stored in the memory. When the processor executes the computer program, it specifically executes the steps in the intelligent monitoring method for power anomaly considering event waves as described in any one of claims 1 to 9.

Images