A smart AI-based power transmission grid fault hybrid diagnosis method and system

Abstract

The application belongs to the technical field of smart grid, and discloses a power transmission network fault hybrid diagnosis method and system based on smart AI; including: collecting group fault electrical quantity data; respectively performing standardization processing on each group of fault electrical quantity data, and constructing a fault data set; extracting a fault feature vector from the fault data set; obtaining protection detection data, fusing the fault feature vector and the protection detection data, and adopting a hybrid diagnosis strategy to locate a fault occurrence position; obtaining power grid topology data, combining the power grid topology data and a geographic information system, correcting the fault occurrence position in geographic coordinates, and obtaining a fault geographic position; according to the fault geographic position and the fault feature vector, calculating a fault influence range and a fault severity, and generating a fault diagnosis report; the application can effectively improve the robustness and response speed of fault diagnosis, accurately locate the fault occurrence position, reduce fault recovery time, and enhance the safety and stability of the power transmission network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart grid, more particularly, the present application relates to a power transmission network fault hybrid diagnosis method and system based on intelligent AI. BACKGROUND

[0002] With the continuous growth of global power demand and the rapid development of smart grid, the scale and complexity of power transmission network are also significantly improved. However, power transmission network often faces challenges such as frequent faults, difficult positioning and slow recovery during operation. Power system failure not only causes large-scale power outages, but also has a serious impact on the safe and stable operation of the power grid. Therefore, quickly and accurately diagnosing power transmission network faults has become a key task in power system management. In traditional power transmission network fault diagnosis, a single data source or a single algorithm is usually used for positioning, which makes it vulnerable to errors in complex fault scenarios and difficult to achieve high-precision, fast and robust diagnosis.

[0003] With the continuous development of intelligent technology, fault diagnosis methods based on artificial intelligence have gradually become a research hotspot. For example, the patent with publication number CN113030647A discloses a power transmission network line three-span section fault active warning system. It includes an accident signal acquisition module, a fault distance calculation module, a power transmission network fault diagnosis module, a power transmission line account maintenance module, a "three-span" section maintenance module, a power grid geographic information display module and a warning notification module. This invention can realize the visualization management of power transmission network and "three-span" data based on geographic information platform, can diagnose fault location and actively warn faults in "three-span" section by summarizing accident signals, and can initiate notifications in time for field operation and maintenance personnel to improve the response speed of power transmission network line "three-span" section fault conditions.

[0004] However, the above technology can realize the fault location diagnosis of power transmission network, but it only relies on the fault distance calculation of accident tripping signals, the data source is single, lacks in-depth analysis and extraction of fault characteristics, and is difficult to accurately describe fault characteristics, resulting in insufficient recognition ability for complex fault patterns. Moreover, it mainly relies on traditional distance calculation methods, which are easily affected by noise interference or line parameter errors in complex power grid structure or multi-section fault conditions, resulting in decreased diagnosis accuracy and insufficient adaptability to different fault types.

[0005] In view of this, the present application proposes a power transmission network fault hybrid diagnosis method and system based on intelligent AI to solve the above problems. SUMMARY

[0006] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned purpose, the present application provides the following technical scheme: a power transmission network fault hybrid diagnosis method based on intelligent AI, comprising:

[0007] collecting group fault electrical quantity data, is an integer greater than 1;

[0008] standardizing each group of fault electrical quantity data respectively to construct a fault data set;

[0009] extracting a fault feature vector from the fault data set by using wavelet transform and deep learning technology, the fault feature vector including fault waveform features, fault transient features and fault spectrum features;

[0010] obtaining protection detection data, fusing the fault feature vector and the protection detection data, and adopting a hybrid diagnosis strategy to quickly locate a fault occurrence position;

[0011] obtaining power grid topology data, correcting the fault occurrence position in geographic coordinates in combination with the power grid topology data and a geographic information system to obtain a fault geographic position;

[0012] calculating a fault influence range and a fault severity according to the fault geographic position and the fault feature vector, and generating a fault diagnosis report.

[0013] Further, the fault electrical quantity data includes voltage waveform data and current waveform data;

[0014] The method for constructing the fault data set includes:

[0015] performing denoising processing on each group of fault electrical quantity respectively to obtain preliminary processing data, the preliminary processing data including preliminary voltage processing data and preliminary current processing data;

[0016] counting the number of data in each group of preliminary voltage processing data respectively and marking as data quantity; comparing each data quantity respectively, marking the data quantity with the largest value as maximum quantity, and marking the preliminary voltage processing data corresponding to the maximum quantity as voltage standard data; marking the preliminary current processing data corresponding to the same fault recording device as the voltage standard data as current standard data; marking all preliminary voltage processing data that are not marked as voltage standard data as to-be-aligned voltage data, and marking all preliminary current processing data that are not marked as current standard data as to-be-aligned current data;

[0017] performing time alignment on each group of to-be-aligned voltage data and the voltage standard data respectively to obtain voltage aligned data; performing time alignment on each group of to-be-aligned current data and the current standard data respectively to obtain current aligned data; constructing the fault data set according to each group of voltage aligned data, current aligned data, voltage standard data and current standard data; wherein the method for performing time alignment on the to-be-aligned voltage data and the voltage standard data is consistent with the method for performing time alignment on the to-be-aligned current data and the current standard data.

[0018] Further, the method for time aligning the voltage data to be aligned with the voltage standard data comprises:

[0019] According to the time sequence, the voltages in the voltage data to be aligned and the voltage standard data are respectively acquired, and corresponding voltage sequences are respectively constructed; the Euclidean distances between each two voltages in the two voltage sequences are respectively calculated and marked as voltage distances; a distance matrix is constructed according to the voltage distances, and the elements in the distance matrix correspond to the voltage distances one by one; an accumulated distance matrix is constructed according to the distance matrix, and the size of the accumulated distance matrix is consistent with the size of the distance matrix; an element in the first row and the first column of the accumulated distance matrix is marked as a starting point element, the elements in the first row of the accumulated distance matrix are marked as head elements, the elements in the first column of the accumulated distance matrix are marked as index elements, and the elements in the accumulated distance matrix which are not marked as the starting point element, the head elements and the index elements are all marked as data elements;

[0020] The starting point element in the accumulated distance matrix is equal to the starting point element in the distance matrix, each head element in the accumulated distance matrix is the previous head element plus the corresponding element in the distance matrix, and each index element in the accumulated distance matrix is the previous index element plus the corresponding element in the distance matrix; each data element in the accumulated distance matrix is the minimum value in the corresponding adjacent elements plus the corresponding element in the distance matrix, and the adjacent elements corresponding to the data element include the elements located on the left side of the data element, the elements located above the data element and the elements located at the upper left of the data element in the accumulated distance matrix;

[0021] Starting from the element corresponding to the lower right corner of the accumulated distance matrix, the optimal path is acquired by backtracking to the starting point element of the accumulated distance matrix; the process of backtracking is that: the element backtracked to is marked as a current element, the minimum value in the corresponding adjacent elements of the current element in the accumulated distance matrix is selected as a backtracking element, and the current element is updated to the backtracking element; according to the optimal path, a voltage matching relationship is established, the voltage matching relationship is the corresponding relationship between the voltages in the voltage data to be aligned and the voltages in the voltage standard data; the voltage data to be aligned is processed by interpolation according to the voltage matching relationship; the method for processing the voltage data to be aligned by interpolation is that: when there are voltages in the voltage standard data corresponding to one voltage in the voltage data to be aligned in the optimal path, the corresponding voltage is marked as an interpolation voltage, and a linear interpolation method is used at the interpolation voltage to generate new voltages, which is an integer greater than 1.

[0022] Further, the fault waveform feature includes an amplitude feature and a waveform change rate; the fault transient feature includes a transient amplitude and a transient duration; and the fault spectrum feature includes a spectrum amplitude and a frequency ratio;

[0023] The voltage waveform data after standardization is marked as a voltage standard waveform, and the current waveform data is marked as a current standard waveform; the amplitude feature includes the maximum amplitude and minimum amplitude of the voltage standard waveform, and the maximum amplitude and minimum amplitude of the current standard waveform; the waveform change rate includes the maximum instantaneous slope of the voltage standard waveform and the maximum instantaneous slope of the current standard waveform; the transient amplitude includes the maximum instantaneous deviation value of the voltage standard waveform and the maximum instantaneous deviation value of the current standard waveform; the transient duration includes the time length of the voltage high amplitude change and the time length of the current high amplitude change; the frequency spectrum amplitude includes the amplitude distribution of the voltage signal at different frequencies and the amplitude distribution of the current signal at different frequencies; the frequency ratio includes the amplitude ratio of the corresponding fundamental wave and harmonic wave of the voltage signal and the amplitude ratio of the corresponding fundamental wave and harmonic wave of the current signal;

[0024] The method for extracting the fault feature vector from the fault data set comprises:

[0025] Wavelet transform is performed on the voltage standard waveform and the current standard waveform in the fault data set respectively to obtain fault transient features; the voltage standard waveform and the current standard waveform in the fault data set are input into the trained feature extraction model respectively to extract fault feature data, wherein the fault feature data includes fault waveform features and fault frequency spectrum features; wherein the feature extraction model includes single feature extraction models, The number of data in the fault feature data, and the single feature extraction model corresponds to the data in the fault feature data one by one; The single feature extraction model is a deep neural network model, and The training process of the single feature extraction model is consistent.

[0026] Further, the protection detection data is a circuit breaker tripping signal;

[0027] The method for positioning the fault occurrence position comprises:

[0028] respectively, and the corresponding fundamental components are extracted; the fundamental components corresponding to each voltage standard waveform are marked as voltage fundamental components, and the fundamental components corresponding to each current standard waveform are marked as current fundamental components; each voltage fundamental component is divided by the corresponding current fundamental component to obtain the corresponding impedance; the line corresponding to the protection detection data of the power transmission network is marked as a fault line, the impedance corresponding to the fault line is obtained and marked as a fault impedance; the number of fault impedances is counted and marked as an impedance number; according to the impedance number, the fault distance is calculated by using a calculated impedance and a distance calculation model respectively, the fault distance calculated by using the calculated impedance is marked as a first distance, and the fault distance calculated by using the distance calculation model is marked as a second distance; wherein, the calculated impedance is the fault impedance used to calculate the fault distance, the fault distance is the distance between the fault occurrence position and the fault recording device corresponding to the calculated impedance, the training process of the distance calculation model is consistent with the training process of the single feature extraction model, and both are deep neural network models;

[0029] obtain historical data, the historical data including data obtained at different historical time points obtain distance error data, the distance error data including calculation error data and model error data, is an integer greater than 1; wherein, the calculation error data is the difference between the fault distance calculated by using the calculated impedance and the corresponding actual distance, the model error data is the difference between the fault distance calculated by using the distance calculation model and the corresponding actual distance, and the actual distance is the distance between the actual fault occurrence position and the fault recording device corresponding to the calculated impedance; the randomness corresponding to the calculation error data and the model error data is calculated respectively, the weight coefficients corresponding to the calculation error data and the model error data are calculated according to the randomness respectively; the first distance is multiplied by the weight coefficient corresponding to the calculation error data, and then the second distance is added to the weight coefficient corresponding to the model error data to obtain a weighted distance, and the fault occurrence position is located according to the weighted distance.

[0030] Further, the method of calculating the fault distance by using the calculated impedance and the distance calculation model respectively includes:

[0031] if the impedance number is equal to 2, one of the fault impedances is randomly selected as the calculated impedance; obtain power grid topology data, obtain the total length of the line corresponding to the fault line and the unit impedance according to the power grid topology data, multiply the total length of the line by the unit impedance to obtain the total impedance of the line; divide the calculated impedance by the total impedance of the line, and then multiply the total length of the line to obtain the first distance;

[0032] If the number of impedances is greater than 2, the fault recording devices located at both ends of the fault line are marked as edge devices, the fault impedances corresponding to the edge devices are edge impedances, one of the edge impedances is randomly selected as a calculation impedance, and a corresponding first distance is calculated; the fault recording device corresponding to the calculation impedance is marked as a calculation device, and the fault recording devices that are not marked as edge devices are marked as intermediate devices; according to the power grid topology, the distances between each intermediate device and the calculation device are obtained and marked as device distances; if there is a device distance less than or equal to twice the first distance, the corresponding intermediate device is marked as a recalculation device, the calculation impedance is replaced by the fault impedance corresponding to the recalculation device, and the corresponding first distance is recalculated; if all device distances are greater than twice the first distance, the first distance is not recalculated.

[0033] If the number of impedances is equal to 2, the fault waveform features, fault transient features and fault spectrum features corresponding to the calculation device are obtained from the fault feature vector and marked as first positioning data; if the number of impedances is greater than 2, the fault waveform features, fault transient features and fault spectrum features corresponding to the recalculation device are obtained from the fault feature vector and marked as second positioning data; the first positioning data or the second positioning data is input into the trained distance calculation model to calculate a corresponding second distance.

[0034] Further, the method for calculating randomness includes:

[0035] The number of each different value in the calculation error data is counted and marked as a value number; the number of all values in the calculation error data is counted and marked as a total value number; each value number is divided by the total value number to obtain a value probability; 2 is taken as the base, and each value probability is taken as a true number in turn to calculate the information amount corresponding to each value probability; each information amount is multiplied by the corresponding value probability to obtain a value contribution; each value contribution is added in turn, and the negative is taken to obtain the randomness corresponding to the calculation error data; the calculation method of the randomness corresponding to the model error data is consistent with the calculation method of the randomness corresponding to the calculation error data;

[0036] The method for calculating the weight coefficient includes:

[0037] The randomness corresponding to the calculation error data is marked as a first randomness, and the randomness corresponding to the model error data is marked as a second randomness; the first randomness is added to the second randomness to obtain a total randomness; the first randomness is divided by the total randomness to obtain the weight coefficient corresponding to the calculation error data; the second randomness is divided by the total randomness to obtain the weight coefficient corresponding to the model error data.

[0038] Further, the method for obtaining the fault geographical position includes:

[0039] The power grid topology data is imported into a geographic information system, each fault recording device in the power transmission grid is marked with coordinates, and is marked as device coordinates; the device coordinates of the fault recording device corresponding to the calculated impedance are obtained, and are marked as standard coordinates; a positioning circle is generated with the standard coordinates as the center and the fault distance as the radius; the intersection of the positioning circle and the fault line is marked as a fault point, the number of fault points is counted, and is marked as the fault number; if the fault number is equal to 1, the coordinates corresponding to the fault point are obtained from the geographic information system, and are taken as the fault geographic location; if the fault number is greater than 1, the device coordinates corresponding to the calculation device are obtained, and are marked as calculation coordinates; an auxiliary circle is generated with the calculation coordinates as the center and the fault distance as the radius; the intersection of the auxiliary circle and the fault line is marked as an auxiliary point; the coordinates of each fault point are obtained, and are marked as first coordinates; the coordinates of each auxiliary point are obtained, and are marked as second coordinates; the Euclidean distance between each first coordinate and each second coordinate is calculated in turn, and is marked as point distance; the first coordinate corresponding to the minimum point distance is taken as the fault geographic location.

[0040] Further, the method for generating a fault diagnosis report comprises:

[0041] Each fault recording device corresponding to each set of fault electrical quantity data is marked as a fault detection device; according to the power grid topology data, the line set between each fault detection device and the fault geographic location is obtained, and the total length of each line corresponding to each line set is obtained; the total length of each line corresponding to each line set is added in turn to obtain the influence range corresponding to each line set; each influence range is sorted from large to small to generate an influence sorting table; the influence range ranked first in the influence sorting table is taken as the fault influence range;

[0042] The fault feature vector and the fault influence range are taken as influence data, and the influence data is input into the trained degree prediction model to predict the corresponding fault severity; the training process of the degree prediction model is consistent with the training process of the single feature extraction model, and both are deep neural network models;

[0043] According to the fault influence range, the fault severity and the fault geographic location, a fault diagnosis report is generated.

[0044] A power transmission grid fault hybrid diagnosis system based on intelligent AI is used to implement the power transmission grid fault hybrid diagnosis method based on intelligent AI, comprising:

[0045] The data acquisition module is used to acquire a set of fault electrical quantity data, which is an integer greater than 1;

[0046] The data processing module is used to perform standardization processing on each set of fault electrical quantity data respectively, and construct a fault data set;

[0047] a feature extraction module configured to extract a fault feature vector from the fault data set by using a wavelet transform and a deep learning technique, the fault feature vector including a fault waveform feature, a fault transient feature and a fault frequency spectrum feature;

[0048] a fault location module configured to obtain protection detection data, fuse the fault feature vector and the protection detection data, and quickly locate a fault occurrence position by using a hybrid diagnosis strategy;

[0049] a position correction module configured to obtain power grid topology data, correct the fault occurrence position in geographic coordinates by combining the power grid topology data and a geographic information system, and obtain a fault geographic position;

[0050] a risk assessment module configured to calculate a fault influence range and a fault severity according to the fault geographic position and the fault feature vector, and generate a fault diagnosis report.

[0051] The technical effects and advantages of the power grid fault hybrid diagnosis method and system based on intelligent AI are as follows:

[0052] By time-aligning the collected multiple sets of fault electrical quantity data, a unified fault data set is established, effectively solving the problem of data time axis deviation caused by factors such as sampling rate and signal propagation delay; the intelligent AI technology is used to realize automatic extraction of various fault feature parameters, which can comprehensively depict the characteristics of the power grid fault; the protection detection data and the fault feature vector are fused, and the hybrid diagnosis strategy is used to locate the fault position, thereby improving the positioning accuracy in complex fault scenarios and reducing the error influence; the fault position is corrected by combining the power grid topology data and the geographic information system, and the accurate geographic position of the fault occurrence is determined; the fault influence range and the severity are calculated based on the fault features and the geographic position, the influence of the fault on the power grid is comprehensively evaluated, and more reliable fault diagnosis basis is provided for power grid dispatchers; the robustness and response speed of fault diagnosis can be effectively improved, the fault occurrence position can be accurately located, the fault recovery time can be reduced, the safety and stability of the power grid can be enhanced, and significant economic and social benefits can be achieved. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 FIG. 1 is a schematic diagram of a power grid fault hybrid diagnosis system based on intelligent AI according to Embodiment 1 of the present application;

[0054] Figure 2 FIG. 2 is a flowchart of a power grid fault hybrid diagnosis method based on intelligent AI according to Embodiment 2 of the present application. DETAILED DESCRIPTION

[0055] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0056] Embodiment 1

[0057] Please refer to Figure 1 As shown in the figure, the power grid fault hybrid diagnosis system based on intelligent AI in the embodiment includes a data acquisition module, a data processing module, a feature extraction module, a fault location module, a position correction module and a risk assessment module. The modules are connected through wired and / or wireless means to realize data transmission between the modules.

[0058] The data acquisition module is configured to acquire a plurality of groups of fault electrical quantity data, which is an integer greater than 1.

[0059] The fault electrical quantity data includes voltage waveform data and current waveform data, which are obtained by a fault recording device installed on the power grid. The voltage waveform data is the waveform of the voltage of each measurement point in the power grid changing with time, which is used to analyze the voltage amplitude, phase angle, drop, etc. The current waveform data is the waveform of the current of each measurement point in the power grid changing with time, which is used to analyze the current amplitude, impact characteristics, short-circuit current, etc.

[0060] The data processing module is configured to respectively perform standardization processing on each group of fault electrical quantity data to construct a fault data set.

[0061] The method for constructing the fault data set includes:

[0062] Each group of fault electrical quantity data is respectively subjected to denoising processing to obtain preliminary processing data, which includes preliminary voltage processing data and preliminary current processing data. The purpose of denoising processing is to remove noise caused by factors such as equipment error, environmental interference, measurement error, etc., to improve data quality. The method for denoising processing may be, for example, wavelet denoising, median filtering, Kalman filtering, etc.

[0063] respectively, and mark as data quantity; compare each data quantity respectively, mark the data quantity with the largest value as maximum quantity, and mark the preliminary voltage processing data corresponding to the maximum quantity as voltage standard data; mark the preliminary current processing data corresponding to the voltage standard data as current standard data; mark all the preliminary voltage processing data not marked as voltage standard data as to-be-aligned voltage data, and mark all the preliminary current processing data not marked as current standard data as to-be-aligned current data;

[0064] align each group of to-be-aligned voltage data with the voltage standard data in time to obtain voltage aligned data; align each group of to-be-aligned current data with the current standard data in time to obtain current aligned data; construct a fault data set according to each group of voltage aligned data, current aligned data, voltage standard data and current standard data; it should be understood that in power grid fault analysis, the data collected by different wave recording devices have deviations on the time axis due to factors such as sampling rate and signal propagation delay, so synchronization processing is needed to adjust the time axis of the waveform to ensure that the data of each measurement point can be correctly matched.

[0065] The method of aligning the to-be-aligned voltage data with the voltage standard data in time is consistent with the method of aligning the to-be-aligned current data with the current standard data in time; the method of aligning the to-be-aligned voltage data with the voltage standard data in time comprises:

[0066] According to the time sequence, the voltages in the to-be-aligned voltage data and the voltage standard data are obtained respectively, and corresponding voltage sequences are constructed respectively; the Euclidean distance between each two voltages in the two voltage sequences is calculated respectively and marked as voltage distance; a distance matrix is constructed according to the voltage distance, and the elements in the distance matrix correspond one-to-one to the voltage distance; the calculation method of Euclidean distance is prior art, which will not be described in detail here; according to the distance matrix, a cumulative distance matrix is constructed, and the size of the cumulative distance matrix is consistent with the size of the distance matrix; the element in the first row and the first column of the cumulative distance matrix is marked as a starting element, the elements in the first row of the cumulative distance matrix are marked as head elements, the elements in the first column of the cumulative distance matrix are marked as index elements, and the elements in the cumulative distance matrix not marked as starting elements, head elements and index elements are all marked as data elements;

[0067] The starting element in the cumulative distance matrix is ​​equal to the starting element in the distance matrix. Each head element in the cumulative distance matrix is ​​the previous head element plus the corresponding element in the distance matrix. Each index element in the cumulative distance matrix is ​​the previous index element plus the corresponding element in the distance matrix. Each data element in the cumulative distance matrix is ​​the minimum value of the corresponding adjacent elements plus the corresponding element in the distance matrix. The adjacent elements corresponding to the data element include the element to the left of the data element, the element above the data element, and the element to the upper left of the data element in the cumulative distance matrix.

[0068] Starting from the element corresponding to the lower right corner of the cumulative distance matrix, backtrack to the starting element of the cumulative distance matrix to obtain the optimal path; the backtracking process is: mark the element backtracked to as the current element, select the minimum value among the adjacent elements corresponding to the current element in the cumulative distance matrix as the backtracking element, and update the current element to the backtracking element; according to the optimal path, establish a voltage matching relationship, which is the corresponding relationship between the voltage in the voltage data to be aligned and the voltage in the voltage standard data; according to the voltage matching relationship, interpolate the voltage data to be aligned so that the voltage data to be aligned is aligned with the voltage standard data on the time axis; the method for interpolating the voltage data to be aligned is: for a voltage in the data to be aligned in the optimal path, the voltage in the voltage standard data is corresponding to the voltage in the voltage standard data. When a voltage is generated, the corresponding voltage is marked as the interpolation voltage, and a linear interpolation method is used at the interpolation voltage to generate A new voltage, is an integer greater than 1; the linear interpolation method is an existing technology and will not be described in detail here.

[0069] For example, the voltage sequences corresponding to the voltage data to be aligned are 101 and 106, and the voltage sequences corresponding to the voltage standard data are 100, 102, and 105; therefore, the distance matrix is: , cumulative distance matrix: ; The optimal path includes 3, 2, and 1 in the cumulative distance matrix, that is, the optimal path is ; The voltage matching relationship is that the first voltage in the voltage data to be aligned corresponds to the first voltage in the voltage standard data, the first voltage in the voltage data to be aligned corresponds to the second voltage in the voltage standard data, and the second voltage in the voltage data to be aligned corresponds to the third voltage in the voltage standard data; since the first voltage in the voltage data to be aligned corresponds to two voltages in the voltage standard data, the first voltage in the voltage data to be aligned is marked as the interpolation voltage, and a linear interpolation method is used to generate a new voltage 103 after the interpolation voltage.

[0070] The feature extraction module is configured to extract a fault feature vector from the fault data set by using wavelet transform and deep learning technology, and the fault feature vector includes fault waveform features, fault transient features and fault spectrum features.

[0071] The fault feature vector helps to comprehensively represent the characteristics and properties of the power grid fault, thereby achieving accurate fault positioning and identification.

[0072] The fault waveform features are the main features of the voltage waveform and the current waveform when the power grid fails, and can reflect the basic form and change trend of the fault signal; the fault waveform features include amplitude features and waveform change rates; the fault transient features are the sudden or transient response of the current and voltage at the moment of the power grid failure or within a short period of time after the failure; the fault transient features include transient amplitude and transient duration; the fault spectrum features are the frequency distribution and energy distribution of the voltage and current during the power grid failure, and the fault spectrum features include spectrum amplitude and frequency ratio.

[0073] The voltage waveform data after standardization is marked as a voltage standard waveform, and the current waveform data is marked as a current standard waveform; the amplitude features include the maximum amplitude (i.e. the value of the highest point in the waveform) and the minimum amplitude (i.e. the value of the lowest point in the waveform) of the voltage standard waveform, and the maximum amplitude and the minimum amplitude of the current standard waveform; the waveform change rate includes the maximum instantaneous slope (i.e. the maximum change rate of the voltage in a unit of time) of the voltage standard waveform and the maximum instantaneous slope of the current standard waveform; the transient amplitude includes the maximum instantaneous deviation value (i.e. the maximum deviation of the waveform from its average value) of the voltage standard waveform and the maximum instantaneous deviation value of the current standard waveform; the transient duration includes the time length of the high amplitude change of the voltage (i.e. the duration of the amplitude deviating from the preset normal threshold, and the normal threshold is pre-set by the person skilled in the art according to the actual situation) and the time length of the high amplitude change of the current; the spectrum amplitude includes the amplitude distribution of the voltage signal at different frequencies and the amplitude distribution of the current signal at different frequencies; the frequency ratio includes the amplitude ratio of the fundamental wave and the harmonic wave corresponding to the voltage signal and the amplitude ratio of the fundamental wave and the harmonic wave corresponding to the current signal.

[0074] The method for extracting the fault feature vector from the fault data set includes:

[0075] The voltage standard waveform and the current standard waveform in the fault data set are respectively subjected to wavelet transform to obtain the fault transient features; the voltage standard waveform and the current standard waveform in the fault data set are respectively input into the trained feature extraction model to extract fault feature data, and the fault feature data includes fault waveform features and fault spectrum features; wherein the feature extraction model includes single feature extraction models, is the number of data in the fault feature data, and the single feature extraction model corresponds to the data in the fault feature data one by one, that is, a single feature extraction model extracts one type of data in the fault feature data. is 5, namely, including the maximum amplitude, minimum amplitude, maximum instantaneous slope, amplitude distribution at different frequencies, and amplitude ratio of fundamental wave to harmonics; illustratively, the single feature extraction model corresponds to the maximum amplitude in the fault feature data, so the voltage standard waveform is input into the single feature extraction model to extract the maximum amplitude of the voltage standard waveform, and the current standard waveform is input into the single feature extraction model to extract the maximum amplitude of the current standard waveform; Each single feature extraction model is a deep neural network model, and The training process of each single feature extraction model is the same.

[0076] The training process of the maximum amplitude corresponding single feature extraction model includes:

[0077] Pre-collection There are different standard waveforms, the standard waveform is voltage standard waveform or current standard waveform, The corresponding maximum amplitude is set for each standard waveform. is an integer greater than 1, and the standard waveform and the corresponding maximum amplitude are converted into a corresponding set of feature vectors; the maximum amplitude corresponding to the standard waveform is collected by those skilled in the art in the process of extracting fault feature data historically. Different standard waveforms are analyzed in turn according to the actual situation, and the corresponding maximum amplitude is extracted from each standard waveform. Set the corresponding maximum amplitude for different standard waveforms in turn;

[0078] Each set of feature vectors is used as the input of the single feature extraction model. The single feature extraction model uses a set of predicted maximum amplitudes corresponding to each set of standard waveforms as output, and the actual maximum amplitude corresponding to each set of standard waveforms as the prediction target. The actual maximum amplitude is the pre-set maximum amplitude corresponding to the standard waveform; the training goal is to minimize the sum of the prediction errors of all standard waveforms; the calculation formula for the prediction error is: ,in is the prediction error, is the group number of the characteristic vector corresponding to the standard waveform, For the The predicted maximum amplitude corresponding to the standard waveform of the group, For the The actual maximum amplitude corresponding to the standard waveform of the group; the single feature extraction model is trained until the sum of the prediction errors reaches convergence and the training is stopped.

[0079] The fault positioning module is configured to acquire protection detection data, fuse the fault feature vector and the protection detection data, and quickly locate the fault occurrence position by using a hybrid diagnosis strategy.

[0080] The protection detection data is a circuit breaker tripping signal, which is acquired by an electrical protection device (circuit breaker) installed on the power transmission network; the circuit breaker tripping signal is a signal emitted when the circuit breaker performs a disconnecting operation, and is used to indicate that the fault line in the power transmission network has been cut off, i.e., effectively isolating the fault in the power transmission network, preventing the fault from spreading to other parts, and ensuring the normal operation of other lines.

[0081] The method for locating the fault occurrence position comprises:

[0082] The Fourier transform is performed on the voltage standard waveform and the current standard waveform in the fault data set respectively, each voltage standard waveform and each current standard waveform is converted from the time domain to the frequency domain, and the corresponding fundamental component is extracted; each voltage fundamental component is marked as a voltage fundamental component, and each current fundamental component is marked as a current fundamental component; each voltage fundamental component is divided by the corresponding current fundamental component to obtain the corresponding impedance; the line (i.e., the power transmission line) corresponding to the protection detection data of the power transmission network is marked as the fault line, the impedance corresponding to the fault line is obtained and marked as the fault impedance; the number of fault impedances is counted and marked as the impedance number; the Fourier transform is a prior art, and the specific process is not described in detail here; according to the impedance number, the fault distance is calculated by using the calculated impedance and the distance calculation model, the fault distance calculated by using the calculated impedance is marked as the first distance, and the fault distance calculated by using the distance calculation model is marked as the second distance; wherein, the calculated impedance is the fault impedance used to calculate the fault distance, the fault distance is the distance between the fault occurrence position and the fault recorder corresponding to the calculated impedance, the training process of the distance calculation model is consistent with the training process of the single feature extraction model, and both are deep neural network models;

[0083] The historical data includes the fault data and the distance error data acquired at different historical time points. The distance error data includes the calculation error data and the model error data, is an integer greater than 1; wherein the calculation error data is a difference between a fault distance calculated by using the calculation impedance and a corresponding actual distance, the model error data is a difference between a fault distance calculated by using the distance calculation model and a corresponding actual distance, and the actual distance is a distance between an actual fault occurrence position and a fault recording device corresponding to the calculation impedance; the randomness corresponding to the calculation error data and the model error data is calculated respectively, the weight coefficient corresponding to the calculation error data and the model error data is calculated according to the randomness respectively; the first distance is multiplied by the weight coefficient corresponding to the calculation error data, and then added to the second distance multiplied by the weight coefficient corresponding to the model error data, to obtain a weight distance, and the fault occurrence position is located according to the weight distance.

[0084] The method for calculating the fault distance by using the calculation impedance and the distance calculation model respectively includes:

[0085] If the impedance quantity is equal to 2, one fault impedance is randomly selected as the calculation impedance; the power grid topology data is obtained, the power grid topology data is information describing the structure of the power grid, the connection relationship of the lines and the parameters of the devices, and reflects the physical and electrical connection relationship between the elements such as the transformer substation, the transmission line, the load, the switching device and the generator in the power transmission grid; the power grid topology data is obtained through the SCADA system of the power grid company, the local power supply company or the power dispatching center; according to the power grid topology data, the total length of the line corresponding to the fault line and the unit impedance are obtained, the total length of the line is multiplied by the unit impedance to obtain the total impedance of the line; the calculation impedance is divided by the total impedance of the line, and then multiplied by the total length of the line to obtain the first distance.

[0086] If the impedance quantity is greater than 2, the fault recording devices located at both ends of the fault line are marked as edge devices, the fault impedances corresponding to the edge devices are marked as edge impedances, one edge impedance is randomly selected as the calculation impedance, and the corresponding first distance is calculated; the fault recording device corresponding to the calculation impedance is marked as a calculation device, and the fault recording devices which are not marked as the edge devices are marked as intermediate devices; according to the power grid topology structure, the distance between each intermediate device and the calculation device is obtained and marked as a device distance; if there is a device distance less than or equal to twice the first distance, the corresponding intermediate device is marked as a recalculation device, the calculation impedance is replaced by the fault impedance corresponding to the recalculation device, and the corresponding first distance is recalculated; if all device distances are greater than twice the first distance, the first distance is not recalculated.

[0087] It should be noted that since at least two fault recording devices will be configured on a power transmission line, there is no case where the number of impedances is less than 2; the reason for recalculating the fault distance is that when the distance of a device is less than or equal to twice the fault distance, it means that the distance between the corresponding intermediate device and the fault location is closer, and the corresponding impedance can better reflect the electrical characteristics of the fault location, so the recalculated fault distance can better accurately locate the actual fault location.

[0088] If the number of impedances is equal to 2, the fault waveform features, fault transient features and fault spectrum features corresponding to the calculation device are obtained from the fault feature vector and marked as first positioning data; if the number of impedances is greater than 2, the fault waveform features, fault transient features and fault spectrum features corresponding to the recalculated device are obtained from the fault feature vector and marked as second positioning data; the first positioning data or the second positioning data is input into the trained distance calculation model to calculate the corresponding second distance.

[0089] The method for calculating randomness includes:

[0090] The number of each different value in the calculation error data is counted and marked as a value number; the number of all values in the calculation error data is counted and marked as a total value number; each value number is divided by the total value number to obtain a value probability; 2 is taken as the base, and each value probability is taken as a true number in turn to calculate the information amount corresponding to each value probability; each information amount is multiplied by the corresponding value probability to obtain a value contribution; each value contribution is added in turn, and the negative is taken to obtain the randomness corresponding to the calculation error data; the calculation method of the randomness corresponding to the model error data is consistent with the calculation method of the randomness corresponding to the calculation error data.

[0091] The method for calculating the weight coefficient includes:

[0092] The randomness corresponding to the calculation error data is marked as a first randomness, and the randomness corresponding to the model error data is marked as a second randomness; the first randomness is added to the second randomness to obtain a total randomness; the first randomness is divided by the total randomness to obtain the weight coefficient corresponding to the calculation error data; the second randomness is divided by the total randomness to obtain the weight coefficient corresponding to the model error data.

[0093] It should be appreciated that when the power transmission network fails, the electrical disturbance will propagate throughout the power transmission network, causing the voltage and current of the lines adjacent to the fault line to fluctuate, thereby triggering the fault recording devices on these lines; but the electrical protection device can usually correctly identify that only the fault line needs to be disconnected, that is, only the electrical protection device on the fault line will generate a circuit breaker tripping signal; therefore, only relying on the fault feature vector cannot accurately determine the fault line in the power transmission network, and the fault line needs to be accurately determined in combination with the protection detection data.

[0094] The position correction module is configured to obtain power grid topology data, correct the geographic coordinates of the fault occurrence position in combination with the power grid topology data and a geographic information system, and obtain the fault geographic position.

[0095] The method for obtaining the fault geographic position comprises the following steps:

[0096] The power grid topology data is imported into the geographic information system, each fault recording device in the power transmission network is marked with coordinates, and the coordinates are marked as device coordinates; the geographic information system is a system for storing, managing, analyzing, processing and visualizing geographic data, which can combine spatial data (such as maps, geographic coordinates, terrain, etc.) with non-spatial data (such as power transmission lines, fault recording devices, etc.), and is helpful to obtain relevant information of the geographic position; the device coordinates of the fault recording device corresponding to the calculated impedance are obtained and marked as standard coordinates; a positioning circle is generated with the standard coordinates as the center and the fault distance as the radius; the intersection of the positioning circle and the fault line is marked as a fault point, the number of fault points is counted, and the number is marked as the fault number; if the fault number is equal to 1, the coordinates of the fault point corresponding to the geographic information system are obtained and used as the fault geographic position; if the fault number is greater than 1, the device coordinates corresponding to the calculation device are obtained and marked as calculation coordinates; an auxiliary circle is generated with the calculation coordinates as the center and the fault distance as the radius; the intersection of the auxiliary circle and the fault line is marked as an auxiliary point; the coordinates of each fault point are obtained and marked as first coordinates; the coordinates of each auxiliary point are obtained and marked as second coordinates; the Euclidean distance between each first coordinate and each second coordinate is calculated in turn, and the Euclidean distance is marked as a point distance; the calculation method of the Euclidean distance is a prior art, and the specific process will not be described in detail here; the first coordinate corresponding to the minimum point distance is used as the fault geographic position; since there is a fault occurrence position, there is no case where the fault number is less than 1.

[0097] The risk assessment module is configured to calculate the fault impact range and the fault severity according to the fault geographic position and the fault feature vector, and generate a fault diagnosis report.

[0098] The method for generating the fault diagnosis report comprises the following steps:

[0099] mark each fault recording device corresponding to each set of fault electrical quantity data as a fault detection device; according to the power grid topology data, obtain the line set between each fault detection device and the fault geographical position, and the total length of each line corresponding to each line set; add the total length of each line corresponding to each line set in turn to obtain the influence range corresponding to each line set; sort each influence range from large to small to generate an influence sorting table; take the influence range ranked first in the influence sorting table as the fault influence range;

[0100] take the fault feature vector and the fault influence range as influence data, and input the influence data into the trained degree prediction model to predict the corresponding fault severity; the training process of the degree prediction model is consistent with the training process of the single feature extraction model, and both are deep neural network models;

[0101] According to the fault influence range, the fault severity and the fault geographical position, a fault diagnosis report is generated.

[0102] The embodiment can effectively solve the problem of data time axis deviation caused by factors such as sampling rate and signal propagation delay by time alignment of the collected multiple sets of fault electrical quantity data to establish a unified fault data set; intelligent AI technology is used to realize automatic extraction of various fault feature parameters, which can fully characterize the characteristics of power transmission network faults; the protection detection data and the fault feature vector are fused, and a hybrid diagnosis strategy is used to locate the fault position, which improves the positioning accuracy in complex fault scenarios and reduces the error influence; combined with the power grid topology data and the geographic information system, the fault position is corrected to determine the accurate geographical position of the fault; based on the fault feature and the geographical position, the fault influence range and the severity are calculated to fully evaluate the influence of the fault on the power grid, and more reliable fault diagnosis basis is provided for power grid dispatchers; the robustness and response speed of fault diagnosis can be effectively improved, the fault occurrence position can be accurately located, the fault recovery time can be reduced, the safety and stability of the power transmission network can be enhanced, and significant economic and social benefits can be achieved.

[0103] Embodiment 2

[0104] Please refer to Figure 2 The embodiment does not describe some parts in detail, which can be seen from the description of embodiment 1, and provides a hybrid fault diagnosis method for power transmission network based on intelligent AI, which comprises:

[0105] collect sets of fault electrical quantity data, is an integer greater than 1;

[0106] Each set of fault electrical quantity data is standardized and processed to construct a fault data set;

[0107] Wavelet transform and deep learning technology are used to extract a fault feature vector from the fault data set, and the fault feature vector includes a fault waveform feature, a fault transient feature and a fault frequency spectrum feature;

[0108] Protection detection data is acquired, the fault feature vector and the protection detection data are fused, and a hybrid diagnosis strategy is used to quickly locate a fault occurrence position;

[0109] Power grid topology data is acquired, the fault occurrence position is corrected in geographic coordinates in combination with the power grid topology data and a geographic information system, and a fault geographic position is acquired;

[0110] According to the fault geographic position and the fault feature vector, a fault influence range and a fault severity are calculated, and a fault diagnosis report is generated.

[0111] Embodiment 3

[0112] The application also provides an electronic device. The electronic device can include one or more processors and one or more memories. The memory stores computer readable code, which, when executed by the one or more processors, can perform an intelligent AI-based hybrid fault diagnosis method for a power transmission network as described above.

[0113] The method or system according to the embodiments of the application can also be implemented by means of the architecture of the electronic device shown in the application. The electronic device can include a bus, one or more CPUs, ROM, RAM, a communication port connected to a network, input / output, a hard disk, etc. The storage device in the electronic device, such as ROM or hard disk, can store an intelligent AI-based hybrid fault diagnosis method for a power transmission network provided by the application. Further, the electronic device can also include a user interface. Of course, the architecture shown in the application is only exemplary, and when different devices are implemented, one or more components of the electronic device shown in the application can be omitted according to actual needs.

[0114] Embodiment 4

[0115] One embodiment of the application discloses a computer readable storage medium. The computer readable storage medium stores computer readable instructions. When the computer readable instructions are executed by a processor, an intelligent AI-based hybrid fault diagnosis method for a power transmission network according to the embodiments of the application described with reference to the above figures can be performed. The storage medium includes but is not limited to, for example, volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and cache memory, etc. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0116] In addition, according to the embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium storing machine-readable instructions executable by a processor to perform instructions corresponding to the method steps provided by the present application, for example: a hybrid fault diagnosis method for power transmission network based on intelligent AI. When this computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0117] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0118] Finally: the above is only a preferred embodiment of the present application and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included within the protection scope of the present application.

Claims

1. A smart AI-based hybrid diagnostic method for power transmission grid faults, characterized by, The method comprises the following steps: collecting group fault electrical quantity data, is an integer greater than 1; The method for constructing the fault data set comprises the following steps: The method for time aligning the to-be-aligned voltage data and the voltage standard data comprises the following steps: According to the time sequence, the voltages in the to-be-aligned voltage data and the voltage standard data are obtained respectively, and corresponding voltage sequences are constructed respectively; the Euclidean distances between every two voltages in the two voltage sequences are calculated respectively and are marked as voltage distances; the cumulative distance matrix is constructed according to the voltage distances; the element in the first row and the first column in the cumulative distance matrix is marked as a starting element, and the optimal path is obtained by backtracking from the element corresponding to the lower right corner of the cumulative distance matrix to the starting element of the cumulative distance matrix; the voltage matching relationship is established according to the optimal path, the voltage matching relationship is the corresponding relationship between the voltage in the to-be-aligned voltage data and the voltage in the voltage standard data; the to-be-aligned voltage data is subjected to interpolation processing according to the voltage matching relationship, so that the to-be-aligned voltage data is aligned with the voltage standard data on the time axis; The fault feature vector is extracted from the fault data set by using wavelet transform and deep learning technology, and the fault feature vector comprises a fault waveform feature, a fault transient feature and a fault frequency spectrum feature; The protection detection data is obtained, the fault feature vector and the protection detection data are fused, and a hybrid diagnosis strategy is adopted to quickly locate the fault occurrence position; The power grid topology data is obtained, the fault occurrence position is corrected in geographical coordinates by combining the power grid topology data and a geographic information system, and the fault geographical position is obtained; The fault influence range and the fault severity are calculated according to the fault geographical position and the fault feature vector, and a fault diagnosis report is generated. The fault electrical quantity data comprises voltage waveform data and current waveform data; The preliminary processing data is obtained by denoising each set of fault electrical quantity, and the preliminary processing data comprises preliminary voltage processing data and preliminary current processing data; 2. The smart AI based hybrid power grid fault diagnostic method as claimed in claim 1 wherein, The number of data in each set of preliminary voltage processing data is counted respectively and is marked as a data number; each data number is compared respectively, the data number with the maximum value is marked as a maximum number, and the preliminary voltage processing data corresponding to the maximum number is marked as voltage standard data; The preliminary current processing data corresponding to the same fault recording device as the voltage standard data is marked as current standard data; the preliminary voltage processing data that is not marked as voltage standard data is marked as to-be-aligned voltage data, and the preliminary current processing data that is not marked as current standard data is marked as to-be-aligned current data; The method comprises the following steps: ​ The method of time aligning the voltage data to be aligned with the voltage standard data is consistent with the method of time aligning the current data to be aligned with the current standard data.

3. The hybrid fault diagnosis method for power transmission grid based on smart AI of claim 2, wherein, The distance matrix is constructed according to the voltage distance, and elements in the distance matrix correspond to the voltage distance one by one; the cumulative distance matrix is constructed according to the distance matrix, and the size of the cumulative distance matrix is consistent with the size of the distance matrix; An element in the first row and the first column of the cumulative distance matrix is marked as a starting point element, elements in the first row of the cumulative distance matrix are marked as head elements, elements in the first column of the cumulative distance matrix are marked as index elements, and elements in the cumulative distance matrix that are not marked as the starting point element, the head element or the index element are marked as data elements; The starting point element in the cumulative distance matrix is equal to the starting point element in the distance matrix, each head element in the cumulative distance matrix is the previous head element plus the corresponding element in the distance matrix, and each index element in the cumulative distance matrix is the previous index element plus the corresponding element in the distance matrix; each data element in the cumulative distance matrix is the minimum value in the corresponding adjacent elements plus the corresponding element in the distance matrix, and the adjacent elements corresponding to the data element include elements located on the left side of the data element, elements located above the data element and elements located on the upper left side of the data element in the cumulative distance matrix; The process of backtracking is as follows: an element backtracked to is marked as a current element, the minimum value in the adjacent elements corresponding to the current element in the cumulative distance matrix is selected as a backtracking element, and the current element is updated to the backtracking element; The method for interpolating the voltage data to be aligned is as follows: for a voltage in the data to be aligned in the optimal path, the voltage in the voltage standard data is When a voltage is generated, the corresponding voltage is marked as the interpolation voltage, and a linear interpolation method is used at the interpolation voltage to generate A new voltage, is an integer greater than 1.

4. The hybrid fault diagnosis method for power transmission grid based on intelligent AI of claim 3, wherein, The fault waveform features include amplitude features and waveform change rates; and the fault transient features include transient amplitudes and transient durations. The fault spectral features include spectral amplitudes and frequency ratios. The voltage waveform data after the normalization processing are all marked as voltage standard waveforms, and the current waveform data are all marked as current standard waveforms. The amplitude features include maximum amplitudes and minimum amplitudes of the voltage standard waveforms and maximum amplitudes and minimum amplitudes of the current standard waveforms; the waveform change rates include maximum instantaneous slopes of the voltage standard waveforms and maximum instantaneous slopes of the current standard waveforms; the transient amplitudes include maximum instantaneous deviation values of the voltage standard waveforms and maximum instantaneous deviation values of the current standard waveforms; and the transient durations include time lengths of high-amplitude changes of the voltage and the current. The spectral amplitudes include amplitude distributions of the voltage signal at different frequencies and amplitude distributions of the current signal at different frequencies; and the frequency ratios include amplitude ratio values of the voltage signal corresponding to a fundamental wave and a harmonic wave and amplitude ratio values of the current signal corresponding to the fundamental wave and the harmonic wave. The method of extracting the fault feature vector from the fault data set includes: Wavelet transform is performed on the voltage standard waveform and the current standard waveform in the fault data set respectively to obtain fault transient characteristics; the voltage standard waveform and the current standard waveform in the fault data set are respectively input into the trained feature extraction model to extract fault feature data, the fault feature data including fault waveform characteristics and fault frequency spectrum characteristics; wherein the feature extraction model includes single feature extraction models, the number of data in the fault feature data, the single feature extraction model corresponding to the data in the fault feature data one by one; each single feature extraction model is a deep neural network model, and the training processes of the single feature extraction models are consistent.

5. The smart AI-based hybrid diagnostic method for power grid fault according to claim 4, wherein, The protection detection data is a circuit breaker tripping signal. The method of locating the fault occurrence position includes: The Fourier transform is performed on the voltage standard waveform and the current standard waveform in the fault data set respectively, each voltage standard waveform and each current standard waveform is converted from time domain to frequency domain, and the corresponding fundamental components are extracted respectively; each voltage fundamental component is marked as a voltage fundamental component, and each current fundamental component is marked as a current fundamental component; each voltage fundamental component is divided by the corresponding current fundamental component to obtain the corresponding impedance; the line corresponding to the protection detection data in the power transmission network is marked as a fault line, the impedance corresponding to the fault line is obtained and marked as a fault impedance; the number of fault impedances is counted and marked as an impedance number; according to the impedance number, the fault distance is calculated by using a calculated impedance and a distance calculation model respectively, the fault distance calculated by using the calculated impedance is marked as a first distance, and the fault distance calculated by using the distance calculation model is marked as a second distance; wherein, the calculated impedance is the fault impedance used to calculate the fault distance, the fault distance is the distance between the fault location and the fault recording device corresponding to the calculated impedance, the training process of the distance calculation model is the same as that of the single feature extraction model, and both are deep neural network models; acquiring historical data, the historical data including distance error data acquired at different historical moments distance error data, the distance error data including calculation error data and model error data, is an integer greater than 1; wherein the calculation error data is a difference between a fault distance calculated by using a calculation impedance and a corresponding actual distance, the model error data is a difference between a fault distance calculated by using a distance calculation model and the corresponding actual distance, and the actual distance is a distance between an actual fault occurrence position and a fault recording device corresponding to the calculation impedance; a randomness corresponding to the calculation error data and the model error data is calculated respectively, a weight coefficient corresponding to the calculation error data and the model error data is calculated according to the randomness respectively; a first distance is multiplied by the weight coefficient corresponding to the calculation error data, and then a second distance is added after being multiplied by the weight coefficient corresponding to the model error data, to acquire a weighted distance, and the fault occurrence position is located according to the weighted distance.

6. The hybrid fault diagnosis method for power transmission grid based on smart AI of claim 5, wherein, The method for calculating the fault distance by using the calculated impedance and the distance calculation model respectively comprises: If the impedance number is equal to 2, one fault impedance is randomly selected as the calculated impedance; the power grid topology data is obtained, the total length of the line corresponding to the fault line and the unit impedance are obtained according to the power grid topology data, the total length of the line is multiplied by the unit impedance to obtain the total impedance of the line; the calculated impedance is divided by the total impedance of the line, and then multiplied by the total length of the line to obtain the first distance; If the impedance number is greater than 2, the fault recording devices located at both ends of the fault line are marked as edge devices, the fault impedances corresponding to the edge devices are marked as edge impedances, one edge impedance is randomly selected as the calculated impedance, and the corresponding first distance is calculated; the fault recording device corresponding to the calculated impedance is marked as a calculation device, and the fault recording devices that are not marked as edge devices are marked as intermediate devices; according to the power grid topology structure, the distance between each intermediate device and the calculation device is obtained and marked as a device distance; if there is a device distance less than or equal to twice the first distance, the corresponding intermediate device is marked as a recalculation device, the calculated impedance is replaced by the fault impedance corresponding to the recalculation device, and the corresponding first distance is recalculated; if all device distances are greater than twice the first distance, the first distance is not recalculated; If the impedance number is equal to 2, the fault waveform features, the fault transient features and the fault spectrum features corresponding to the calculation device are obtained from the fault feature vector and marked as first positioning data; if the impedance number is greater than 2, the fault waveform features, the fault transient features and the fault spectrum features corresponding to the recalculation device are obtained from the fault feature vector and marked as second positioning data; the first positioning data or the second positioning data is input into the trained distance calculation model to calculate the corresponding second distance.

7. The hybrid fault diagnosis method for power grid based on intelligent AI of claim 6, wherein, The method for calculating randomness comprises: The number of each different value in the error data is counted and marked as a value number; the number of all values in the error data is counted and marked as a total value number; each value number is divided by the total value number to obtain a value probability; the value probability is taken as a true number to calculate the information amount corresponding to each value probability; each information amount is multiplied by the corresponding value probability to obtain a value contribution; each value contribution is added in sequence, and the negative is taken to obtain the randomness corresponding to the calculation error data; the calculation method of the randomness corresponding to the model error data is consistent with the calculation method of the randomness corresponding to the calculation error data; The method for calculating the weight coefficient comprises: The randomness corresponding to the calculation error data is marked as a first randomness, and the randomness corresponding to the model error data is marked as a second randomness; the first randomness is added to the second randomness to obtain a total randomness; the first randomness is divided by the total randomness to obtain a weight coefficient corresponding to the calculation error data; the second randomness is divided by the total randomness to obtain a weight coefficient corresponding to the model error data.

8. The hybrid fault diagnosis method for power transmission grid based on smart AI of claim 7, wherein, The method for obtaining the fault geographical position comprises: The power grid topology data is imported into a geographic information system, each fault recording device in the power transmission network is coordinate-labeled, and is marked as a device coordinate; the device coordinate of the fault recording device corresponding to the calculation impedance is obtained and is marked as a standard coordinate; a positioning circle is generated with the standard coordinate as the center and the fault distance as the radius; the intersection of the positioning circle and the fault line is marked as a fault point, the number of fault points is counted and is marked as a fault number; if the fault number is equal to 1, the coordinate corresponding to the fault point is obtained from the geographic information system and is taken as the fault geographical position; if the fault number is greater than 1, the device coordinate corresponding to the calculation device is obtained and is marked as a calculation coordinate; an auxiliary circle is generated with the calculation coordinate as the center and the fault distance as the radius; the intersection of the auxiliary circle and the fault line is marked as an auxiliary point; the coordinate of each fault point is obtained and is marked as a first coordinate; the coordinate of each auxiliary point is obtained and is marked as a second coordinate; the Euclidean distance between each first coordinate and each second coordinate is calculated in sequence and is marked as a point distance; the first coordinate corresponding to the minimum point distance is taken as the fault geographical position.

9. The hybrid fault diagnosis method for power grid based on intelligent AI of claim 8, wherein, The method for generating a fault diagnosis report comprises: Each fault recording device corresponding to each group of fault electrical quantity data is marked as a fault detection device; according to the power grid topology data, a line set between each fault detection device and the fault geographical position is obtained, and the total length of each line in each line set is obtained; the total length of each line in each line set is added in sequence to obtain the influence range corresponding to each line set; each influence range is sorted from large to small to generate an influence sorting table; the influence range ranked first in the influence sorting table is taken as the fault influence range; The fault feature vector and the fault influence range are taken as influence data, the influence data are input into the trained degree prediction model, and corresponding fault severity is predicted; the training process of the degree prediction model is consistent with the training process of the single feature extraction model, and both are deep neural network models; According to the fault influence range, the fault severity and the fault geographic location, a fault diagnosis report is generated.

10. An intelligent AI-based hybrid power grid fault diagnosis system for implementing the intelligent AI-based hybrid power grid fault diagnosis method according to any one of claims 1-9, characterized in that, It comprises: The data acquisition module is configured to acquire group fault electrical quantity data, is an integer greater than 1. a data processing module configured to perform standardization processing on each group of fault electrical quantity data respectively, and to construct a fault data set; a feature extraction module configured to extract a fault feature vector from the fault data set by using wavelet transform and deep learning technology, the fault feature vector comprising a fault waveform feature, a fault transient feature and a fault frequency spectrum feature; a fault positioning module configured to acquire protection detection data, fuse the fault feature vector and the protection detection data, and quickly locate a fault occurrence position by using a hybrid diagnosis strategy; a position correction module configured to acquire power grid topology data, correct the fault occurrence position in geographic coordinates by combining the power grid topology data and a geographic information system, and acquire a fault geographic location; a risk assessment module configured to calculate a fault influence range and a fault severity according to the fault geographic location and the fault feature vector, and to generate a fault diagnosis report.

Images