A method for fusion analysis of live detection data in online monitoring of substation

Abstract

The invention relates to the technical field of cable fault detection, and discloses a data fusion and analysis method for on-line monitoring of a substation, comprising the following steps: step S102, extracting an abnormal traveling wave signal from a traveling wave signal of a monitoring device C; step S103, extracting an abnormal traveling wave signal from traveling wave signals of a monitoring device A and a monitoring device B; step S104, preprocessing the abnormal traveling wave signals of the monitoring devices A, B and C respectively to obtain a first time-frequency matrix, a second time-frequency matrix and a third time-frequency matrix; step S105, inputting the first time-frequency matrix and the third time-frequency matrix into a fault location model, and outputting values ​​representing the positions of a plurality of fault points between the monitoring device A and the monitoring device C; the invention establishes a nonlinear mapping relationship between traveling wave signals superimposed by a plurality of fault points and the actual fault point positions through a fault location model, and combines the idea of ​​dichotomy to improve the accuracy and speed of cable fault point location.

Description

Technical Field

[0001] The present invention relates to the technical field of cable fault detection, and more specifically, to a method for fusion and analysis of live detection data of online monitoring of a substation. Background Art

[0002] As the core hub of the power system, the substation undertakes the important task of power conversion and distribution. The cable is an important facility connecting the key equipment in the substation, and its operating status directly affects the stability of the power system. Partial discharge is one of the key factors affecting the insulation performance of the cable. It often indicates the early degradation or damage of the cable insulation layer. Partial discharge will not only accelerate the aging of the insulation material, but also may cause the cable to fail, thus affecting the normal operation of the power system. Therefore, it is particularly important to monitor the partial discharge of wires and cables.

[0003] When a cable generates partial discharge, a transient electromagnetic wave will be generated at the fault point. This electromagnetic wave will propagate to both ends of the cable and be captured by the monitoring devices at both ends of the cable. The monitoring devices record the arrival time of the traveling wave, and then calculate the distance of the fault point relative to the monitoring device based on the propagation speed of the traveling wave in the cable, thereby realizing the fault location of the cable partial discharge. However, when there are multiple fault points in the cable at the same time, each fault point will generate a transient electromagnetic wave. When these electromagnetic waves encounter the cable terminal or other impedance discontinuity points, reflection and transmission phenomena will occur, causing the traveling wave signal to form a complex waveform after multiple reflections and superpositions, resulting in inaccurate positioning of the cable fault point. Summary of the invention

[0004] The present invention provides a method for fusion and analysis of live detection data of online monitoring of a substation, which solves the technical problems in the above-mentioned background technology.

[0005] The present invention provides a method for fusion analysis of live detection data of substation online monitoring, comprising the following steps:

[0006] Step S101, installing a monitoring device A and a monitoring device B at both ends of the cable to be monitored, respectively, and installing a monitoring device C at the center point of the cable to be monitored;

[0007] Step S102, obtaining a traveling wave signal of the monitoring device C, and extracting an abnormal traveling wave signal from the traveling wave signal;

[0008] The traveling wave signal is represented by a time domain waveform diagram, the horizontal axis of the time domain waveform diagram represents the time point, and the vertical axis represents the signal strength, and the signal strength is represented by voltage, and the unit is volt;

[0009] The abnormal traveling wave signal includes the signal strength at U time points, where U is a custom parameter;

[0010] Step S103, extracting abnormal traveling wave signals from the traveling wave signals of the monitoring device A and the monitoring device B according to the abnormal traveling wave signal of the monitoring device C;

[0011] Step S104, preprocessing the abnormal traveling wave signals of the monitoring device A, the monitoring device B and the monitoring device C respectively to obtain a first time-frequency matrix, a second time-frequency matrix and a third time-frequency matrix;

[0012] The first time-frequency matrix includes U rows and V columns, and the element value of the i-th row and j-th column represents the power spectrum density of the j-th frequency value of the abnormal traveling wave signal of the monitoring device A at the i-th time point, where 1≤i≤U, 1≤j≤V, and V is a custom parameter;

[0013] The second time-frequency matrix and the third time-frequency matrix have the same representation as the first time-frequency matrix;

[0014] Step S105, inputting the first time-frequency matrix and the third time-frequency matrix into the fault location model, and the output values ​​represent the locations of multiple fault points between the monitoring device A and the monitoring device C;

[0015] Step S106: input the second time-frequency matrix and the third time-frequency matrix into the fault location model, and the output value represents the location of multiple fault points between monitoring device B and monitoring device C.

[0016] Furthermore, a time domain waveform diagram corresponding to the traveling wave signal of the monitoring device C that meets the preset rules is obtained until the number of time points at which the time domain waveform diagram is obtained is equal to U, and the time domain waveform diagram is used as the abnormal traveling wave signal of the monitoring device C.

[0017] Furthermore, the preset rules include: within a preset time window, the first indicator is greater than or equal to a first threshold; within the preset time window, the second indicator is greater than or equal to a second threshold; within the preset time window, the third indicator is greater than or equal to a third threshold; the signal strength of the time domain waveform is greater than or equal to a fourth threshold; wherein the number of time points within the preset time window, the first threshold, the second threshold, the third threshold and the fourth threshold are all custom parameters;

[0018] First indicator The calculation formula is as follows:

[0019] ;

[0020] in represents the number of time points in which the signal strength in the time domain waveform graph within the preset time window is greater than or equal to the fifth threshold, , Y represents the number of time points in the preset time window, where the fifth threshold is a custom parameter;

[0021] Second indicator The calculation formula is as follows:

[0022] ;

[0023] The third indicator The calculation formula is as follows:

[0024] ;

[0025] in Indicates the signal strength at the yth time point within the preset time window, and Respectively represent the mean and standard deviation of the signal intensity at Y time points within the preset time window.

[0026] Furthermore, the traveling wave signals of the monitoring device A and the monitoring device B at the same time point as the abnormal traveling wave signal of the monitoring device C are extracted as the abnormal traveling wave signals of the monitoring device A and the monitoring device B.

[0027] Furthermore, the abnormal traveling wave signal is preprocessed to obtain a time-frequency matrix, including the following steps:

[0028] Step S201, removing noise from the abnormal traveling wave signal through a high-pass filter;

[0029] Step S202, converting the time domain waveform corresponding to the abnormal traveling wave signal into U frequency domain waveforms by fast Fourier transform;

[0030] The horizontal axis of the frequency domain waveform graph represents the frequency value, and the vertical axis represents the power spectrum density;

[0031] Step S203, respectively extracting power spectrum densities corresponding to V frequency values ​​in U frequency domain waveform graphs;

[0032] Step S204: Normalize the U frequency domain waveforms respectively by Min-Max normalization to obtain a time-frequency matrix.

[0033] Further, extracting the power spectrum density corresponding to V frequency values ​​in the frequency domain waveform diagram includes the following steps:

[0034] Step S301, obtaining a traveling wave signal of a cable to be monitored in a normal state, and marking it as a first traveling wave signal;

[0035] Step S302, obtaining a traveling wave signal of the cable to be monitored in a partial discharge state, and marking it as a second traveling wave signal;

[0036] The first traveling wave signal and the second traveling wave signal both include signal strengths at U time points;

[0037] Step S303, converting the time domain waveforms corresponding to the first traveling wave signal and the second traveling wave signal into U first frequency domain waveforms and U second frequency domain waveforms respectively through fast Fourier transform;

[0038] The first frequency domain waveform diagram and the second frequency domain waveform diagram both include power spectrum densities corresponding to K frequency values, where K is greater than V;

[0039] Step S304, calculating the correlation coefficient of the power spectrum density corresponding to K frequency values ​​in the U first frequency domain waveform graphs and the U second frequency domain waveform graphs;

[0040] Correlation coefficient of the power spectral density corresponding to the kth frequency value The calculation formula is as follows:

[0041] ;

[0042] Where 1≤k≤K, represents the power spectral density corresponding to the kth frequency value in the uth first frequency domain waveform, represents the average value of the power spectrum density corresponding to the kth frequency value in the U first frequency domain waveforms, represents the power spectral density corresponding to the kth frequency value in the uth second frequency domain waveform, represents the average value of the power spectrum density corresponding to the kth frequency value in the U second frequency domain waveform graphs;

[0043] Step S305 , sorting the correlation coefficients of the power spectrum densities corresponding to the K frequency values ​​in descending order, and obtaining the power spectrum densities corresponding to the first V frequency values.

[0044] Further, the fault location model includes: a feature fusion layer, a feature extraction layer, a feature update layer, a first classifier, a second classifier, a third classifier and a fourth classifier;

[0045] The feature fusion layer is used to fuse the two input time-frequency matrices to obtain fusion features;

[0046] The fusion feature is represented by a matrix of U rows and V columns. The element value of the u-th row and v-th column of the fusion feature is represented by a combination vector. The combination vector includes two dimensional values, which correspond to the element values ​​of the u-th row and v-th column of the two time-frequency matrices, where 1≤u≤U, 1≤v≤V;

[0047] The feature extraction layer includes U first units, the uth first unit inputs the combination vector of the 1st column to the Vth column of the uth row of the fusion feature, and outputs the first updated feature;

[0048] U first units share weight parameters and bias parameters;

[0049] The feature update layer includes U second units, the uth second unit inputs the first updated feature output by the uth first unit, and outputs the second updated feature;

[0050] The second updated features output by the U-th second unit are respectively input into the first classifier and the second classifier;

[0051] The classification space of the first classifier represents the distances between the monitoring device A and the monitoring device C and the multiple fault points.

[0052] The classification space of the second classifier represents the distances between the monitoring device B and the monitoring device C and the multiple fault points.

[0053] Furthermore, the calculation formula of the u-th first unit is as follows:

[0054] ;

[0055] in represents the first updated feature of the u-th first unit output, Represents the combined vector of the 1st column to the Vth column of the uth row of the fused feature input of the uth first unit, and denote the first weight parameter and the second weight parameter of the u-th first unit respectively, and They respectively represent the first bias parameter and the second bias parameter of the u-th first unit, and Swish represents the Swish activation function.

[0056] Furthermore, the calculation formula of the u-th second unit includes:

[0057] ;

[0058] ;

[0059] ;

[0060] ;

[0061] ;

[0062] ;

[0063] ;

[0064] ;

[0065] ;

[0066] ;

[0067] in represents the second updated feature of the u-th second unit output, represents the first updated feature of the u-th first unit output of the u-th second unit input, d represents the number of dimensions of the first updated feature, , and denote the first conversion feature, the second conversion feature and the third conversion feature of the u-th second unit respectively, and They represent the first gating value and the second gating value of the u-th second unit respectively, and both are real values. , , and denote the first intermediate feature, the second intermediate feature, the third intermediate feature and the fourth intermediate feature of the u-th second unit respectively, and They represent the first intermediate feature and the second intermediate feature of the u-1th second unit respectively, and All are assigned a value of 0. , , , , and denote respectively the first weight parameter, the second weight parameter, the third weight parameter, the fourth weight parameter, the fifth weight parameter and the sixth weight parameter of the u-th second unit, , , , , and They respectively represent the first bias parameter, the second bias parameter, the third bias parameter, the fourth bias parameter, the fifth bias parameter and the sixth bias parameter of the u-th second unit, T represents the transposition operation, and max represents the maximum value. represents point-by-point multiplication, sigmoid represents the sigmoid activation function, and exp represents the power function of a natural constant.

[0068] Furthermore, obtaining training samples for training the fault location model through a simulation platform includes the following steps:

[0069] Step S401, constructing a cable model through a simulation platform, and installing a monitoring device A, a monitoring device B, and a monitoring device C at both ends and a center of the cable model respectively;

[0070] Step S402, generating a first time-frequency matrix, a second time-frequency matrix and a third time-frequency matrix as training data of a training sample according to steps S102 to S104;

[0071] Step S403, generating R fault points at random positions between monitoring device A and monitoring device C and between monitoring device B and monitoring device C of the cable model through a simulation platform, where R is a custom parameter and is a positive integer greater than or equal to 0;

[0072] Step S404, recording the distance between each fault point between monitoring device A and monitoring device C and monitoring device A, and recording the distance between each fault point between monitoring device B and monitoring device C and monitoring device B as a sample label of a training sample;

[0073] Step S405, repeating steps S402 to S403 until G training samples are obtained, where G is a custom parameter.

[0074] The beneficial effect of the present invention is that the present invention establishes a nonlinear mapping relationship between the traveling wave signals superimposed by multiple fault points and the actual fault point position through a fault location model, and combines the idea of ​​dichotomy to improve the accuracy and speed of cable fault point location. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 It is a flow chart of a method for fusion and analysis of live detection data of substation online monitoring of the present invention;

[0076] Figure 2 It is a flow chart of the present invention for preprocessing abnormal traveling wave signals to obtain a time-frequency matrix;

[0077] Figure 3 is a flow chart of extracting power spectrum density corresponding to V frequency values ​​in a frequency domain waveform diagram of the present invention;

[0078] Figure 4 It is a flow chart of obtaining training samples for training a fault location model through a simulation platform of the present invention;

[0079] Figure 5 It is a comparison chart of the test results of the present invention. DETAILED DESCRIPTION

[0080] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0081] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0082] like Figure 1 to Figure 5 As shown, a method for fusion analysis of live detection data of substation online monitoring includes the following steps:

[0083] Step S101, installing a monitoring device A and a monitoring device B at both ends of the cable to be monitored, respectively, and installing a monitoring device C at the center point of the cable to be monitored;

[0084] Step S102, obtaining a traveling wave signal of the monitoring device C, and extracting an abnormal traveling wave signal from the traveling wave signal;

[0085] The traveling wave signal is represented by a time domain waveform diagram, the horizontal axis of the time domain waveform diagram represents the time point, and the vertical axis represents the signal strength, and the signal strength is represented by voltage, and the unit is volt;

[0086] The abnormal traveling wave signal includes the signal strength at U time points, where U is a custom parameter;

[0087] Step S103, extracting abnormal traveling wave signals from the traveling wave signals of the monitoring device A and the monitoring device B according to the abnormal traveling wave signal of the monitoring device C;

[0088] Step S104, preprocessing the abnormal traveling wave signals of the monitoring device A, the monitoring device B and the monitoring device C respectively to obtain a first time-frequency matrix, a second time-frequency matrix and a third time-frequency matrix;

[0089] The first time-frequency matrix includes U rows and V columns, and the element value of the i-th row and j-th column represents the power spectrum density of the j-th frequency value of the abnormal traveling wave signal of the monitoring device A at the i-th time point, where 1≤i≤U, 1≤j≤V, and V is a custom parameter;

[0090] The second time-frequency matrix and the third time-frequency matrix have the same representation as the first time-frequency matrix;

[0091] Step S105, inputting the first time-frequency matrix and the third time-frequency matrix into the fault location model, and the output values ​​represent the locations of multiple fault points between the monitoring device A and the monitoring device C;

[0092] Step S106: input the second time-frequency matrix and the third time-frequency matrix into the fault location model, and the output value represents the location of multiple fault points between monitoring device B and monitoring device C.

[0093] It should be noted that the monitoring device can be a traveling wave sensor or an ultra-high frequency (UHF) sensor, etc. The UHF sensor can detect electromagnetic waves with a frequency range of 300MHz to 3GHz. The cable fault location model provided by the present invention can not only monitor the location of the cable partial discharge fault point, but also be used to locate the cable disconnection fault, grounding fault and short circuit fault, etc. For example, when the cable is broken, the breaking point will also generate transient electromagnetic waves; when the insulation layer of the cable is damaged and contacts the ground, a grounding fault will be formed, and current leakage may occur at the fault point, accompanied by the generation of transient electromagnetic waves; a phase-to-phase short circuit or a phase-to-ground short circuit in the cable will cause a large instantaneous current to flow, generating a strong electromagnetic field, and then generating detectable electromagnetic waves.

[0094] In one embodiment of the present invention, a time domain waveform diagram corresponding to a traveling wave signal of a monitoring device C that satisfies preset rules is obtained until the number of time points at which the time domain waveform diagram is obtained is equal to U, and the time domain waveform diagram is used as an abnormal traveling wave signal of the monitoring device C.

[0095] In one embodiment of the present invention, the preset rules include: within a preset time window, the first indicator is greater than or equal to a first threshold; within the preset time window, the second indicator is greater than or equal to a second threshold; within the preset time window, the third indicator is greater than or equal to a third threshold; the signal strength of the time domain waveform is greater than or equal to a fourth threshold; wherein the number of time points within the preset time window, the first threshold, the second threshold, the third threshold and the fourth threshold are all custom parameters;

[0096] First indicator The calculation formula is as follows:

[0097] ;

[0098] in represents the number of time points in which the signal strength in the time domain waveform graph within the preset time window is greater than or equal to the fifth threshold, , Y represents the number of time points in the preset time window, where the fifth threshold is a custom parameter;

[0099] Second indicator The calculation formula is as follows:

[0100] ;

[0101] The third indicator The calculation formula is as follows:

[0102] ;

[0103] in Indicates the signal strength at the yth time point within the preset time window, and Respectively represent the mean and standard deviation of the signal intensity at Y time points within the preset time window.

[0104] In one embodiment of the present invention, the traveling wave signals of monitoring device A and monitoring device B at the same time point as the abnormal traveling wave signal of monitoring device C are extracted as the abnormal traveling wave signals of monitoring device A and monitoring device B.

[0105] In one embodiment of the present invention, Figure 2 As shown, the abnormal traveling wave signal is preprocessed to obtain a time-frequency matrix, including the following steps:

[0106] Step S201, removing noise from the abnormal traveling wave signal through a high-pass filter;

[0107] Step S202, converting the time domain waveform corresponding to the abnormal traveling wave signal into U frequency domain waveforms by fast Fourier transform;

[0108] The horizontal axis of the frequency domain waveform graph represents the frequency value, and the vertical axis represents the power spectrum density;

[0109] Step S203, respectively extracting power spectrum densities corresponding to V frequency values ​​in U frequency domain waveform graphs;

[0110] Step S204: Normalize the U frequency domain waveforms respectively by Min-Max normalization to obtain a time-frequency matrix.

[0111] In one embodiment of the present invention, Figure 3As shown, extracting the power spectrum density corresponding to V frequency values ​​in the frequency domain waveform diagram includes the following steps:

[0112] Step S301, obtaining a traveling wave signal of a cable to be monitored in a normal state, and marking it as a first traveling wave signal;

[0113] Step S302, obtaining a traveling wave signal of the cable to be monitored in a partial discharge state, and marking it as a second traveling wave signal;

[0114] The first traveling wave signal and the second traveling wave signal both include signal strengths at U time points;

[0115] Step S303, converting the time domain waveforms corresponding to the first traveling wave signal and the second traveling wave signal into U first frequency domain waveforms and U second frequency domain waveforms respectively through fast Fourier transform;

[0116] The first frequency domain waveform diagram and the second frequency domain waveform diagram both include power spectrum densities corresponding to K frequency values, where K is greater than V;

[0117] Step S304, calculating the correlation coefficient of the power spectrum density corresponding to K frequency values ​​in the U first frequency domain waveform graphs and the U second frequency domain waveform graphs;

[0118] Correlation coefficient of the power spectral density corresponding to the kth frequency value The calculation formula is as follows:

[0119] ;

[0120] Where 1≤k≤K, represents the power spectral density corresponding to the kth frequency value in the uth first frequency domain waveform, represents the average value of the power spectrum density corresponding to the kth frequency value in the U first frequency domain waveforms, represents the power spectral density corresponding to the kth frequency value in the uth second frequency domain waveform, represents the average value of the power spectrum density corresponding to the kth frequency value in the U second frequency domain waveform graphs;

[0121] Step S305 , sorting the correlation coefficients of the power spectrum densities corresponding to the K frequency values ​​in descending order, and obtaining the power spectrum densities corresponding to the first V frequency values.

[0122] It should be noted that the correlation coefficient can assist in identifying the frequency values ​​in the abnormal traveling wave signal that are highly correlated with the known traveling wave signal of cable partial discharge, thereby reducing the calculation amount of the fault location model and improving the speed of subsequent cable fault location.

[0123] In one embodiment of the present invention, the fault location model includes: a feature fusion layer, a feature extraction layer, a feature update layer, a first classifier, a second classifier, a third classifier and a fourth classifier;

[0124] The feature fusion layer is used to fuse the two input time-frequency matrices to obtain fusion features;

[0125] The fusion feature is represented by a matrix of U rows and V columns. The element value of the u-th row and v-th column of the fusion feature is represented by a combination vector. The combination vector includes two dimensional values, which correspond to the element values ​​of the u-th row and v-th column of the two time-frequency matrices, where 1≤u≤U, 1≤v≤V;

[0126] For example, if the element value of the first time-frequency matrix in the uth row and vth column is 1.1, and the element value of the third time-frequency matrix in the uth row and vth column is 1.2, then the two dimension values ​​of the combined vector in the uth row and vth column of the fusion feature are 1.1 and 1.2 respectively;

[0127] The feature extraction layer includes U first units, the uth first unit inputs the combination vector of the 1st column to the Vth column of the uth row of the fusion feature, and outputs the first updated feature;

[0128] U first units share weight parameters and bias parameters;

[0129] The feature update layer includes U second units, the uth second unit inputs the first updated feature output by the uth first unit, and outputs the second updated feature;

[0130] The second updated features output by the U-th second unit are respectively input into the first classifier and the second classifier;

[0131] The classification space of the first classifier represents the distances between the monitoring device A and the monitoring device C and the multiple fault points.

[0132] The classification space of the second classifier represents the distances between the monitoring device B and the monitoring device C and the multiple fault points.

[0133] It should be noted that the second updated features output by the Uth second unit can also be input into the third classifier and the fourth classifier respectively. The classification space of the third classifier represents the distances between multiple fault points between monitoring device A and monitoring device C and monitoring device C, and the classification space of the fourth classifier represents the distances between multiple fault points between monitoring device B and monitoring device C and monitoring device C.

[0134] In one embodiment of the present invention, the calculation formula of the u-th first unit is as follows:

[0135] ;

[0136] in represents the first updated feature of the u-th first unit output, Represents the combined vector of the 1st column to the Vth column of the uth row of the fused feature input of the uth first unit, and denote the first weight parameter and the second weight parameter of the u-th first unit respectively, and They respectively represent the first bias parameter and the second bias parameter of the u-th first unit, and Swish represents the Swish activation function.

[0137] It should be noted that the weight parameters and bias parameters in the first unit are all learnable hyperparameters, where If it is a matrix of size 2×V, then the first weight parameter of the first unit can be designed as a matrix of size 8×2, and the matrix of size 8×2 is multiplied by the matrix of size 2×V to obtain a matrix of size 8×V. The second weight parameter of the first unit can be designed as a vector of size 1×8, and the vector of size 1×8 is multiplied by the matrix of size 8×V to obtain a vector of size 1×V, that is, the first updated feature is a vector of size 1×V.

[0138] In one embodiment of the present invention, the calculation formula of the u-th second unit includes:

[0139] ;

[0140] ;

[0141] ;

[0142] ;

[0143] ;

[0144] ;

[0145] ;

[0146] ;

[0147] ;

[0148] ;

[0149] in represents the second updated feature of the u-th second unit output, represents the first updated feature of the u-th first unit output of the u-th second unit input, d represents the number of dimensions of the first updated feature, , and denote the first conversion feature, the second conversion feature and the third conversion feature of the u-th second unit respectively, and They represent the first gating value and the second gating value of the u-th second unit respectively, and both are real values. , , and denote the first intermediate feature, the second intermediate feature, the third intermediate feature and the fourth intermediate feature of the u-th second unit respectively, and They represent the first intermediate feature and the second intermediate feature of the u-1th second unit respectively, and All are assigned a value of 0. , , , , and denote respectively the first weight parameter, the second weight parameter, the third weight parameter, the fourth weight parameter, the fifth weight parameter and the sixth weight parameter of the u-th second unit, , , , , and They respectively represent the first bias parameter, the second bias parameter, the third bias parameter, the fourth bias parameter, the fifth bias parameter and the sixth bias parameter of the u-th second unit, T represents the transposition operation, and max represents the maximum value. represents point-by-point multiplication, sigmoid represents the sigmoid activation function, and exp represents the power function of a natural constant.

[0150] It should be noted that the weight parameters and bias parameters in the second unit are both learnable hyperparameters, the first updated feature is a vector of size 1×V, the first weight parameter, the second weight parameter and the third weight parameter of the second unit can all be designed as matrices of size V×8, then the first conversion feature, the second conversion feature and the third conversion feature are vectors of size 1×8, the fourth weight parameter and the fifth weight parameter of the second unit are both vectors of size V×1, then the first gating value and the second gating value are both real values, the first intermediate feature obtained by multiplying the third conversion feature by the transpose of the second conversion feature is also a real value, the second intermediate feature is a vector of size 1×8, the third intermediate feature is also a vector of size 1×8, then the sixth weight parameter of the second unit is designed as a matrix of size V×8, the fourth intermediate feature is a vector of size 1×8, then the second update feature obtained by point-by-point multiplication of the two 1×8 vectors is a vector of size 1×8.

[0151] In one embodiment of the present invention, Figure 4 As shown, obtaining training samples for training a fault location model through a simulation platform includes the following steps:

[0152] Step S401, constructing a cable model through a simulation platform, and installing a monitoring device A, a monitoring device B, and a monitoring device C at both ends and a center of the cable model respectively;

[0153] Step S402, generating a first time-frequency matrix, a second time-frequency matrix and a third time-frequency matrix as training data of a training sample according to steps S102 to S104;

[0154] Step S403, generating R fault points at random positions between monitoring device A and monitoring device C and between monitoring device B and monitoring device C of the cable model through a simulation platform, where R is a custom parameter and is a positive integer greater than or equal to 0;

[0155] Step S404, recording the distance between each fault point between monitoring device A and monitoring device C and monitoring device A, and recording the distance between each fault point between monitoring device B and monitoring device C and monitoring device B as a sample label of a training sample;

[0156] Step S405, repeating steps S402 to S403 until G training samples are obtained, where G is a custom parameter, and preferably, G is set to 3000.

[0157] It should be noted that the difference between the value output by the fault localization model at each iteration and the sample label of the training sample is specified as the loss function, the gradient information of the weight parameters and bias parameters in the fault localization model is calculated by the chain rule, and the weight parameters and bias parameters in the fault localization model are updated by the gradient descent algorithm, which will not be elaborated here.

[0158] It should be noted that the simulation platform is Simulink, and the cable model has the same parameters as the cable to be tested, such as geometric dimensions (including the outer diameter, inner diameter, conductor diameter, etc. of the cable), material properties (including the resistivity of the conductor, the dielectric constant and loss tangent of the insulation layer, the resistivity of the shielding layer, etc.), structural layout (including single-core or multi-core cables, the positional relationship between conductors and shielding layers, etc.), etc. Building a cable model through a simulation platform is a conventional technical means and will not be elaborated here.

[0159] like Figure 5As shown, 10 groups of cable fault location tests are completed by the fault location model and traveling wave method provided by the present invention respectively, the distance between monitoring device A and monitoring device C is 10 km, and the test result is the distance between the fault point (taking local discharge as an example) and monitoring device A (it can also be the distance between the fault point and monitoring device B), in km. It is found through observation that the fault location model provided by the present invention is more accurate than the traveling wave method.

[0160] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A method for fusion analysis of live detection data for online monitoring of substations, characterized in that: The following steps are involved: Step S101, installing a monitoring device A and a monitoring device B at both ends of the cable to be monitored, respectively, and installing a monitoring device C at the center point of the cable to be monitored; Step S102, obtaining a traveling wave signal of the monitoring device C, and extracting an abnormal traveling wave signal from the traveling wave signal; The traveling wave signal is represented by a time domain waveform diagram, the horizontal axis of the time domain waveform diagram represents the time point, and the vertical axis represents the signal strength, and the signal strength is represented by voltage, and the unit is volt; The abnormal traveling wave signal includes the signal strength at U time points, where U is a custom parameter; Step S103, extracting abnormal traveling wave signals from the traveling wave signals of the monitoring device A and the monitoring device B according to the abnormal traveling wave signal of the monitoring device C; Step S104, preprocessing the abnormal traveling wave signals of the monitoring device A, the monitoring device B and the monitoring device C respectively to obtain a first time-frequency matrix, a second time-frequency matrix and a third time-frequency matrix; The first time-frequency matrix includes U rows and V columns, and the element value of the i-th row and j-th column represents the power spectrum density of the j-th frequency value of the abnormal traveling wave signal of the monitoring device A at the i-th time point, where 1≤i≤U, 1≤j≤V, and V is a custom parameter; The second time-frequency matrix and the third time-frequency matrix have the same representation as the first time-frequency matrix; Step S105, inputting the first time-frequency matrix and the third time-frequency matrix into the fault location model, and the output values ​​represent the locations of multiple fault points between the monitoring device A and the monitoring device C; Step S106, inputting the second time-frequency matrix and the third time-frequency matrix into the fault location model, and the output value represents the location of multiple fault points between the monitoring device B and the monitoring device C; Preprocessing the abnormal traveling wave signal to obtain a time-frequency matrix includes the following steps: Step S201, removing noise from the abnormal traveling wave signal through a high-pass filter; Step S202, converting the time domain waveform corresponding to the abnormal traveling wave signal into U frequency domain waveforms by fast Fourier transform; The horizontal axis of the frequency domain waveform graph represents the frequency value, and the vertical axis represents the power spectrum density; Step S203, respectively extracting power spectrum densities corresponding to V frequency values ​​in U frequency domain waveform graphs; Step S204, normalizing the U frequency domain waveforms respectively by Min-Max normalization to obtain a time-frequency matrix; Extracting the power spectrum density corresponding to V frequency values ​​in the frequency domain waveform diagram includes the following steps: Step S301, obtaining a traveling wave signal of a cable to be monitored in a normal state, and marking it as a first traveling wave signal; Step S302, obtaining a traveling wave signal of the cable to be monitored in a partial discharge state, and marking it as a second traveling wave signal; The first traveling wave signal and the second traveling wave signal both include signal strengths at U time points; Step S303, converting the time domain waveforms corresponding to the first traveling wave signal and the second traveling wave signal into U first frequency domain waveforms and U second frequency domain waveforms respectively through fast Fourier transform; The first frequency domain waveform diagram and the second frequency domain waveform diagram both include power spectrum densities corresponding to K frequency values, where K is greater than V; Step S304, calculating the correlation coefficient of the power spectrum density corresponding to K frequency values ​​in the U first frequency domain waveform graphs and the U second frequency domain waveform graphs; Correlation coefficient of the power spectral density corresponding to the kth frequency value The calculation formula is as follows: ； Where 1≤k≤K, represents the power spectral density corresponding to the kth frequency value in the uth first frequency domain waveform, represents the average value of the power spectrum density corresponding to the kth frequency value in the U first frequency domain waveforms, represents the power spectral density corresponding to the kth frequency value in the uth second frequency domain waveform, represents the average value of the power spectrum density corresponding to the kth frequency value in the U second frequency domain waveform graphs; Step S305 , sorting the correlation coefficients of the power spectrum densities corresponding to the K frequency values ​​in descending order, and obtaining the power spectrum densities corresponding to the first V frequency values.

2. A method for fusion and analysis of live detection data for online monitoring of substations according to claim 1, characterized in that: A time domain waveform diagram corresponding to the traveling wave signal of the monitoring device C that meets the preset rules is obtained until the number of time points of the obtained time domain waveform diagram is equal to U, and the time domain waveform diagram is used as the abnormal traveling wave signal of the monitoring device C.

3. A method for fusion and analysis of live detection data for online monitoring of substations according to claim 2, characterized in that: The preset rules include: within the preset time window, the first indicator is greater than or equal to the first threshold; within the preset time window, the second indicator is greater than or equal to the second threshold; within the preset time window, the third indicator is greater than or equal to the third threshold; the signal strength of the time domain waveform is greater than or equal to the fourth threshold; wherein the number of time points within the preset time window, the first threshold, the second threshold, the third threshold and the fourth threshold are all custom parameters; First indicator The calculation formula is as follows: ； in represents the number of time points in which the signal strength in the time domain waveform graph within the preset time window is greater than or equal to the fifth threshold, , Y represents the number of time points in the preset time window, where the fifth threshold is a custom parameter; Second indicator The calculation formula is as follows: ； The third indicator The calculation formula is as follows: ； in Indicates the signal strength at the yth time point within the preset time window, and Respectively represent the mean and standard deviation of the signal intensity at Y time points within the preset time window.

4. A method for fusion and analysis of live detection data for online monitoring of substations according to claim 1, characterized in that: The traveling wave signals of the monitoring device A and the monitoring device B at the same time point as the abnormal traveling wave signal of the monitoring device C are extracted as the abnormal traveling wave signals of the monitoring device A and the monitoring device B.

5. The method for fusion and analysis of live detection data for online monitoring of substation according to claim 1 is characterized in that: The fault location model includes: a feature fusion layer, a feature extraction layer, a feature update layer, a first classifier, a second classifier, a third classifier and a fourth classifier; The feature fusion layer is used to fuse the two input time-frequency matrices to obtain fusion features; The fusion feature is represented by a matrix of U rows and V columns. The element value of the u-th row and v-th column of the fusion feature is represented by a combination vector. The combination vector includes two dimensional values, which correspond to the element values ​​of the u-th row and v-th column of the two time-frequency matrices, where 1≤u≤U, 1≤v≤V; The feature extraction layer includes U first units, the uth first unit inputs the combination vector of the 1st column to the Vth column of the uth row of the fusion feature, and outputs the first updated feature; U first units share weight parameters and bias parameters; The feature update layer includes U second units, the uth second unit inputs the first updated feature output by the uth first unit, and outputs the second updated feature; The second updated features output by the U-th second unit are respectively input into the first classifier and the second classifier; The classification space of the first classifier represents the distances between the monitoring device A and the monitoring device C and the multiple fault points. The classification space of the second classifier represents the distances between the monitoring device B and the monitoring device C and the multiple fault points.

6. A method for fusion and analysis of live detection data for online monitoring of substations according to claim 5, characterized in that: The calculation formula for the uth first unit is as follows: ； in represents the first updated feature of the u-th first unit output, Represents the combined vector of the 1st column to the Vth column of the uth row of the fused feature input of the uth first unit, and denote the first weight parameter and the second weight parameter of the u-th first unit respectively, and They respectively represent the first bias parameter and the second bias parameter of the u-th first unit, and Swish represents the Swish activation function.

7. A method for fusion and analysis of live detection data for online monitoring of substations according to claim 5, characterized in that: The calculation formula for the u-th second unit include: ； ； ； ； ； ； ； ； ； ； in represents the second updated feature of the u-th second unit output, represents the first updated feature of the u-th first unit output of the u-th second unit input, d represents the number of dimensions of the first updated feature, , and denote the first conversion feature, the second conversion feature and the third conversion feature of the u-th second unit respectively, and They represent the first gating value and the second gating value of the u-th second unit respectively, and both are real values. , , and denote the first intermediate feature, the second intermediate feature, the third intermediate feature and the fourth intermediate feature of the u-th second unit respectively, and They represent the first intermediate feature and the second intermediate feature of the u-1th second unit respectively, and All are assigned a value of 0. , , , , and denote respectively the first weight parameter, the second weight parameter, the third weight parameter, the fourth weight parameter, the fifth weight parameter and the sixth weight parameter of the u-th second unit, , , , , and They respectively represent the first bias parameter, the second bias parameter, the third bias parameter, the fourth bias parameter, the fifth bias parameter and the sixth bias parameter of the u-th second unit, T represents the transposition operation, and max represents the maximum value. represents point-by-point multiplication, sigmoid represents the sigmoid activation function, and exp represents the power function of a natural constant.

8. The method for fusion and analysis of live detection data for online monitoring of substation according to claim 1 is characterized in that: Obtaining training samples for training a fault location model through a simulation platform includes the following steps: Step S401, constructing a cable model through a simulation platform, and installing a monitoring device A, a monitoring device B, and a monitoring device C at both ends and a center of the cable model respectively; Step S402, generating a first time-frequency matrix, a second time-frequency matrix and a third time-frequency matrix as training data of a training sample according to steps S102 to S104; Step S403, generating R fault points at random positions between monitoring device A and monitoring device C and between monitoring device B and monitoring device C of the cable model through a simulation platform, where R is a custom parameter and is a positive integer greater than or equal to 0; Step S404, recording the distance between each fault point between monitoring device A and monitoring device C and monitoring device A, and recording the distance between each fault point between monitoring device B and monitoring device C and monitoring device B as a sample label of a training sample; Step S405, repeating steps S402 to S403 until G training samples are obtained, where G is a custom parameter.

Images