Line fault reason identification method and system based on intelligent model

Abstract

The invention discloses a line fault reason identification method and system based on an intelligent model. The method comprises the following steps: acquiring fault identification data; assigning a fault judgment item of each fault type based on the fault identification data, and obtaining a fault judgment result of each fault type based on a preset weight and assignment of the fault judgment item; generating a fault mechanism judgment result score matrix based on the fault judgment result, and processing the fault mechanism judgment result score matrix to generate a normalized mechanism judgment matrix; establishing a fault data set, and outputting a discrimination probability distribution matrix through the trained intelligent model; processing the normalization mechanism discrimination matrix and the discrimination probability distribution matrix to generate a fusion matrix; establishing a meteorological incidence matrix based on the incidence relation between the meteorological information and the fault reason during the line fault; and through multiplication of the fusion matrix and the meteorological incidence matrix, a final fault reason identification matrix is established, and line fault reasons are identified based on the final fault reason identification matrix.

Description

Technical Field

[0001] The present invention relates to the technical field of relay protection for AC power transmission lines, and more specifically, to a method and system for identifying the cause of a line fault based on an intelligent model. Background Art

[0002] As energy transformation accelerates, the pressure to ensure supply continues to increase. In recent years, the power supply load has continued to rise, and the peak load period is prone to overlap with extreme weather. Equipment failure and shutdown seriously weaken the grid structure, threatening the safety of the power grid and the reliability of power supply. As an important asset of the company and the main road for power transmission, the high-voltage transmission line can quickly and accurately identify the type and cause of the fault after it fails and stops operating, which helps to assist on-site operation and maintenance personnel and dispatching and operation personnel to carry out fault handling and equipment repair, effectively improve the work quality and efficiency of front-line production employees, and further consolidate the foundation of the inherent safety of the power grid.

[0003] Prior art 1, (Application name: Transmission line fault cause identification method based on high- and low-frequency wavelet feature association), proposes a transmission line fault cause identification method based on high- and low-frequency wavelet feature association, by establishing an association model between the wavelet features of high-frequency traveling wave data and the wavelet features of low-frequency recording data, and using low-frequency recording data to identify the fault cause. First, extract a variety of fault phase current samples and establish a database, and extract low-frequency and high-frequency feature vectors through wavelet transform and BP neural network. Secondly, establish a correlation relationship model between fault feature quantity and fault type based on the above results. Finally, the feature quantity of the sample of the fault cause to be tested is sequentially brought into the correlation relationship model of multiple fault types, and the output result with the largest fitting degree and greater than the set value is selected, and the sample data of the fault cause to be tested is judged to be the fault type in the model corresponding to the output result. However, prior art 1 identifies the fault cause through fault current information, and among various AC line fault types, similar fault currents may exist between different faults, and the accuracy of judging the fault cause using only fault current information is not high.

[0004] Since the fault identification methods in the prior art mainly rely on manual analysis, it often takes a long time from on-site line inspection to system power supply restoration. In addition, when the fault phenomenon is not obvious, the fault identification method based on manual analysis is difficult to accurately distinguish.

[0005] Therefore, a technology is needed to identify the cause of line faults based on an intelligent model. Summary of the invention

[0006] The technical solution of the present invention provides a method and system for identifying the cause of a line fault based on an intelligent model, so as to solve the problem of how to identify the cause of a line fault based on an intelligent model.

[0007] In order to solve the above problems, the present invention provides a line fault cause identification method based on an intelligent model, the method comprising: Obtaining fault identification data of the faulty line; Assigning values ​​to the fault judgment items of each fault type based on the fault identification data, and obtaining the fault judgment result of each fault type based on the preset weights and values ​​of the fault judgment items of different types of faults; Generate a fault mechanism discrimination result score matrix based on the fault judgment result, and perform normalization processing on the fault mechanism discrimination result score matrix to generate a normalized mechanism discrimination matrix; Establishing a fault data set, and based on the fault data set, outputting a discriminant probability distribution matrix through a trained intelligent model; Performing fusion processing on the normalized mechanism discriminant matrix and the discriminant probability distribution matrix to generate a fusion matrix; Based on the correlation between meteorological information and fault causes during line faults, a meteorological correlation matrix is ​​established; A final fault cause identification matrix is ​​established by multiplying the fusion matrix and the meteorological correlation matrix, and the cause of the line fault is identified based on the final fault cause identification matrix.

[0008] Preferably, the fault types include: lightning strike fault, wind deflection fault, low resistance foreign body fault, high resistance foreign body fault, wildfire fault, pollution flashover fault, ice dancing fault, ice flashover fault, tree flashover fault, bird damage fault and mechanical external force fault; The fault judgment items include: fault type item, fault nature item, coincident power item, non-periodic component item, harmonic content item, fault time item, season item, time item and load item.

[0009] Preferably, generating a fault mechanism discrimination result score matrix based on the fault judgment result, normalizing the fault mechanism discrimination result score matrix to generate a normalized mechanism discrimination matrix, includes: The fault mechanism discrimination result score matrix M s =[ M 1,..., M i ,..., M 9] T ,Ms i ∈[0,100] , si=1…9; Normalize the fault mechanism discrimination result score matrix to obtain the normalized mechanism discrimination matrix M s : Preferably, the fault data set includes electrical quantity characteristic data and a label corresponding to the fault type, and establishing the fault data set includes: The data window length of the preset time before the fault and the preset time after the fault is intercepted, the fault data is sampled based on the preset sampling rate, and the fault data set D=[(x1,y1),...(x i ,y i ),...,(x n ,y n )], where x i (i=1,2,3...n) is the electrical characteristic data of all channels of the i-th fault data recording, y i (i=1,2,3...n) is the label corresponding to the actual fault type of the i-th fault data recording, y i ∈[1,2,3,4,5,6,7,8,9].

[0010] Preferably, it also includes establishing an intelligent model: The intelligent model is an inception deep convolutional neural network model including k inception deep modules, global average pooling and a fully connected layer; Randomly initialize the fully connected layer parameters w fc ,b fc , randomly initialize all convolution kernels ck of the inception deep convolutional neural network model; Design the loss function of the inception deep convolutional neural network model: in, represents the probability value predicted by the model, represents the true category label of the sample, j For the j convolution kernels, c is the total number of convolution kernels, and z is the final classification label.

[0011] Preferably, the method further comprises training the intelligent model through the following steps: In forward propagation, the feature map is input, and multiple sets of feature maps are generated through convolution kernels and pooling operations; the generated multiple sets of feature maps are spliced ​​in the depth direction to generate the final output feature map; the output feature map is processed through a fully connected layer to output the final classification label z; Calculate the weight w of the fully connected layer fc and bias b fc The gradient is: in, L is the loss function; Calculate the gradient of all convolution kernels ck. When there is a convolution kernel whose output is c o When , the loss function is calculated for the output c o Convolution kernel weight w ck and bias b ck The gradient is: in, L is the loss function, w is the weight, and b is the bias; Back propagation, update the convolution kernel and fully connected layer parameters according to the gradient: in, is the learning rate, w old is the weight before iteration, w new is the weight after iteration, b old is the bias before iteration, b new is the post-iteration bias; The step of training the intelligent model is repeated until the loss function converges, and the training of the intelligent model is completed.

[0012] Preferably, the fusing the normalized mechanism discriminant matrix and the discriminant probability distribution matrix to generate a fusion matrix comprises: The discriminant probability distribution matrix m output by the intelligent model AI =[ m AI1 ,..., m AIi ,..., m AI9 ] T , m AIi ∈[0,1], with the normalized mechanism discriminant matrix m s To perform the fusion: Among them, m merge is the fusion matrix.

[0013] Preferably, the method of establishing a final fault cause identification matrix by multiplying the fusion matrix and the meteorological correlation matrix comprises: m mergenew =Am merge Among them, A is the meteorological correlation matrix, m mergenew It is the final fault cause identification matrix.

[0014] According to another aspect of the present invention, the present invention provides a method for identifying the cause of a line fault based on an intelligent model, the method comprising: A mechanism judgment unit is used to obtain fault identification data of the fault line; assign values ​​to fault judgment items of each fault type based on the fault identification data, and obtain fault judgment results of each fault type based on preset weights and values ​​of fault judgment items of different types of faults; generate a fault mechanism discrimination result score matrix based on the fault judgment results, and normalize the fault mechanism discrimination result score matrix to generate a normalized mechanism discrimination matrix; A probability judgment unit is used to establish a fault data set, and based on the fault data set, output a judgment probability distribution matrix through a trained intelligent model; The result unit is used to fuse the normalized mechanism discriminant matrix and the discriminant probability distribution matrix to generate a fusion matrix; establish a meteorological correlation matrix based on the correlation between meteorological information and fault causes when the line fails; establish a final fault cause identification matrix by multiplying the fusion matrix and the meteorological correlation matrix, and identify the cause of the line fault based on the final fault cause identification matrix.

[0015] Preferably, the fault types include: lightning strike fault, wind deflection fault, low resistance foreign body fault, high resistance foreign body fault, wildfire fault, pollution flashover fault, ice dancing fault, ice flashover fault, tree flashover fault, bird damage fault and mechanical external force fault; The fault judgment items include: fault type item, fault nature item, coincident power item, non-periodic component item, harmonic content item, fault time item, season item, time item and load item.

[0016] Preferably, the mechanism judgment unit is used to generate a fault mechanism discrimination result score matrix based on the fault judgment result, normalize the fault mechanism discrimination result score matrix to generate a normalized mechanism discrimination matrix, and further to: The fault mechanism discrimination result score matrix M s =[ M 1,..., M i ,..., M 9] T ,Ms i ∈[0,100] , si=1…9; Normalize the fault mechanism discrimination result score matrix to obtain the normalized mechanism discrimination matrix M s :

[0017] Preferably, the fault data set includes electrical quantity characteristic data and labels corresponding to the fault types. The probability judgment unit is used to establish the fault data set and is also used to: The data window length of the preset time before the fault and the preset time after the fault is intercepted, the fault data is sampled based on the preset sampling rate, and the fault data set D=[(x1,y1),...(x i ,y i ),...,(x n ,y n )], where x i (i=1,2,3...n) is the electrical characteristic data of all channels of the i-th fault data recording, y i (i=1,2,3...n) is the label corresponding to the actual fault type of the i-th fault data recording, y i ∈[1,2,3,4,5,6,7,8,9].

[0018] Preferably, the probability judgment unit is also used to establish an intelligent model: The intelligent model is an inception deep convolutional neural network model including k inception deep modules, global average pooling and a fully connected layer; Randomly initialize the fully connected layer parameters w fc ,b fc , randomly initialize all convolution kernels ck of the inception deep convolutional neural network model; Design the loss function of the inception deep convolutional neural network model: in, represents the probability value predicted by the model, represents the true category label of the sample, j For the j convolution kernels, c is the total number of convolution kernels, and z is the final classification label. Preferably, the probability judgment unit is also used to train the intelligent model through the following steps: In forward propagation, the feature map is input, and multiple sets of feature maps are generated through convolution kernels and pooling operations; the generated multiple sets of feature maps are spliced ​​in the depth direction to generate the final output feature map; the output feature map is processed through a fully connected layer to output the final classification label z; Calculate the weight w of the fully connected layer fc and bias b fc The gradient is: in, L is the loss function; Calculate the gradient of all convolution kernels ck. When there is a convolution kernel whose output is c o When , the loss function is calculated for the output c o Convolution kernel weight w ck and bias b ck The gradient is: in, L is the loss function, w is the weight, and b is the bias; Back propagation, update the convolution kernel and fully connected layer parameters according to the gradient: in, is the learning rate, w old is the weight before iteration, w new is the weight after iteration, b old is the bias before iteration, b new is the post-iteration bias; The step of training the intelligent model is repeated until the loss function converges, and the training of the intelligent model is completed.

[0019] Preferably, the result unit is used to fuse the normalized mechanism discriminant matrix and the discriminant probability distribution matrix to generate a fusion matrix, and is also used to: The discriminant probability distribution matrix m output by the intelligent model AI =[ m AI1 ,..., m AIi ,..., m AI9 ] T , m AIi ∈[0,1], with the normalized mechanism discriminant matrix m s To perform the fusion: Among them, m merge is the fusion matrix.

[0020] Preferably, the result unit is used to establish a final fault cause identification matrix by multiplying the fusion matrix and the meteorological correlation matrix, and is also used to: m mergenew =Am merge Among them, A is the meteorological correlation matrix, m mergenew It is the final fault cause identification matrix.

[0021] The technical solution of the present invention provides a method and system for identifying the cause of a line fault based on an intelligent model, wherein the method includes: obtaining fault identification data of a faulty line; assigning values ​​to fault judgment items of each fault type based on the fault identification data, and obtaining fault judgment results of each fault type based on preset weights and values ​​of fault judgment items of different types of faults; generating a fault mechanism discrimination result score matrix based on the fault judgment results, normalizing the fault mechanism discrimination result score matrix, and generating a normalized mechanism discrimination matrix; establishing a fault data set, and based on the fault data set, outputting a discrimination probability distribution matrix through a trained intelligent model; fusing the normalized mechanism discrimination matrix and the discrimination probability distribution matrix to generate a fusion matrix; establishing a meteorological correlation matrix based on the correlation between meteorological information and the cause of the fault when the line fails; establishing a final fault cause identification matrix by multiplying the fusion matrix and the meteorological correlation matrix, and identifying the cause of the line fault based on the final fault cause identification matrix. The technical solution of the present invention provides a method and system for identifying the cause of line faults combined with artificial intelligence technology. It comprehensively utilizes artificial intelligence technology, fault mechanism analysis and meteorological-fault cause correlation knowledge, integrates the "data" of historical power grid faults and the "knowledge" of fault mechanism analysis, and collaboratively realizes rapid and accurate identification of the cause of transmission line faults, effectively ensuring power grid safety, power supply and clean energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings: Figure 1 A flow chart of a line fault cause identification method based on an intelligent model according to a preferred embodiment of the present invention; Figure 2 A flow chart of a method for identifying the cause of a line fault in combination with artificial intelligence technology according to a preferred embodiment of the present invention; Figure 3 Schematic diagram of a double-terminal PI type transmission line according to a preferred embodiment of the present invention; Figure 4 A schematic diagram of an inception convolutional neural network architecture suitable for fault cause classification according to a preferred embodiment of the present invention; Figure 5 A schematic diagram of an inception basic module according to a preferred embodiment of the present invention; and Figure 6 A structural diagram of a line fault cause identification system based on an intelligent model according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0023] Now, exemplary embodiments of the present invention are described with reference to the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided to disclose the present invention in detail and completely and to fully convey the scope of the present invention to those skilled in the art. The terms used in the exemplary embodiments shown in the accompanying drawings are not intended to limit the present invention. In the accompanying drawings, the same units / elements are marked with the same reference numerals.

[0024] Unless otherwise specified, the terms (including technical terms) used herein have the commonly understood meanings to those skilled in the art. In addition, it is understood that the terms defined in commonly used dictionaries should be understood to have the same meanings as those in the context of the relevant fields, and should not be understood as idealized or overly formal meanings.

[0025] Figure 1 The present invention is a flowchart of a method for identifying the cause of a line fault based on an intelligent model according to a preferred embodiment of the present invention.

[0026] Due to the shortcomings of the existing technology that the fault electrical characteristics are single and the identification accuracy is not high, the present invention proposes a line fault cause identification method combined with artificial intelligence technology. The present invention comprehensively utilizes fault mechanism analysis, artificial intelligence technology and meteorological-fault cause correlation knowledge, and integrates the "data" of historical power grid faults and the "knowledge" of fault mechanism analysis to jointly realize the rapid and accurate identification of the causes of transmission line faults, effectively ensuring the safety of the power grid, power supply and clean energy consumption.

[0027] like Figure 1 As shown, the present invention provides a line fault cause identification method based on an intelligent model, the method comprising: Step 101: Obtain fault identification data of a faulty line; The present invention initializes the data required for fault identification based on the on-site recording data, including the AC line voltage level, AC line length, AC line unit length resistance, reactance, inductance and capacitance, the distance between the fault point and the faults on both sides, the line voltage transformer ratio, the line current transformer ratio and other information. Figure 2 shown.

[0028] The present invention performs data processing based on the initialization reading, and converts the data collected by the voltage and current transformer into a primary value. The data conversion expression on the M side of the fault line is: (1) (2) (3) In the formula, I AM , I BM , I CM are the sampling values ​​of the current transformer in the AC line, I AMP1 , I BMP1 , I CMP1 are the primary values ​​of the current on the M side of the AC line, U AM , U BM , U CM are the sampling values ​​of the voltage transformer in the AC line, U AMP1 , U BMP1 , U CMP1 are the primary values ​​of the voltage on the M side of the AC line, I 03MP1 and U 03MP1 are the zero-sequence current and zero-sequence voltage at the fault point of the AC fault line, CT_M and PT_M are the current and voltage transformer ratios on the M side of the AC line, respectively.

[0029] The data reduction expression of the N side of the fault line is: (4) (5) (6) In the formula, I AN , I BN , I CN are the sampling values ​​of the current transformer in the AC line, I ANP1 , I BNP1 , I CNP1 are the primary values ​​of the current on the N side of the AC line, U AN , U BN , U CN are the sampling values ​​of the voltage transformer in the AC line, UANP1 , U BNP1 , U CNP1 are the primary values ​​of the voltage on the N side of the AC line, I 03NP1 and U 03NP1 are the zero-sequence current and zero-sequence voltage at the fault point of the AC fault line, CT_N and PT_N are the current and voltage transformer ratios on the N side of the AC line, respectively.

[0030] In addition, if there is a high-voltage reactor in the AC line, it is necessary to further calculate the AC line high-voltage reactor current and subtract the high-voltage reactor current from the above data reduction expression. The high-voltage reactor current calculation expression is: (7) (8) In the formula, I GKAMP1 , I GKBMP1 , I GKCMP1 They are the primary value of the high withstand current on the M side of the AC line, I GKAM , I GKBM , I GKCM They are respectively the transformation ratio of the high-resistance current transformer on the M side, I GKANP1 , I GKBNP1 , I GKCNP1 They are the primary value of the high withstand current on the N side of the AC line, I GKAN , I GKBN , I GKCN are the transformation ratios of the high-resistance current transformer on the N side respectively.

[0031] Based on the above data processing results, the present invention uses a backward Fourier algorithm to calculate the voltage, current and differential current phasor on both sides of the AC fault line. In the backward Fourier algorithm, 12 points are sampled in each power frequency cycle.

[0032] The present invention is based on the fault electrical quantities on both sides, and calculates the voltage, current and transition resistance at the fault point by solving differential equations, and calculates the effective value and standard deviation of the transition resistance at the fault stage by using the instantaneous transition resistance. Figure 3 The double-terminal PI transmission line shown in FIG. 1 is used as an example to illustrate the voltage and current expressions at the fault point.

[0033] (9) (10) (11) In the formula, L is the length of the AC transmission line, r , l , c They are the resistance, inductance and capacitance per unit length of the AC transmission line respectively. u f , i f and R f are the fault point voltage, current and transition resistance respectively, x is the distance between the fault point and the M side of the AC transmission line, u m , u n , i n and i n are the voltage and current on the M and N sides of the AC transmission line respectively.

[0034] Combine equations (9)-(11) and substitute the voltage and current phasors on both sides calculated in the above steps into the above expressions to obtain the voltage at the fault point: u f ( t ) and fault point current i f ( t ). The calculation expression of instantaneous transition resistance at the fault point is: (12) The average value and standard deviation of transition resistance are calculated based on instantaneous transition resistance. The calculation expression is: (13) (14) In the formula, R avg and R E are the mean value and standard deviation of transition resistance, N is the total number of sampling points in one power frequency cycle, R f ( i) is the instantaneous value of the transition resistance calculated for each sampling point.

[0035] The present invention calculates the harmonic components of the three-phase fault current respectively based on the Fourier algorithm, and calculates the harmonics of each phase and the total harmonics of the three phases.

[0036] The present invention calibrates the fault stage based on the voltage and current phasors calculated in the above steps.

[0037] (1) The phase current mutation detection method is used to detect the first fault moment. To avoid the influence of data disturbance, three consecutive points are judged. If three consecutive elements of the phase current mutation are greater than the threshold, it is judged that a fault has occurred. The phase current magnitude is used to detect the two tripping moments. If more than 95% of the elements are less than the threshold, it is judged that a trip has been achieved.

[0038] (2) The phase current mutation detection method is used to detect the start time of the second fault. The start time of the second fault is identified starting from 100ms after the first trip.

[0039] The present invention searches for the fault voltage peak moment closest to the fault occurrence moment and the voltage peak value corresponding to the moment based on the fault moment calculation result in the above steps.

[0040] The present invention extracts information such as fault phase, reclosing, fault time, etc. based on the on-site recording file, and extracts weather information at the time of fault occurrence in combination with the meteorological system.

[0041] Step 102: assigning values ​​to the fault judgment items of each fault type based on the fault identification data, and obtaining the fault judgment result of each fault type based on the preset weights and values ​​of the fault judgment items of different types of faults; Preferably, the fault types include: lightning strike fault, wind deflection fault, low-resistance foreign body fault, high-resistance foreign body fault, wildfire fault, pollution flashover fault, ice dancing fault, ice flashover fault, tree flashover fault, bird damage fault and mechanical external force fault; Fault judgment items include: fault type item, fault nature item, coincident power item, non-periodic component item, harmonic content item, fault time item, season item, time item and load item.

[0042] Based on the above calculation results, the present invention realizes fault identification through multi-factor weighted logic judgment, which is as follows: (1) Lightning fault logic judgment and weight setting According to the analysis of lightning fault mechanism, fault judgment is divided into 7 items: The first item is the fault type item. If the fault type is a single-phase grounding fault, the first item is assigned 100 points. If it is other faults, the first item is assigned 10 points. The second item is the fault nature item. If the fault nature is a metallic fault, the second item is assigned 100 points, and the other points are linearly reduced according to the transition resistance value, with the lowest score being 0 points. The third item is the reclosing success item. If the reclosing is successful, the third item is assigned 100 points. If the reclosing is unsuccessful, the third item is assigned 0 points. The fourth item is the non-periodic component item. The larger the non-periodic component, the higher the score. The highest score is 100 points and the lowest score is 0 points. The fifth item is the harmonic content item. The higher the harmonic content, the higher the score. The highest score is 100 points and the lowest score is 0 points. The sixth item is the fault time item. If the fault time is at the voltage peak, the sixth item is 100 points. The scores of other items decrease linearly as they are farther from the peak, and the lowest score is 0 points. Item 7 is the season item. If the fault occurs in spring or summer, Item 7 is assigned 90 points. If the fault occurs in autumn, Item 7 is assigned 70 points. If the fault occurs in winter, Item 7 is assigned 10 points.

[0043] The weights of the above 7 items are 0.15, 0.15, 0.1, 0.2, 0.2, 0.1, 0.1 respectively. The logical judgment score of lightning fault is the sum of the product of each weight and the score of the item.

[0044] (2) Logical judgment and weight setting of windage fault According to the analysis of windage fault mechanism, fault judgment is divided into 7 items: The first item is the fault type item. If the fault type is a single-phase grounding fault, the first item is assigned 90 points. If the fault type is a phase-to-phase grounding fault, the first item is assigned 70 points. If the fault type is a phase-to-phase fault, the first item is assigned 60 points. If it is other faults, the first item is assigned 10 points. The second item is the fault nature item. If the fault nature is a metallic fault, the second item is assigned 100 points, and the others are assigned 0 points. The third item is the reclosing success item. If the reclosing is successful, the third item is assigned 90 points. If the reclosing is unsuccessful, the third item is assigned 10 points. The fourth item is the non-periodic component item. The smaller the non-periodic component, the higher the score. The highest score is 100 points and the lowest score is 0 points. The fifth item is the harmonic content item. The smaller the harmonic content, the higher the score. The highest score is 100 points and the lowest score is 0 points. The sixth item is the fault time item. If the fault time is at the voltage peak, the sixth item is 100 points. The scores of other items decrease linearly as they are farther from the peak, and the lowest score is 0 points. Item 7 is the season item. If the fault occurs in spring or winter, Item 7 is assigned 90 points. If the fault occurs in summer, Item 7 is assigned 70 points. If the fault occurs in autumn, Item 7 is assigned 30 points.

[0045] The weights of the above 7 items are 0.2, 0.2, 0.15, 0.05, 0.05, 0.2, and 0.15 respectively. The logical judgment score of wind deviation fault is the sum of the product of each weight and the score of the item.

[0046] (3) Low-resistance foreign object fault logic judgment and weight setting According to the low-resistance foreign body fault mechanism analysis, fault judgment is divided into five items: The first item is the fault type item. If the fault type is a phase-to-phase fault, the first item is assigned 70 points. If the fault type is a ground fault, the first item is assigned 50 points. If it is other faults, the first item is assigned 10 points. The second item is the fault nature item. If the fault nature is a metallic fault, the second item is assigned 100 points, and the other points are linearly reduced according to the transition resistance value, with the lowest score being 0 points. The third item is the reclosing success item. If the reclosing is successful, the third item is assigned 80 points. If the reclosing is unsuccessful, the third item is assigned 20 points. The fourth item is the harmonic content item. The smaller the harmonic content, the higher the score. The highest score is 100 points and the lowest score is 0 points. The fifth item is the fault time item. If the fault time is at the voltage peak, the sixth item is 80 points, and the others are assigned 50 points.

[0047] The weights of the above five items are 0.1, 0.3, 0.1, 0.3, and 0.2 respectively. The logical judgment score of low-resistance foreign body fault is the sum of the product of each weight and the score of the item.

[0048] (4) High-resistance foreign body fault logic judgment and weight setting According to the low-resistance foreign body fault mechanism analysis, fault judgment is divided into five items: The first item is the fault type item. If the fault type is a phase-to-phase fault, the first item is assigned 10 points. If the fault type is a ground fault, the first item is assigned 100 points. If it is other faults, the first item is assigned 10 points. The second item is the fault nature item. If the fault nature is a metallic fault, the second item is assigned 0 points. The other points are linearly increased according to the transition resistance value, with the highest score being 100 points. The third item is the reclosing success item. If the reclosing is successful, the third item is assigned 80 points. If the reclosing is unsuccessful, the third item is assigned 20 points. The fourth item is the harmonic content item. The greater the harmonic content, the higher the score. The highest score is 100 points and the lowest score is 0 points. Item 5 is the fault time item. If the fault time is at the voltage peak, item 6 is 60 points, and the others are assigned 40 points.

[0049] The weights of the above five items are 0.1, 0.3, 0.1, 0.3, and 0.2 respectively. The logical judgment score of low-resistance foreign body fault is the sum of the product of each weight and the score of the item.

[0050] (5) Wildfire fault logic judgment and weight setting According to the analysis of wildfire failure mechanism, fault judgment is divided into 8 items: The first item is the fault type item. If the fault type is a single-phase grounding fault, the first item is assigned 90 points. If the fault type is a phase-to-phase grounding fault, the first item is assigned 70 points. If the fault type is a phase-to-phase fault, the first item is assigned 50 points. If it is other faults, the first item is assigned 10 points. The second item is the fault nature item. If the fault nature is a metallic fault, the second item is assigned 0 points. The other points are linearly increased according to the transition resistance value, with the highest score being 100 points. The third item is the reclosing success item. If the reclosing is successful, the third item is assigned 0 points. If the reclosing is unsuccessful, the third item is assigned 100 points. The fourth item is the non-periodic component item. The smaller the non-periodic component, the higher the score. The highest score is 100 points and the lowest score is 0 points. The fifth item is the harmonic content item. The greater the harmonic content, the higher the score. The highest score is 100 points and the lowest score is 0 points. The sixth item is the fault time item. If the fault time is at the voltage peak, the sixth item is 100 points. The scores of other items decrease linearly as they are farther from the peak, and the lowest score is 0 points. Item 7 is the season item. If the fault occurs in spring or summer, Item 7 is assigned 90 points. If the fault occurs in autumn, Item 7 is assigned 60 points. If the fault occurs in winter, Item 7 is assigned 10 points. If the fault occurs around Qingming Festival, Item 7 is assigned 100 points. Item 8 is the time item. If the fault occurs in the morning, Item 8 is assigned 70 points. If the fault occurs in the morning, Item 8 is assigned 60 points. If the fault occurs in the afternoon, Item 8 is assigned 30 points. If the fault occurs in the evening, Item 8 is assigned 30 points.

[0051] The weights of the above 8 items are 0.1, 0.2, 0.2, 0.05, 0.2, 0.1, 0.1, and 0.05 respectively. The score of forest fire fault logic judgment is the sum of the product of each weight and the score of the item.

[0052] (6) Logic judgment and weight setting of pollution flashover fault According to the analysis of the pollution flashover fault mechanism, fault judgment is divided into 7 items: The first item is the fault type item. If the fault type is a single-phase grounding fault, the first item is assigned 100 points. If it is other faults, the first item is assigned 0 points. The second item is the fault nature item. If the fault nature is a metallic fault, the second item is assigned 10 points, and the others are assigned 90 points. The third item is the reclosing success item. If the reclosing is successful, the third item is assigned 20 points. If the reclosing is unsuccessful, the third item is assigned 90 points. The fourth item is the non-periodic component item. The smaller the non-periodic component, the higher the score. The highest score is 100 points and the lowest score is 0 points. The fifth item is the harmonic content item. The greater the harmonic content, the higher the score. The highest score is 100 points and the lowest score is 0 points. The sixth item is the fault time item. If the fault time is at the voltage peak, the sixth item is 100 points. The scores of other items decrease linearly as they are farther from the peak, and the lowest score is 0 points. Item 7 is the season item. If the fault occurs in spring, item 7 is assigned 80 points. If the fault occurs in summer, item 7 is assigned 30 points. If the fault occurs in autumn, item 7 is assigned 20 points. If the fault occurs in winter, item 7 is assigned 90 points. Item 8 is the time item. If the fault occurs in the morning, Item 8 is assigned 70 points. If the fault occurs in the morning, Item 8 is assigned 30 points. If the fault occurs in the afternoon, Item 8 is assigned 30 points. If the fault occurs in the evening, Item 8 is assigned 70 points.

[0053] The weights of the above 8 items are 0.15, 0.15, 0.2, 0.1, 0.1, 0.1, 0.1, 0.1, and 0.1 respectively. The logical judgment score of the pollution flashover fault is the sum of the product of each weight and the score of the item.

[0054] (6) Logical judgment and weight setting of ice dancing fault According to the analysis of ice dancing fault mechanism, fault judgment is divided into 7 items: The first item is the fault type item. If the fault type is a single-phase ground fault, the first item is assigned 70 points. If the fault type is a phase-to-phase fault, the first item is assigned 90 points. If it is other faults, the first item is assigned 10 points. The second item is the fault nature item. If the fault nature is a metallic fault, the second item is assigned 100 points, and the other points are linearly reduced according to the transition resistance value, with the lowest score being 0 points. The third item is the reclosing success item. If the reclosing is successful, the third item is assigned 0 points. If the reclosing is unsuccessful, the third item is assigned 100 points. The fourth item is the non-periodic component item. The smaller the non-periodic component, the higher the score. The highest score is 100 points and the lowest score is 0 points. The fifth item is the harmonic content item. The smaller the harmonic content, the higher the score. The highest score is 100 points and the lowest score is 0 points. The sixth item is the fault time item. If the fault time is at the voltage peak, the sixth item is 100 points. The scores of other items decrease linearly as they are farther from the peak, and the lowest score is 0 points. Item 7 is the season item. If the fault occurs in winter, item 7 is assigned 100 points, and the others are assigned 90 points. The weights of the above 7 items are 0.15, 0.15, 0.2, 0.05, 0.05, 0.1, and 0.3 respectively. The logical judgment score of ice dancing fault is the sum of the product of each weight and the score of the item.

[0055] (7) Icing flashover fault logic judgment and weight setting According to the analysis of ice flashover fault mechanism, fault judgment is divided into 7 items: The first item is the fault type item. If the fault type is a single-phase grounding fault, the first item is assigned 90 points. If it is other faults, the first item is assigned 30 points. The second item is the fault nature item. If the fault nature is a metallic fault, the second item is assigned 90 points, and the others are assigned 50 points. The third item is the reclosing success item. If the reclosing is successful, the third item is assigned 100 points. If the reclosing is unsuccessful, the third item is assigned 0 points. The fourth item is the non-periodic component item. The smaller the non-periodic component, the higher the score. The highest score is 100 points and the lowest score is 0 points. The fifth item is the harmonic content item. The smaller the harmonic content, the higher the score. The highest score is 100 points and the lowest score is 0 points. The sixth item is the fault time item. If the fault time is at the voltage peak, the sixth item is 100 points. The scores of other items decrease linearly as they are farther from the peak, and the lowest score is 0 points. Item 7 is the season item. If the fault occurs in spring, Item 7 is assigned 70 points. If the fault occurs in winter, Item 7 is assigned 100 points. The rest are assigned 10 points.

[0056] The weights of the above 7 items are 0.1, 0.1, 0.1, 0.05, 0.1, 0.15, and 0.4 respectively. The logical judgment score of ice flashover fault is the sum of the product of each weight and the score of the item.

[0057] (8) Logical judgment and weight setting of tree flash fault According to the analysis of tree flash failure mechanism, fault judgment is divided into 8 items: The first item is the fault type item. If the fault type is a single-phase grounding fault, the first item is assigned 100 points. If it is other faults, the first item is assigned 0 points. The second item is the fault nature item. If the fault nature is a metallic fault, the second item is assigned 0 points. The other points are linearly increased according to the transition resistance value, with the highest score being 100 points. The third item is the reclosing success item. If the reclosing is successful, the third item is assigned 0 points. If the reclosing is unsuccessful, the third item is assigned 100 points. The fourth item is the non-periodic component item. The smaller the non-periodic component, the higher the score. The highest score is 100 points and the lowest score is 0 points. The fifth item is the harmonic content item. The greater the harmonic content, the higher the score. The highest score is 100 points and the lowest score is 0 points. The sixth item is the fault time item. If the fault time is at the voltage peak, the sixth item is 100 points. The scores of other items decrease linearly as they are farther from the peak, and the lowest score is 0 points. Item 7 is the season item. If the fault occurs in spring, item 7 is assigned 70 points; if the fault occurs in summer, item 7 is assigned 90 points; if the fault occurs in autumn, item 7 is assigned 30 points; if the fault occurs in winter, item 7 is assigned 10 points; Item 8 is the load item. If the fault occurs in the morning, Item 8 is assigned 70 points. If the fault occurs in the morning, Item 8 is assigned 30 points. The greater the load at the time of the fault, the higher the score. The highest score is 100 points and the lowest score is 0 points.

[0058] The weights of the above 8 items are 0.15, 0.15, 0.2, 0.1, 0.1, 0.1, 0.1, 0.1, and 0.1 respectively. The logical judgment score of tree flash fault is the sum of the product of each weight and the score of the item.

[0059] (9) Logical judgment and weight setting of bird damage fault According to the analysis of bird damage failure mechanism, fault judgment is divided into 8 items: The first item is the fault type item. If the fault type is a single-phase grounding fault, the first item is assigned 80 points. If the fault type is a phase-to-phase fault, the first item is assigned 50 points. If it is any other fault, the first item is assigned 0 points. The second item is the fault nature item. If the fault nature is a metallic fault, the second item is assigned 0 points. The other points are linearly increased according to the transition resistance value, with the highest score being 100 points. The third item is the reclosing success item. If the reclosing is successful, the third item is assigned 100 points. If the reclosing is unsuccessful, the third item is assigned 0 points. The fourth item is the non-periodic component item. The larger the non-periodic component, the higher the score. The highest score is 100 points and the lowest score is 0 points. The fifth item is the harmonic content item. The greater the harmonic content, the higher the score. The highest score is 100 points and the lowest score is 0 points. The sixth item is the fault time item. If the fault time is at the voltage peak, the sixth item is 100 points. The scores of other items decrease linearly as they are farther from the peak, and the lowest score is 0 points. Item 7 is the season item. If the fault occurs in spring, item 7 is assigned 90 points. If the fault occurs in summer, item 7 is assigned 60 points. If the fault occurs in autumn, item 7 is assigned 70 points. If the fault occurs in winter, item 7 is assigned 10 points. Item 8 is the time item. If the fault occurs at night, Item 8 is assigned 90 points, and other times are assigned 20 points.

[0060] The weights of the above 8 items are 0.15, 0.15, 0.2, 0.1, 0.1, 0.1, 0.1, 0.1, and 0.1 respectively. The logical judgment score of bird damage fault is the sum of the product of each weight and the score of the item.

[0061] (10) Logical judgment and weight setting of mechanical external force fault According to the analysis of mechanical external force failure mechanism, fault judgment is divided into 9 items: The first item is the fault type item. If the fault type is a single-phase grounding fault, the first item is assigned 100 points. If the fault type is a phase-to-phase fault, the first item is assigned 50 points. If it is any other fault, the first item is assigned 0 points. The second item is the fault nature item. If the fault nature is a metallic fault, the second item is assigned 100 points, and the other points are linearly reduced according to the transition resistance value, with the lowest score being 0 points. The third item is the reclosing success item. If the reclosing is successful, the third item is assigned 30 points. If the reclosing is unsuccessful, the third item is assigned 100 points. The fourth item is the transition resistance change item. If the transition resistance standard deviation is greater than or equal to 10, the fourth item is assigned 0 points. If the transition resistance standard deviation is less than 10, the fourth item is assigned 100 points. The fifth item is the non-periodic component item. The larger the non-periodic component, the higher the score. The highest score is 100 points and the lowest score is 0 points. The sixth item is the harmonic content item. The greater the harmonic content, the higher the score. The highest score is 100 points and the lowest score is 0 points. The seventh item is the fault time item. If the fault time is at the voltage peak, the seventh item is 100 points. The scores of other items decrease linearly as they are farther from the peak, and the lowest score is 0 points. Item 8 is the seasonal item. If the fault occurs in spring, Item 8 is assigned 80 points. If the fault occurs in summer, Item 8 is assigned 60 points. If the fault occurs in autumn and winter, Item 8 is assigned 10 points. Item 9 is the time item. If the fault occurs during the day, item 9 is assigned 90 points, and 0 points are assigned at other times.

[0062] The weights of the above 9 items are 0.2, 0.2, 0.2, 0.1, 0.05, 0.05, 0.05, 0.1, 0.05 respectively. The logical judgment score of mechanical external force failure is the sum of the product of each weight and the score of the item.

[0063] Step 103: generating a fault mechanism discrimination result score matrix based on the fault judgment result, normalizing the fault mechanism discrimination result score matrix, and generating a normalized mechanism discrimination matrix; Preferably, a fault mechanism discrimination result score matrix is ​​generated based on the fault judgment result, and the fault mechanism discrimination result score matrix is ​​normalized to generate a normalized mechanism discrimination matrix, including: Fault mechanism identification result score matrix M s =[ M 1,..., M i ,..., M 9] T ,Ms i ∈[0,100] , si=1…9; Normalize the fault mechanism discrimination result score matrix to obtain the normalized mechanism discrimination matrix M s :

[0064] Based on the above logic judgment results, the present invention sorts the scores of all fault types. The fault type with the highest score is the line fault cause identification result based on the fault mechanism and fault electrical characteristics. Suppose the fault mechanism judgment result score matrix M s =[ M 1,..., M i ,..., M 9] T ,m si ∈[0,100], normalize the matrix to get the normalized mechanism discriminant matrix M s : (15) Step 104: Establish a fault data set, and output a discriminant probability distribution matrix through the trained intelligent model based on the fault data set; Preferably, the fault data set includes electrical quantity characteristic data and a label corresponding to the fault type, and establishing the fault data set includes: The data window length of the preset time before the fault and the preset time after the fault is intercepted, the fault data is sampled based on the preset sampling rate, and the fault data set D=[(x1,y1),...(x i ,y i ),...,(xn ,y n )], where x i (i=1,2,3...n) is the electrical characteristic data of all channels of the i-th fault data recording, y i (i=1,2,3...n) is the label corresponding to the actual fault type of the i-th fault data recording, y i ∈[1,2,3,4,5,6,7,8,9].

[0065] Preferably, it also includes establishing an intelligent model: The intelligent model is an inception deep convolutional neural network model that includes k inception deep modules, global average pooling, and a fully connected layer; Randomly initialize the fully connected layer parameters w fc ,b fc , randomly initialize all convolution kernels ck of the inception deep convolutional neural network model; Design the loss function of the inception deep convolutional neural network model: in, represents the probability value predicted by the model, represents the true category label of the sample, j For the j convolution kernels, c is the total number of convolution kernels, and z is the final classification label.

[0066] Preferably, the method further comprises training the intelligent model through the following steps: In the forward propagation, the feature map is input, and multiple sets of feature maps are generated through convolution kernels and pooling operations; the generated multiple sets of feature maps are spliced ​​in the depth direction to generate the final output feature map; the output feature map is processed through the fully connected layer to output the final classification label z; Calculate the weight w of the fully connected layer fc and bias b fc The gradient is: in, L is the loss function; Calculate the gradient of all convolution kernels ck. When there is a convolution kernel whose output is c o When , the loss function is calculated for the output c o Convolution kernel weight w ck and bias b ck The gradient is: in,L is the loss function, w is the weight, and b is the bias; Back propagation, update the convolution kernel and fully connected layer parameters according to the gradient: in, is the learning rate, w old is the weight before iteration, w new is the weight after iteration, b old is the bias before iteration, b new is the post-iteration bias; The steps of training the intelligent model are repeated until the loss function converges, and the intelligent model is trained.

[0067] The present invention collects transmission line fault recording files (set to n) containing fault causes in the entire network, uses all electrical characteristic quantities and data calculated in the above steps, and forms a fusion fault recording file of m electrical characteristic quantity channels with the original voltage and current raw data, intercepts the data window length of 200ms before the fault and 1280ms after the fault, fixes the sampling rate of all recording files to 5kHz by interpolation sampling, forms a fault data set with the fault cause corresponding to each fault recording file, and forms a fault data set D=[(x1,y1),...(x i ,y i ),...,(x n ,y n )], where x i (i=1,2,3...n) is the m*6500-dimensional electrical quantity characteristic data of all channels of the i-th fault recording, y i (i=1,2,3...n) is the label corresponding to the actual fault cause of the i-th fault recording, y i ∈[1,2,3,4,5,6,7,8,9], corresponding to lightning strike, wind deflection, low-resistance foreign body, wildfire, pollution flashover, ice dancing, ice flashover, bird damage and mechanical force. 10% of the data set in data set D is extracted as the test set, and 90% is used as the training set.

[0068] The present invention designs an inception convolutional neural network model including k inception modules, as shown in the attached figure. Figure 4 Each inception module is shown in the attached Figure 5 As shown in Figure 1, multiple convolution kernels of different sizes are used in parallel in the same layer to process the input feature map.

[0069] The present invention randomly initializes the fully connected layer parameters w fc ,b fc , randomly initialize all convolution kernels ck.

[0070] The present invention designs the loss function: (16) In the formula, represents the probability value predicted by the model, Represents the true category label of the sample.

[0071] In the forward propagation of the present invention, the input feature map is passed through multiple convolution kernels of different sizes and pooling operations to generate multiple sets of feature maps, which are then spliced ​​in the depth direction to form the final output feature map. These feature maps are then processed through a fully connected layer to output the final classification result z.

[0072] The present invention calculates the gradient of the fully connected layer, and the output of the fully connected layer is the final classification label z, so the loss function fully connected layer weight w can be calculated fc and bias b fc The gradient of is: (17) The present invention calculates the gradient of all convolution kernels ck. Assuming that the output of a convolution kernel is co, the loss function can be calculated for the convolution kernel weight w ck and bias b ck The gradient of is: (18) The present invention back propagates and updates the convolution kernel and fully connected layer parameters according to the gradient: (19) Repeat the above steps and perform multiple forward propagation and back propagation until the loss function converges and the model training is completed.

[0073] Step 105: fusing the normalized mechanism discriminant matrix and the discriminant probability distribution matrix to generate a fusion matrix; Preferably, the normalized mechanism discriminant matrix and the discriminant probability distribution matrix are fused to generate a fusion matrix, including: The discriminant probability distribution matrix m output by the intelligent model AI =[ m AI1 ,..., m AIi ,..., m AI9 ] T , m AIi ∈[0,1], and the normalized mechanism discriminant matrix m s To perform the fusion: Among them, m merge is the fusion matrix. The present invention assumes that the discriminant probability distribution m output by the artificial intelligence model is AI =[ m AI1 ,..., m AIi ,..., m AI9 ] T , m AIi ∈[0,1], which is the normalized mechanism discriminant matrix of the above step m s Perform fusion processing: (20) Step 106: Establishing a meteorological correlation matrix based on the correlation between meteorological information and fault causes when the line fails; Step 107: Establish a final fault cause identification matrix by multiplying the fusion matrix and the meteorological correlation matrix, and identify the cause of the line fault based on the final fault cause identification matrix.

[0074] Preferably, the final fault cause identification matrix is ​​established by multiplying the fusion matrix and the meteorological correlation matrix, including: m mergenew =Am merge Among them, A is the meteorological correlation matrix, m mergenew It is the final fault cause identification matrix.

[0075] The present invention integrates the correlation between meteorological information and fault causes when a line fault occurs, sets a meteorological correlation matrix A, as shown in Appendix 1, and probabilistically reorganizes the mechanism fusion results and the artificial intelligence discrimination results to exclude impossible fault causes under specific meteorological conditions, and further improves the recognition accuracy through human knowledge and experience of meteorological-fault causes.

[0076] (twenty one) Appendix 1 Meteorological correlation matrix Aiming at the shortcomings of the existing AC fault identification technology, which uses a single fault electrical characteristic and has low identification accuracy, the present invention proposes a line fault cause identification method and system combined with artificial intelligence technology, which comprehensively utilizes artificial intelligence technology, fault mechanism analysis and meteorological-fault cause correlation knowledge, integrates the "data" of historical power grid faults and the "knowledge" of fault mechanism analysis, and collaboratively realizes the rapid and accurate identification of the cause of transmission line faults, effectively ensuring the safety of the power grid, power supply and clean energy consumption. It can be promoted and applied in 220kV, 500kV, and 100kV AC lines.

[0077] Figure 6 A structural diagram of a line fault cause identification system based on an intelligent model according to a preferred embodiment of the present invention.

[0078] Figure 6 A structural diagram of a line fault cause identification system based on an intelligent model according to a preferred embodiment of the present invention.

[0079] like Figure 6 As shown, the present invention provides a line fault cause identification method based on an intelligent model, the method comprising: The mechanism judgment unit 601 is used to obtain fault identification data of the fault line; assign values ​​to fault judgment items of each fault type based on the fault identification data, and obtain fault judgment results of each fault type based on the preset weights and values ​​of the fault judgment items of different types of faults; generate a fault mechanism discrimination result score matrix based on the fault judgment results, and normalize the fault mechanism discrimination result score matrix to generate a normalized mechanism discrimination matrix; Preferably, the fault types include: lightning strike fault, wind deflection fault, low-resistance foreign body fault, high-resistance foreign body fault, wildfire fault, pollution flashover fault, ice dancing fault, ice flashover fault, tree flashover fault, bird damage fault and mechanical external force fault; Fault judgment items include: fault type item, fault nature item, coincident power item, non-periodic component item, harmonic content item, fault time item, season item, time item and load item.

[0080] Preferably, the mechanism judgment unit 601 is used to generate a fault mechanism discrimination result score matrix based on the fault judgment result, normalize the fault mechanism discrimination result score matrix to generate a normalized mechanism discrimination matrix, and further to: Fault mechanism identification result score matrix M s =[ M 1,..., M i ,..., M 9] T ,Ms i∈[0,100] , si=1…9; Normalize the fault mechanism discrimination result score matrix to obtain the normalized mechanism discrimination matrix M s :

[0081] The probability judgment unit 602 is used to establish a fault data set, and output a judgment probability distribution matrix based on the fault data set through a trained intelligent model; Preferably, the fault data set includes electrical quantity characteristic data and labels corresponding to the fault types. The probability judgment unit is used to establish the fault data set and is also used to: The data window length of the preset time before the fault and the preset time after the fault is intercepted, the fault data is sampled based on the preset sampling rate, and the fault data set D=[(x1,y1),...(x i ,y i ),...,(x n ,y n )], where x i (i=1,2,3...n) is the electrical characteristic data of all channels of the i-th fault data recording, y i (i=1,2,3...n) is the label corresponding to the actual fault type of the i-th fault data recording, y i ∈[1,2,3,4,5,6,7,8,9].

[0082] Preferably, the probability judgment unit 602 is also used to establish an intelligent model: The intelligent model is an inception deep convolutional neural network model that includes k inception deep modules, global average pooling, and a fully connected layer; Randomly initialize the fully connected layer parameters w fc ,b fc , randomly initialize all convolution kernels ck of the inception deep convolutional neural network model; Design the loss function of the inception deep convolutional neural network model: in, represents the probability value predicted by the model, represents the true category label of the sample, j For the j convolution kernels, c is the total number of convolution kernels, and z is the final classification label.

[0083] Preferably, the probability judgment unit 602 is further used to train the intelligent model through the following steps: In the forward propagation, the feature map is input, and multiple sets of feature maps are generated through convolution kernels and pooling operations; the generated multiple sets of feature maps are spliced ​​in the depth direction to generate the final output feature map; the output feature map is processed through the fully connected layer to output the final classification label z; Calculate the weight w of the fully connected layer fc and bias b fc The gradient is: in, L is the loss function; Calculate the gradient of all convolution kernels ck. When there is a convolution kernel whose output is c o When , the loss function is calculated for the output c o Convolution kernel weight w ck and bias b ck The gradient is: in, L is the loss function, w is the weight, and b is the bias; Back propagation, update the convolution kernel and fully connected layer parameters according to the gradient: in, is the learning rate, w old is the weight before iteration, w new is the weight after iteration, b old is the bias before iteration, b new is the post-iteration bias; The steps of training the intelligent model are repeated until the loss function converges, and the intelligent model is trained.

[0084] The result unit 603 is used to fuse the normalized mechanism discriminant matrix and the discriminant probability distribution matrix to generate a fusion matrix; establish a meteorological correlation matrix based on the correlation between meteorological information and the cause of the fault when the line fails; establish a final fault cause identification matrix by multiplying the fusion matrix and the meteorological correlation matrix, and identify the cause of the line fault based on the final fault cause identification matrix.

[0085] Preferably, the result unit 603 is used to fuse the normalized mechanism discriminant matrix and the discriminant probability distribution matrix to generate a fusion matrix, and is also used to: The discriminant probability distribution matrix m output by the intelligent model AI =[ m AI1 ,..., m AIi ,..., mAI9 ] T , m AIi ∈[0,1], and the normalized mechanism discriminant matrix m s To perform the fusion: Among them, m merge is the fusion matrix.

[0086] Preferably, the result unit 603 is used to establish a final fault cause identification matrix by multiplying the fusion matrix and the meteorological correlation matrix, and is also used to: m mergenew =Am merge Among them, A is the meteorological correlation matrix, m mergenew It is the final fault cause identification matrix.

[0087] A line fault cause identification system based on an intelligent model in a preferred embodiment of the present invention corresponds to a line fault cause identification method based on an intelligent model in another preferred embodiment of the present invention, and will not be described in detail here.

[0088] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The solutions in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and interpreted scripting language JavaScript, etc.

[0089] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0090] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0091] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0092] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0093] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

[0094] The invention has been described above with reference to a few embodiments. However, it is readily apparent to a person skilled in the art that other embodiments than the ones disclosed above are equally within the scope of the invention, as defined by the appended patent claims.

[0095] Generally, all terms used in the claims are to be interpreted according to their ordinary meaning in the technical field, unless explicitly defined otherwise herein. All references to "a / said / the [means, components, etc.]" are to be openly interpreted as at least one instance of the means, components, etc., unless explicitly stated otherwise. The steps of any method disclosed herein do not necessarily have to be performed in the exact order disclosed, unless explicitly stated otherwise.

Claims

1. A method for identifying the cause of a line fault based on an intelligent model, the method comprising: Obtaining fault identification data of the faulty line; Assigning values ​​to the fault judgment items of each fault type based on the fault identification data, and obtaining the fault judgment result of each fault type based on the preset weights and values ​​of the fault judgment items of different types of faults; Generate a fault mechanism discrimination result score matrix based on the fault judgment result, and perform normalization processing on the fault mechanism discrimination result score matrix to generate a normalized mechanism discrimination matrix; Establishing a fault data set, and based on the fault data set, outputting a discriminant probability distribution matrix through a trained intelligent model; Performing fusion processing on the normalized mechanism discriminant matrix and the discriminant probability distribution matrix to generate a fusion matrix; Based on the correlation between meteorological information and fault causes during line faults, a meteorological correlation matrix is ​​established; A final fault cause identification matrix is ​​established by multiplying the fusion matrix and the meteorological correlation matrix, and the cause of the line fault is identified based on the final fault cause identification matrix.

2. The method according to claim 1, wherein the fault type comprises: Lightning fault, wind deviation fault, low resistance foreign body fault, high resistance foreign body fault, mountain fire fault, pollution flashover fault, ice dancing fault, ice flashover fault, tree flashover fault, bird damage fault and mechanical external force fault; The fault judgment items include: fault type item, fault nature item, coincident power item, non-periodic component item, harmonic content item, fault time item, season item, time item and load item.

3. The method according to claim 1, wherein generating a fault mechanism discrimination result score matrix based on the fault judgment result, normalizing the fault mechanism discrimination result score matrix to generate a normalized mechanism discrimination matrix, comprises: The fault mechanism discrimination result score matrix M s =[ M 1,..., M i ,..., M 9] T ,Ms i ∈[0,100] , si=1…9; Normalize the fault mechanism discrimination result score matrix to obtain the normalized mechanism discrimination matrix M s : 。 4. The method according to claim 1, wherein the fault data set comprises electrical quantity characteristic data and a label corresponding to the fault type; establishing the fault data set comprises: The data window length of the preset time before the fault and the preset time after the fault is intercepted, the fault data is sampled based on the preset sampling rate, and the fault data set D=[(x1,y1),...(x i ,y i ),...,(x n ,y n )], where x i (i=1,2,3...n) is the electrical characteristic data of all channels of the i-th fault data recording, y i (i=1,2,3...n) is the label corresponding to the actual fault type of the i-th fault data recording, y i ∈[1,2,3,4,5,6,7,8,9].

5. The method according to claim 1, further comprising establishing an intelligent model: The intelligent model is an inception deep convolutional neural network model including k inception deep modules, global average pooling and a fully connected layer; Randomly initialize the fully connected layer parameters w fc ,b fc , randomly initialize all convolution kernels ck of the inception deep convolutional neural network model; Design the loss function of the inception deep convolutional neural network model: in, represents the probability value predicted by the model, represents the true category label of the sample, j For the j convolution kernels, c is the total number of convolution kernels, and z is the final classification label.

6. The method according to claim 5 further comprises training the intelligent model by: In forward propagation, the feature map is input, and multiple sets of feature maps are generated through convolution kernels and pooling operations; the generated multiple sets of feature maps are spliced ​​in the depth direction to generate the final output feature map; Process the output feature map through a fully connected layer and output the final classification label z; Calculate the weight w of the fully connected layer fc and bias b fc The gradient is: in, L is the loss function; Calculate the gradient of all convolution kernels ck. When there is a convolution kernel whose output is c o When , the loss function is calculated for the output c o Convolution kernel weight w ck and bias b ck The gradient is: in, L is the loss function, w is the weight, and b is the bias; Back propagation, update the convolution kernel and fully connected layer parameters according to the gradient: in, is the learning rate, w old is the weight before iteration, w new is the weight after iteration, b old is the bias before iteration, b new is the post-iteration bias; The step of training the intelligent model is repeated until the loss function converges, and the training of the intelligent model is completed.

7. According to the method of claim 1, the fusing of the normalized mechanism discriminant matrix and the discriminant probability distribution matrix to generate a fusion matrix comprises: The discriminant probability distribution matrix m output by the intelligent model AI =[ m AI1 ,..., m AIi ,..., m AI9 ] T , m AIi ∈[0,1], with the normalized mechanism discriminant matrix m s To perform the fusion: Among them, m merge is the fusion matrix.

8. The method according to claim 7, wherein the step of establishing a final fault cause identification matrix by multiplying the fusion matrix and the meteorological correlation matrix comprises: m mergenew =Am merge Among them, A is the meteorological correlation matrix, m mergenew It is the final fault cause identification matrix.

9. A line fault cause identification system based on an intelligent model, the system comprising: A mechanism judgment unit, used to obtain fault identification data of the faulty line; Assigning values ​​to the fault judgment items of each fault type based on the fault identification data, and obtaining the fault judgment result of each fault type based on the preset weights and values ​​of the fault judgment items of different types of faults; Generate a fault mechanism discrimination result score matrix based on the fault judgment result, and perform normalization processing on the fault mechanism discrimination result score matrix to generate a normalized mechanism discrimination matrix; A probability judgment unit is used to establish a fault data set, and based on the fault data set, output a judgment probability distribution matrix through a trained intelligent model; The result unit is used to fuse the normalized mechanism discriminant matrix and the discriminant probability distribution matrix to generate a fusion matrix; establish a meteorological correlation matrix based on the correlation between meteorological information and fault causes when the line fails; establish a final fault cause identification matrix by multiplying the fusion matrix and the meteorological correlation matrix, and identify the cause of the line fault based on the final fault cause identification matrix.

10. The system according to claim 9, wherein the fault types include: Lightning fault, wind deviation fault, low resistance foreign body fault, high resistance foreign body fault, mountain fire fault, pollution flashover fault, ice dancing fault, ice flashover fault, tree flashover fault, bird damage fault and mechanical external force fault; The fault judgment items include: fault type item, fault nature item, coincident power item, non-periodic component item, harmonic content item, fault time item, season item, time item and load item.

11. The system according to claim 9, wherein the mechanism judgment unit is used to generate a fault mechanism discrimination result score matrix based on the fault judgment result, normalize the fault mechanism discrimination result score matrix to generate a normalized mechanism discrimination matrix, and further to: The fault mechanism discrimination result score matrix M s =[ M 1,..., M i ,..., M 9] T ,Ms i ∈[0,100] , si=1…9; Normalize the fault mechanism discrimination result score matrix to obtain the normalized mechanism discrimination matrix M s : 。 12. The system according to claim 9, wherein the fault data set comprises electrical quantity characteristic data and a label corresponding to the fault type; the probability judgment unit is used to establish the fault data set, and is also used to: The data window length of the preset time before the fault and the preset time after the fault is intercepted, the fault data is sampled based on the preset sampling rate, and the fault data set D=[(x1,y1),...(x i ,y i ),...,(x n ,y n )], where x i (i=1,2,3...n) is the electrical characteristic data of all channels of the i-th fault data recording, y i (i=1,2,3...n) is the label corresponding to the actual fault type of the i-th fault data recording, y i ∈[1,2,3,4,5,6,7,8,9].

13. The system according to claim 9, wherein the probability judgment unit is further used to establish an intelligent model: The intelligent model is an inception deep convolutional neural network model including k inception deep modules, global average pooling and a fully connected layer; Randomly initialize the fully connected layer parameters w fc ,b fc , randomly initialize all convolution kernels ck of the inception deep convolutional neural network model; Design the loss function of the inception deep convolutional neural network model: in, represents the probability value predicted by the model, represents the true category label of the sample, j For the j convolution kernels, c is the total number of convolution kernels, and z is the final classification label.

14. The system according to claim 13, wherein the probability judgment unit is further configured to train the intelligent model through the following steps: In forward propagation, the feature map is input, and multiple sets of feature maps are generated through convolution kernels and pooling operations; the generated multiple sets of feature maps are spliced ​​in the depth direction to generate the final output feature map; Process the output feature map through a fully connected layer and output the final classification label z; Calculate the weight w of the fully connected layer fc and bias b fc The gradient is: in, L is the loss function; Calculate the gradient of all convolution kernels ck. When there is a convolution kernel whose output is c o When , the loss function is calculated for the output c o Convolution kernel weight w ck and bias b ck The gradient is: in, L is the loss function, w is the weight, and b is the bias; Back propagation, update the convolution kernel and fully connected layer parameters according to the gradient: in, is the learning rate, w old is the weight before iteration, w new is the weight after iteration, b old is the bias before iteration, b new is the post-iteration bias; The step of training the intelligent model is repeated until the loss function converges, and the training of the intelligent model is completed.

15. The system according to claim 9, wherein the result unit is used to fuse the normalized mechanism discriminant matrix and the discriminant probability distribution matrix to generate a fusion matrix, and is also used to: The discriminant probability distribution matrix m output by the intelligent model AI =[ m AI1 ,..., m AIi ,..., m AI9 ] T , m AIi ∈[0,1], with the normalized mechanism discriminant matrix m s To perform the fusion: in, m merge is the fusion matrix.

16. The system according to claim 15, wherein the result unit is used to establish a final fault cause identification matrix by multiplying the fusion matrix and the meteorological correlation matrix, and is also used to: m mergenew =Am merge in, A is the meteorological correlation matrix, m mergenew It is the final fault cause identification matrix.

Images