A distribution network fault accurate positioning method and system based on random forest

Abstract

The present invention discloses a distribution network fault accurate positioning method and system based on random forest, which fully combines the actual field operation, considers the influence of the commonly existing measurement error and line parameter error on the positioning error of the double-end synchronization algorithm, uses the additive Gaussian white noise function to simulate the measurement error and line parameter error, and at the same time, selects four basic algorithms with differences to obtain four positioning results respectively, then processes the four positioning results of the basic algorithm to form training sample data, and trains the random forest model, and uses the trained random forest fault positioning model to accurately locate the distribution network fault. The present invention solves the technical problem that the prior art uses a plurality of double-end fault positioning algorithms based on synchronization information to locate distribution network faults, does not fully consider the measurement error and line parameter error, and affects the distribution network fault positioning accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault monitoring, and in particular to a distribution network fault accurate positioning method and system based on random forest. Background Art

[0002] With the access of distributed generation (DG) to the distribution network, the scale of the distribution network continues to expand, and the network topology becomes more and more complex, and the probability of distribution network failure is also increasing. Fault location after a distribution network failure is a basic link in distribution network fault handling. After a fault occurs, quickly and accurately completing the precise location of the distribution line fault will help improve the speed of fault isolation and power supply recovery, and help enhance power supply reliability.

[0003] The existing distribution network fault location methods include the traveling wave method and the fault analysis method. The traveling wave method measures the traveling wave generated by the fault to accurately locate the fault. The traveling wave method has high requirements for sampling rate, large data transmission, and high requirements for the communication system. In addition, the use of the traveling wave method requires the installation of a large number of special equipment, which has high economic costs. Therefore, in actual application scenarios, the fault analysis method is more inclined to be used for distribution network fault location. The fault analysis method uses system parameters and measurement data to establish a loop equation, and directly or iteratively calculates the fault distance. It can be mainly divided into a single-terminal method and a double-terminal method. The single-terminal method is affected by the transition resistance and has a slightly poor reliability. In the trend of intelligence and informatization, more and more devices with synchronous measurement functions are configured in the distribution network. For example, the synchronous phasor measurement device can synchronously collect voltage and current signals and transmit them to the dispatching center. After these collected synchronous data are substituted into the double-terminal fault location algorithm based on synchronous information, the location result can be obtained. Although the addition of synchronous information greatly reduces the principle error of the location algorithm based on the double-terminal quantity, the influence of measurement error and line parameter error on the location result cannot be ignored.

[0004] The present invention aims to solve the problem that the measurement errors of various double-terminal fault location algorithms based on synchronization information and line parameter errors in the existing fault location methods affect the distribution network fault location accuracy, and provides a distribution network fault accurate location method and system based on random forest. Summary of the invention

[0005] The present invention provides a method and system for accurately locating distribution network faults based on random forests, which is used to solve the technical problem that in the prior art, multiple double-end fault location algorithms based on synchronization information are used to locate distribution network faults, but measurement errors and line parameter errors are not fully considered, thus affecting the accuracy of distribution network fault location.

[0006] In view of this, a first aspect of the present invention provides a method for accurately locating distribution network faults based on random forests, comprising the following steps:

[0007] S1. Build a distribution network model, perform fault simulation on the distribution network model, and obtain the three-phase voltage and current at both ends of the fault line;

[0008] S2. Use an additive Gaussian white noise function to simulate the measurement errors of the three-phase voltage and current to obtain the three-phase voltage and current containing the measurement errors, and perform error simulation on the line parameters to obtain the line parameters containing the errors;

[0009] S3, perform full-cycle Fourier transform and phase sequence transform on the three-phase voltage and current containing measurement errors to obtain the sequence electrical quantities at both ends of the line;

[0010] S4, respectively substituting the sequence electrical quantities at both ends of the line and the line parameters containing errors into the positive sequence voltage and current method based on the lumped parameter line model, the positive and negative sequence impedance equal method based on the lumped parameter line model, the modulus solution method based on the distributed parameter line model, and the phase solution method based on the distributed parameter line model to obtain the corresponding first positioning result, second positioning result, third positioning result, and fourth positioning result;

[0011] S5. Calculate the difference between the first positioning result, the second positioning result, the third positioning result and the fourth positioning result respectively to obtain a first difference, a second difference, a third difference, a fourth difference, a fifth difference and a sixth difference;

[0012] S6. Using the first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference as input parameters and the actual fault location as a label to construct a training sample;

[0013] S7. Use the training samples to train the random forest model to obtain a trained random forest model, and use the trained random forest model as a distribution network fault location model;

[0014] S8. Obtain the three-phase voltage and current at both ends of the fault line collected when a fault occurs in the distribution network, execute steps S3 to S5, obtain a first positioning result, a second positioning result, a third positioning result, a fourth positioning result, a first difference, a second difference, a third difference, a fourth difference, a fifth difference and a sixth difference, and input the first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference into a distribution network fault location model to obtain a distribution network fault location result.

[0015] Optionally, step S2 specifically includes:

[0016] The three-phase voltage and current are respectively multiplied by the first random number generated by the additive white Gaussian noise function to obtain the three-phase voltage and current containing measurement errors;

[0017] The line parameter is multiplied by a second random number generated by an additive white Gaussian noise function to obtain a line parameter containing an error.

[0018] Optionally, step S3 specifically includes:

[0019] After obtaining the three-phase voltage and current containing measurement errors, take the instantaneous values ​​of the three-phase voltage and current in the second cycle after the fault occurs and perform full-cycle Fourier transform to obtain the phasor values ​​of the three-phase voltage and current at both ends of the fault line;

[0020] The symmetrical component method is used to transform the phase values ​​of the three-phase voltage and current into phase sequence to obtain the sequence electrical quantities at both ends of the line.

[0021] Optionally, the number of CART trees of the random forest model is 100, the maximum depth of the CART tree is 20, and the minimum number of samples of leaf nodes is 5.

[0022] Optionally, the first positioning result is:

[0023]

[0024] Where x1 is the first positioning result, l is the total length of the fault line MN, z1 is the unit positive sequence impedance of the line, is the positive sequence voltage at the N end of the fault line, is the positive sequence current at the N end of the fault line, is the positive sequence voltage at the fault line M terminal, is the positive sequence current at terminal M of the fault line.

[0025] Optionally, the second positioning result is:

[0026]

[0027] Among them, x2 is the second positioning result, is the negative sequence voltage at the M end of the fault line, is the negative sequence current at the fault line M terminal, is the negative sequence voltage at the N end of the fault line, is the negative sequence current at the N end of the fault line.

[0028] Optionally, the third positioning result is:

[0029] x3=ln(A 2 +B 2 ) / 4α1

[0030]

[0031] Among them, x3 is the third positioning result, γ1 is the positive sequence propagation coefficient of the fault line MN, and Z c1 is the line positive sequence characteristic impedance, y1 is the admittance per unit length of the fault line MN, α1 is the attenuation constant, β1 is the phase constant, A is the real part, and B is the imaginary part.

[0032] Optionally, the fourth positioning result is:

[0033] x4=arctan(B / A) / 2β1

[0034] Among them, x4 is the fourth positioning result.

[0035] The second aspect of the present invention provides a distribution network fault accurate positioning system based on random forest, comprising the following modules:

[0036] Modeling module, used to build a distribution network model, perform fault simulation on the distribution network model, and obtain the three-phase voltage and current at both ends of the fault line;

[0037] The error simulation module is used to use the additive Gaussian white noise function to perform measurement error simulation on the three-phase voltage and current to obtain the three-phase voltage and current containing measurement errors, and to perform error simulation on the line parameters to obtain the line parameters containing errors;

[0038] The sequence electrical quantity module is used to perform full-cycle Fourier transform and phase sequence transformation on the three-phase voltage and current containing measurement errors to obtain the sequence electrical quantities at both ends of the line;

[0039] An initial positioning module is used to substitute the sequence electrical quantities at both ends of the line and the line parameters containing errors into the positive sequence voltage and current method based on the lumped parameter line model, the positive and negative sequence impedance equality method based on the lumped parameter line model, the module value solution method based on the distributed parameter line model, and the phase solution method based on the distributed parameter line model, respectively, to obtain the corresponding first positioning result, second positioning result, third positioning result, and fourth positioning result;

[0040] A difference calculation module, used to calculate the difference between the first positioning result, the second positioning result, the third positioning result and the fourth positioning result, respectively, to obtain a first difference, a second difference, a third difference, a fourth difference, a fifth difference and a sixth difference;

[0041] A training sample construction module, used to construct a training sample by taking the first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference as input parameters and the actual fault location as a label;

[0042] A model training module is used to train a random forest model using training samples to obtain a trained random forest model, and use the trained random forest model as a distribution network fault location model;

[0043] A fault location module is used to obtain the three-phase voltage and current at both ends of the fault line collected when a fault occurs in the distribution network, and after being processed by the sequence electrical quantity module, the initial positioning module and the difference calculation module, a first positioning result, a second positioning result, a third positioning result, a fourth positioning result, a first difference, a second difference, a third difference, a fourth difference, a fifth difference and a sixth difference are obtained. The first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference are input into a distribution network fault location model to obtain a distribution network fault location result.

[0044] Optionally, the error simulation module is specifically used for:

[0045] The three-phase voltage and current are respectively multiplied by the first random number generated by the additive white Gaussian noise function to obtain the three-phase voltage and current containing measurement errors;

[0046] The line parameter is multiplied by a second random number generated by an additive white Gaussian noise function to obtain a line parameter containing an error.

[0047] It can be seen from the above technical solutions that the distribution network fault accurate positioning method and system based on random forest provided by the present invention has the following advantages:

[0048] The distribution network fault accurate positioning method based on random forest provided by the present invention fully combines the actual field operation, considers the influence of the commonly existing measurement error and line parameter error on the positioning error of the double-end synchronization algorithm, uses the additive Gaussian white noise function to simulate the measurement error of the three-phase voltage and current, obtains the three-phase voltage and current containing the measurement error, and simulates the line parameters to obtain the line parameters containing the error. At the same time, taking the measurement results and line parameters considering the error as the fault analysis factors, four basic algorithms with differences are selected for calculation, and four positioning results are obtained respectively. Subsequently, the four positioning results of the basic algorithm are processed to form training sample data, and the random forest model is trained. Finally, after the distribution network fails, the trained random forest fault location model is used to complete the accurate positioning. The present invention utilizes the random forest in ensemble learning to construct a fault location model, and the model has high positioning accuracy, which is beneficial to reducing maintenance time, accelerating fault recovery speed, and improving system operation reliability. It also fully considers the influence of the commonly existing measurement errors and line parameter errors on the positioning error of the double-end synchronization algorithm, and the positioning effect is not affected by factors such as transition resistance, fault start angle, and line length. It is suitable for a new distribution network containing high-penetration distributed power sources, and solves the technical problem that in the prior art, a plurality of double-end fault location algorithms based on synchronization information are used to locate distribution network faults, but measurement errors and line parameter errors are not fully considered, which affects the distribution network fault location accuracy.

[0049] A distribution network fault precise positioning system based on random forest provided in the present invention is used to execute a distribution network fault precise positioning method based on random forest provided in the present invention. Its principle and technical effects are the same as those of a distribution network fault precise positioning method based on random forest provided in the present invention, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 A schematic diagram of a flow chart of a method for accurately locating distribution network faults based on random forests provided in the present invention;

[0052] Figure 2 This is a schematic diagram of the distribution network model constructed in the present invention;

[0053] Figure 3 The concentrated parameter line model sequence circuit provided in the present invention;

[0054] Figure 4 It is the positive sequence circuit of the distributed parameter line model provided in the present invention;

[0055] Figure 5 A schematic diagram of the overall construction process of the random forest model for accurate fault location provided in the present invention;

[0056] Figure 6 A schematic diagram showing the relationship between the mean absolute error of the random forest model provided in the present invention and the number of CART trees;

[0057] Figure 7 A schematic diagram of the structure of a distribution network fault precise positioning system based on random forest provided in the present invention. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] For easier understanding, see Figure 1 The present invention provides an embodiment of a method for accurately locating distribution network faults based on random forests, comprising the following steps:

[0060] Step 101: construct a distribution network model, perform fault simulation on the distribution network model, and obtain the three-phase voltage and current at both ends of the fault line.

[0061] It should be noted that, in the embodiment of the present invention, a distribution network model is first constructed, a fault is set in the distribution network model, and the three-phase voltage and current at both ends of the fault line are obtained, such as Figure 2 As shown in Figure 1, the distribution network model line adopts the Bergeron model, the ideal voltage source line voltage is 35kV, the transformer ratio is 35kV / 10.5kV, the load is a three-phase line-to-ground fixed load, and a phase-to-phase fault is set on line B1M6 (line length 6460 meters). A large number of fault simulations are performed by setting different line parameters, line types, fault locations, transition resistances, and fault start angles. The line parameters are shown in Table 1. The fault location is 5%-95% from the head end of the line (interval 5%, a total of 19 values); the transition resistance is 0Ω~55Ω (interval 5Ω, a total of 12 values); the fault start angle is 0°~324° (interval 36°, a total of 12 values). After the fault occurs, the measuring device installed at the head and end of the line can synchronously measure the three-phase voltage and current at the installation location and send it to the control center.

[0062] Table 1 Line parameter settings

[0063]

[0064] Step 102: Use an additive white Gaussian noise function to perform measurement error simulation on the three-phase voltage and current to obtain the three-phase voltage and current containing measurement errors, and perform error simulation on the line parameters to obtain the line parameters containing errors.

[0065] It should be noted that for a single fault scenario generated by simulation, the instantaneous values ​​of the three-phase voltage and current at both ends of the fault line, as well as the line parameters (such as line length, impedance, etc.) are multiplied by random numbers to simulate the measurement error and line parameter error. Considering that the measurement error and line parameter error are randomly distributed, the random number used to simulate the error in the embodiment of the present invention is generated by an additive Gaussian white noise function, that is, the three-phase voltage and current are respectively multiplied by the first random number generated by the additive Gaussian white noise function to obtain the three-phase voltage and current containing the measurement error, and the line parameters are multiplied by the second random number generated by the additive Gaussian white noise function to obtain the line parameters containing the error. The amplitude of the additive Gaussian white noise function obeys the Gaussian distribution, and the power spectrum density is uniformly distributed. When the signal-to-noise ratio is 35dB, the average absolute error of the generated random number is 1.4%, which is consistent with the actual situation of the measurement error and line parameter error. In order to expand the amount of data, the error simulation is repeated 10 times for each fault scenario.

[0066] Step 103: Perform full-cycle Fourier transform and phase sequence transform on the three-phase voltage and current containing measurement errors to obtain sequence electrical quantities at both ends of the line.

[0067] It should be noted that after completing the simulation of measurement errors and line parameter errors, the three-phase voltage and current containing measurement errors are subjected to full-cycle Fourier transform and phase sequence transformation to obtain the sequence electrical quantities at both ends of the line. Specifically, the instantaneous values ​​of the three-phase voltage and current in the second cycle after the fault occurs are taken for full-cycle Fourier transform to obtain the phasor values ​​of the three-phase voltage and current at both ends of the faulty line. The symmetrical component method is used to perform phase sequence transformation on the phasor values ​​of the three-phase voltage and current to obtain the sequence electrical quantities at both ends of the line. At this point, the electrical quantity containing the error is obtained. (positive sequence voltage at terminal M of fault line), (positive sequence voltage at the N end of the fault line), (positive sequence current at terminal M of fault line), (positive sequence current at the N end of the fault line), (negative sequence voltage at terminal M of fault line), (negative sequence voltage at the N end of the fault line), (negative sequence current at terminal M of fault line) and (negative sequence current at the N end of the fault line), and the line length l containing the error, the positive sequence resistance per unit length r1, and the positive sequence inductance per unit length x L1 and positive sequence capacitive reactance to ground x C1 .

[0068] Step 104, substitute the sequence electrical quantities at both ends of the line and the line parameters containing errors into the positive sequence voltage and current method based on the lumped parameter line model, the positive and negative sequence impedance equality method based on the lumped parameter line model, the module value solution method based on the distributed parameter line model and the phase solution method based on the distributed parameter line model, respectively, to obtain the corresponding first positioning result, second positioning result, third positioning result and fourth positioning result.

[0069] It should be noted that, in combination with the development trend of the new distribution network and the fault characteristics of the distribution line, four basic algorithms are selected in the embodiment of the present invention, namely: the positive sequence voltage and current method based on the concentrated parameter line model, the positive and negative sequence impedance equal method based on the concentrated parameter line model, the module value solution method based on the distributed parameter line model and the phase solution method based on the distributed parameter line model. The sequence electrical quantities at both ends of the line and the line parameters containing errors are respectively substituted into the positive sequence voltage and current method based on the concentrated parameter line model, the positive and negative sequence impedance equal method based on the concentrated parameter line model, the module value solution method based on the distributed parameter line model and the phase solution method based on the distributed parameter line model to obtain the corresponding first positioning result, second positioning result, third positioning result and fourth positioning result.

[0070] Specifically, the implementation principle of the first positioning result obtained using the positive sequence voltage and current method based on the lumped parameter line model is as follows:

[0071] like Figure 3 As shown, l is the total length of the fault line MN, x is the distance from the fault to the M end, is the sequence voltage at the fault location, and are the sequence voltage and current at the fault line M terminal, and are the sequence voltage and current at the N end of the fault line, respectively, i is the sequence impedance per unit length of the fault line, where i=1 represents positive sequence and i=2 represents negative sequence. After a fault occurs, the positive sequence voltage at the fault location can be calculated using the positive sequence voltage and positive sequence current at the M end of the fault line MN, or the positive sequence voltage and positive sequence current at the N end of the fault line MN. Combined with the uniqueness of the positive sequence voltage at the fault location, the following equation can be written:

[0072]

[0073] Where l is the total length of the fault line MN, z1 is the unit positive sequence impedance of the line, z1=r1+jx L1 , r1 is the positive sequence resistance per unit length of the line, x L1 is the positive sequence inductance per unit length of the line, is the positive sequence voltage at the N end of the fault line, is the positive sequence current at the N end of the fault line, is the positive sequence voltage at the fault line M terminal, is the positive sequence current at terminal M of the fault line.

[0074] Solving equation (1) can obtain the fault distance. For the convenience of distinction, the location result of the positive sequence voltage and current method based on the lumped parameter line model is recorded as the first location result x1:

[0075]

[0076] The implementation principle of the second positioning result obtained by using the positive and negative sequence impedance equality method based on the lumped parameter line model is as follows:

[0077] Considering that the distribution line is a static three-phase symmetrical element, the positive sequence impedance z1 is equal to the negative sequence impedance z2, that is:

[0078] z1=z2 (3)

[0079] Similar to the positive sequence voltage and current method, the negative sequence voltage and negative sequence current at both ends of the line can be used to represent the negative sequence voltage at the fault, and the following equation is listed:

[0080]

[0081] Formulas (1), (3) and (4) are combined to obtain the positioning result of the positive and negative sequence impedance equality method based on the lumped parameter line model, which is recorded as the second positioning result x2:

[0082]

[0083] in, is the negative sequence voltage at the M end of the fault line, is the negative sequence current at the fault line M terminal, is the negative sequence voltage at the N end of the fault line, is the negative sequence current at the N end of the fault line.

[0084] The implementation principles of the third positioning result and the fourth positioning result obtained by using the modulus solution method based on the distributed parameter line model and the phase solution method based on the distributed parameter line model are as follows:

[0085] Using distributed parameter line models, such as Figure 4As shown, according to the transmission line theory, considering the distribution characteristics of line parameters, the voltage and current distribution at each point along the line, as well as the relationship between the voltage and current vectors at both ends of the line can be obtained by solving the equation. Using this theory, the positive sequence voltage at the fault can be expressed by the positive sequence voltage and current at the M end or the positive sequence voltage and current at the N end, as shown in equation (6):

[0086]

[0087] Among them, ch is the hyperbolic cosine function, sh is the hyperbolic sine function, γ1 is the positive sequence propagation coefficient of the fault line MN, Z c1 is the line positive sequence characteristic impedance, y1 is the admittance per unit length of the fault line MN, α1 is the attenuation constant, and β1 is the phase constant.

[0088] According to formula (6), equation (7) about the fault distance can be obtained:

[0089]

[0090] Among them, A is the real part and B is the imaginary part. Combined with γ1=α1+jβ1, we can get formula (8):

[0091]

[0092] You can get:

[0093] Therefore, the third positioning result x3 obtained by the modulus solution method based on the distributed parameter line model is:

[0094] x3=ln(A 2 +B 2 ) / 4α1 (9)

[0095] The phase solution method based on the distributed parameter line model uses the phase constant β1 and the phase angle information of formula (8) to obtain tan(2β1x)=B / A. Therefore, the fourth positioning result x4 obtained by the phase solution method based on the distributed parameter line model is:

[0096] x4=arctan(B / A) / 2β1

[0097] Since α1<β1, the phase solution method has better stability than the module value solution method.

[0098] The above four basic algorithms have the following differences in line models, solution ideas, and solution parameters:

[0099] ① Line model. The positive sequence voltage and current method and the positive and negative sequence impedance equal method use a concentrated parameter model with a small amount of calculation. The modulus value solution method and the phase solution method use a distributed parameter line model, which has a higher theoretical positioning accuracy.

[0100] ② Solution ideas. The positive and negative sequence impedance equality method solves the fault distance by combining the positive and negative sequence network equations and the information that "the positive and negative sequence impedance of the line are equal", which is suitable for extreme scenarios where line parameters are missing. The distributed parameter modulus solution method uses the attenuation constant and modulus information for solution, and the distributed parameter phase solution method uses the phase constant and phase information for solution.

[0101] ③ Parameter solution. Compared with the positive-sequence voltage and current method, the positive-negative sequence impedance equality method uses the negative-sequence voltage and current information but does not use the line impedance parameters. The distributed parameter modulus solution method and the phase solution method consider the positive-sequence capacitance parameter more finely. Although more parameter information needs to be obtained, the theoretical positioning accuracy is higher.

[0102] Step 105 , respectively calculate the pairwise differences of the first positioning result, the second positioning result, the third positioning result and the fourth positioning result to obtain a first difference, a second difference, a third difference, a fourth difference, a fifth difference and a sixth difference.

[0103] It should be noted that after step 104, the positioning results corresponding to the four basic algorithms are obtained: the first positioning result x1, the second positioning result x2, the third positioning result x3 and the fourth positioning result x4. Compared with only using the positioning results of the four basic algorithms as input features, if the pairwise differences are also used as input features, the trained model can achieve better results on the test samples. Therefore, in the embodiment of the present invention, the pairwise differences of the first positioning result, the second positioning result, the third positioning result and the fourth positioning result are calculated respectively (the number of differences is: indivual).

[0104] Step 106: Use the first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference as input parameters and the actual fault location as a label to construct a training sample.

[0105] It should be noted that for the i-th fault sample, the positioning results of the above four basic algorithms are x i,1 , x i,2 , x i,3 and x i,4 , the actual fault location is y i After step 105, the i-th training sample can be obtained and recorded as in, is the input, y i is the output. To simplify the model calculation process, the 10 input quantities of the first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference can be normalized, and the normalized data can be used as training samples. Among them, the normalization method can be divided by the line length, that is, the first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference are all divided by the line length for normalization.

[0106] Step 107: Use the training samples to train the random forest model to obtain a trained random forest model, and use the trained random forest model as a distribution network fault location model.

[0107] It should be noted that when the training samples are constructed, they can form a training sample set. Use the training samples to train the random forest model. Randomly select a part of the samples from the training sample set as the validation set, and determine the hyperparameters by observing the performance of the model on the validation set. Then use all the training sample set data to train the random forest model once. Finally, save the trained model.

[0108] Random forest is an ensemble learning algorithm based on Classification and Regression Tree (CART). It has the advantages of being insensitive to outliers, fast computing speed, and not prone to overfitting. It can complete classification and regression tasks. For regression problems, the training sample set Composed of input space, CART recursively divides each region into two sub-regions and determines the output value on each sub-region. The specific steps are as follows:

[0109] 1. Constructing a random forest model

[0110] (1) Traverse feature j and partition variable s, and select the variable pair (j, s) that achieves the minimum in equation (11).

[0111]

[0112] Among them, R1 and R2 are two sub-regions, and c1 and c2 are output values ​​of the two sub-regions.

[0113] (2) Using the selected variable pair (j, s), the data is divided into two sub-regions and the output value of each sub-region is calculated.

[0114]

[0115]

[0116] in, is the prediction result of the regression tree, N m For region R m The number of samples in .

[0117] (3) For the two sub-regions, repeat steps (1) and (2) until the stopping condition is met.

[0118] (4) Divide the input space into M regions, R1, R2, …, R M After that, the constructed CART regression tree is obtained:

[0119]

[0120] Here, I is a function that is 1 when the condition is met and 0 when the condition is not met.

[0121] The random forest algorithm that can integrate multiple CART trees often has better performance. When using random forests to complete the fusion of the positioning results of the basic algorithm, the overall process is: first, random sampling with replacement is used to generate several sub-datasets from the training data set. Then, multiple CART regression trees are generated independently and in parallel using the sub-datasets. It is worth noting that when dividing the left and right subtrees, the random forest will randomly select a part of the features from all the features on the node, and then find the optimal features and partitioning variables in this selected part of the features. Finally, the test sample is input into multiple regression trees, and the arithmetic average of the multiple regression trees is taken to obtain the positioning result of the random forest. The overall construction process of the fault location model based on random forests is as follows: Figure 5 shown.

[0122] 2. Adjust the parameters to determine the hyperparameters of the random forest model

[0123] Randomly take 95% of the training sample set for model training, and take the remaining 5% as the validation set for model verification. By observing the performance of the model on the validation set, adjust and determine the hyperparameters. Due to the small number of sample features, the minimum number of samples required for internal node re-division, the maximum number of features, the maximum number of leaf nodes and other hyperparameters take the default values. In this way, there are only three hyperparameters that need to be adjusted: the number of CART trees, the maximum depth of the CART tree, and the minimum number of leaf nodes.

[0124] Taking the determination of the number of CART trees as an example, the random forest model is trained with 95% of the training set data, and the mean absolute error on the validation set is recorded as the number of CART trees changes, as shown in Figure 2. Figure 6 As shown. Figure 6Three pieces of information can be obtained from the results: ① Compared with 4 input features, the average absolute error of the model trained with 10 input features is smaller and the positioning accuracy is higher; ② As the number of CART trees increases, the average absolute error gradually decreases; ③ When the number of trees reaches a certain number, the positioning accuracy will no longer be significantly improved. In order to balance the model effect and training expenses, the number of CART trees is taken as 100. Using the same method, taking into account the positioning accuracy and model complexity, it is finally determined that the maximum depth of the CART tree is 20 and the minimum number of leaf node samples is 5.

[0125] 3. After the hyperparameters are determined, train again using the data from the entire training sample set

[0126] The random forest model is trained again using all training samples to finally obtain the distribution network fault location model, which is then saved.

[0127] Step 108, obtain the three-phase voltage and current at both ends of the fault line collected when the distribution network fails, execute steps 103 to 105, obtain the first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference, input the first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference into the distribution network fault location model to obtain the distribution network fault location result.

[0128] It should be noted that the distribution network fault location model obtained in step 107 is applied to a fault location scenario in actual application. After a fault occurs in a real distribution line, the three-phase voltage and current at both ends of the fault line are collected, and then the corresponding first positioning result, second positioning result, third positioning result, fourth positioning result, first difference, second difference, third difference, fourth difference, fifth difference and sixth difference are obtained according to the processing flow of steps 103 to 105. Then, the first positioning result, second positioning result, third positioning result, fourth positioning result, first difference, second difference, third difference, fourth difference, fifth difference and sixth difference (whether normalization is required according to actual needs to obtain normalized data) are input into the distribution network fault location model to obtain the distribution network fault location result.

[0129] For the distribution network fault location model obtained in the embodiment of the present invention, the following provides a demonstration of the application effect of testing the generalization ability of the distribution network fault location model in terms of fault location, line parameters, fault start angle, transition resistance, load, line length, etc., and its applicability in fault location of a distribution network containing high penetration distributed power sources:

[0130] Generalization ability on fault location:

[0131] As shown in Table 2, the fault locations of the test set are different from those of the training set, which are 807.5m, 1453.5m, 2099.5m, 2745.5m, 3391.5m, 4037.5m, 4683.5m, 5329.5m, and 5975.5m. The mean absolute errors of the four basic algorithms and the random forest model on the corresponding test sets are calculated.

[0132] Due to the existence of measurement errors and line parameter errors, the positioning errors of the basic algorithms are large and do not meet the requirements of the distribution network. Comparing the positioning errors of the algorithms in Table 2, it is found that compared with the basic algorithm, the positioning error of the trained random forest model is significantly reduced and is less than 200 meters, which meets the requirements of field operation. This shows that the constructed model has good generalization ability in fault location.

[0133] Table 2 Generalization performance of random forest model on fault location

[0134]

[0135] In Table 2, the numbers are the mean absolute errors of different algorithms on the corresponding test sets, in meters.

[0136] Generalization capability on line parameters:

[0137] Different distribution networks and different lines often have differences in line parameters. The generalization ability of the random forest model on line parameters is verified. Four sets of line parameters are randomly generated within a reasonable range, which are different from the line parameters in the training set, as shown in Table 3.

[0138] Similarly, under four sets of line parameters, the mean absolute error between the random forest model and the basic algorithm on the test set is calculated, as shown in Table 4. It is found that the positioning error of the trained model is still the smallest and is less than 200m, which also meets the requirements of field operation.

[0139] Table 3. Training set line parameter settings

[0140]

[0141] Table 4 Generalization performance of random forest model on line parameters

[0142]

[0143]

[0144] In Table 4, the numbers are the mean absolute errors of different algorithms on the corresponding test sets, in meters.

[0145] Generalization capability in terms of fault starting angle, transition resistance, load, line length, etc.:

[0146] By changing the fault starting angle and transition resistance, line load, line length and other factors, the generalization performance of the random forest fault location model is tested, as shown in Table 5, Table 6 and Table 7 respectively.

[0147] The results show that the random forest model, which integrates the positioning results of the four basic algorithms, has significantly reduced the positioning error, all of which are less than 200m. The random forest positioning model has higher positioning accuracy and has good generalization ability in terms of fault starting angle and transition resistance.

[0148] Table 5 Generalization performance of random forest model on fault onset angle and transition resistance

[0149]

[0150] Table 6 Generalization performance of random forest model on line load

[0151]

[0152]

[0153] Table 7 Generalization performance of random forest model on line length

[0154]

[0155] In Tables 5-7, the numbers are the mean absolute errors of different algorithms on the corresponding test sets, in meters.

[0156] Adaptability when distributed power supply is connected:

[0157] Under the background of the "dual carbon" strategy and the construction of a new power system, more and more distributed power sources will be connected to the distribution network. After the distributed power sources are connected, the fault accurate location algorithm in the distribution network should still have high positioning accuracy and good adaptability. In order to verify the adaptability of the constructed random forest model when a large number of distributed power sources are connected, corresponding tests were carried out.

[0158] The two types of distributed power sources, rotary and inverter, are connected to the end of the distribution line respectively, and then a phase-to-phase fault is set on the line. The electrical quantities at both ends of the line are collected and processed to finally form a test set sample. The average absolute error of the positioning results of the four basic algorithms and the random forest algorithm on the corresponding test set is calculated, as shown in Table 8. The rotary distributed power source is simulated by the series connection of an ideal voltage source and an internal resistance. The inverter distributed power source is photovoltaic. Low penetration refers to the power of the distributed power source of the fault feeder / the load of the line ≈30%, and high penetration refers to the power of the distributed power source of the fault feeder / the load of the line ≈60%.

[0159] Table 8 Adaptability of the random forest model to a distribution network containing distributed power sources

[0160]

[0161]

[0162] In Table 8, the numbers are the mean absolute errors of different algorithms on the corresponding test sets, in meters.

[0163] It can be found that when different types of distributed power sources are connected, compared with the basic algorithm, the constructed random forest fault location model can still achieve the minimum positioning error, and it is still less than 200m. This is because the basic algorithm used in constructing the random forest location model uses the fault steady-state power frequency electrical quantity, avoiding the complex transient process of the distributed power source after the fault occurs. Therefore, the constructed random forest fault location model is still applicable in the new distribution network with high penetration distributed power sources.

[0164] The distribution network fault accurate positioning method based on random forest provided in the embodiment of the present invention fully combines the actual field operation, considers the influence of the commonly existing measurement error and line parameter error on the positioning error of the double-end synchronization algorithm, uses the additive Gaussian white noise function to simulate the measurement error of the three-phase voltage and current, obtains the three-phase voltage and current containing the measurement error, and simulates the error of the line parameter to obtain the line parameter containing the error. At the same time, taking the measurement results and line parameters considering the error as the fault analysis factors, four basic algorithms with differences are selected for calculation, and four positioning results are obtained respectively. Subsequently, the four positioning results of the basic algorithm are processed to form training sample data, and the random forest model is trained. Finally, after the distribution network fails, the trained random forest fault location model is used to complete the accurate positioning. The present invention utilizes the random forest in ensemble learning to construct a fault location model, and the model has high positioning accuracy, which is beneficial to reducing maintenance time, accelerating fault recovery speed, and improving system operation reliability. It also fully considers the influence of the commonly existing measurement errors and line parameter errors on the positioning error of the double-end synchronization algorithm, and the positioning effect is not affected by factors such as transition resistance, fault start angle, and line length. It is suitable for a new distribution network containing high-penetration distributed power sources, and solves the technical problem that in the prior art, a plurality of double-end fault location algorithms based on synchronization information are used to locate distribution network faults, but measurement errors and line parameter errors are not fully considered, which affects the distribution network fault location accuracy.

[0165] For easier understanding, see Figure 7 The present invention provides an embodiment of a distribution network fault accurate positioning system based on random forest, comprising the following modules:

[0166] Modeling module, used to build a distribution network model, perform fault simulation on the distribution network model, and obtain the three-phase voltage and current at both ends of the fault line;

[0167] The error simulation module is used to use the additive Gaussian white noise function to perform measurement error simulation on the three-phase voltage and current to obtain the three-phase voltage and current containing measurement errors, and to perform error simulation on the line parameters to obtain the line parameters containing errors;

[0168] The sequence electrical quantity module is used to perform full-cycle Fourier transform and phase sequence transformation on the three-phase voltage and current containing measurement errors to obtain the sequence electrical quantities at both ends of the line;

[0169] An initial positioning module is used to substitute the sequence electrical quantities at both ends of the line and the line parameters containing errors into the positive sequence voltage and current method based on the lumped parameter line model, the positive and negative sequence impedance equality method based on the lumped parameter line model, the module value solution method based on the distributed parameter line model, and the phase solution method based on the distributed parameter line model, respectively, to obtain the corresponding first positioning result, second positioning result, third positioning result, and fourth positioning result;

[0170] A difference calculation module, used to calculate the difference between the first positioning result, the second positioning result, the third positioning result and the fourth positioning result, respectively, to obtain a first difference, a second difference, a third difference, a fourth difference, a fifth difference and a sixth difference;

[0171] A training sample construction module, used to construct a training sample by taking the first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference as input parameters and the actual fault location as a label;

[0172] A model training module is used to train a random forest model using training samples to obtain a trained random forest model, and use the trained random forest model as a distribution network fault location model;

[0173] A fault location module is used to obtain the three-phase voltage and current at both ends of the fault line collected when a fault occurs in the distribution network, and after being processed by the sequence electrical quantity module, the initial positioning module and the difference calculation module, a first positioning result, a second positioning result, a third positioning result, a fourth positioning result, a first difference, a second difference, a third difference, a fourth difference, a fifth difference and a sixth difference are obtained. The first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference are input into a distribution network fault location model to obtain a distribution network fault location result.

[0174] The error simulation module is specifically used for:

[0175] The three-phase voltage and current are respectively multiplied by the first random number generated by the additive white Gaussian noise function to obtain the three-phase voltage and current containing measurement errors;

[0176] The line parameter is multiplied by a second random number generated by an additive white Gaussian noise function to obtain a line parameter containing an error.

[0177] The sequence electrical quantity module is specifically used for:

[0178] After obtaining the three-phase voltage and current containing measurement errors, take the instantaneous values ​​of the three-phase voltage and current in the second cycle after the fault occurs and perform full-cycle Fourier transform to obtain the phasor values ​​of the three-phase voltage and current at both ends of the fault line;

[0179] The symmetrical component method is used to transform the phase values ​​of the three-phase voltage and current into phase sequence to obtain the sequence electrical quantities at both ends of the line.

[0180] The number of CART trees in the random forest model is 100, the maximum depth of the CART tree is 20, and the minimum number of samples for a leaf node is 5.

[0181] The first positioning result is:

[0182]

[0183] Where x1 is the first positioning result, l is the total length of the fault line MN, z1 is the unit positive sequence impedance of the line, is the positive sequence voltage at the N end of the fault line, is the positive sequence current at the N end of the fault line, is the positive sequence voltage at the fault line M terminal, is the positive sequence current at terminal M of the fault line.

[0184] The second positioning result is:

[0185]

[0186] Among them, x2 is the second positioning result, is the negative sequence voltage at the M end of the fault line, is the negative sequence current at the fault line M terminal, is the negative sequence voltage at the N end of the fault line, is the negative sequence current at the N end of the fault line.

[0187] The third positioning result is:

[0188] x3=ln(A 2 +B 2 ) / 4α1

[0189]

[0190] Among them, x3 is the third positioning result, γ1 is the positive sequence propagation coefficient of the fault line MN, and Z c1 is the line positive sequence characteristic impedance, y1 is the admittance per unit length of the fault line MN, α1 is the attenuation constant, β1 is the phase constant, A is the real part, and B is the imaginary part.

[0191] The fourth positioning result is:

[0192] x4=arctan(B / A) / 2β1

[0193] Among them, x4 is the fourth positioning result.

[0194] The distribution network fault accurate positioning system based on random forest provided in the embodiment of the present invention fully combines the actual field operation, considers the influence of the commonly existing measurement error and line parameter error on the positioning error of the double-end synchronization algorithm, uses the additive Gaussian white noise function to simulate the measurement error of the three-phase voltage and current, obtains the three-phase voltage and current containing the measurement error, and simulates the line parameters to obtain the line parameters containing the error. At the same time, taking the measurement results and line parameters considering the error as the fault analysis factors, four basic algorithms with differences are selected for calculation, and four positioning results are obtained respectively. Subsequently, the four positioning results of the basic algorithm are processed to form training sample data, and the random forest model is trained. Finally, after the distribution network fails, the trained random forest fault location model is used to complete the accurate positioning. The present invention utilizes the random forest in ensemble learning to construct a fault location model, and the model has high positioning accuracy, which is beneficial to reducing maintenance time, accelerating fault recovery speed, and improving system operation reliability. It also fully considers the influence of the commonly existing measurement errors and line parameter errors on the positioning error of the double-end synchronization algorithm, and the positioning effect is not affected by factors such as transition resistance, fault start angle, and line length. It is suitable for a new distribution network containing high-penetration distributed power sources, and solves the technical problem that in the prior art, a plurality of double-end fault location algorithms based on synchronization information are used to locate distribution network faults, but measurement errors and line parameter errors are not fully considered, which affects the distribution network fault location accuracy.

[0195] A distribution network fault precise positioning system based on random forest provided in the present invention is used to execute a distribution network fault precise positioning method based on random forest provided in the present invention. Its principle and technical effects are the same as those of a distribution network fault precise positioning method based on random forest provided in the present invention, and will not be repeated here.

[0196] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A distribution network fault accurate location method based on random forest, characterized in that: The following steps are involved: S1. Build a distribution network model, perform fault simulation on the distribution network model, and obtain the three-phase voltage and current at both ends of the fault line; S2. Use an additive Gaussian white noise function to simulate the measurement errors of the three-phase voltage and current to obtain the three-phase voltage and current containing the measurement errors, and perform error simulation on the line parameters to obtain the line parameters containing the errors; S3, perform full-cycle Fourier transform and phase sequence transform on the three-phase voltage and current containing measurement errors to obtain the sequence electrical quantities at both ends of the line; S4. Substitute the sequence electrical quantities at both ends of the line and the line parameters containing errors into the positive-sequence voltage and current method based on the lumped parameter line model, the positive- and negative-sequence impedance equality method based on the lumped parameter line model, the module value solution method based on the distributed parameter line model, and the phase solution method based on the distributed parameter line model, respectively, to obtain a first positioning result corresponding to the positive-sequence voltage and current method based on the lumped parameter line model, a second positioning result corresponding to the positive- and negative-sequence impedance equality method based on the lumped parameter line model, a third positioning result corresponding to the module value solution method based on the distributed parameter line model, and a fourth positioning result corresponding to the phase solution method based on the distributed parameter line model; S5. Calculate the difference between the first positioning result, the second positioning result, the third positioning result and the fourth positioning result respectively to obtain a first difference, a second difference, a third difference, a fourth difference, a fifth difference and a sixth difference; S6. Using the first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference as input parameters and the actual fault location as a label to construct a training sample; S7. Use the training samples to train the random forest model to obtain a trained random forest model, and use the trained random forest model as a distribution network fault location model; S8. Obtain the three-phase voltage and current at both ends of the fault line collected when a fault occurs in the distribution network, execute steps S3 to S5, obtain a first positioning result, a second positioning result, a third positioning result, a fourth positioning result, a first difference, a second difference, a third difference, a fourth difference, a fifth difference and a sixth difference, and input the first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference into a distribution network fault location model to obtain a distribution network fault location result; The first positioning result is: Where x1 is the first positioning result, l is the total length of the fault line MN, z1 is the unit positive sequence impedance of the line, is the positive sequence voltage at the N end of the fault line, is the positive sequence current at the N end of the fault line, is the positive sequence voltage at the fault line M terminal, is the positive sequence current at terminal M of the fault line; The second positioning result is: Among them, x2 is the second positioning result, is the negative sequence voltage at the M end of the fault line, is the negative sequence current at the fault line M terminal, is the negative sequence voltage at the N end of the fault line, is the negative sequence current at the N end of the fault line; The third positioning result is: x3=ln(A 2 +B 2 ) / 4α1 Among them, x3 is the third positioning result, γ1 is the positive sequence propagation coefficient of the fault line MN, and Z c1 is the line positive sequence characteristic impedance, y1 is the admittance per unit length of the fault line MN, α1 is the attenuation constant, β1 is the phase constant, A is the real part, and B is the imaginary part; The fourth positioning result is: x4=arctan(B / A) / 2β1 Among them, x4 is the fourth positioning result.

2. The method for accurately locating distribution network faults based on random forest according to claim 1 is characterized in that: Step S2 specifically includes: The three-phase voltage and current are respectively multiplied by the first random number generated by the additive white Gaussian noise function to obtain the three-phase voltage and current containing measurement errors; The line parameter is multiplied by a second random number generated by an additive white Gaussian noise function to obtain a line parameter containing an error.

3. The method for accurately locating distribution network faults based on random forest according to claim 1 is characterized in that: Step S3 specifically includes: After obtaining the three-phase voltage and current containing measurement errors, take the instantaneous values ​​of the three-phase voltage and current in the second cycle after the fault occurs and perform full-cycle Fourier transform to obtain the phasor values ​​of the three-phase voltage and current at both ends of the fault line; The symmetrical component method is used to transform the phase values ​​of the three-phase voltage and current into phase sequence to obtain the sequence electrical quantities at both ends of the line.

4. The method for accurately locating distribution network faults based on random forest according to claim 1, characterized in that: The number of CART trees in the random forest model is 100, the maximum depth of the CART tree is 20, and the minimum number of samples for a leaf node is 5.

5. A distribution network fault accurate positioning system based on random forest, characterized in that: Includes the following modules: Modeling module, used to build a distribution network model, perform fault simulation on the distribution network model, and obtain the three-phase voltage and current at both ends of the fault line; The error simulation module is used to use the additive Gaussian white noise function to perform measurement error simulation on the three-phase voltage and current to obtain the three-phase voltage and current containing measurement errors, and to perform error simulation on the line parameters to obtain the line parameters containing errors; The sequence electrical quantity module is used to perform full-cycle Fourier transform and phase sequence transformation on the three-phase voltage and current containing measurement errors to obtain the sequence electrical quantities at both ends of the line; An initial positioning module is used to substitute the sequence electrical quantities at both ends of the line and the line parameters containing errors into the positive sequence voltage and current method based on the lumped parameter line model, the positive and negative sequence impedance equality method based on the lumped parameter line model, the module value solution method based on the distributed parameter line model, and the phase solution method based on the distributed parameter line model, respectively, to obtain a first positioning result corresponding to the positive sequence voltage and current method based on the lumped parameter line model, a second positioning result corresponding to the positive and negative sequence impedance equality method based on the lumped parameter line model, a third positioning result corresponding to the module value solution method based on the distributed parameter line model, and a fourth positioning result corresponding to the phase solution method based on the distributed parameter line model; A difference calculation module, used to calculate the difference between the first positioning result, the second positioning result, the third positioning result and the fourth positioning result, respectively, to obtain a first difference, a second difference, a third difference, a fourth difference, a fifth difference and a sixth difference; A training sample construction module, used to construct a training sample by taking the first positioning result, the second positioning result, the third positioning result, the fourth positioning result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference as input parameters and the actual fault location as a label; A model training module is used to train a random forest model using training samples to obtain a trained random forest model, and use the trained random forest model as a distribution network fault location model; a fault location module, for acquiring the three-phase voltage and current at both ends of the fault line collected when a fault occurs in the distribution network, and obtaining a first location result, a second location result, a third location result, a fourth location result, a first difference, a second difference, a third difference, a fourth difference, a fifth difference and a sixth difference after being processed by the sequence electrical quantity module, the initial location module and the difference calculation module; and inputting the first location result, the second location result, the third location result, the fourth location result, the first difference, the second difference, the third difference, the fourth difference, the fifth difference and the sixth difference into a distribution network fault location model to obtain a distribution network fault location result; The first positioning result is: Where x1 is the first positioning result, l is the total length of the fault line MN, z1 is the unit positive sequence impedance of the line, is the positive sequence voltage at the N end of the fault line, is the positive sequence current at the N end of the fault line, is the positive sequence voltage at the fault line M terminal, is the positive sequence current at terminal M of the fault line; The second positioning result is: Among them, x2 is the second positioning result, is the negative sequence voltage at the M end of the fault line, is the negative sequence current at the fault line M terminal, is the negative sequence voltage at the N end of the fault line, is the negative sequence current at the N end of the fault line; The third positioning result is: x3=ln(A 2 +B 2 ) / 4α1 Among them, x3 is the third positioning result, γ1 is the positive sequence propagation coefficient of the fault line MN, and Z c1 is the line positive sequence characteristic impedance, y1 is the admittance per unit length of the fault line MN, α1 is the attenuation constant, β1 is the phase constant, A is the real part, and B is the imaginary part; The fourth positioning result is: x4=arctan(B / A) / 2β1 Among them, x4 is the fourth positioning result.

6. The distribution network fault accurate positioning system based on random forest according to claim 5 is characterized in that: The error simulation module is specifically used for: The three-phase voltage and current are respectively multiplied by the first random number generated by the additive white Gaussian noise function to obtain the three-phase voltage and current containing measurement errors; The line parameter is multiplied by a second random number generated by an additive white Gaussian noise function to obtain a line parameter containing an error.

Images