Distribution Line Fault Location Method Using Multi-Algorithm Location Result Data Fusion

Abstract

The present invention discloses a fault location method for distribution lines by fusing the positioning result data of multiple algorithms. An electromagnetic transient simulation model of the distribution network is established and simulated under multiple fault scenarios, and the three-phase voltage and current are synchronously recorded; the measurement error and line parameter error are simulated; the fundamental frequency phasor extraction and phase sequence transformation of the voltage and current are carried out; several precise positioning algorithms adapted to the development trend and fault characteristics of the distribution network are selected as the basic algorithms, and several positioning results are obtained by substituting the data for calculation; the implementation method and input and output of the data fusion model are determined, and the data fusion model is trained using the normalized training data set. After a fault occurs in the distribution line, the relevant data is collected, processed and input into the proposed data fusion model, and the fault location result can be obtained. The present invention can reduce the influence of measurement error and line parameter error on the precise fault location of the distribution line and improve the positioning accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of distribution network fault location applications, and particularly relates to a distribution line fault location method using data fusion of positioning result data of multiple algorithms. Background Art

[0002] With the continuous development of society and economy, power companies and power users have higher and higher requirements for power supply reliability. The distribution network has the characteristics of large scale, close connection with users, and high fault probability. After a fault occurs, the data collected by the measuring devices in the distribution network should be fully utilized to carry out fault location work, laying a foundation for subsequent fault isolation and power supply restoration. Completing the accurate fault location of the distribution line quickly and accurately is crucial for accelerating fault recovery, shortening the power outage time, improving power supply reliability and user satisfaction. Although the measuring devices with synchronization functions provide a basis for the two-terminal synchronization algorithms, inevitable measurement errors and line parameter errors will lead to large positioning errors of these algorithms, making it difficult to meet the requirements of power supply reliability.

[0003] With the increase in investment in the distribution network and the development trend of intelligentization and informatization, more and more measuring devices with synchronization functions are deployed in the distribution network. For example, a synchronized phasor measurement unit can synchronously collect analog voltage and current signals and transmit them to the data concentrator in the dispatching center. Based on the lumped parameter or distributed parameter line model, using these collected synchronized voltage and current signals and line parameters, a variety of two-terminal synchronization fault location algorithms can be used for the accurate fault location of the distribution line.

[0004] When actually carrying out accurate fault location, factors such as transfer error and quantization error will cause measurement errors, and factors such as environmental changes and imperfect management systems will cause line parameter errors. When these inevitable measurement errors and line parameter errors are large, the positioning errors of the two-terminal synchronization fault location algorithms are also large, and the positioning accuracy is affected.

[0005] Due to the differences in the line models, electrical parameters, line parameters, solution strategies, etc. used in various existing two-terminal synchronization fault location algorithms, their positioning results have differences and complementarities. If the information contained in this difference and complementarity can be fully exploited and the results of multiple two-terminal synchronization fault location algorithms are fused, it may be beneficial to reduce the influence of measurement errors and line parameter errors on the positioning accuracy and reduce the positioning error.

[0006] Therefore, it is very necessary to fully utilize the positioning results of multiple algorithms and propose an accurate distribution line fault location method based on data fusion, which is beneficial to reducing the influence of measurement errors and line parameter errors on the positioning accuracy. Summary of the Invention

[0007] The object of the present invention is to provide a distribution line fault location method using multi-algorithm location result data fusion to solve the problem that measurement errors and line parameter errors affect the accuracy of distribution network fault location in the existing background. The present invention can reduce the influence of the commonly existing measurement errors and line parameter errors on the accurate fault location of distribution lines, improve the location accuracy, and is of great significance for accelerating fault recovery, shortening power outage time, and improving power supply reliability and user satisfaction.

[0008] To achieve the above object, the present invention adopts the following technical solutions:

[0009] The present invention first generates a large number of distribution line fault samples through the electromagnetic transient simulation software PSCAD / EMTDC and the data processing software MATLAB; then, in combination with the development trend and fault characteristics of the distribution network, basic location algorithms are selected, and the input, output, and implementation method of the data fusion model for accurate fault location are determined; then, using the generated training data set, the internal structure and parameters of the data fusion model are trained; finally, when a fault occurs in the actual distribution network, the trained data fusion model is used to complete the accurate fault location.

[0010] Specifically, it includes the following steps:

[0011] Step 1: Use PSCAD / EMTDC to establish an electromagnetic transient simulation model of the distribution network, set multiple fault scenarios for a large number of simulations, and synchronously record the three-phase voltages and currents at both ends of the fault line to form a large number of fault samples.

[0012] Step 2: For each fault sample, simulate the measurement errors of the three-phase voltages and currents and the line parameter errors to obtain the three-phase voltages and currents containing measurement errors and the line parameters containing errors.

[0013] Step 3: Extract the fundamental frequency phasors of the three-phase voltages and currents containing measurement errors, and then perform phase sequence transformation on the fundamental frequency phasors to obtain the positive and negative sequence voltage and current at both ends of the fault line.

[0014] Step 4: Select several accurate location algorithms that adapt to the development trend and fault characteristics of the distribution network as the basic algorithms, and substitute the line parameters containing errors and the positive and negative sequence voltage and current into the basic algorithms to calculate several location results.

[0015] Step 5: Use the artificial neural network as the implementation method of the data fusion model. Use several location results and the line resistance, inductance, and capacitance as the input of the data fusion model, and use the fault location as the output; perform normalization processing on the input and output to obtain the training data set. So far, the implementation method and input and output of the data fusion model have been determined.

[0016] Step 6: Use the training dataset obtained in Step 5 to train the data fusion model, obtain the internal structure and parameters of the data fusion model, and save the trained data fusion model.

[0017] Step 7: After a real distribution line fails, use the data fusion model trained in Step 6 for fault location.

[0018] Furthermore, when setting multiple fault scenarios in Step 1, different line parameters, line types, fault locations, transition resistances, and fault starting angles are considered and set, so that the number of fault samples is rich enough to be used for the training of the data fusion model.

[0019] Furthermore, when simulating measurement errors and line parameter errors in Step 2, the Additive White Gaussian Noise (AWGN) function is used. This function is a basic noise and interference model, whose amplitude follows a Gaussian distribution and the power spectral density is uniformly distributed. By reasonably setting the signal-to-noise ratio according to the actual situation, the measurement errors and line parameter errors in the distribution network can be well simulated.

[0020] Furthermore, when extracting fundamental frequency phasors in Step 3, the data of the second power frequency cycle after the fault is used, and the fast Fourier transform is adopted. When performing phase sequence transformation, the symmetrical component method is used.

[0021] Furthermore, several fault precise location algorithms can be selected in Step 4. Taking four as an example for illustration, these four algorithms are the positive sequence voltage and current method and the positive and negative sequence impedance equality method based on the lumped parameter line model, and the modulus value solving method and the phase solving method based on the distributed parameter line model, which are respectively denoted as basic algorithm one, basic algorithm two, basic algorithm three, and basic algorithm four.

[0022] All four basic algorithms are based on the fault steady-state power frequency synchronous information, which fully considers the development trend and fault characteristics of the distribution network: ① With the increase in investment in intelligentization and informatization, more and more measurement devices in the distribution network provide available synchronous information for precise fault location of lines; ② Different from transmission lines, faults in distribution lines do not need to be cleared within an extremely short time, so the fault steady-state information can be used for fault location; ③ Algorithms for processing power frequency information, including Fourier transform and phase sequence transformation, are simple and reliable and have low requirements for the sampling rate of equipment.

[0023] Introduce the location principles of the four basic algorithms.

[0024] Basic algorithm one is the positive sequence voltage and current method based on the lumped parameter line model. Using the positive sequence voltage and positive sequence current at the head of the fault line, the positive sequence voltage at the fault point can be calculated Utilize the positive-sequence voltage at the end of the faulty line and the positive-sequence current to also calculate the positive-sequence voltage at the fault location. The system of equations is as follows:

[0025]

[0026] where l is the length of the faulty line, x is the distance from the fault point to the beginning of the line, z1 is the positive-sequence impedance per unit length of the line, r1 is the positive-sequence resistance per unit length of the line, and x L1 is the positive-sequence reactance per unit length of the line.

[0027] Since the system of equations contains two equations and only has two unknowns, x and it can thus be solved. Denote the positioning result of Basic Algorithm 1 as x1:

[0028]

[0029] Basic Algorithm 2 is the method of equal positive and negative sequence impedances based on the lumped-parameter line model. If the positive-sequence impedance per unit length z1 of the line is unknown, the following approach can be taken. Similar to Basic Algorithm 1, the negative-sequence voltage at the fault location can be deduced either from the negative-sequence voltage at the beginning of the line and the negative-sequence current or from the negative-sequence voltage at the end of the line and the negative-sequence current The system of equations is as follows:

[0030]

[0031] where z2 is the negative-sequence impedance per unit length of the line.

[0032] Since the distribution line is a three-phase static component, the positive-sequence impedance is equal to the negative-sequence impedance, i.e., there is a relationship:

[0033] z1 = z2

[0034] By combining this relationship and the two systems of equations, a system of equations containing 5 equations is obtained:

[0035]

[0036] This system of equations contains 5 unknowns (x, z1, and z2), so it can be solved. Denote the positioning result of Basic Algorithm 2 as x2:

[0037]

[0038] Basic Algorithm Three and Basic Algorithm Four are the modulus solution method and the phase solution method based on the distributed parameter line model. The positive sequence voltage at the fault point can be calculated either from the positive sequence voltage and positive sequence current at the head of the faulty line, or from the positive sequence voltage and positive sequence current at the end of the faulty line. The equations are written as follows:

[0039]

[0040] where is the positive sequence propagation coefficient, y1 is the positive sequence admittance to ground of the line, y1 = j / x C1 , x C1 is the reciprocal of the positive sequence susceptance per unit length, α1 is the positive sequence attenuation constant, β1 is the positive sequence phase constant, is the positive sequence wave impedance.

[0041] By simplifying the equations, we can get:

[0042]

[0043] where A and B represent the real part and the imaginary part respectively. Combining γ1 = α1 + jβ1, we can get:

[0044]

[0045] Basic Algorithm Three uses α1 related to the modulus information to solve for the fault location, denoted as x3:

[0046] x3 = ln(A 2 + B 2 ) / 4α1

[0047] Basic Algorithm Four uses β1 related to the phase angle information to solve for the fault location, denoted as x4:

[0048] x4 = arctan(B / A) / 2β1

[0049] If the line parameters and electrical quantities with errors are substituted into the calculation formulas of the four basic algorithms, four positioning results with errors can be obtained: x′1, x′2, x′3, x′4.

[0050] Furthermore, in step 5, determine the implementation method and input / output of the data fusion model, and perform normalization processing on the data.

[0051] Due to the randomness of measurement errors and line parameter errors, the positioning error of the basic algorithm is random. Since the four basic algorithms use different line models (lumped parameter model / distributed parameter model), electrical quantities, and line parameters, there are redundancy and complementarity among the positioning results of the basic algorithms. Therefore, the accuracy of distribution line fault positioning can be improved by mining the information contained in the redundancy and complementarity.

[0052] Let the true value of the electrical quantity be M k (k = 1, 2,..., 8), where M1 to M8 represent and Let the relative error of the electrical quantity be ε k (k = 1, 2,..., 8); let the true value of the line parameter be P m (m = 1, 2, 3, 4), where P1 to P4 represent l, r1, x L1 and x C1 ; let the relative error of the line parameter be δ m (m = 1, 2, 3, 4).

[0053] When actually performing fault positioning, the electrical quantities and line parameters used by the basic algorithm all contain errors. The electrical quantities and line parameters with errors are denoted as M' k and P' m , and the expressions are:

[0054] M' k = ε k M k (k = 1, 2,..., 8)

[0055] P' m = δ m P m (m = 1, 2, 3, 4)

[0056] Taking the basic algorithm one as an example, the expression of the positioning result x'1 with errors is:

[0057]

[0058] The positioning error Δx1 can be obtained by subtracting the positioning result x1 without errors from the positioning result x'1 with errors:

[0059]

[0060] It can be seen from this that the positioning error of the basic algorithm one is related to the following factors: the true value of the electrical quantity and the relative errors ε1, ε2, ε3, and ε4 of the electrical quantity; the true values of the line parameters l, r1, xL1 ; The relative errors of line parameters are δ1, δ2, and δ3. Therefore, the complex relationship of the positioning error Δx1 can be expressed as a function:

[0061]

[0062] Similarly, the functional relationships of the positioning errors of the other three basic algorithms are respectively denoted as:

[0063]

[0064]

[0065]

[0066] During actual fault location, the true values of electrical quantities and their relative errors, as well as the true values of line parameters and their relative errors, are all unknowable. For example, the true value of the electrical quantity and the specific value of the relative error ε1 can only be obtained through measurement and calculation and is equal to Therefore, the independent variables should be changed to quantities that can be obtained. Taking the positioning error Δx1 of the first basic algorithm as an example, the following changes are made to its independent variables: ① Use x′1 to represent the relationship between the independent variables; ② Delete the electrical quantity-related parameters in the independent variables and retain the measured line parameters l′, x′1, x′ containing errors L1 . The following can be obtained:

[0067]

[0068] The same operations are performed on the other three basic algorithms, and the following can be obtained:

[0069]

[0070]

[0071]

[0072] Combined with Equation The following system of equations can be obtained:

[0073]

[0074] Each electrical quantity and its relative error, line parameter and its relative error are interrelated and coupled with the positioning results of the four basic algorithms, and the function f ΔiThe analytical expressions for (i = 1, 2, 3, 4) cannot be accurately obtained. Since artificial neural networks can better fit complex functional relationships, the artificial neural network is used as the implementation method of the data fusion model. As the most basic artificial neural network, the training algorithm of the Multi-Layer Perceptron (MLP) is not complex, with a short training time and high accuracy. Therefore, an artificial neural network model is constructed using "MLPRegressor" in the Python toolkit Scikit-learn.

[0075] Since the positioning results x′ of the four basic algorithms i (i = 1, 2, 3, 4) and the line parameters r′1, x′ L1 , x′ C1 can be obtained through measurement calculation or query, these 7 data can be used as inputs and the actual fault location y as the output.

[0076] Since the positioning results of the four basic algorithms and the line parameters have different magnitudes and dimensions, increasing the training difficulty, normalization should be performed on them. The normalization method is: divide the positioning results of several basic algorithms by the fault line length, and divide the line resistance, inductive reactance, and capacitive reactance by the corresponding reference values.

[0077] The expression of the data fusion model is:

[0078] x fuse* = f(x 1* , x 2* , x 3* , x 4* , r 1* , x L1* , x C1* )

[0079] Where: x fuse* = y / l′, x 1* = x′1 / l′, x 2* = x′2 / l′, x 3* = x′3 / l′, x 4* = x′4 / l′, r 1* = r′1 / r 1,base , x L1* = x′ L1 / x L1,base , x C1* = x′ C1 / x C1,base ; l′ is the fault line length, y is the actual location, x′1, x′2, x′3, and x′4 are the positioning results of the four basic algorithms respectively, r′1 is the resistance per unit length of the line, x′ L1 is the inductive reactance per unit length of the line, x′C1 is the capacitive reactance per unit length of the line, r 1,base =0.2Ω / km, x L1,base =0.2Ω / km, x C1,base =0.1MΩ·km;x fuse* 、x 1* 、x 2* 、x 3* 、x 4* 、r 1* 、x L1* and x C1* are y, x′1, x′2, x′3, x′4, r′1, x′ L1 and x′ C1 The normalized result.

[0080] After a large number of simulation samples are normalized, 1* ,x 2* ,x 3* ,x 4* ,r 1* ,x L1* ,x C1* As input (feature), x fuse* A large number of samples for output (label) constitute a training set, which can be used to train data fusion models based on artificial neural networks.

[0081] Furthermore, in step 6, the training dataset obtained in step 5 is used to train the artificial neural network-based data fusion model. First, 1% of the samples from the training dataset are randomly selected as the validation set. The remaining 99% of the samples are used to train the artificial neural network model. By observing the model's performance on the validation set, hyperparameters such as the number of hidden layers, the number of nodes in each hidden layer, the activation function, the optimizer, and the loss function are adjusted and determined, thereby determining the model's internal structure. The model is then trained once using all the training samples to determine the model's internal parameters. At this point, the internal structure and parameters of the artificial neural network-based data fusion model have been determined, and the data fusion model training is complete.

[0082] Furthermore, in step 7, after a fault occurs on an actual distribution line, the three-phase voltages and three-phase currents recorded by the voltage transformers at both ends of the faulty line are collected. The voltages and currents are subjected to fundamental frequency phasor extraction and phase sequence transformation to obtain the positive and negative sequence voltages and currents at both ends of the faulty line. The positive and negative sequence voltages and currents, as well as the line resistance, inductive reactance, and capacitive reactance, are substituted into four basic location algorithms to obtain four location results. The four normalized location results and the line parameters are then input into the data fusion model trained in step 6 to obtain the fault location result.

[0083] Compared with the prior art, the present invention has the following beneficial technical effects:

[0084] First, the method of the present invention uses PSCAD / EMTDC to establish a distribution network model, simulates and generates a large number of phase-to-phase fault samples, and constructs a database that can be used for model training. Then, error simulation, fundamental frequency extraction, and phase sequence transformation operations are performed on the fault data, and four basic positioning results are obtained by substituting them into the positioning formulas of four basic algorithms. Secondly, the present invention determines the implementation method and input / output of the data fusion model, and trains the model using the normalized data. Finally, the trained model can be used for actual fault location of distribution lines. The data fusion model of the invention takes into account the influence of measurement errors and line parameter errors on the positioning accuracy of existing positioning algorithms, has practical significance and good application prospects; the model has high positioning accuracy, reduces the maintenance time, speeds up the fault recovery speed, and improves the reliability of system operation; compared with the traveling wave method, the method of the present invention has low requirements for the sampling rate, does not require expensive measurement equipment, and has good economy; the present invention makes full use of the double-end synchronous power frequency information, is not affected by the system operation mode and transition resistance, and does not require iterative calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] The accompanying drawings of the specification are used to provide a further understanding of the present invention, and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention, and do not constitute an improper limitation of the present invention.

[0086] Figure 1 For the generation steps and usage method of the data fusion model for precise fault location of distribution lines in the present invention;

[0087] Figure 2 For the schematic diagram of the distribution network structure used when simulating and generating fault samples in the present invention;

[0088] Figure 3 For the positive sequence circuit of the lumped parameter line model;

[0089] Figure 4 For the negative sequence circuit of the lumped parameter line model;

[0090] Figure 5 For the positive sequence circuit of the distributed parameter line model;

[0091] Figure 6 For the schematic diagram of the data fusion model structure based on artificial neural network in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0092] To enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0093] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0094] The present invention is a precise fault location method for distribution lines that utilizes the data fusion of multi-algorithm location results. As Figure 1 shown, it specifically includes the following steps:

[0095] I. Figure 2 is a schematic diagram of the distribution network structure. Measuring devices are installed at both the head and the end of the line. These measuring devices can synchronously measure and upload the three-phase voltage and current signals at the installation location. The line adopts the Berliou model, the line voltage of the voltage source is 35 kV; the transformer ratio is 35 kV / 10.5 kV; the load adopts a three-phase line-to-ground fixed load (Fixed Load, 3-Phase); the fault type is an interphase fault. The simulation parameters of the training set are set as shown in Table 1. When a fault occurs on line B1M6, after the measuring device detects the mutation, it synchronously samples the three-phase voltage and three-phase current and uploads them to the processing center.

[0096] Table 1 Simulation Parameter Settings of the Training Set

[0097]

[0098] Second, for each fault sample, the three-phase voltage and current sampling values and line parameters are multiplied by random coefficients to simulate measurement errors and line parameter errors. The random coefficients are generated by the AWGN function in MATLAB. The random coefficients are generated with the help of the Additive White Gaussian Noise (AWGN) function in MATLAB. The AWGN function is a basic noise and interference model. Its amplitude distribution follows a Gaussian distribution, while the power spectrum density is uniformly distributed. The signal-to-noise ratio is 35dB, and 10,000 random numbers are generated using the AWGN function. The maximum value of these 10,000 random numbers is 1.0635, the minimum value is 0.9335, the average value is 1.0001, and the average absolute error is 1.40%. This basically conforms to the distribution of measurement errors and line parameter errors.

[0099] 3. After the error simulation, the sampling values of the three-phase voltage and current in the second power frequency cycle after the fault are selected, and the three-phase voltage and current are fast Fourier transformed to obtain the three-phase voltage and current phasors. Then, the symmetrical component method is used to transform the phase sequence of the voltage and current phasors to obtain the positive and negative sequence voltages and currents at both ends of the fault line. The beginning and end of the fault line are marked as M and N respectively. At this point, the electrical quantity containing the error is obtained. and And the line length l' containing errors, the positive sequence resistance per unit length r'1, the positive sequence inductive reactance per unit length x' L1 and the positive sequence capacitive reactance to ground x′ C1 .

[0100] Fourth, taking into full consideration the development trends and fault characteristics of the distribution network, four basic algorithms were selected. These four basic algorithms are the positive sequence voltage and current method and the positive and negative sequence impedance equality method based on the lumped parameter line model, and the modulus value solution method and phase solution method based on the distributed parameter line model. They are denoted as Basic Algorithm 1, Basic Algorithm 2, Basic Algorithm 3, and Basic Algorithm 4, respectively.

[0101] When calculating the positioning results of the four basic algorithms below, the electrical quantities and line parameters used all contain errors, so the parameters involved in the formula are marked with "'".

[0102] The first basic algorithm is the positive sequence voltage and current method based on the lumped parameter line model. Figure 3 is the positive sequence circuit of the lumped parameter line model, and the positive sequence voltage at the fault point is The positive sequence voltage at the first end of the fault line can be used and positive sequence current Calculated, or the positive sequence voltage at the end of the fault line can be used and positive sequence current The calculation results show that the equation group is:

[0103]

[0104] Where l' is the length of the fault line, x is the distance between the fault point and the line head end, z'1 is the positive sequence impedance per unit length of the line, z'1 = r'1 + jx' L1 , r′1 is the positive sequence resistance per unit length of the line, x′ L1 is the positive sequence inductive reactance per unit length of the line.

[0105] Solving this set of equations yields the positioning result of basic algorithm 1, which is denoted as x′1 for easy distinction:

[0106]

[0107] The second basic algorithm is the positive and negative sequence impedance equality method based on the lumped parameter line model. Figure 4 The negative sequence circuit of the lumped parameter line model is used to model the negative sequence voltage at the head end of the fault line. and negative sequence current The negative sequence voltage at the fault point can be calculated Using the negative sequence voltage at the end of the fault line and negative sequence current The negative sequence voltage at the fault can also be calculated The system of equations is:

[0108]

[0109] Where z′2 is the negative sequence impedance per unit length of the line.

[0110] Since the distribution line is a three-phase static element, the positive and negative sequence impedances are equal:

[0111] z′1=z′2

[0112] The combined equations result in a system containing 5 equations and 5 unknown quantities (x, The system of equations for z′1 and z′2):

[0113]

[0114] Solving this set of equations yields the positioning result of basic algorithm 2, denoted as x′2:

[0115]

[0116] Basic Algorithm 3 and Basic Algorithm 4 use distributed parameter line models. Figure 5 is the positive sequence circuit of the distributed parameter line model, and the positive sequence voltage at the fault point is It can be determined by the positive sequence voltage at the first end of the fault line and positive sequence current It can also be calculated from the positive sequence voltage at the end of the fault line and the positive-sequence current It is calculated that the following system of equations is written:

[0117]

[0118] where is the positive-sequence propagation coefficient, y′1 is the positive-sequence admittance of the line to the ground, and y′1 = j / x′ C1 , x′ C1 is the reciprocal of the positive-sequence susceptance per unit length, α′1 is the positive-sequence attenuation constant, β1′ is the positive-sequence phase constant, is the positive-sequence wave impedance.

[0119] Simplifying the system of equations, we can get:

[0120]

[0121] where A′ and B′ represent the real part and the imaginary part respectively. Combining γ′1 = α′1 + jβ′1, we can get:

[0122]

[0123] The basic algorithm three uses α′1 related to the modulus information to solve for the fault location, denoted as x′3:

[0124] x′3 = ln(A′ 2 + B′ 2 ) / 4α′1

[0125] The basic algorithm four uses β′1 related to the phase angle information to solve for the fault location, denoted as x′4:

[0126] x′4 = arctan(B′ / A′) / 2β′1

[0127] Thus, the location results x′1, x′2, x′3, and x′4 of the four algorithms are obtained.

[0128] V. Determine the implementation method, input and output of the data fusion model, and normalize the data.

[0129] Each electrical quantity, relative error, line parameter, and relative error are interrelated and coupled with the location results of the four basic algorithms, and the functional relationship between them is complex and difficult to obtain. Since the artificial neural network can better fit complex functional relationships, the artificial neural network is used as the implementation method of the data fusion model. As the most basic artificial neural network model, the training algorithm of the multi-layer perceptron is not complex, and the training time is short and the accuracy is high. Therefore, the "MLPRegressor" in the Python toolkit Scikit-learn is used to construct the artificial neural network.

[0130] Due to the positioning results x′ of the four basic algorithms i (i = 1, 2, 3, 4), and the line parameters r′1, x′ L1 , x′ C1 which can be obtained through measurement calculation or query, these 7 data can be used as inputs. As a model for precise fault location of distribution lines, the actual fault location y is naturally used as the output. The data fusion model structure based on artificial neural network is as Figure 6 shown.

[0131] Since the positioning results of the four basic algorithms and the line parameters have different magnitudes and dimensions, it increases the training difficulty and should be normalized. Specifically, divide the positioning results of the four algorithms by the line length l′, x 1* = x′1 / l′, x 2* = x′2 / l′, x 3* = x′3 / l′, x 4* = x′4 / l′; divide the line parameters by the corresponding reference values, r 1* = r′1 / r 1,base , x L1* = x′ L1 / x L1,base , x C1* = x′ C1 / x C1,base , and r 1,base = 0.2Ω / km, x L1,base = 0.2Ω / km, x C1,base = 0.1MΩ·km; divide the actual fault location by the line length, x fuse* = y / l′. After normalization, a large number of samples are formed with x 1* , x 2* , x 3* , x 4* , r 1* , x L1* , x C1* as inputs (features) and x fuse* as the output (label). The training set composed of a large number of samples can be used for the training of the data fusion model based on artificial neural network.

[0132] VI. Use a large number of normalized samples to train a data fusion model based on an artificial neural network (ANN). First, randomly select 1% of the samples from the training set as the validation set. By observing the test effect of the model on the validation set, adjust the number of hidden layers and the number of neurons in each layer of the artificial neural network. Finally, it is determined that the number of hidden layers is 3, the number of nodes in each hidden layer is 15, 12, and 7 respectively, the activation function is Tanh, the optimizer is Adam, and the loss function is the mean squared error. Then, use all the test set samples to train the artificial neural network. Finally, the internal parameters of the data fusion model based on the artificial neural network are obtained, and this model can be used for accurate fault location of distribution lines. Computer configuration: Processor Intel(R) Core(TM) i5-7500 CPU @ 3.4GHz, 3480MHz Physical Memory (RAM) 8.00GB.

[0133] VII. When a new fault occurs in the distribution line, similar to steps I, III, IV, and V, record the three-phase voltages and currents at both ends of the faulty line, extract the fundamental frequency phasors and perform phase sequence transformation, calculate the positioning results of the four basic algorithms, and normalize the line parameters and the positioning results of the four basic algorithms. Then, input the normalized data into the data fusion model to obtain the fused positioning result.

[0134] Using the control variable method, test samples with differences in factors such as fault location, line parameters, transition resistance, fault starting angle, line load, line length, noise intensity, distribution network grounding method, and fault type compared with the training set samples are generated by PSCAD / EMTDC, and the positioning effects of the basic algorithms and the obtained data fusion model are compared.

[0135] The line parameters in Table 2 can be selected during testing, and these line parameters are different from the training set. The default values of other factors of the test samples are listed in Table 3. If a certain factor is different from the default setting during testing, this factor will be listed separately.

[0136] Table 2 Selectable line parameters of the test set

[0137]

[0138]

[0139] Table 3 Default settings of test samples

[0140]

[0141] The positioning effect is measured by the mean absolute error (MAE) and the mean relative error (MRE), and the calculation methods are as follows:

[0142]

[0143]

[0144] Among them, N represents the total number of test samples, and x′ j represents the positioning result of the algorithm on the j-th test sample, and y j represents the actual fault location on the j-th test sample, and l represents the actual total length of the line.

[0145] By analyzing the test results in Tables 4 to 10, it can be found that, compared with the basic algorithm, the positioning errors output by the data fusion model of the invention are significantly reduced. This shows that the data fusion model makes full use of the complementarity of the positioning results of multiple basic algorithms and can output more accurate positioning results. Moreover, the data fusion model of the invention is not affected by factors such as fault location, line parameters, transition resistance, and fault starting angle, and has strong adaptability.

[0146] Table 4 Test Results for Different Fault Locations

[0147]

[0148] Table 5 Test Results for Different Line Parameters and Transition Resistances

[0149]

[0150]

[0151] Table 6 Test Results for Different Fault Starting Angles

[0152]

[0153] Table 7 Test Results for Different Line Loads

[0154]

[0155] Table 8 Test Results for Different Line Lengths

[0156]

[0157] Table 9 Test Results for Different Noise Intensities

[0158]

[0159] Table 10 Test Results for Different Grounding Methods and Fault Types

[0160]

[0161]

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than limit the scope of its protection. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that after reading the present invention, various changes, modifications or equivalent replacements can still be made to the specific implementation manners of the invention. However, these changes, modifications or equivalent replacements are all within the scope of protection of the pending claims of the invention.

Claims

1. A method for fault location of distribution lines using data fusion of multi-algorithm location results, characterized in that, Including: Step 1: Establish an electromagnetic transient simulation model of the distribution network, set multiple fault scenarios and conduct simulations, synchronously record the three-phase voltages and currents at both ends of the fault line, and form several fault samples. Step 2: For each fault sample, simulate the measurement errors of the three-phase voltages and currents and the line parameter errors to obtain the three-phase voltages and currents with measurement errors, and the line parameters with errors. Step 3: Extract the fundamental frequency phasors from the three-phase voltages and currents with measurement errors, and then perform phase sequence transformation on the fundamental frequency phasors to obtain the positive and negative sequence voltage and current at both ends of the fault line. Step 4: Combine the development trend and fault characteristics of the distribution network, select several precise location algorithms as the basic algorithms, and then substitute the line parameters with errors and the positive and negative sequence voltage and current into the basic algorithms to calculate several location results. Step 5: Use several location results and the line resistance, inductive reactance, and capacitive reactance as the inputs of the data fusion model, and use the fault location as the output of the data fusion model; perform normalization processing on the inputs and outputs. A large number of normalized data form the training data set. Step 6: Use the training data set obtained in Step 5 to train the data fusion model to determine the internal structure and parameters of the data fusion model. After a fault occurs in the actual distribution line, collect the three-phase voltages at both ends of the fault line and the three-phase currents recorded by the current transformers, extract the fundamental frequency phasors and perform phase sequence transformation on the voltages and currents to obtain the positive and negative sequence voltage and current at both ends of the fault line. Substitute the positive and negative sequence voltage and current and the line parameters into several basic location algorithms to obtain several location results, and input the normalized several location results and the line resistance, inductive reactance, and capacitive reactance into the data fusion model trained in Step 6 to obtain the fault location result.

2. The distribution line fault location method using multi-algorithm location result data fusion according to claim 1, characterized in that When setting multiple fault scenarios in Step 1, different line parameters, line types, fault locations, transition resistances, and fault starting angles are considered and set.

3. The distribution line fault location method using multi-algorithm location result data fusion according to claim 1, characterized in that, When simulating the measurement errors and line parameter errors in Step 2, the additive Gaussian white noise function is adopted.

4. The method for locating faults in a distribution line by fusing multi-algorithm location result data according to claim 1, characterized in that, The four basic algorithms selected in Step 4 are the positive sequence voltage and current method and the positive and negative sequence impedance equality method based on the lumped parameter line model, the modulus solution method and the phase solution method based on the distributed parameter line model. Substitute the line parameters with errors and the positive and negative sequence voltage and current into the four basic algorithms to obtain four location results with errors \(x'_1\), \(x'_2\), \(x'_3\), \(x'_4\) correspondingly.

5. The method for locating faults in a distribution line by fusing multi-algorithm location result data according to claim 4, characterized in that The method for normalizing the inputs and outputs in Step 5 is: divide the location results of several basic algorithms by the length of the fault line, and divide the line resistance, inductive reactance, and capacitive reactance by the corresponding reference values. After determining the inputs and outputs and normalizing the inputs and outputs, the expression of the data fusion model is obtained as: x fuse* = f(x 1* , x 2* , x 3* , x 4* , r 1* , x L1* , x C1* ) where: x fuse* = y / l′, x 1* = x′1 / l′, x 2* = x′2 / l′, x 3* = x′3 / l′, x 4* = x′4 / l′, r 1* = r′1 / r 1,base , x L1* = x′ L1 / x L1,base , x C1* = x′ C1 / x C1,base ; l′ is the length of the faulty line, y is the actual fault location, x′1, x′2, x′3, and x′4 are the location results of four basic algorithms respectively, r1′ is the resistance per unit length of the line, x′ L1 is the inductive reactance per unit length of the line, x′ C1 is the capacitive reactance per unit length of the line, x fuse* , x 1* , x 2* , x 3* , x 4* , r 1* , x L1* and x C1* are the normalized values of y, x′1, x′2, x′3, x′4, r1′, x′ L1 and x′ C1 respectively.

6. The method for locating faults in a distribution line by using data fusion of multi-algorithm location results according to claim 5, characterized in that r 1,base = 0.2 Ω / km, x L1,base = 0.2 Ω / km, x C1,base = 0.1 MΩ·km.

7. The method for fault location of distribution lines using data fusion of multi-algorithm location results according to claim 5, characterized in that The data fusion model described in step 5 uses an artificial neural network, and the training data set consists of a large number of samples with x 1* , x 2* , x 3* , x 4* , r 1* , x L1* , x C1* as the input and x fuse* as the output.

8. The method for fault location of distribution lines using data fusion of multi-algorithm location results according to claim 7, characterized in that When training the data fusion model in Step 6, first randomly select 1% of the samples from the training dataset as the validation set, and use the remaining 99% of the samples to train the artificial neural network. By observing and comparing the performance of the artificial neural network on the validation set, hyperparameters such as the number of hidden layers and the number of neurons in each layer of the artificial neural network are adjusted to obtain the structure of the artificial neural network. Then, use all the data in the training set to train the artificial neural network with the determined structure once to obtain the internal parameters of the artificial neural network. At this point, both the internal structure and parameters of the artificial neural network have been obtained, that is, the training of the data fusion model is completed.

Images