Asynchronous fault location method for two-terminal transmission lines based on artificial neural network and network migration

Abstract

The present invention is a two-terminal transmission line asynchronous fault ranging method based on artificial neural networks and network migration, including S1, fault ranging analysis; S2, ANN1 design and training; S3, ANN2 input calculation; S4, ANN2 design and training; S5, fault ranging. The present invention uses the positive sequence voltage and positive sequence current sampled asynchronously at both ends during normal operation and short circuit faults of the transmission line to achieve two-terminal fault ranging. The ranging accuracy is not affected by factors such as fault location, fault type, transition resistance, fault initial phase angle, load current, and asynchronous data, and the ranging accuracy is high. When performing fault ranging, it does not rely on line models and line parameters and is not affected by line modeling and parameter errors. It still has high ranging accuracy when using a small amount of fault data to train the network. Based on the concept of network migration, normal operation data can be used to update line parameter information in the network in real time. When the line parameter changes do not exceed ±1%, high ranging accuracy can still be achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system relay protection, and in particular relates to an asynchronous fault location method for a double-terminal transmission line based on artificial neural network and network migration. Background Art

[0002] High-voltage transmission lines often experience short-circuit faults due to factors such as lightning, storms, vandalism, vegetation, insulation breakdown, and damage to towers or other line components. If a permanent fault occurs, the line must be inspected and repaired before power can be restored. To minimize the economic losses caused by power outages, transmission lines must be restored to meet load demand as quickly as possible. Accurate fault location allows maintenance personnel to quickly reach the fault location for inspection and repair, thereby minimizing outage duration. Therefore, transmission line fault location algorithms have long been a key research topic in the power system field.

[0003] In principle, fault location methods for transmission lines can be categorized as traveling wave methods and fault analysis methods. The traveling wave method faces wave head identification challenges during the distance measurement process and requires specialized high-speed sampling equipment, making its implementation relatively complex. Compared to the traveling wave method, the fault analysis method has been widely used in practical projects due to its advantages such as lower equipment requirements and lower investment. However, in actual operation, line parameters may deviate from the parameters set at the time of commissioning due to factors such as ambient temperature, climate conditions, geological conditions, and operating time. Furthermore, inadequate parameter management and incomplete parameter records make it difficult to accurately obtain line distribution parameters in real time. The fault analysis method's reliance on the accuracy of line parameters can lead to significant errors in distance measurement results. While fault location algorithms based on line parameter identification exist, these algorithms all rely on establishing line parameter equations based on a specific line model to achieve fault location. Inaccurate line modeling can lead to significant distance measurement errors.

[0004] With the widespread application of artificial intelligence algorithms in power systems, neural networks can be used to mine line fault characteristics, thereby proposing a method for fault location in transmission lines that does not require line models or line parameters. However, existing neural network-based fault location algorithms often use transient signals after a line fault, which places high demands on sampling equipment and requires a large amount of historical fault data to learn the mapping relationship between electrical characteristics and fault distances. However, in actual operation, such a large amount of historical fault data is difficult to obtain. Summary of the Invention

[0005] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a two-terminal transmission line asynchronous fault location method based on artificial neural networks and network migration. The two-terminal fault location is realized by using the positive sequence voltage and positive sequence current sampled asynchronously at both ends during normal operation and short circuit faults of the transmission line. The ranging accuracy is not affected by factors such as fault location, fault type, transition resistance, fault initial phase angle, load current and asynchronous data, and has high ranging accuracy.

[0006] The present invention solves the technical problem by the following technical solutions:

[0007] A method for asynchronous fault location of a double-terminal transmission line based on artificial neural network and network migration is characterized in that the method comprises the following steps:

[0008] S1. Fault distance analysis

[0009] When a double-terminal transmission line fails,

[0010]

[0011] in: and They are respectively the positive sequence voltage and positive sequence current at the M terminal and the positive sequence voltage and positive sequence current at the N terminal after the line fault;

[0012] A is a second-order square matrix, whose value depends only on the line parameters and δ under different line models, and is independent of factors such as fault type and fault location.

[0013] After fault analysis and formula derivation, the ranging function is obtained:

[0014]

[0015] Where: f(x) is an expression including line parameters and fault location;

[0016] When a double-terminal transmission line operates normally, the voltage and current at both ends of the line have the following relationship:

[0017]

[0018] in: and They are respectively the positive sequence voltage and current at the M terminal and the positive sequence voltage and current at the N terminal when the line is in normal operation;

[0019] S2. Design and training of normal network ANN1

[0020] (1) ANN1 is a three-layer feedforward neural network structure. The number of neurons in the input layer and the output layer are both 4. Considering the output error and overfitting phenomenon of the neural network, the number of neurons in the hidden layer is selected as 10, and the activation function is a linear function.

[0021] (2) A large amount of normal operation data of the line is collected, and the real and imaginary parts of the positive sequence voltage and positive sequence current at the M end of the line are used as the input of ANN1, and the real and imaginary parts of the positive sequence voltage and positive sequence current at the N end of the line are used as the output of ANN1. ANN1 is trained to fit the mapping relationship between the positive sequence voltage and positive sequence current at the M end and the positive sequence voltage and positive sequence current at the N end in equation (3), and the mapping relationship is stored in ANN1 in the form of connection weights and thresholds between neurons;

[0022] S3. Input calculation of fault location network ANN2

[0023] Collect data when a line fault occurs. As shown in formula (1), the positive sequence voltage and positive sequence current at the line M end are used as the input of the trained ANN1, and its output is subtracted from the positive sequence voltage and positive sequence current at the line N end to obtain the input of ANN2.

[0024] S4. Design and training of fault location network ANN2

[0025] (1) The number of neurons in the input layer and output layer of ANN2 is 4 and 1 respectively. The network parameters of ANN1 are transferred to ANN2 and fixed. A hidden layer is added to form the fault location network ANN2. Considering the output error and overfitting of the neural network, the number of neurons in the new hidden layer is selected as 10, and the activation function is selected as the tansig function.

[0026] (2) Using an artificial neural network to fit the functional relationship in formula (2), the calculation result in step S3 is used as the input of ANN2, the corresponding fault location is used as the output of ANN2, ANN2 is trained, and the remaining network parameters of ANN2 are adjusted;

[0027] S5. Fault distance measurement

[0028] After a line fault occurs, the positive-sequence voltage and positive-sequence current at both ends of the line are collected. The data at the M end of the line is used as the input of ANN1, and its output is subtracted from the data at the N end of the line to obtain the input of ANN2. The corresponding ANN2 output result is the fault location of the line.

[0029] The advantages and beneficial effects of the present invention are:

[0030] 1. The present invention is based on an asynchronous fault location method for two-terminal transmission lines using artificial neural networks and network migration. Based on the trained ANN1 and ANN2, accurate fault location can be achieved by utilizing the positive-sequence voltage and positive-sequence current sampled asynchronously at both ends after a transmission line fault. The ranging accuracy is not affected by factors such as the fault location, fault type, transition resistance, initial fault phase angle, load current, and asynchronous data, and has high ranging accuracy.

[0031] 2. The present invention's asynchronous fault location method for two-terminal transmission lines, based on artificial neural networks and network migration, does not rely on line models and line parameters when performing fault location, and is therefore unaffected by line modeling and parameter errors. It also achieves high ranging accuracy even when training the network with a small amount of fault data, consistent with the reality that historical fault data is relatively small during line operation. Based on the concept of network migration, normal operation data can be used to update line parameter information in the network in real time, achieving high ranging accuracy when line parameter changes do not exceed ±1%. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Schematic diagram of a double-terminal power transmission line fault system applicable to the present invention;

[0033] Figure 2 Schematic diagram of the structure of the artificial neural network used for fault location in the present invention;

[0034] Figure 3 Flowchart of the present invention. DETAILED DESCRIPTION

[0035] The present invention will be further described in detail below through specific examples. The following examples are only illustrative and not restrictive, and the scope of protection of the present invention cannot be limited thereto.

[0036] Figure 1 The schematic diagram of the system when the double-terminal transmission line of the present invention is faulty is shown in FIG. The distance between the fault point f and the M terminal is x, the total length of the line is l, and the voltage and current at each point of the line are as follows: Figure 1 The present invention provides an asynchronous fault location method for a double-terminal transmission line based on an artificial neural network and network migration, which mainly includes two parts: fault location principle analysis and fault location algorithm implementation.

[0037] 1. Analysis of fault distance measurement principle

[0038] The distance measurement principle of the double-terminal transmission line is analyzed based on the distributed parameter model and the π-type line model, where the positive sequence wave impedance and positive sequence propagation constant of the line in the distributed parameter model are Z c In the γ- and Π-type line models, the positive-sequence impedance and positive-sequence admittance of the line are Z and Y respectively, and a detailed description is mainly given by taking the distributed parameter model as an example.

[0039] The present invention takes into account the asynchrony of the two-end data sampling. Taking the time of collecting data at the N end as a reference, the asynchronous angle of the data collected at the M end relative to the data collected at the N end is δ, that is, when considering the asynchrony of the data collected at both ends, if the positive sequence voltage and positive sequence current of the N end are respectively and Then the positive sequence voltage and positive sequence current at terminal M are and

[0040] When a fault occurs at point f, according to the uniform transmission line equation, the positive sequence voltage at terminal M is and positive sequence current Calculate the positive sequence voltage at the fault point and the positive sequence current flowing from the M side to the fault point for:

[0041]

[0042] The positive sequence voltage at the fault point and the positive sequence current flowing from the fault point to the N side Calculate the positive sequence voltage at the N terminal and positive sequence current for:

[0043]

[0044] according to in is the positive sequence current flowing into the fault point, and it can be obtained from formula (5):

[0045]

[0046] Substituting formula (4) into formula (6) yields:

[0047]

[0048] Where:

[0049]

[0050] make

[0051]

[0052] Simplifying formula (7) we can get:

[0053]

[0054] Divide the two sides of the equation (10) up and down to eliminate the unknown quantity Then, after equation transformation, the ranging equation can be obtained as follows:

[0055]

[0056] When the line operates normally, the positive sequence voltage and current at both ends of the line have the following relationship:

[0057]

[0058] Similarly, based on the π-type line model, the distance measurement principle of the double-terminal transmission line is analyzed. When the line is operating normally, the positive sequence voltage and current at both ends of the line have the following relationship:

[0059]

[0060] Where:

[0061] When a line fault occurs, the ranging equation can be obtained as:

[0062]

[0063] Where:

[0064] Summarizing the above fault location principle, let:

[0065]

[0066] Where, and are the positive sequence voltage and current at the M terminal and the positive sequence current at the N terminal after the line fault respectively; A is a second-order square matrix, whose value depends on the line parameters and δ under different line models, and is independent of factors such as the fault type and fault location.

[0067] When a fault occurs on a double-terminal transmission line, equations (11) and (14) can be uniformly transformed into the following ranging function:

[0068]

[0069] Where f(x) is an expression that includes line parameters and fault location, and f(x) varies for different line models.

[0070] When the double-terminal transmission line operates normally, equations (12) and (13) can be unified into the following expression:

[0071]

[0072] Where, and They are respectively the positive sequence voltage and current at the M terminal and the positive sequence voltage and current at the N terminal when the line is operating normally.

[0073] From Equation (17), we can see that the matrix A can be obtained using the positive-sequence voltage and current at both ends of the transmission line during normal operation. Combining Equations (15) and (16), we can see that the fault location of a two-terminal transmission line can be achieved using the positive-sequence voltage and current at both ends after the fault, as well as the matrix A. However, the ranging function f(x) varies for different line models. To eliminate the influence of line models and line parameters on the ranging results, this paper proposes a data-driven transmission line fault location algorithm using the function fitting capability of artificial neural networks.

[0074] 2. Fault distance measurement

[0075] Based on the above-mentioned distance measurement principle analysis, the present invention aims to utilize a large amount of normal operation data and a small amount of historical fault data at both ends of a two-terminal transmission line, and fit equations (16) and (17) through artificial neural networks to establish a fault distance measurement model for the two-terminal transmission line.

[0076] With the real and imaginary parts of the positive sequence voltage and current at the M terminal as input and the real and imaginary parts of the positive sequence voltage and current at the N terminal as output, a normal network ANN1 is formed to fit the linear mapping relationship contained in the matrix A in equation (17). This mapping relationship only depends on the line parameters and the asynchronous angle δ. Combining equations (15) and (16), it can be seen that the distance function can be obtained by using ANN1 and And the matrix A plays a role in the ranging function again. Therefore, ANN1 can be migrated to the new neural network, and a hidden layer is added after it to form the fault ranging network ANN2, which is used to fit the ranging function in formula (23), that is, - and The real and imaginary parts of are taken as input, and the corresponding fault location is taken as output.

[0077] The network structure and migration relationship between ANN1 and ANN2 are as follows Figure 2 As shown, the overall process of the double-terminal transmission line fault location algorithm of the present invention is as follows Figure 3 The specific steps are as follows:

[0078] (1) Design and training of ANN1

[0079] like Figure 2 As shown, ANN1 is a three-layer feedforward neural network structure. The number of input layer neurons and output layer neurons is 4. Taking into account the output error and overfitting phenomenon of the neural network, the number of hidden layer neurons is selected as 10, and the activation function is a linear function.

[0080] A large amount of normal line operation data is collected, with the real and imaginary parts of the positive-sequence voltage and current at the M terminal serving as input to ANN1, and the real and imaginary parts of the positive-sequence voltage and current at the N terminal serving as output. ANN1 is trained to fit the mapping relationship between the positive-sequence voltage and current at the M terminal and the positive-sequence voltage and current at the N terminal, and this mapping relationship is stored in ANN1 in the form of connection weights and thresholds between neurons.

[0081] (2) ANN2 input calculation

[0082] The data when the line fails is collected, and the positive sequence voltage and positive sequence current at the M terminal are used as the input of the trained ANN1. Its output is subtracted from the positive sequence voltage and positive sequence current at the N terminal to obtain the input of ANN2.

[0083] (3) Design and training of ANN2

[0084] like Figure 2 As shown, the number of neurons in the input layer and output layer of ANN2 are 4 and 1 respectively. The network parameters of ANN1 are migrated to ANN2 and fixed. Taking into account the output error and overfitting phenomenon of the neural network, the number of neurons in the new hidden layer is selected as 10, and the activation function is the tansig function.

[0085] The calculation result in step (2) is used as the input of ANN2, and the corresponding fault location is used as the output of ANN2. Since network migration provides some information required for the ranging function, only a small amount of fault data is needed to adjust the remaining network parameters of ANN2 to fit the ranging function.

[0086] (4) Fault distance measurement

[0087] When a line fault occurs, the positive sequence voltage and positive sequence current at both ends of the line are collected, the M-end data is used as the input of ANN1, and its output is calculated with the N-end data to obtain the input of ANN2. The corresponding ANN2 output result is the fault location of the line.

[0088] The present invention's two-terminal transmission line fault location method based on artificial neural networks and network migration learns the relationship between the line's positive-sequence voltage and current and the fault location from a large amount of normal operating data and a small amount of historical fault data, without having to establish a complex ranging function expression based on a specific line model and accurate line parameters in advance. Therefore, the present invention's fault location method is not affected by line modeling and parameter errors. When performing fault location, the present invention first uses normal operating data to adjust the network parameters of ANN1, and then migrates the network parameters to ANN2. The fault data is then input into ANN1 and ANN2 to obtain the ranging results. Therefore, when the line parameters change, the adjustment of ANN1 and ANN2 using normal operating data can reduce the fault location error, and based on the concept of network migration, the number of fault samples required for network training can also be reduced.

[0089] Although the embodiments and drawings of the present invention are disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, changes and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A method for asynchronous fault location of a two-terminal transmission line based on artificial neural network and network migration, characterized by: The steps of the method are: S1. Fault distance analysis When a double-terminal transmission line fails, in: and They are respectively the positive sequence voltage and positive sequence current at the M terminal and the positive sequence voltage and positive sequence current at the N terminal after the line fault; A is a second-order square matrix, whose value depends only on the line parameters and δ under different line models. The asynchronous angle of the data collected at the M end relative to the data collected at the N end is δ, which is independent of the fault type and fault location. After fault analysis and formula derivation, the ranging function is obtained: Where: f(x) is an expression including line parameters and fault location; When a double-terminal transmission line operates normally, the voltage and current at both ends of the line have the following relationship: in: and They are respectively the positive sequence voltage and current at the M terminal and the positive sequence voltage and current at the N terminal when the line is in normal operation; S2. Design and training of normal network ANN1 (1) ANN1 is a three-layer feedforward neural network structure. The number of neurons in the input layer and the output layer are both 4. Considering the output error and overfitting phenomenon of the neural network, the number of neurons in the hidden layer is selected as 10, and the activation function is a linear function. (2) A large amount of normal operation data of the line is collected, and the real and imaginary parts of the positive sequence voltage and positive sequence current at the M end of the line are used as the input of ANN1, and the real and imaginary parts of the positive sequence voltage and positive sequence current at the N end of the line are used as the output of ANN1. ANN1 is trained to fit the mapping relationship between the positive sequence voltage and positive sequence current at the M end and the positive sequence voltage and positive sequence current at the N end in equation (3), and the mapping relationship is stored in ANN1 in the form of connection weights and thresholds between neurons; S3. Input calculation of fault location network ANN2 Collect data when a line fault occurs. As shown in formula (1), the positive sequence voltage and positive sequence current at the line M end are used as the input of the trained ANN1, and its output is subtracted from the positive sequence voltage and positive sequence current at the line N end to obtain the input of ANN2. S4. Design and training of fault location network ANN2 (1) The number of neurons in the input layer and output layer of ANN2 is 4 and 1 respectively. The network parameters of ANN1 are transferred to ANN2 and fixed. A hidden layer is added to form the fault location network ANN2. Considering the output error and overfitting of the neural network, the number of neurons in the new hidden layer is selected as 10, and the activation function is selected as the tansig function. (2) Using an artificial neural network to fit the functional relationship in formula (2), the calculation result in step S3 is used as the input of ANN2, the corresponding fault location is used as the output of ANN2, ANN2 is trained, and the remaining network parameters of ANN2 are adjusted; S5. Fault distance measurement After a line fault occurs, the positive-sequence voltage and positive-sequence current at both ends of the line are collected. The data at the M end of the line is used as the input of ANN1, and its output is subtracted from the data at the N end of the line to obtain the input of ANN2. The corresponding ANN2 output result is the fault location of the line.

Images