Intelligent station secondary system fault location method based on optimized BP neural network

Abstract

The application discloses a kind of intelligent station secondary system fault location methods based on optimization BP neural network, the method is: the classification of intelligent station secondary system fault;With the way of experiment simulates the fault of intelligent station secondary system, establishes fault database;Fault data are collected, preprocessed;To distinguish fault type, design intelligent station secondary system fault location matrix rule, the rule will be combined with binary code, through the generation of hexadecimal fault location coding of binary conversion, form corresponding fault location table;For fault location data, realize data dimension reduction using low variance filtering method and principal component analysis method, introduce dynamic self-adaptive inertia weight factor to improve particle swarm optimization ability, and then optimize the fault location training model based on BP neural network, obtain complete intelligent positioning method.The application can help intelligent station operation and maintenance personnel to correctly locate fault, shorten intelligent station secondary system fault location time, greatly improve work efficiency.

Description

Technical Field

[0001] This invention relates to the field of fault location in intelligent substation secondary systems, and specifically to a fault location method for intelligent substation secondary systems based on an optimized BP neural network. Background Technology

[0002] With the advancement of the State Grid's construction, smart substations have gradually replaced conventional substations. In smart substations, digital communication transmission modes, represented by Ethernet and fiber optics, have completely replaced traditional cable-based wiring circuits. The networking of communication platforms, the intelligence of digital information, and the standardization of information sharing have ensured that the various subsystems of the entire station are no longer isolated information islands, but interconnected, and the types of data that can be collected are becoming increasingly rich. Currently, research on primary equipment maintenance is relatively mature, while condition-based maintenance of secondary equipment started relatively late and its development speed is also slower. Therefore, the formulation of a work plan for fault diagnosis of secondary equipment in smart substations and the establishment of fault simulation tests are becoming increasingly urgent. With the rise of artificial intelligence technology and the rapid development of microprocessor computing power, more and more intelligent algorithms are being applied to the field of fault diagnosis, greatly improving the speed and accuracy of diagnosis, and providing a more in-depth, detailed, and comprehensive research direction and solution for the development of secondary system fault diagnosis.

[0003] Currently, troubleshooting for intelligent substations still primarily relies on manual fault location. When a secondary circuit fails, maintenance personnel meticulously check all communication links based on their communication status, or spend considerable time and effort analyzing massive amounts of alarm information to determine the cause of the fault. This tedious and inefficient process takes 2-4 hours to resolve. Furthermore, as the scale of intelligent substations expands, the volume of real-time updated data increases significantly, leading to the omission of crucial information and further complicating fault location. While the introduction of artificial intelligence can enable rapid processing of fault data, the numerous types of faults in the secondary systems of intelligent substations, the limited criteria for judgment, the weak correlation between fault characteristic values, the complex connections within the secondary systems, and the strong subjectivity in the collection and selection of fault characteristic information ultimately result in low accuracy in fault location. Summary of the Invention

[0004] The purpose of this invention is to provide a fault location method for the secondary system of an intelligent substation based on an optimized BP neural network, so as to achieve efficient and accurate fault location for the secondary system of the intelligent substation.

[0005] The technical solution to achieve the purpose of this invention is: a fault location method for a secondary system of an intelligent substation based on an optimized BP neural network, comprising the following steps:

[0006] Step 1: Collect historical data on secondary system faults of the intelligent substation, analyze the fault location and causes, and classify the secondary system faults of the intelligent substation.

[0007] Step 2: Set up a fault simulation experiment for the secondary system of the intelligent station, simulate the fault types identified in Step 1, collect data from the fault simulation experiment, and integrate the fault data to form a fault database.

[0008] Step 3: Use binary encoding to number the experiments in Step 2. This encoding rule can combine the fault type and corresponding fault point of the intelligent station secondary system with the binary code to realize the correspondence between the binary code and the fault point. Then, through base conversion, hexadecimal fault location code is generated to form the corresponding fault location table. This table is then combined with the fault location database in Step 2 to form a complete fault location database with fault type encoding.

[0009] Step 4: The complete fault location database is filtered by low variance filtering. The filtered data is then processed by principal component analysis to reduce the dimensionality of the data. The processed data is then imported into the BP neural network intelligent location algorithm based on improved particle swarm optimization. The hexadecimal fault location code is used as the output result to complete the learning and training of the fault database.

[0010] Step 5: Import the fault data of the intelligent substation secondary system into the BP neural network intelligent positioning algorithm in Step 4, output the hexadecimal fault location result, compare the output result with the fault location table, find the actual fault point, and complete the fault elimination work of the intelligent substation secondary system.

[0011] Compared with the prior art, the significant advantages of this invention are: (1) This fault location method and fault classification method improve the fault analysis of the secondary system of the intelligent station, which is conducive to the rapid analysis and location of on-site accidents; (2) A set of fault location matrix rules for the secondary system of the intelligent station is proposed. The intelligent algorithm is combined with the fault location table formed by the rules, which improves the efficiency of finding fault points after the algorithm is located, shortens the fault elimination time of the intelligent station maintenance personnel, and meets the requirements of safe and stable operation and maintenance of the intelligent station; (3) A dynamic adaptive inertial weight factor is designed to improve the optimization ability of the particle swarm algorithm, thereby generating a fault location model of the secondary system of the intelligent station using the modified particle swarm algorithm to optimize the BP neural network, which improves the convergence speed of the model and the accuracy of fault prediction. Attached Figure Description

[0012] Figure 1 This is a flowchart of the intelligent substation secondary system fault location method based on optimized BP neural network of the present invention.

[0013] Figure 2 This is a typical link model structure diagram of an intelligent station.

[0014] Figure 3 This is a schematic diagram of the fault location table for the intelligent station.

[0015] Figure 4 The iterative process diagram for optimizing the BP neural network algorithm model.

[0016] Figure 5 A schematic diagram illustrating the optimization of the BP neural network algorithm model's accuracy. Detailed Implementation

[0017] like Figure 1 As shown, the present invention provides a fault location method for a secondary system of an intelligent substation based on an optimized BP neural network, comprising the following steps:

[0018] Step 1: Collect historical data on secondary system faults of the intelligent substation, analyze the fault location and causes, and classify the secondary system faults of the intelligent substation.

[0019] Step 2: Set up a fault simulation experiment for the secondary system of the intelligent station, simulate the fault types identified in Step 1, collect data from the fault simulation experiment, and integrate the fault data to form a fault database.

[0020] Step 3: Use binary encoding to number the experiments in Step 2. This encoding rule can combine the fault type and corresponding fault point of the intelligent station secondary system with the binary code to realize the correspondence between the binary code and the fault point. Then, through base conversion, hexadecimal fault location code is generated to form the corresponding fault location table. This table is then combined with the fault location database in Step 2 to form a complete fault location database with fault type encoding.

[0021] Step 4: The complete fault location database is filtered by low variance filtering. The filtered data is then processed by principal component analysis to reduce the dimensionality of the data. The processed data is then imported into the BP neural network intelligent location algorithm based on improved particle swarm optimization. The hexadecimal fault location code is used as the output result to complete the learning and training of the fault database.

[0022] Step 5: Import the fault data of the intelligent substation secondary system into the BP neural network intelligent positioning algorithm in Step 4, output the hexadecimal fault location result, compare the output result with the fault location table, find the actual fault point, and complete the fault elimination work of the intelligent substation secondary system.

[0023] Furthermore, in step 1, the fault types of the intelligent station's secondary system are classified into the following three main categories:

[0024] The first category is intelligent station sampling circuit failure, including current sampling circuit and voltage sampling circuit failure. The causes of failure include external wiring errors.

[0025] The second category is intelligent station equipment failures, including failures of protection devices, merging units, and intelligent terminals. The causes of failures include abnormal software operation, power board failure, CPU board failure, DI board failure, and DO board failure.

[0026] The third category is secondary communication loop failures, including network link, direct acquisition link, and direct jump link failures. The causes of these failures include failures in the corresponding merging unit, smart terminal, switch fiber optic cable, or fusion splice interface.

[0027] Furthermore, in step 2, the fault types identified in step 1 are simulated, data from the fault simulation experiment are collected, and the fault data is integrated to form a fault database, as detailed below:

[0028] The simulated intelligent substation secondary system fault types include: simulated normal operation, voltage transformer (TV) wiring error, current transformer (TA) wiring error, voltage transformer (TV) disconnection, current transformer (TA) disconnection, intelligent substation equipment functional module board fault, and fiber optic or fusion splice fault. The intelligent substation equipment includes protection devices, merging units, and intelligent terminals.

[0029] The specific fault simulation methods for the intelligent station secondary system fault experiment in step 2 are as follows:

[0030] (1) Current transformer to terminal block disconnection: Simulate the disconnection of phase A of the current transformer. A fault disconnection at a certain point of the line from phase A to the terminal block of the merging unit, or a loose or disconnected connection at the contact point can cause a single-phase disconnection. In the experiment, the phase A connection is removed to simulate a single-phase disconnection. After entering the same voltage and current values ​​as under normal conditions, the monitoring background collects the voltage and current data values ​​under the fault condition and records the fault experiment data.

[0031] (2) Incorrect wiring from current transformer to terminal block: The A and B phases of the simulated current transformer are reversed. During the experiment, the A phase current output of the relay protection device should be connected to the B phase current input of the merging unit terminal block, and the B phase current output should be connected to the A phase current input of the merging unit terminal block. After entering the voltage and current data, record the fault experiment data collected by the monitoring background.

[0032] (3) Intelligent station equipment failure: Simulate the failure of the merging unit. During the experiment, the sampling board module of the line merging unit is pulled out from the back of the protection device. After the voltage and current data are entered, the monitoring background sends the corresponding alarm information of the merging unit and records the failure test data.

[0033] (4) Fiber optic or fusion splice failure: During the experiment, fiber optic breaks and TX and RX reversal are simulated by plugging, unplugging or replacing the fiber optic cable to simulate communication link failure. After the voltage and current data are entered, the monitoring background issues alarm information for the corresponding link and records the fault experimental data.

[0034] Furthermore, step 3, combined with step 2, forms a fault location table, as detailed below:

[0035] After collecting fault data, each set of data needs to be numbered, and the corresponding intelligent station secondary system fault type and its corresponding fault point are combined with the binary code. The specific sixteen-bit binary rules are as follows:

[0036] D15-D14: Indicates the fault type (00 for no fault; 01 for a first-class intelligent station sampling circuit fault; 10 for a second-class intelligent station equipment fault; 11 for a third-class secondary communication circuit fault);

[0037] D13-D12: Indicate the specific sampling fault type (D14 indicates voltage transformer sampling fault; D13 indicates current transformer fault);

[0038] D11-D10: Indicates the specific device type (00 no device fault; 01 merging unit fault; 10 smart terminal fault; 11 protection device fault);

[0039] D9-D8: Indicates the specific link type (00 no link failure; 01 network link failure; 10 direct acquisition link failure; 11 direct jump link failure);

[0040] D7: Indicates a sampling circuit breakage fault;

[0041] D6-D4: Indicates the fault phase sequence of the sampling circuit;

[0042] D3-D0: Indicates the fault point number (numbered from 1 to 15, which can represent 15 fault points under each type. For large intelligent stations, if it is necessary to add fault point numbers, simply shift the entire binary code to the left, and the extra bits together form a new position number).

[0043] This establishes the correspondence between binary codes and fault points, and generates hexadecimal fault location codes through base conversion, forming a corresponding fault location table. The relationship between the fault location matrix code and the matrix points can then be determined from the fault location table.

[0044] Furthermore, the filtering and dimensionality reduction of the fault database in step 4 are as follows:

[0045] Low variance is used to filter data. Variance is used to calculate the difference between each variable and the population mean. When the data distribution is relatively dispersed, the sum of the squares of the differences between each data point and the mean is large, and the variance is large. When the data distribution is relatively concentrated, the sum of the squares of the differences between each data point and the mean is small. Generally speaking, variables with low variance are considered to carry little information and can be directly deleted.

[0046] Principal component analysis (PCA) is used to reduce the dimensionality of data. PCA uses orthogonal transformations to convert observed data represented by linearly correlated variables into data represented by a few linearly independent variables, called principal components. In PCA, it's crucial to ensure that the cumulative contribution rate of the first few extracted principal components reaches a high level. Furthermore, these extracted principal components must be able to provide interpretations consistent with the actual context and meaning. If the original variables have a high correlation, the cumulative contribution rate of the first few principal components will usually reach a high level; in other words, the cumulative contribution rate is usually easier to achieve in this case.

[0047] Furthermore, the improved particle swarm optimization algorithm in step 4 optimizes the BP neural network, as follows:

[0048] Optimizing a BP neural network model using an improved particle swarm optimization algorithm essentially leverages the excellent global search and optimization capabilities of the particle swarm optimization algorithm to continuously optimize the initial weights and thresholds of the BP network model. Experimental data is input into the BP network model, and the classification accuracy is used as the fitness function of the BP algorithm. A set of parameter combinations that minimizes the fitness function value is found and used as the parameter settings for the BP network model.

[0049] Particle swarm optimization treats individuals in a swarm as massless and volumeless particles in a D-dimensional search space. Each particle has its own position x and velocity v, and it continuously moves towards its historical best position P in the solution space through a fitness function. best and the best location cluster across the entire domain g best :

[0050]

[0051] in, Let be the velocity of particle i in the d-th dimension at the k-th iteration; p represents the velocity of particle i in the d-th dimension during the (k-1)th iteration. id p represents the optimal value found so far for the i-th particle. gd This is the optimal value found so far in the entire particle swarm; This represents the position of particle i in the d-th dimension during the k-th iteration. ω represents the position of particle i in the d-th dimension during the (k-1)th iteration; ω is the inertia weight factor; c1 and c2 are learning factors; r1 and r2 increase the random search capability.

[0052] The fitness function is expressed as:

[0053]

[0054] Where F(i) is the fitness function of the i-th particle; n is the number of samples in the training set; y j This represents the actual output value of the j-th sample data; o j Let be the predicted output value of the j-th sample data in the classification model; function 1(y j ==o j ) is y j with o j The output is 1 when the values ​​are the same, and 0 when they are different.

[0055] The evolutionary dispersion k(t) is defined as the ratio of the standard deviation of the fitness values ​​of the t-th generation population to that of the (t-1)-th generation population.

[0056]

[0057] Where: k(t) is the degree of evolutionary discretization of generation t relative to generation t-1; F(t) is the fitness value of the population in generation t; F(t-1) is the fitness value of the population in generation t-1;

[0058] In neural networks, the sigmoid function is used to construct the neuron activation function, defined as:

[0059]

[0060] Combining the evolutionary discreteness k(t) and the Sigmoid function, the following formula for calculating the nonlinear dynamic adaptive inertia weight factor is given:

[0061]

[0062] Where: ω max ω min is the maximum and minimum inertia weight; the function exp represents an exponential function with base e; t represents the current iteration number; T represents the maximum number of evolution iterations; b is the damping factor, which takes the value [0,1]; k(t) is the degree of evolutionary discretization of generation t relative to generation t-1.

[0063] The BP neural network is optimized using an improved particle swarm optimization algorithm to obtain a BP neural network intelligent localization algorithm. The specific steps are as follows:

[0064] (1) Initialize the BP neural network structure and initialize the particle swarm optimization algorithm, including parameters such as particle velocity, position, PSO optimization termination condition, and number of iterations;

[0065] (2) Based on the velocity and position of each particle, obtain the fitness, individual extreme value and population extreme value of each particle through the fitness function;

[0066] (3) Fitness of all particles in the population and population optimum g best Compare the particle's fitness and its individual optimal P. best Replace the better one, calculate the inertial weight ω according to equation (5), and update the velocity and position of the particle;

[0067] (4) After PSO optimization is terminated, the optimal weights and thresholds of the BP neural network are obtained and substituted into the BP neural network to initialize the initial values ​​and thresholds of the BP neural network.

[0068] (5) Input sample data, and train the model using the backpropagation of error;

[0069] (6) Determine the error between the output value and the true value of the BP neural network. When the error value is large, calculate the backward error and backpropagate to update the connection weights and thresholds between each layer until the termination condition is met, that is, the error value is not greater than the threshold or the maximum number of iterations is reached.

[0070] Furthermore, in step 5, the fault data is imported into the intelligent positioning algorithm output from step 4 and compared with the fault location table to find the actual fault point, as follows:

[0071] In the application, only the corresponding data that changes under the fault state needs to be collected for fault data. This data is directly used as input. For other data, the normal operating state value is used as the default input. The fault data is imported into the intelligent fault location algorithm, and finally the fault hexadecimal code is output. This code is compared with the fault location table to determine the fault type and fault point, guiding the intelligent station operation and maintenance personnel to complete the troubleshooting work.

[0072] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0073] Example

[0074] To verify the effectiveness of the present invention, Figure 2 The typical link model of the intelligent station shown is an example. Fault simulation experiments are conducted according to steps 1 and 2. Step 3 involves selecting a portion of the faults to obtain the following results: Figure 3The fault location table is shown. Following step 4, the algorithm model initialization parameters are set as follows: low variance filtering threshold is 0.01; principal component analysis contribution is 0.98; particle swarm optimization algorithm population size is 5; population update times are 30; maximum velocity is 1.0; minimum velocity is -1.0; maximum boundary is 2.0; minimum boundary is -2.0; learning factors c1 and c2 are both 4.5; maximum and minimum inertia weights are 0.9 and 0.4; damping factor is 0.2; BP neural network hidden layer node count is 6; training iterations are 1000; target error is 1e-5; learning rate is 0.01. The fault data is imported into the algorithm model for training, and the number of iterations and model accuracy are shown below. Figure 4 , Figure 5 As shown.

[0075] If we assume that the voltage transformer of the #1 main transformer merging unit is disconnected from the terminal block B phase, and the input voltage of phases A and C is 57.7V, then the data collected by the monitoring backend is as follows: the voltage amplitude of phase B changes from the input value to 0V, while the voltages of the other two phases remain unchanged; the self-generated zero-sequence voltage value changes from about 0.01V to about 38.49V; the negative-sequence voltage value increases from about 0V to about 19.25V; the amplitudes of line voltage Uab and line voltage Ubc both decrease to the phase voltage amplitude, and the angle changes from 120 degrees to a non-fixed value; the active power decreases to 2 / 3 of the original value.

[0076] The fault data is imported into the fault location model of the intelligent substation's secondary system. After calculation, the hexadecimal code for fault location is 60A2H, corresponding to the binary code 0110000010100010. Following step 5, the output result is compared with the fault location table to locate the actual fault point, namely, the open circuit from the voltage transformer of the #1 main transformer merging unit to phase B of the terminal block. At this point, the fault type and its location can be identified, allowing the intelligent substation maintenance personnel to eliminate the fault and ensure the safe operation of the intelligent substation.

[0077] The embodiments described above are merely typical implementations of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A fault location method for a secondary system of an intelligent substation based on an optimized BP neural network, characterized in that, Includes the following steps: Step 1: Collect historical data on secondary system faults of the intelligent substation, analyze the fault location and causes, and classify the secondary system faults of the intelligent substation. Step 2: Set up a fault simulation experiment for the secondary system of the intelligent station, simulate the fault types identified in Step 1, collect data from the fault simulation experiment, and integrate the fault data to form a fault database. Step 3: Use binary encoding to number the experiments in Step 2. This encoding rule can combine the fault type and corresponding fault point of the intelligent station secondary system with the binary code to realize the correspondence between the binary code and the fault point. Then, through base conversion, hexadecimal fault location code is generated to form the corresponding fault location table. This table is then combined with the fault location database in Step 2 to form a complete fault location database with fault type encoding. Step 4: The complete fault location database is filtered by low variance filtering. The filtered data is then processed by principal component analysis to reduce the dimensionality of the data. The processed data is then imported into the BP neural network intelligent location algorithm based on improved particle swarm optimization. The hexadecimal fault location code is used as the output result to complete the learning and training of the fault database. Step 5: Import the fault data of the intelligent station secondary system into the BP neural network intelligent positioning algorithm in step 4, output the hexadecimal fault positioning result, compare the output result with the fault positioning table, find the actual fault point, and complete the fault elimination work of the intelligent station secondary system. Step 4 uses the improved particle swarm optimization-based BP neural network intelligent localization algorithm, as detailed below: Particle swarm optimization treats individuals in a swarm as massless and volumeless particles in a D-dimensional search space, each with its own position. and speed In the solution space, it continuously moves towards its historical best position through the fitness function. and the best location in the entire field : （1） in, For the first Particles in the next iteration In the Dimensional speed; For the first Particles in the next iteration In the Dimensional speed; For the first The best value found so far for each particle; This is the optimal value found so far in the entire particle swarm; For the first Particles in the next iteration In the The position of the dimension; For the first Particles in the next iteration In the The position of the dimension; Inertia weighting factor; , For learning factors; , Increase random search capability; The fitness function is expressed as: （2） in, For the first Fitness function for each particle; The number of samples in the training set; For the first The actual output value of each sample data; For the classification model Predicted output value for each sample data; function for and The output is 1 when the values ​​are the same, and 0 when they are different; The first Generation and the first The ratio of the standard deviations of the fitness values ​​of different generations of a population is defined as the evolutionary dispersion. : （3） in: For the first The generation relative to the first The degree of discretization in generational evolution; For the first Fitness values ​​of the generation population; For the first Fitness values ​​of the generation population; In neural networks, the sigmoid function is used to construct the neuron activation function, defined as: （4） Combining evolutionary dispersion Based on the Sigmoid function, the following formula is given for calculating the nonlinear dynamic adaptive inertia weight factor: （5） in: , The maximum and minimum inertia weights; function This represents an exponential function with base e; Indicates the current iteration number; Indicates the maximum number of evolution iterations; The damping factor is [0,1]. For the first The generation relative to the first The degree of discretization in generational evolution; The BP neural network is optimized using an improved particle swarm optimization algorithm to obtain a BP neural network intelligent localization algorithm. The specific steps are as follows: (1) Initialize the BP neural network structure and initialize the particle swarm optimization algorithm, including parameters such as particle velocity, position, PSO optimization termination condition, and number of iterations; (2) Based on the velocity and position of each particle, obtain the fitness, individual extreme value and population extreme value of each particle through the fitness function; (3) Fitness of all particles in the population and population optimality Compare the fitness and individual optimality of the particles. Replace it with the better one, and calculate the inertia weight according to equation (5). Update the particle's velocity and position; (4) After PSO optimization terminates, the optimal weights and thresholds of the BP neural network are obtained, and the initial values ​​and thresholds of the BP neural network are substituted into the BP neural network to initialize the BP neural network. (5) Input sample data, and train the model using the backpropagation of error; (6) Determine the error between the output value and the true value of the BP neural network. When the error value is large, calculate the backward error and backpropagate to update the connection weights and thresholds between each layer until the termination condition is met, that is, the error value is not greater than the threshold or the maximum number of iterations is reached.

2. The method for fault location of intelligent substation secondary system based on optimized BP neural network according to claim 1, characterized in that, Step 1 categorizes the fault types of the intelligent substation secondary system into the following three main categories: The first category is intelligent station sampling circuit failure, including current sampling circuit and voltage sampling circuit failure. The causes of failure include external wiring errors. The second category is intelligent station equipment failures, including failures of protection devices, merging units, and intelligent terminals. The causes of failures include abnormal software operation, power board failure, CPU board failure, DI board failure, and DO board failure. The third category is secondary communication loop failures, including network link, direct acquisition link, and direct jump link failures. The causes of these failures include failures in the corresponding merging unit, smart terminal, switch fiber optic cable, or fusion splice interface.

3. The method for fault location of intelligent substation secondary system based on optimized BP neural network according to claim 1, characterized in that, In step 2, the fault types identified in step 1 are simulated, and data from the fault simulation experiment are collected. The fault data is then integrated to form a fault database, as detailed below: The simulated intelligent substation secondary system fault types include: simulated normal operation, voltage transformer (TV) wiring error, current transformer (TA) wiring error, voltage transformer (TV) disconnection, current transformer (TA) disconnection, intelligent substation equipment functional module board fault, and fiber optic or fusion splice fault. The intelligent substation equipment includes protection devices, merging units, and intelligent terminals.

4. The method for fault location of intelligent substation secondary system based on optimized BP neural network according to claim 1, characterized in that, The specific fault simulation methods for the intelligent station secondary system fault experiment in step 2 are as follows: (1) Current transformer to terminal block disconnection: Simulate the disconnection of phase A of the current transformer. A fault disconnection at a certain point of the line from phase A to the terminal block of the merging unit, or a loose or loose connection at the contact point can cause a single-phase disconnection. In the experiment, the phase A connection is removed to simulate a single-phase disconnection. After entering the same voltage and current values ​​as under normal conditions, the monitoring background collects the voltage and current data values ​​under the fault condition and records the fault experiment data. (2) Incorrect wiring from current transformer to terminal block: The A and B phases of the simulated current transformer are reversed. During the experiment, the A phase current output of the relay protection device should be connected to the B phase current input of the merging unit terminal block, and the B phase current output should be connected to the A phase current input of the merging unit terminal block. After entering the voltage and current data, record the fault experiment data collected by the monitoring background. (3) Intelligent station equipment failure: simulate the failure of the merging unit. During the experiment, the sampling board module of the line merging unit is pulled out from the back of the protection device. After the voltage and current data are entered, the monitoring background issues the corresponding alarm information of the merging unit and records the failure test data. (4) Fiber optic or fusion splice failure: During the experiment, fiber optic breakage and TX and RX reversal are simulated by plugging, unplugging or replacing the fiber optic cable to simulate communication link failure. After the voltage and current data are entered, the monitoring background issues alarm information for the corresponding link and records the fault experimental data.

5. The method for fault location of intelligent substation secondary system based on optimized BP neural network according to claim 1, characterized in that, The formation of the fault location table in step 3 is as follows: After collecting fault data, each set of data needs to be numbered, and the corresponding intelligent station secondary system fault type and its corresponding fault point are combined with the binary code. The specific sixteen-bit binary rules are as follows: D15-D14: Indicates the fault type; 00 indicates no fault; 01 indicates a first-type intelligent station sampling circuit fault; 10 indicates a second-type intelligent station equipment fault; 11 indicates a third-type secondary communication circuit fault. D13-D12: Indicate the specific sampling fault type; D14 indicates a voltage transformer sampling fault; D13 indicates a current transformer fault. D11-D10: Indicates the specific device type; 00 indicates no device fault; 01 indicates merging unit fault; 10 indicates smart terminal fault; 11 indicates protection device fault. D9-D8: Indicates the specific link type; 00 indicates no link failure; 01 indicates network link failure; 10 indicates direct acquisition link failure; 11 indicates direct jump link failure; D7: Indicates a sampling circuit breakage fault; D6-D4: Indicates the fault phase sequence of the sampling circuit; D3-D0: Indicates the fault point number; the numbering is from 1 to 15, representing 15 fault points under each type. If a fault point number needs to be added, the entire binary code is shifted to the left, and the extra bits are used together to form a new position number. The above rules establish the correspondence between binary codes and fault points. Hexadecimal fault location codes are generated through base conversion, forming a corresponding fault location table. The relationship between the fault location matrix code and the matrix points is then determined from the fault location table.

6. The method for fault location of intelligent substation secondary system based on optimized BP neural network according to claim 1, characterized in that, The dimensionality reduction process for the faulty database in step 4 is as follows: Low variance is used to filter data; variance is used to calculate the difference between each variable and the population mean. Principal component analysis (PCA) is used to reduce the dimensionality of data. PCA uses orthogonal transformation to convert observed data represented by linearly correlated variables into data represented by a few linearly independent variables. The linearly independent variables are called principal components. In PCA, it is ensured that the cumulative contribution rate of the first few extracted principal components reaches a set value. Secondly, these extracted principal components must be able to provide an interpretation that is consistent with the actual background and meaning. If the original variables have a correlation higher than the threshold, the cumulative contribution rate of the first few principal components reaches the set value.

Images