A method for identifying distribution network line parameters based on GA-SVR

Abstract

A distribution network line parameter identification method based on GA-SVR includes: S1, setting the parameters of each branch, including the resistance and reactance of each branch, and obtaining the index data of the voltage amplitude, active power, reactive power, active loss and reactive loss of the head and end nodes of each branch through the node flow calculation program; S2, extracting each index data and sorting them into a line parameter data group; S3, normalizing the data in the line parameter data group; S4, combining the normalized line parameter data group with the GA algorithm to optimize the penalty factor c and kernel function parameter g of the support vector regression machine SVR to form a GA-SVR; S5, identifying the distribution network line parameters through GA-SVR. Effective identification of distribution network line parameters is achieved; and simulation results show that compared with the traditional SVR method, this method greatly improves the identification accuracy of line parameters and has a better line parameter identification effect.

Description

Technical Field

[0001] The present invention relates to the field of power technology of distribution network, and in particular to a distribution network line parameter identification method based on GA-SVR. Background Art

[0002] Refined identification of distribution network line parameters is helpful for subsequent work such as state estimation, accurate calculation of power flow, accurate calculation of line loss, relay protection setting and fault location. It is not only an important basis for achieving higher-level and more reasonable access to energy storage equipment and other devices, but also a prerequisite for optimized operation of distribution networks. In recent years, with the access of large-scale distributed power sources, energy storage equipment, electric vehicles and nonlinear loads to distribution networks, the structure of distribution networks has become increasingly complex. At the same time, line parameters are affected by factors such as distribution network operating conditions, line temperature, line environment, overvoltage, etc., and there are differences between the actual values ​​and the recorded values. Therefore, accurate identification of line parameters is very important for distribution network analysis applications such as power flow calculation, relay protection setting and fault location.

[0003] In previous studies, the line parameter identification is based on the known identification equations, but the line parameters in actual situations often have certain differences from the theoretical mathematical model, which may cause the identification results to fail to reflect the true value of the line parameters in a true sense; sometimes, in order to ensure the accuracy of the established model, the order of the model will be greatly improved, and the complexity of parameter identification based on the high-order system model will also rise linearly. Line parameter identification without a line model is the key to data-driven line parameter identification. Support vector regression (SVR) can effectively build a line parameter identification model, but the values ​​of its penalty factor and kernel function parameters have a great influence on the regression prediction model. Genetic algorithm (GA) is designed and proposed according to the evolution law of organisms in nature, and it has unique advantages for parameter optimization problems. Based on the current research on these issues at home and abroad, the present invention further studies the method of using GA-SVR algorithm to improve the accuracy of distribution network line parameter identification, mainly studies the use of GA to optimize the penalty factor and kernel function parameters of SVR, and builds a GA-SVR regression prediction processor, thereby achieving high-precision identification of distribution network line parameters. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a distribution network line parameter identification method based on GA-SVR. By setting the parameters between each branch and substituting them into a 33-node power flow calculation program, the voltage amplitude, active power, reactive power, active loss and reactive loss data of the head and terminal nodes of each branch are obtained, GA is used to obtain the optimal values ​​of the SVR penalty factor and the kernel function parameters, a GA-SVR regression prediction processor is constructed, the data samples are normalized and divided into a training set and a test set by a 5-fold cross validation, and the data are input into the GA-SVR regression prediction processor. Combined with the optimal parameters, high-precision identification of the distribution network line parameters is achieved.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A distribution network line parameter identification method based on GA-SVR includes the following steps:

[0007] S1. Set the parameters of each branch, including the resistance and reactance of each branch, and obtain the index data of voltage amplitude, active power, reactive power, active loss and reactive loss of the head and end nodes of each branch through the node flow calculation program;

[0008] S2, extracting various index data and arranging them to form a line parameter data group;

[0009] S3, normalizing each index data in the line parameter data group;

[0010] S4, combining the normalized line parameter data group with the GA algorithm to optimize the penalty factor c and kernel function parameter g of the support vector regression machine SVR, thereby forming a GA-SVR regression prediction processor;

[0011] S5. Identify the distribution network line parameters through the GA-SVR regression prediction processor.

[0012] To optimize the above technical solutions, the specific measures taken also include:

[0013] Furthermore, the content of step S4 includes:

[0014] S4.1. Obtain each indicator data in the normalized line parameter data set and divide it into a training sample set and a test sample set;

[0015] S4.2. Initialize the GA algorithm and reset its original parameters; specifically, initialize the population and randomly generate O chromosomes, which are numbered m1, m2, m3, ..., m o ;

[0016] S4.3. Each chromosome has its own gene encoding k iCorresponding to, and the gene encoding k i Each has its own corresponding penalty factor c i and kernel function parameter g i ; The penalty factor c i and kernel function parameter g i These are the parameters of the support vector regression machine SVR;

[0017] Based on the characteristics of support vector regression machine SVR that can build vector regression model, by combining gene encoding k i The corresponding penalty factor c i , kernel function parameter g i and training sample sets to train a support vector regression model, and using a test sample set to evaluate the performance of the support vector regression model;

[0018] S4.4. Calculate the gene code k corresponding to each chromosome i The fitness f i , and according to the fitness f i Calculate the probability p that the corresponding chromosome is the parent chromosome i ; and select the two chromosomes with the highest probability as the parent chromosomes of the next generation population; among them:

[0019]

[0020] In the formula, f i Represents the fitness of any chromosome; M is the number of samples in the test sample set, is the regression value of the jth test sample, is the feature vector of the jth test sample, Support vector regression model for test samples The predicted value, p i represents the probability of any corresponding chromosome being selected as the parent chromosome, f k is the kth chromosome m among o chromosomes k The fitness of , o is the number of chromosomes;

[0021] S4.5, iterating the population to update the gene coding information in the chromosome, specifically: performing selection, crossover, and mutation operations on the two selected parent chromosomes to generate O chromosomes;

[0022] The selection method adopts the roulette method; the mutation method adopts the uniform mutation method, and the crossover method adopts the linear combination method. The specific formula is: Where m p and m q represents the two parent chromosomes selected in step S4.4, m ph 、m qhThey correspond to the two parental chromosomes after gene recombination; p represents the crossover probability;

[0023] S4.6, loop steps S4.3-S4.5 to complete repeated iterations of the population, and set the number of iterations to T. After reaching the number of iterations T, proceed to step S4.7;

[0024] S4.7. When the number of iterations T is reached, the penalty factor c corresponding to the chromosome with the best fitness in the current population is * With g * As the optimal parameters of the support vector regression machine SVR, a GA-SVR regression prediction processor is formed; at the same time, the GA-SVR regression prediction processor is trained and verified by combining the training sample set and the test sample set.

[0025] Furthermore, the value of the iteration number T is 50 times.

[0026] Furthermore, the method of dividing the training sample set and the test sample set adopts a 5-fold cross-validation method.

[0027] Furthermore, the ratio of the training sample set to the test sample set is 9:1.

[0028] The beneficial effects of the present invention are:

[0029] 1. Support vector regression (SVR) can effectively build a line parameter identification model, but the values ​​of its penalty factor and kernel function parameters have a great influence on the regression prediction model; Genetic algorithm (GA) is designed and proposed according to the evolution law of organisms in nature, and it has unique advantages in parameter optimization problems; Based on the current research on these issues at home and abroad, the present invention further studies the method of using GA-SVR algorithm to improve the accuracy of distribution network line parameter identification, mainly studies the use of GA to optimize the penalty factor and kernel function parameters of SVR, and builds a GA-SVR regression prediction processor, thereby achieving high-precision identification of distribution network line parameters.

[0030] 2. Traditionally, in distribution network calculations, workers calculate flow data using line parameters, whereas this application uses flow data to reversely predict (derive) line parameters, thereby making the identification of line parameters more accurate. The idea of ​​this application is to: (1) perform simulations to obtain multiple sets of flow data; (2) use the multiple sets of data obtained to train a prediction model; and (3) use the trained prediction model to verify the effect.

[0031] 3. Compared with the traditional SVR, the present invention can greatly improve the identification accuracy of line parameters and has a better line parameter identification effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a structural diagram of a 33-node distribution network system in an embodiment of the present invention.

[0033] Figure 2 Schematic diagram of line resistance parameters of a 33-node distribution network system identified by SVR in an embodiment of the present invention.

[0034] Figure 3 Schematic diagram of line reactance parameters of a 33-node distribution network system identified by SVR in an embodiment of the present invention.

[0035] Figure 4 It is a schematic diagram of the genetic algorithm optimization process in an embodiment of the present invention.

[0036] Figure 5 It is a schematic diagram of line resistance parameters of a 33-node distribution network system identified by GA-SVR in an embodiment of the present invention.

[0037] Figure 6 Schematic diagram of GA-SVR identifying line reactance parameters of a 33-node distribution network system in an embodiment of the present invention.

[0038] Figure 7 Schematic diagram of line resistance error identification of a 33-node distribution network system by SVR and GA-SVR in an embodiment of the present invention.

[0039] Figure 8 Schematic diagram of line reactance error identification of a 33-node distribution network system by SVR and GA-SVR in an embodiment of the present invention.

[0040] Fig. 9 It is a schematic diagram of the overall construction of GA-SVR of the present invention. DETAILED DESCRIPTION

[0041] The present invention will now be described in further detail with reference to the accompanying drawings.

[0042] refer to Fig. 9 The main scheme of this application is as follows: a distribution network line parameter identification method based on GA-SVR, including:

[0043] Step 1: Set the parameters between each branch of the 33-node distribution network system, and obtain the voltage amplitude, active power, reactive power, active loss, reactive loss and other data of the head and end nodes of each branch through the 33-node power flow calculation program.

[0044] Step 2, extract the characteristic index parameters calculated by the calculation program, and organize the index data accordingly; that is, organize the voltage amplitude, active power, reactive power, active loss, reactive loss corresponding to each branch node, and the resistance, reactance and other data corresponding to each branch.

[0045] Step 3: normalize the sorted distribution network flow and the corresponding line parameter data group.

[0046] Step 4: Optimize the penalty factor c and kernel function parameter g of SVR through the GA algorithm to build a GA-SVR regression prediction processor.

[0047] Support vector regression (SVR) can build a regression model through training sample data, but the selection of penalty factor and kernel function parameters has a significant impact on its regression effect. Genetic algorithm (GA) can effectively optimize the corresponding parameters globally by imitating the natural competition law of survival of the fittest in nature, using operations such as selection, crossover, and mutation, thereby optimizing the penalty factor and kernel function parameters of SVR.

[0048] In step 4 above, GA is used to optimize the penalty factor and kernel function parameters of SVR, including:

[0049] 1) Obtain the normalized and sorted distribution network flow and corresponding line parameter data set in step 3.

[0050] 2) Parameter initialization. Initialize the population and randomly generate o chromosomes, which are labeled m1, m2, m3, ..., m o , each chromosome m i Both are related to the gene encoding k i Corresponding to, where c i With g i The SVR parameters corresponding to the i-th chromosome are the penalty factor c and the kernel function parameter g.

[0051] 3) Calculate fitness. Combined with gene encoding k i The SVR parameters in the training sample set are used to train the support vector regression model, and the test sample set is used to evaluate the model performance and calculate the gene encoding k i The fitness f in i , according to the fitness, the probability p of selecting two chromosomes as parent chromosomes is calculated i , and then according to the probability p i Select 2 chromosomes from the population as parent chromosomes;

[0052]

[0053] In the formula, For training the model, test the sample set The predicted value, p i Represents chromosome m i The probability of being selected as a parent chromosome, f k Chromosome m k The fitness of, M is the number of evaluation samples, is the feature vector of the jth evaluation sample, y j e is the regression value of the jth evaluation sample, and o is the number of chromosomes.

[0054] 4) Set the number of iterations to 50. If the number of iterations is not reached, repeat the selection, crossover and mutation operations on the parent chromosomes. Perform the mutation operation using uniform mutation and the crossover operation using linear combination. Perform the crossover operation on the chromosomes with a crossover probability of p (p∈[0,1]).

[0055]

[0056] In the formula, m p 、m q They correspond to the two parental chromosomes before gene recombination, m ph 、m qh They correspond to the two parental chromosomes after gene recombination.

[0057] 5) If the number of iterations is reached, the penalty factor c corresponding to the chromosome with the best fitness in the current population is * With g * As the optimal parameters of the support vector regression machine, the final model is trained and verified in combination with the sample set.

[0058] Step 5, use the 5-fold cross-validation method to divide the 320 groups of feature index data normalized in step 3 into training set and test set (288 groups for training set and 32 groups for test set), where the training set is used as the training sample of the GA-SVR regression prediction processor, and the test set is used as the test sample of the GA-SVR regression prediction processor for testing. Combined with the optimal penalty factor and kernel function parameters, high-precision identification of distribution network line parameters can be achieved.

[0059] In the embodiment, the data in step 1 include line resistance and line reactance, and the data obtained through power flow calculation are voltage amplitude, active power and reactive power at the head and end nodes of the branch.

[0060] In the embodiment, step 2 extracts the characteristic index parameters calculated by the calculation program, and organizes the index data accordingly, that is, the voltage amplitude, active power, reactive power corresponding to each branch node, and the resistance and reactance corresponding to each branch are sorted out, totaling 320 groups.

[0061] In the embodiment, step 3 normalizes the sorted distribution network flow and the corresponding line parameter data, and the normalization formula is:

[0062]

[0063] Where: X is the original data, X minis the minimum value in the original data, X max is the maximum value in the original data.

[0064] In the embodiment, in step 4, the penalty factor and kernel function parameters of SVR are optimized by GA, and the optimization result is that the penalty factor is 99.8424 and the kernel function parameter is 1.1934.

[0065] The method of the present invention is verified by simulation as follows:

[0066] Based on MATLAB, a line parameter identification regression prediction model for a 33-node distribution system is established and a simulation experiment is carried out. The structure of the 33-node distribution system is as follows: Figure 1 As shown ( Figure 1 The numbers in represent node numbers and have no other special meanings; and the 33-node power distribution system is a conventional system. This application uses this system to verify the parameter identification method based on GA-SVR proposed in the article).

[0067] Through simulation, 320 groups of sample data such as voltage amplitude, active power, reactive power, etc. of the head and terminal nodes of each branch are obtained. The 320 groups of normalized feature index data are divided into training set and test set (288 groups for training set and 32 groups for test set) by 5-fold cross-validation method. The training set is used as training sample of GA-SVR regression prediction processor, and the test set is used as test sample of GA-SVR regression prediction processor for testing.

[0068] Firstly, the unoptimized traditional SVR regression prediction processor is used to train and verify the distribution network line parameter identification model. The SVR model is built using the default parameters, with the penalty factor c as 1 and the kernel function parameter g as 1. The simulation results are as follows: Figure 2 , Figure 3 As shown, it can be seen that the predicted values ​​of line resistance and reactance identified by SVR are very different from the actual values.

[0069] Secondly, in order to find the global optimal point and improve the accuracy of the model, the GA algorithm is used to optimize the penalty factor c and kernel function parameter g in the SVR model. The optimization process is as follows: Figure 4 As shown, the optimization result is the penalty factor c * is 99.8424, and the kernel function parameter g * It is 1.1934.

[0070] Finally, the optimal penalty factor and kernel function parameters are combined to achieve high-precision identification of distribution network line parameters. Figure 5 , Figure 6 shown.

[0071] The error of resistance and reactance parameters identified by SVR and GA-SVR is as follows Figure 7 , 8 As shown in the figure, the SVR identification resistance absolute percentage mean error (MAPE) is 57.75%, and the reactance absolute percentage mean error (MAPE) is 63.70%; while the support vector regression machine optimized by genetic algorithm identification resistance absolute percentage mean error (MAPE) is 5.01%, and the reactance absolute percentage mean error (MAPE) is 6.58%, which are 52.74% and 57.12% lower than the SVR identification method. In addition, the resistance and reactance root mean square error (RMSE) of the SVR method are 0.277 and 0.223 respectively, and the resistance and reactance root mean square error (RMSE) of the improved SVR parameter identification method are 0.0327 and 0.0298 respectively. Compared with SVR, the RMSE value of the improved SVR parameter identification method is reduced by 0.2443 and 0.1932 respectively in the identification of resistance and reactance. It can be seen that the improved SVR parameter identification model has the advantages in distribution network parameter identification.

[0072] The above is only an embodiment of the present application. The protection scope of the present invention is not limited to the above embodiment, and is not used to limit the present application. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as belonging to the protection scope of the present invention.

[0073] It should be noted that the terms such as "upper", "lower", "left", "right", "front", "back", etc. cited in the invention are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments in their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.

[0074] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A distribution network line parameter identification method based on GA-SVR, characterized in that: The following steps are involved: S1. Set the parameters of each branch, including the resistance and reactance of each branch, and obtain the index data of voltage amplitude, active power, reactive power, active loss and reactive loss of the head and end nodes of each branch through the node flow calculation program; S2, extracting various index data and arranging them to form a line parameter data group; S3, normalizing each index data in the line parameter data group; S4, combining the normalized line parameter data group with the GA algorithm to optimize the penalty factor c and kernel function parameter g of the support vector regression machine SVR, thereby forming a GA-SVR regression prediction processor; The content of step S4 includes: S4.

1. Obtain each indicator data in the normalized line parameter data set and divide it into a training sample set and a test sample set; S4.

2. Initialize the GA algorithm and reset its original parameters; specifically, initialize the population and randomly generate O chromosomes, which are numbered m1, m2, m3, ..., m o ; S4.

3. Each chromosome has its own gene encoding k i Corresponding to, and the gene encoding k i Each has its own corresponding penalty factor c i and kernel function parameter g i ; The penalty factor c i and kernel function parameter g i These are the parameters of the support vector regression machine SVR; Based on the characteristics of support vector regression machine SVR that can build vector regression model, by combining gene encoding k i The corresponding penalty factor c i , kernel function parameter g i and training sample sets to train a support vector regression model, and using a test sample set to evaluate the performance of the support vector regression model; S4.

4. Calculate the gene code k corresponding to each chromosome i The fitness fi of the next generation is calculated, and the probability pi of the corresponding chromosome as the parent chromosome is calculated according to the fitness fi; and the two chromosomes with the highest probability are selected as the parent chromosomes of the next generation population; where: In the formula, fi represents the fitness of any chromosome; M is the number of samples in the test sample set, is the regression value of the jth test sample, is the feature vector of the jth test sample, Support vector regression model for test samples The predicted value of pi represents the probability of any corresponding chromosome being selected as the parent chromosome, and fk is the kth chromosome m among o chromosomes. k The fitness of , o is the number of chromosomes; S4.5, iterating the population to update the gene coding information in the chromosome, specifically: performing selection, crossover, and mutation operations on the two selected parent chromosomes to generate O chromosomes; The selection method adopts the roulette method; the mutation method adopts the uniform mutation method, and the crossover method adopts the linear combination method. The specific formula is: Where mp and mq represent the two parent chromosomes selected in step S4.4, m ph 、m qh They correspond to the two parental chromosomes after gene recombination; p represents the crossover probability; S4.6, loop steps S4.3-S4.5 to complete repeated iterations of the population, and set the number of iterations to T. After reaching the number of iterations T, proceed to step S4.7; S4.

7. When the number of iterations T is reached, the penalty factor c corresponding to the chromosome with the best fitness in the current population is * With g * As the optimal parameters of the support vector regression machine SVR, a GA-SVR regression prediction processor is formed; and the GA-SVR regression prediction processor is trained and verified by combining the training sample set and the test sample set; S5. Identify the distribution network line parameters through the GA-SVR regression prediction processor.

2. A distribution network line parameter identification method based on GA-SVR according to claim 1, characterized in that: The value of the iteration number T is 50 times.

3. A distribution network line parameter identification method based on GA-SVR according to claim 1, characterized in that: The method of dividing the training sample set and the test sample set adopts the 5-fold cross-validation method.

4. A distribution network line parameter identification method based on GA-SVR according to claim 1, characterized in that: The ratio of training sample set to test sample set is 9:1.

Images