High photovoltaic permeability power distribution network single-phase earth fault line selection method and device

Abstract

The invention provides a high photovoltaic permeability power distribution network single-phase earth fault line selection method and device, and belongs to the technical field of power grid fault diagnosis. The method comprises the following steps: taking the minimum envelope entropy as an optimization target, and optimizing the number of decomposition layers and penalty factors of a VMD model based on an IGGO algorithm to obtain an optimized VMD model; based on the optimized VMD model, carrying out noise reduction decomposition on the transient zero-sequence current of the fault of the power distribution network to obtain a plurality of IMF components of the transient zero-sequence current; performing signal reconstruction on each IMF component based on the distribution entropy to obtain a fault feature matrix of the power distribution network; and inputting the fault feature matrix into a trained CNN-SVM model to obtain a fault line selection result of the power distribution network. The method can improve the accuracy of the fault line selection technology, and guarantees the safe and stable operation of the power distribution network and the reliability of power supply.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid fault diagnosis, and particularly to a single-phase grounding fault line selection method and device for a distribution network with high photovoltaic penetration rate. Background Art

[0002] As a key link in the power system that directly faces users, the safe and stable operation of the distribution network plays a crucial role in ensuring reliable power supply and meeting the electricity demand of social production and life. In recent years, with the strong promotion and application of clean energy globally, the penetration rate of photovoltaic power generation in the distribution network has been continuously increasing due to its advantages of being renewable, pollution-free, and widely distributed. The access of a large number of distributed photovoltaic power sources to the distribution network has changed the structure and operating characteristics of the traditional distribution network, bringing many new challenges to aspects such as fault diagnosis and protection configuration of the distribution network.

[0003] Among various faults in the distribution network, the single-phase grounding fault is one of the most common fault types. Under normal operating conditions, the three-phase system of the distribution network is in a balanced state. However, when a single-phase grounding fault occurs, the voltage of the fault phase drops sharply, and the voltages of the non-fault phases increase, which may cause the weak insulation links to be damaged, further expanding the fault range and affecting the power supply reliability. In the distribution network environment with high photovoltaic penetration rate represented by the former Nanyu area, due to the intermittency, volatility of photovoltaic power sources, and the diversity of their access positions and capacities, the electrical characteristics of single-phase grounding faults become more complex. There are significant differences in the magnitude, direction of the fault current, and the fault transient process compared with the traditional distribution network. This makes the line selection method based on traditional fault feature analysis prone to misjudgment. At the same time, the fault transient signals under high photovoltaic penetration rate contain more harmonic components, high-frequency components, and complex attenuation characteristics, which all increase the difficulty of accurately extracting effective fault features for fault line selection. Summary of the Invention

[0004] An embodiment of the present invention provides a single-phase grounding fault line selection method and device for a distribution network with high photovoltaic penetration rate to solve the problem of improving the accuracy of fault line selection for a distribution network with high photovoltaic penetration rate.

[0005] In a first aspect, an embodiment of the present invention provides a single-phase grounding fault line selection method for a distribution network with high photovoltaic penetration rate, including:

[0006] Taking the minimum envelope entropy as the optimization objective, optimizing the decomposition layer number and penalty factor of the VMD model based on the IGGO algorithm to obtain an optimized VMD model;

[0007] Based on the optimized VMD model, performing noise reduction decomposition on the transient zero-sequence current of the distribution network when a fault occurs to obtain multiple IMF components of the transient zero-sequence current;

[0008] Signal reconstruction is performed on each IMF component based on distribution entropy to obtain the fault feature matrix of the distribution network;

[0009] The fault feature matrix is input into the trained CNN-SVM model to obtain the fault line selection result of the distribution network.

[0010] In a possible implementation, taking the minimum envelope entropy as the optimization objective, the decomposition layer number and penalty factor of the VMD model are optimized based on the IGGO algorithm to obtain the optimized VMD model, including:

[0011] Initialize the population; where the population includes multiple individuals, and the position of each individual corresponds to a set of decomposition layer number and penalty factor of the VMD model;

[0012] Divide the population into an exploration group and a development group, and update the positions of the individuals in the exploration group and the development group based on the current living environment and the current iteration number respectively;

[0013] Take the envelope entropy as the fitness function, calculate the fitness function value based on the position of each individual, and determine the optimal individual in the population based on the fitness function value;

[0014] Repeat the above steps until the convergence condition is met, take the position of the current optimal individual as the optimization result, and apply it to the VMD model to obtain the optimized VMD model.

[0015] In a possible implementation, initialize the population, including:

[0016] Initialize the population based on the improved Tent chaotic mapping; where the formula for initializing the population is:

[0017]

[0018] where, x n is the initial position of the nth individual, N is the population size, U b is the upper bound of the position, L b is the lower bound of the position, rand(0,1) is a random number.

[0019] In a possible implementation, taking the minimum envelope entropy as the optimization objective, the decomposition layer number and penalty factor of the VMD model are optimized based on the IGGO algorithm to obtain the optimized VMD model, and it also includes:

[0020] Perturb the position of the optimal individual based on the current iteration number and the adaptive t-distribution; where the formula for perturbation is:

[0021]

[0022] where, H bis the position of the optimal individual after mutation, X Flock1 is the position of the optimal individual, g(n) is the adaptive t-distribution with degree of freedom n, L + is a row vector with all dimensions being 1, t is the current iteration number, t max is the iteration number threshold.

[0023] In a possible implementation, updating the positions of individuals in the exploration group based on the current living environment and the current iteration number includes:

[0024] When t % 2 = 0, r3 < 0.5, and |A| < 1, update the positions of each individual in the exploration group based on the first formula; where the first formula is:

[0025] X Paddle (t + 1) = X * (t) - A|C.X * (t) - X(t)|

[0026] A = 2ar1 - a

[0027] C = 2r2

[0028] r1 = 0.5 + 0.5cos(tπ / T)

[0029] r2 = Levy(λ)

[0030] where t is the iteration number, r3 is the living environment, X * (t) is the position of the optimal individual, A is the search step parameter, C is the search amplitude parameter, X(t) is the position of an individual at the t-th iteration, X(t + 1) is the updated position of this individual, a is the linear change parameter, r1 is the step adjustment parameter, r2 is the amplitude adjustment parameter, and λ is the preset value;

[0031] When t % 2 = 0, r3 < 0.5, and |A| ≥ 1, update the positions of other individuals in the exploration group based on the position of a randomly selected individual in the exploration group and the second formula; where the second formula is:

[0032] X(t + 1) = w1*X Paddle1 +z*w2*(X Paddle2 -X Paddle3 )+(1 - z)*w3*(X(t) - X Pa ddl e1 )

[0033] where w1, w2, and w3 are random parameters, z is the decreasing parameter, X Paddle1 、X Paddle2 and X Pdddle3 are the positions of randomly selected individuals in the exploration group;

[0034] When t % 2 = 0 and r3 ≥ 0.5, update the positions of each individual in the exploration group based on the third formula; where the third formula is:

[0035] X(t + 1) = w4 * |X * (t) - X(t)|.e bl .cos(2πl) + [2w1(r4 + r5)] * X * (t)

[0036] where b is a constant, and l, w4, r4, and r5 are random parameters;

[0037] When t % 2 ≠ 0, randomly select multiple individuals in the exploration group as individuals who have lost their partners, and update the positions of these individuals based on the patrol formula; where the patrol formula is:

[0038] X(t + 1) = X(t) + D(1 + z) * w * (X(t) - X Flock1 )

[0039]

[0040] where X Flock1 is the position of the optimal individual, D is the step size control function, w is a random parameter, and d is the dimension.

[0041] In a possible implementation, update the positions of individuals in the exploitation group based on the current living environment and the current iteration number, including:

[0042] When t % 2 ≠ 0, update the positions of each individual in the exploitation group based on a randomly selected individual in the exploitation group and the fourth formula; where the fourth formula is:

[0043] X1 = X Sentry1 - A1.|C1.X Sentry1 - X|

[0044] X2 = X Sentry2 - A2.|C2.X Sentry2 - X|

[0045] X3 = X Sentry3 - A3.|C3.X Sentry3 - X|

[0046]

[0047] A = 2ar1 - a

[0048] C = 2r2

[0049] r1 = 0.5 + 0.5cos(tπ / T)

[0050] r2 = Levy(λ)

[0051] where t1, X2, and X3 are intermediate variables, and X Sentry1 , X Sentry2 , and X Sentry3 are the positions of individuals randomly selected from the development group, A is the search step parameter, C is the search amplitude parameter, X is the position of an individual at the t-th iteration, X(t + 1) is the updated position of this individual, a is the linear change parameter, r1 is the step adjustment parameter, r2 is the amplitude adjustment parameter, and λ is a preset value;

[0052] When t % 2 ≠ 0, update the positions of each individual in the development group based on the patrol formula; where the patrol formula is:

[0053] X(t + 1) = X(t) + D(1 + z) * w * (X(t) - X Flock1 )

[0054]

[0055] where X Flock1 is the position of the optimal individual, D is the step control function, w is a random parameter, and d is the dimension.

[0056] In a possible implementation, based on the distribution entropy, signal reconstruction is performed on each IMF component to obtain the fault feature matrix of the distribution network, including:

[0057] Calculate the distribution entropy of each IMF component;

[0058] Use the IMF components with distribution entropy values in the interval [δ, 1] as effective IMF signals for reconstruction to obtain the fault feature matrix of the distribution network; where δ is a preset value.

[0059] In a possible implementation, before inputting the fault feature matrix into the trained CNN - SVM model, it further includes:

[0060] Build a distribution network simulation model considering the access of a high - proportion photovoltaic power source;

[0061] Based on the distribution network simulation model, simulate various fault conditions and collect the corresponding transient zero - sequence currents;

[0062] Use the fault feature matrix of each transient zero - sequence current as a training sample and the corresponding fault condition as a sample label to train the initial CNN - SVM model to obtain the trained CNN - SVM model.

[0063] In a second aspect, an embodiment of the present invention provides a single - phase grounding fault line selection device for a distribution network with a high photovoltaic penetration rate, including:

[0064] A parameter optimization module, which is used to minimize the envelope entropy as the optimization objective, and optimize the decomposition layer number and penalty factor of the VMD model based on the IGGO algorithm to obtain an optimized VMD model;

[0065] A noise reduction decomposition module, which is used to perform noise reduction decomposition on the transient zero-sequence current of the distribution network with a fault based on the optimized VMD model to obtain multiple IMF components of the transient zero-sequence current;

[0066] A signal reconstruction module, which is used to perform signal reconstruction on each IMF component based on the distribution entropy to obtain a fault feature matrix of the distribution network;

[0067] A fault line selection module, which is used to input the fault feature matrix into a trained CNN-SVM model to obtain the fault line selection result of the distribution network.

[0068] The embodiment of the present invention provides a single-phase grounding fault line selection method and device for a distribution network with a high photovoltaic penetration rate. By optimizing the parameters of the VMD model through the IGGO algorithm and constructing a fault feature matrix through the distribution entropy, it can accurately extract the zero-sequence current in the complex transient process of the single-phase grounding fault for the stronger waveform and randomness of distributed photovoltaics. Through the CNN-SVM model, for the small sample problem of the distribution network grounding fault, automatic feature learning is performed on the extracted effective signals, and automatic fault classification is realized, which can improve the accuracy of the fault line selection technology and ensure the safe and stable operation of the distribution network and the reliability of power supply. Description of the Drawings

[0069] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0070] Figure 1 It is a flowchart of the implementation of the single-phase grounding fault line selection method for a distribution network with a high photovoltaic penetration rate provided by an embodiment of the present invention;

[0071] Figure 2 It is a waveform diagram of the single-phase grounding transient zero-sequence current of a fault feeder provided by an embodiment of the present invention;

[0072] Figure 3 It is a time-domain diagram of the modal components obtained by processing the fault transient current signal provided by an embodiment of the present invention;

[0073] Figure 4A It is a distribution diagram of the individual positions of the randomly initialized population provided by an embodiment of the present invention;

[0074] Figure 4B It is the distribution diagram of the positions of individuals in the Tent chaotic mapping population provided by an embodiment of the present invention;

[0075] Figure 5 It is the iterative curve graph of the optimization process of the IGGO algorithm provided by an embodiment of the present invention;

[0076] Figure 6 It is the equivalent model diagram of a distribution network with high photovoltaic penetration provided by an embodiment of the present invention;

[0077] Figure 7 It is the training effect diagram of the CNN - SVM model provided by an embodiment of the present invention;

[0078] Figure 8 It is the implementation flow chart of the single - phase grounding fault line selection method for a distribution network with high photovoltaic penetration provided by another embodiment of the present invention;

[0079] Figure 9 It is the structural schematic diagram of the single - phase grounding fault line selection device for a distribution network with high photovoltaic penetration provided by an embodiment of the present invention. Detailed implementation manners

[0080] In the following description, specific details such as specific system structures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present invention. However, those skilled in the art should clearly understand that the present invention can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well - known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present invention.

[0081] To make the purpose, technical solutions, and advantages of the present invention clearer, the following will be described through specific embodiments in conjunction with the accompanying drawings.

[0082] Figure 1 The implementation flow chart of the single - phase grounding fault line selection method for a distribution network with high photovoltaic penetration provided by the embodiments of the present invention is described in detail as follows:

[0083] Step 101: Taking the minimum envelope entropy as the optimization objective, optimize the decomposition layer number and penalty factor of the VMD model based on the IGGO algorithm to obtain an optimized VMD model.

[0084] In this embodiment, compared with traditional signal extraction algorithms, VMD decomposition can adaptively decompose a signal into multiple modal components with specific center frequencies according to the characteristics of the signal itself, effectively extracting each modal information closely related to the fault characteristics and providing a good basis for accurate fault diagnosis. VMD can more accurately identify fault characteristics, avoid diagnostic errors caused by modal aliasing, and is less affected by the endpoint effect compared with other modal decompositions, and can process the fault signals of the distribution network more stably. In the face of noise interference, VMD exhibits good robustness. During the decomposition process, it has a certain ability to suppress noise through mechanisms such as variational constraints. VMD only performs decomposition processing based on the characteristics of the signal itself and does not require prior construction of a complex system model to adapt to the fault signals of the distribution network.

[0085] In conventional methods such as the variational mode decomposition method, the central frequency method, the instantaneous frequency mean method, and the singular value kurtosis difference spectrum method, there is a large degree of subjectivity in the selection of the K value, and the analysis and selection speed is slow. Using intelligent algorithms to optimize the parameters of VMD has more obvious advantages compared with these traditional methods based on manual observation, and can avoid errors caused by human subjective experience. The Greylag Goose Optimization (GGO) algorithm is a meta-heuristic optimization algorithm. Through the evaluation of 29 different objective functions, comparison with 7 well-known algorithms, and solving 2 practical engineering design problems, the effectiveness of the algorithm in solving optimization problems is verified.

[0086] Step 102: Denoise and decompose the transient zero-sequence current when a fault occurs in the distribution network based on the optimized VMD model to obtain multiple IMF components of the transient zero-sequence current.

[0087] In this embodiment, first, the collected transient zero-sequence current signal containing noise is input into IGGO-VMD. The waveform of the transient zero-sequence current of the single-phase grounding of the fault feeder is as Figure 2 shown; then, use the IGGO algorithm to optimize the decomposition layer number k and the penalty factor α in VMD, select the envelope entropy as the fitness function, and select the [k, α] with the smallest envelope entropy as the optimal k and α parameter combination; third, input the optimized k and α into VMD to obtain the optimized VMD model, and use IGGO-VMD to perform denoising decomposition on the fault transient zero-sequence current to obtain k IMF components as shown in Figure 3 shown.

[0088] Step 103: Reconstruct the signal based on the distribution entropy for each IMF component to obtain the fault feature matrix of the distribution network.

[0089] In this embodiment, Distribution Entropy (DistEn) is a method for quantifying the complexity of time series data. It is particularly suitable for short data sequences and has significant applications in analyzing physiological time series, such as heart rate variability (HRV). Compared with other entropy measurement methods such as sample entropy and fuzzy entropy, distribution entropy does not require arbitrarily defining a distance threshold when evaluating the complexity of a signal, which makes it more effective and accurate when dealing with short data sequences. Effective IMF components can be selected through distribution entropy for fault line selection.

[0090] Step 104: Input the fault feature matrix into the trained CNN-SVM model to obtain the fault line selection result of the distribution network.

[0091] In this embodiment, the network architecture of the CNN-SVM single-phase grounding fault identification model includes: an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer.

[0092] The input layer is used for data input. Since the data features of influencing factors in the project do not have a two-dimensional structure, the model uses one-dimensional convolution for calculation. In the convolutional layer, a convolutional kernel is used to perform feature extraction and feature mapping on the influencing factor data after RF processing. The convolutional kernel performs the following operations:

[0093]

[0094] where is the feature map output after convolution; is the feature map output by the previous convolutional layer; is the convolutional kernel; is the bias term; x(*) is the activation function. The Relu function is used as the activation function to assist in expressing the complex features of the input.

[0095] The pooling layer samples the feature data obtained by convolution through a sliding window in the maximum pooling manner, and takes the maximum value within the window as the output to form the influencing factor feature sequence of the pooling layer. The maximum pooling is as follows:

[0096]

[0097] The fully connected layer is usually located in the last few layers of the network, flattens the selected and filtered influencing factor features, and performs non-linear combination. The output layer mainly realizes the classification of single-phase grounding fault line selection for the distribution network with high PV penetration rate.

[0098] In the embodiment of the present invention, the parameters of the VMD model are optimized by the IGGO algorithm, and a fault feature matrix is constructed through distribution entropy. For the stronger waveform and randomness of distributed photovoltaics, the zero-sequence current in the complex transient process of single-phase grounding faults can be accurately extracted. Through the CNN-SVM model, for the small sample problem of distribution network grounding faults, automatic feature learning is carried out on the extracted effective signals, and automatic fault classification is realized, which can improve the accuracy of fault line selection technology, ensure the safe and stable operation of the distribution network and the reliability of power supply.

[0099] In a possible implementation manner, taking the minimum envelope entropy as the optimization objective, the decomposition layer number and penalty factor of the VMD model are optimized based on the IGGO algorithm to obtain an optimized VMD model, including:

[0100] Initializing the population; where the population includes multiple individuals, and the position of each individual corresponds to a set of decomposition layer numbers and penalty factors of the VMD model;

[0101] Dividing the population into an exploration group and a development group, and updating the positions of the individuals in the exploration group and the development group respectively based on the current living environment and the current iteration number;

[0102] Taking the envelope entropy as the fitness function, calculating the fitness function value based on the position of each individual, and determining the optimal individual in the population based on the fitness function value;

[0103] Repeating the above steps until the convergence condition is met, taking the position of the current optimal individual as the optimization result, and applying it to the VMD model to obtain an optimized VMD model.

[0104] In this embodiment, the main principle of the IGGO optimization algorithm includes: The GGO algorithm adopts an exploration and development mechanism, simulating the foraging and patrolling behaviors of a flock of greylag geese. The population is divided into an exploration group (n1) responsible for finding new environments, and a development group (n2) that builds nests, forages, and guards at the current best location. The individual with the best fitness will be appointed as the "leader" to guide other individuals in foraging and migration. At the same time, some widowed greylag geese will leave the flock and live alone.

[0105] The exploration behavior in the IGGO algorithm includes: The flock of greylag geese works during the day and rests at night, repeatedly searching the nearby areas, looking forward to a better solution.

[0106] The developer behavior in the IGGO algorithm includes: The flock of greylag geese circles around the optimal solution, foraging during the day and patrolling at night in the vicinity. When t%2 = 0, the flock of greylag geese forages.

[0107] The gray goose flock distribution stage includes: At the beginning of the GGO algorithm, the gray goose flock is evenly divided into an exploration group (n1) and a development group (n2), each accounting for 50%. Subsequently, the number of individuals in each group is dynamically adjusted, reducing the exploration group and increasing the development group. If the objective function value of the optimal solution remains unchanged for three consecutive iterations, the number of individuals in the exploration group is increased to jump out of the local optimum.

[0108] The following improvement measures are added to the GGO algorithm in this embodiment: improving the Tent chaotic mapping to initialize the population, dynamic search based on Lévy flight and non-linear step size strategy, and optimal individual perturbation based on adaptive t-distribution.

[0109] Furthermore, in this embodiment, the envelope entropy is used as the fitness function of the IGGO algorithm. The envelope entropy can reflect the complexity of the signal amplitude change. In many actual signal processing scenarios, the amplitude change of the signal contains rich information; the concept of envelope entropy is based on information entropy and measures the uncertainty of the signal envelope amplitude change from the perspective of information theory. Its physical meaning is clear, making it easier to understand the direction and goal of optimization when searching for parameters according to the envelope entropy. The [k, α] combination with the minimum envelope entropy is selected as the optimal parameter combination, and the envelope entropy calculation formula is as follows:

[0110]

[0111] where E p is the envelope entropy, a(j) is the envelope signal sequence obtained by Hilbert demodulation of the signal x(j), where j = 1, 2, 3…, m, m is a positive integer, HI is the Hilbert transform of the signal, and p j is the normalized form of a(j).

[0112] In a possible implementation manner, initializing the population includes:

[0113] Initializing the population based on the improved Tent chaotic mapping; where the formula for initializing the population is:

[0114]

[0115] where x n is the initial position of the nth individual, N is the population size, U b is the upper bound of the position, L b is the lower bound of the position, and rand(0,1) is a random number.

[0116] In this embodiment, when the grey goose algorithm solves the optimization problem, it usually initializes the population randomly. Randomly initializing the population may cause the algorithm to focus on certain regions in the search space, thus falling into a local optimal solution. Moreover, for a distribution network with a high photovoltaic penetration rate, the volatility and randomness of photovoltaic power are stronger, and the transient process of the grounding fault is more complex when the light changes sharply. Therefore, more sufficient space search is required. For this reason, an improved Tent chaotic mapping is proposed to initialize the population in this embodiment. By combining random number generation and piecewise linear mapping and introducing the population boundary, the population initialization is made relevant to the boundary, helping the population to cover the search space more comprehensively, thereby enhancing the global search ability and population diversity. The specific formula is as follows:

[0117]

[0118] In the formula: N is the number of particles; L b and U b respectively represent the lower bound and the upper bound of the optimization problem; rand(0, 1) is a random number. The population position distribution obtained by random initialization is as shown in Figure 4A and the population position distribution obtained by initializing through the Tent chaotic mapping is as shown in Figure 4B . By comparing Figure 4A and Figure 4B , it can be seen that the population position distribution obtained by initializing through the Tent chaotic mapping is more dispersed.

[0119] In a possible implementation manner, taking the minimum envelope entropy as the optimization goal, the decomposition layer number and penalty factor of the VMD model are optimized based on the IGGO algorithm to obtain an optimized VMD model. It further includes:

[0120] perturbing the position of the optimal individual based on the current iteration number and the adaptive t-distribution; among them, the formula for perturbation is:

[0121]

[0122] where H b is the position of the mutated optimal individual, X Flock1 is the position of the optimal individual, g(n) is the adaptive t-distribution with n degrees of freedom, L + is a row vector with all dimensions being 1, t is the current iteration number, and t max is the iteration number threshold.

[0123] In this embodiment, aiming at the problem that the basic GGO algorithm is prone to falling into a local optimum during the iteration process, resulting in rapid assimilation of the population, the present invention introduces an adaptive t-distribution to perturb the individuals, enabling the algorithm to jump out of the local optimal solution, thereby improving the accuracy of optimization. The probability density calculation formula of the adaptive t-distribution is as follows:

[0124]

[0125] In the formula, n is the degree of freedom; X flock1 is the current global optimal value of the population; Γ is the gamma function.

[0126] Construct a piecewise function to describe the variation relationship between the iteration times t and the degree of freedom n in different stages. A linear function is adopted in the early stage of iteration, and a quadratic function is adopted in the later stage of iteration. Its expression is as follows:

[0127]

[0128] In the formula, t is the current iteration number; t c is the iteration number threshold.

[0129] Furthermore, considering the randomness of mutation and the computational complexity, only the globally optimal individual is perturbed. The perturbation equation of the optimal individual based on the adaptive t-distribution is as follows:

[0130] H b = X Flock1 + g(n)·L + (7)

[0131] In the formula, H b is the mutated optimal individual; X Flock1 is the current global optimal position; n is the degree of freedom of the adaptive t-distribution, calculated according to formula (13); L + is a row vector with all dimensions being 1. To avoid the possibility that the fitness of the optimal individual deteriorates due to the randomness of mutation, a greedy rule is introduced to retain the better individual before and after mutation to the next generation.

[0132] In a possible implementation manner, update the positions of the individuals in the exploration group based on the current living environment and the current iteration number, including:

[0133] When t % 2 = 0, r3 < 0.5, and |A| < 1, update the positions of each individual in the exploration group based on the first formula; where the first formula is:

[0134] X Paddle (t + 1) = X * (t) - A|C.X * (t) - X(t)|

[0135] A = 2ar1 - a

[0136] c = 2r2

[0137] r1 = 0.5 + 0.5cos(tπ / T)

[0138] r2 = Levy(λ)

[0139] Among them, t is the number of iterations, r3 is the living environment, and X * (t) is the position of the optimal individual, A is the search step parameter, C is the search amplitude parameter, X(t) is the position of an individual at the t-th iteration, X(t + 1) is the updated position of this individual, a is the linear change parameter, r1 is the step adjustment parameter, r2 is the amplitude adjustment parameter, and λ is a preset value;

[0140] When t % 2 = 0, r3 < 0.5, and |A| ≥ 1, update the positions of other individuals in the exploration group based on the positions of randomly selected individuals in the exploration group and the second formula; among them, the second formula is:

[0141] X(t + 1) = w1 * X Paddle1 + z * w2 * (X Paddle2 - X Paddle3 ) + (1 - z) * w3 * (X(t) - X Paddle1 )

[0142] Among them, w1, w2, and w3 are random parameters, z is a decreasing parameter, and X Paddle1 , X Paddle2 and X Paddle3 are the positions of randomly selected individuals in the exploration group;

[0143] When t % 2 = 0 and r3 ≥ 0.5, update the positions of each individual in the exploration group based on the third formula; among them, the third formula is:

[0144] X(t + 1) = w4 * |X * (t) - X(t)|.e bl .cos(2πl) + [2w1(r4 + r5)] * X * (t)

[0145] Among them, b is a constant, and l, w4, r4, and e5 are random parameters;

[0146] When t % 2 ≠ 0, randomly select multiple individuals in the exploration group as individuals that have lost their mates, and update the positions of these individuals based on the patrol formula; among them, the patrol formula is:

[0147] X(t + 1) = X(t) + D(1 + z) * w * (X(t) - X Flock1 )

[0148]

[0149] Among them, X Flock1 is the position of the optimal individual, D is the step control function, w is a random parameter, and d is the dimension.

[0150] In this embodiment, when t % 2 = 0 (t % 2 means dividing the iteration algebra by 2 and then taking the remainder), the grey goose flock starts to work. When winter comes, that is, r3 < 0.5, the grey goose flock will migrate to a suitable place to survive. Among them, the value of r3 represents the change of the living environment, which randomly changes within [0, 1].

[0151] When |A| < 1, the position update of the agents in the exploration group is as follows:

[0152] X(t + 1) = X * (t) - A|C.X * (t) - X(t)| (8)

[0153] Among them, X(t) is the position of an agent based on t, t is the current iteration number, X * (t) is the position of the current optimal solution (leader), X(t + 1) is the position of the agent after update, A and C are two parameters, and the update method is as follows:

[0154]

[0155] Among them, the parameter a linearly changes from 2 to 0, and the values of r1 and r2 randomly change within [0, 1].

[0156] To prevent the grey goose flock from being completely dominated by the current agent, the grey goose flock will randomly select 3 individuals, namely X Paddle1 、X Paddle2 and X Paddle3 , to jointly decide the migration direction of the grey goose flock.

[0157] When |A| ≥ 1, the position update of the agents in the exploration group is as follows:

[0158] X(t + 1) = w1*X Paddle1 +z*w2*(X Paddle2 -X Paddle3 )+(1 - z)*w3*(X(t) - X Paddle1 )(10)

[0159] Among them, the values of w1, w2, and w3 are updated within [0, 2], and the parameter z decreases exponentially. The update method is as follows:

[0160]

[0161] Among them, t is the current iteration number, t max is the maximum iteration number.

[0162] When finding a more suitable living environment, that is, r3 ≥ 0.5, the grey goose flock will build nests and forage nearby. During the update, the values of a and A show a decreasing trend. At this time, the update process of the agents in the exploration group is as follows:

[0163] X(t + 1) = w4 * |X * (t) - X(t)|.e bl .cos(2πl) + [2w1(r4 + r5)] * X * (t) (12)

[0164] where b is a constant, l is a random value in [-1, 1]; the value of w4 is updated in [0, 2], and r4 and r5 are updated in [0, 1].

[0165] When t % 2 ≠ 0, the flock of greylag geese will rest. However, a greylag goose individual that has lost its mate will leave the flock and live alone, and needs to work continuously, day and night. Randomly select a greylag goose individual that has lost its mate in the exploration group, and the process of updating its position is as follows:

[0166] X(t + 1) = X(t) + D(1 + z) * w * (X(t) - X Flock1 ) (13)

[0167]

[0168] where, X Flock1 is the place where the flock of greylag geese inhabits, that is, the optimal solution.

[0169] In GGO, the migration route of the flock of greylag geese is mainly based on the flight direction of the leader (optimal solution), and it is more suitable to search for a living environment near this direction. And its influence parameters are the values of A and C, and the influence parameters of the values of A and C are the values of r1 and r2. The values of r1 and r2 randomly change in [0, 1], so the values of A and C are full of uncertainties, resulting in unstable optimization ability of the algorithm, low optimization efficiency, and slow convergence speed. The present invention uses Levy flight to keep the search with a large amplitude. When searching for a new position, it is necessary to mainly take the leader direction as the main, and make the change of the search amplitude mainly based on the new position. The calculation formula of r2 is:

[0170] r2 = Levy(λ) (15)

[0171] In the formula: λ is the Levy scaling parameter, and the Monte Carlo method is used to generate a random number r2 that follows the Levy distribution. The Levy distribution is a long-tailed distribution, with large jumps existing, and most jumps are relatively small, abandoning the characteristic that the tail of the normal distribution decays relatively fast. In the present invention, λ = 0.2, so that the search amplitude of the exploration group is less affected by the leader position as much as possible, so as to always maintain a large amplitude base.

[0172] In addition, parameter A is used in GGO to control the update step size of the population. The update is performed using A = 2ar1 - a, where a linearly varies from 2 to 0, and the value of A shows a decreasing trend, making the exploration direction extend towards the leader's position as much as possible. However, if parameter A decreases too quickly in the early stage, it will lead to insufficient global search ability of the algorithm, and if it decreases too slowly in the later stage, it will result in low local development accuracy. Therefore, the present invention proposes a non-linear cosine curve to solve this problem. The updated method of r1 after improvement is as follows:

[0173] r1 = 0.5 + 0.5cos(tπ / T) (16)

[0174] At this time, the value of A will non-linearly decrease from 2 to 1×10 -5 . This enables the algorithm to maintain a relatively large search step size in the early stage, improving the global search ability of the algorithm; and have a relatively small step size in the later stage of the search, improving the accuracy of local development of the algorithm.

[0175] Figure 5 is the iteration curve of the optimization process of the IGGO algorithm. It can be seen that the fitness values of individuals converge rapidly in the early stage of iteration and fluctuate slowly in the later stage.

[0176] In a possible implementation, the positions of individuals in the development group are updated based on the current survival environment and the current iteration number, including:

[0177] When t%2 ≠ 0, the positions of each individual in the development group are updated based on a randomly selected individual in the development group and the fourth formula; where the fourth formula is:

[0178] X1 = X Sentry1 - A1.|C1.X Sentry1 - X|

[0179] X2 = X Sentry2 - A2.|C2.X Sentry2 - X|

[0180] X3 = X Sentry3 - A3.|C3.X Sentry3 - X|

[0181]

[0182] A = 2ar1 - a

[0183] C = 2r2

[0184] r1 = 0.5 + 0.5cos(tπ / T)

[0185] r2 = Levy(λ)

[0186] Among them, X1, X2, and X3 are intermediate variables, X 1entry1 、XSentry2 and X Sentry3 is the position of an individual randomly selected from the development group, A is the search step parameter, C is the search amplitude parameter, X is the position of an individual at the t-th iteration, X(t + 1) is the position of the individual after update, a is the linearly varying parameter, r1 is the step adjustment parameter, r2 is the amplitude adjustment parameter, and λ is a preset value;

[0187] When t % 2 ≠ 0, update the position of each individual in the development group based on the patrol formula; where the patrol formula is:

[0188] X(t + 1) = X(t) + D(1 + z) * w * (X(t) - X Flock1 )

[0189]

[0190] where X Flock1 is the position of the optimal individual, D is the step control function, w is a random parameter, and d is the dimension.

[0191] In this embodiment, the individuals in the development group gather around the optimal individual in the population, foraging during the day and patrolling at night. When t % 2 = 0, the gray goose flock forages, and the development group will send out 3 individuals (sentinels), namely X Sentry1 , X Sentry2 and X Sentry3 , to guide the other individuals in the development group (X NonSentry ) to change their positions towards the estimated position of the prey. The update process is as follows:

[0192]

[0193] where the values of the variables A1, A2, A3, C1, C2, C3 are all calculated by formula (2).

[0194] When t % 2 ≠ 0, the gray goose flock will patrol. The gray goose flock will patrol around the habitat to prevent interference from natural enemies. The update method is as follows:

[0195] X(t + 1) = X(t) + D(1 + z) * w * (X(t) - X Flock1 ) (18)

[0196] where X Flock1 is the place where the gray goose flock is located, that is, the optimal solution.

[0197] In a possible implementation, based on the distribution entropy, signal reconstruction is performed on each IMF component to obtain the fault feature matrix of the distribution network, including:

[0198] Calculate the distribution entropy of each IMF component;

[0199] The IMF components with distribution entropy values in the interval [δ, 1] are used as effective IMF signals for reconstruction to obtain the fault feature matrix of the distribution network; where δ is a preset value.

[0200] In this embodiment, the distribution entropy (DistEn) is used to screen out the effective IMF components for signal reconstruction. The distribution entropy can well characterize the complexity of each intrinsic mode function (IMF) component and capture the change law of the IMF component. The calculation of the distribution entropy is based on the probability distribution and is not directly affected by the magnitude of the signal. As long as the probability distribution characteristics remain unchanged, the value of the distribution entropy is relatively stable. The formula for calculating the distribution entropy is as follows:

[0201]

[0202] In the formula, DistEn is the distribution entropy of the signal, and x i is the i-th signal of the signal sequence; n is the number of signals; p(xi) is the occurrence probability of the sample x i . The IMF components with distribution entropy values in the interval [δ, 1] are used as effective signals to participate in data reconstruction; the IMF components with distribution entropy values in the interval [0, δ] are directly discarded as invalid signals containing a large amount of random noise. In the present invention, δ is taken as 0.2.

[0203] In a possible implementation manner, before inputting the fault feature matrix into the trained CNN-SVM model, it further includes:

[0204] Building a distribution network simulation model considering the access of a high proportion of photovoltaic power sources;

[0205] Based on the distribution network simulation model, simulating various fault conditions and collecting the corresponding transient zero-sequence currents;

[0206] Using the fault feature matrices of the transient zero-sequence currents as training samples and the corresponding fault conditions as sample labels to train the initial CNN-SVM model to obtain the trained CNN-SVM model.

[0207] In this embodiment, building the simulation model includes: considering the access of a high proportion of photovoltaic power (setting the access quantity, location, etc. of the photovoltaic power sources), simulating the fault transient information under various conditions (such as changing the fault initial phase angle, transition resistance, fault location, etc.), obtaining the fault transient zero-sequence current, and the equivalent model is as Figure 6 shown.

[0208] After obtaining the training samples, use the fault labels to train the convolutional neural network to obtain the parameters of the convolutional neural network; then, transfer the parameters from the input layer to the flattening layer in the convolutional neural network, and use the output data of the Flatten layer to train the support vector machine; finally, replace the Softmax classifier in the convolutional neural network with the support vector machine to obtain a distribution network fault classification model. The training effect of the CNN-SVM model is as Figure 7 shown.

[0209] Among them, the key of SVM is to select a suitable kernel function. The Gaussian kernel function is easy to implement, widely used and has high accuracy. Therefore, the Gaussian kernel function is adopted to improve the classification performance of the fault feeder. The Gaussian kernel function is as follows:

[0210] k(x i ,x j )=exp[-g(x i -x j ) 2 (g>0) (20)

[0211] In the formula, is the kernel function variance, g=1 / (2σ 2 ), σ is the width parameter of the kernel function, and x j is the center of the Gaussian kernel function. The two main hyperparameters that determine the prediction performance of SVM are the penalty factor C and the kernel parameter g.

[0212] In a specific embodiment, as Figure 8 shown, the overall process of training the CNN-SVM and completing fault line selection includes:

[0213] Step 1: Build a distribution network simulation model considering the access of a high proportion of photovoltaic power sources.

[0214] Step 2: Simulate different working conditions and obtain the zero-sequence current of single-phase grounding faults.

[0215] Step 3: Use the IGGO-VMD algorithm to denoise and decompose the transient zero-sequence current of single-phase grounding faults, obtain the IMF components, use the distribution entropy (DistEn) to screen out the effective IMF components for signal reconstruction, and construct a fault feature matrix based on the effective IMF components.

[0216] Step 4: Divide the fault feature matrix into a training set and a test set, input the training set and the test set into the CNN for feature learning, and use the SVM to replace the softmax classifier in the CNN to obtain a classification model with excellent performance.

[0217] Step 5: Input the transient zero-sequence current of the actual distribution network fault into the CNN-SVM model to finally complete the fault line selection.

[0218] As can be seen from the above, the present invention has the following beneficial effects:

[0219] 1. The signal extraction technology adopted by the present invention for single-phase grounding fault line selection in a distribution network with high photovoltaic penetration rate is variational mode decomposition. This method has better effect and stronger robustness compared with traditional signal extraction methods. At the same time, an improved grey goose algorithm is used to optimize the k and α parameters in variational mode decomposition to make its decomposition effect optimal. By improving the Tent chaotic mapping to initialize the population, dynamic search based on Levy flight and non-linear step size strategy, and improved method of optimal individual perturbation based on adaptive t-distribution, the algorithm can overcome the problem that the grounding fault transient process caused by high photovoltaic penetration rate is more complex, and the performance of the algorithm is effectively improved.

[0220] 2. The present invention abandons the link of manual feature vector extraction, and the convolutional neural network in deep learning performs feature self-learning on the transient zero-sequence current of the fault. Compared with the method of manually extracting feature vectors, it can obtain higher classification accuracy; an SVM is used to replace the softmax classifier in the CNN to construct a CNN-SVM model. Compared with the softmax classifier, the SVM is more suitable for small sample problems such as distribution network faults, and when dealing with non-linear problems, the SVM has better generalization ability.

[0221] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution is prior or subsequent. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0222] The following is an apparatus embodiment of the present invention. For the details not described in detail therein, reference can be made to the corresponding method embodiments above.

[0223] Figure 9 The structural schematic diagram of the single-phase grounding fault line selection device for a distribution network with high photovoltaic penetration rate provided by the embodiment of the present invention is shown. For the convenience of description, only the parts related to the embodiment of the present invention are shown, and are described in detail as follows:

[0224] As Figure 9 shown, the single-phase grounding fault line selection device 9 for a distribution network with high photovoltaic penetration rate includes:

[0225] A parameter optimization module 91, configured to use the minimum envelope entropy as the optimization target, and optimize the decomposition layer number and penalty factor of the VMD model based on the IGGO algorithm to obtain an optimized VMD model;

[0226] A noise reduction decomposition module 92, configured to perform noise reduction decomposition on the transient zero-sequence current of the distribution network when a fault occurs based on the optimized VMD model to obtain multiple IMF components of the transient zero-sequence current;

[0227] The signal reconstruction module 93 is used to reconstruct the signals of each IMF component based on the distribution entropy to obtain the fault feature matrix of the distribution network;

[0228] The fault line selection module 94 is used to input the fault feature matrix into the trained CNN-SVM model to obtain the fault line selection result of the distribution network.

[0229] In a possible implementation manner, the parameter optimization module 91 is specifically used for:

[0230] Initializing the population; wherein, the population includes multiple individuals, and the position of each individual corresponds to a set of decomposition layers and penalty factors of the VMD model;

[0231] Dividing the population into an exploration group and a development group, and updating the positions of the individuals in the exploration group and the development group respectively based on the current living environment and the current iteration number;

[0232] Taking the envelope entropy as the fitness function, calculating the fitness function value based on the position of each individual, and determining the optimal individual in the population based on the fitness function value;

[0233] Repeating the above steps until the convergence condition is satisfied, taking the position of the current optimal individual as the optimization result, and applying it to the VMD model to obtain the optimized VMD model.

[0234] In a possible implementation manner, the parameter optimization module 91 is specifically used for:

[0235] Initializing the population based on the improved Tent chaotic mapping; wherein, the formula for initializing the population is:

[0236]

[0237] wherein, x n is the initial position of the nth individual, N is the population size, U b is the upper bound of the position, L b is the lower bound of the position, and rand(0,1) is a random number.

[0238] In a possible implementation manner, the parameter optimization module 91 is further used for:

[0239] Perturbing the position of the optimal individual based on the current iteration number and the adaptive t-distribution; wherein, the formula for perturbation is:

[0240]

[0241] wherein, H b is the position of the mutated optimal individual, X Flock1 is the position of the optimal individual, g(n) is the adaptive t-distribution with n degrees of freedom, L +A row vector with all dimensions being 1, t is the current iteration number, t max is the iteration number threshold.

[0242] In a possible implementation, the parameter optimization module 91 is specifically configured to:

[0243] When t % 2 = 0, r3 < 0.5, and |A| < 1, update the positions of each individual in the exploration group based on the first formula; where the first formula is:

[0244] X Paddle (t + 1) = X * (t) - A|C.X * (t) - X(t)|

[0245] A = 2ar1 - a

[0246] C = 2r2

[0247] r1 = 0.5 + 0.5cos(tπ / T)

[0248] r2 = Levy(λ)

[0249] where t is the iteration number, r3 is the living environment, X * (t) is the position of the optimal individual, A is the search step parameter, C is the search amplitude parameter, X(t) is the position of an individual at the t-th iteration, X(t + 1) is the updated position of this individual, a is the linear change parameter, r1 is the step adjustment parameter, r2 is the amplitude adjustment parameter, and λ is a preset value;

[0250] When t % 2 = 0, r3 < 0.5, and |A| ≥ 1, update the positions of other individuals in the exploration group based on the position of a randomly selected individual in the exploration group and the second formula; where the second formula is:

[0251] X(t + 1) = w1*X Paddle1 +z*w2*(X Paddle2 -X Paddle3 )+(1 - z)*w3*(X(t) - X Paddle1 )

[0252] where w1, w2, and w3 are random parameters, z is a decreasing parameter, X Paddle1 、X Paddle2 and X Paddle3 are the positions of randomly selected individuals in the exploration group;

[0253] When t % 2 = 0, r3 ≥ 0.5, update the positions of each individual in the exploration group based on the third formula; where the third formula is:

[0254] X(t + 1) = w4*|X* (t)-X(t)|.e bl .cos(2πl)+[2w1(r4+r5)]*X * (t)

[0255] Where b is a constant, and l, w4, r4, and r5 are random parameters;

[0256] When t % 2 ≠ 0, randomly select multiple individuals in the exploration group as individuals who have lost their partners, and update the positions of these individuals based on the patrol formula; where the patrol formula is:

[0257] X(t + 1) = X(t) + D(1 + z)*w*(X(t) - X Flock1 )

[0258]

[0259] Where X Flock1 is the position of the optimal individual, D is the step size control function, w is a random parameter, and d is the dimension.

[0260] In a possible implementation, the parameter optimization module 91 is specifically configured to:

[0261] When t % 2 ≠ 0, update the position of each individual in the development group based on the individuals randomly selected in the development group and the fourth formula; where the fourth formula is:

[0262] X1 = X Sentry1 -A1.|C1.X Sentry1 -X|

[0263] X2 = X Sentry2 -A2.|C2.X Sentry2 -X|

[0264] X3 = X Sentry3 -A3.|C3.X Sentry3 -X|

[0265]

[0266] A = 2ar1 - a

[0267] C = 2r2

[0268] r1 = 0.5 + 0.5cos(tπ / T)

[0269] r2 = Levy(λ)

[0270] Where X1, X2, and X3 are intermediate variables, X Sentry1 、X Sentry2 and X Sentry3is the position of an individual randomly selected from the development group, A is the search step parameter, C is the search amplitude parameter, X is the position of an individual at the t-th iteration, X(t + 1) is the position of the individual after update, a is the linearly varying parameter, r1 is the step adjustment parameter, r2 is the amplitude adjustment parameter, and λ is the preset value;

[0271] When t % 2 ≠ 0, update the positions of each individual in the development group based on the patrol formula; where the patrol formula is:

[0272] X(t + 1) = X(t) + D(1 + z) * w * (X(t) - X Flock1 )

[0273]

[0274] where, X Flock1 is the position of the optimal individual, D is the step control function, w is the random parameter, and d is the dimension.

[0275] In a possible implementation, the signal reconstruction module 93 is specifically used for:

[0276] Calculate the distribution entropy of each IMF component;

[0277] Use the IMF components with distribution entropy values in the interval [δ, 1] as effective IMF signals for reconstruction to obtain the fault feature matrix of the distribution network; where δ is the preset value.

[0278] In a possible implementation, the fault line selection module 94 is further used for:

[0279] Before inputting the fault feature matrix into the trained CNN - SVM model, build a distribution network simulation model considering the access of a high - proportion photovoltaic power source;

[0280] Simulate various fault conditions based on the distribution network simulation model and collect the corresponding transient zero - sequence currents;

[0281] Use the fault feature matrices of the transient zero - sequence currents as training samples and the corresponding fault conditions as sample labels to train the initial CNN - SVM model to obtain the trained CNN - SVM model.

[0282] In the embodiments of the present invention, the parameters of the VMD model are optimized by the IGGO algorithm, and a fault feature matrix is constructed through distribution entropy, which can accurately extract the zero-sequence current in the complex transient process of single-phase grounding faults for the stronger waveform and randomness of distributed photovoltaics. Through the CNN-SVM model, for the small sample problem of distribution network grounding faults, automatic feature learning is performed on the extracted effective signals, and automatic fault classification is realized, which can improve the accuracy of fault line selection technology and ensure the safe and stable operation of the distribution network and the reliability of power supply.

[0283] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For the parts not described in detail or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0284] Those of ordinary skill in the art can realize that the templates, units, and algorithm steps of the various examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0285] If the module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the methods of the above embodiments of the present invention, it can also be completed by a computer program instructing the relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above embodiments of the single-phase grounding fault line selection method for high photovoltaic penetration distribution networks can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0286] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A single-phase grounding fault line selection method for a distribution network with high photovoltaic penetration rate, characterized in that, Including: Taking the minimum envelope entropy as the optimization objective, optimizing the decomposition layer number and penalty factor of the VMD model based on the IGGO algorithm to obtain an optimized VMD model; Based on the optimized VMD model, performing noise reduction decomposition on the transient zero-sequence current when a fault occurs in the distribution network to obtain multiple IMF components of the transient zero-sequence current; Based on the distribution entropy, performing signal reconstruction on each IMF component to obtain the fault feature matrix of the distribution network; Inputting the fault feature matrix into a trained CNN-SVM model to obtain the fault line selection result of the distribution network.

2. The single-phase grounding fault line selection method for a distribution network with high photovoltaic penetration rate according to claim 1, wherein The step of taking the minimum envelope entropy as the optimization objective, optimizing the decomposition layer number and penalty factor of the VMD model based on the IGGO algorithm to obtain an optimized VMD model includes: Initializing the population; wherein, the population includes multiple individuals, and the position of each individual corresponds to a set of the decomposition layer number and penalty factor of the VMD model; Dividing the population into an exploration group and a development group, and updating the positions of the individuals in the exploration group and the development group respectively based on the current living environment and the current iteration number; Taking the envelope entropy as the fitness function, calculating the fitness function value based on the position of each individual, and determining the optimal individual in the population based on the fitness function value; Repeating the above steps until the convergence condition is satisfied, taking the position of the current optimal individual as the optimization result and applying it to the VMD model to obtain an optimized VMD model.

3. The single-phase grounding fault line selection method for a distribution network with high photovoltaic penetration rate according to claim 2, characterized in that, The step of initializing the population includes: Initializing the population based on the improved Tent chaotic mapping; wherein, the formula for initializing the population is: where x n is the initial position of the nth individual, N is the population size, U b is the upper bound of the position, L b is the lower bound of the position, and rand(0, 1) is a random number.

4. The single-phase grounding fault line selection method for a distribution network with high photovoltaic penetration rate according to claim 2, characterized in that, The step of taking the minimum envelope entropy as the optimization objective, optimizing the decomposition layer number and penalty factor of the VMD model based on the IGGO algorithm to obtain an optimized VMD model further includes: Perturbing the position of the optimal individual based on the current iteration number and the adaptive t-distribution; wherein, the formula for perturbation is: H b = X Flock1 + g(n)·L + Among them, H b is the position of the optimal individual after mutation, X Flock1 is the position of the optimal individual, g(n) is an adaptive t-distribution with n degrees of freedom, L + is a row vector with all dimensions being 1, t is the current iteration number, t max is the iteration number threshold.

5. The single-phase grounding fault line selection method for a distribution network with high photovoltaic penetration according to claim 2, characterized in that, Updating the positions of the individuals in the exploration group based on the current living environment and the current iteration number, including: When t%2 = 0, r3 < 0.5, |A| < 1, updating the position of each individual in the exploration group based on the first formula; wherein, the first formula is: X Paddle (t + 1)=X * (t)-A|C.X * (t)-X(t)| A = 2ar1 - a C＝2r2 r1 = 0.5 + 0.5cos(tπ / T) r2 = Levy(λ) where t is the number of iterations, r3 is the survival environment, and X * (t) is the position of the optimal individual, A is the search step parameter, C is the search amplitude parameter, X(t) is the position of an individual at the t-th iteration, X(t + 1) is the position of the individual after update, a is the linear change parameter, r1 is the step adjustment parameter, r2 is the amplitude adjustment parameter, and λ is a preset value; When t%2 = 0, r3 < 0.5, |A| ≥ 1, updating the positions of the other individuals in the exploration group based on the position of a randomly selected individual in the exploration group and the second formula; wherein, the second formula is: X(t + 1) = w1 * X Paddle1 + z * w2 * (X Paddle2 - X Paddle3 ) + (1 - z) * w3 * (X(t) - X Paddle1 ) where w1, w2, and w3 are random parameters, z is a decreasing parameter, and X Paddle1 , X Paddle2 , and X Pdddle3 are the positions of individuals randomly selected from the exploration group; When t%2 = 0, r3 ≥ 0.5, updating the position of each individual in the exploration group based on the third formula; wherein, the third formula is: X(t + 1) = w4 * |X * (t) - X(t)|.e bl .cos(2πl) + [2w1(r4 + r5)] * X * (t) Wherein, b is a constant, and l, w4, r4, and r5 are random parameters; When t%2 ≠ 0, randomly selecting multiple individuals in the exploration group as individuals losing their partners, and updating the positions of these individuals based on the patrol formula; wherein, the patrol formula is: X(t + 1) = X(t) + D(1 + z)*w*(X(t) - X Flock1 ) Among them, X Flock1 is the position of the optimal individual, D is the step size control function, w is the random parameter, and d is the dimension.

6. The single-phase grounding fault line selection method for a distribution network with high photovoltaic penetration rate according to claim 2, characterized in that, Updating the positions of the individuals in the development group based on the current living environment and the current iteration number, including: When t%2 ≠ 0, updating the position of each individual in the development group based on a randomly selected individual in the development group and the fourth formula; wherein, the fourth formula is: X1 = X Sentry1 -A1.|C1.X Sentry1 -X| X2 = X Sentry2 -A2·|C2·X Sentry2 -X X3 = X Sentry3 -A3·|C3·X Sentry3 -X A = 2ar1 - a C＝2r2 r1 = 0.5 + 0.5cos(tπ / T) r2 = Levy(λ) where t1, X2, and X3 are intermediate variables, X Sentry1 , X Sentry2 , and X Sentry3 are the positions of individuals randomly selected from the development group, A is the search step size parameter, C is the search amplitude parameter, X is the position of an individual at the t-th iteration, X(t + 1) is the position of the individual after update, a is the linearly varying parameter, r1 is the step size adjustment parameter, r2 is the amplitude adjustment parameter, and λ is a preset value; When t % 2 ≠ 0, update the position of each individual in the development group based on the patrol formula; where, the patrol formula is: X(t + 1) = X(t) + D(1 + z)*w*(X(t) - X Flock1 ) Among them, X Flock1 is the position of the optimal individual, D is the step size control function, w is a random parameter, and d is the dimension.

7. The single-phase grounding fault line selection method for a distribution network with high photovoltaic penetration rate according to claim 1, characterized in that, The signal reconstruction of each IMF component based on the distribution entropy to obtain the fault feature matrix of the distribution network includes: Calculate the distribution entropy of each IMF component; Use the IMF components with the distribution entropy value in the interval [δ, 1] as effective IMF signals for reconstruction to obtain the fault feature matrix of the distribution network; where, δ is a preset value.

8. The single-phase grounding fault line selection method for a distribution network with high photovoltaic penetration rate according to claim 1, wherein, Before inputting the fault feature matrix into the trained CNN - SVM model, it also includes: Build a distribution network simulation model considering the access of a high - proportion photovoltaic power source; Simulate various fault conditions based on the distribution network simulation model and collect the corresponding transient zero - sequence currents; Use the fault feature matrices of each transient zero - sequence current as training samples and the corresponding fault conditions as sample labels to train the initial CNN - SVM model to obtain the trained CNN - SVM model.

9. A single-phase grounding fault line selection device for a distribution network with high photovoltaic penetration rate, characterized in that, It includes: A parameter optimization module, which is used to take the minimum envelope entropy as the optimization goal and optimize the decomposition layer number and penalty factor of the VMD model based on the IGGO algorithm to obtain an optimized VMD model; A noise reduction decomposition module, which is used to perform noise reduction decomposition on the transient zero - sequence current when the distribution network fails based on the optimized VMD model to obtain multiple IMF components of the transient zero - sequence current; A signal reconstruction module, which is used to perform signal reconstruction on each IMF component based on the distribution entropy to obtain the fault feature matrix of the distribution network; A fault line selection module, which is used to input the fault feature matrix into the trained CNN - SVM model to obtain the fault line selection result of the distribution network.

10. The single-phase grounding fault line selection device for a distribution network with high photovoltaic penetration rate according to claim 9, characterized in that, The parameter optimization module is specifically used for: The process of taking the minimum envelope entropy as the optimization goal and optimizing the decomposition layer number and penalty factor of the VMD model based on the IGGO algorithm to obtain an optimized VMD model includes: Initialize the population; where, the population includes multiple individuals, and the position of each individual corresponds to a set of the decomposition layer number and penalty factor of the VMD model; Divide the population into an exploration group and a development group, and update the positions of the individuals in the exploration group and the development group respectively based on the current living environment and the current iteration number; Take the envelope entropy as the fitness function, calculate the fitness function value based on the position of each individual, and determine the optimal individual in the population based on the fitness function value; Repeat the above steps until the convergence condition is met, take the position of the current optimal individual as the optimization result, and apply it to the VMD model to obtain an optimized VMD model.

Images