A distribution network fault analysis method based on multi-source information fusion

Abstract

The present invention provides a distribution network fault analysis method based on multi-source information fusion, comprising: step 1, distribution network fault area division: fault area division is performed based on the static topology information of the distribution network in the equipment asset management system and in combination with the distribution automation information; step 2, switch quantity information analysis of the distribution network fault area: by constructing a fault diagnosis model, parameter optimization is performed based on a quantum genetic algorithm to obtain the fuzzy fault degree of the switch quantity information; step 3, electrical quantity information analysis of the distribution network fault area: feature extraction of current information in the fault area is performed based on the Hilbert-Huang Transform (HHT) to obtain the amplitude distortion degree and the fault energy degree; step 4, distribution network fault information fusion decision: based on the spectral clustering algorithm, the fuzzy fault degree, amplitude distortion degree and fault energy degree of the component are comprehensively analyzed to realize the analysis of the fault component.

Description

Technical Field

[0001] The present invention relates to the field of distribution network fault diagnosis, and in particular to a distribution network fault analysis method based on multi-source information fusion. Background Art

[0002] Compared with the current power supply construction and power transmission and transformation construction, the construction of distribution network is relatively lagging behind, which is not in line with the high requirements of distribution network for the current large-scale urbanization development in my country. As an important link in the production, transmission and use of electric energy, the distribution system is the key link between the actual user demand side and the power generation and transmission system. Therefore, after a failure occurs in the distribution network, rapid and reasonable positioning to effectively guide the maintenance work is a key part of the current safe and sustainable development of the urban power grid.

[0003] At present, the basis for judging the fault component information in power grid fault diagnosis mainly comes from the basic topology information of the power grid, protection action information, etc. This type of judgment method is mostly based on the basic information of the switch quantity to judge the fault component, while ignoring the influence of the electrical quantity information on the fault diagnosis during the fault. Due to the risk of malfunction of the switchgear in the relay protection system, only using the switch information quantity to judge the fault information has certain limitations, which will lead to the inaccuracy of the fault diagnosis results and even possible misjudgment. Therefore, it is of great practical significance to carry out research on the distribution network fault judgment method based on multi-source data to provide method support for improving the power supply reliability of the distribution network. To realize the distribution network fault judgment based on multi-source data, the following problems need to be solved: 1) How to extract the different fault information characteristics of the distribution network based on multi-source information; 2) How to fuse the multi-source fault judgment information to achieve the final decision of fault diagnosis. Summary of the invention

[0004] In order to make up for the deficiencies in the prior art, the present invention provides a distribution network fault analysis method based on multi-source information fusion.

[0005] The method comprises:

[0006] Step 1: Division of distribution network fault areas

[0007] When a fault occurs in the distribution network, the search scope is mainly reflected in the actions of related switch elements and protection devices. The fault area can be obtained by retrieving the static topology information of the distribution network in the equipment asset management system for comparison before and after the fault and combining the switch information of the distribution automation system. The fault area is a passive area due to the isolation effect of the switch elements and protection devices. The faulty element is also included in this passive power-off area. Therefore, the location and identification of the faulty element is mainly carried out in this passive area.

[0008] Step 2 Analysis of switch quantity information in the fault area of ​​the distribution network

[0009] Aiming at the switch quantity information of the distribution network, a diagnostic model based on optimization technology is used to creatively transform the power grid fault diagnosis problem into a 0-1 integer programming problem for finding extreme values, thus having a rigorous mathematical foundation. Then, a quantum genetic algorithm is used to find the optimal solution of the model, effectively realizing the distribution network fault diagnosis.

[0010] The step 2 comprises the following steps:

[0011] Step 21: Construct a fault diagnosis model based on the actual and expected values ​​of the component action. The specific model is:

[0012]

[0013] Where: r k,m and are the actual state and expected state of the main protection of a certain device; r k,s and are the actual state and expected state of the backup protection of a certain device respectively; k,l and are the actual state and expected state of the remote backup protection of a certain device respectively; C i and C i * are the actual state and expected state of the circuit breaker respectively.

[0014] Step 2.2: Use quantum genetic algorithm to solve the fault diagnosis model and obtain the fault component set;

[0015] Step 2.2.1: Initialize the fault region component population Q(t0) and randomly generate n chromosomes encoded with quantum bits;

[0016] Step 2.2.2: Perform a decoding measurement on each individual in the initialized population Q(t0) to obtain a set of corresponding definite solutions in is the jth solution in the t0th generation population, expressed as a binary string of length m, where each bit is 0 or 1, according to the probability of the quantum bit ( or ) is selected. The measurement process is to randomly generate a number in the interval [0,1]. If it is greater than the square of the probability amplitude, the measurement result is 1, otherwise it is 0. The fitness of the determined solution is evaluated, and the optimal individual and the corresponding fitness are recorded; the fitness function adopts the objective function, that is, Θ(x) = E(x).

[0017] Step 2.2.3: Use the quantum rotating door U(t) to adjust the individuals and obtain a new population Q(t); this adjustment strategy is to adjust the individuals The fitness of the current measured value Θ(x) is equal to the fitness value of the current best individual in the population Θ(best i ) is compared, if Θ(x)>Θ(best i ), then adjust The corresponding quantum bits in the i ,β i ) towards the direction that is favorable to x i Evolution in the direction of appearance, on the contrary, adjustment The corresponding quantum bits in the i ,β i ) Towards the best i Evolution in the direction of emergence.

[0018] Step 2.2.4: Increase the number of iterations t by 1 until the set number of iterations is reached. The optimal solution is obtained, thereby determining the faulty component set.

[0019] Step 3: Analysis of electrical quantity information in the fault area of ​​the distribution network

[0020] When a distribution network fails, the electrical quantities of the components in the system will have transient components. Therefore, the electrical quantity failure degree of the components is obtained by analyzing the electrical quantities of the set of suspected fault components obtained in step 1.

[0021] The step 3 comprises the following steps:

[0022] Step 3.1: Perform a fast eigenmode decomposition of the current information through the component

[0023] Step 3.1.1: Let the input current signal be f(t), and calculate the extreme point R(t) of the signal:

[0024] R(t),t0≤t j ≤t m-1 (2)

[0025] In the formula, t0, t j and t m-1 They represent the sampling moments of different extreme points respectively.

[0026] Step 3.1.2: Use the linear transformation method to transform the extreme points and use the residual r(t) in pairs:

[0027]

[0028] In the formula, t j+1 is the sampling time corresponding to the j+1 extreme points.

[0029] Step 3.1.3: Subtract the residual from the original signal to obtain the intrinsic modal function IMF of the original signal. Therefore, the original input signal can be decomposed into multiple unique intrinsic modal components and a residual component.

[0030] Step 3.2: Hilbert transform the modal function IMF

[0031] Assume that a certain modal function IMF real signal is u(t), and its corresponding Hilbert transform is:

[0032]

[0033] The corresponding inverse transform is:

[0034]

[0035] Combining u(t) and v(t) gives the following complex signal:

[0036] x(t)=u(t)+jv(t)=a(t)e jθt (6)

[0037] According to formula (6), we can get:

[0038]

[0039] Then its instantaneous frequency can be defined as:

[0040]

[0041] Therefore, the instantaneous frequency of the original signal u(t) is the phase derivative of the corresponding complex signal x(t).

[0042] Step 3.3: Calculate the amplitude distortion of the component current information

[0043] Through formula (7), a certain modal function can be obtained, and after Hilbert transformation, the jth FIMD component of the current effective value amplitude of the i-th line before and after the fault occurs can be obtained as follows:

[0044]

[0045] Where N is the number of sampling points, and l is the number of sampling points corresponding to the fault moment.

[0046] Then the distortion degree of the fault line current signal can be further obtained as:

[0047]

[0048] Step 3.4: Calculate the component current information fault energy metric;

[0049] According to formula (8), the energy spectrum H(ω,t) of the fault current signal can be obtained, and then the corresponding marginal spectrum can be obtained as follows:

[0050]

[0051] The marginal energy spectrum of Hilbert transform is further obtained as:

[0052]

[0053] Then the current fault energy measurement value of the i-th line at fault can be expressed as:

[0054]

[0055] In the formula, f s Represents the signal sampling frequency.

[0056] Step 4. Distribution network fault information fusion decision

[0057] Through steps 2 and 3, we obtain three kinds of information: component fuzzy fault degree, current information amplitude distortion degree, and current information fault energy degree, which are used as the fault characteristics of each component in the fault area. We use spectral clustering to fuse the fault information and achieve the final decision of fault diagnosis. The specific steps are as follows:

[0058] Step 4.1: Input sample set V = (x1, x2, x3…, x n ), construct the degree matrix D n×n and the adjacency matrix W.

[0059]

[0060] w ij For point x i and point x j Since it is an undirected graph, w ij =w ji .d i For x i The sum of the weights of all connected edges. Matrix D is a diagonal matrix, and only the main diagonal has values, corresponding to the degree d of the i-th point in the i-th row i W is an n×n matrix, and the jth value in the i-th row corresponds to the weight w ij .

[0061] Step 4.2: Use Ncut theory to perform graph cutting clustering and identify faulty components.

[0062] Step 4.2.1: Construct the graph-cutting clustering target E:

[0063]

[0064] where is the complement of A, |A i | is the number of sample points in the sub - graph, and A i is the set of each sub - graph point. i

[0065] Step 4.2.2: Perform the transformation and solution of the objective function E:

[0066]

[0067] Step 4.2.3: Construct the standardized Laplacian matrix D -1 / 2 LD -1 / 2 .

[0068] Step 4.2.4: Calculate the eigenvectors f corresponding to the smallest k1 (k1 << n) eigenvalues of D -1 / 2 LD -1 / 2 respectively.

[0069] Step 4.2.5: Standardize the matrix composed of the eigenvectors f corresponding to each row, and finally form an n×k1 - dimensional feature matrix F.

[0070] Step 4.2.6: Take each row in F as a k1 - dimensional sample, with a total of n samples, and use the traditional k - means clustering method for clustering, with the clustering dimension being k2, to obtain the cluster partition C(c1, c2,... c k2 ), and the outliers are the faulty components.

[0071] Compared with the closest prior art, the excellent effects of the present invention are:

[0072] 1. By establishing a mathematical model for fault diagnosis and using the quantum genetic algorithm for model solution, the present invention obtains the fuzzy fault degree of the faulty component set, realizes the quantification of suspicious faulty components, and has certain applicability in practical applications.

[0073] 2. By using the HHT algorithm for analyzing the electrical quantity information of faulty components, the present invention makes up for the limitation of fault diagnosis only from the perspective of switch components. At the same time, by using the spectral clustering algorithm to fuse multi - source information for fault judgment and decision - making, the objectivity and accuracy of the decision - making are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 is the flow chart of the present invention.

[0075] Figure 2 is the distribution network topology diagram used in the implementation case of the present invention.

[0076] Figure 3 ​This is the optimization graph of the quantum genetic algorithm in the implementation case of the present invention.

[0077] Figure 4 This is a clustering diagram of multi-source data fusion decision in an implementation case of the present invention. DETAILED DESCRIPTION

[0078] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is described in detail below in conjunction with the accompanying drawings and implementation examples. It should be understood that the specific implementation examples described herein are only used to explain the present invention and are not used to limit the invention.

[0079] like Figure 1 As shown in the figure, a distribution network fault analysis method based on multi-source information fusion is shown in the figure. Figure 1 As shown, the specific steps include:

[0080] Step 1: Division of distribution network fault areas

[0081] When a fault occurs in the distribution network, the search scope is mainly reflected in the actions of related switch elements and protection devices. The fault area can be obtained by retrieving the static topology information of the distribution network in the equipment asset management system for comparison before and after the fault and combining the switch information of the distribution automation system. The fault area is a passive area due to the isolation effect of the switch elements and protection devices. The faulty element is also included in this passive power-off area. Therefore, the location and identification of the faulty element is mainly carried out in this passive area.

[0082] Step 2 Analysis of switch quantity information in the fault area of ​​the distribution network

[0083] Aiming at the switch quantity information of the distribution network, a diagnostic model based on optimization technology is used to creatively transform the power grid fault diagnosis problem into a 0-1 integer programming problem for finding extreme values, thus having a rigorous mathematical foundation. Then, a quantum genetic algorithm is used to find the optimal solution of the model, effectively realizing the distribution network fault diagnosis.

[0084] The step 2 comprises the following steps:

[0085] Step 21: Construct a fault diagnosis model based on the actual and expected values ​​of the component action. The specific model is:

[0086]

[0087] Where: r k,m and are the actual state and expected state of the main protection of a certain device; r k,s and are the actual state and expected state of the backup protection of a certain device respectively; k,l and are the actual state and expected state of the remote backup protection of a certain device respectively; C i and are the actual state and expected state of the circuit breaker respectively.

[0088] Step 2.2: Use quantum genetic algorithm to solve the fault diagnosis model and obtain the fault component set;

[0089] Step 2.2.1: Initialize the fault region component population Q(t0) and randomly generate n chromosomes encoded with quantum bits;

[0090] Step 2.2.2: Perform a decoding measurement on each individual in the initialized population Q(t0) to obtain a set of corresponding definite solutions in is the jth solution in the t0th generation population, expressed as a binary string of length m, where each bit is 0 or 1, according to the probability of the quantum bit ( or ) is selected. The measurement process is to randomly generate a number in the interval [0,1]. If it is greater than the square of the probability amplitude, the measurement result is 1, otherwise it is 0. The fitness of the determined solution is evaluated, and the optimal individual and the corresponding fitness are recorded; the fitness function adopts the objective function, that is, Θ(x) = E(x).

[0091] Step 2.2.3: Use the quantum rotating door U(t) to adjust the individuals and obtain a new population Q(t); this adjustment strategy is to adjust the individuals The fitness of the current measured value Θ(x) is equal to the fitness value of the current best individual in the population Θ(best i ) is compared, if Θ(x)>Θ(best i ), then adjust The corresponding quantum bits in the i ,β i ) towards the direction that is favorable to x i Evolution in the direction of appearance, on the contrary, adjustment The corresponding quantum bits in the i ,β i ) Towards the best i Evolution in the direction of emergence.

[0092] Step 2.2.4: Increase the number of iterations t by 1 until the set number of iterations is reached. The optimal solution is obtained, thereby determining the faulty component set.

[0093] Step 3: Analysis of electrical quantity information in the fault area of ​​the distribution network

[0094] When a distribution network fails, the electrical quantities of the components in the system will have transient components. Therefore, the electrical quantity failure degree of the components is obtained by analyzing the electrical quantities of the set of suspected fault components obtained in step 1.

[0095] The step 3 comprises the following steps:

[0096] Step 3.1: Perform a fast eigenmode decomposition of the current information through the component

[0097] Step 3.1.1: Let the input current signal be f(t), and calculate the extreme point R(t) of the signal:

[0098] R(t),t0≤t j ≤t m-1 (2)

[0099] In the formula, t0, t j and t m-1 They represent the sampling moments of different extreme points respectively.

[0100] Step 3.1.2: Use the linear transformation method to transform the extreme points and use the residual r(t) in pairs:

[0101]

[0102] In the formula, t j+1 is the sampling time corresponding to the j+1 extreme points.

[0103] Step 3.1.3: Subtract the residual from the original signal to obtain the intrinsic modal function IMF of the original signal. Therefore, the original input signal can be decomposed into multiple unique intrinsic modal components and a residual component.

[0104] Step 3.2: Hilbert transform the modal function IMF

[0105] Assume that a certain modal function IMF real signal is u(t), and its corresponding Hilbert transform is:

[0106]

[0107] The corresponding inverse transform is:

[0108]

[0109] Combining u(t) and v(t) gives the following complex signal:

[0110] x(t)=u(t)+jv(t)=a(t)e jθt (6)

[0111] According to formula (6), we can get:

[0112]

[0113] Then its instantaneous frequency can be defined as:

[0114]

[0115] Therefore, the instantaneous frequency of the original signal u(t) is the phase derivative of the corresponding complex signal x(t).

[0116] Step 3.3: Calculate the amplitude distortion of the component current information

[0117] Through formula (7), a certain modal function can be obtained, and after Hilbert transformation, the jth FIMD component of the current effective value amplitude of the i-th line before and after the fault occurs can be obtained as follows:

[0118]

[0119] Where N is the number of sampling points, and l is the number of sampling points corresponding to the fault moment.

[0120] Then the distortion degree of the fault line current signal can be further obtained as:

[0121]

[0122] Step 3.4: Calculate the component current information fault energy metric;

[0123] According to formula (8), the energy spectrum H(ω,t) of the fault current signal can be obtained, and then the corresponding marginal spectrum can be obtained as follows:

[0124]

[0125] The marginal energy spectrum of Hilbert transform is further obtained as:

[0126]

[0127] Then the current fault energy measurement value of the i-th line at fault can be expressed as:

[0128]

[0129] In the formula, f s Represents the signal sampling frequency.

[0130] Step 4. Distribution network fault information fusion decision

[0131] The component fuzzy fault degree, the amplitude distortion degree of the current information, and the fault energy degree of the current information are obtained through Step 2 and Step 3, and these three types of information are used as the fault characteristics of each component in the fault area. Spectral clustering is used to fuse the fault information to achieve the final decision of fault diagnosis. The specific steps are as follows:

[0132] Step 4.1: Input the sample set V = (x1, x2, x3…, x n ), and construct the degree matrix D n×n and the adjacency matrix W.

[0133]

[0134] w ij is the weight between point x i and point x j . Since it is an undirected graph, so w ij = w ji . d i is the sum of the weights of all the edges connected to x i . The matrix D is a diagonal matrix with values only on the main diagonal, corresponding to the degree d i of the i-th point in the i-th row. W is an n×n matrix, and the value of the j-th element in the i-th row corresponds to the weight w ij .

[0135] Step 4.2: Use the Ncut theory to perform graph cutting clustering to identify the faulty components.

[0136] Step 4.2.1: Construct the graph cutting clustering objective E:

[0137]

[0138] where is the complement of A i , |A i | is the number of sample points in the subgraph, and A i is the set of points in each subgraph.

[0139] Step 4.2.2: Perform the transformation and solution of the objective function E:

[0140]

[0141] Step 4.2.3: Construct the normalized Laplacian matrix D -1 / 2 LD -1 / 2 .

[0142] Step 4.2.4: Calculate the k1 (k1 << n) smallest eigenvalues of D -1 / 2 LD -1 / 2 and their corresponding eigenvectors f.

[0143] Step 4.2.5: Normalize the matrix composed of the corresponding eigenvectors f by row to finally form an n×k1-dimensional feature matrix F.

[0144] Step 4.2.6: Take each row in F as a k1-dimensional sample, a total of n samples, and cluster them using the traditional k-means clustering method with a clustering dimension of k2 to obtain the cluster partition C (c1, c2, ... c k2 ), the abnormal point is the faulty component.

[0145] Implementation Cases

[0146] 1) Division of distribution network fault areas

[0147] When a distribution network fails, the search scope is mainly reflected in the actions of related switching elements and protection devices. The fault area can be obtained by retrieving the static topology information of the distribution network in the equipment asset management system for comparison before and after the failure, and combining it with the distribution automation information.

[0148] The distribution network topology used in this case is as follows Figure 2 In the topology diagram, T1~T6 are main transformers, B1~B6 are 10KV busbars, L1~L 21 For feeder, CB1~CB 40 The feeder connection mode in the topology diagram is "hand in hand". Each line includes remote backup protection, so the topology diagram includes 99 protections in total. Including the main protection T 1m ~T 6m， B 1m ~B 6m ,L 1m ~L 21m , the near backup protection and the far backup protection are the same, and are represented by s and l. When a fault occurs in the distribution network, the protection action information T collected by the distribution automation information 1m ,T 3s ,B 3l ,L 9s Action, circuit breaker CB1, CB 11 ,CB 12 ,CB 17 ,CB 20 ,CB 21 ,CB 34 Combined with the topology information and switch information, the fault diagnosis components determined are T1, T3, B1, B3, L1~L4, L9, L 10 , L 12 , L 13 , the state vector corresponding to the above elements is E = [e1, e2, e3, ···, e 12 The circuit breakers that need to be diagnosed are: CB1, CB4~CB7, CB16 ~CB 21 The actual state vector of the above circuit breaker is C = [c1, c2, c3, ···c 12 ]=[1,0,0,0,0,0,0,1,0,0,1,1]. The protection that needs to be diagnosed for fault is, T 1m ,T 1s ,T 1l ,T 3m ,T 3s ,T 3l ,B 1m ,B 1s ,B 1l ,B 3m ,B 3s ,B 3l ,L 1m ~L 4m ,L 1s ~L 4s ,L 1l ~L 4l ,L 9m , L 9s , L 9l , L 10m , L 10s , L 10l , L 12m , L 12l , L 12s , L 13m ,L 13s ,L 13l , the corresponding protection actual state vector R = [r1, r2, r3···r 36 ]=[1,0,0,0,1,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0].

[0149] Step 2 Analysis of switch quantity information in the fault area of ​​the distribution network

[0150] Step 21: Construct a fault diagnosis model based on the actual and expected values ​​of the component action. The specific model is:

[0151]

[0152] Where: r k,m and are the actual state and expected state of the main protection of a certain device respectively; k,s and are the actual state and expected state of the backup protection of a certain device respectively; k,l and are the actual state and expected state of the remote backup protection of a certain device respectively; Ci and are the actual state and expected state of the circuit breaker respectively.

[0153] Main protection Near backup protection Remote backup protection * k,l =1-[1-the state of the associated element i×(1-the state of the circuit breaker j on the associated path)]; Circuit breaker C i * =max{action expectation of circuit breaker related protection × actual state value of the protection}.

[0154] Based on the above calculation formula of the expected value of the state, the expected value of the protection that needs to be diagnosed in the case is shown in the following table:

[0155]

[0156] Step 2.2: Use quantum genetic algorithm to solve the fault diagnosis model and obtain the fault component set;

[0157] Step 2.2.1: Initialize the fault region component population Q(t0) and randomly generate n chromosomes encoded with quantum bits;

[0158] Step 2.2.2: Perform a decoding measurement on each individual in the initialized population Q(t0) to obtain a set of corresponding definite solutions in is the jth solution in the t0th generation population, expressed as a binary string of length m, where each bit is 0 or 1, according to the probability of the quantum bit ( or ) is selected. The measurement process is to randomly generate a number in the interval [0,1]. If it is greater than the square of the probability amplitude, the measurement result is 1, otherwise it is 0. The fitness of the determined solution is evaluated, and the optimal individual and the corresponding fitness are recorded; the fitness function adopts the objective function, that is, Θ(x) = E(x).

[0159] Step 2.2.3: Use the quantum rotating door U(t) to adjust the individuals and obtain a new population Q(t); this adjustment strategy is to adjust the individuals The fitness of the current measured value Θ(x) is equal to the fitness value of the current best individual in the population Θ(best i ) is compared, if Θ(x)>Θ(best i ), then adjust The corresponding quantum bits in the i ,β i ) towards the direction that is favorable to x i Evolution in the direction of appearance, on the contrary, adjustment The corresponding quantum bits in the i ,β i ) Towards the best i Evolution in the direction of emergence.

[0160] Step 2.2.4: Increase the number of iterations t by 1 until the set number of iterations is reached. The optimal solution is obtained, thereby determining the faulty component set.

[0161] like Figure 3 The figure shows the changes of the optimal fitness, average fitness, and median fitness during the iteration process of quantum optimization. After 250 iterations, the quantum genetic algorithm was used to solve the problem. According to the optimal fitness function, the minimum value of E(x) is 8. At the same time, there are 8 groups of optimal solutions through the iteration process. According to the frequency of each component in the optimal solution, regularization can be performed to obtain suspicious components T1, T3, B1, B3, L1~L4, L9, L 10 , L 12 , L 13 The failure degree is [1,0.8,0.2,0,0,0,0.4,0,0.8,0,0,0].

[0162] Step 3: Analysis of electrical quantity information in the fault area of ​​the distribution network

[0163] When a distribution network fails, the electrical quantities of the components in the system will have transient components. Therefore, the electrical quantity failure degree of the components is obtained by analyzing the electrical quantities of the set of suspected fault components obtained in step 1.

[0164] The step 3 comprises the following steps:

[0165] Step 3.1: Perform a fast eigenmode decomposition of the current information through the component

[0166] Step 3.1.1: Let the input current signal be f(t), and calculate the extreme point R(t) of the signal:

[0167] R(t),t0≤t j ≤t m-1 (2)

[0168] Where t0, t j and t m-1 They represent the sampling moments of different extreme points respectively.

[0169] Step 3.1.2: Use the linear transformation method to transform the extreme points and use the residual r(t) in pairs:

[0170]

[0171] Where, tj+1 is the sampling time corresponding to the j+1 extreme points.

[0172] Step 3.1.3: Subtract the residual from the original signal to obtain the intrinsic modal function IMF of the original signal. Therefore, the original input signal can be decomposed into multiple unique intrinsic modal components and a residual component.

[0173] Step 3.2: Hilbert transform the modal function IMF

[0174] Assume that a certain modal function IMF real signal is u(t), and its corresponding Hilbert transform is:

[0175]

[0176] The corresponding inverse transform is:

[0177]

[0178] Combining u(t) and v(t) gives the following complex signal:

[0179] x(t)=u(t)+jv(t)=a(t)e jθt (6)

[0180] According to formula (6), we can get:

[0181]

[0182] Then its instantaneous frequency can be defined as:

[0183]

[0184] Therefore, the instantaneous frequency of the original signal u(t) is the phase derivative of the corresponding complex signal x(t).

[0185] Step 3.3: Calculate the amplitude distortion of the component current information

[0186] Through formula (7), a certain modal function can be obtained, and after Hilbert transformation, the jth FIMD component of the current effective value amplitude of the i-th line before and after the fault occurs can be obtained as follows:

[0187]

[0188] Where N is the number of sampling points, and l is the number of sampling points corresponding to the fault moment.

[0189] Then the distortion degree of the fault line current signal can be further obtained as:

[0190]

[0191] Step 3.4: Calculate the component current information fault energy metric;

[0192] According to formula (8), the energy spectrum H(ω,t) of the fault current signal can be obtained, and then the corresponding marginal spectrum can be obtained as follows:

[0193]

[0194] The marginal energy spectrum of Hilbert transform is further obtained as:

[0195]

[0196] Then the current fault energy measurement value of the i-th line at fault can be expressed as:

[0197]

[0198] In the formula, f s Represents the signal sampling frequency.

[0199] By passing through the faulty components T1, T3, B1, B3, L1~L4, L9, L 10 , L 12 , L 13 After HHT transformation of the current information, the amplitude distortion degree and fault energy degree of each component can be obtained. The results obtained after normalization are shown in the following table:

[0200] Suspected fault component <![CDATA[T1]]> <![CDATA[T3]]> <![CDATA[B1]]> <![CDATA[B3]]> <![CDATA[L1]]> <![CDATA[L2]]> <![CDATA[L3]]> <![CDATA[L4]]> <![CDATA[L9]]> <![CDATA[L 10 ]]> <![CDATA[L 12 ]]> <![CDATA[L 13 ]]> Amplitude distortion 0.856 0.798 0.263 0.162 0.013 0.032 0.062 0.103 0.861 0.023 0.121 0.034 Fault Energy Measurement 0.726 0.869 0.164 0.127 0.049 0.036 0.012 0.121 0.914 0.101 0.034 0.026

[0201] Step 4. Distribution network fault information fusion decision

[0202] Through steps 2 and 3, we obtain three kinds of information: component fuzzy fault degree, current information amplitude distortion degree, and current information fault energy degree, which are used as the fault characteristics of each component in the fault area. We use spectral clustering to fuse the fault information and achieve the final decision of fault diagnosis. The specific steps are as follows:

[0203] Step 4.1: Input sample set V = (x1, x2, x3…, x n ), construct the degree matrix D n×n and the adjacency matrix W.

[0204]

[0205]

[0206] w ij For point x i and point x j Since it is an undirected graph, w ij =wji 。d i is x i The sum of the weights of all the edges connected. Matrix D is a diagonal matrix with values only on the main diagonal, corresponding to the degree d of the i-th point in the i-th row i 。W is an n×n matrix, and the value of the j-th element in the i-th row corresponds to the weight w ij 。

[0207] Step 4.2: Use the Ncut theory for graph cutting clustering to identify faulty components.

[0208] Step 4.2.1: Construct the graph cutting clustering objective E:

[0209]

[0210] where is the complement of A i , |A i | is the number of sample points in the subgraph, and A i is the set of points in each subgraph.

[0211] Step 4.2.2: Perform the transformation and solution of the objective function E:

[0212]

[0213] Step 4.2.3: Construct the normalized Laplacian matrix D -1 / 2 LD -1 / 2 。

[0214] Step 4.2.4: Calculate the eigenvectors f corresponding to the k1 (k1 << n) smallest eigenvalues of D -1 / 2 LD -1 / 2 respectively.

[0215] Step 4.2.5: Normalize the matrix composed of the eigenvectors f corresponding to each row, and finally form an n×k1-dimensional eigenmatrix F.

[0216] Step 4.2.6: Take each row in F as a k1-dimensional sample, with a total of n samples, and use the traditional k-means clustering method for clustering, with the clustering dimension being k2, to obtain the cluster partition C(c1, c2,... c k2 ), and the outliers are the faulty components. The clustering result is as shown in Figure 4 . The final fault diagnosis result shows that the faulty components are T1, T3, and L9.

[0217] The above implementation scheme is only the technical idea of ​​the present invention, and the protection scope of the present invention cannot be limited thereto. Any changes made on the basis of the technical scheme according to the technical idea proposed by the present invention shall fall within the protection scope of the present invention. The above implementation scheme is only the technical idea of ​​the present invention, and the protection scope of the present invention cannot be limited thereto. Any changes made on the basis of the technical scheme according to the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A distribution network fault analysis method based on multi-source information fusion, the method comprising: Step 1: Division of distribution network fault areas: When a fault occurs in the distribution network, the search scope is mainly reflected in the actions of related switch elements and protection devices. The fault area can be obtained by retrieving the static topology information of the distribution network in the equipment asset management system for comparison before and after the fault and combining it with the switch information of the distribution automation system. The fault area is a passive area due to the isolation effect of the switch elements and protection devices. The faulty element is also included in this passive power-off area. Therefore, the location and identification of the faulty element is mainly carried out in the passive area. Step 2 Analysis of switch quantity information in the fault area of ​​the distribution network: Aiming at the switch quantity information of the distribution network, the diagnostic model based on optimization technology is used to creatively transform the power grid fault diagnosis problem into a 0-1 integer programming problem of finding extreme values, thus having a rigorous mathematical foundation. Then, the quantum genetic algorithm is used to find the optimal solution of the model, effectively realizing the distribution network fault diagnosis. The step 2 comprises the following steps: Step 2.1: Construct a fault diagnosis model based on the actual and expected values ​​of component actions. The specific model is: Where: r k,m and r * k,m are the actual state and expected state of the main protection of a certain device; r k,s and r * k,s are the actual state and expected state of the backup protection of a certain device respectively; k,l and r * k,l are the actual state and expected state of the remote backup protection of a certain device respectively; C i and C i * are the actual state and expected state of the circuit breaker respectively; Step 2.2: Use quantum genetic algorithm to solve the fault diagnosis model and obtain the fault component set; The step 2.2 comprises the following steps: Step 2.2.1: Initialize the fault region component population Q(t0) and randomly generate n chromosomes encoded with quantum bits; Step 2.2.2: Perform a decoding measurement on each individual in the initialized population Q(t0) to obtain a set of corresponding definite solutions in is the jth solution in the t0th generation population, expressed as a binary string of length m, where each bit is 0 or 1, according to the probability of the quantum bit ( or i=1,2,···,m) is selected; the measurement process is to randomly generate a number in the interval [0,1]. If it is greater than the square of the probability amplitude, the measurement result is 1, otherwise it is 0; the fitness of the determined solution is evaluated, and the optimal individual and the corresponding fitness are recorded; the fitness function adopts the objective function, that is, Θ(x)=E(x); Step 2.2.3: Use the quantum rotating door U(t) to adjust the individuals and obtain a new population Q(t); this adjustment strategy is to adjust the individuals The fitness of the current measured value Θ(x) is equal to the fitness value of the current best individual in the population Θ(best i ) is compared, if Θ(x)>Θ(best i ), then adjust The corresponding quantum bits in the i ,β i ) towards the direction that is favorable to x i Evolution in the direction of appearance, on the contrary, adjustment The corresponding quantum bits in the i ,β i ) Towards the best i The direction of emergence evolves; Step 2.2.4: Increase the number of iterations t by 1 until the set number of iterations is reached; obtain the optimal solution, thereby determining the fault component set; Step 3: Analysis of electrical quantity information in the fault area of ​​the distribution network: When a distribution network fails, the electrical quantities of the components in the system will have transient components; therefore, the electrical quantity fault degree of the components is obtained by analyzing the electrical quantities of the set of suspected fault components obtained in step 1; The step 3 comprises the following steps: Step 3.1: Perform a fast eigenmode decomposition of the current information through the component The step 3.1 comprises the following steps: Step 3.1.1: Let the input current signal be f(t), and calculate the extreme point R(t) of the signal: R(t),t0≤t j ≤t m-1 (2) In the formula, t0, t j and t m-1 Respectively represent the sampling moments of different extreme points; Step 3.1.2: Use the linear transformation method to transform the extreme points and use the residual r(t) in pairs: Where, t j+1 is the sampling time corresponding to the j+1 extreme value points; Step 3.1.3: Subtract the residual from the original signal to obtain the intrinsic modal function IMF of the original signal; therefore, the original input signal can be decomposed into multiple unique intrinsic modal components and a residual component; Step 3.2: Hilbert transform the modal function IMF Assume that a certain modal function IMF real signal is u(t), and its corresponding Hilbert transform is: The corresponding inverse transform is: Combining u(t) and v(t) gives the following complex signal: x(t)=u(t)+jv(t)=a(t)e jθt (6) According to formula (6), we can get: Then its instantaneous frequency can be defined as: Therefore, the instantaneous frequency of the original signal u(t) is the phase derivative of the corresponding complex signal x(t); Step 3.3: Calculate the amplitude distortion of the component current information Through formula (7), a certain modal function can be obtained, and after Hilbert transformation, the jth FIMD component of the current effective value amplitude of the i-th line before and after the fault occurs can be obtained as follows: In the formula, N is the number of sampling points, l is the number of sampling points corresponding to the fault moment; Then the distortion degree of the fault line current signal can be further obtained as: Step 3.4: Calculate the component current information fault energy metric; According to formula (8), the energy spectrum H(ω,t) of the fault current signal can be obtained, and then the corresponding marginal spectrum can be obtained as follows: The marginal energy spectrum of Hilbert transform is further obtained as: Then the current fault energy measurement value of the i-th line at fault can be expressed as: In the formula, f s The signal sampling frequency represented by Step 4. Distribution network fault information fusion decision: Through steps 2 and 3, three kinds of information, namely, component fuzzy fault degree, current information amplitude distortion degree, and current information fault energy degree, are obtained as the fault characteristics of each component in the fault area. Spectral clustering is used to fuse the fault information to achieve the final decision of fault diagnosis. The step 4 comprises the following specific steps: Step 4.1: Input sample set V = (x1, x2, x3…, x n ), construct the degree matrix D n×n and the adjacency matrix W; w ij For point x i and point x j The weight between Since it is an undirected graph, w ij =w ji ;d i For x i The sum of the weights of all connected edges; The matrix D is a diagonal matrix, and only the main diagonal has values, corresponding to the degree d of the i-th point in the i-th row i ; W is an n×n matrix, and the jth value in the i-th row corresponds to the weight w ij ; Step 4.2: Use Ncut theory to perform graph clustering and identify faulty components; The step 4.2 comprises the following specific steps: Step 4.2.1: Construct the graph-cutting clustering target E: in A i The complement of |A i | is the number of sample points in the subgraph, A i is the set of points for each subgraph; Step 4.2.2: Transform and solve the objective function E: Step 4.2.3: Construct the normalized Laplacian matrix D -1 / 2 LD -1 / 2 ; Step 4.2.4: Calculate D -1 / 2 LD -1 / 2 The eigenvectors f corresponding to the smallest k1 (k1 << n) eigenvalues respectively Step 4.2.5: Normalize the matrix composed of the corresponding eigenvectors f by row, and finally form an n×k1-dimensional eigenmatrix F; Step 4.2.6: Take each row in F as a k1-dimensional sample, a total of n samples, and cluster them using the traditional k-means clustering method with a clustering dimension of k2 to obtain the cluster partition C (c1, c2, ... c k2 ), the abnormal point is the faulty component.

Images