Distribution network fault location method based on multi-information fusion of particle filter algorithm

Abstract

A distribution network fault location method based on multi-information fusion using a particle filter algorithm includes the following steps: extracting transient information and using discrete wavelet transform to extract the transient zero-sequence current high-frequency component; calculating the similarity entropy and skewness coefficient of the transient zero-sequence current high-frequency component at each measurement point to obtain the corresponding preliminary fault section location; fusing the two preliminary fault section locations using a particle filter algorithm to obtain the fault section location; determining the steady-state zero-sequence voltage and power at the beginning and end of the fault section; segmenting the fault section and using a particle filter algorithm to obtain the zero-sequence voltage and power at the next segmentation point; iterating the segmentation until the threshold conditions are met, and outputting the fault location. This method uses the transient zero-sequence current high-frequency component and the steady-state zero-sequence voltage and power as fault signatures, achieving fault location in active distribution networks under different fault locations, fault closing angles, and white noise interference.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault location, and in particular to a distribution network fault location method based on multi-information fusion of a particle filtering algorithm. Background Art

[0002] Distributed power sources (DGs), with their flexible power supply and green, low-carbon nature, have become a core component of the next-generation distribution network. Traditional single-ended distribution networks are evolving into active distribution networks with the continuous integration of DGs. Due to the fluctuating and random nature of DG output, the fault equivalent circuits and parameters of active distribution networks are complex and highly variable. This not only weakens fault signatures, but also makes it difficult to accurately locate the faulty section using a single fault signature.

[0003] Currently, there are three main fault location methods: impedance, traveling wave, and signal injection. Impedance methods can be divided into single-ended and double-ended methods. The disadvantage of the impedance distance measurement method is that the impedance equation no longer holds when a branch appears on the fault line, making it unsuitable for complex lines. The advantage of the traveling wave method is its high ranging accuracy, but measuring fault traveling waves requires specialized recording equipment. The signal injection method allows for the injection of different signals based on system characteristics, offering greater flexibility. However, it requires supporting signal generation and collection equipment, leading to higher costs. Summary of the Invention

[0004] In response to the problem that inaccurate fault location is prone to occur when using a single fault characteristic quantity for fault location, the present invention provides a distribution network fault location method based on multi-information fusion of a particle filtering algorithm, aiming to solve the problem that inaccurate fault location is prone to occur when a single fault characteristic quantity is used. The transient zero-sequence current high-frequency component and the steady-state zero-sequence voltage power are used as fault characteristic quantities to realize the fault location function under different fault locations, fault closing angles, and white noise interference conditions in the active distribution network.

[0005] The technical solution adopted by the present invention is:

[0006] The distribution network fault location method based on multi-information fusion of particle filter algorithm includes the following steps:

[0007] Step 1: Extract transient information and use discrete wavelet transform to extract the high-frequency component of transient zero-sequence current;

[0008] Step 2: Calculate the similarity entropy and skewness coefficient of the high-frequency component of the transient zero-sequence current at each measurement point to obtain the corresponding preliminary fault section location;

[0009] Step 3: Use the particle filter algorithm to fuse the two preliminary fault section locations to obtain the fault section location;

[0010] Step 4: Determine the steady-state zero-sequence voltage and power at the first and last faults of the fault section;

[0011] Step 5: Cut the fault section and use the particle filter algorithm to obtain the zero-sequence voltage and power of the next cutting point. Iterate the cutting until the threshold condition is met and output the fault location.

[0012] In step 1, the problem that the short-circuit fault must have a continuous steady-state process when using steady-state fault information for fault location is difficult to apply to the neutral point grounding system through the arc suppression coil is solved. The transient information is introduced by discrete wavelet transform (DWT) to improve the positioning accuracy. The wavelet basis function Perform translation and scale transformation to obtain the mother wavelet function coefficient W φ (a,τ), as shown in formula (1):

[0013]

[0014] Where a>0 is the scale factor, which reflects the change of the wavelet basis function itself; τ is the parameter reflecting the translation, which indicates the position of the wavelet basis function in the time domain of the signal; t is the starting time of the signal data window to be analyzed after the short circuit fault; φ a,τ (t) is the mother wavelet function.

[0015] Then, the mother wavelet function is inner-producted with the signal to be analyzed x(t) to obtain the wavelet coefficient W of the signal to be analyzed. a,τ , as shown in formula (2):

[0016] Wa, τ = <x(t),φa, τ (t)>=W(a,τ) (2);

[0017] in, <x(t),φ a,τ (t)> represents the inner product of the signal to be analyzed and the mother wavelet function; W(a,τ) represents that the wavelet coefficient of the signal to be analyzed is a function of a and τ.

[0018] At this time, the transient component X(t) of the signal to be analyzed is shown in formula (3):

[0019]

[0020] Substituting the intercepted zero-sequence current data window into equations (1), (2), and (3), the extracted transient zero-sequence current high-frequency component can be obtained as shown in equation (4):

[0021]

[0022] The step 2 comprises the following steps:

[0023] S2.1: Based on the transient zero-sequence current high-frequency component extracted in step 1, the random variable X={x1,x2,…,x n}、Y={y1,y2,…,y n} is the transient zero-sequence current state of any two measurement points on the fault line; p(X=i)(i=1,2,…,n) and p(Y=i)(i=1,2,…,n) are the corresponding probabilities of each measurement point. The relative entropy between any two measurement points on the fault line is calculated as shown in Equation (5):

[0024]

[0025] Among them, D(X||Y) represents the relative entropy value of state X relative to state Y; D(Y||X) represents the relative entropy value of state Y relative to state X; p(x i ) represents the probability of fault line measurement point X=i; p(y i ) represents the probability of the faulty line measurement point Y=i.

[0026] Finally, the relative entropy matrix is ​​formed as shown in formula (6).

[0027]

[0028] Where D represents the relative entropy matrix of any two measurement points on the fault line; that is, the matrix D a 、D d The elements in the matrix D are compared b 、D c The middle element is obviously smaller. Based on this, we can preliminarily determine the location of the fault section (m,n);

[0029] D a Represents the matrix in which the element values ​​in matrix D are obviously smaller and located in the upper left corner; D b Indicates the matrix with significantly larger element values ​​in the matrix D and located in the upper right corner; D c Indicates the matrix with significantly larger element values ​​in D and located in the lower left corner; D d Represents the matrix in which the element values ​​in matrix D are significantly smaller and located in the lower right corner.

[0030] D(1||1) represents the relative entropy of measurement point 1 relative to measurement point 1; D(m||1) represents the relative entropy of measurement point m relative to measurement point 1; D(1m) represents the relative entropy of measurement point 1 relative to measurement point m; D(m||m) represents the relative entropy of measurement point m relative to measurement point m; D(n||1) represents the relative entropy of measurement point n relative to measurement point 1; D(k||1) represents the relative entropy of measurement point k relative to measurement point 1; D(n||m) represents the relative entropy of measurement point n relative to measurement point m; D(k||m) represents the relative entropy of measurement point k relative to measurement point m ; D(1n) represents the relative entropy of measurement point 1 relative to measurement point n; D(m||n) represents the relative entropy of measurement point m relative to measurement point n; D(1k) represents the relative entropy of measurement point 1 relative to measurement point k; D(m||k) represents the relative entropy of measurement point m relative to measurement point k; D(n||n) represents the relative entropy of measurement point n relative to measurement point n; D(k||n) represents the relative entropy of measurement point k relative to measurement point n; D(n||k) represents the relative entropy of measurement point n relative to measurement point k; D(k||k) represents the relative entropy of measurement point k relative to measurement point k. S2.2: Calculate the skewness coefficient of the transient component, as shown in formula (7):

[0031]

[0032] Where S represents the skewness coefficient; x i The i-th value of the sequence to be analyzed is represented by; represents the average value of the sequence to be analyzed; n represents the number of values ​​in the sequence to be analyzed.

[0033] The magnitude of the skewness coefficient reflects the degree of asymmetry in the data distribution. When the skewness coefficient is greater than zero, the data distribution is skewed to the right, meaning that there are more items with larger values. When the skewness coefficient is less than zero, the data distribution is skewed to the left, meaning that there are more items with smaller values. The larger the absolute value of the skewness coefficient, the more pronounced the data skewness. In practical applications, data distribution characteristics can be determined by observing data distribution graphs or performing quantitative analysis using the skewness coefficient.

[0034] S2.3: Obtain two preliminary fault section locations through the two transient fault information, namely matrix D a 、D d The elements in the matrix D are compared b 、D c The elements in the middle are obviously smaller. Based on this, the fault section location (m,n) can be preliminarily determined; m represents the fault line measurement point m; n represents the fault line measurement point n;

[0035] From formula (7), we can get the skewness coefficient of each measurement point of the fault line as S = {S1, S2, ..., S m ,Sn ,…,S q}, S1, S2, ..., Sm represent the skewness coefficients of the fault line measurement points 1, 2, ..., m respectively; S n ,…,S q They represent the skewness coefficients of the fault line measurement points n, ..., q respectively.

[0036] If the skewness coefficients from measurement point 1 to measurement point m are all less than zero, and the skewness coefficients from measurement point n to measurement point q are all greater than zero, the fault section can be preliminarily determined to be (m, n).

[0037] In step 3, in order to improve the accuracy of fault section location, the particle filter algorithm is introduced to correct the preliminary fault section location result obtained by the relative entropy coefficient after preprocessing by the skewness coefficient location result to obtain the final fault section location. The particle filter algorithm process includes the following steps:

[0038] Step 3.1: Assume the system state model equation and observation model equation as shown in equations (8) and (9):

[0039] x t =f(x t-1 )+Q t (8);

[0040] y t =h(x t )+R t (9);

[0041] Among them, x t 、y t represents a random variable, f(·) and h(·) are functions of the variable x; t is the number of iterations; f(x t-1 ) means f(x t ) Function of variable x before iteration; Q t R represents the predicted noise; t represents the noise of the observation.

[0042] Step 3.2: According to the system model of formula (8), the state of each particle is predicted to obtain the particle sample at the next moment. The prediction equation is as follows:

[0043]

[0044] Where n represents the total number of particles; represents the i-th particle; represents the prediction weight of the i-th particle; δ(·) represents the Dirac function; f t - (x) represents the particle sample at the next moment; x represents the variable to be predicted; The unit impulse representing the difference between the current particle sample and the i-th particle.

[0045] Step 3.3: Based on the observation model and observation data, calculate the weight of each particle and perform normalization. The weight represents the degree of consistency between each particle and the observation data. The update equation is shown in Equation (11):

[0046]

[0047] in, represents the weight of the i-th particle; f t + (x) represents the posterior probability density of variable X; η represents the inverse of the sum of the weights of the i-th particle; represents the prediction weight of the i-th particle; n represents the total number of particles; f R [y t -h(x)] represents the observation noise probability density.

[0048] Bring the above particle filter algorithm process into the distribution network measurement data;

[0049] First, initialize the distribution network measurement points by pre-processing the skewness coefficient positioning results and perform particle initialization to obtain represents the i-th particle at the t-th step, i∈(1,k), k is the total number of measurement points in the distribution network, represents the i-th particle at the measurement point m at the t-th step; represents the i-th particle at the measurement point n at the t-th step.

[0050] Assign initial weight values ​​to particles Then predict and generate samples at time t represents the particle sample before particle initialization; Q represents the noise during prediction.

[0051] And calculate the corresponding weight value of the particle at time t: as shown in formula (12)

[0052]

[0053] Among them, f t - (P) represents the weight value of the particle at time t; represents the weight value before particle initialization; P represents the sample at time t after particle initialization; P t -(i) represents the sample at time t generated after prediction; δ(PP t -(i) ) represents the pulse difference between the sample particle after initialization and the predicted one; k represents the total number of measurement points in the distribution network.

[0054] Then, the result obtained by formula (12) is used as the observation quantity to correct the relative entropy coefficient fault location result after distribution network preprocessing according to formula (13).

[0055]

[0056] Among them, f t + (P) represents the sample corrected after particle prediction; f W (P t -P t - ) represents the difference between the particle weight value after the relative entropy coefficient positioning result and the particle weight value before the prediction at time t; P t Represents the relative entropy coefficient positioning result particle initialization sample; P t - Represents the particle weight value before the relative entropy coefficient positioning result prediction; η represents the inverse of the sum of the weights of the i-th particle.

[0057] Final resampling: Assume that the particle at step t is P t =[P t 1 ,P t 2 ,…,P t i ,…,P t k ], P t 1 ,P t 2 ,…,P t i ,…,P t k They represent the particle samples at measurement points 1, 2, ..., k respectively.

[0058] P t i The next step of particles generated after iteration is When detected The weight value When it is less than the set threshold ζ, a random number j from 1 to k is generated, and P t i =P t i , and iterate again This process repeats until In step 4, since the transformer connected to the DG has a delta connection, the impact of the DG on the zero-sequence component can be ignored. The zero-sequence current and voltage distribution of the infinitesimal interval in the fault section of the active distribution network is as follows: Figure 2 As shown. Among them, U 0(K) 、U 0(K+1)is the zero-sequence voltage at the beginning and end of the Kth infinite interval in the upstream section of the fault; U 0(I) 、U 0(I+1) is the zero-sequence voltage at the beginning and end of the Ith infinite interval in the downstream section of the fault; I 0(K) , I 0(I) is the zero-sequence current flowing through the wireless cell K and I; L0, R0, C0, G0 are the zero-sequence inductance, resistance, capacitance, and conductance per unit length of the line; the voltage U on the left and right sides of the fault point 0(K+1) 、U 0(I) Satisfying formula (14):

[0059] U 0(K+1) =U 0(K) +I 0(K) Z0

[0060] U 0(I) =U 0(I+1) +I 0(I) Z0 (14);

[0061] Z0=R0+jωL0

[0062] Where Z0 represents the line zero-sequence impedance; ω represents the angular frequency.

[0063] From formula (14), the zero-sequence voltage amplitude distribution of the fault section can be obtained as follows: Figure 3 As shown in Figure 2, the zero-sequence voltage amplitude at the fault point is the largest. Based on the differential idea, the fault section (m, n) is divided into infinite sub-intervals, U 0(st) 、U 0(ed) is the zero-sequence voltage at the beginning and end of the sub-interval.

[0064] At this time, the calculation of the fault point position is converted into seeking the sub-interval with the highest zero-sequence voltage amplitude. When the left and right sides of the sub-interval are infinitely close, the interval can be regarded as the point where the fault occurs.

[0065] In step 5, the final fault segment is cut based on the iterative algorithm and particle filtering principle:

[0066] Define the flow from the head end to the end as forward, and the flow from the end end to the head end as reverse. Assuming that the fault section (m, n) is known, it is cut as follows Figure 4 As shown. Among them: S m is the zero-sequence power at the head end of the fault section, V m is the zero-sequence voltage at the head end of the fault section, S n is the zero-sequence power at the end of the fault section, V nis the zero-sequence voltage at the end of the fault section; △V1 is the positive zero-sequence transverse voltage drop of the mk section, δV1 is the positive zero-sequence longitudinal voltage drop of the mk section, △S1 is the positive zero-sequence power loss of the mk section; △V2 is the positive zero-sequence transverse voltage drop of the In section, δV2 is the positive zero-sequence longitudinal voltage drop of the In section, △S2 is the positive zero-sequence power loss of the In section; △V2′ is the negative zero-sequence transverse voltage drop of the kI section, δV2′ is the negative zero-sequence longitudinal voltage drop of the kI section, △S2′ is the negative zero-sequence power loss of the kI section; △V3 is the negative zero-sequence transverse voltage drop of the In section, δV is the negative zero-sequence longitudinal voltage drop of the In section, △S3 is the negative zero-sequence power loss of the In section; R+jX is the zero-sequence impedance of the line corresponding to the cut section.

[0067] Extract the zero-sequence voltage and power at both ends of the fault section. At this time, forward calculation is performed on the finite number of cut points k and I in the fault section (m, n). k 、V I , the first and last zero-sequence voltages and powers of the next cutting point are obtained through the forward iterative method as shown in equations (15), (16), and (17):

[0068]

[0069] Among them, P m Indicates the active power at the measuring point m; Q m Represents the reactive power at the measuring point m; V m Represents the voltage at the measuring point m; P k Indicates the active power at the measuring point k; Q k Represents the reactive power at the measuring point k; V k represents the voltage at the measuring point k; R represents the resistance between the measuring points m and k; X represents the reactance between the measuring points m and k.

[0070]

[0071] Reverse calculation is performed to calculate the zero-sequence voltage of a finite number of cut points k and I in the fault section (m,n), as shown in equations (18), (19), and (20):

[0072]

[0073] Among them, P I ′ represents the active power at point I during reverse calculation; Q I ′ represents the reactive power at point I during reverse calculation; V I ′ shows the voltage at point I during reverse calculation.

[0074]

[0075] Wherein, δV3 represents the longitudinal voltage drop of the In segment.

[0076]

[0077] Among them, V I ′ represents the voltage at point I during reverse calculation; V k ′ represents the voltage at point k during reverse calculation.

[0078] The forward and reverse results of each cutting of the fault section are fused using the particle filter algorithm to obtain the power and voltage information of the next cutting point; then the iterative cutting is continued until the precise fault location is obtained.

[0079] The present invention provides a distribution network fault location method based on multi-information fusion of particle filtering algorithm, and the technical effects are as follows:

[0080] 1) The present invention proposes a fault location method that combines transient information and steady-state information, which makes up for the fault location limitations brought about by a single amount of fault information and improves the accuracy of fault location.

[0081] 2) The present invention utilizes the particle filter algorithm to perform information fusion, thereby increasing the utilization rate of fault characteristic quantities with obvious fault characteristics, and facilitating the identification of system fault points.

[0082] 3) To address the problem that a single fault characteristic quantity is difficult to accurately reflect the fault location, this method proposes a particle filter algorithm multi-information fusion model based on multiple fault characteristic quantities. The transient zero-sequence current high-frequency component and steady-state zero-sequence voltage are used as fault characteristic quantities, and the particle filter algorithm model is used to perform multi-information fusion to obtain the fault location.

[0083] 4) The present invention proposes a new method for applying multi-information fusion in fault location based on the particle filter algorithm, using the relative entropy and skewness coefficient of the high-frequency component of the zero-sequence current as the fault characteristic quantity for locating the fault section; using the particle filter algorithm to fuse the two transient high-frequency component information to obtain the fault section position; using the zero-sequence voltage of the fault section as the fault characteristic quantity, using the forward iteration and particle filter algorithm fusion to obtain the steady-state zero-sequence voltage and power distribution, thereby realizing effective and accurate fault location of the active distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The present invention will be further described below with reference to the accompanying drawings and examples:

[0085] Figure 1 Fault location flowchart

[0086] Figure 2 Schematic diagram of zero-sequence current and voltage distribution between infinitesimal intervals in the fault section.

[0087] Figure 3 Schematic diagram of zero-sequence voltage amplitude distribution in the fault section.

[0088] Figure 4 Schematic diagram of the cutting point distribution in the fault section.

[0089] Figure 5 Schematic diagram of the active distribution network system model.

[0090] Figure 6 Schematic diagram of transient information when a fault occurs on the upstream side of the DG.

[0091] Figure 7 Schematic diagram of particle weight iteration for accurate fault location when a fault occurs on the upstream side of a DG.

[0092] Figure 8 Schematic diagram of transient information when a fault occurs between two DGs.

[0093] Figure 9 Schematic diagram of particle weight iteration fault precise location when a fault occurs between two DGs.

[0094] Figure 10 Schematic diagram of transient information when a fault occurs on the downstream side of the DG.

[0095] Figure 11 Schematic diagram of particle weight iteration for accurate fault location when a fault occurs on the downstream side of a DG.

[0096] Figure 12 Schematic diagram of transient information when the fault closing angle is 60°.

[0097] Figure 13 Schematic diagram of accurate fault location using particle weight iteration at a 60° fault closing angle.

[0098] Figure 14 Schematic diagram of transient information when a fault occurs on the upstream side of the DG with 25dB white noise.

[0099] Figure 15 Schematic diagram of particle weight iteration for accurate fault location when a fault occurs on the upstream side of a DG with 25dB white noise. DETAILED DESCRIPTION

[0100] A new method for locating distribution network faults based on multi-information fusion using a particle filter model. This method addresses the problem that a single fault characteristic cannot accurately reflect the fault location. Starting from multiple fault characteristics, a particle filter algorithm multi-information fusion model is proposed. Using the transient zero-sequence current high-frequency component and steady-state zero-sequence voltage as fault characteristics, the particle filter algorithm model is used to fuse the multi-information to determine the fault location.

[0101] This method extracts the high-frequency component of transient zero-sequence current after the fault through discrete wavelet transform; defines the similarity entropy and skewness coefficient of the high-frequency component of transient zero-sequence current to obtain the preliminary fault section; implements the fusion of similarity entropy and skewness coefficient information based on the multi-information fusion model of particle filter algorithm to obtain the location of the fault section; adopts the forward iteration method to cut the fault section, and uses particle filter to fuse the weights of zero-sequence voltage and power at the cutting point at different fault moments to obtain the zero-sequence voltage and power at the next cutting point, until the cutting stops when the threshold is met, and the fault point location is obtained. The flow chart of this method is as follows Figure 1 As shown, the principle is simple, the fault location speed is fast, and it is easy to implement, and it can be well applied to active distribution networks.

[0102] The following steps are involved:

[0103] Step 1. Transient information extraction:

[0104] According to formula (2), the wavelet coefficient of the zero-sequence current signal I0(t) can be obtained by taking the inner product with the mother wavelet function, as shown below:

[0105] W a,τ = <I0(t),φ a,τ (t)>=W(a,τ) (2);

[0106] Using the wavelet coefficients obtained by equation (2), the zero-sequence current signal I0(t) can be expressed as

[0107]

[0108] Since the Daubechies wavelet family (dbN) has better wavelet expansion performance and can control the boundary distortion caused by energy concentration by adjusting the support length, the present invention adopts the Daubechies wavelet as the wavelet basis function.

[0109] Step 2. Transient information preprocessing:

[0110] 1) Similarity entropy calculation: The transient zero-sequence current of each measuring point corresponding to the fault line is regarded as a measurement system, and the transient zero-sequence current state of any two measuring points of the fault line is expressed by random variables X = {x1, x2, ..., x n}、Y={y1,y2,…,y n}, the corresponding probabilities are p(X=i)(i=1,2,…,n) and p(Y=i)(i=1,2,…,n). The entropy of the transient zero-sequence current at the above two measurement points is defined as follows:

[0111]

[0112] In order to highlight the trend and size of the relative entropy value change, the absolute value is used for calculation. The relative entropy value of state X relative to state Y D(X||Y) and the relative entropy value of state Y relative to state X D(Y||X) are as follows:

[0113]

[0114] Assuming that the fault line has k measurement points and the fault section is (m, n), the fault line measurement system contains k different states. The relative entropy value of each measurement point state relative to the k states is calculated to form the following matrix.

[0115]

[0116] The transient zero-sequence current waveforms of the measurement points at the upstream and downstream of the fault are consistent, and the relative entropy values ​​are small; on the contrary, the relative entropy values ​​are large; that is, the matrix D a 、D d The elements in the matrix D are compared b 、D c The elements in the middle are obviously smaller. Based on this, we can preliminarily determine the location of the fault section (m,n).

[0117] 2) Calculation of skewness coefficient: The transient zero-sequence current state at any point on the fault line is expressed as a random variable X = {x1, x2, ..., x n} indicates that if there are q measurement points in the fault line, its skewness coefficient is calculated as follows.

[0118]

[0119] The peak of the frequency distribution of transient zero-mode current sampling values ​​upstream of the fault point is on the right, the long tail is on the left of the mode, and the data obeys a negative skew distribution, S < 0; the peak of the frequency distribution of transient zero-mode current sampling values ​​in the non-fault section is on the left, the long tail is on the right of the mode, and the data obeys a positive skew distribution, S > 0. From formula (30), the skew coefficient of each measurement point on the fault line is S = {S1, S2, ..., S m ,S n ,…,S q If the skewness coefficients from measurement point 1 to measurement point m are all less than zero, and the skewness coefficients from measurement point n to measurement point q are all greater than zero, the fault section can be preliminarily determined to be (m, n).

[0120] Step 3. Steady-state information analysis:

[0121] Extract the zero-sequence voltage and power at both ends of the fault section. At this time, forward calculation is performed on the finite number of cut points k and I in the fault section (m, n). k 、V I , as shown in formulas (15), (16), and (17):

[0122]

[0123] The zero-sequence voltages of a finite number of cut points k and I in the fault section (m, n) are calculated by reverse calculation, as shown in equations (18), (19), and (20):

[0124]

[0125] The forward and reverse results of each cutting of the fault section are fused using the particle filter algorithm to obtain the power and voltage information of the next cutting point; then the iterative cutting is continued until the precise fault location is obtained.

[0126] Step 4. Locate the fault section:

[0127] 1) Definition of particle filter algorithm parameters. This paper analyzes the fault location results based on the density of particles after particle filtering, and presents the fault location process in a 3D graph, where the X and Y axes are the fault measurement points n1, n2, ..., n s , Z is the fault location coefficient r, which is defined as follows

[0128]

[0129] Among them, l mn is the length of the fault section line, l f The distance between the fault point and the measuring point near the busbar outlet in the fault section is n m The line length, l m is the measurement point n m The line length from the busbar outlet.

[0130] The segment positioning result in the particle filter information fusion is displayed as an XY plane projection. When the particle projection falling into a certain grid p in the XY plane is the densest, the segment corresponding to the grid p is the fault segment. To measure the density of particle projection in the XY plane, the present invention defines the plane density N p , which is defined as follows

[0131]

[0132] Where N1 is the number of particle projections falling into the XY plane grid p, S p is the area of ​​the grid p in the XY plane.

[0133] Subsequent work will continue to incorporate other information and present the results in XZ plane projection. The particles whose projection falls on the XZ plane and have the highest linear density on the Z axis are considered as the failure points. The linear density is defined as follows

[0134]

[0135] Where, l is the preset unit length of the Z axis, l=l th , N2 is the number of particle projections that fall into the XZ plane after the first projection of the particle filter, and then the second projection is made on the Z axis and falls into the line segment l.

[0136] 2) Initialization: The skewness coefficient positioning results of each measurement point in the distribution network after skewness coefficient preprocessing are initialized by particles to obtain Represents the i-th particle at step t, i∈(1,k), k is the total number of measurement points in the distribution network. Assign the particle an initial weight value

[0137] 3) Prediction, generating samples at time t And calculate the corresponding weight value of the particle at time t: as shown in formula (12)

[0138]

[0139] 4) Update: Use the result obtained by formula (12) as the observation quantity to correct the relative entropy coefficient fault location result after distribution network preprocessing according to formula (13).

[0140]

[0141] (5) Resampling: Assume that the particle at step t is P t =[P t 1 ,P t 2 ,…,P t i ,…,P t k ], P t i The next step of particles generated after iteration is When detected The weight value When it is less than the set threshold ζ, a random number j from 1 to k is generated, and P t i =P t i , and iterate again This process repeats until

[0142] (6) Calculate the plane density, stop resampling, and calculate the particle projection plane density N on the XY plane p , the plane density N p The largest segment is regarded as the fault segment and the result is output.

[0143] Step 5. Accurately locate the fault:

[0144] (1) Obtain the fault sections (m, n) of k groups of fault times four power frequency cycles after the fault time i The zero-sequence power and voltage at the beginning and end, namely S m(t) ={S m(1) ,S m(2) ,…,S m(k)}、V m(t) 、S n(t) 、V n(t) , at this time the number of iterations i = 0;

[0145] (2) Assume that the number of subintervals obtained in each iteration is N, the total number of iterations is K, and the length of the fault section is l mn , then at the i-th iteration

[0146] (3) S m(t) 、V m(t) Substitute into equations (15), (16), and (17) to calculate the power and voltage S of the next cutting point m+1 at different times for the m point k group. m+1(t) 、V m+1(t) ; S n(t) 、V n(t) Substitute into equations (18), (19), and (20) to calculate the power and voltage S′ at the cutting point n-1 on point n. n-1(t) , V′ n-1(t) ; Use the particle filter algorithm to filter out the zero-sequence power and voltage at the fault time with larger particle weights in the forward and reverse cutting processes to obtain a new fault interval (m, n) i+1 .

[0147] (4) The new fault interval (m,n) i+1 Continue cutting iteration and particle filter algorithm screening until l mn ≤l th , l th Set a threshold for the iteration interval. At this time, the fault interval can be regarded as a fault point and output, ending the iteration.

[0148] Simulation verification:

[0149] In order to verify the fault location method proposed in this invention (hereinafter referred to as location method), the PSCAD / EMTDC simulation platform was used to build the following Figure 5 The active distribution network model shown. The left side shows the system power supply; each measurement point is separated by a 5km overhead line, for a total of 19 measurement points. Fault location verification was performed under different fault locations, different fault closing angles, and noise-containing fault scenarios.

[0150] 1) Simulation verification of fault location on the upstream side of the DG: The fault is set upstream of the two DGs, between measurement points 4 and 5, and 3906m away from measurement point 4. For example, a single-phase grounding fault on phase A, with a fault closing angle of 0° and white noise of 0dB, is used. The fault time is set to 0.1s and lasts for 0.2s. The zero-sequence current of each monitoring point after the fault occurs is read, and its high-frequency zero-sequence current relative entropy and skewness coefficient are calculated. The particle filter algorithm is used to fuse transient information to obtain the fault section location. The zero-sequence voltage and power at the beginning and end of the fault section are extracted and iteratively cut to obtain the fault location. The simulation results are shown as follows: Figure 6 、 Figure 7 shown.

[0151] Depend on Figure 6 It can be seen that the different colored dotted lines in the zero-sequence current high-frequency component represent the fault upstream measurement points 1-4, and the different colored solid lines represent the fault downstream measurement points 5-12; the zero-sequence current high-frequency component waveform on the upstream side of the fault point is consistent and the relative entropy value is small, while the zero-sequence current high-frequency component waveform on the downstream side of the fault point is consistent and the relative entropy value is large; it can be preliminarily determined that the fault section is (4,5). In the zero-sequence current high-frequency component frequency distribution diagram, the black, red, blue, and green columns represent the fault upstream measurement points 1-4, and the remaining color columns represent the fault downstream measurement points 5-12. Figure 10 It can be seen that the frequency distribution peak of measurement point 1 close to the busbar is on the right side, and the long tail is located to the left of the mode, which conforms to the negative skewed distribution, and the skewness coefficient is less than zero. The same is true for measurement points 2, 3, and 4. The frequency distribution peak of measurement points 5-12 in the non-fault section on the downstream side of the fault is on the left side, and the long tail is located to the right of the mode, which conforms to the positive skewed distribution, and the skewness coefficient is greater than zero. The skewness coefficients between measurement points 4 and 5 have different signs, and it can be preliminarily concluded that the fault section is (4,5).

[0152] Depend on Figure 7 It can be seen that using the fault transient information according to Figure 6 The particle filter algorithm information fusion process shown in the figure fuses the above preliminary segment positioning results and obtains the fault segment as (4,5). It iterates the cutting point and obtains the following: Figure 7 The particle weighted iteration method accurately locates the upstream fault. The X and Y axes represent the measurement points in the section, and the Z axis represents the accurate positioning coefficient. First, the fault section location results are projected onto the XY plane, as indicated by the green triangles. The density of each grid plane is calculated, and the maximum plane density corresponds to the fault section (4, 5). The accurate positioning results are further projected onto the XZ plane, and the linear density on the Z axis is calculated. The accurate positioning coefficient r = 0.7745 corresponding to the maximum linear density is obtained. The calculated fault point is located in section (4, 5), 3872.3 meters away from measurement point 4, with an absolute positioning error of 33.7 meters.

[0153] 2) Fault location simulation verification between two DGs: The fault is set between the two DGs, between measurement points 7 and 8, and 2818.5m away from measurement point 7; take phase A single-phase grounding fault, the fault closing angle is 0°, and the white noise is 0dB as an example. The fault time is set to 0.1s and lasts for 0.2s. The simulation results are shown below. Figure 8 、 Figure 9 shown.

[0154] The transient component of fault information is as follows: Figure 8 As shown. Figure 8 It can be seen that the waveforms of the high-frequency components of the zero-sequence currents at measuring points 1-7 are consistent, and the relative entropy values ​​are small. The waveforms of the high-frequency components of the zero-sequence currents at measuring points 8-12 are consistent, and the relative entropy values ​​are large. The fault section can be preliminarily obtained as (7, 8). The frequency distribution peak of measuring points 1-7 is on the right, and the long tail is located to the left of the mode, which conforms to the negatively skewed distribution, and the skewness coefficient is less than zero. The frequency distribution peak of measuring points 8-12 is on the left, and the long tail is located to the right of the mode, which conforms to the positively skewed distribution, and the skewness coefficient is greater than zero. The skewness coefficients between measuring points 7 and 8 are of different signs, and the fault section can be preliminarily obtained as (7, 8).

[0155] The above preliminary segment positioning results are fused to obtain the fault segment (7,8); the cutting point is iterated in the fault segment (7,8) to obtain the following: Figure 9 When a fault occurs between the two DGs shown in the figure, particle weight iteration is used to precisely locate the fault. The plane density of each grid is calculated, and the fault segment corresponding to the maximum plane density is (7, 8). The precise positioning results are further projected onto the XZ plane, and the linear density on the Z axis is calculated. The precise positioning coefficient r = 0.5694 for the fault point corresponding to the maximum linear density is obtained. The calculated fault point is located in segment (7, 8), 2847.06 meters away from measurement point 7, with an absolute positioning error of 28.56 meters.

[0156] 3) Simulation verification of fault location on the downstream side of the DG: The fault is located on the downstream side of the two DGs, between measurement points 10 and 11, and 1598m away from measurement point 10. A single-phase grounding fault on phase A, a fault closing angle of 60°, and white noise of 0dB are used as an example. The fault duration is set to 0.1s and lasts for 0.2s. The simulation results are shown below. Figure 10 、 Figure 11 shown.

[0157] Depend on Figure 10It can be seen that the waveforms of the high-frequency components of the zero-sequence current at measuring points 1-10 are consistent, and the relative entropy values ​​are small. The waveforms of the high-frequency components of the zero-sequence current at measuring points 11 and 12 are consistent, and the relative entropy values ​​are large. The fault section can be preliminarily obtained as (10, 11). The frequency distribution peak of measuring points 1-10 is on the right, and the long tail is on the left side of the mode, which conforms to the negatively skewed distribution, and the skewness coefficient is less than zero. The frequency distribution peak of measuring points 11 and 12 is on the left, and the long tail is on the right side of the mode, which conforms to the positively skewed distribution, and the skewness coefficient is greater than zero. The skewness coefficients between measuring points 10 and 11 are of different signs, and the fault section can be preliminarily obtained as (10, 11).

[0158] The above preliminary segment positioning results are fused to obtain the fault segment (10,11); the cutting point is iterated in the fault segment (10,11), and the following is obtained: Figure 11 The particle weighted iteration method accurately locates the fault downstream of the DG shown in the figure. The plane density of each grid is calculated, and the fault segment corresponding to the maximum plane density is (10, 11). The accurate positioning results are further projected onto the XZ plane and the linear density on the Z axis is calculated. The accurate positioning coefficient r = 0.3269 corresponding to the maximum linear density is obtained. The calculated fault point is located in the segment (10, 11), 1634.27 meters away from measurement point 7, and the absolute positioning error is 36.27 meters.

[0159] 4) Simulation Verification of a 60° Fault Closing Angle: The fault was set upstream of the two DGs, between measurement points 4 and 5, and 3906m away from measurement point 4. For example, a single-phase grounding fault on phase A was used, with a 60° fault closing angle and 0dB white noise. A symmetrical three-phase ABC fault was set between measurement points 10 and 11. The fault occurred at 0.02s. The simulation results are shown below. Figure 12 、 Figure 13 shown. Figure 12 Analysis and Figure 6 The same, no further elaboration will be given.

[0160] The particle weight iteration fault accurate positioning at 60° closing angle is as follows Figure 13 The fault point precise positioning coefficient r = 0.7904, the calculated fault point is located in section (4,5), 3951.8m away from measurement point 4, and the absolute positioning error is 45.8m.

[0161] 3) Adaptability simulation verification under white noise: The fault is set upstream of the DG, between measurement points 4 and 5, and 3906m away from measurement point 4. Phase A single-phase grounding is used as an example. The fault closing angle is 0°, and 25dB Gaussian white noise is added. The simulation results are shown below. Figure 14 、 Figure 15 shown.

[0162] Depend on Figure 14It can be seen that under the influence of 25dB Gaussian white noise, the transient component of the fault information has a power frequency periodic waveform distortion at the beginning of the intercepted data window. However, based on the waveform of the high-frequency component of the zero-sequence current and its frequency distribution at this time, it can still be preliminarily obtained that the fault section is (4,5).

[0163] Particle weighted iterative fault accurate positioning under the influence of 25dB white noise Figure 15 The precise positioning coefficient of the fault point corresponding to the maximum linear density is r = 0.7907. The calculated fault point is located in section (4, 5), 3953.7m away from measurement point 4, and the absolute positioning error is 47.7m.

Claims

1. A distribution network fault location method based on multi-information fusion of particle filter algorithm, characterized by The following steps are involved: Step 1: Extract transient information and use discrete wavelet transform to extract the high-frequency component of transient zero-sequence current; Step 2: Calculate the relative entropy and skewness coefficient of the high-frequency component of the transient zero-sequence current at each measurement point to obtain the corresponding preliminary fault section location; Step 3: Use the particle filter algorithm to fuse the two preliminary fault section locations to obtain the fault section location; Step 4: Determine the steady-state zero-sequence voltage and power at the first and last faults of the fault section; Step 5: Use the forward iteration method to cut the fault section, use the particle filter to fuse the weights of the zero-sequence voltage and power at the cutting points at different fault times, and obtain the zero-sequence voltage and power at the next cutting point. Iterate the cutting until the threshold condition is met and output the fault location.

2. The distribution network fault location method based on multi-information fusion of particle filter algorithm according to claim 1 is characterized by: In step 1, the transient information is introduced by using discrete wavelet transform (DWT) to transform the wavelet basis function Perform translation and scaling to obtain the mother wavelet function coefficients , as shown in formula (1): (1)； in, , is the scale factor, reflecting the change of the wavelet basis function itself; To reflect the parameters of translation, it represents the position of wavelet basis function in the time domain of the signal; The starting time of the signal data window to be analyzed after the short circuit fault; is the mother wavelet function; Then combine the mother wavelet function with the signal to be analyzed Perform inner product to obtain the wavelet coefficients of the signal to be analyzed , as shown in formula (2): (2) ； in, Represents the inner product of the signal to be analyzed and the mother wavelet function; The wavelet coefficients of the signal to be analyzed are about function; At this time, the transient component of the signal to be analyzed As shown in formula (3): (3) ； Substituting the intercepted zero-sequence current data window into equations (1), (2), and (3), the extracted transient zero-sequence current high-frequency component can be obtained as shown in equation (4): (4)。 3. The distribution network fault location method based on multi-information fusion of particle filter algorithm according to claim 1 is characterized by: The step 2 comprises the following steps: S2.1: Based on the transient zero-sequence current high-frequency component extracted in step 1, the random variable 、 is the transient zero-sequence current state of any two measuring points on the fault line; calculate the relative entropy between any two measuring points on the fault line, as shown in formula (5): (5) ； in, Indicates status Relative to the state relative entropy; Indicates status Relative to the state relative entropy; Indicates the fault line measurement point probability; Indicates the fault line measurement point The probability of ; finally the relative entropy matrix is ​​formed as shown in formula (6): (6) ； in, Represents the relative entropy matrix of any two measurement points on the fault line; that is, the matrix 、 The elements in the matrix are compared 、 The middle element is obviously smaller; based on this, the fault section location (m, n) can be preliminarily determined; Representation matrix The matrix with significantly smaller elements and located in the upper left corner; Representation matrix The matrix with obviously larger element values ​​and located in the upper right corner; express The matrix with obviously larger element values ​​and located in the lower left corner; Representation matrix The matrix with obviously smaller elements and located in the lower right corner; Indicates the relative entropy of measurement point 1 relative to measurement point 1; Represents the relative entropy of measurement point m relative to measurement point 1; Represents the relative entropy of measurement point 1 relative to measurement point m; It represents the relative entropy of the measurement point m relative to the measurement point m; Represents the relative entropy of measurement point n relative to measurement point 1; Represents the relative entropy of measurement point k relative to measurement point 1; Represents the relative entropy of measurement point n relative to measurement point m; Represents the relative entropy of measurement point k relative to measurement point m; Represents the relative entropy of measurement point 1 relative to measurement point n; Represents the relative entropy of measurement point m relative to measurement point n; Represents the relative entropy of measurement point 1 relative to measurement point k; Represents the relative entropy of measurement point m relative to measurement point k; It represents the relative entropy of measurement point n relative to measurement point n; Represents the relative entropy of measurement point k relative to measurement point n; Represents the relative entropy of measurement point n relative to measurement point k; It represents the relative entropy of measurement point k relative to measurement point k; S2.2: Calculate the transient component skewness coefficient, as shown in formula (7): (7) ； in, represents the skewness coefficient; The i-th value of the sequence to be analyzed is represented by; represents the mean value of the sequence to be analyzed; Indicates the number of values ​​in the sequence to be analyzed; S2.3: Get two preliminary fault section locations through two transient fault information, namely matrix 、 The elements in the matrix are compared 、 The middle element is obviously smaller; based on this, the fault section location (m, n) can be preliminarily determined; From formula (7), the skewness coefficient of each measurement point of the fault line can be obtained as , They represent the skewness coefficients of the fault line measurement points 1, 2, ..., m respectively; They represent the skewness coefficients of the fault line measurement points n, ..., q respectively; If the skewness coefficients from measurement point 1 to measurement point m are all less than zero, and the skewness coefficients from measurement point n to measurement point q are all greater than zero, it can be preliminarily determined that the fault section is (m, n).

4. The distribution network fault location method based on multi-information fusion of particle filter algorithm according to claim 2 is characterized by: The size of the skewness coefficient reflects the degree of asymmetry of the data distribution; when the skewness coefficient is greater than zero, it means that the data distribution is skewed to the right, that is, there are more items with larger values; when the skewness coefficient is less than zero, it means that the data distribution is skewed to the left, that is, there are more items with smaller values; the larger the absolute value of the skewness coefficient, the more obvious the skewness of the data; the distribution characteristics of the data can be judged through data distribution graphs or quantitative analysis using the skewness coefficient.

5. The distribution network fault location method based on multi-information fusion of particle filter algorithm according to claim 1 is characterized by: In step 3, a particle filter algorithm is introduced to correct the preliminary fault section location result obtained by the relative entropy coefficient using the skewness coefficient location result after preprocessing to obtain the final fault section location; The particle filter algorithm process includes the following steps: Step 3.1: Assume the system state model equation and observation model equation as shown in Equations (8) and (9): (8)； (9)； in, 、 represents a random variable, 、 For variables function; is the number of iteration steps; express About variables before iteration function; represents the predicted noise; represents the noise of the observation; Step 3.2: According to the system model of formula (8), the state of each particle is predicted to obtain the particle sample at the next moment. The prediction equation is as follows: (10)； in, represents the total number of particles; Indicates the particles; Indicates the Particle prediction weight; represents the Dirac function; Represents the particle sample at the next moment; represents the variable being predicted; Indicates the particle's current particle sample and the The unit impulse of the particle difference; Step 3.3: Based on the observation model and observation data, calculate the weight of each particle and perform normalization. The weight represents the degree of consistency between each particle and the observation data. The update equation is shown in Equation (11): (11) ； in, Indicates the Particle weights; Representing variables The posterior probability density of ; Indicates the The inverse of the sum of the weights of the particles; Indicates the Particle prediction weight; represents the total number of particles; represents the probability density of observation noise; Bring the above particle filter algorithm process into the distribution network measurement data; First, initialize the distribution network measurement points by pre-processing the skewness coefficient positioning results and perform particle initialization to obtain , indicating the Step 1 particles, , is the total number of measurement points in the distribution network, Indicates the measurement point m Step 1 particles; Indicates the measurement point n Step 1 particles; Assign initial weight values ​​to particles ; Then predict and generate samples at time t , Represents the particle sample before particle initialization; represents the noise during prediction; And calculate the corresponding weight value of the particle at time t: as shown in formula (12) (12) ； in, Indicates the weight value of the particle at time t; Represents the weight value of the particle before initialization; represents the sample at time t after particle initialization; Represents the sample at time t generated after prediction; A pulse representing the difference between the sample particle after initialization and the predicted value; Indicates the total number of measurement points in the distribution network; Then update and use the result obtained by formula (12) as the observation quantity to correct the relative entropy coefficient fault location result after distribution network preprocessing according to formula (13); (13) ； in, Represents the sample corrected after particle prediction; It represents the difference between the particle weight value after the relative entropy coefficient positioning result at time t and before the prediction; Represents the relative entropy coefficient positioning result particle initialization sample; Represents the particle weight value before the relative entropy coefficient positioning result prediction; Indicates the The inverse of the sum of the weights of the particles; Final resampling: Assume The particle at step , represent the particle samples at measurement points 1, 2, …, k respectively; The next step of particles generated after iteration is , when detected The weight value Less than the set threshold When Random number ,make , and iterate again , and repeat this process until .

6. The distribution network fault location method based on multi-information fusion of particle filter algorithm according to claim 1 is characterized by: In step 4, since the transformer connected to the DG has a delta connection, the impact of the DG on the zero-sequence component can be ignored; 、 The first The zero-sequence voltage at the beginning and end of an infinite interval; 、 is the first section in the downstream section of the fault The zero-sequence voltage at the beginning and end of an infinite interval; 、 Wireless cell 、 Zero-sequence current flowing through; 、 、 、 is the zero-sequence inductance, resistance, capacitance, and conductance per unit length of the line; the voltage on the left and right sides of the fault point 、 Satisfying formula (14): (14)； in, Indicates the line zero-sequence impedance; represents the angular frequency; From formula (14), we can get the zero-sequence voltage amplitude distribution of the fault section. The zero-sequence voltage amplitude of the fault point is the largest. Divide into infinite subintervals, 、 is the zero-sequence voltage at the beginning and end of the sub-interval.

7. The distribution network fault location method based on multi-information fusion of particle filter algorithm according to claim 1 is characterized by: In step 5, the final fault segment is cut based on the iterative algorithm and particle filtering principle: Define the flow from the head end to the end as forward, and the flow from the end end to the head end as reverse. Assuming the fault section is known ,in: is the zero-sequence power at the head end of the fault section, is the zero-sequence voltage at the head end of the fault section, is the zero-sequence power at the end of the fault section, is the zero-sequence voltage at the end of the fault section; is the positive zero-sequence transverse voltage drop of segment mk, is the positive zero-sequence longitudinal voltage drop of segment mk is the positive zero-sequence power loss of segment mk; is the positive zero-sequence transverse voltage drop of kI segment, is the positive zero-sequence longitudinal voltage drop of kI segment, is the positive zero-sequence power loss of kI segment; is the reverse zero-sequence transverse voltage drop of kI segment, is the reverse zero-sequence longitudinal voltage drop of kI segment, is the reverse zero-sequence power loss of kI segment; is the reverse zero-sequence transverse voltage drop of In segment, is the reverse zero-sequence longitudinal voltage drop of In segment, is the reverse zero-sequence power loss of In segment; is the zero-sequence impedance of the line corresponding to the cut section; Take the zero-sequence voltage and power at both ends of the fault section; at this time, forward calculate the fault section There are finite cutting points k, I zero sequence voltage 、 , the first and last zero-sequence voltage and power of the next cutting point are obtained through the forward iterative method as shown in Equations (15), (16), and (17): (15)； in, Represents the active power at the measuring point m; Represents the reactive power at the measuring point m; Represents the voltage at the measuring point m; represents the active power at the measuring point k; Represents the reactive power at the measuring point k; represents the voltage at the measuring point k; Indicates the resistance between measuring points m and k; Indicates the reactance between the measuring points m and k; (16)； (17)； Reverse calculation of fault section There are a finite number of cutting points k and zero-sequence voltage I, as shown in Equations (18), (19), and (20): (18)； in, Indicates the active power at point I during reverse calculation; Indicates the reactive power at point I during reverse calculation; The voltage at point I when the table is reversed; (19)； in, Indicates the longitudinal voltage drop in the In segment; (20)； in, Indicates the voltage at point I during reverse calculation; Indicates the voltage at point k during reverse calculation; The forward and reverse results of each cutting of the fault section are fused using the particle filter algorithm to obtain the power and voltage information of the next cutting point; then the iterative cutting is continued until the precise fault location is obtained.

Images