VSC short-circuit fault identification method and system based on wavelet approximate entropy

Abstract

The present invention provides a VSC short-circuit fault identification method and system based on wavelet approximate entropy, wherein the method includes constructing a three-phase multi-level VSC DC power distribution system; collecting capacitor voltage after metallic short-circuit fault of the DC power distribution system; performing wavelet decomposition on the capacitor voltage after the metallic short-circuit fault to obtain wavelet approximate entropy of each layer of decomposition coefficients; using wavelet approximate entropy as fault signal feature vector to form a training set; setting fault code according to fault type to form an output set; training BP neural network according to the training set and the output set to obtain the trained BP neural network; obtaining the feature vector of the line to be tested when the DC distribution network actually has a ground fault; inputting the feature vector of the line to be tested into the trained BP neural network, outputting the fault code, and realizing fault identification. Wavelet approximate entropy combines the time-frequency localization characteristics of wavelet analysis and the characteristics of approximate entropy characterizing transient signals, and can accurately detect and identify metallic short-circuit faults.

Description

Technical Field

[0001] The present invention belongs to the technical field of direct current fault detection in power systems, and in particular relates to a method and system for identifying VSC short-circuit faults based on wavelet approximate entropy. Background Art

[0002] Voltage Source Converter (VSC) is a power electronic conversion method based on fully controlled power semiconductor devices. It is the core component of the voltage source commutated DC transmission system. It usually uses turn-off devices such as IGBT, IGCT and anti-parallel diodes to form its basic unit. It is widely used in two-terminal high-voltage DC transmission systems and multi-terminal DC transmission systems. It is a key part that affects the performance, operation mode, equipment cost and operation loss of the entire conversion system. Research on control strategies for DC transmission systems based on voltage source converters has always been a very important link, which directly affects the power quality and economy.

[0003] DC distribution networks have certain advantages over AC distribution networks: (1) AC distribution networks are unable to meet my country's growing electricity demand, while DC networks have the advantages of high transmission capacity and suitability for long-distance transmission; (2) With the increase in DC loads and the development of distributed energy, pure copper AC distribution networks have high losses and poor flexibility, while DC networks do not need to consider voltage phase and frequency, thereby further improving their controllability and reliability; (3) The transmission cost of DC distribution networks is lower than that of AC distribution networks.

[0004] Faults in DC distribution networks are a key issue that affects the development of DC power grids. AC systems have been developed for a long time and already have relatively mature and complete fault identification technologies. Because the DC system started late, the fault identification and diagnosis technology of DC distribution networks is still in the theoretical research stage and has many defects. When a DC distribution network fails, the transient characteristics of the fault are more complex than those of the AC system, and the fault identification is more difficult; in addition, DC distribution networks are mostly cable lines, which are deeply buried underground, and it is difficult to implement manual inspection methods for faults. Therefore, it is of great significance to conduct rapid and reliable fault identification research on DC distribution networks. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a VSC short-circuit fault identification method and system based on wavelet approximate entropy.

[0006] In a first aspect, the present invention provides a VSC short circuit fault identification method based on wavelet approximate entropy, comprising:

[0007] Build a three-phase multi-level VSC DC distribution system;

[0008] Collect capacitor voltage after metallic short circuit fault in DC power distribution system;

[0009] The capacitor voltage after the metallic short circuit fault is decomposed by wavelet, and the wavelet approximate entropy of each layer of decomposition coefficients is obtained;

[0010] The wavelet approximate entropy is used as the fault signal feature vector to form a training set;

[0011] Set the fault code according to the fault type to form an output set;

[0012] Train the BP neural network according to the training set and the output set to obtain the trained BP neural network;

[0013] Obtaining the characteristic vector of the line to be tested when the DC distribution network actually has a ground fault;

[0014] The characteristic vector of the line to be tested is input into the trained BP neural network, and the fault code is output to realize fault identification.

[0015] Furthermore, the collecting of capacitor voltage after a metallic short circuit fault in the DC power distribution system includes:

[0016] The fault is determined according to the following formula:

[0017]

[0018] D i is the derivative of the DC side capacitor voltage at point x(i); x(i+n) and x(i) are the i+nth and ith sampling values ​​of the DC side capacitor voltage respectively; T is the signal sampling period; D th is the set fault threshold.

[0019] Furthermore, when the DC side capacitor voltage is collected and exceeds the fault threshold for the first time, the value of i is recorded and the data in the signal [iH, i+H] is intercepted as the fault period intercepted signal for fault feature extraction; wherein H is the intercepted data range.

[0020] Furthermore, the capacitor voltage after the metallic short circuit fault is subjected to wavelet decomposition to obtain the wavelet approximate entropy of each layer of decomposition coefficients, including:

[0021] Set the number of signal decomposition layers to 5, and select db5 as the wavelet basis for wavelet decomposition;

[0022] Reconstruct the wavelet decomposition coefficients and extract the signal characteristics of each frequency component in the five layers from low frequency to high frequency;

[0023] Calculate the approximate entropy of the signal in each frequency band.

[0024] Furthermore, the BP neural network sets the number of input layer nodes to 5, the number of output layer nodes to 1, and the number of neurons in the middle layer to 10; the middle layer uses the tansig function, and the output layer uses the logsig function; the training function is the trainlm function.

[0025] In a second aspect, the present invention provides a VSC short circuit fault identification system based on wavelet approximate entropy, comprising:

[0026] Building blocks for building a three-phase multi-level VSC DC power distribution system;

[0027] A collection module, used to collect capacitor voltage after a metallic short circuit fault in a DC power distribution system;

[0028] The wavelet decomposition module is used to perform wavelet decomposition on the capacitor voltage after the metallic short circuit fault to obtain the wavelet approximate entropy of each layer of decomposition coefficients;

[0029] A training set composition module is used to form a training set using wavelet approximate entropy as a fault signal feature vector;

[0030] An output set composition module is used to set the fault code according to the fault type to form an output set;

[0031] A neural network training module is used to train the BP neural network according to the training set and the output set to obtain the trained BP neural network;

[0032] An acquisition module is used to acquire a characteristic vector of a line to be tested when a DC distribution network actually has a ground fault;

[0033] The fault identification module is used to input the characteristic vector of the line to be tested into the trained BP neural network and output the fault code to realize fault identification.

[0034] Furthermore, the acquisition module includes:

[0035] The fault determination unit is used to determine the occurrence of a fault according to the following formula:

[0036]

[0037] D i is the derivative of the DC side capacitor voltage at point x(i); x(i+n) and x(i) are the i+nth and ith sampling values ​​of the DC side capacitor voltage respectively; T is the signal sampling period; D th is the set fault threshold.

[0038] Furthermore, the wavelet decomposition module comprises:

[0039] The decomposition layer number setting unit is used to set the signal decomposition layer number to 5 and select db5 as the wavelet basis for wavelet decomposition;

[0040] The coefficient reconstruction unit is used to reconstruct the wavelet decomposition coefficients and extract the signal characteristics of each frequency component in the five layers from low frequency to high frequency;

[0041] The calculation unit is used to calculate the approximate entropy of the signal in each frequency band range.

[0042] The present invention provides a VSC short-circuit fault identification method and system based on wavelet approximate entropy, wherein the method collects the capacitor voltage after the metallic short-circuit fault of the DC power distribution system as the original data, and trains the BP neural network for identification by obtaining the wavelet approximate entropy of the DC side capacitor voltage as the fault signal feature vector; the wavelet approximate entropy combines the time-frequency localization characteristics of wavelet analysis and the characteristics of approximate entropy characterizing transient signals, and can accurately detect and identify metallic short-circuit faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 A flow chart of a VSC short circuit fault identification method based on wavelet approximate entropy provided in an embodiment of the present invention;

[0045] Figure 2 A topological diagram of a three-phase two-level VSC DC power distribution system provided in an embodiment of the present invention;

[0046] Figure 3 A wavelet coefficient diagram of a fault signal provided by an embodiment of the present invention;

[0047] Figure 4 The MSE variation trend diagram of the BP neural network provided by the embodiment of the present invention;

[0048] Figure 5 An output result error distribution diagram provided by an embodiment of the present invention;

[0049] Figure 6 A BP neural network fitting relationship diagram provided by an embodiment of the present invention;

[0050] Figure 7 A diagram showing the fault type identification results provided by an embodiment of the present invention;

[0051] Figure 8 A schematic structural diagram of a VSC short-circuit fault identification system based on wavelet approximate entropy provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] like Figure 1 As shown, a VSC short circuit fault identification method based on wavelet approximate entropy is provided in some embodiments of the present invention, including:

[0054] Step 101: construct a three-phase multi-level VSC direct current distribution system.

[0055] like Figure 2 As shown, a three-phase two-level VSC DC distribution system simulation model is built on the simulation software to simulate various metallic short circuits through circuit breakers.

[0056] Step 102: collecting capacitor voltage after a metallic short circuit fault occurs in the DC power distribution system.

[0057] Collect the DC side capacitor voltage u when various short circuits occur c , the sampling time interval is 0.001s.

[0058] When a fault occurs in the DC power distribution system, the capacitor voltage on the DC side will change suddenly, and the derivative value of the capacitor voltage will be very large. Because when the DC power distribution network is operating normally, the capacitor voltage on the DC side also fluctuates to a certain extent. In order to prevent misdiagnosis of faults, the fault is determined by setting the derivative threshold reasonably.

[0059] The fault is determined according to the following formula:

[0060]

[0061] D i is the derivative of the DC side capacitor voltage at point x(i); x(i+n) and x(i) are the i+nth and ith sampling values ​​of the DC side capacitor voltage respectively; T is the signal sampling period; D th is the set fault threshold. The value of the fault threshold is related to the voltage fluctuation of the DC capacitor during normal operation of the power grid and the sampling period T. Therefore, D th The size is set according to the sampling frequency. In this embodiment, D th The value is 100. When the DC side capacitor voltage signal is collected and satisfies the above formula for the first time, the line fails, the value of i is recorded, and the data in the signal [iH, i+H] is intercepted as the fault period intercepted signal for fault feature extraction; where H is the intercepted data range, and the value of H in this embodiment is 500.

[0062] Step 103: Perform wavelet decomposition on the capacitor voltage after a metallic short - circuit fault to obtain the wavelet approximate entropy of the decomposition coefficients at each layer.

[0063] Through wavelet decomposition, the fault signal can be decomposed into different independent frequency bands at different levels. Suppose the signal decomposition level is 5. The dbN wavelet has good regularity. As the order N increases, the localization ability of the signal in the frequency domain becomes stronger, and the tight support in the time domain weakens. In order to obtain more excellent detail coefficients in time - frequency resolution, db5 is selected as the wavelet basis for wavelet decomposition.

[0064] Reconstruct the wavelet decomposition coefficients. The wavelet coefficient diagrams of different fault signals are as Figure 3 shown. Extract the signal features of each frequency component in the 5 layers from low - frequency to high - frequency.

[0065] The approximate entropy of the signal in each frequency band range contains a large amount of unbalanced and non - linear information. Calculate the approximate entropy of the signal in each frequency band range to form the fault signal feature vector. The approximate entropy calculation method is as follows:

[0066] ① Suppose the original data collected is u(1), u(2), …, u(N), a total of N points. Predetermine the pattern dimension m = 2 and the similarity tolerance r = 0.15STD, where STD is the standard deviation of the original data u(i). The sequence {u(i)} is composed into an m - dimensional vector in order:

[0067] X(i)=[u(i), u(i + 1), …, u(i + m - 1)] i = 1, 2, …, N - m - 1

[0068] ② Calculate the distance between the vector X(i) and the remaining vectors X(j) for each value.

[0069]

[0070] ③ Given a threshold r (r > 0), count the number of d[X(i), X(j)] < r for each value and the ratio of this number to the total number of vectors N - m + 1, denoted as

[0071]

[0072] ④ Take the logarithm and then calculate its average value, denoted as φ m (r).

[0073]

[0074] ⑤ Then increase the dimension m by 1 to become m + 1, and repeat steps ① - ④ to obtain c i m+1(r) and φ m+1 (r).

[0075] ⑥ The approximate entropy ApEn of this sequence:

[0076] ApEn(m,r)=φ m (r)-φ m+1 (r).

[0077] Step 104, using wavelet approximate entropy as the fault signal feature vector to form a training set.

[0078] Step 105, setting a fault code according to the fault type, and forming an output set from the fault codes.

[0079] The corresponding relationship between fault types and output fault codes is shown in Table 1.

[0080] Table 1 Fault code table

[0081]

[0082]

[0083] Step 106, training the BP neural network according to the training set and the output set to obtain a trained BP neural network.

[0084] Number of input layer nodes: The fault signal is decomposed into 5 layers and the wavelet approximate entropy is calculated, so the number of input layer nodes is 5.

[0085] Number of nodes in the output layer: The number of nodes is 1. Since the output result is a decimal, the output is rounded off and used as the result.

[0086] Determination of the number of nodes in the middle layer: Set the number of neurons in the middle layer to 10.

[0087] Selection of intermediate layer and output layer functions: Generally, the intermediate layer uses the tansig function and the output layer uses the logsig function.

[0088] The training function is selected as the trainlm function.

[0089] 180 sets of fault signal features were obtained, of which 120 sets were used as training data to train the network. The training data, verification data, and test data were allocated in a ratio of 70%, 15%, and 15%. The MSE of the training results changes with the number of iterations as shown in the figure below: Figure 4 As shown. Figure 4 It can be seen that after 40 iterations, the MSE change trend tends to be stable.

[0090] The error histogram between the output value and the actual fault code is shown in Figure 5As shown in the figure, when the error is less than 0.5, the output result is rounded to the nearest integer and becomes the fault type code. Figure 5 It can be seen that among the 120 sets of data, 118 sets of output results are consistent with the actual situation.

[0091] The closer the R value is to 1, the better the fit is, and the closer it is to 0, the worse the correlation of the data is. Figure 6 As shown in the figure, the R of the data fitting for training, validation, and testing in the model is very close to 1, and this model can be used to diagnose fault categories.

[0092] Step 107, obtaining a characteristic vector of the line to be tested when the DC distribution network actually has a ground fault.

[0093] Step 108: input the feature vector of the line to be tested into the trained BP neural network, and output the fault code to realize fault identification.

[0094] In order to verify the accuracy of the diagnosis results of the fault diagnosis system, the VSC DC side voltage is collected as a test sample, and the DC side fault and AC side fault are simulated by changing the fault occurrence conditions. 20 sets of data are taken for each fault type, for a total of 120 sets of data. Figure 7 shown.

[0095] The fault identification results are summarized in Table 2. The total recognition accuracy of the test samples is above 90%. The DC distribution network fault identification accuracy of this model is low and the effect is poor. It has higher diagnostic accuracy and robustness. It verifies that the approximate entropy of each level calculated by wavelet decomposition as the feature vector as the training and verification sample of the fault pattern recognition network can effectively determine the VSC DC and AC fault categories.

[0096] Table 2 Fault identification results

[0097]

[0098] like Figure 8 As shown, an embodiment of the present invention further provides a VSC short circuit fault identification system based on wavelet approximate entropy, comprising:

[0099] A construction module 10 is used to construct a three-phase multi-level VSC DC power distribution system;

[0100] The acquisition module 20 is used to acquire the capacitor voltage after the metallic short circuit fault of the DC power distribution system;

[0101] The wavelet decomposition module 30 is used to perform wavelet decomposition on the capacitor voltage after the metallic short circuit fault to obtain the wavelet approximate entropy of each layer of decomposition coefficients;

[0102] A training set forming module 40 is used to form a training set using wavelet approximate entropy as a fault signal feature vector;

[0103] An output set composition module 50 is used to set a fault code according to the fault type to form an output set;

[0104] A neural network training module 60 is used to train the BP neural network according to the training set and the output set to obtain a trained BP neural network;

[0105] An acquisition module 70 is used to acquire a characteristic vector of a line to be tested when a DC distribution network actually has a ground fault;

[0106] The fault identification module 80 is used to input the characteristic vector of the line to be tested into the trained BP neural network and output the fault code to realize fault identification.

[0107] Optionally, the acquisition module includes:

[0108] The fault determination unit is used to determine the occurrence of a fault according to the following formula:

[0109]

[0110] D i is the derivative of the DC side capacitor voltage at point x(i); x(i+n) and x(i) are the i+nth and ith sampling values ​​of the DC side capacitor voltage respectively; T is the signal sampling period; D th is the set fault threshold.

[0111] Optionally, the wavelet decomposition module includes:

[0112] The decomposition layer number setting unit is used to set the signal decomposition layer number to 5 and select db5 as the wavelet basis for wavelet decomposition;

[0113] The coefficient reconstruction unit is used to reconstruct the wavelet decomposition coefficients and extract the signal characteristics of each frequency component in the five layers from low frequency to high frequency;

[0114] The calculation unit is used to calculate the approximate entropy of the signal in each frequency band range.

[0115] In this specification, the same or similar parts between the various embodiments can be referred to each other. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description in the method embodiment.

[0116] The present invention has been described in detail above in conjunction with specific implementations and exemplary examples, but these descriptions cannot be understood as limiting the present invention. Those skilled in the art understand that, without departing from the spirit and scope of the present invention, a variety of equivalent substitutions, modifications or improvements may be made to the technical solution of the present invention and its implementation methods, all of which fall within the scope of the present invention. The scope of protection of the present invention shall be subject to the attached claims.

Claims

1. A VSC short-circuit fault identification method based on wavelet approximate entropy, characterized in that: include: Build a three-phase multi-level VSC DC distribution system; Collect capacitor voltage after metallic short circuit fault in DC power distribution system; The capacitor voltage after the metallic short circuit fault is decomposed by wavelet, and the wavelet approximate entropy of each layer of decomposition coefficients is obtained; The wavelet approximate entropy is used as the fault signal feature vector to form a training set; Set the fault code according to the fault type to form an output set; Train the BP neural network according to the training set and the output set to obtain the trained BP neural network; Obtaining the characteristic vector of the line to be tested when the DC distribution network actually has a ground fault; Input the characteristic vector of the line to be tested into the trained BP neural network, and output the fault code to realize fault identification; The collecting of capacitor voltage after a metallic short circuit fault in the DC power distribution system includes: The fault is determined according to the following formula: D i is the derivative of the DC side capacitor voltage at point x(i); x(i+n) and x(i) are the i+nth and ith sampling values ​​of the DC side capacitor voltage respectively; T is the signal sampling period; D th is the set fault threshold.

2. The VSC short circuit fault identification method according to claim 1, characterized in that: When the DC side capacitor voltage exceeds the fault threshold for the first time, the value of i is recorded and the data in the signal [iH, i+H] is intercepted as the fault period interception signal to extract fault features; Where H is the intercepted data range.

3. The VSC short circuit fault identification method according to claim 1, characterized in that: The capacitor voltage after the metallic short circuit fault is subjected to wavelet decomposition to obtain the wavelet approximate entropy of each layer of decomposition coefficients, including: Set the number of signal decomposition layers to 5, and select db5 as the wavelet basis for wavelet decomposition; Reconstruct the wavelet decomposition coefficients and extract the signal characteristics of each frequency component in the five layers from low frequency to high frequency; Calculate the approximate entropy of the signal in each frequency band.

4. The VSC short circuit fault identification method according to claim 1, characterized in that: The BP neural network sets the number of input layer nodes to 5, the number of output layer nodes to 1, and the number of neurons in the middle layer to 10; the middle layer uses the tansig function, and the output layer uses the logsig function; The training function is selected as the trainlm function.

5. A VSC short-circuit fault identification system based on wavelet approximate entropy, characterized in that: include: Building blocks for building a three-phase multi-level VSC DC power distribution system; A collection module, used to collect capacitor voltage after a metallic short circuit fault in a DC power distribution system; The wavelet decomposition module is used to perform wavelet decomposition on the capacitor voltage after the metallic short circuit fault to obtain the wavelet approximate entropy of each layer of decomposition coefficients; A training set composition module is used to form a training set using wavelet approximate entropy as a fault signal feature vector; An output set composition module is used to set the fault code according to the fault type to form an output set; A neural network training module is used to train the BP neural network according to the training set and the output set to obtain the trained BP neural network; An acquisition module is used to acquire a characteristic vector of a line to be tested when a DC distribution network actually has a ground fault; The fault identification module is used to input the characteristic vector of the line to be tested into the trained BP neural network and output the fault code to realize fault identification; The acquisition module comprises: The fault determination unit is used to determine the occurrence of a fault according to the following formula: D i is the derivative of the DC side capacitor voltage at point x(i); x(i+n) and x(i) are the i+nth and ith sampling values ​​of the DC side capacitor voltage respectively; T is the signal sampling period; D th is the set fault threshold.

6. The VSC short circuit fault identification system according to claim 5, characterized in that: The wavelet decomposition module comprises: The decomposition layer number setting unit is used to set the signal decomposition layer number to 5 and select db5 as the wavelet basis for wavelet decomposition; The coefficient reconstruction unit is used to reconstruct the wavelet decomposition coefficients and extract the signal characteristics of each frequency component in the five layers from low frequency to high frequency; The calculation unit is used to calculate the approximate entropy of the signal in each frequency band range.

Images