Three-terminal hybrid direct current transmission line protection method based on line mode power characteristics

Abstract

A three-terminal hybrid direct current transmission line protection method based on line mode power characteristics comprises the following steps: building a three-terminal hybrid direct current transmission model, and collecting direct current line mode voltage and line mode power signals by analyzing fault characteristics of different fault areas; calculating the line mode power break variable, the line mode power high-frequency energy and the voltage break variable of the measurement point during the fault as fault characteristic quantities; building a BiTCN network fault identification model, and carrying out offline training and online verification on the BiTCN network fault identification model; and taking the fault characteristic quantity as an input quantity of a BiTCN network fault identification model, taking the fault type as an output quantity, and constructing a line protection criterion based on the BiTCN network fault identification model. According to the method, the input characteristic quantity of the BiTCN network model is identified by analyzing the line mode power characteristics when the three-terminal hybrid direct current transmission line fails, and the difficulty of manual threshold setting can be avoided. And the output of the model can be more easily associated with the input sequence, so that the calculation efficiency is higher, and the protection speed is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of direct current line protection, and in particular to a three-terminal hybrid direct current transmission line protection method based on line mode power characteristics. Background Art

[0002] In recent years, the three-terminal hybrid DC transmission system, as a new type of power transmission method, has attracted widespread attention in the electrical field at home and abroad. It has the advantages of mature traditional DC technology, low cost, good flexible DC regulation capability, and no commutation failure, making it one of the effective ways of long-distance and large-capacity transmission.

[0003] As the first line of defense, line protection is of great significance for maintaining the reliability and safety of DC transmission systems. Since the topology and boundary conditions of the three-terminal hybrid DC transmission system are significantly different from those of traditional DC transmission and flexible DC transmission, and there are problems such as the wide span of overhead lines, harsh working environment, and high probability of failure, fast and reliable line protection is required to ensure safe and stable operation of the system.

[0004] In recent years, the emergence of artificial intelligence algorithms has provided new ideas for fault identification. For example, the document "Flexible DC Distribution Line Protection Scheme Based on GRU Network [J]. Power System Technology, 2021, 45(10): 3885-3896" selects the high-frequency transient energy and bus voltage of DC voltage as input feature quantities, constructs a gated cyclic unit network model, realizes the protection of transmission lines, and meets the reliability and speed requirements of DC protection. However, the model used in this document has difficulties in extracting features from sequences, and it is difficult to capture the dependencies between data features. The model has poor generalization ability, and there is a risk of false operation and refusal to operate the protection.

[0005] Many scholars use the current-limiting reactors installed at both ends of the line as boundaries, and then construct the protection principle of DC lines based on the different characteristics of the boundaries. However, compared with the traditional two-terminal DC transmission method, the hybrid multi-terminal DC system has a more complex topology and a more flexible operation mode. The boundaries between the rectifier and inverter sides are asymmetric, and the T-zone structure formed by connecting adjacent lines through the busbar has no obvious boundaries. The above schemes are difficult to apply directly, which brings new challenges to the protection design of hybrid multi-terminal DC lines.

[0006] Although the above DC line protection method has certain advantages, it generally has the problem that feature extraction is complex and threshold analysis and adjustment are required to accurately identify faults, while the adjustment process is difficult and the workload is large. Summary of the invention

[0007] In view of the above problems, the present invention proposes a three-terminal hybrid DC transmission line protection method based on line mode power characteristics, which can identify full line faults, has strong anti-interference ability, and is not affected by AC faults and weak boundary elements. The technical solution adopted by the present invention is:

[0008] A three-terminal hybrid DC transmission line protection method based on line mode power characteristics comprises the following steps:

[0009] Step 1: Build a three-terminal hybrid DC transmission model, analyze the fault characteristics of different fault areas, and collect DC line mode voltage and line mode power signals;

[0010] Step 2: Calculate the line mode power mutation, line mode power high frequency energy, and voltage mutation at the measurement point during the fault as fault characteristic quantities;

[0011] Step 3: Build a BiTCN network fault identification model, train it offline, and verify it online;

[0012] Step 4: Take the fault feature quantity as the input quantity of the BiTCN network fault identification model, and the fault type as the output quantity, and construct the line protection criterion based on the BiTCN network fault identification model.

[0013] In step 1, PSCAD / EMTDC is used to build Figure 1 The three-terminal hybrid DC transmission model shown in the figure, the fault characteristics of different areas are analyzed as follows:

[0014] 1) When a fault occurs in the T zone and the lines on both sides, the line mode power mutation is as follows:

[0015]

[0016] Where: They are the line mode current mutations of lines T1 and T2 on both sides of the T zone, and the positive direction is the direction during normal operation; when a fault occurs in the T zone, Opposite polarity, Same polarity;

[0017] When the left side of the T zone fails, and The polarity is the same;

[0018] When the right side of the T zone fails, and The polarities are opposite.

[0019] Therefore: When the T zone fails: When the left side of the T zone fails: When the right side of the T zone fails:

[0020] 2) The fault analysis inside and outside the line area I is as follows:

[0021] When the frequency is greater than 2000Hz, the high-frequency component of the DC power of line I will be blocked by the boundary elements and thus attenuated; when the current flows through the converter equivalent impedance Z LCC When the voltage and current frequency at both ends of Z are high, LCC Contains only the reactance component, i.e. Z LCC =jωL LCC , the DC filter can be equivalent to a fixed inductor L in the high frequency band filter Therefore, there is a relationship between the current and voltage at both ends of the rectifier side boundary:

[0022]

[0023] In formula (2): Z filter is the filter equivalent impedance; Z Ldc is the equivalent impedance of the current limiting reactor; Z LCC is the equivalent impedance of the LCC converter; Measure the voltage in area I (Kunbei side) of the line; Measure voltage outside line zone I (Kunbei side); To measure the current in area I (Kunbei side) of the line; The current is measured outside line zone I (Kunbei side).

[0024] It can be seen from formula (2) that when a fault occurs in the line I, the high-frequency fault components of the current and voltage on both sides of the boundary satisfy: That is, P m >P a , P m is the external line mode power of line area I (Kunbei side); P a It is the line mode power in line area I (Kunbei side).

[0025] The relationship between the current and voltage components at the line boundary measurement point during an out-of-zone fault is as follows:

[0026]

[0027] Where: Z MMC1 is the equivalent impedance of the MMC converter.

[0028] In the high frequency band, the equivalent inductance of the DC filter is much smaller than the inductance of the smoothing reactor. It can be seen that: That is, P m <P a .

[0029] 3) The fault analysis inside and outside the line zone II is as follows:

[0030] Since there is no filter at the inverter side MMC end in the three-terminal hybrid DC transmission system, only the current limiting reactor is used as the boundary element of the line. In order to make the protection proposed by the present invention have sufficient reliability and sensitivity, according to the capacitance between the conductor and the earth, a section of overhead line to ground capacitance and the current limiting reactor are used to form the line boundary in the converter station of line II area (Longmen side). When the frequency is greater than 5000Hz, the high-frequency component of the DC power of line II will also be blocked by the boundary, resulting in attenuation;

[0031] There is a relationship between the current and voltage fault components at both ends of the inverter side boundary:

[0032]

[0033] In formula (4): Measure the voltage in line zone II (gantry side); Measure voltage outside line zone II (gantry side); To measure the current outside the line zone II (gantry side); To measure the current in line zone II (gantry side).

[0034] Therefore, when an internal fault occurs in line II, the boundary fault current and voltage satisfy: When an out-of-zone fault occurs on line II, the current and voltage meet the following conditions: Therefore, when the fault occurs in the line zone II, the following conditions are satisfied: P n >P c ; When the fault occurs outside the zone: P n <P c , P n P is the line mode power in line zone II (gantry side); c It is the external line mode power of line zone II (gantry side).

[0035] DC line mode voltage, line mode power signal:

[0036] In order to avoid the electromagnetic coupling relationship between the positive and negative poles of the DC line, phase mode transformation is used to convert various electrical quantities into line mode and ground mode components.

[0037]

[0038] Where U p and U d are the measured positive and negative fault voltages respectively; U0 and U1 are the ground mode and line mode components of the fault voltage respectively.

[0039] Define line mode power as:

[0040] P1=I1*U1 (6);

[0041] Where I1 and U1 are the line mode components of the fault current and voltage respectively.

[0042] In step 2, the fault characteristic quantity is selected as follows:

[0043] During system operation, electrical quantities will continue to fluctuate. When a line fails, voltages of both lines will change suddenly. The one with the larger amplitude of the sudden change will be selected as the starting criterion to improve the protection sensitivity.

[0044] (1) Fault direction identification feature:

[0045] In order to facilitate analysis, the discrete line mode power mutation data points are integrated and summed, and the expression is:

[0046]

[0047] Where: N is the number of discrete points in the data window; They are the line mode power mutation values ​​at the measuring points T1 and T2 on both sides of the busbar in the T zone; is the integral value of the mutation amount.

[0048] When a fault occurs in the T zone, The polarity is opposite; when the DC line fails, the polarity is the same, and when the line on the left side of the T zone fails All are positive poles, and all are negative poles when the lines on the right side of the T zone fail. The polarity difference of the line mode power mutation when faults occur in different directions can be used to form the fault direction identification feature.

[0049] (2) Fault identification characteristics inside and outside the zone:

[0050] The sampling frequency of the present invention is set to 20kHz, and a filter is used to extract the voltage and current fault components in the range of 5 to 7.5kHz to calculate the line mode power high-frequency component. Since wavelet transform, as a signal processing tool, can analyze signals at different scales and better locate signal characteristics, wavelet transform is used to extract the line mode power high-frequency component to distinguish the in-zone and out-of-zone faults of different lines. The fault transient current obtained at the measurement point is decomposed by db4 wavelet 5 layers, and the first layer of wavelet coefficients corresponding to the line mode power high-frequency component are extracted. The wavelet energy of the line mode power high-frequency component at the line boundary measurement point is defined as:

[0051]

[0052] Where P a , P m , P n , P c E is the line mode high frequency power at the measurement points on both sides of the two line boundaries, which is equal to the product of the line mode current fault component and the line mode voltage fault component in the corresponding frequency band.i Represents the wavelet energy of the high-frequency component of the line mode power; P i (n) represents the line mode power of the measurement point; a represents the measurement point outside the line area I (Kunbei side); m represents the measurement point inside the line area I (Kunbei side); n represents the measurement point inside the line area II (Longmen side); c represents the measurement point outside the line area II (Longmen side); N represents the number of wavelet coefficients.

[0053] When a line fails, only the energy detected at the measurement point close to the fault fluctuates greatly, while the other fluctuations are small. Therefore, the high-frequency wavelet energy characteristics of the line mode power can be used to form the characteristic quantity for identifying faults inside and outside the area.

[0054] (3) Fault selection:

[0055] Since the system adopts a true bipolar operation structure, when a single-pole grounding fault occurs, the voltage amplitude of the fault pole line changes significantly compared to the non-fault pole line; when a bipolar grounding fault occurs, both the positive and negative pole voltages will fluctuate, and the amplitude changes are similar. Therefore, the magnitude of the voltage mutation amplitude can be used to achieve fault pole selection. For ease of analysis, the discrete voltage mutation data points are integrated and summed, and the expression is:

[0056]

[0057] Where: △U p (n), △U d (n) are the voltage mutation of positive and negative electrodes respectively; U p , U d is the integral value of voltage mutation;

[0058] When the positive pole of the DC line fails, the voltage mutation of the positive pole is greater than that of the negative pole; when the negative pole of the DC line fails, the voltage mutation of the negative pole is greater than that of the positive pole; when the DC line has a bipolar fault, the voltage amplitudes of the positive and negative poles change greatly, and the voltage changes of bipolar faults are roughly the same. Therefore, the voltage mutation characteristics can be used to form the fault pole selection characteristics.

[0059] In step 3, the BiTCN network fault identification model is constructed as follows:

[0060] The core of TCN is the convolution operation, which processes time series data through convolution layers. Compared with CNN, the size of the convolution kernel of TCN is usually variable, and the size of the receptive field is increased by stacking multiple convolution layers.

[0061] Assume that the convolution kernel is K = (k1, k2, ..., k m ), k1, k2, ..., k m The specific weight values ​​in the convolution kernel are respectively; m is the size of the convolution kernel. The input time series is α=(h1,h2,...,h n), h1,h2,...,h n Represent the value of each time point respectively; n is the sequence size.

[0062] Then α n The causal convolution of is represented as:

[0063]

[0064] Where: p n-m+1 represents input data; k i Represent the weight of each convolution kernel respectively.

[0065] Using dilated convolutions to avoid the vanishing gradient problem can increase the receptive field without increasing the number of parameters. A key feature of dilated convolutions is that they can capture information of different scales at each level, which helps capture long-range dependencies in temporal data. By reducing the number of convolutional neurons between two adjacent layers by the value of the dilation coefficient l, α n The dilated convolution can be expressed as:

[0066]

[0067] Where: d represents the expansion factor, which grows exponentially with 2; p n-(m-1)d Represents input data.

[0068] In order to strengthen network training and improve model performance, residual connections are usually used in TCN. The output of each convolutional layer is added to its input to form a residual block, which makes the network easier to optimize. Part of the residual block is composed of dilated causal convolution and nonlinear mapping, which can be expressed as:

[0069] K(α i )=ω2(Relu(ω1α i )) (12);

[0070] Where: ω1, ω2 represent the weights of the convolutional layer; Relu represents the activation function, which activates the output of the residual module; Relu(ω1α i ) represents the activation function of each convolutional layer.

[0071] The other part is a one-dimensional full convolution operation, which makes the input and output sequences have the same length. The output can be expressed as

[0072] P(α i )=K(α i )+H i (13);

[0073] P i+1 =Relu(P(α i )) (14);

[0074] Where: P(α i ) represents the convolution output; K(α i ) represents the residual block; H i Represents the input sequence.

[0075] P i+1 Represents the full convolution output result; Relu(P(α i )) represents the activation function of the convolution output.

[0076] However, the traditional TCN model only performs forward convolution calculation on the input sequence, ignoring the implicit information in the backward direction. In order to effectively extract local and global features, the present invention adopts a BiTCN network model with strong feature capture capability, which can mine deep fault characteristics from complex time series samples and is suitable for processing information processing problems that need to consider multiple factors, conditions and fuzzy information. The BiTCN network model structure is as follows: Figure 2 As shown in the figure. It includes inputting fault feature matrix, bidirectional sliding window processing, two-layer TCN model, fusing bidirectional features, and finally outputting the result through the Softmax output layer. Causal dilation convolution and residual connection are retained on the basis of TCN. On the one hand, the dependency relationship of data from front to back is learned through forward data, and on the other hand, the dependency relationship of data from back to front is learned through reverse data. Finally, the data learned from the two directions are fused to obtain the dependency relationship between the data in the forward and reverse propagation directions.

[0077] BiTCN network model training:

[0078] Different fault characteristics are used as input of the BiTCN network model They represent the line mode power mutation on both sides of the busbar in the T zone; E m ,E a ,E n ,E c Respectively represent the wavelet energy of the high-frequency components of the line mode power on both sides of the line Ⅰ and Ⅱ boundaries; U p ,U d Respectively represent the integral value of the voltage mutation of the positive and negative poles of the line;

[0079] The output results are various fault types, including positive pole fault in line zone I, negative pole fault in line zone I, bipolar fault in line zone I, fault outside line zone I, positive pole fault in line zone II, negative pole fault in line zone II, bipolar fault in line zone II, fault outside line zone II, positive pole fault in T zone, and negative pole fault in T zone.

[0080] With reference to actual projects such as Wudongde, in order to facilitate the comparison of feature changes before and after the fault and the need for fault data analysis, data 2ms before the fault and 3ms after the fault were extracted. The acquisition of the data set comprehensively considered influencing factors such as fault distance (collecting 1%, 25%, 50%, and 75% of the line respectively), transition resistance (the interval was 0.01~600Ω, and the change step was 10Ω) to ensure sufficient samples for offline training of the BiTCN network model.

[0081] A BiTCN network model was constructed on the MATLAB platform. The normalized data was divided into training set and validation set in a ratio of 4:1. Parameters such as iterative learning and number of iterations were adjusted. After the adjustment, the fault diagnosis accuracy of the BiTCN network model could achieve accurate output of 10 different types of faults.

[0082] In step 4, the three-terminal hybrid DC transmission protection process is as follows:

[0083] S4101. Build a three-terminal hybrid DC transmission model;

[0084] S4102, collecting fault voltage and current signals at the measuring point;

[0085] S4103: Determine whether the start criterion is met based on the obtained signal △U Ⅰ , △U Ⅱ Indicates the line mode voltage mutation of the two lines; Indicates the start criterion for the setting; max(△U Ⅰ ,△U Ⅱ ) represents the maximum value of the two; considering a certain margin, set If satisfied, proceed to S4104;

[0086] S4104, calculating fault characteristic quantities: line mode power mutation quantity, line mode high frequency wavelet energy, line mode voltage mutation quantity;

[0087] S4105. Build a BiTCN network model and perform offline training on it;

[0088] S4106, taking the fault feature quantity as the input quantity of the BiTCN network model, and taking the fault type as the output quantity;

[0089] S4107, when the fault occurs, data is input and the BiTCN network model outputs the corresponding fault type;

[0090] S4108, corresponding protection action.

[0091] Line protection criterion: The voltage mutation is used as the starting criterion. When a DC fault occurs, the voltage change rate at both ends of the line is significantly greater than the normal fluctuation range. After the protection criterion is activated, when a fault occurs in the T area and the DC lines on both sides, the line mode power mutation quantity has obvious characteristics, which constitutes the fault direction identification criterion; when a fault occurs inside or outside the line area, the fault energy difference between the corresponding measurement points inside and outside the line area is obvious, which constitutes the fault identification inside and outside the line area; when a fault occurs at different poles of the line, the voltage mutation quantity has obvious characteristics, which constitutes the fault pole selection criterion.

[0092] The present invention provides a three-terminal hybrid DC transmission line protection method based on line mode power characteristics, which has the following beneficial effects:

[0093] 1) The present invention uses line mode power characteristics to construct protection criteria and trains the BiTCN network model to identify fault types. This method can identify full-line faults and has strong anti-interference performance and is not affected by AC faults and weak boundary elements.

[0094] 2) Since traditional protection algorithms have problems such as difficulty in threshold setting and difficulty in detecting high-resistance faults, the present invention establishes a BiTCN network model, realizes fault diagnosis through an intelligent algorithm, avoids the problem of complex manual threshold setting, and the method of the present invention has good high-resistance resistance.

[0095] 3) In a three-terminal hybrid DC transmission system, when a line fails, the signal will have strong nonlinearity and be easily affected by sampling frequency, weak boundaries, etc. Therefore, the present invention jointly determines the fault type based on directional elements and boundary conditions, so as to achieve being unaffected by sampling frequency and weak boundaries. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0097] Figure 1 This is a diagram of a three-terminal hybrid DC transmission system.

[0098] Figure 2 This is the BiTCN network model structure diagram.

[0099] Figure 3 This is the protection flow chart of three-terminal hybrid DC transmission.

[0100] Figure 4 This is the fault simulation result diagram under different transition resistances.

[0101] Figure 5 This is the fault simulation result diagram under weak boundary conditions.

[0102] Figure 6 This is the fault simulation result diagram under AC line fault. DETAILED DESCRIPTION

[0103] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0104] Taking the ±800kV Wudongde three-terminal DC transmission project as a reference, the system uses LCC converters at the sending end and MMC converters with fault self-clearing capabilities are connected in parallel at the receiving end. Current-limiting reactors are installed on all lines, and DC filters are also provided on the LCC side.

[0105] The goal of the three-terminal hybrid DC transmission protection scheme is to quickly extract fault characteristics, formulate safe and reliable protection criteria, accurately identify the fault area, and enable the protection to operate correctly;

[0106] The goal of fault area identification is to be able to identify faults in the entire line, have a good ability to tolerate transition resistance, and meet the requirements of speed and reliability.

[0107] A three-terminal hybrid DC transmission line protection scheme based on line mode power characteristics,

[0108] First, analyze the differences in fault area characteristics. Secondly, extract the line mode power mutation, high-frequency energy and voltage mutation during the fault as fault characteristics through wavelet transform, train the BiTCN network model offline, and verify it online. Finally, the identification of line selection and pole selection for different faults of three-terminal hybrid DC lines is realized. On the basis of traditional methods, artificial intelligence algorithms are introduced to mine deep abstract features in the data through training, so as to handle various working conditions and abnormal situations and improve the reliability of protection. The specific steps include the following:

[0109] S1. Build a three-terminal hybrid DC transmission model through PSCAD / EMTDC, analyze the fault characteristics of different fault areas, and collect the DC line line mode voltage and line mode power signals.

[0110] The fault characteristics of different areas are analyzed as follows:

[0111] 1) When a fault occurs in the T zone and the lines on both sides, the line mode power mutation is as follows:

[0112]

[0113] Where: It is the sudden change of line mode current of lines T1 and T2 on both sides of zone T, and the positive direction is the direction during normal operation;

[0114] T zone failure Opposite polarity, Same polarity; when fault occurs on the left side of T zone and The polarity is the same; when the fault occurs on the right side of the T zone and The polarities are opposite. Therefore: When the T zone fails: When the left side of the T zone fails: When the right side of the T zone fails:

[0115] 2) The fault analysis inside and outside the line area I is as follows:

[0116] When the frequency is greater than 2000Hz, the high-frequency component of the DC power of line I will be blocked by the boundary elements and attenuated. LCC When the voltage and current frequency at both ends of Z are high, LCC Contains only the reactance component, i.e. Z LCC =jωL LCC , the DC filter can be equivalent to a fixed inductor L in the high frequency band filter Therefore, there is a relationship between the current and voltage at both ends of the rectifier side boundary:

[0117]

[0118] Where: Z filter is the filter equivalent impedance; Z Ldc is the equivalent impedance of the current limiting reactor; Z LCC is the equivalent impedance of the LCC converter.

[0119] It can be seen from formula (2) that when a fault occurs in the line I, the high-frequency fault components of the current and voltage on both sides of the boundary satisfy: That is, P m >P a .

[0120] The relationship between the current and voltage components at the line boundary measurement point during an out-of-zone fault is as follows:

[0121]

[0122] Where: Z MMC1 is the equivalent impedance of the MMC converter.

[0123] In the high frequency band, the equivalent inductance of the DC filter is much smaller than the inductance of the smoothing reactor. It can be seen that: That is, P m <P a .

[0124] 3) The fault analysis inside and outside the line area I is as follows:

[0125] Since there is no filter at the inverter side MMC end in the three-terminal hybrid DC transmission system, only the current limiting reactor is used as the boundary element of the line. In order to make the protection proposed by the present invention have sufficient reliability and sensitivity, according to the capacitance between the conductor and the earth, a section of overhead line to ground capacitance and the current limiting reactor are used on the MMC2 side to form the line boundary. When the frequency is greater than 5000Hz, the high-frequency component of the DC power of line II will also be blocked by the boundary, resulting in attenuation.

[0126] There is a relationship between the current and voltage fault components at both ends of the inverter side boundary:

[0127]

[0128] Therefore, when an internal fault occurs on line II, the boundary fault current and voltage satisfy: When an out-of-zone fault occurs on line II, the current and voltage meet the following conditions: Therefore, when a fault occurs in line zone II, the following conditions are satisfied: P n >P c ; When the fault occurs outside the zone: P n <P c .

[0129] S2. Calculate the line mode power mutation, line mode power high frequency energy and voltage mutation at the measuring point during the fault as fault characteristic quantities.

[0130] The fault characteristic quantities are selected as follows:

[0131] During system operation, electrical quantities will continue to fluctuate. When a line fails, voltages of both lines will change suddenly. The one with the larger amplitude of the sudden change will be selected as the starting criterion to improve the protection sensitivity.

[0132] 1) Fault direction identification feature

[0133] In order to facilitate analysis, the discrete line mode power mutation data points are integrated and summed, and the expression is:

[0134]

[0135] Where: N is the number of discrete points in the data window; They are the line mode power mutation values ​​at the measuring points T1 and T2 on both sides of the busbar in the T zone; is the integral value of the mutation amount.

[0136] When a fault occurs in the T zone, The polarity is opposite; when the DC line fails, the polarity is the same, and when the line on the left side of the T zone fails All are positive, and all are negative when the line on the right side of the T area fails. This feature can be used to form a fault direction identification feature.

[0137] 2) Fault identification characteristics inside and outside the area:

[0138] The sampling frequency of the present invention is set to 20kHz, and the voltage and current fault components in the range of 5 to 7.5kHz are extracted by using a filter to calculate the high-frequency component of the line mode power. Since wavelet transform, as a signal processing tool, can analyze signals at different scales and can better locate signal characteristics, wavelet transform is used to extract high-frequency energy to distinguish faults in and out of the area of ​​different lines. The fault transient current obtained at the measuring device is decomposed into 5 layers by db4 wavelet, and the first layer of wavelet coefficients corresponding to the high-frequency component are extracted. The wavelet energy of the high-frequency component of the line mode power at the line boundary measurement point is defined as:

[0139]

[0140] Where P a , P m , P n , P c They are the line mode high frequency powers at the measuring points on both sides of the boundaries of the two lines, and their magnitudes are equal to the product of the line mode current fault component and the line mode voltage fault component in the corresponding frequency band.

[0141] When a line fails, only the energy detected at the measurement point close to the fault fluctuates greatly, while the other fluctuations are small. Therefore, the high-frequency wavelet energy characteristics of the line mode power can be used to form the characteristic quantity for identifying faults inside and outside the area.

[0142] 3) Fault selection:

[0143] Since the system adopts a true bipolar operation structure, when a single-pole grounding fault occurs, the voltage amplitude of the fault pole line changes significantly compared to the non-fault pole line; when a bipolar grounding fault occurs, both the positive and negative pole voltages will fluctuate, and the amplitude changes are similar. Therefore, the magnitude of the voltage mutation amplitude can be used to achieve fault pole selection. For ease of analysis, the discrete voltage mutation data points are integrated and summed, and the expression is:

[0144]

[0145] Where: △U p (n), △U n (n) is the voltage mutation of positive and negative electrodes; U p , U n is the integral value of the voltage mutation.

[0146] When the positive pole of the DC line fails, the voltage mutation of the positive pole is greater than that of the negative pole; when the negative pole of the DC line fails, the voltage mutation of the negative pole is greater than that of the positive pole; when the DC line has a bipolar fault, the voltage amplitudes of the positive and negative poles change greatly, and the voltage changes of bipolar faults are roughly the same. Therefore, the voltage mutation characteristics can be used to form the fault pole selection characteristics.

[0147] S3. Build a BiTCN fault recognition model, train it offline, and verify it online.

[0148] The BiTCN network model is constructed as follows:

[0149] The core of TCN is the convolution operation, which processes time series data through convolution layers. Compared with CNN, the size of the convolution kernel of TCN is usually variable, and the size of the receptive field is increased by stacking multiple convolution layers.

[0150] Assume that the convolution kernel is K = (k1, k2, ..., k m ), the input time series is α=(h1,h2,...,h n ), then α n The causal convolution of is represented as:

[0151]

[0152] Where: p n-m+1 Represents input data.

[0153] Using dilated convolutions to avoid the vanishing gradient problem can increase the receptive field without increasing the number of parameters. A key feature of dilated convolutions is that they can capture information of different scales at each level, which helps capture long-range dependencies in temporal data. By reducing the number of convolutional neurons between two adjacent layers by the value of the dilation coefficient l, α n The dilated convolution can be expressed as:

[0154]

[0155] Where: d represents the expansion factor, which grows exponentially with 2.

[0156] In order to strengthen network training and improve model performance, residual connections are usually used in TCN. The output of each convolutional layer is added to its input to form a residual block, which makes the network easier to optimize. Part of the residual block is composed of dilated causal convolution and nonlinear mapping, which can be expressed as:

[0157] K(α i )=ω2(Relu(ω1α i )) (12)

[0158] Where: ω1 and ω2 represent the weights of the convolutional layer; Relu represents the activation function, which activates the output of the residual module.

[0159] The other part is a one-dimensional full convolution operation, which makes the input and output sequences have the same length. The output can be expressed as:

[0160] P(αi )=K(α i )+H i (13)

[0161] P i+1 =Relu(P(α i )) (14)

[0162] Where: P i+1 Represents the full convolution output result.

[0163] However, the traditional TCN model only performs forward convolution calculation on the input sequence, ignoring the implicit information in the backward direction. In order to effectively extract local and global features, the present invention adopts a BiTCN network model with strong feature capture capability, which can mine deep fault characteristics from complex time series samples and is suitable for processing information processing problems that require consideration of multiple factors, conditions and fuzzy information. The structure is as follows: Figure 2 shown.

[0164] Model training:

[0165] Different fault characteristics are used as input of BiTCN The output results are various fault types. Referring to the actual projects such as Wudongde, in order to facilitate the comparison of feature changes before and after the fault and the need for fault data analysis, the data of 2ms before the fault and 3ms after the fault are extracted. The acquisition of the data set comprehensively considers the influencing factors such as the fault distance (collecting 1%, 25%, 50%, and 75% of the line respectively), the transition resistance (the interval is 0.01~600Ω, and the change step is 10Ω), etc., to ensure that there are sufficient samples to train the BiTCN network model offline.

[0166] A BiTCN network model was constructed on the MATLAB platform. The normalized data was divided into training set and validation set in a ratio of 4:1. Parameters such as iterative learning and number of iterations were adjusted. After the model was adjusted, the fault diagnosis accuracy could achieve accurate output of 10 different types of faults.

[0167] like Figure 3 As shown in the figure, the process of constructing a three-terminal hybrid DC transmission protection based on BiTCN is as follows:

[0168] S4101. Build a three-terminal hybrid DC transmission model;

[0169] S4102, collecting fault voltage and current signals at the measuring point;

[0170] S4103: Determine whether the start criterion is met based on the obtained signal Considering a certain margin, set If satisfied, proceed to step S4104);

[0171] S4104, calculating fault characteristic quantities: line mode power mutation quantity, line mode high frequency wavelet energy, line mode voltage mutation quantity;

[0172] S4105. Build a BiTCN network model and perform offline training on it;

[0173] S4106, taking the fault feature quantity as the input quantity of BiTCN, and taking the fault type as the output quantity;

[0174] S4107, when the fault occurs, data is input and BiTCN outputs the corresponding fault type;

[0175] S4108, corresponding protection action.

[0176] The present invention takes the ±800kV Wudongde three-terminal DC transmission project as a reference object, and the model parameters are shown in Table 1.

[0177] Table 1 Parameters

[0178]

[0179] In order to verify the reliability of the method proposed in the present invention, metallic faults are set at 30% and 80% of the distance from the line I and II protection installation points and at the fault outside the zone, and the transition resistance is set to 200Ω. The protection discrimination results at different fault locations are shown in Table 2.

[0180] Table 2 Protection results at different fault locations

[0181]

[0182] It can be seen from Table 2 that the algorithm proposed in the present invention can correctly identify the faults in the text, and the protection can act correctly with a high accuracy.

[0183] In general, the higher the sampling frequency, the higher the requirements for the protection device. In order to study the influence of different sampling frequencies on the method proposed in the present invention, taking the positive protection as an example, the T zone fault and the inner and outer faults of line I are set, and the sampling frequencies are set to 5 and 10 kHz, respectively. The simulation results are shown in Table 3.

[0184] Table 3 Simulation results under the influence of sampling frequency

[0185]

[0186] It can be seen from Table 3 that under different sampling frequencies, the method proposed in the present invention can accurately identify the fault type and provide reliable protection. Therefore, the method proposed in the present invention is not affected by the sampling frequency.

[0187] The simulation results of the method proposed in the present invention when it fails under different interference conditions are as follows: Figures 4 to 6 shown.

[0188] Figure 4 is the fault simulation result diagram under different transition resistances; Figure 4 It can be seen that: with the increase of transition resistance, the line mode power wavelet energy will decrease, but in the range of transition resistance reaching 600Ω, the results obtained meet the proposed protection criteria, and the BiTCN model can correctly identify the fault type and make the corresponding protection action, meeting the sensitivity of protection.

[0189] Figure 5 is the fault simulation result diagram under weak boundary conditions; Figure 5 It can be seen that the high-frequency components of the power obtained at the detection point during internal and external faults can meet the protection criteria. Therefore, in a weak boundary environment, the BiTCN model can correctly detect the fault type and the protection can operate reliably.

[0190] Figure 6 is the fault simulation result diagram under AC line fault; Figure 6 It can be seen that according to the division of the boundary, the AC side fault also belongs to the outside of the zone. When the transition resistance is set to 0.01Ω, the three-phase grounding fault of the AC line, the voltage fluctuation measured when the AC line fails is small, does not meet the start criterion, and the protection will not malfunction.

[0191] The present invention identifies the input characteristic quantity of the BiTCN network model by analyzing the line mode power characteristics of the three-terminal hybrid DC transmission line when it fails, which can avoid the difficulty of manual threshold setting. It can also more easily associate the output of the model with the input sequence, has higher calculation efficiency, and improves the protection speed.

Claims

1. A three-terminal hybrid DC transmission line protection method based on line mode power characteristics, characterized in that The following steps are involved: Step 1: Build a three-terminal hybrid DC transmission model, analyze the fault characteristics of different fault areas, and collect DC line mode voltage and line mode power signals; Step 2: Calculate the line mode power mutation amount, line mode power high frequency energy, and voltage mutation amount at the measurement point during the fault as fault characteristic quantities; Step 3: Build a BiTCN network fault identification model, train it offline, and verify it online; Step 4: Take the fault feature quantity as the input quantity of the BiTCN network fault identification model, and the fault type as the output quantity, and construct the line protection criterion based on the BiTCN network fault identification model.

2. The three-terminal hybrid DC transmission line protection method based on line mode power characteristics according to claim 1 is characterized in that: In step 1, a three-terminal hybrid DC transmission model is built, and the fault characteristics of different areas are analyzed as follows: 1) When a fault occurs in the T zone and the lines on both sides, the line mode power mutation is as follows: Where: They are the line mode current mutations of lines T1 and T2 on both sides of the T zone, and the positive direction is the direction during normal operation; when a fault occurs in the T zone, Opposite polarity, Same polarity; When the left side of the T zone fails, and The polarity is the same; When the right side of the T zone fails, and The polarities are opposite; Therefore: When the T zone fails: When the left side of the T zone fails: When the right side of the T zone fails: 2) The fault analysis inside and outside the line area I is as follows: When the frequency is greater than 2000Hz, the high-frequency component of the DC power of line I will be blocked by the boundary elements and thus attenuated; when the current flows through the converter equivalent impedance Z LCC When the voltage and current frequency at both ends of Z are high, LCC Contains only the reactance component, i.e. Z LCC =jωL LCC , the DC filter can be equivalent to a fixed inductor L in the high frequency band filter Therefore, there is a relationship between the current and voltage at both ends of the rectifier side boundary: In formula (2): Z filter is the filter equivalent impedance; is the equivalent impedance of the current limiting reactor; Z LCC is the equivalent impedance of the LCC converter; Measure the voltage in zone I of the line; Measure voltage outside zone I of the line; To measure the current in zone I of the line; Measure current outside zone I of the line; It can be seen from formula (2) that when a fault occurs in the line I, the high-frequency fault components of the current and voltage on both sides of the boundary satisfy: That is, P m >P a , P m is the external line mode power in line zone I; P a is the line mode power in zone I of the line; The relationship between the current and voltage components at the line boundary measurement point during an out-of-zone fault is as follows: Where: Z MMC1 is the equivalent impedance of the MMC converter; In the high frequency band, the equivalent inductance of the DC filter is much smaller than the inductance of the smoothing reactor. It can be seen that: That is, P m <P a ; 3) The fault analysis inside and outside the line zone II is as follows: In the converter station of line II, a section of overhead line-to-ground capacitance and a current-limiting reactor are used to form the line boundary. When the frequency is greater than 5000Hz, the high-frequency component of the DC power of line II will also be blocked by the boundary, resulting in attenuation. There is a relationship between the current and voltage fault components at both ends of the inverter side boundary: In formula (4): Measure the voltage in zone II of the line; Measure voltage outside zone II of the line; Measure current outside zone II of the line; To measure the current in zone II of the line; Therefore, when an internal fault occurs in line II, the boundary fault current and voltage satisfy: When an out-of-zone fault occurs on line II, the current and voltage meet the following conditions: Therefore, when the fault occurs in the line zone II, the following conditions are satisfied: P n >P c ; When the fault occurs outside the zone: P n <P c , P n is the line mode power in zone II of the line; P c The line mode power outside the line zone II; DC line line mode voltage, line mode power signal: In order to avoid the electromagnetic coupling relationship between the positive and negative poles of the DC line, the phase mode transformation is used to convert each electrical quantity into line mode and ground mode components; Where U p and U d are the measured positive and negative fault voltages respectively; U0 and U1 are the ground mode and line mode components of the fault voltage respectively; Define line mode power as: P1=I1*U1 (6); Where I1 and U1 are the line mode components of the fault current and voltage respectively.

3. The three-terminal hybrid DC transmission line protection method based on line mode power characteristics according to claim 1 is characterized in that: In step 2, the fault characteristic quantity is selected as follows: (1) Fault direction identification feature: In order to facilitate analysis, the discrete line mode power mutation data points are integrated and summed, and the expression is: Where: N is the number of discrete points in the data window; They are the line mode power mutation values ​​at the measuring points T1 and T2 on both sides of the busbar in the T zone; is the integral value of the mutation amount; When a fault occurs in the T zone, The polarity is opposite; when the DC line fails, the polarity is the same, and when the line on the left side of the T zone fails All are positive poles, and all are negative poles when the line on the right side of the T zone fails. The polarity difference of the line mode power mutation when the fault direction is different constitutes the fault direction identification feature; (2) Fault identification characteristics inside and outside the zone: The high-frequency components of line mode power are extracted by wavelet transform to distinguish the faults inside and outside the area of ​​different lines. The transient fault current obtained at the measurement point is decomposed by wavelet to extract the first-layer wavelet coefficients corresponding to the high-frequency components of line mode power. The wavelet energy of the high-frequency components of line mode power at the line boundary measurement point is defined as: In the formula, E i Represents the wavelet energy of the high-frequency component of the line mode power; P i (n) represents the line mode power of the measurement point; a represents the measurement point outside the line zone I; m represents the measurement point inside the line zone I; n represents the measurement point inside the line zone II; c represents the measurement point outside the line zone II; N represents the number of wavelet coefficients; When a line fails, only the energy detected at the measurement point close to the fault fluctuates greatly, while the other fluctuations are small. Therefore, the high-frequency wavelet energy characteristics of the line mode power are used to form the fault identification feature quantity inside and outside the area. (3) Fault selection: When a single-pole ground fault occurs, the voltage amplitude of the fault pole line changes significantly compared to the non-fault pole line; When a bipolar ground fault occurs, both the positive and negative voltages will fluctuate, with similar amplitude changes; Therefore, the magnitude of the voltage mutation is used to realize fault pole selection; the discrete voltage mutation data points are integrated and summed, and the expression is: Where: △U p (n), △U d (n) are the voltage mutation of positive and negative electrodes respectively; U p , U d is the integral value of voltage mutation; When the positive pole of the DC line fails, the voltage mutation of the positive pole is greater than that of the negative pole; when the negative pole of the DC line fails, the voltage mutation of the negative pole is greater than that of the positive pole; when the DC line has a bipolar fault, the voltage amplitudes of the positive and negative poles change greatly, and the voltage changes of bipolar faults are roughly the same. Therefore, the voltage mutation characteristics are used to form the fault pole selection characteristics.

4. The three-terminal hybrid DC transmission line protection method based on line mode power characteristics according to claim 1 is characterized in that: In step 3, the BiTCN network fault identification model is constructed as follows: The core of TCN is the convolution operation. Let the convolution kernel be K = (k1, k2, ..., k m ), k1, k2, ..., k m The specific weight values ​​in the convolution kernel respectively; m is the size of the convolution kernel; The input time series is α=(h1,h2,...,h n ), h1,h2,...,h n Respectively represent the value of each time point; n is the sequence size; Then α n The causal convolution of is represented as: Where: p n-m+1 represents input data; k i Represent the weight of each convolution kernel respectively; Use dilated convolution to avoid the gradient vanishing problem. Reduce the number of convolution neurons between two adjacent layers by the dilation coefficient l. Then α n The dilated convolution is expressed as: Where: d represents the expansion factor; p n-(m-1)d Represents input data; Residual connections are used in TCN; the output of each convolutional layer is added to its input to form a residual block, which makes the network easier to optimize; part of the residual block is composed of dilated causal convolution and nonlinear mapping, expressed as: K(a i )=ω2(Relu(ω1a i )) (12); Where: ω1, ω2 represent the weights of the convolutional layer; Relu represents the activation function, which activates the output of the residual module; Relu(ω1α i ) represents the activation function of each convolutional layer; The other part is a one-dimensional full convolution operation, which makes the input and output sequences have the same length. The output can be expressed as P(a i )=K(a i )+H i (13); P i+1 =Relu(P(α i )) (14): Where: P(α i ) represents the convolution output; K(α i ) represents the residual block; H i represents the input sequence; P i+1 Represents the full convolution output result; Relu(P(α i )) represents the activation function of the convolution output.

5. The three-terminal hybrid DC transmission line protection method based on line mode power characteristics according to claim 4 is characterized in that: Bi The TCN network model includes input fault feature matrix, bidirectional sliding window processing, and two-layer TCN model; BiTCN network model training: Different fault characteristics are used as input of the BiTCN network model They represent the line mode power mutation on both sides of the busbar in the T zone; E m ,E a ,E n ,E c Respectively represent the wavelet energy of the high-frequency components of the line mode power on both sides of the line Ⅰ and Ⅱ boundaries; U p ,U d Respectively represent the integral value of the voltage mutation of the positive and negative poles of the line.

6. The three-terminal hybrid DC transmission line protection method based on line mode power characteristics according to claim 5 is characterized in that: The output results are various fault types, including positive pole fault in line zone I, negative pole fault in line zone I, bipolar fault in line zone I, fault outside line zone I, positive pole fault in line zone II, negative pole fault in line zone II, bipolar fault in line zone II, fault outside line zone II, positive pole fault in T zone, and negative pole fault in T zone.

7. The three-terminal hybrid DC transmission line protection method based on line mode power characteristics according to claim 1 is characterized in that: In step 4, the three-terminal hybrid DC transmission protection process is as follows: S4101. Build a three-terminal hybrid DC transmission model; S4102, collecting fault voltage and current signals at the measuring point; S4103: Determine whether the start criterion is met based on the obtained signal △U Ⅰ , △U Ⅱ Indicates the line mode voltage mutation of the two lines; Indicates the start criteria set; max(△U Ⅰ ,△U Ⅱ ) represents the maximum value of the two; considering a certain margin, set If satisfied, proceed to S4104; S4104, calculating fault characteristic quantities: line mode power mutation quantity, line mode high frequency wavelet energy, line mode voltage mutation quantity; S4105. Build a BiTCN network model and perform offline training on it; S4106, taking the fault feature quantity as the input quantity of the BiTCN network model, and taking the fault type as the output quantity; S4107, when the fault occurs, data is input and the BiTCN network model outputs the corresponding fault type; S4108, corresponding protection action.

8. The three-terminal hybrid DC transmission line protection method based on line mode power characteristics according to claim 7 is characterized in that: Taking the voltage mutation as the starting criterion, when a DC fault occurs, the voltage change rate at both ends of the line is significantly greater than the normal fluctuation range; when the protection criterion is activated, when a fault occurs in the T zone and the DC lines on both sides, the line mode power mutation amount has obvious characteristics, which constitutes the fault direction identification criterion; When there is a fault inside or outside the line area, the fault energy difference between the corresponding measuring points inside and outside the line area is obvious, which constitutes the identification of the fault inside and outside the line area; When faults occur at different poles of the line, the voltage mutation characteristics are obvious, which constitutes the fault pole selection criterion.

Images