A commutation failure prediction method based on a dual-model adaptive fusion algorithm

Abstract

The present invention relates to a commutation failure prediction method based on a dual-model adaptive fusion algorithm, belonging to the technical field of high-voltage direct current transmission operation analysis, including: determining whether a fault occurs and the fault type; S2: extracting U based on the Park transformation model m1 and φ1; extracting U based on the commutation voltage fitting model m2 and φ2, and the adaptive fusion module performs weighting to obtain the final commutation voltage amplitude U m and the final offset angle φ; calculating the area S provided by the system during the commutation process based on U m and φ pro ; calculating the commutation demand area S of the system need , and comparing it with the area S provided by the system pro . When S need > S pro , the system will experience commutation failure; otherwise, the system will not experience commutation failure. The present invention can accurately and quickly predict commutation failure, improve the extraction accuracy of the system commutation voltage during asymmetric faults, and improve the influence of harmonics on the extraction of the system commutation voltage.

Description

Technical Field

[0001] The present invention relates to a commutation failure prediction method based on a dual-model adaptive fusion algorithm, belonging to the technical field of high-voltage direct current (HVDC) transmission operation analysis. Background Art

[0002] Line-commutated converter based high-voltage direct current (LCC-HVDC) transmission has become a widely used solution in inter-regional power transmission systems and back-to-back asynchronous interconnections. Due to low power losses, low capital costs, and fast emergency response capabilities, LCC-HVDC has better performance in large HVDC networks.

[0003] Thyristors are the core devices for realizing AC / DC power conversion. Since thyristors do not have self-turn-off capabilities, there are some inherent drawbacks. Commutation failure (CF) is one of them, and commutation failure occurs due to a decrease in the voltage amplitude of the commutation bus voltage on the AC side. It will bring serious consequences to the HVDC system, such as voltage waveform distortion, sudden increase in DC current, and significant active and reactive power impacts. Commutation failure is a frequently occurring dynamic event; for example, in 2019, a three-phase short-circuit fault in Suizeng caused commutation failure in 9 circuits, and the DC power loss accounted for about 90% of the total DC capacity. In addition, if timely and effective control measures cannot be taken, CF will lead to AC grid frequency oscillations and DC blocking.

[0004] Line-commutated converter based high-voltage direct current (LCC-HVDC) technology has been widely applied in the power system due to its advantages in power loss and investment cost. It has currently been applied to the fields of inter-regional, long-distance, and large-capacity power transmission, providing a good solution to the problem of reverse distribution of energy and load centers in China. However, since the thyristors used in LCC-HVDC do not have self-turn-off capabilities, there are some inherent defects. Commutation failure is one of the most common phenomena in LCC-HVDC systems, which can lead to a decrease in DC voltage and a sharp increase in DC current. If effective control measures are not taken in time, it may lead to subsequent commutation failures and even DC blocking, threatening the safe and stable operation of the power system.

[0005] Accurately and quickly predicting the occurrence of commutation failure is the basis for solving the commutation failure problem in HVDC systems. Therefore, proposing an accurate and fast commutation failure prediction method is of great significance for providing sufficient time margin for subsequent protection control and reducing the impact of commutation failure on the power system. However, the existing methods for predicting commutation failure have the following defects: slow prediction speed for the first commutation failure in the face of asymmetric faults, poor accuracy in extracting commutation voltage characteristic quantities after asymmetric faults, and easy occurrence of missed or false judgments in commutation failure prediction. Summary of the Invention

[0006] In view of the above problems, the present invention provides a commutation failure prediction method based on a dual-model adaptive fusion algorithm, which can accurately and quickly predict commutation failures, improve the extraction accuracy of the commutation voltage of the system under asymmetric faults, and mitigate the impact of harmonics on the extraction of the commutation voltage of the system.

[0007] The commutation voltage amplitude extracted by the dual-model adaptive fusion algorithm of the present invention is: U m = U m1 * f 1(ωt) + U m2 * f 2(ωt) , and the extracted commutation voltage offset angle is: Among them, U m1 , are the commutation voltage amplitude and offset angle extracted by the Park transformation model; U m2 , are the commutation voltage amplitude and offset angle extracted by the fitting model; f 1(ωt) , f 2(ωt) are the respective adaptive factors of the two models. Then calculate the system commutation demand area commutation process system-provided area Among them, X r is the equivalent commutation reactance; γ min is the minimum turn-off angle; β is the lead trigger angle; t0 is the commutation start time, and t1 is the commutation end time. If the system commutation demand area is greater than the commutation process system-provided area, it is predicted that the system will experience commutation failure; otherwise, it is predicted that the system will not experience commutation failure.

[0008] The technical solution of the present invention is as follows:

[0009] A commutation failure prediction method based on a dual-model adaptive fusion algorithm, comprising the following steps:

[0010] S1: Determine whether a fault has occurred and the type of fault;

[0011] S2: Extract the commutation voltage amplitude U m1 and the offset angle

[0012] based on the Park transformation model m2 and the offset angle and input the obtained commutation voltage amplitudes U m1 , U m2 and the offset angle into the adaptive fusion module to obtain the final commutation voltage amplitude U m and the final offset angle

[0013] S4: Based on Um and The commutation process calculation system provides an area S pro ;

[0014] S5: Calculate the commutation demand area S of the system need , and compare it with the area S provided by the system pro for comparison. When S need > S pro , the system will experience commutation failure; otherwise, the system will not experience commutation failure.

[0015] First, collect the commutation voltage on the inverter side of the system, and detect whether the system has a fault and whether the fault is an asymmetric fault by superposition and transformation. If the system has an asymmetric fault, extract the commutation voltage amplitude and offset angle; if a symmetric fault occurs, only extract the commutation voltage amplitude. Secondly, use the extracted commutation voltage related characteristic quantities to calculate the commutation demand area of the system and the area provided by the system during the commutation process. Finally, accurate and rapid prediction of commutation failure can be achieved by comparing the magnitudes of the two. Different from traditional methods, the present invention can distinguish fault types, improve the extraction accuracy of the system commutation voltage when an asymmetric fault occurs, and at the same time improve the influence of harmonics on the system voltage extraction, improve the commutation failure prediction accuracy, reduce the misjudgment probability of the system commutation failure prediction, and there is also a certain improvement in the voltage extraction speed. Extracting the commutation voltage after a fault faster can predict whether commutation failure occurs earlier, enabling the DC control system to have more time to control and adjust the trigger angle size to maintain the safe and economic operation of the system.

[0016] Preferably, in step S1, a fault detection module is used to detect the occurrence and type of the fault;

[0017] The fault detection module includes zero-sequence detection and Clark transformation-based detection, which are respectively used to detect whether an asymmetric fault and a symmetric fault occur; it can accurately detect the occurrence of the fault, distinguish between symmetric and asymmetric faults, and ensure the rapidity of the action.

[0018] When the system is operating normally, there is no zero-sequence component in the three-phase voltage; when an asymmetric fault occurs in the system, a zero-sequence voltage component will appear in the three-phase voltage of the inverter side bus. The three-phase voltage zero-sequence component U0 is:

[0019] U0 = U a + U b + U c (1)

[0020] where Ua is the instantaneous voltage of phase a; Ub is the instantaneous voltage of phase b; Uc is the instantaneous voltage of phase c;

[0021] U0 is used as an output signal, and its value will be compared with a preset value. If this value is greater than the preset value, a single-phase fault is considered to have occurred; otherwise, it is considered that an asymmetric fault has not occurred. Under normal conditions, the zero-sequence component of the three-phase voltage is 0, that is, U0 = 0; when an asymmetric fault occurs, a zero-sequence component will appear in the three-phase voltage, that is, U0 > 0;

[0022] Three-phase faults are detected based on the Clark transformation. The Clark transformation is a prior art. It is based on the sinusoidal characteristics of the three-phase voltage and uses a rotating vector coordinate to describe the instantaneous voltage drop value. The vector value rotates in the αβ plane at an angular velocity ω. When detecting based on the Clark transformation, first calculate the transformed vector value |U αβ |:

[0023]

[0024] Wherein,

[0025]

[0026] In the formula, |U αβ | is the transformed vector value, and U α 、U β are the vectors obtained by mapping the transformed vector value to the α and β axes respectively;

[0027] When the system is operating normally, |U αβ | will be stable within a certain range. When a three-phase fault occurs in the system, this value will show a certain waveguide and decrease accordingly.

[0028] Subtract the value of |U αβ | at the current sampling moment from its value at the previous sampling moment. When the difference is less than 0.01, the system is in a normal working state; when the difference is greater than or equal to 0.01, it is determined that a symmetrical fault has occurred.

[0029] Preferably, in step S2, the principle of the Park transformation model is as follows: The three-phase commutation voltages are extracted in real time in the system. After the rotation transformation, virtual three-phase voltages are constructed for each phase. After the Park transformation, the output voltages U xd 、U xq mapped on the virtual d and q axes are calculated by formulas (5) and (7) respectively to calculate the commutation voltage amplitude U m1 and the offset angle

[0030]

[0031]

[0032] Wherein, j represents an imaginary number; ω represents the phase-locked loop frequency; t represents a time variable.

[0033] The Park transformation model is an existing model, which can overcome the problem that the traditional method cannot correctly reflect the change of the commutation voltage of the system under asymmetric fault conditions, and reduces the misjudgment probability of commutation failure prediction.

[0034] Preferably, in step S3, the principle of commutation voltage fitting is to obtain the amplitude, frequency and phase angle of the sine curve by sampling the sine curve. Assuming that the frequency of the AC system remains unchanged before and after the fault, the magnitude and phase angle of the commutation voltage can be obtained through two or more sampling points. The commutation voltage u extracted by the fitting model after the fault is:

[0035]

[0036] where k1 and k2 are parameters representing sinusoidal quantities;

[0037] Assume that the sampling points are (t c1 , u c1 ), (t c2 , u c2 )…(t cn , u cn ), where t c1 , t c2 , …, t cn are the times at different sampling moments respectively, and u c1 , u c2 , …, u cn are the commutation voltage amplitudes at different sampling moments respectively. It can be obtained that:

[0038]

[0039] k1 and k2 can be obtained through matrix inverse operation respectively:

[0040]

[0041] The commutation voltage amplitude U m2 and the offset angle obtained by fitting are:

[0042]

[0043] where Un represents the commutation voltage amplitude when the system is operating normally.

[0044] Preferably, in step S3, the adaptive fusion module is composed of a Park transformation model and a commutation voltage fitting model. In the Park transformation model, the three-phase commutation voltages are extracted in real time. After the rotation transformation, virtual three-phase voltages are constructed for each phase respectively. After the Park transformation, the commutation voltage amplitude U m1 and the offset angle In the commutation voltage fitting model, the commutation voltage amplitude U is obtained by fitting two points after delaying the commutation voltage on the inverter side m2 and the offset angle

[0045] The adaptive fusion module assigns respective adaptive factors f 1(ωt) 、f 2(ωt) to the Park transformation model and the commutation voltage fitting model according to the system natural frequency ωt, and weights them to obtain the final commutation voltage amplitude U m and the offset angle as follows:

[0046] U m =U m1 *f 1(ωt) +U m2 *f 2(ωt) (13)

[0047]

[0048] where the adaptive factors f 1(ωt) 、f 2(ωt) are shown as follows. The values of the adaptive factors are related to the system frequency and take the form of an exponential function:

[0049]

[0050] f 2(ωt) =1-f 1(ωt) (16)

[0051] where C1 - C4 represent different exponential function parameters. After repeated simulation verification, when C1 = 14, C2 = 3.4, C3 = 2, and C4 = -3.3, the proposed dual-model adaptive fusion extraction algorithm for commutation voltage has the best extraction effect on the commutation voltage.

[0052] Preferably, in step S4, the area S pro provided by the system during the commutation process is calculated, and the formula is as follows:

[0053]

[0054] where γ min is the minimum turn-off angle; β is the lead trigger angle, and U m 、 are the commutation voltage amplitude and the offset angle finally extracted in step 2, respectively.

[0055] Preferably, in step S5, the area S need required for system commutation is calculated, and the formula is as follows:

[0056]

[0057] Among them, X r is the equivalent commutation reactance; t0 is the commutation start time; t1 is the commutation end time; represents the DC side current at the commutation start time; represents the DC side current at the commutation end time.

[0058] For the parts not detailed in the present invention, reference can be made to the prior art.

[0059] The beneficial effects of the present invention are as follows:

[0060] Through the analysis of the commutation process in the LCC-HVDC system, there are a series of problems in the extraction of commutation voltage at present, such as low extraction accuracy, slow speed, and poor voltage extraction effect in the face of asymmetric faults. Therefore, the present invention proposes a commutation failure prediction method based on a dual-model adaptive fusion algorithm. Compared with the traditional method, the present invention can distinguish the fault type, improve the extraction accuracy of the commutation voltage of the system when an asymmetric fault occurs, and at the same time improve the influence of harmonics on the system voltage extraction, improve the commutation failure prediction accuracy, reduce the misjudgment probability of the system commutation failure prediction, and also have a certain improvement in the voltage extraction speed, which has greater engineering significance. It lays a foundation for the subsequent timely action and adjustment of the control system, and has important significance for reducing the impact of commutation failure on the power system. Description of the Drawings

[0061] The specification drawings forming a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application.

[0062] Figure 1 is the topological structure diagram of a 6-pulse Graetz bridge inverter;

[0063] Figure 2 is the wiring diagram of the standard test model;

[0064] Figure 3 are the extraction effect diagrams of the commutation voltage amplitude and the extraction effect diagrams of the commutation voltage waveform during symmetric faults, where (a) is the extraction effect diagram of the commutation voltage amplitude; (b) is the extraction effect diagram of the commutation voltage waveform;

[0065] Figure 4 are the extraction effect diagrams of the commutation voltage amplitude, the offset angle, and the extraction effect diagrams of the commutation voltage waveform during asymmetric faults; where (a) is the extraction effect diagram of the commutation voltage amplitude, (b) is the extraction effect diagram of the offset angle, and (c) is the extraction effect diagram of the commutation voltage waveform;

[0066] Figure 5For the commutation voltage and DC current waveforms when a symmetrical fault occurs on the inverter side and the grounding impedance is 0.18H; where (a) is the commutation voltage waveform and (b) is the DC current waveform;

[0067] Figure 6 For the first prediction case of the proposed method;

[0068] Figure 7 For the commutation voltage and DC current waveforms when a symmetrical fault occurs on the inverter side and the grounding impedance is 0.22H, where (a) is the commutation voltage waveform, (b) is the first DC current waveform, and (c) is the second DC current waveform;

[0069] Figure 8 For the second prediction case of the proposed method;

[0070] Figure 9 For the commutation voltage and DC current waveforms when an asymmetrical fault occurs on the inverter side and the grounding impedance is 0.125H, where (a) is the commutation voltage waveform and (b) is the DC current waveform;

[0071] Figure 10 For the third prediction case of the proposed method;

[0072] Figure 11 For the commutation voltage and DC current waveforms when an asymmetrical fault occurs on the inverter side and the grounding impedance is 0.175H, where (a) is the commutation voltage waveform, (b) is the first DC current waveform, and (c) is the second DC current waveform;

[0073] Figure 12 For the fourth prediction case of the proposed method.

[0074] Figure 13 For the fault type detection control block diagram;

[0075] Figure 14 For the commutation voltage dual - model adaptive fusion extraction algorithm control block diagram;

[0076] Figure 15 For the flowchart of the commutation failure prediction method based on the dual - model adaptive fusion algorithm. Specific implementation manner

[0077] In order to enable those skilled in the art to better understand the technical solutions in this specification, the following describes the technical solutions in the embodiments of the present invention clearly and completely in conjunction with the accompanying drawings in the embodiments of this specification, but not limited thereto. For those parts not elaborated in the present invention, the conventional techniques in the art are adopted.

[0078] Figure 1The topological structure diagram of a 6-pulse Graetz bridge inverter is shown in FIG. The following calculation methods are all based on this topological structure. The present invention can be applied to a 6-pulse or 12-pulse Graetz bridge inverter. The 12-pulse inverter structure is composed of two 6-pulse converters connected in series on the DC side.

[0079] V1~V6 are the 1st to 6th bridge arms; I d is the DC current; U d is the DC side voltage; the 6-pulse Graetz bridge inverter consists of six thyristor bridge arms, which are turned on in the order of 1-6. In normal operation, in a 60° repetitive cycle, 2 bridge arms and 3 bridge arms are turned on in turn. Before V1 and V3 are switched, V1 and V2 are turned on, and the DC side is connected to the two-phase AC power supply A and C. The port voltage is u ac , DC current i d Flows through V1 and V2. When the V3 trigger pulse arrives, if the voltage across V3 u ba is positive, it is turned on immediately and V1 starts to commutate to V3. Figure 1 Medium d Indicates the equivalent impedance of the AC measurement system; L r Represents the equivalent impedance of the DC line; i a 、i b 、i c They represent the current flowing through the three phases a, b, and c respectively.

[0080] Example 1

[0081] A commutation failure prediction method based on dual-model adaptive fusion algorithm, such as Figure 15 As shown, the following steps are included:

[0082] S1: Determine whether a fault occurs and the type of fault;

[0083] S2: Extracting the commutation voltage amplitude U based on the Park transformation model m1 and offset angle

[0084] S3: Extract the commutation voltage amplitude U based on the commutation voltage fitting model m2 and offset angle The obtained commutation voltage amplitude U m1 , U m2 and offset angle Input to the adaptive fusion module to obtain the final commutation voltage amplitude U m and the final offset angle

[0085] S4: Based on U m and Calculate the area S provided by the commutation process system pro ;

[0086] S5: Calculate the required commutation area S of the computing system need , and compare it with the provided area S of the system pro . When S need > S pro , the system will experience commutation failure; otherwise, the system will not experience commutation failure.

[0087] Different from the traditional method, the present invention can distinguish the types of faults, improve the accuracy of system commutation voltage extraction when asymmetric faults occur, and at the same time improve the influence of harmonics on system voltage extraction, improve the accuracy of commutation failure prediction, reduce the misjudgment probability of system commutation failure prediction, and also have a certain improvement in the voltage extraction speed.

[0088] Embodiment 2

[0089] A commutation failure prediction method based on a dual-model adaptive fusion algorithm, as described in Embodiment 1, except that in step S1, a fault detection module is used to detect the occurrence and type of faults, such as Figure 1 ;

[0090] The fault detection module includes zero-sequence detection and Clark-transform-based detection, which are respectively used to detect whether asymmetric faults and symmetric faults occur; it can accurately detect the occurrence of faults, distinguish symmetric faults and asymmetric faults, and ensure the rapidity of actions.

[0091] When the system is operating normally, there is no zero-sequence component in the three-phase voltages; when an asymmetric fault occurs in the system, a zero-sequence voltage component will appear in the three-phase voltages of the inverter-side bus. The zero-sequence component U0 of the three-phase voltages is:

[0092] U0 = U a + U b + U c (1)

[0093] where Ua is the instantaneous voltage of phase a; Ub is the instantaneous voltage of phase b; Uc is the instantaneous voltage of phase c;

[0094] U0 is used as the output signal, and its value will be compared with a preset value. If this value is greater than the preset value, it is considered that a single-phase fault has occurred; otherwise, it is considered that no asymmetric fault has occurred. Under normal circumstances, the zero-sequence component of the three-phase voltages is 0, that is, U0 = 0; when an asymmetric fault occurs, a zero-sequence component will appear in the three-phase voltages, that is, U0 > 0;

[0095] To detect three-phase faults based on Clark transform, Clark transform is a prior art. It is based on the sine characteristics of three-phase voltages and uses a rotating vector coordinate to describe the instantaneous voltage drop value. The vector value rotates at an angular velocity ω in the αβ plane. When detecting based on Clark transform, first calculate the transformed vector value |Uαβ |:

[0096]

[0097] Wherein,

[0098]

[0099] In the formula, |U αβ | is the transformed vector value, and U α , U β are respectively the vectors obtained by mapping the transformed vector value onto the α and β axes;

[0100] When the system is operating normally, |U αβ | will be stable within a certain range. When a three-phase fault occurs in the system, this value will show a certain fluctuation and decrease accordingly.

[0101] Take the difference between the value of |U αβ | at the current sampling moment and the value at the previous sampling moment. When the difference is less than 0.01, the system is in a normal working state. When the difference is greater than or equal to 0.01, it is determined that a symmetrical fault has occurred.

[0102] Embodiment 3

[0103] A commutation failure prediction method based on a dual-model adaptive fusion algorithm, as described in Embodiment 2. The difference is that in step S2, the principle of the Park transformation model is as follows: in the system, three-phase commutation voltages are extracted in real time. After the rotation transformation, virtual three-phase voltages are constructed for each phase. After the Park transformation, the output voltages U xd , U xq mapped on the virtual d and q axes are obtained. The commutation voltage amplitude U m1 and the offset angle

[0104]

[0105]

[0106] are calculated through formulas (5) and (7) respectively. Here, j represents the imaginary number; ω represents the PLL frequency; t represents the time variable.

[0107] The Park transformation model is an existing model, which can overcome the problem that the traditional method cannot fully and correctly reflect the change of the system commutation voltage under asymmetric fault conditions, and reduces the misjudgment probability of commutation failure prediction.

[0108] Embodiment 4

[0109] A commutation failure prediction method based on a dual - model adaptive fusion algorithm, as described in Embodiment 3, except that in step S3, the principle of commutation voltage fitting is to obtain the amplitude, frequency, and phase angle of a sine curve by sampling the sine curve. Assuming that the frequency of the AC system remains unchanged before and after a fault, the magnitude and phase angle of the commutation voltage can be obtained through two or more sampling points. The commutation voltage u extracted by the fitting model after the fault is:

[0110]

[0111] where k1 and k2 are parameters representing sinusoidal quantities;

[0112] Assume that the sampling points are respectively (t c1 , u c1 ), (t c2 , u c2 ) … (t cn , u cn ), where t c1 , t c2 , …, t cn are the times at different sampling moments respectively, and u c1 , u c2 , …, u cn are the commutation voltage amplitudes at different sampling moments respectively. Then we can get:

[0113]

[0114] k1 and k2 can be obtained respectively through matrix inverse operation:

[0115]

[0116] The commutation voltage amplitude U m2 and the offset angle obtained by fitting are:

[0117]

[0118] where Un represents the commutation voltage amplitude during normal operation of the system.

[0119] The adaptive fusion module consists of a Park transformation model and a commutation voltage fitting model. In the Park transformation model, the three - phase commutation voltages are extracted in real - time. After the rotation transformation, virtual three - phase voltages are constructed for each phase respectively. After the Park transformation, the commutation voltage amplitude U m1 and the offset angle are calculated. While in the commutation voltage fitting model, the commutation voltage on the inverter side is extracted, and after two - point fitting with a time delay, the commutation voltage amplitude U m2 and the offset angle

[0120] The adaptive fusion module assigns respective adaptive factors f to the Park transformation model and the commutation voltage fitting model according to the system natural frequency ωt 1(ωt) 、f 2(ωt) ,and obtains the final commutation voltage amplitude U m and the offset angle as follows:

[0121] U m =U m1 *f 1(ωt) +U m2 *f 2(ωt) (13)

[0122]

[0123] where the adaptive factors f 1(ωt) 、f 2(ωt) are shown as follows. The values of the adaptive factors are related to the system frequency and take the form of an exponential function:

[0124]

[0125] f 2(ωt) =1-f 1(ωt) (16)

[0126] where C1 - C4 represent different exponential function parameters. After repeated simulation verification, when C1 = 14, C2 = 3.4, C3 = 2, and C4 = -3.3, the proposed dual-model adaptive fusion extraction algorithm for commutation voltage has the best extraction effect on the commutation voltage.

[0127] Example 5

[0128] A commutation failure prediction method based on a dual-model adaptive fusion algorithm, as described in Example 4, except that in step S4, the system-provided area S during the commutation process is calculated pro ,and the formula is as follows:

[0129]

[0130] where γ min is the minimum turn-off angle; β is the lead trigger angle, and U m 、 are respectively the commutation voltage amplitude and the offset angle finally extracted in step 2.

[0131] Example 6

[0132] A commutation failure prediction method based on a dual-model adaptive fusion algorithm, as described in Example 4, except that in step S5, the system commutation demand area S is calculated need ,and the formula is as follows:

[0133]

[0134] wherein, X r is the equivalent commutation reactance; t0 is the commutation start time; t1 is the commutation end time; represents the DC-side current at the commutation start time; represents the DC-side current at the commutation end time.

[0135] Method verification:

[0136] Based on the CIGRE BENCH MARK HVDC standard test model, a commutation failure identification module was built in MATLAB / Simulink. The rated voltage of the DC system is 500 kV and the rated current is 2 kA. The wiring diagram is as Figure 2 shown. A fault was set on the AC bus on the inverter side (i.e., Figure 2 the bus connected to the AC system on the inverter side in ) to verify the feasibility of the proposed commutation failure identification method.

[0137] A symmetrical fault was set. The fault occurrence time was 0.7 s, the fault duration was 0.1 s, and the grounding reactance was 0.1 H. As Figure 3 shown, after the symmetrical fault occurred in the system at 0.7 s, the commutation voltage amplitude dropped. The amplitude dropped from 1 pu to about 0.74 pu. The proposed method accurately extracted that the commutation voltage amplitude dropped to about 0.73 pu, which shows that the amplitude extracted by the method of the present invention has good consistency with the true amplitude. Moreover, the commutation voltage waveform extracted by the proposed method can correctly reflect the true change of the commutation voltage curve after the fault, and better predict whether commutation failure occurs.

[0138] An asymmetrical fault was set. The fault occurrence time was 0.7 s, the fault duration was 0.1 s, and the grounding reactance was 0.08 H. As Figure 4 shown, after the symmetrical fault occurred in the system at 0.7 s, the commutation voltage amplitude dropped. The amplitude dropped from 1 pu to about 0.85 pu. As Figure 4 (a), after commutation failure, the voltage further dropped to about 0.8 pu. It can be shown that the amplitude extracted by the method of the present invention has good consistency with the true amplitude. The proposed method can accurately and quickly extract the commutation voltage amplitude after the fault. Moreover, the commutation voltage waveform extracted by the proposed method can correctly reflect the true change of the commutation voltage curve after the fault, and better predict whether commutation failure occurs. At the same time, the extraction of the shift angle can more accurately describe the change of the commutation voltage after the fault, improving the accuracy of the subsequent prediction of the first commutation failure.

[0139] Set a symmetrical fault with the fault occurrence time at 0.7 s, the fault duration at 0.1 s, and the grounding reactance at 0.18 H. After the symmetrical fault occurs in the system at 0.7 s, the first commutation failure occurs at 0.7146 s, and the commutation voltage drops, as Figure 5 shown. After the fault occurs, the commutation demand area S need increases, and the area S pro provided by the system decreases. Eventually, the commutation area provided by the system is less than the demand area, and the successful commutation condition cannot be met, resulting in commutation failure. The proposed method predicts the occurrence of the first commutation failure 12 milliseconds in advance at 0.7027 s, as Figure 6 shown.

[0140] Set a symmetrical fault with the fault occurrence time at 0.7 s, the fault duration at 0.1 s, and the grounding reactance at 0.22 H. After a slight symmetrical fault occurs in the system at 0.7 s, the system does not experience commutation failure, Figure 7 as shown. Based on the method proposed in the present invention, after the fault, the commutation voltage drops and the DC current rises. The commutation area provided by the system decreases but is still greater than the commutation demand area. Therefore, it is successfully predicted that the system will not experience commutation failure, Figure 8 as shown.

[0141] Set an asymmetrical fault with the fault occurrence time at 0.7 s, the fault duration at 0.1 s, and the grounding reactance at 0.125 H. After the symmetrical fault occurs in the system at 0.7 s, the first commutation failure occurs at 0.7095 s, and the commutation voltage drops, as Figure 9 shown. After the fault occurs, the commutation demand area S need increases, and the area S pro provided by the system decreases. Eventually, the commutation area provided by the system is less than the demand area, and the successful commutation condition cannot be met, resulting in commutation failure. The proposed method predicts the occurrence of the first commutation failure 1.5 milliseconds in advance at 0.708 s, as Figure 10 shown.

[0142] Set an asymmetrical fault with the fault occurrence time at 0.7 s, the fault duration at 0.1 s, and the grounding reactance at 0.175 H. After the symmetrical fault occurs in the system at 0.7 s, the system does not experience commutation failure, as Figure 11 shown. The commutation area provided by the system decreases but is still greater than the commutation demand area. The proposed method successfully predicts that the system will not experience commutation failure, as Figure 12 shown.

[0143] In the Figures 3 to 12 appendix, the abscissa is time for all;

[0144] Figure 14Among them, the upper box represents the commutation voltage amplitude and offset angle extracted by the Park transformation model, the lower left represents the commutation voltage amplitude and offset angle extracted by the commutation voltage fitting model, and the lower right represents the commutation voltage amplitude and offset angle obtained by final weighting.

[0145] The above are the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A commutation failure prediction method based on a dual-model adaptive fusion algorithm, characterized in that: The steps include: S1: Determine whether a fault occurs and the type of fault; S2: Extracting the commutation voltage amplitude U based on the Park transformation model m1 and offset angle S3: Extract the commutation voltage amplitude U based on the commutation voltage fitting model m2 and offset angle The obtained commutation voltage amplitude U m1 , U m2 and offset angle Input to the adaptive fusion module to obtain the final commutation voltage amplitude U m and the final offset angle S4: Based on U m and Calculate the area S provided by the commutation process system pro ; S5: Calculate the system switching required area S need , and the system provides area S pro For comparison, when S need >S pro , then the system will experience commutation failure, otherwise, the system will not experience commutation failure; Step S1 detects the occurrence and type of a fault through a fault detection module; The fault detection module includes zero-sequence detection and Clark transformation-based detection, which are used to detect whether an asymmetric fault or a symmetric fault occurs, respectively; When the system operates normally, the three-phase voltage has no zero-sequence component; when an asymmetric fault occurs in the system, the zero-sequence voltage component will appear on the three-phase voltage of the inverter-side busbar, and the three-phase voltage zero-sequence component U0 is: U0=U a +U b +U c (1) Among them, Ua is the instantaneous voltage of phase a; Ub is the instantaneous voltage of phase b; Uc is the instantaneous voltage of phase c; Under normal circumstances, the zero-sequence component of the three-phase voltage is 0, that is, U0 = 0; when an asymmetric fault occurs, a zero-sequence component will appear in the three-phase voltage, that is, U0>0; When detecting based on Clark transform, first calculate the transformed vector value |U αβ |: in, In the formula, |U αβ | is the transformed vector value, U α , U β are the vectors on the α and β axes mapped to the transformed vector values ​​respectively; Will|U αβ The value at the current sampling time is subtracted from the value at the previous sampling time. When the difference is less than 0.01, the system is in normal working state. When the difference is greater than or equal to 0.01, it is judged that a symmetrical fault has occurred. In step S3, the adaptive fusion module is composed of a Park transformation model and a commutation voltage fitting model. The adaptive fusion module assigns the Park transformation model and the commutation voltage fitting model their respective adaptive factors f according to the system natural frequency ωt. 1(ωt) 、f 2(ωt) , weighted to obtain the final commutation voltage amplitude U m and offset angle As follows: And m =U m1 *f 1(ωt) +U m2 *f 2(ωt) (13) The adaptive factor f 1(ωt) 、f 2(ωt) As shown below: f 2(ωt) =1-f 1(ωt) (16) Among them, C1=14, C2=3.4, C3=2, C4=-3.

3.

2. The commutation failure prediction method based on dual-model adaptive fusion algorithm according to claim 1 is characterized in that: In step S2, the principle of the Park transformation model is: extract the three-phase commutation voltage in the system in real time, construct a virtual three-phase voltage for each phase after rotation transformation, and output the voltage U mapped on the virtual d and q axes after Park transformation. xd , U xq , calculate the commutation voltage amplitude U by formula (5) and formula (7) respectively m1 and offset angle Wherein, j represents an imaginary number; ω represents the phase-locked loop frequency; and t represents the time variable.

3. The commutation failure prediction method based on dual-model adaptive fusion algorithm according to claim 2 is characterized in that: In step S4, the area S provided by the commutation process is calculated pro , the formula is as follows: where γ min is the minimum turn-off angle; β is the advance trigger angle.

4. The commutation failure prediction method based on dual-model adaptive fusion algorithm according to claim 3 is characterized in that: In step S5, the system commutation required area S is calculated. need , the formula is as follows: Among them, X r is the equivalent commutation reactance; t0 is the start time of commutation; t1 is the end time of commutation; Indicates the DC side current at the start of commutation; Indicates the DC side current at the end of commutation.

Images