High-voltage direct-current line protection method utilizing active response of data driving and control system

Abstract

The invention relates to a high-voltage direct-current line protection method utilizing active response of a data driving and control system, and belongs to the technical field of ultrahigh-voltage direct-current transmission line protection. The method comprises the following steps: carrying out principal component analysis on voltage waveform data of obtained fault data, and constructing a two-dimensional PCA (Principal Component Analysis) plane; fitting a main distribution axis through linear regression, and vertically projecting each fault sample point onto the main distribution axis to obtain a projection scalar; using kernel density estimation to obtain probability density distribution of a projection scalar of an intra-region fault sample, and obtaining a fuzzy region; continuously monitoring the change rate of the direct-current voltage, if the change rate exceeds a preset threshold, judging that the system has a fault, starting a protection device, acquiring real-time waveform data, projecting the real-time waveform data to a two-dimensional PCA plane to obtain real-time feature points, and normalizing the real-time feature points to obtain a real-time projection scalar; and different control strategies are triggered based on the real-time projection scalar. The technical problem that in the prior art, converter valve protection action locking is easily caused, and a large amount of power loss is caused is solved.

Description

Technical Field

[0001] This invention relates to a high-voltage direct current (HVDC) line protection method that utilizes data-driven and control system active response, belonging to the field of ultra-high voltage direct current (UHVDC) transmission line protection technology. Background Technology

[0002] With the increasing use of renewable energy sources such as wind and solar power, these power generation centers are located in relatively remote areas far from load centers. Therefore, a power transmission technology with large capacity and long transmission distance is needed to transmit electricity. Ultra-high voltage direct current (UHVDC) transmission has many advantages over AC transmission, such as large transmission capacity, long transmission distance, convenient grid interconnection, and easy power regulation, and has therefore received widespread attention and development.

[0003] Given the more complex structure and diverse operating modes of ultra-high voltage direct current (UHVDC) transmission systems, the requirements for control and protection are also higher. Furthermore, UHVDC transmission lines are long, traversing complex terrains and surrounding environments, making them more prone to failure compared to other components. Transmission line failures account for approximately 50% of all UHVDC system faults, making the protection of UHVDC transmission lines particularly crucial.

[0004] Currently, traveling wave protection operates rapidly, but it is susceptible to interference, requires high sampling frequency, is insensitive to high-resistance grounding faults, and lacks mature setting principles. Differential undervoltage protection has higher reliability than traveling wave protection, but its shortcomings are similar. Frequency-domain based protection requires high sampling frequency, has complex algorithms, and involves a large amount of computation. Longitudinal differential protection can detect high-resistance grounding faults, but it has high communication requirements and is greatly affected by the distributed capacitance current of the line, resulting in a slow response speed. As the transition resistance and the distance from the fault point to the protection installation point gradually increase, the differences in fault electrical characteristics between faults within and outside the zone gradually become blurred. This makes single-ended quantity protection prone to failure to operate under high-resistance faults. Due to their long delay, backup protections such as longitudinal differential protection cause converter protection to operate before them, blocking the DC transmission system and resulting in a significant loss of transmission power. Therefore, this invention proposes a high-voltage DC line protection method that utilizes data-driven and control system active response to improve the performance of transmission line protection. Summary of the Invention

[0005] The purpose of this invention is to provide a high-voltage direct current (HVDC) line protection method that utilizes data-driven and control system active response. This method aims to solve the problems of high-resistance fault down-sweep protection failure, excessive delay in low-voltage protection and longitudinal differential protection, which easily lead to converter valve protection blocking and cause significant power loss. This invention aims to improve the reliability and sensitivity of HVDC transmission line protection.

[0006] To achieve the above objectives, the technical solution of the present invention is: a high-voltage direct current line protection method utilizing data-driven and control system active response, comprising the following steps: Step 1: Perform principal component analysis on the voltage waveform data of the acquired fault data, and project the high-dimensional waveform data onto the two-dimensional PCA plane composed of the first two principal components. Step 2: In the two-dimensional PCA plane, the faults in the area are continuously distributed along an axis. The main distribution axis is fitted by linear regression. Each fault sample point is vertically projected onto the main distribution axis, and the projected points are normalized to obtain a projection scalar between 0 and 1. Step 3: Use kernel density estimation to obtain the probability density distribution of the projected scalar of the fault samples within the region. By setting a density threshold, the tail of the density distribution curve is defined as the starting point of the fuzzy region, and the minimum value of the projected scalar of the fault samples outside the region is defined as the ending point of the fuzzy region, thus obtaining the fuzzy region. Step 4: Continuously monitor the DC voltage change rate. If it exceeds the preset threshold, it is determined to be a system fault, the protection device is activated, and real-time waveform data is collected. Step 5: Project the real-time waveform data onto the two-dimensional PCA plane to obtain real-time feature points and project them onto the principal distribution axis, then normalize them to obtain real-time projection scalars; Step 6: Trigger different control strategies based on the real-time projected scalar; for fault zones within the zone, execute a fast isolation strategy, trip the circuit breaker and start a restart sequence; for fuzzy zones, start a collaborative discrimination strategy; for fault zones outside the zone, directly execute a blocking strategy.

[0007] Optionally, Step 1 specifically includes: Step 1.1: Normalize the fault voltage curve cluster using the DC voltage during normal operation as a reference. The expression is:

[0008] In the formula, u ( t ) indicates the pole voltage at the protection installation location; U N This indicates the rated DC voltage during normal operation. This represents the normalized voltage value. Step 1.2: Select 10 sampling points (1 before the fault and 9 after the fault) as the fault sample data and perform PCA clustering: m Data from the fault voltage curve u Composition m The original data matrix is ​​10 × 10 dimensional. P The expression is:

[0009] The signal is represented using orthogonal basis functions, and the expression is:

[0010] In the formula, f i ( t ) are orthogonal basis functions. k ij For signal u i ( t In orthogonal basis functions f i ( t Projection onto ) Original data matrix P Converted to matrix form, the expression is:

[0011] In the formula, Q for m× 10th order coefficient matrix, F =[ f 1( t ), f 2( t ), ..., f 10 ( t )] is 10 × 10th order matrix; Let the eigenvector matrix be... C The expression is:

[0012] In the formula, for P The matrix formed by subtracting the mean from each element has an eigenvalue corresponding to an eigenvector, and satisfies the following:

[0013]

[0014] In the formula, λ p Eigenvector matrix C The corresponding eigenvalues, It is a diagonal matrix; Diagonal matrix The eigenvalues ​​are transformed into a descending order from largest to smallest, and the eigenvectors corresponding to the two largest eigenvalues ​​are selected as the first principal component and the second principal component, thereby constructing a two-dimensional PCA plane.

[0015] Optionally, Step 2 specifically includes: Step 2.1: Based on the constructed two-dimensional PCA plane, linear regression is performed on the internal fault samples to fit the principal distribution axis in order to capture the continuous changing trend of fault characteristics from the beginning to the end of the line, and a matrix is ​​designed. X = [1, PC 1int ], response variable Y = PC 2int ,in, PC 1int The value of the first principal component of the internal fault sample point. PC 2int The second principal component value and regression coefficient of the internal fault sample points. β The expression is:

[0016] The principal axis equation expression obtained from the regression coefficients is as follows:

[0017] in, β 1 is the intercept. β 2 represents the slope; PC 1 represents the value of the first principal component. PC 2 represents the value of the second principal component; Step 2.2: Project all fault samples onto the principal distribution axis, projection values p The calculation formula is:

[0018] The projection values ​​are normalized using the following expression:

[0019] In the formula, S For the projected values p The projected scalar after normalization; S min and S max These are the minimum and maximum values ​​of the internal fault projection, respectively.

[0020] Optionally, Step 3 specifically includes: Step 3.1: Use kernel density estimation (KDE) to obtain the S-value distribution of internal fault samples, the expression of which is:

[0021] In the formula, s Let be any point on the principal axis of projection; S iFor internal fault sample projection scalar; K The Gaussian kernel function; h For bandwidth parameters; For position s The estimated probability density at; Step 3.2: The definition of the fuzzy region is achieved through density change detection, where the starting point... S start Defined as the proportion of density that first drops to its peak value α At that point, the expression is:

[0022] In the formula, inf{…} represents the infm; max( ) represents the probability density curve The maximum value in the domain; end S end Take the minimum value of the external fault sample projection scalar: S end =min( S ext ),in, S ext For external fault sample projection scalars, the final blurred area Defined as:

[0023] In the formula, R 2 This represents the region defined in the PCA two-dimensional plane; S ( PC 1, PC 2) Represents the projection scalar of the fault sample point in the PCA plane onto the principal axis.

[0024] Optionally, the expression for determining a system fault is:

[0025] in, u dc The instantaneous value of the pole voltage measured at the line outlet; Δ u set To protect the startup threshold value.

[0026] Optionally, Step 6 specifically includes: Step 6.1: Determine the fault type based on the real-time projected scalar, expressed as:

[0027] in, S It is a projected scalar; Sstart and S end These are the start and end points of the fuzzy region, respectively. Step 6.2: Construct the electrode selection criterion based on the magnitude and direction of the grounding electrode line current. The expression is:

[0028] in, τ This represents a sampling time point, and its window range is the current time. t The previous 6ms cycle; i g ( τ () represents the grounding electrode current within the sampling window; I g,set The threshold value for the grounding electrode current; i m ( t ( ) represents the current at the measurement point; I dc,set The threshold for measuring point current is ∀, which means that the logical condition must be satisfied simultaneously for all sampling points, and ∧ means logical AND. Step 6.3: Implement different control strategies based on different fault types, specifically as follows: Fault within the zone: The control system will force a phase shift restart by changing the firing angle of the rectifier side to 18°-120°-160°-18°. External fault: The control system will force phase shifting and blocking of the rectifier side firing angle at 18°-160°-90°; For fuzzy zone faults: the control system will prioritize a forced phase shift restart of the rectifier side firing angle at 18°-120°-160°-18°; during this period, the real-time judgment result of the other end will be continuously monitored. If the other end judges the fault type as an internal fault, the control strategy will not be changed; if the other end judges the fault type as an external fault, the control strategy will be modified to a forced phase shift lockout of 18°-160°-90°.

[0029] The beneficial effects of this invention are as follows: This invention can be divided into two stages: offline training and online protection. In the offline stage, a training set is first constructed using simulation and field data, and the fault voltage waveform is projected onto a low-dimensional PCA space through principal component analysis. Then, based on the distribution of fault points within the zone in the PCA space, the principal distribution axis is fitted, the projected scalar is calculated and normalized. Finally, kernel density estimation is applied to analyze the probability density distribution of the fault scalar within the zone, scientifically delineating the boundaries of the fuzzy zone, thereby establishing a data-driven adaptive discrimination benchmark. In the online protection stage, the system monitors the DC voltage change rate in real time to initiate protection, collects fault waveforms and projects them onto the PCA space to calculate real-time scalars. Based on the interval into which the scalar falls—clearly defined fault zone, fuzzy zone, or clearly defined fault zone outside the zone—differentiated strategies are dynamically triggered, namely rapid isolation, collaborative discrimination, or blocking, improving the performance of the high-voltage direct current transmission system line protection system. Attached Figure Description

[0030] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments mentioned in this application without creative work are within the scope of protection of this application.

[0031] Figure 1 This is a flowchart of the steps of the present invention; Figure 2 This is a simulation circuit diagram of a conventional DC power transmission system according to the present invention; Figure 3 This is a cluster of metallic fault voltage curves obtained from the simulation experiments of this invention; Figure 4 This is a set of fault voltage curves with transition resistance obtained from the simulation experiments of this invention; Figure 5 This is the PCA space constructed by principal component analysis in this invention; Figure 6 This invention defines the fuzzy region and the process of defining it using kernel density estimation. Figure 7 The voltage waveform and start-up criteria for the fault were set up in the verification test for this invention; Figure 8 Real-time feature projection and projection scalar of the fault set for the verification test of this invention; Figure 9 The grounding electrode line current waveform and corresponding control strategy were set up for the verification test of this invention. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly and thoroughly described below with reference to the accompanying drawings and specific embodiments. For ease of description, it is obvious that the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0033] It should be understood that spatial relative terms, such as "above," "below," "left," and "right," may be used in the text. These spatial relative terms are used to indicate the different positions of related devices or signals in the diagram.

[0034] Example 1: As Figure 1 As shown, a high-voltage direct current line protection method utilizing data-driven and control system active response includes the following steps: Step 1: Perform principal component analysis on the voltage waveform data of the acquired fault data, and project the high-dimensional waveform data onto the two-dimensional PCA plane composed of the first two principal components. Step 1.1: Normalize the fault voltage curve cluster using the DC voltage during normal operation as a reference. The expression is:

[0035] In the formula, u ( t ) indicates the pole voltage at the protection installation location; U N This indicates the rated DC voltage during normal operation. This represents the normalized voltage value. Step 1.2: Select 10 sampling points (1 before the fault and 9 after the fault) as the fault sample data and perform PCA clustering: m Data from the fault voltage curve u Composition m The original data matrix is ​​10 × 10 dimensional. P The expression is:

[0036] The signal is represented using orthogonal basis functions, and the expression is:

[0037] In the formula, f i ( t ) are orthogonal basis functions. k ij For signal u i ( t In orthogonal basis functionsf i ( t Projection onto ) Original data matrix P Converted to matrix form, the expression is:

[0038] In the formula, Q for m× 10th order coefficient matrix, F =[ f 1( t ), f 2( t ), ..., f 10 ( t )] is 10 × 10th order matrix; Let the eigenvector matrix be... C The expression is:

[0039] In the formula, for P The matrix formed by subtracting the mean from each element has an eigenvalue corresponding to an eigenvector, and satisfies the following:

[0040]

[0041] In the formula, λ p Eigenvector matrix C The corresponding eigenvalues; the larger the eigenvalue, the more feature information the corresponding eigenvector contains from the original data matrix. It is a diagonal matrix; Diagonal matrix The eigenvalues ​​in the data are transformed into a descending order from largest to smallest. The eigenvectors corresponding to the two largest eigenvalues ​​are selected as the first principal components and the second principal components, thereby constructing a two-dimensional PCA plane to achieve dimensionality reduction while preserving the original feature information of the data to a large extent.

[0042] Optionally, the fault data is obtained by simulating various operating conditions through electromagnetic transient simulation, specifically: setting up external faults with a transition resistance of 0Ω at different locations from the near end to the far end of the transmission line, as well as various external faults at locations such as converter valves and AC systems on the rectifier side and inverter side.

[0043] Specifically, in this embodiment, based on actual traditional DC transmission system parameters, an electromagnetic transient simulation model of an LCC-HVDC true bipolar transmission system is built on the PSCAD / EMTDC platform. The rated voltage is ±800kV, the rated capacity is 5000MW, the converter station has a dual 12-pulse series valve group series structure, and the outlet is equipped with a smoothing reactor and a DC filter. The transmission line is 1500km long. Considering the symmetry of the HVDC transmission system, the fault analysis and simulation verification in this embodiment only focus on the protection of the DC line on the rectifier side.

[0044] Furthermore, assuming a ground fault occurs on the positive line of the DC system, 300 metallic ground fault points are set every 5 km along the entire 1500 km length of the UHVDC line, from farthest to near. Outside the fault zone, rectifier-side and inverter-side faults are also included. Rectifier-side faults include rectifier-side output faults and rectifier-side AC system faults; inverter-side faults include inverter-side output faults and inverter-side AC system faults, totaling 10 fault points. With a sampling frequency of 6.4 kHz, a cluster of line fault voltage curves under the aforementioned faults is obtained through PSCAD / EMTDC electromagnetic transient simulation.

[0045] Step 2: In the two-dimensional PCA plane, the faults in the area are continuously distributed along an axis. The main distribution axis is fitted by linear regression. Each fault sample point is vertically projected onto the main distribution axis, and the projected points are normalized to obtain a projection scalar between 0 and 1. Step 2.1: Based on the constructed two-dimensional PCA plane, linear regression is performed on the internal fault samples to fit the principal distribution axis in order to capture the continuous changing trend of fault characteristics from the beginning to the end of the line, and a matrix is ​​designed. X = [1, PC 1int ], response variable Y = PC 2int ,in, PC 1int The value of the first principal component of the internal fault sample point. PC 2int The second principal component value and regression coefficient of the internal fault sample points. β The expression is:

[0046] The principal axis equation expression obtained from the regression coefficients is as follows:

[0047] in, β 1 is the intercept. β 2 represents the slope; PC 1 represents the value of the first principal component. PC 2 represents the value of the second principal component; Step 2.2: Project all fault samples onto the principal distribution axis, projection values p The calculation formula is:

[0048] The projection values ​​are normalized using the following expression:

[0049] In the formula, S For the projected values p The projected scalar after normalization; S min and S max These are the minimum and maximum values ​​of the internal fault projection, respectively.

[0050] Step 3: Use kernel density estimation to obtain the probability density distribution of the projected scalar of the fault samples in the region. By setting a density threshold, the tail of the density distribution curve (i.e. the region where the high resistance fault causes feature ambiguity) is defined as the starting point of the ambiguity region. Then, the minimum value of the projected scalar of the fault samples outside the region is defined as the ending point of the ambiguity region, thus obtaining the ambiguity region. Step 3.1: Use kernel density estimation (KDE) to obtain the S-value distribution of internal fault samples, the expression of which is:

[0051] In the formula, s Let be any point on the principal axis of projection; S i For internal fault sample projection scalar; K The Gaussian kernel function; h For bandwidth parameters; For position s The estimated probability density at; Step 3.2: The definition of the fuzzy region is achieved through density change detection, where the starting point... S start Defined as the proportion of density that first drops to its peak value α At that point, the expression is:

[0052] In the formula, inf{…} represents the infm; max( ) represents the probability density curve The maximum value in the domain; end S end Take the minimum value of the external fault sample projection scalar: Send =min( S ext ),in, S ext For external fault sample projection scalars, the final blurred area Defined as:

[0053] In the formula, R 2 This represents the region defined in the PCA two-dimensional plane; S ( PC 1, PC 2) Represents the projection scalar of the fault sample point in the PCA plane onto the principal axis.

[0054] Optionally, in this embodiment, the key parameters are optimized through statistical learning; wherein, the bandwidth is calculated using the Scott rule: ,in for S Standard deviation of values; density threshold α The coefficients were determined through cross-validation grid search to maximize the comprehensive evaluation index of protection performance.

[0055] It is understandable that Step 3 of this embodiment is naturally immune to sample imbalance because the starting point of the fuzzy region depends only on the density distribution of internal faults, and external faults are only used to determine the boundary endpoints, effectively solving the performance degradation problem of traditional machine learning methods when samples are imbalanced.

[0056] It is understood that Steps 1-3 above constitute the first stage of this embodiment, namely the offline training stage, which is used to construct the two-dimensional PCA plane and divide the fuzzy region.

[0057] Step 4: Continuously monitor the DC voltage change rate. If it exceeds the preset threshold, it is determined to be a system fault, the protection device is activated, and real-time waveform data is collected. Optionally, the expression for determining a system fault is:

[0058] in, u dc The instantaneous value of the pole voltage measured at the line outlet; Δ u set To protect the startup threshold value.

[0059] Specifically, in this embodiment, the system continuously collects the line voltage signal at a fixed sampling rate of 6.4kHz to monitor the line voltage change rate. When the protection device is activated, it collects one sampling point of the voltage waveform before activation and nine sampling points after activation as fault data.

[0060] Step 5: Project the real-time waveform data onto the two-dimensional PCA plane to obtain real-time feature points and project them onto the principal distribution axis, then normalize them to obtain real-time projection scalars; Optionally, after normalizing the fault voltage waveform acquired in a short real-time time window, the real-time waveform is projected onto a two-dimensional feature plane composed of the first principal component and the second principal component to obtain real-time feature points.

[0061] Step 6: Trigger different control strategies based on the real-time projected scalar; for fault zones within the zone, execute a fast isolation strategy, trip the circuit breaker and start a restart sequence; for fuzzy zones, start a collaborative discrimination strategy; for fault zones outside the zone, directly execute a blocking strategy.

[0062] Step 6.1: Determine the fault type based on the real-time projected scalar, expressed as:

[0063] in, S It is a projected scalar; S start and S end These are the start and end points of the fuzzy region, respectively. Step 6.2: Construct the electrode selection criterion based on the magnitude and direction of the grounding electrode line current. The expression is:

[0064] in, τ This represents a sampling time point, and its window range is the current time. t The previous 6ms cycle; i g ( τ () represents the grounding electrode current within the sampling window; I g,set The threshold value for the grounding electrode current; i m ( t ( ) represents the current at the measurement point; I dc,set The threshold for measuring point current is ∀, which means that the logical condition must be satisfied simultaneously for all sampling points, and ∧ means logical AND. Step 6.3: Implement different control strategies based on different fault types, specifically as follows: Fault within the zone: The control system will force a phase shift restart by changing the firing angle of the rectifier side to 18°-120°-160°-18°. External fault: The control system will force phase shifting and blocking of the rectifier side firing angle at 18°-160°-90°; For fuzzy zone faults: the control system will prioritize a forced phase shift restart of the rectifier side firing angle at 18°-120°-160°-18°; during this period, the real-time judgment result of the other end will be continuously monitored. If the other end judges the fault type as an internal fault, the control strategy will not be changed; if the other end judges the fault type as an external fault, the control strategy will be modified to a forced phase shift lockout of 18°-160°-90°.

[0065] It is understood that Steps 4-6 above constitute the second stage of this embodiment, namely the online protection stage, which is used to make real-time judgments and execute different control strategies according to different fault types when a line fault occurs.

[0066] The effectiveness of the present invention is further demonstrated below through specific examples. The specific implementation process is as follows: Phase 1: Offline Training S1: Based on actual traditional DC transmission system parameters, a system is built in PSCAD / EMTDC as follows: Figure 2 The simulation model shown.

[0067] LCC-HVDC is a true bipolar transmission system based on grid-commutated converters. Taking a traditional DC transmission system as the research object, it has a rated voltage of ±800kV and a rated capacity of 5000MW. The converter station has a dual 12-pulse series valve group series structure, and the outlet is equipped with a smoothing reactor and a DC filter. The transmission system is as follows: Figure 2 As shown, considering the symmetry of the high-voltage direct current transmission system, the fault analysis and simulation verification in this embodiment only focuses on the protection of the DC line on the rectifier side.

[0068] S2: In the established simulation model, set up in-zone and out-of-zone faults with different fault conditions.

[0069] Assuming a ground fault occurs on the positive line of the DC system, 300 metallic ground fault points are set every 5 km along the entire 1500 km length of the UHVDC line, from farthest to near. Outside the fault zone, rectifier-side and inverter-side faults are included. Rectifier-side faults include rectifier-side output faults and rectifier-side AC system faults; inverter-side faults include inverter-side output faults and inverter-side AC system faults, totaling 10 fault points. Assuming a sampling frequency of 6.4 kHz, the line fault voltage curves under the above faults are obtained through PSCAD / EMTDC electromagnetic transient simulation, and the differences in fault voltage between faults within and outside the fault zone are indicated as follows: Figure 3 As shown.

[0070] S3: Further set the fault over-resistance to reveal the trend of fault voltage as the fault resistance increases.

[0071] Transition resistors of 50 Ω, 100 Ω, 150 Ω, 200 Ω, 250 Ω, and 300 Ω are sequentially installed based on the fault conditions within the zone. The fault voltage waveform is as follows: Figure 4 As shown, under the position-resistance coupling effect, the differences in voltage waveform characteristics between faults within and outside the zone gradually become blurred, which is the fundamental reason why traveling wave protection fails to operate under high-resistance faults.

[0072] S4: Analyze the fault voltage waveform using principal component analysis.

[0073] Principal component analysis (PCA) was used to select the principal components that best represent the data characteristics, constructing a two-dimensional plane to transform the complex fault voltage waveform into an interpretable distribution on the two-dimensional plane. Internal faults exhibit a banded distribution as resistance increases. For example... Figure 5 As shown.

[0074] S5: Fit the principal distribution axis and calculate the projection scalar Based on the established PCA model, a principal distribution axis is fitted by performing linear regression on the internal fault samples to capture the continuous changing trend of fault characteristics from the beginning to the end of the line. Each fault sample point is then vertically projected onto this principal distribution axis, and finally, the projected points are normalized to obtain a projection scalar between 0 and 1.

[0075] S6: Scientifically define the fuzzy region using kernel density estimation.

[0076] The probability density distribution of the projected scalar of fault samples within all regions was analyzed using kernel density estimation. The test samples are as follows: Figure 6 As shown in (a). A density threshold is set, defining the tail of the density distribution curve as the starting point of the fuzzy region, and the vicinity of the minimum value of the projection scalar of the fault samples outside the region as the ending point of the fuzzy region, as shown in (a). Figure 6 As shown in (b); the final defined fuzzy region, as... Figure 6 As shown in (c).

[0077] Phase Two: Online Protection Phase S7: Fault setting and protection activation.

[0078] like Figure 7 As shown, a ground fault with a transition resistance of 200Ω is installed at the midpoint of the positive transmission line. The fault occurs at 0ms, and the fault traveling wave reaches the protection installation point after 2.5ms. The fault voltage waveform is as follows. Figure 7 As shown in (a), the activation criterion (d) is satisfied. u dc / d t >Δ u set The protection is activated after 0.6ms, and the voltage change rate waveform is as follows: Figure 7 As shown in (b).

[0079] S8: Real-time feature projection and scalar computation.

[0080] After normalizing the acquired fault voltage waveform within a short real-time window, the real-time waveform is projected onto a two-dimensional feature plane composed of the first and second principal components, yielding real-time feature points (-0.49, -0.13). These real-time feature points are then projected onto the principal distribution axis determined in the offline phase and normalized to obtain the real-time projection scalar. S =1.2543), such as Figure 8 As shown in (a), based on the real-time projected scalar, the fault type is initially determined to be a fuzzy region fault. S start ≤ S ≤ S end ),like Figure 8 As shown in (b).

[0081] S9: Fault Selection and Control Strategy Based on the magnitude and direction of the grounding electrode line current, ( The fault is determined to be a positive circuit fault, such as... Figure 9 As shown in (a). During the fault selection process, signals from the other end are constantly received to correct the operating strategy. Since no blocking signal from the other end is received, the protection system ultimately determines that it is a line fault, and therefore performs a forced phase shift restart according to 18°-120°-160°-18°, as shown in (a). Figure 9 As shown in (b).

[0082] Furthermore, this embodiment also sets up respectively F 1 (Metallic grounding at the midpoint of the line). F 2 (Inverter-side converter valve grounding). F 3 (Rectifier-side converter valve grounding), specific data are shown in Table 1.

[0083] Table 1 Judgment Results

[0084] In summary, this invention consists of two phases: offline training and online protection. In the offline phase, a training set is first constructed using simulation and field data. Principal component analysis (PCA) is then used to project the fault voltage waveform onto a low-dimensional PCA space. Next, the principal distribution axis is fitted based on the distribution of fault points within the zone in the PCA space, and the projected scalar is calculated and normalized. Finally, kernel density estimation is applied to analyze the probability density distribution of the fault scalar within the zone, scientifically delineating the boundaries of the fuzzy zone, thereby establishing a data-driven adaptive discrimination benchmark. In the online protection phase, the system monitors the DC voltage change rate in real time to initiate protection, collects fault waveforms, and projects them onto the PCA space to calculate real-time scalars. Based on the interval the scalar falls into—whether it falls within a clearly defined fault zone, a fuzzy zone, or a clearly defined fault zone outside the zone—differentiated strategies are dynamically triggered, namely rapid isolation, collaborative discrimination, or blocking, improving the performance of the high-voltage direct current transmission system's line protection system.

[0085] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for HVDC line protection using data driven and actively responding system, characterized in that, The method comprises the following steps: Step 1: principal component analysis is performed on the acquired voltage waveform data of the fault data, and high-dimensional waveform data is projected onto a two-dimensional PCA plane composed of the first two principal components; Step 2: in the two-dimensional PCA plane, the in-zone fault is continuously distributed along an axis, a main distribution axis is fitted by linear regression, each fault sample point is vertically projected onto the main distribution axis, and the projection point is normalized to obtain a projection scalar between 0 and 1; Step 3: the probability density distribution of the projection scalar of the in-zone fault sample is obtained by using kernel density estimation, the tail of the density distribution curve is defined as the starting point of the fuzzy zone by setting a density threshold, and the minimum value of the projection scalar of the out-zone fault sample is defined as the end point of the fuzzy zone, so as to obtain the fuzzy zone; Step 4: the rate of change of the direct current voltage is continuously monitored, and if it exceeds a preset threshold, it is determined that the system is faulty, the protection device is started, and real-time waveform data is collected; Step 5: the real-time waveform data is projected onto the two-dimensional PCA plane to obtain real-time feature points and project them onto the main distribution axis, and then normalized to obtain real-time projection scalars; Step 6: different control strategies are triggered based on the real-time projection scalars; for the in-zone fault zone, a rapid isolation strategy is executed, tripping and starting a restart sequence; for the fuzzy zone, a collaborative discrimination strategy is started; and for the out-zone fault zone, a lockout strategy is directly executed.

2. A method for HVDC line protection using data-driven and control system active response according to claim 1, characterized in that, The Step 1 is specifically: Step 1.1: the fault voltage curve cluster is normalized with respect to the direct current voltage during normal operation, and the expression is: ； In the formula, u t U represents the polar voltage at the protection installation site; N U represents the DC voltage rating in normal operation; U represents the normalized voltage value;​ Step 1.2: Select 1 sample point before the fault, 9 sample points after the fault, a total of 10 sample point data as fault sample data, and perform PCA clustering: divide the data into 10 clusters m Data of the fault voltage curve u is constructed as m a 10-dimensional original data matrix P , and the expression is: ； The orthogonal basis function is used to represent the signal, and the expression is: ； wherein f i ( t ) is an orthogonal basis function, k ij is a signal u i ( t ) is a projection of the signal f i ( t ) on the orthogonal basis function The original data matrix P is converted into a matrix form, expressed as: ； wherein Q is m× 10th order coefficient matrix, F = [ f 1( t ), f 2( t ), f 10 ( t ) is a 10 × 10th order matrix; Let Feature vector matrix C The expression is: ； In the formula, is P The matrix formed by subtracting the mean value from each element in the matrix, each eigenvalue corresponds to a characteristic vector, and satisfies: ； ； wherein The Step 2 is specifically: p is a matrix of eigenvectors C corresponding eigenvalues, is a diagonal matrix; The eigenvalues in the diagonal matrix are converted into descending order from large to small, and the eigenvectors corresponding to the largest two eigenvalues are selected as the first principal component and the second principal component, thereby constructing a two-dimensional PCA plane.

3. The method of claim 2, wherein the method is characterized by, PC Step2.1: Based on the constructed two-dimensional PCA plane, the main distribution axis is fitted by linear regression on the internal fault samples to capture the continuous change trend of fault characteristics from the beginning to the end of the line, and the design matrix X = [1, PC 1int ] is obtained, and the response variable Y = PC 2int , wherein PC 1int is the first principal component value of the internal fault sample point, The main axis equation expression is obtained by the regression coefficient: 2int is the second principal component value of the internal fault sample point, and the regression coefficient β is expressed as: ； PC ； wherein β 1 is the intercept, β 2 is the slope; PC 1 is the first principal component value, The projection value is normalized, and the expression is: 2 is the second principal component value; Step 2.2: Project all faulty samples to the principal distribution axis, the projection value p The calculation formula is: ； The Step 3 is specifically: ； In the formula, S is the projection value of the internal fault after normalization processing; p is the projection value of the internal fault after normalization processing; S min and S max are the minimum and maximum values of the internal fault projection value, respectively.

4. The method of claim 3, wherein the method further comprises: Step 3.1: the S value distribution of the internal fault sample is obtained by using kernel density estimation KDE, and the expression is: PC ； where s is the projection of any point on the principal axis of projection; S i is the projection of the internal fault sample; K is the Gaussian kernel function; h is the bandwidth parameter; is the estimated probability density at location s ​ Step 3.2: Definition of fuzzy region is achieved by density change detection, where the start point S start is defined as the first drop of density to the peak ratio α , which is expressed as: ； where inf{...} denotes the infimum; max{...} denotes the maximum ) denotes the probability density curve the maximum value in the domain of definition; end S end Take the minimum value of the external fault sample projection scalar: S end =min( S ext ),in, S ext For external fault sample projection scalars, the final blurred area Defined as: ； wherein R 2 represents the area delimited in the PCA two-dimensional plane; S PC 1, The expression for determining the system fault is: 2) represents the projection scalar of the fault sample point projected onto the principal axis in the PCA plane.​ 5. The method of claim 1, wherein the method is characterized by, The Step 6 is specifically: ； wherein u dc is the polar line voltage instantaneous value measured at the line outlet; Δ u set is the protection activation threshold value.

6. The method of claim 1, wherein the method is characterized by, Step 6.1: the fault type is determined according to the real-time projection scalar, and the expression is: Step 6.2: a pole selection criterion is constructed according to the size and direction of the ground pole line current, and the expression is: ； wherein S is a projection scalar; S start and S end are the start and end points of the blur zone, respectively. Tau ； wherein, Tau represents a sampling time point, whose window range is the current time t 6 ms period before the current time; i g ( Step 6.3: different control strategies are executed according to different fault types, specifically: ) is the ground electrode current in the sampling window; I g,set is the threshold value of the ground electrode current; i m ( t ) is the measurement point current; I dc,set is the threshold value of the measurement point current, and represents that all sampling points satisfy the logical condition simultaneously, and represents logical AND. In-zone fault: the control system forcibly shifts the phase of the rectifier side trigger angle by 18°-120°-160°-18° and restarts; Out-zone fault: the control system forcibly shifts the phase of the rectifier side trigger angle by 18°-160°-90° and locks; Fuzzy zone fault: the control system forcibly shifts the phase of the rectifier side trigger angle by 18°-120°-160°-18° and restarts; the real-time discrimination result of the opposite end is continuously monitored during the process, if the opposite end discriminates the fault type as an internal fault, the control strategy is not changed; if the opposite end discriminates the fault type as an out-zone fault, the control strategy is modified to forcibly shift the phase of the rectifier side trigger angle by 18°-160°-90° and lock. ​

Images