A power distribution network fault line identification method based on zero sequence current transient oscillation characteristics adaptive fusion

Abstract

The application discloses a distribution network fault line identification method based on zero sequence current transient oscillation characteristics adaptive fusion, relates to distribution network fault diagnosis and identification technology, and triggers fault sampling by bus zero sequence voltage exceeding threshold value, adopts adaptive time window synchronous collection of each line zero sequence current based on transient oscillation period dynamic extension; adaptive decomposition and denoising reconstruction of signals are completed through variational mode decomposition combined with permutation entropy, four-dimensional transient characteristic parameters of modified transient peak amplitude, instantaneous dominant oscillation frequency, logarithmic decay rate and transient energy entropy are extracted; based on fault working condition estimation factor, dynamic adaptive matching of feature criterion weight is realized through a particle swarm optimized fuzzy logic inference machine, and comprehensive fault confidence index of each line is calculated; finally, a double-criterion discrimination function is constructed in combination with a zero sequence current-voltage polarity correlation coefficient, accurate fault line identification and bus fault discrimination are completed, and the method can be widely applied to 35kV and below medium-voltage distribution network single-phase ground fault line selection scenes.

Description

Technical Field

[0001] This invention relates to distribution network fault diagnosis and identification technology, specifically a method for identifying faulty lines in distribution networks based on adaptive fusion of zero-sequence current transient oscillation characteristics. Background Technology

[0002] Medium-voltage distribution networks of 35kV and below are the core hubs connecting the power generation side and end users in the power system. Their power supply reliability directly determines the order of industrial and agricultural production and the safety of electricity use for people's livelihood. Single-phase grounding faults are the most frequent type of fault in medium-voltage distribution networks, accounting for more than 80% of all faults in the distribution network. If the faulty line cannot be quickly and accurately identified and isolated in time after a fault occurs, it will cause problems such as insulation breakdown, equipment burnout, and expansion of the fault range, and even cause large-scale power outages. Overvoltages generated by operating with faults for a long time will also continue to threaten the insulation safety of the entire power grid system. Therefore, high-precision and highly adaptable fault line identification technology is the core foundation for ensuring the safe and stable operation of the distribution network.

[0003] Current fault location technologies for distribution networks are mainly divided into two categories: steady-state fault location and transient fault location. Steady-state fault location methods rely on steady-state characteristics such as the amplitude and phase of the zero-sequence current to identify faults. While simple in principle and easy to implement in engineering, in neutral-point grounded systems with arc-suppression coils, the characteristics of the steady-state zero-sequence current of the faulty line are significantly weakened after compensation. When facing high-resistance grounding faults, the steady-state characteristics are almost completely drowned out by on-site electromagnetic noise, leading to serious problems of false tripping and failure to trip, severely limiting its applicability. Transient fault location methods, on the other hand, utilize the rich characteristics of the transient zero-sequence current after a fault for identification. Unaffected by arc-suppression coil compensation, they are more adaptable to high-resistance faults and have become the mainstream direction in industry research and engineering applications.

[0004] However, existing transient signal selection technologies still have many technical shortcomings that are difficult to overcome: First, they mostly use fixed time windows to collect signals, which cannot adapt to the differences in the duration of transient oscillations under different fault conditions, and are prone to effective feature truncation or invalid data redundancy; Second, the signal decomposition and denoising methods have insufficient anti-interference capabilities, and are prone to mode mixing, which leads to fault feature distortion and poor feature fidelity in strong noise environments; Third, they generally use fixed-weight fault criteria, which cannot be adaptively adjusted according to fault distance, grounding resistance and other conditions, and are only effective for near-end low-resistance faults, while the identification accuracy drops sharply in far-end high-resistance fault scenarios; Fourth, the single-criteria decision mode is susceptible to interference, cannot effectively distinguish between line faults and bus faults, has insufficient anti-maloperation capabilities, and is difficult to adapt to the complex and ever-changing field operating environment of distribution networks. Summary of the Invention

[0005] The purpose of this invention is to provide a method for identifying faulty lines in distribution networks based on adaptive fusion of zero-sequence current transient oscillation characteristics. By taking the zero-sequence current transient oscillation characteristics as the core, and through five core links—adaptive time window synchronous acquisition, VMD-PE adaptive signal reconstruction, four-dimensional transient feature extraction, PSO optimized fuzzy logic weight adaptive matching driven by multi-condition factors, and polarity correlation-confidence dual criterion fusion line selection—it achieves accurate and rapid identification of faulty lines under different fault conditions. At the same time, it has the ability to identify bus faults and prevent maloperation blocking. It effectively solves the industry pain points of traditional line selection methods, such as weak transient features, poor anti-interference ability, and low identification accuracy in high-resistance grounding and remote fault scenarios.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for identifying faulty lines in distribution networks based on adaptive fusion of zero-sequence current transient oscillation characteristics includes the following steps:

[0008] S1: Real-time monitoring of the zero-sequence voltage of the distribution network bus. When the absolute value of the zero-sequence voltage of the bus is greater than the set start threshold, each outgoing terminal is triggered to synchronously collect the zero-sequence current within the adaptive time window after the fault. The adaptive time window is dynamically extended according to the transient oscillation period of the detected zero-sequence current.

[0009] S2: Decompose the collected zero-sequence current into several intrinsic mode components, determine the optimal number of modes, calculate the center frequency and arrangement entropy value of each intrinsic mode component, select the intrinsic mode component with the smallest arrangement entropy value and center frequency within a preset range that matches the neutral point grounding method of the distribution network to reconstruct the signal, and obtain the transient oscillation reconstruction signal.

[0010] S3: Extract four-dimensional feature parameters from the transient oscillation reconstruction signal, including the corrected transient peak amplitude, instantaneous dominant oscillation frequency, logarithmic decay rate, and transient energy entropy, and construct a four-dimensional transient feature vector;

[0011] S4: Based on the four-dimensional feature parameters, construct the fault distance estimation factor, grounding resistance estimation factor, and oscillation characteristic estimation factor. Input the three estimation factors into the fuzzy logic inference engine optimized by the optimization algorithm. Through the fuzzy logic inference engine, dynamically match the weight coefficients of the criteria corresponding to the four-dimensional feature parameters. After normalizing the four-dimensional feature parameters, calculate the comprehensive fault confidence index of each line by combining the weight coefficients.

[0012] S5: Calculate the polarity correlation coefficient between the first half-wave of the transient zero-sequence current of each line and the first half-wave of the zero-sequence voltage of the bus. Combine the comprehensive fault confidence index and the polarity correlation coefficient to construct the final discrimination function. Execute the fault line selection logic based on the final discrimination function value and the preset threshold, and output the fault line identification result.

[0013] Furthermore, the start-up threshold mentioned in step S1 is set to 15% of the phase voltage of the distribution network.

[0014] Furthermore, the initial value of the adaptive time window in step S1 is 20ms, which is dynamically extended according to the detected transient oscillation period of the zero-sequence current, with the extension range being 1 to 2 times the oscillation period. The logic of the dynamic extension is as follows: the transient oscillation period of the zero-sequence current is detected by Fourier transform, and the window is extended by adding 1 to 2 times the amplitude of the oscillation period to the initial value of 20ms.

[0015] Furthermore, in step S2, the collected zero-sequence current is decomposed into several intrinsic mode components by variational mode decomposition algorithm, and the optimal number of modes is determined adaptively by permutation entropy; the logic for determining the optimal number of modes is as follows: calculate the permutation entropy of each intrinsic mode component under different number of modes, and select the number of modes with the smallest permutation entropy value as the optimal decomposition number.

[0016] Furthermore, the preset center frequency range mentioned in step S2 is adaptively adjusted according to the neutral point grounding method of the distribution network. The specific adjustment rules are as follows: in a distribution network where the neutral point is grounded through an arc suppression coil, an intrinsic mode component with a center frequency of not less than 150Hz is selected; in a distribution network where the center point is not grounded through an arc suppression coil, an intrinsic mode component with a center frequency in the range of 200Hz to 3000Hz is selected.

[0017] Furthermore, the specific calculation method for the four-dimensional feature parameters mentioned in step S3 is as follows:

[0018] (1) Correcting transient peak amplitude: Select the maximum absolute value of the reconstructed signal as the original peak amplitude, calculate the amplitude correction coefficient, the amplitude correction coefficient is the ratio of the original peak amplitude to the average value of 5 consecutive peaks, and the product of the original peak amplitude and the amplitude correction coefficient is the corrected transient peak amplitude;

[0019] (2) Instantaneous dominant oscillation frequency: The instantaneous phase of the reconstructed signal is obtained by using Hilbert transform. The ratio of the first derivative of the instantaneous phase with respect to time to 2π is taken as the instantaneous frequency. The average value of all instantaneous frequencies is selected as the instantaneous dominant oscillation frequency.

[0020] (3) Logarithmic decay rate: It is calculated by the average of the natural logarithm ratios of consecutive peaks, and the number of consecutive peaks is 8 to 10.

[0021] (4) Transient energy entropy: It is calculated based on the information theory entropy value theory. First, the total energy of the reconstructed signal is calculated, the total energy is divided into 10 to 20 intervals and the energy ratio of each interval is calculated, and then the transient energy entropy is obtained by the entropy value calculation method.

[0022] Furthermore, the specific construction method of the three estimation factors mentioned in step S4 is as follows:

[0023] (1) Fault distance estimation factor: It is calculated based on the instantaneous dominant oscillation frequency. It is obtained by taking the near-end fault reference oscillation frequency as the reference and the ratio of the difference between the instantaneous dominant oscillation frequency and the reference oscillation frequency to the reference oscillation frequency. The range of the fault distance estimation factor is 0 to 1.

[0024] (2) Grounding resistance estimation factor: It is calculated based on the logarithmic decay rate. It is obtained by taking the low-resistance fault reference decay rate and the high-resistance fault maximum decay rate as the reference, and by the ratio of the difference between the logarithmic decay rate and the low-resistance fault reference decay rate to the difference between the high-resistance fault maximum decay rate and the low-resistance fault reference decay rate. The value range of the grounding resistance estimation factor is 0 to 1.

[0025] (3) Oscillation characteristic estimation factor: is the ratio of transient oscillation duration to adaptive time window. The transient oscillation duration is the time interval length where the absolute value of the reconstructed signal is not less than 0.1 times the corrected transient peak amplitude. The oscillation characteristic estimation factor is obtained by normalizing the ratio.

[0026] Furthermore, the reasoning process of the fuzzy logic inference engine in step S4 is as follows: first, the three estimation factors are fuzzified, then reasoning is performed according to the preset fuzzy rules, and finally the dynamic weight coefficients of each feature criterion are output through defuzzification. The fuzzy rules follow the principle that amplitude weight is high when the near end is low and the oscillation is obvious, and energy and frequency weight is high when the far end is high and the oscillation is weak.

[0027] Furthermore, the optimization algorithm described in step S4 is a particle swarm optimization algorithm, and its optimization logic is as follows: using the fault identification accuracy of historical fault data as the fitness function, iteratively optimizing the Gaussian membership function parameters of the fuzzy logic inference engine, wherein the Gaussian membership function parameters include the mean and variance, and iteratively optimizing until the fault identification accuracy tends to stabilize.

[0028] Furthermore, step S5 specifically includes the following steps:

[0029] S51: The polarity correlation coefficient is the Pearson correlation coefficient; the calculation takes half of the transient oscillation period as the time range, and uses the average value of the first half wave of the transient zero-sequence current reconstruction signal of each line and the average value of the first half wave of the bus zero-sequence voltage as the calculation benchmark, and obtains it through the Pearson correlation calculation method; among which, the Pearson correlation coefficient of the faulty line is less than 0, and the Pearson correlation coefficient of the non-faulty line is greater than 0.

[0030] S52: The final discriminant function is the product of the comprehensive fault confidence index and the corresponding expression of the Pearson correlation coefficient;

[0031] S53: The fault selection logic is as follows: The preset threshold is determined by simulation and field data calibration, and is set to 0.6; the final discrimination function value of all lines is traversed to obtain the maximum value; if the maximum value is greater than the preset threshold and the Pearson correlation coefficient of the corresponding line is less than 0, then the line is determined to be a faulty line and a trip output is triggered; if the Pearson correlation coefficient of all lines is greater than 0, then a bus fault is determined, an alarm signal is issued and the trip output is blocked; if the maximum value of the final discrimination function value is not greater than the preset threshold, then the fault characteristics are not obvious, a fault warning signal is issued and the distribution network status is continuously monitored.

[0032] Furthermore, in step S2, the acquired zero-sequence current is decomposed into several intrinsic mode components using an empirical mode decomposition algorithm, an ensemble empirical mode decomposition algorithm, or an intrinsic timescale decomposition algorithm.

[0033] Furthermore, the optimization algorithm mentioned in step S4 is a genetic algorithm, a simulated annealing algorithm, or a neural network optimization algorithm.

[0034] Furthermore, the polarity correlation coefficient mentioned in step S5 is either the Spearman correlation coefficient or the Kendall rank correlation coefficient.

[0035] When a single-phase ground fault occurs in a distribution network, the zero-sequence voltage of the bus will be distorted and raised. This method uses this as a fault initiation criterion to accurately capture the time of fault occurrence. Furthermore, considering the significant differences in the duration and period of transient oscillations under different fault conditions, an adaptive time window based on the dynamic extension of the transient oscillation period is adopted, combined with high-precision multi-terminal synchronization technology, to completely capture the core transient zero-sequence current information of each line fault. This avoids feature truncation or data redundancy caused by a fixed window, laying a precise data foundation for subsequent analysis. Addressing the problem that the power frequency steady-state component and high-frequency random noise in the original zero-sequence current can easily cause distortion and loss of fault features, this method adaptively decomposes the zero-sequence current into multiple independent intrinsic mode components through variational mode decomposition. Combining the permutation entropy's ability to quantify the regularity of the sequence, the effective component with the lowest noise content is selected. Simultaneously, based on the neutral grounding method of the distribution network, the center frequency selection range is adaptively matched to accurately remove interference components and reconstruct a pure signal that retains only the core transient characteristics of the fault, ensuring the accuracy of feature extraction from the root. Based on this, this method extracts four-dimensional characteristic parameters: modified transient peak amplitude, instantaneous dominant oscillation frequency, logarithmic decay rate, and transient energy entropy. It comprehensively characterizes the core characteristics of fault transient oscillation from four dimensions: amplitude, frequency, damping, and energy. Among them, the modified transient peak amplitude can eliminate the error of single-peak abnormal fluctuation and accurately characterize the transient impact intensity; the instantaneous dominant oscillation frequency is strongly correlated with the fault distance; the logarithmic decay rate can quantify the zero-sequence loop damping characteristics that are positively correlated with the grounding resistance; and the transient energy entropy can characterize the distribution concentration of fault transient energy. This method achieves full coverage characterization of different fault distances and grounding resistance conditions. To address the core challenges of significant differences in fault characterization capabilities across various fault conditions and the inability of fixed-weight criteria to adapt to all conditions, this method constructs three estimation factors—fault distance, grounding resistance, and oscillation characteristics—based on four-dimensional features. This enables accurate quantitative characterization of fault conditions. These factors are then input into a fuzzy logic inference engine optimized by the particle swarm optimization algorithm. Following the core rule of "high amplitude weight when near-end resistance is low and oscillation is significant, and high energy and frequency weight when far-end resistance is high and oscillation is weak," the weight coefficients of each feature criterion are dynamically matched. This allows features with strong fault characterization capabilities to receive higher weights, weakens the interference of ineffective features, and achieves adaptive optimization of the criteria for fault conditions.Ultimately, this method combines the stable physical law that faulty and non-faulty lines have a fixed polarity difference—the first half-wave of the transient zero-sequence current of the faulty line has opposite polarities to the first half-wave of the zero-sequence voltage of the bus, while the non-faulty line has the same polarity. It integrates the comprehensive fault confidence index obtained by adaptive weighting with the Pearson correlation coefficient, which characterizes polarity correlation, to construct the final discriminant function. This not only quantifies the strength of fault characteristics through the confidence index but also locks in the essential laws of the faulty line through the polarity correlation coefficient. It amplifies the discriminant characteristics of the faulty line and suppresses the interference signals of the non-faulty line. At the same time, it achieves accurate differentiation in multiple scenarios, including faulty line identification, bus fault discrimination, and early warning of unclear fault characteristics. While ensuring the accuracy of line selection, it minimizes false tripping and improves the safety and reliability of power distribution network supply.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] 1. Significantly improved integrity of fault transient signal acquisition, solving the feature loss problem at its source. Existing technologies use fixed time windows to acquire zero-sequence current, which is prone to truncation of transient features of near-end faults and redundancy of data for high-resistance faults at the far end. This invention designs an adaptive time window based on the dynamic extension of the transient oscillation period. It can match the optimal acquisition duration according to the oscillation characteristics of zero-sequence current under different fault conditions, and completely capture the core transient information of the fault. It avoids the loss of effective features caused by the fixed window, and provides accurate and complete raw data support for subsequent fault analysis. It is especially suitable for various fault scenarios with large differences in transient duration.

[0038] 2. Significantly enhanced signal anti-interference capability and feature fidelity, adaptable to complex and noisy field environments. Addressing the issues of mode aliasing and inability to effectively separate power frequency steady-state components from high-frequency random noise in existing technologies, leading to fault feature distortion, this invention employs variational mode decomposition combined with permutation entropy algorithm to adaptively determine the optimal number of mode decompositions. Simultaneously, based on the neutral grounding method of the distribution network, it adaptively matches the center frequency selection rule of the effective components, accurately removing interference components and reconstructing a pure transient oscillation signal. Even in scenarios with strong electromagnetic interference and weak fault features, it can still completely preserve core fault features, fundamentally improving the accuracy and reliability of feature extraction.

[0039] 3. Adaptive Compatibility Across All Fault Conditions, Completely Overcoming the Adaptability Shortcomings of Traditional Fixed-Weight Fault Criteria: Addressing the industry pain point that existing fixed-weight fault criteria are only effective for near-end low-resistance faults, resulting in a sharp drop in accuracy for far-end faults and high-resistance grounding scenarios, this invention constructs three estimation factors based on four-dimensional transient features: fault distance, grounding resistance, and oscillation characteristics. Through a fuzzy logic inference engine optimized by particle swarm optimization, it achieves dynamic adaptive matching of feature criterion weights, following the core rule of "high weight for near-end low-resistance amplitude and high weight for far-end high-resistance energy frequency." This ensures that the features with the strongest representational ability under different fault conditions receive the highest weight, significantly improving the generalization ability of fault identification across all scenarios. It also maintains an extremely high recognition accuracy for 10kΩ and above high-resistance grounding faults and far-end faults at the end of the line.

[0040] 4. Dual-criteria fusion decision-making balances accuracy and anti-maloperation capability, improving the reliability of power distribution networks. Addressing the shortcomings of existing single-criteria line selection technologies, which are susceptible to interference leading to maloperation and failure to operate, and cannot effectively distinguish between line faults and bus faults, this invention fuses a comprehensive fault confidence index that quantifies the strength of fault characteristics with a zero-sequence current-voltage polarity correlation coefficient that characterizes the essential polarity of the fault to construct the final discriminant function. This not only quantifies the probability of the fault through an adaptively weighted confidence index but also identifies the essential differences between faulty and non-faulty lines through the polarity correlation coefficient. It can automatically amplify the discriminant characteristics of faulty lines and suppress interference signals from non-faulty lines. Simultaneously, a closed-loop discriminant logic is provided to accurately distinguish between three scenarios: line faults, bus faults, and faults with unclear characteristics. Combined with a hardware interlocking loop, this ensures high fault identification accuracy while minimizing false tripping and cascading tripping, significantly improving the power supply safety and continuity of the distribution network.

[0041] 5. Strong engineering adaptability, wide applicability and low implementation cost. The method of this invention can adaptively adapt to various mainstream medium-voltage distribution network grounding methods such as neutral point ungrounded, grounded through arc suppression coil, and grounded through small resistor, without the need to significantly adjust algorithm parameters for different distribution network architectures. At the same time, the algorithm complexity is controllable, and the computational load is adapted to the conventional computing power level of distribution network field terminals. There is no need to add a large number of additional hardware devices. It can be directly embedded into existing distribution network fault recording terminals and feeder terminal units (FTUs). The engineering implementation difficulty is low, and the applicable scope covers medium-voltage distribution networks of 35kV and below, which has extremely high promotion and application value. Attached Figure Description

[0042] Figure 1 This is a flowchart of a method for identifying faulty lines in a distribution network based on adaptive fusion of the transient oscillation characteristics of zero-sequence current, according to the present invention. Detailed Implementation

[0043] The technical solutions of the present invention will now be described in detail with reference to the accompanying drawings.

[0044] like Figure 1 As shown, a method for identifying faulty lines in a distribution network based on adaptive fusion of zero-sequence current transient oscillation characteristics includes the following steps:

[0045] S1: Real-time monitoring of the zero-sequence voltage of the distribution network bus. When the absolute value of the zero-sequence voltage of the bus is greater than the set start threshold, each outgoing terminal is triggered to synchronously collect the zero-sequence current within the adaptive time window after the fault. The adaptive time window is dynamically extended according to the transient oscillation period of the detected zero-sequence current.

[0046] S2: Decompose the collected zero-sequence current into several intrinsic mode components, determine the optimal number of modes, calculate the center frequency and arrangement entropy value of each intrinsic mode component, select the intrinsic mode component with the smallest arrangement entropy value and center frequency within a preset range that matches the neutral point grounding method of the distribution network to reconstruct the signal, and obtain the transient oscillation reconstruction signal.

[0047] S3: Extract four-dimensional feature parameters from the transient oscillation reconstruction signal, including the corrected transient peak amplitude, instantaneous dominant oscillation frequency, logarithmic decay rate, and transient energy entropy, and construct a four-dimensional transient feature vector;

[0048] S4: Based on the four-dimensional feature parameters, construct the fault distance estimation factor, grounding resistance estimation factor, and oscillation characteristic estimation factor. Input the three estimation factors into the fuzzy logic inference engine optimized by the optimization algorithm. Through the fuzzy logic inference engine, dynamically match the weight coefficients of the criteria corresponding to the four-dimensional feature parameters. After normalizing the four-dimensional feature parameters, calculate the comprehensive fault confidence index of each line by combining the weight coefficients.

[0049] S5: Calculate the polarity correlation coefficient between the first half-wave of the transient zero-sequence current of each line and the first half-wave of the zero-sequence voltage of the bus. Combine the comprehensive fault confidence index and the polarity correlation coefficient to construct the final discrimination function. Execute the fault line selection logic based on the final discrimination function value and the preset threshold, and output the fault line identification result.

[0050] Furthermore, the start-up threshold mentioned in step S1 is set to 15% of the phase voltage of the distribution network.

[0051] Furthermore, the initial value of the adaptive time window in step S1 is 20ms, which is dynamically extended according to the detected transient oscillation period of the zero-sequence current, with the extension range being 1 to 2 times the oscillation period. The logic of the dynamic extension is as follows: the transient oscillation period of the zero-sequence current is detected by Fourier transform, and the window is extended by adding 1 to 2 times the amplitude of the oscillation period to the initial value of 20ms.

[0052] Furthermore, in step S2, the collected zero-sequence current is decomposed into several intrinsic mode components by variational mode decomposition algorithm, and the optimal number of modes is determined adaptively by permutation entropy; the logic for determining the optimal number of modes is as follows: calculate the permutation entropy of each intrinsic mode component under different number of modes, and select the number of modes with the smallest permutation entropy value as the optimal decomposition number.

[0053] Furthermore, the preset center frequency range mentioned in step S2 is adaptively adjusted according to the neutral point grounding method of the distribution network. The specific adjustment rules are as follows: in a distribution network where the neutral point is grounded through an arc suppression coil, an intrinsic mode component with a center frequency of not less than 150Hz is selected; in a distribution network where the center point is not grounded through an arc suppression coil, an intrinsic mode component with a center frequency in the range of 200Hz to 3000Hz is selected.

[0054] Furthermore, the specific calculation method for the four-dimensional feature parameters mentioned in step S3 is as follows:

[0055] (1) Correcting transient peak amplitude: Select the maximum absolute value of the reconstructed signal as the original peak amplitude, calculate the amplitude correction coefficient, the amplitude correction coefficient is the ratio of the original peak amplitude to the average value of 5 consecutive peaks, and the product of the original peak amplitude and the amplitude correction coefficient is the corrected transient peak amplitude;

[0056] (2) Instantaneous dominant oscillation frequency: The instantaneous phase of the reconstructed signal is obtained by using Hilbert transform. The ratio of the first derivative of the instantaneous phase with respect to time to 2π is taken as the instantaneous frequency. The average value of all instantaneous frequencies is selected as the instantaneous dominant oscillation frequency.

[0057] (3) Logarithmic decay rate: It is calculated by the average of the natural logarithm ratios of consecutive peaks, and the number of consecutive peaks is 8 to 10.

[0058] (4) Transient energy entropy: It is calculated based on the information theory entropy value theory. First, the total energy of the reconstructed signal is calculated, the total energy is divided into 10 to 20 intervals and the energy ratio of each interval is calculated, and then the transient energy entropy is obtained by the entropy value calculation method.

[0059] Furthermore, the specific construction method of the three estimation factors mentioned in step S4 is as follows:

[0060] (1) Fault distance estimation factor: It is calculated based on the instantaneous dominant oscillation frequency. It is obtained by taking the near-end fault reference oscillation frequency as the reference and the ratio of the difference between the instantaneous dominant oscillation frequency and the reference oscillation frequency to the reference oscillation frequency. The range of the fault distance estimation factor is 0 to 1.

[0061] (2) Grounding resistance estimation factor: It is calculated based on the logarithmic decay rate. It is obtained by taking the low-resistance fault reference decay rate and the high-resistance fault maximum decay rate as the reference, and by the ratio of the difference between the logarithmic decay rate and the low-resistance fault reference decay rate to the difference between the high-resistance fault maximum decay rate and the low-resistance fault reference decay rate. The value range of the grounding resistance estimation factor is 0 to 1.

[0062] (3) Oscillation characteristic estimation factor: is the ratio of transient oscillation duration to adaptive time window. The transient oscillation duration is the time interval length where the absolute value of the reconstructed signal is not less than 0.1 times the corrected transient peak amplitude. The oscillation characteristic estimation factor is obtained by normalizing the ratio.

[0063] Furthermore, the reasoning process of the fuzzy logic inference engine in step S4 is as follows: first, the three estimation factors are fuzzified, then reasoning is performed according to the preset fuzzy rules, and finally the dynamic weight coefficients of each feature criterion are output through defuzzification. The fuzzy rules follow the principle that amplitude weight is high when the near end is low and the oscillation is obvious, and energy and frequency weight is high when the far end is high and the oscillation is weak.

[0064] Furthermore, the optimization algorithm described in step S4 is a particle swarm optimization algorithm, and its optimization logic is as follows: using the fault identification accuracy of historical fault data as the fitness function, iteratively optimizing the Gaussian membership function parameters of the fuzzy logic inference engine, wherein the Gaussian membership function parameters include the mean and variance, and iteratively optimizing until the fault identification accuracy tends to stabilize.

[0065] Furthermore, step S5 specifically includes the following steps:

[0066] S51: The polarity correlation coefficient is the Pearson correlation coefficient; the calculation takes half of the transient oscillation period as the time range, and uses the average value of the first half wave of the transient zero-sequence current reconstruction signal of each line and the average value of the first half wave of the bus zero-sequence voltage as the calculation benchmark, and obtains it through the Pearson correlation calculation method; among which, the Pearson correlation coefficient of the faulty line is less than 0, and the Pearson correlation coefficient of the non-faulty line is greater than 0.

[0067] S52: The final discriminant function is the product of the comprehensive fault confidence index and the corresponding expression of the Pearson correlation coefficient;

[0068] S53: The fault selection logic is as follows: The preset threshold is determined by simulation and field data calibration, and is set to 0.6; the final discrimination function value of all lines is traversed to obtain the maximum value; if the maximum value is greater than the preset threshold and the Pearson correlation coefficient of the corresponding line is less than 0, then the line is determined to be a faulty line and a trip output is triggered; if the Pearson correlation coefficient of all lines is greater than 0, then a bus fault is determined, an alarm signal is issued and the trip output is blocked; if the maximum value of the final discrimination function value is not greater than the preset threshold, then the fault characteristics are not obvious, a fault warning signal is issued and the distribution network status is continuously monitored.

[0069] Furthermore, in step S2, the acquired zero-sequence current is decomposed into several intrinsic mode components using an empirical mode decomposition algorithm, an ensemble empirical mode decomposition algorithm, or an intrinsic timescale decomposition algorithm.

[0070] Furthermore, the optimization algorithm mentioned in step S4 is a genetic algorithm, a simulated annealing algorithm, or a neural network optimization algorithm.

[0071] Furthermore, the polarity correlation coefficient mentioned in step S5 is the Spearman correlation coefficient or the Kendall rank correlation coefficient. Further, in another embodiment, step 1: fault triggering and adaptive synchronization sampling

[0072] The zero-sequence voltage of the distribution network bus is monitored in real time. When the absolute value of the zero-sequence voltage of the bus is greater than the set start threshold, each outgoing terminal is triggered to synchronously collect the zero-sequence current within the adaptive time window after the fault. The adaptive time window is dynamically extended according to the transient oscillation period of the detected zero-sequence current.

[0073] Fault Trigger Determination

[0074] A bus zero-sequence voltage triggering threshold model is constructed. In this embodiment, a start-up threshold is set. Real-time monitoring of zero-sequence voltage of distribution network bus ,when When a single-phase ground fault is detected in the distribution network, a sampling command is triggered.

[0075] in, This is the threshold value for initiating a single-phase ground fault in the distribution network. For real-time monitoring of the zero-sequence voltage of the distribution network bus; It is the absolute value of the zero-sequence voltage of the bus.

[0076] Synchronous sampling control

[0077] The high-speed synchronous acquisition and control model with GPS / BeiDou microsecond-level timing is activated, triggering each outgoing intelligent terminal to operate at a sampling frequency. Synchronous acquisition of zero-sequence current Initialize the adaptive time window The transient oscillation period of zero-sequence current is detected by Fourier transform. , and according to The time window is dynamically extended, with a maximum length not exceeding 100ms, to ensure complete capture of transient oscillation processes.

[0078] in, Adaptive time window length; The transient oscillation period of the zero-sequence current detected by Fourier transform; For the first The raw zero-sequence current signal is synchronously acquired from each outgoing line. It is a time variable.

[0079] Data output

[0080] The raw zero-sequence current signals of each line were collected. The signal is transmitted to the VMD signal processing module.

[0081] Step 2: Adaptive Decomposition and Reconstruction of Zero-Sequence Current Signal Based on VMD

[0082] The collected zero-sequence current is decomposed into several intrinsic mode components. The optimal number of modes is adaptively determined. The center frequency and arrangement entropy value of each intrinsic mode component are calculated. The intrinsic mode component with the smallest arrangement entropy value and center frequency within a preset range that matches the neutral grounding method of the distribution network is selected to reconstruct the signal and obtain the transient oscillation reconstruction signal.

[0083] VMD algorithm initialization

[0084] The original signal of zero-sequence current Input the VMD algorithm and construct the VMD decomposition mathematical model:

[0085] ;

[0086] in, For the first The zero-sequence current raw signal is input to the VMD algorithm via a single line; The number of modes in the VMD decomposition is used to adaptively determine the optimal value. The index of the modal component is 1, and its value ranges from 1 to 1. ; For the first The first line obtained after zero-sequence current decomposition One intrinsic mode component; This is the residual signal after VMD decomposition.

[0087] Adaptive determination of the optimal number of modes

[0088] Set the number of modalities and the traversal range Calculate different The average permutation entropy of each intrinsic mode component under the given value is selected, and the value with the smallest average permutation entropy is chosen. The value is used as the optimal decomposition mode number; the smaller the permutation entropy, the stronger the regularity of the component signals and the more obvious the transient fault characteristics.

[0089] Modal component screening and reconstruction

[0090] The adaptive modal component selection algorithm based on grounding method is activated to calculate the center frequency of each intrinsic mode component. Differential screening rules are applied based on the neutral grounding method of the distribution network:

[0091] Arc suppression coil grounding system: screening The high-frequency modal components avoid the compensation frequency band of the fundamental wave by the arc suppression coil;

[0092] Non-arc suppression coil grounding system (neutral point ungrounded / grounded through a small resistor): Screening The modal components avoid the 50Hz power frequency interference zone and the high-frequency random noise band.

[0093] The filtered modal components are selected for signal reconstruction to obtain a pure transient oscillation reconstructed signal. It filters out power frequency steady-state components, high-frequency random noise, and invalid interference components.

[0094] Data output

[0095] Reconstruct the signal Transmitted to the feature fusion calculation module.

[0096] Step 3: Extraction of multi-dimensional feature parameters of fault transient state

[0097] The four core feature parameters of the transient oscillation reconstruction signal are extracted: corrected transient peak amplitude, instantaneous dominant oscillation frequency, logarithmic decay rate, and transient energy entropy. At the same time, the transient oscillation coefficient auxiliary feature is extracted. After all features are normalized, a transient feature vector is constructed.

[0098] To reconstruct the signal Based on this, the calculation of each feature parameter is completed, and finally the feature vector is constructed. The specific calculation process is as follows:

[0099] Corrected transient peak amplitude and normalization

[0100] Original peak value calculation:

[0101] ;

[0102] Amplitude correction factor calculation:

[0103] ;

[0104] in, This represents the average of five consecutive peak values ​​during the transient process of the reconstructed signal. This is the amplitude correction factor, used to correct amplitude deviations caused by single-peak disturbances.

[0105] Corrected peak value calculation:

[0106] ;

[0107] Normalization process:

[0108] ;

[0109] in, For the first The original peak amplitude of the signal reconstructed from the transient oscillation of the line; For the first The corrected transient peak amplitude of the reconstructed signal of the line; This is the normalized corrected transient peak amplitude; , These represent the maximum and minimum values ​​of the corrected transient peak amplitude for all lines in the distribution network.

[0110] Instantaneous dominant oscillation frequency and frequency difference normalization

[0111] right Perform Hilbert transform to obtain the analytic signal, and extract the instantaneous phase. Calculate the instantaneous frequency:

[0112] ;

[0113] Calculate the instantaneous dominant oscillation frequency:

[0114] ;

[0115] Its theoretical model satisfies ,in For the zero-sequence equivalent inductance of the distribution network, This refers to the total capacitance to ground of the distribution network lines.

[0116] Frequency difference normalization processing:

[0117] ;

[0118] In this embodiment, a near-end fault reference frequency is set. .

[0119] in, For the first The instantaneous frequency of the reconstructed signal of the line; For the first The instantaneous dominant oscillation frequency of the reconstructed signal of the line; For the first Normalized value of frequency difference of each line; It is the absolute value of the difference between the reference frequency and the dominant frequency of the line.

[0120] Logarithmic decay rate

[0121] The logarithmic decay rate is calculated using the following formula. This parameter is positively correlated with the grounding resistance; the higher the grounding resistance, the better. The larger the value:

[0122] ;

[0123] in, For the first The logarithmic attenuation rate of the reconstructed signal for each line; In this embodiment, the number of consecutive peaks involved in the calculation is... Take 8~10; The peak index is 1 to 1. ; For the first step in the transient process of reconstructing the signal One peak; For the first step in the transient process of reconstructing the signal A number of adjacent peaks.

[0124] Transient energy entropy and normalization

[0125] The total transient energy of the reconstructed signal within the time window is calculated as follows:

[0126] ;

[0127] Total energy Divided into equal parts Calculate the energy in each interval. and energy percentage:

[0128] ;

[0129] Calculate the original transient energy entropy value:

[0130] ;

[0131] Normalization process:

[0132] ;

[0133] The faulty line has concentrated energy. Smaller; energy is dispersed in non-faulty lines. Larger.

[0134] in, For the first The total transient energy of the reconstructed signal of a line within the time window; For the first The energy percentage of each energy range; After the total energy is divided equally, the first Transient energy in each interval; For the first The original transient energy entropy value of the reconstructed signal of the line; The normalized transient energy entropy; , These represent the maximum and minimum values ​​of the original transient energy entropy of all lines in the distribution network, respectively.

[0135] Transient oscillation coefficient and normalization

[0136] Calculate the duration of transient oscillation:

[0137] ;

[0138] Calculate the original transient oscillation coefficients:

[0139] ;

[0140] Normalization process:

[0141] ;

[0142] In this embodiment, the faulty line Non-faulty lines .

[0143] in, The duration of the transient oscillation is the length of time during which the amplitude determination condition is met. The amplitude threshold coefficient for determining the duration of oscillation; For the first The original transient oscillation coefficients of the line; These are the normalized transient oscillation coefficients; , These represent the maximum and minimum values ​​of the original transient oscillation coefficients for all lines in the distribution network.

[0144] Feature output

[0145] All normalized feature parameters are transmitted to the weight calculation unit of the feature fusion calculation module.

[0146] Step 4: Dynamic weight matching and comprehensive fault confidence calculation based on fuzzy logic reasoning

[0147] Based on the four-dimensional core feature parameters, a fault distance estimation factor, a grounding resistance estimation factor, and an oscillation characteristic estimation factor are constructed. The three estimation factors are input into a fuzzy logic inference engine optimized by an optimization algorithm. The fuzzy logic inference engine dynamically matches the weight coefficients of the criteria corresponding to each feature parameter, and the comprehensive fault confidence index of each line is obtained by weighted calculation based on the weight coefficients.

[0148] This step constructs a fault distance estimation factor based on four-dimensional core features. Grounding resistance estimation factor Oscillation characteristic estimation factor Three input variables are used to optimize the membership function parameters of the fuzzy logic inference engine using the Particle Swarm Optimization (PSO) algorithm, and the magnitude weights are dynamically output. Frequency weight Energy weight Oscillation coefficient weight Finally, the comprehensive fault confidence index is calculated. The specific implementation process is as follows:

[0149] Constructing three estimation factors

[0150] The range of values ​​for the three estimation factors is 100%. This comprehensively characterizes the distance, grounding resistance, and oscillation characteristics of the fault. The specific calculation formula is as follows:

[0151] Fault distance estimation factor Based on the instantaneous dominant oscillation frequency, the farther the fault distance, The bigger

[0152] ;

[0153] Grounding resistance estimation factor Based on the logarithmic decay rate, the higher the grounding resistance, The larger;

[0154] ;

[0155] Oscillation characteristic estimation factor Based on transient oscillation coefficients, the more pronounced the oscillation characteristics, the better. The bigger

[0156] ;

[0157] PSO-optimized fuzzy logic inference engine

[0158] Membership function selection: The Gaussian membership function is selected as the basic membership function of the fuzzy inference engine, and the formula is as follows:

[0159] ;

[0160] in, Let be the mean of the Gaussian function. Let be the variance of the Gaussian function.

[0161] PSO optimization objective: Using the fault identification accuracy of historical fault data as the fitness function, iteratively optimize the Gaussian membership function. and The parameters are adjusted until the recognition accuracy stabilizes (≥99%), thus solving the problem that traditional fuzzy inference parameters rely on human experience.

[0162] Fuzzy rule base establishment: Fuzzy rules are established based on optimization results and engineering experience. The rules follow the core principle of "high amplitude weight for obvious near-end low-resistance oscillations and high energy / frequency weight for weak far-end high-resistance oscillations". Examples of core rules are as follows:

[0163] Rule ①: If Small and Small and Large, then high, Low, Low, Low;

[0164] Rule ②: If large and large and Small, then Low, middle, high, high;

[0165] Rule 3: If and and In the middle, then middle, middle, middle, middle.

[0166] Dynamic weight calculation

[0167] Will , , The optimized fuzzy logic inference engine is input, and after fuzzification, rule-based reasoning, and defuzzification, it outputs weight coefficients. , , , The weights satisfy the constraints:

[0168] ;

[0169] Calculation of Comprehensive Fault Confidence Index

[0170] Calculate the comprehensive fault confidence index for each line using the following formula. , The higher the value, the higher the probability that the line is faulty.

[0171] ;

[0172] Output Results

[0173] Each line Transmitted to the intelligent decision engine.

[0174] Step 5: Fault Route Selection Decision Based on Polarity-Confidence Dual Constraints

[0175] Calculate the polarity correlation coefficient between the first half-wave of the transient zero-sequence current of each line and the first half-wave of the zero-sequence voltage of the bus. Combine the comprehensive fault confidence index and the polarity correlation coefficient to construct the final discrimination function. Execute the fault line selection logic based on the final discrimination function value and the preset threshold, and output the fault line identification result.

[0176] This step combines physical polarity constraints (the zero-sequence current of the faulty line and the zero-sequence voltage of the busbar have opposite polarities) and statistical energy constraints (comprehensive fault confidence) to construct a dual-constraint discrimination model, achieving accurate identification of faulty lines. The specific implementation process is as follows:

[0177] Calculate the Pearson polarity correlation coefficient

[0178] Within the first half-wave time window after the fault, the first... Transient zero-sequence current after VMD reconstruction of the line With bus zero-sequence voltage Calculate the Pearson polarity correlation coefficient:

[0179] ;

[0180] in, This refers to the number of sampling points within the first half-wave time window; This is the average value of the zero-sequence current sequence; It is the average value of the zero-sequence voltage sequence of the bus.

[0181] Core judgment rule: faulty circuit (Current and voltage have opposite polarities), non-faulty circuit (Current and voltage have the same polarity).

[0182] Construct the final discriminant function

[0183] The final discriminant function is constructed using the following formula to achieve a deep fusion of polarity physical constraints and confidence statistical constraints:

[0184] ;

[0185] The core function of this function: faulty circuit... , It will be magnified The numerical values ​​enhance the fault characteristics; non-faulty lines are due to , It will shrink The value helps to suppress the risk of misjudgment.

[0186] Fault route selection decision execution

[0187] This embodiment sets a fault detection threshold. The route selection decision algorithm is initiated, and the following logic is executed:

[0188] Traverse all routes Find the maximum value ;

[0189] like And the corresponding line If the line is determined to be faulty, the intelligent decision engine will trigger a trip output, and simultaneously output the estimated fault distance and grounding resistance.

[0190] If all lines If a busbar fault is detected, the intelligent decision engine will issue an audible and visual alarm signal and simultaneously block the trip output to prevent false tripping.

[0191] like If the fault characteristics are not obvious, a fault warning signal is issued, and the operating status of the distribution network is continuously monitored.

[0192] Results Feedback

[0193] The fault identification results and tripping commands / alarm signals are transmitted to the distribution network automation monitoring platform. At the same time, the fault data is stored in the historical database of the self-learning optimization module for iterative optimization of the fuzzy inference engine.

[0194] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. These embodiments are merely descriptions of preferred embodiments and are not intended to limit the scope or concept of the invention. The specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. Such combinations, as long as they do not violate the spirit of the present invention, should also be considered as part of this disclosure. To avoid unnecessary repetition, the present invention will not further describe the various possible combinations.

[0195] This invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this invention and without departing from the design idea of ​​this invention, all modifications and improvements made by those skilled in the art to the technical solutions of this invention should fall within the protection scope of this invention. The technical content for which protection is sought in this invention has been fully described in the claims.

Claims

1. A method for identifying faulty lines in a distribution network based on adaptive fusion of zero-sequence current transient oscillation characteristics, characterized in that, Includes the following steps: S1: Real-time monitoring of the zero-sequence voltage of the distribution network bus. When the absolute value of the zero-sequence voltage of the bus is greater than the set start threshold, each outgoing terminal is triggered to synchronously collect the zero-sequence current within the adaptive time window after the fault. The adaptive time window is dynamically extended according to the transient oscillation period of the detected zero-sequence current. S2: Decompose the collected zero-sequence current into several intrinsic mode components, determine the optimal number of modes, calculate the center frequency and arrangement entropy value of each intrinsic mode component, select the intrinsic mode component with the smallest arrangement entropy value and center frequency within a preset range that matches the neutral point grounding method of the distribution network to reconstruct the signal, and obtain the transient oscillation reconstruction signal. S3: Extract four-dimensional feature parameters from the transient oscillation reconstruction signal, including the corrected transient peak amplitude, instantaneous dominant oscillation frequency, logarithmic decay rate, and transient energy entropy, and construct a four-dimensional transient feature vector; S4: Based on the four-dimensional feature parameters, construct the fault distance estimation factor, grounding resistance estimation factor, and oscillation characteristic estimation factor. Input the three estimation factors into the fuzzy logic inference engine optimized by the optimization algorithm. Through the fuzzy logic inference engine, dynamically match the weight coefficients of the criteria corresponding to the four-dimensional feature parameters. After normalizing the four-dimensional feature parameters, calculate the comprehensive fault confidence index of each line by combining the weight coefficients. S5: Calculate the polarity correlation coefficient between the first half-wave of the transient zero-sequence current of each line and the first half-wave of the zero-sequence voltage of the bus. Combine the comprehensive fault confidence index and the polarity correlation coefficient to construct the final discrimination function. Execute the fault line selection logic based on the final discrimination function value and the preset threshold, and output the fault line identification result.

2. The method for identifying faulty lines in a distribution network based on adaptive fusion of zero-sequence current transient oscillation characteristics according to claim 1, characterized in that, The start threshold in step S1 is set to 15% of the phase voltage of the distribution network.

3. The method for identifying faulty lines in a distribution network based on adaptive fusion of zero-sequence current transient oscillation characteristics according to claim 1, characterized in that, The initial value of the adaptive time window in step S1 is 20ms. It is dynamically extended according to the detected transient oscillation period of the zero-sequence current, and the extension range is 1 to 2 times the oscillation period. The logic of the dynamic extension is as follows: the transient oscillation period of the zero-sequence current is detected by Fourier transform, and the window is extended by adding 1 to 2 times the oscillation period to the initial value of 20ms.

4. The method for identifying faulty lines in a distribution network based on adaptive fusion of zero-sequence current transient oscillation characteristics according to claim 1, characterized in that, In step S2, the collected zero-sequence current is decomposed into several intrinsic mode components by variational mode decomposition algorithm, and the optimal number of modes is determined adaptively by permutation entropy. The logic for determining the optimal number of modes is as follows: calculate the permutation entropy of each intrinsic mode component under different number of modes, and select the number of modes with the smallest permutation entropy value as the optimal decomposition number.

5. The method for identifying faulty lines in a distribution network based on adaptive fusion of zero-sequence current transient oscillation characteristics according to claim 4, characterized in that, The preset center frequency range mentioned in step S2 is adaptively adjusted according to the neutral point grounding method of the distribution network. The specific adjustment rules are as follows: in the distribution network with the neutral point grounded through the arc suppression coil, the intrinsic mode component with a center frequency of not less than 150Hz is selected; in the distribution network without the arc suppression coil grounding, the intrinsic mode component with a center frequency in the range of 200Hz to 3000Hz is selected.

6. The method for identifying faulty lines in a distribution network based on adaptive fusion of zero-sequence current transient oscillation characteristics according to claim 1, characterized in that, The specific calculation method for the four-dimensional feature parameters mentioned in step S3 is as follows: (1) Correcting transient peak amplitude: Select the maximum absolute value of the reconstructed signal as the original peak amplitude, calculate the amplitude correction coefficient, the amplitude correction coefficient is the ratio of the original peak amplitude to the average value of 5 consecutive peaks, and the product of the original peak amplitude and the amplitude correction coefficient is the corrected transient peak amplitude; (2) Instantaneous dominant oscillation frequency: The instantaneous phase of the reconstructed signal is obtained by using Hilbert transform. The ratio of the first derivative of the instantaneous phase with respect to time to 2π is taken as the instantaneous frequency. The average value of all instantaneous frequencies is selected as the instantaneous dominant oscillation frequency. (3) Logarithmic decay rate: It is calculated by the average of the natural logarithm ratios of consecutive peaks, and the number of consecutive peaks is 8 to 10. (4) Transient energy entropy: It is calculated based on the information theory entropy value theory. First, the total energy of the reconstructed signal is calculated, the total energy is divided into 10 to 20 intervals and the energy ratio of each interval is calculated, and then the transient energy entropy is obtained by the entropy value calculation method.

7. The method for identifying faulty lines in a distribution network based on adaptive fusion of zero-sequence current transient oscillation characteristics according to claim 1, characterized in that, The specific construction method of the three estimation factors mentioned in step S4 is as follows: (1) Fault distance estimation factor: It is calculated based on the instantaneous dominant oscillation frequency. It is obtained by taking the near-end fault reference oscillation frequency as the reference and the ratio of the difference between the instantaneous dominant oscillation frequency and the reference oscillation frequency to the reference oscillation frequency. The range of the fault distance estimation factor is 0 to 1. (2) Grounding resistance estimation factor: It is calculated based on the logarithmic decay rate. It is obtained by taking the low-resistance fault reference decay rate and the high-resistance fault maximum decay rate as the reference, and by the ratio of the difference between the logarithmic decay rate and the low-resistance fault reference decay rate to the difference between the high-resistance fault maximum decay rate and the low-resistance fault reference decay rate. The value range of the grounding resistance estimation factor is 0 to 1. (3) Oscillation characteristic estimation factor: is the ratio of transient oscillation duration to adaptive time window. The transient oscillation duration is the time interval length where the absolute value of the reconstructed signal is not less than 0.1 times the corrected transient peak amplitude. The oscillation characteristic estimation factor is obtained by normalizing the ratio.

8. The method for identifying faulty lines in a distribution network based on adaptive fusion of zero-sequence current transient oscillation characteristics according to claim 7, characterized in that, The reasoning process of the fuzzy logic inference engine in step S4 is as follows: First, the three estimation factors are fuzzified, then reasoning is performed according to the preset fuzzy rules, and finally the dynamic weight coefficients of each feature criterion are output through defuzzification. The fuzzy rules follow the principle that when the near end is low-resistance and the oscillation is obvious, the amplitude weight is high, and when the far end is high-resistance and the oscillation is weak, the energy and frequency weights are high.

9. The method for identifying faulty lines in a distribution network based on adaptive fusion of zero-sequence current transient oscillation characteristics according to claim 8, characterized in that, The optimization algorithm described in step S4 is the particle swarm optimization algorithm. Its optimization logic is as follows: using the fault identification accuracy of historical fault data as the fitness function, iteratively optimize the Gaussian membership function parameters of the fuzzy logic inference engine. The Gaussian membership function parameters include the mean and variance. Iterative optimization continues until the fault identification accuracy tends to stabilize.

10. The method for identifying faulty lines in a distribution network based on adaptive fusion of zero-sequence current transient oscillation characteristics according to claim 1, characterized in that, Step S5 specifically includes the following steps: S51: The polarity correlation coefficient is the Pearson correlation coefficient; the calculation takes half of the transient oscillation period as the time range, and uses the average value of the first half wave of the transient zero-sequence current reconstruction signal of each line and the average value of the first half wave of the bus zero-sequence voltage as the calculation benchmark, and obtains it through the Pearson correlation calculation method; among which, the Pearson correlation coefficient of the faulty line is less than 0, and the Pearson correlation coefficient of the non-faulty line is greater than 0. S52: The final discriminant function is the product of the comprehensive fault confidence index and the corresponding expression of the Pearson correlation coefficient; S53: The fault selection logic is as follows: The preset threshold is determined by simulation and field data calibration, and is set to 0.6; the final discrimination function value of all lines is traversed to obtain the maximum value; if the maximum value is greater than the preset threshold and the Pearson correlation coefficient of the corresponding line is less than 0, then the line is determined to be a faulty line and a trip output is triggered; if the Pearson correlation coefficient of all lines is greater than 0, then a bus fault is determined, an alarm signal is issued and the trip output is blocked; if the maximum value of the final discrimination function value is not greater than the preset threshold, then the fault characteristics are not obvious, a fault warning signal is issued and the distribution network status is continuously monitored.

Images