Fault monitoring management method and system for 35kV wind power plant overhead current collection line

Abstract

The invention provides a fault monitoring management method and system for a 35kV wind power plant overhead current collection line. The method comprises the following steps: acquiring and simplifying a current collection line diagram into a topological structure; and acquiring fault traveling wave data through the fault data detection terminal. And decoupling fault traveling wave data by using a Karen-Boolean transform method. And adopting a mode decomposition algorithm to obtain traveling wave mode component signals of different frequency domains. And calculating a signal energy spectrum based on an NTOO energy operator, and calibrating an initial wave head of the fault traveling wave. And calculating a fault coefficient according to the topological structure and the initial wave head, and positioning a fault point. Energy spectrum features are extracted, and fault types are identified through a deep learning model. And generating a monitoring report containing the fault point and type. The method has the effects of accurately positioning the fault point in the current collection line of the wind power plant and accurately predicting the fault type.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind power line fault management, and particularly relates to a fault monitoring and management method and system for 35 kV overhead collector lines in a wind farm. Background Art

[0002] With the continuous growth of the global demand for clean energy, wind power generation, as an important form of renewable energy, has witnessed rapid development in recent years. The construction of large-scale wind farms has not only made important contributions to the transformation of the energy structure but also brought a series of technical challenges, especially in terms of operation and maintenance.

[0003] As the "vascular system" of a wind farm, the collector line is responsible for collecting and transmitting the electric energy generated by each wind turbine generator to the substation, and its reliability directly affects the power generation efficiency and economic benefits of the entire wind farm. Traditional fault monitoring methods for wind farm collector lines mainly rely on regular manual inspections and simple remote monitoring. These methods may be sufficient in small-scale wind farms, but with the continuous expansion of the scale and increasing complexity of wind farms, their limitations have become increasingly apparent. However, some large wind farms usually cover a vast area, and the collector lines may extend for dozens or even hundreds of kilometers and are often distributed in environments with complex terrains and variable climates. Manual inspections are not only time-consuming and labor-intensive but may also be unable to be carried out in a timely manner under adverse weather conditions. Therefore, in the prior art, fault diagnosis methods based on steady-state analysis are usually adopted instead of manual inspections. However, the complex topological structure of a wind farm with multiple branch nodes and the complex dynamic operating characteristics affected by weather will significantly weaken the accuracy of the steady-state analysis fault diagnosis method, ultimately making it difficult to accurately locate the line fault points in the wind farm and accurately analyze the fault types of line faults. Summary of the Invention

[0004] The present invention provides a fault monitoring and management method and system for 35 kV overhead collector lines in a wind farm to solve the problems of difficultly accurately locating the line fault points in the wind farm and accurately analyzing the fault types of line faults.

[0005] In a first aspect, the present invention provides a fault monitoring and management method for 35 kV overhead collector lines in a wind farm, and the method includes the following steps:

[0006] Obtain the collector line diagram of the wind farm collector line;

[0007] Simplify the collector line diagram into a collector line topological structure, where the collector line topological structure includes a collector main line directly connected to the wind farm power transmission system and a plurality of collector branch lines connected to the collector main line;

[0008] Obtain the fault traveling wave data in the wind farm collector line through the fault data detection terminal deployed in the main collector line;

[0009] Use the Karenbauer transform method to decouple the fault traveling wave data into fault traveling wave components;

[0010] Adopt a modal decomposition algorithm to decompose the fault traveling wave components to obtain traveling wave modal component signals in multiple different frequency domains;

[0011] Based on the NTEO energy operator, calculate the signal energy spectrum of all the traveling wave modal component signals, and calibrate the initial wave head of the fault traveling wave of the fault traveling wave data according to the signal energy spectrum of the highest frequency traveling wave modal component signal;

[0012] Based on the initial wave head of the fault traveling wave and according to the topology structure of the collector line, calculate the fault coefficients of all the collector branch lines, and combine all the fault coefficients to locate the line fault point in the topology structure of the collector line;

[0013] Extract the energy spectrum features of all the signal energy spectra, and based on the energy spectrum features, identify the fault type of the line fault point through a deep learning model;

[0014] Generate a fault monitoring report by combining the line fault point and the fault type.

[0015] Optionally, the simplification of the collector line diagram into the collector line topology structure includes the following steps:

[0016] Identify the key components in the collector line diagram and the collector lines between the key components;

[0017] Simplify all the key components into key nodes, and simplify the collector lines into key node edges between the key nodes to form multiple topological lines. The key nodes include wind turbine generator nodes, collector station nodes, fault data detection terminal nodes, and wind farm power transmission system nodes. The number of fault data detection terminal nodes is 2;

[0018] Add node edge attributes to the key node edges according to the actual length of the collector line, and add key node attributes to the wind turbine generator nodes according to the unit status information of the wind turbine generators corresponding to the wind turbine generator nodes;

[0019] For any one of the topological lines, if both ends of the topological line are the fault data detection terminal nodes, and all nodes in the topological line except the fault data detection terminal nodes are the collector station nodes, then mark the topological line as the main collector line, and all the collector station nodes are connected to the wind farm power transmission system nodes through the main collector line;

[0020] If both ends of the topological line are the substation node and the wind turbine node respectively, then mark the topological line as a collector branch line;

[0021] After all the topological lines are marked, the simplified collector line topology structure of the collector line diagram is obtained.

[0022] Optionally, the steps of decomposing the fault traveling wave component by using the modal decomposition algorithm to obtain multiple traveling wave modal component signals in different frequency domains are as follows:

[0023] Collect the wind farm meteorological information in the same time period as the fault traveling wave data through the weather monitoring module configured in the wind farm power transmission system;

[0024] Calculate the meteorological severity degree of the period when the fault traveling wave data is obtained based on the wind farm meteorological information and according to the preset meteorological condition rules;

[0025] Integrate the key node attributes of all the wind turbine nodes, and calculate the unit state complexity of all the wind turbine nodes according to all the key node attributes;

[0026] Calculate the signal smoothness of the fault traveling wave data by combining the meteorological severity degree and the unit state complexity;

[0027] If the signal smoothness is less than the preset smoothness threshold, then decompose the fault traveling wave component by using the empirical mode decomposition algorithm to obtain multiple traveling wave modal component signals in different frequency domains;

[0028] If the signal smoothness is greater than or equal to the smoothness threshold, then decompose the fault traveling wave component by using the variational mode decomposition algorithm to obtain multiple traveling wave modal component signals in different frequency domains.

[0029] Optionally, the steps of decomposing the fault traveling wave component by using the empirical mode decomposition algorithm to obtain multiple traveling wave modal component signals in different frequency domains are as follows:

[0030] Calculate the signal maximum value and the signal minimum value of the fault traveling wave component, and combine the signal maximum value and the signal minimum value and use the cubic spline interpolation method to obtain the upper envelope line and the lower envelope line of the fault traveling wave component;

[0031] Perform an average calculation on the upper envelope line and the lower envelope line to obtain the average component of the fault traveling wave component;

[0032] Take the difference between the fault traveling wave component and the average component as the preliminary IMF component;

[0033] If the number of zeros of the prepared IMF component is equal to the number of extreme values, or the average value of the upper and lower envelope lines of the prepared IMF component at any time is 0, then the prepared IMF component is used as the traveling wave mode component signal in one of the frequency domains;

[0034] Calculate the remaining signal component after subtracting the traveling wave mode component signal from the fault traveling wave component;

[0035] Use the remaining signal component as a new round of fault traveling wave components and repeat the above component decomposition steps until the component decomposition cut-off condition is met;

[0036] After completing the component decomposition of the fault traveling wave component, use the soft limit function to filter out the signal noise of all the traveling wave mode component signals in the corresponding frequency domain;

[0037] The expression formula of the component decomposition cut-off condition is as follows:

[0038]

[0039] In the formula: M represents the number of signal quantities of the traveling wave mode component signals that have completed decomposition, represents the traveling wave mode component signal obtained in the m-th round of the component decomposition step, represents the traveling wave mode component signal obtained in the (m - 1)-th round of the component decomposition step, and ε represents the cut-off threshold.

[0040] Optionally, the step of using the variational mode decomposition algorithm to decompose the fault traveling wave component to obtain multiple traveling wave mode component signals in different frequency domains includes the following steps:

[0041] Construct a decomposition constraint model for the fault traveling wave component based on the decomposition constraint conditions of the variational mode decomposition algorithm. The decomposition constraint conditions are as follows: the deviation between the reconstructed signal and the fault traveling wave component is minimized after reconstructing the N traveling wave mode component signals obtained by decomposing the fault traveling wave component using the variational mode decomposition algorithm, and the sum of the estimated bandwidths of all the traveling wave mode component signals is minimized;

[0042] Introduce the Lagrange operator and the penalty factor into the decomposition constraint model to construct the model augmented expression of the decomposition constraint model;

[0043] Based on the model augmented expression and using the alternating direction method of the Lagrange operator, perform alternating iterative optimization in the frequency domain until the Wiener filtering of all the traveling wave modes satisfies the convergence condition. In each round of alternating iterative optimization process, a traveling wave mode component signal in a different frequency domain is decomposed from the fault traveling wave component;

[0044] The expression formula of the convergence condition is as follows:

[0045]

[0046] where: K represents the number of signal of the traveling wave mode component signals that have been decomposed, represents the Wiener filter of the traveling wave mode component signal obtained during the s-th round of alternating iterative optimization, represents the Wiener filter of the traveling wave mode component signal obtained during the (s + 1)-th round of alternating iterative optimization, ||·||2 represents the Euclidean norm, and ζ represents the convergence accuracy.

[0047] Optionally, calculating the signal energy spectra of all the traveling wave mode component signals based on the NTEO energy operator, and calibrating the initial wave head of the fault traveling wave of the fault traveling wave data according to the signal energy spectrum of the highest-frequency traveling wave mode component signal includes the following steps:

[0048] Convert all the traveling wave mode component signals into discrete component signals;

[0049] Use the NTEO energy operator to calculate the signal energy spectra of each of the discrete component signals respectively. The specific calculation formula of the signal energy spectrum is as follows:

[0050]

[0051] where: Ψ[h i (x)] represents the signal energy spectrum of the i-th discrete component signal h i (x), T represents the resolution parameter, and M represents the number of signals of the discrete component signals;

[0052] Select the signal energy spectrum corresponding to the traveling wave mode component signal in the highest frequency domain as the target signal energy spectrum, and calibrate the time point with the highest energy value in the target signal energy spectrum as the initial wave head of the fault traveling wave of the fault traveling wave data.

[0053] Optionally, the fault traveling wave components include a fault traveling wave line mode component and a fault traveling wave zero mode component. The fault traveling wave line mode component and the fault traveling wave zero mode component share the initial wave head of the fault traveling wave. Calculating the fault coefficients of all the collector branch lines based on the initial wave head of the fault traveling wave and according to the collector line topology structure, and locating the line fault point in the collector line topology structure includes the following steps:

[0054] Take the fault data detection terminal node closest to the wind farm power transmission system node in the collector line topology structure as the starting node of the main line, and take the fault data detection terminal node farthest from the wind farm power transmission system node as the ending node of the main line;

[0055] Define the direction from the starting node of the main line to the ending node of the main line as the positive direction of the traveling wave;

[0056] Combined with the initial wavefront of the fault traveling wave and the first propagation speed of the line mode component of the fault traveling wave, calculate respectively the first propagation time for the line mode component of the fault traveling wave to propagate to the starting node of the main line, and the second propagation time for the line mode component of the fault traveling wave to propagate to the ending node of the main line;

[0057] Combined with the initial wavefront of the fault traveling wave and the second propagation speed of the zero mode component of the fault traveling wave, calculate respectively the third propagation time for the zero mode component of the fault traveling wave to propagate to the starting node of the main line, and the fourth propagation time for the zero mode component of the fault traveling wave to propagate to the ending node of the main line;

[0058] Combined with the first propagation time, the second propagation time, the third propagation time, the fourth propagation time and the first propagation speed, calculate respectively the first fault distance between the imaginary fault point and the starting node of the main line, and the second fault distance between the imaginary fault point and the ending node of the main line;

[0059] Combined with the first fault distance, the second fault distance and the node-edge attribute in the collector line topology structure, and calculate according to the positive direction of the traveling wave to obtain the fault coefficients of all the collector branch lines;

[0060] If the fault coefficient of any one of the collector branch lines is 0, it is determined that the line fault point generating the fault traveling wave data is located in the collector branch line with the fault coefficient of 0, and the midpoint of the faulty collector branch line is used as the line fault point in the collector line topology structure;

[0061] If all the fault coefficients are not 0, it is determined that the line fault point is located in the collector main line, calculate by combining the initial wavefront of the fault traveling wave, the propagation speed of the fault traveling wave component and the node-edge attribute, and use the double-ended traveling wave location method to determine the line fault point in the collector line topology structure.

[0062] Optionally, the calculation formula of the first fault distance is as follows:

[0063]

[0064] In the formula: represents the first fault distance between the imaginary fault point G and the starting node Q1 of the main line, t1 represents the first propagation time, t2 represents the second propagation time, t3 represents the third propagation time, t4 represents the fourth propagation time, and v1 represents the first propagation speed;

[0065] The calculation formula for the second fault distance is as follows:

[0066]

[0067] In the formula: represents the second fault distance between the imaginary fault point G and the termination node Q of the main line n therebetween.

[0068] Optionally, the steps of extracting the energy spectrum features of all the signal energy spectra, identifying the fault type of the line fault point based on the energy spectrum features through a deep learning model include the following:

[0069] For the signal energy spectrum of each of the traveling wave mode component signals, energy spectrum segments of different time lengths are intercepted from the signal energy spectrum through time windows of different time scales, and energy time domain features are extracted from each of the energy spectrum segments respectively. The energy time domain features include energy peak value, energy decay rate, energy rise time, energy duration, and energy phase difference feature;

[0070] Statistical methods are used to extract the energy frequency domain features of the signal energy spectrum in the frequency domain. The energy frequency domain features include energy ratio, energy mean, energy variance, and energy kurtosis;

[0071] Correlation analysis method is used to screen out an energy feature subset with fault type discriminative power from the energy time domain features and the energy frequency domain features;

[0072] The energy feature subset is input into a pre-trained energy feature recognition model, and the fault type of the line fault point is output through the energy feature recognition model. The energy feature recognition model is a convolutional neural network model including a multi-head attention mechanism.

[0073] In a second aspect, the present invention also provides a fault monitoring and management system for a 35 kV wind farm overhead collector line, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the fault monitoring and management method for the 35 kV wind farm overhead collector line as described in the first aspect is implemented.

[0074] The beneficial effects of the present invention are:

[0075] The present invention lays a solid foundation for subsequent fault location and analysis by obtaining and simplifying the collector line diagram into a topological structure. This method can better reflect the actual structure of the collector lines in a wind farm and improve the accuracy of fault location. Advanced signal processing techniques, such as the Karenbauer transform method and the modal decomposition algorithm, are adopted to effectively decouple and decompose the fault traveling wave data, thereby extracting more accurate fault characteristic information. In particular, the method of using the NTEO energy operator to calculate the signal energy spectrum and calibrate the initial wavefront of the fault traveling wave greatly improves the accuracy and reliability of fault location. In addition, the present invention innovatively combines the collector line topological structure and the fault coefficient calculation method. This method can not only accurately locate the fault point but also adapt to the complex structure of the collector lines in a wind farm, overcoming the problem of difficult fault location in complex networks by traditional methods. On the other hand, the present invention also introduces a deep learning model to identify the fault type. This method can make full use of big data and artificial intelligence technologies, continuously learn and optimize the fault identification model, and improve the accuracy and adaptability of fault type identification. Finally, an accurate fault monitoring report is generated by combining the fault location and the fault type, providing comprehensive and accurate fault information for the operation and maintenance personnel of the wind farm, helping to quickly formulate maintenance strategies, reduce downtime, and improve the overall operation efficiency of the wind farm. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a schematic flowchart of the fault monitoring and management method for the 35 kV overhead collector lines in a wind farm in one embodiment of the present application.

[0077] Figure 2 It is a schematic structural diagram of the collector line topological structure in one embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0078] The technical solutions in the embodiments of the present application will be clearly described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application belong to the scope of protection of the present application.

[0079] The terms "first", "second", etc. in the description and claims of this application are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of this application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are usually of one category, and the number of objects is not limited. For example, the first object can be one or multiple. In addition, "and / or" in the description and claims means at least one of the connected objects, and the character " / ", generally represents an "or" relationship between the associated objects before and after.

[0080] Figure 1 It is a schematic flow chart of a fault monitoring and management method for a 35kV wind farm overhead collector line in an embodiment. It should be understood that although Figure 1 the steps in the flow chart are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, Figure 1 at least a part of the steps in Figure 1 may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least a part of other steps or sub-steps or stages of other steps. As

[0081] S101. Obtain the collector line diagram of the wind farm collector line.

[0082] Among them, this step is completed by collecting the engineering design drawings, on-site survey data and geographic information system (GIS) data of the wind farm. The collector line diagram contains the location information of key equipment such as wind turbines, collector stations, and power transmission systems, as well as the route and length of the overhead lines connecting these equipment. Obtaining a complete and accurate collector line diagram is crucial for subsequent fault monitoring and location. In actual operation, professional power system drawing software can be used to draw and store the collector line diagram. This electronic collector line diagram is not only convenient for updating and management, but also can be integrated with other systems such as the SCADA (Supervisory Control and Data Acquisition) system to improve the overall operation and maintenance efficiency.

[0083] S102. Simplify the collector line diagram into a collector line topology structure.

[0084] Among them, key components in the collector line diagram are first identified, including wind turbine generators, collector substations, fault data detection terminals, and wind farm power transmission systems. Then, these key components are simplified into nodes, and the lines connecting these nodes are simplified into edges, forming a network structure composed of nodes and edges. In this structure, the lines directly connected to the wind farm power transmission system are marked as main collector lines, and the other lines connected to the main collector lines are marked as branch collector lines. Each edge is given the attribute of the actual line length, and each node representing a wind turbine generator is given the attribute of the unit status information. This simplified topological structure retains the key information of the original collector line while greatly reducing the complexity of subsequent calculations. For example, a wind farm with 50 wind turbines may have its collector line diagram simplified into a topological structure with 53 nodes (50 wind turbine nodes, 2 collector substation nodes, and 1 power transmission system node) and 52 edges.

[0085] S103. Obtain the fault traveling wave data in the wind farm collector line through the fault data detection terminals deployed in the main collector lines.

[0086] Among them, the fault data detection terminal is a high-precision measuring device installed on the main collector line. Usually using GPS clock synchronization technology, it can accurately capture the instantaneous values of voltage and current when a fault occurs. When a fault occurs on the line, a high-frequency electromagnetic transient phenomenon, that is, a fault traveling wave, will be generated at the fault point. These traveling waves propagate along the line at a speed close to the speed of light and are recorded by the fault data detection terminals. The typical sampling frequency of the fault data detection terminal can reach above 1 MHz, and it can capture signal changes at the microsecond level. The obtained fault traveling wave data usually includes the instantaneous values of three-phase voltage and current, as well as the corresponding timestamps. For example, in the case of a single-phase ground fault, the fault data detection terminal may record that the voltage of phase A suddenly drops, while the current of phase A rises sharply, and these changes are not obvious in the other two phases. These high-precision and high-time-resolution fault traveling wave data provide key information for subsequent fault location and type identification and are the basic data source for the entire fault monitoring and management method.

[0087] S104. Use the Karenbauer transform method to decouple the fault traveling wave data into fault traveling wave components.

[0088] Among them, the Karenbauer transform is a mathematical method for converting three-phase quantities into modal quantities, which can decouple the coupled three-phase fault traveling wave data into independent modal components. Specifically, the Karenbauer transform converts three-phase voltage or current data into three modal components: zero mode, positive mode, and negative mode. The transformation matrix T is usually defined as:

[0089]

[0090] Among them, a = e(j2π / 3) Assume the three-phase voltages are Va, Vb, and Vc, then the modal voltages are as follows:

[0091]

[0092] Among them, V0 is the zero-mode component, V1 is the positive-mode component, and V2 is the negative-mode component.

[0093] The advantage of this transformation is that it can decompose the complex fault information of three-phase coupling into independent modal components, simplifying the subsequent analysis process. For example, in a single-phase grounding fault, the zero-mode component will increase significantly, while in a three-phase fault, the positive-mode component will dominate. By observing the characteristics of different modal components, the type and severity of the fault can be preliminarily judged. In addition, modal decomposition can also reduce the influence of mutual inductance between phases and improve the accuracy of fault location. This step lays the foundation for subsequent modal decomposition and feature extraction and is a key link in achieving accurate fault location and type identification.

[0094] S105. Decompose the fault traveling wave components using a modal decomposition algorithm to obtain multiple traveling wave modal component signals in different frequency domains.

[0095] Among them, the purpose of this step is to further decompose the complex fault traveling wave components into sub-signals with different frequency characteristics for more refined analysis of fault characteristics. Selecting an appropriate modal decomposition algorithm is crucial for improving the accuracy of subsequent fault identification. According to the characteristics of the fault traveling wave data, the empirical mode decomposition (EMD) algorithm or the variational mode decomposition (VMD) algorithm can be selected. Specifically, during implementation, first collect meteorological information such as wind speed, temperature, and humidity in the same time period as the fault traveling wave data through the weather monitoring module of the wind farm. Then, calculate a meteorological severity index based on this meteorological information. At the same time, calculate the complexity of the unit state according to the operating state of the wind turbines (such as the number of units in operation, power output, etc.). Considering the meteorological severity and the complexity of the unit state comprehensively, obtain the signal smoothness index of the fault traveling wave data. For the monitoring of the collector line under normal operating conditions, the VMD algorithm is preferably selected. Because in this case, the signal usually contains the stable operating characteristics of multiple wind turbines, and VMD can better separate these characteristics. In case of severe weather conditions (such as thunderstorm weather) and in the presence of multiple newly connected wind farms or new types of wind power equipment, the EMD algorithm is preferably selected. Because strong transient signals may be generated in this case, and EMD is more suitable for processing such non-linear and non-stationary signals.

[0096] The EMD algorithm decomposes a signal into multiple Intrinsic Mode Functions (IMFs) and a residual term through an iterative screening process. Each IMF represents the oscillation mode of the original signal at different time scales. The VMD algorithm, on the other hand, decomposes a signal into a predetermined number of band-limited mode functions through an optimization process. Each mode function has its own central frequency and finite bandwidth. Regardless of which algorithm is used, multiple traveling wave mode component signals in different frequency domains will ultimately be obtained. These component signals respectively reflect the characteristics of the fault traveling wave in different frequency ranges, providing richer and more refined information for subsequent feature extraction and fault identification.

[0097] S106. Calculate the signal energy spectrum of all traveling wave mode component signals based on the NTEO energy operator, and calibrate the initial wavefront of the fault traveling wave data according to the signal energy spectrum of the highest-frequency traveling wave mode component signal.

[0098] Among them, the NTEO (Normalized Teager Energy Operator) energy operator is an improved energy calculation method. Compared with the traditional Teager energy operator, it has better noise suppression ability and time resolution. For each traveling wave mode component signal, first convert it into a discrete time series, and then apply the NTEO energy operator to calculate its energy spectrum. This process will obtain multiple energy spectra in different frequency domains. Next, select the energy spectrum corresponding to the highest-frequency traveling wave mode component signal as the target energy spectrum. In this energy spectrum, find the time point with the highest energy value and calibrate it as the initial wavefront of the fault traveling wave. For example, assume that the sampling frequency of the highest-frequency mode component is 1 MHz, and the energy value of the 1000th sampling point in the energy spectrum is the highest. Then it can be determined that the initial wavefront of the fault appears 1 ms after the fault occurs. The advantage of this method is that the highest-frequency mode component is usually the most sensitive to the fault starting moment, so the initial wavefront of the fault can be located more accurately. At the same time, the use of the NTEO energy operator improves the positioning accuracy in a noisy environment. Accurately calibrating the initial wavefront of the fault traveling wave is crucial for subsequent fault location, as it directly affects the accuracy of fault distance calculation.

[0099] S107. Calculate the fault coefficients of all collector branch lines based on the initial wavefront of the fault traveling wave and according to the topology structure of the collector line, and locate the line fault point in the topology structure of the collector line by combining all the fault coefficients.

[0100] Among them, the fault data detection terminal node closest to the node of the wind farm power transmission system in the collector line topology is defined as the starting node of the main line, and the farthest one is defined as the ending node of the main line. The direction from the starting node to the ending node is defined as the positive direction of the traveling wave. Then, calculate the time when the line-mode component and zero-mode component of the fault traveling wave propagate to the two terminal nodes respectively. Next, according to these distances and the node-edge attributes in the topology, calculate the fault coefficient of each collector branch line. The fault coefficient can be defined as the ratio of the distance from the imaginary fault point on this branch line to the connection point of the main line to the total length of the branch line. If there is a branch line with a fault coefficient of 0, it is determined that the fault point is located on this branch line, and the midpoint of the branch line is taken as the fault point. If all fault coefficients are not 0, it is determined that the fault point is located on the main line, and the double-ended traveling wave location method is used to determine the specific location. This method combines the topology information and the traveling wave propagation characteristics, can effectively distinguish the main branch line faults, and give a relatively accurate fault point location.

[0101] S108. Extract the energy spectrum features of all signal energy spectra, and identify the fault type of the line fault point based on the energy spectrum features and through a deep learning model.

[0102] Among them, first extract the features of the energy spectrum of each traveling wave mode component signal. In the time domain, use time windows of different sizes to intercept energy spectrum segments, and extract features such as energy peak value, energy decay rate, energy rise time, energy duration, and energy phase difference. In the frequency domain, calculate statistical features such as energy proportion, energy mean, energy variance, and energy kurtosis. For example, the energy decay rate can be obtained by fitting the decay curve after the energy peak value: E(t) = E0 * e (-αt) , where α is the decay rate. The energy phase difference can be quantified by calculating the correlation coefficient of the energy spectra between different phases. Next, use the correlation analysis method to screen out the most effective feature subset for fault type discrimination from these features. This step can significantly reduce the feature dimension, improve the training efficiency and generalization ability of the model. Finally, input the screened features into a pre-trained deep learning model. This model adopts a convolutional neural network structure containing a multi-head attention mechanism, which can effectively capture the complex relationships between energy features. The output of the model is the prediction result of the fault type.

[0103] The selection of the modal decomposition algorithm in step S105 and the improvement of the energy operator in step S106 have significantly improved the accuracy of the model in identifying fault types. By dynamically selecting the Empirical Mode Decomposition (EMD) or Variational Mode Decomposition (VMD) algorithm according to the signal stationarity, it is possible to better adapt to the signal characteristics under different fault conditions, thereby obtaining more accurate traveling wave modal component signals. These finely decomposed modal components can better reflect the characteristics of fault signals in different frequency domains, providing a high-quality data basis for subsequent feature extraction. At the same time, the Normalized Teager Energy Operator (NTEO) energy operator adopted in S106 has stronger noise suppression ability and higher time resolution compared with the traditional Teager energy operator. This improvement makes the calculation of the energy spectrum more accurate, especially in a complex electromagnetic environment, and can more accurately capture the energy change characteristics of fault traveling waves. The use of the NTEO energy operator not only improves the positioning accuracy of the initial wavehead of the fault but also provides more reliable energy spectrum data for subsequent feature extraction. The optimization of these two key steps works together to significantly enhance the model's ability to distinguish different types of faults. Especially when dealing with complex faults or multiple fault situations, it can provide more accurate and reliable identification results, thus greatly improving the performance and reliability of the entire fault monitoring and management system.

[0104] S109. Generate a fault monitoring report by combining the fault location point and the fault type.

[0105] Among them, the fault location result and the fault type identification result obtained in the foregoing steps are integrated to form a comprehensive and detailed fault monitoring report. The content of the report usually includes the following parts: Fault overview: including basic information such as the time of fault occurrence, duration, and affected range. Fault location: Describe in detail the location of the fault point, including its specific location in the topological structure (such as the main line or a certain branch line) and the distance from the nearest substation. Fault type: Explain the identified fault type (such as single-phase grounding, two-phase short circuit, three-phase short circuit, etc.) and give the judgment basis. Fault feature analysis: including technical details such as the main features of the fault traveling wave and the energy spectrum features.

[0106] In one implementation, the steps for simplifying the collector line diagram into a collector line topological structure are as follows:

[0107] Identify the key components in the collector line diagram and the collector lines between the key components;

[0108] Simplify all key components into key nodes, and simplify the collector lines into key node edges between key nodes to form multiple topological lines. The key nodes include wind turbine generator nodes, substation nodes, fault data detection terminal nodes, and wind farm power transmission system nodes. The number of fault data detection terminal nodes is 2;

[0109] Add node edge attributes to the key node edges according to the actual length of the collector lines, and add key node attributes to the wind turbine generator nodes according to the unit status information of the wind turbine generators corresponding to the wind turbine generator nodes;

[0110] For any topological line, if both ends of the topological line are fault data detection terminal nodes, and all nodes in the topological line except the fault data detection terminal nodes are substation nodes, then mark the topological line as the main collector line, and all substation nodes are connected to the wind farm power transmission system nodes through the main collector line;

[0111] If the two ends of the topological line are a substation node and a wind turbine generator node respectively, then mark the topological line as a branch collector line;

[0112] After all topological lines are marked, the collector line topological structure after simplifying the collector line diagram is obtained.

[0113] In this embodiment, the key components generally include wind turbine generators, substations, fault data detection terminals, and wind farm power transmission systems. Wind turbine generators are the main source of electrical energy. Substations are responsible for collecting and converting electrical energy. Fault data detection terminals are used to monitor the line status, and the power transmission system connects the wind farm to the external power grid. Identifying the collector lines between these components involves tracing the power transmission path, including underground cables and overhead lines. The key to this step is to accurately record the starting point, ending point, and path of each line, laying a foundation for subsequent topological structure simplification. Next, the concepts of graph theory can be used: each wind turbine generator, substation, fault data detection terminal, and power transmission system is represented as a node, and the collector lines connecting these nodes are represented as edges.

[0114] Then, it is necessary to add node-edge attributes to the edges of key nodes and key node attributes to the wind turbine generator nodes. The node-edge attributes mainly refer to the actual length of the collector line, which is crucial for subsequent fault location calculations. For example, if the line length between two substations is 5 kilometers, then the attribute value of this edge is set to 5. These length data are usually obtained from engineering drawings or GPS measurements. For wind turbine generator nodes, the added key node attributes include the operating status information of the units, such as whether it is currently operating, output power, wind turbine speed, etc. These information can be obtained in real-time from the SCADA system of the wind farm. For example, the attributes of a wind turbine node can be: {grid connection time: "2020-01-01", status: "operating", power: 2.5MW, speed: 15rpm}. These attributes not only reflect the real-time operating conditions of the wind farm but also can reflect whether each wind turbine generator is a newly grid-connected one.

[0115] Next, line identification is carried out. If both ends of a topological line are fault data detection terminal nodes and only substation nodes are included in the middle, it is marked as the main collector line. This reflects the characteristics of the main power transmission channels in the actual system. Then, the branch collector lines are identified. When one end of a topological line is a substation node and the other end is a wind turbine generator node, it is marked as a branch collector line. This represents the power transmission path from a single wind turbine to the substation. After all the lines are marked, the final collector line topology structure is obtained. This structure clearly shows the power transmission path from the wind turbines to the substations and then to the power transmission system.

[0116] In one implementation, the steps of decomposing the fault traveling wave component using the modal decomposition algorithm to obtain multiple traveling wave modal component signals in different frequency domains are as follows:

[0117] Collect the wind farm meteorological information in the same time period as the fault traveling wave data through the weather monitoring module configured in the wind farm power transmission system;

[0118] Calculate the meteorological severity of the fault traveling wave data acquisition period based on the wind farm meteorological information and according to the preset meteorological condition rules;

[0119] Integrate the key node attributes of all wind turbine generator nodes and calculate the unit state complexity of all wind turbine generator nodes based on all the key node attributes;

[0120] Calculate the signal smoothness of the fault traveling wave data by combining the meteorological severity and the unit state complexity;

[0121] If the signal smoothness is less than the preset smoothness threshold, use the empirical mode decomposition algorithm to decompose the fault traveling wave component to obtain multiple traveling wave modal component signals in different frequency domains;

[0122] If the signal stationarity is greater than or equal to the stationarity threshold, the variational mode decomposition algorithm is used to decompose the fault traveling wave component, and multiple traveling wave mode component signals in different frequency domains are obtained.

[0123] In this embodiment, it is a key step to collect the wind farm meteorological information in the same time period as the fault traveling wave data through the weather monitoring module configured in the wind farm transmission system. The weather monitoring module usually includes devices such as anemometers, thermometers, humidity sensors, and barometers. These devices are distributed at different locations in the wind farm to comprehensively capture local meteorological changes. When the fault traveling wave data is detected, the system will synchronously record the current meteorological data. For example, if the fault occurs at 14:30:25 on July 15, 2023, then data such as the wind speed (e.g., 12 m / s), temperature (e.g., 28 °C), humidity (e.g., 65%), and air pressure (e.g., 101.3 kPa) at this precise moment will be recorded. These real-time meteorological data are crucial for subsequent analysis of the fault cause and signal characteristics because different meteorological conditions may have a significant impact on the operating state and fault characteristics of the power system. In this way, the correlation between the fault phenomenon and environmental factors can be established, providing an important basis for more accurate fault diagnosis and prevention.

[0124] Based on the collected wind farm meteorological information, the meteorological severity during the period when the fault traveling wave data is obtained is calculated according to the preset meteorological condition rules. This step aims to quantify the potential impact of meteorological conditions on the operation of the power system. The calculation of meteorological severity usually involves the weighted synthesis of multiple factors. For example, the following formula can be used:

[0125] Meteorological severity = a * (wind speed / wind speed threshold) + b * |temperature - suitable temperature| / temperature range + c * (humidity / humidity threshold) + d * |air pressure - standard air pressure| / pressure range.

[0126] Among them, a, b, c, and d are weight coefficients, which are determined according to the importance of each factor's impact on the system. Parameters such as the wind speed threshold, suitable temperature, humidity threshold, and standard air pressure are determined according to equipment specifications and historical data. For example, if the wind speed is 25 m / s (exceeding the threshold of 20 m / s), the temperature is 35 °C (deviating from the suitable temperature of 25 °C), the humidity is 90% (exceeding the threshold of 80%), and the air pressure is 98 kPa (deviating from the standard air pressure of 101.3 kPa), then the calculated meteorological severity may be relatively high, such as 0.8 (assuming a full score of 1). This indicator intuitively reflects the severity of the current meteorological conditions and provides an important reference for subsequent signal processing and fault analysis.

[0127] Next, integrate the key node attributes of all wind turbine generator nodes and calculate the unit state complexity of all wind turbine generator nodes based on these attributes. This step aims to quantify the complexity of the overall operation state of the wind farm. The key node attributes usually include the operation state and grid connection time of each wind turbine. The operation state can be divided into several categories, such as normal operation, fault shutdown, maintenance, etc. The grid connection time reflects the degree of impact of new wind turbines on the fluctuations of the wind farm lines. Therefore, the unit state complexity can be calculated using the following method:

[0128] Unit state complexity = Σ(Wi * Si * Ti) / N

[0129] Where Wi is the weight of each operation state, Si is the number of wind turbines in that state, Ti is the average grid connection time factor of this group of wind turbines, and N is the total number of wind turbines. This index reflects the overall complexity of the current operation state of the wind farm and provides important background information for subsequent signal analysis.

[0130] Next, calculate the signal smoothness of the fault traveling wave data by combining the meteorological severity and the unit state complexity. The purpose of this step is to evaluate the overall characteristics of the fault traveling wave data and provide a basis for selecting an appropriate signal processing method. The signal smoothness can be calculated using the following formula:

[0131] Signal smoothness = 1 - (α * meteorological severity + β * unit state complexity) / (α + β)

[0132] Where α and β are weight coefficients used to adjust the relative influence of meteorological factors and unit state on signal characteristics. For example, if the meteorological severity is 0.8, the unit state complexity is 1.11, α = 0.6, and β = 0.4, then the signal smoothness is: 1 - (0.6 * 0.8 + 0.4 * 1.11) / (0.6 + 0.4) = 0.0734. The closer this value is to 1, the smoother the signal; the closer it is to 0, the less smooth the signal. According to the calculated signal smoothness, select an appropriate modal decomposition algorithm to process the fault traveling wave component. If the signal smoothness is less than the preset smoothness threshold (e.g., 0.5), then use the empirical mode decomposition (EMD) algorithm; if it is greater than or equal to the threshold, then use the variational mode decomposition (VMD) algorithm. Both of these algorithms aim to decompose the fault signal into multiple single-frequency components, but are suitable for different signal characteristics. For the monitoring of the collector line under normal operation conditions, the VMD algorithm is preferably selected. Because in this case, the signal usually contains the stable operation characteristics of multiple wind turbines, and VMD can better separate these characteristics. In case of severe weather conditions (such as thunderstorm weather) and in the presence of multiple newly grid-connected wind farms or new types of wind power equipment, the EMD algorithm is preferably selected. Because in this case, strong transient signals may be generated, and EMD is more suitable for processing such non-linear and non-stationary signals.

[0133] The EMD algorithm decomposes a signal into several intrinsic mode functions (IMFs) and a residual term through an iterative "sifting" process. It is particularly suitable for processing non-linear and non-stationary signals. For example, for a fault traveling wave signal containing multiple frequency components, EMD may decompose it into 5 - 10 IMFs, and each IMF represents an oscillation mode within a specific frequency range. The VMD algorithm, on the other hand, decomposes a signal into a predetermined number of band-limited mode functions through an optimization process. It may be more stable than EMD when processing certain types of signals, especially when dealing with signals close to being stationary. For example, VMD may decompose the same fault traveling wave signal into 3 - 5 mode functions, each with a clearly defined central frequency and bandwidth. Regardless of which algorithm is used, multiple traveling wave modal component signals in different frequency domains will ultimately be obtained. These component signals respectively reflect the characteristics of the fault traveling wave within different frequency ranges, providing a more refined and reliable information basis for subsequent feature extraction and fault identification. Through this adaptive algorithm selection, various complex fault traveling wave data can be processed more accurately, improving the accuracy and reliability of subsequent analysis.

[0134] In one implementation, decomposing the fault traveling wave component using the empirical mode decomposition algorithm to obtain multiple traveling wave modal component signals in different frequency domains includes the following steps:

[0135] Calculate the signal maximum and signal minimum of the fault traveling wave component, and combine the signal maximum and signal minimum and use the cubic spline interpolation method to obtain the upper envelope and lower envelope of the fault traveling wave component;

[0136] Perform an average calculation on the upper envelope and the lower envelope to obtain the average component of the fault traveling wave component;

[0137] Take the difference between the fault traveling wave component and the average component as the preliminary IMF component;

[0138] If the number of zeros and the number of extrema of the preliminary IMF component are equal, or the average value of the upper and lower envelopes of the preliminary IMF component at any moment is 0, then take the preliminary IMF component as one of the traveling wave modal component signals in the frequency domain;

[0139] Calculate the remaining signal component after removing the traveling wave modal component signal from the fault traveling wave component;

[0140] Take the remaining signal component as the new fault traveling wave component and repeat the above component decomposition steps until the component decomposition cut-off condition is met;

[0141] After completing the component decomposition of the fault traveling wave component, use the soft limiter function to filter out the signal noise of all traveling wave modal component signals in the corresponding frequency domain.

[0142] In this embodiment, first, the signal maximum and minimum values of the fault traveling wave component are calculated, and the upper and lower envelopes are obtained by using cubic spline interpolation. Specifically, the fault traveling wave component data is traversed to find all local maximum and minimum points. For example, for a signal containing 1000 sampling points, 50 maximum points and 49 minimum points may be found. Then, the cubic spline interpolation method is used to connect these extreme points to form the upper and lower envelopes. Cubic spline interpolation can generate a smooth curve, avoiding the sawtooth effect that may be brought by linear interpolation. Then, the upper and lower envelopes are averaged to obtain the average component of the fault traveling wave component. The specific operation is to add the values of the upper envelope and the lower envelope at each time point and then divide by 2. For example, if the value of the upper envelope at a certain moment is 10 and the value of the lower envelope is -8, the average component value at this moment is 1. This average component reflects the overall trend of the signal, and removing this trend can highlight the oscillatory components in the signal.

[0143] Next, the difference between the fault traveling wave component and the average component is used as the preliminary IMF component. This step is to extract the oscillatory components in the signal. The specific operation is to subtract the just-calculated average component from the original fault traveling wave component. For example, if the original signal value at a certain moment is 5 and the average component value is 1, the value of the preliminary IMF component at this moment is 4. This difference represents the local oscillation in the signal after removing the overall trend and is a potential intrinsic mode function (IMF). Therefore, it is necessary to determine whether the preliminary IMF component meets the definition conditions of the IMF. The IMF needs to meet two conditions: the number of extreme points is equal to or differs by no more than 1 from the number of zero points; the average value of the upper and lower envelopes is close to 0 at any moment. If these conditions are met, the preliminary IMF component is determined as a traveling wave mode component signal in the frequency domain. For example, if the preliminary IMF component has 100 extreme points and 99 zero points, and the average value of its upper and lower envelopes is less than a preset threshold (such as 0.01) at all times, it can be regarded as a valid IMF.

[0144] Next, the remaining signal component after removing the traveling wave mode component signal from the fault traveling wave component is calculated. The specific operation is to subtract the just-determined IMF from the original fault traveling wave component. This remaining component contains the low-frequency components and trends that have not been extracted. The remaining signal component is used as the new fault traveling wave component and the above component decomposition steps are repeated until the component decomposition cut-off condition is met. This process is the iterative core of the EMD algorithm. The expression formula of the component decomposition cut-off condition is as follows:

[0145]

[0146] where: M represents the number of signal of the traveling wave mode component signals that have been decomposed, Denote the traveling wave mode component signal obtained in the m-th round of the component decomposition step. Denote the traveling wave mode component signal obtained in the (m - 1)-th round of the component decomposition step, and ε represents the cut-off threshold. This iterative process ensures that each frequency component in the signal is extracted one by one.

[0147] After completing the component decomposition of the fault traveling wave component, use the soft-limiting function to filter out the signal noise of all traveling wave mode component signals in the corresponding frequency domain. This step aims to improve the signal-to-noise ratio of the signal. The soft-limiting function can be defined as:

[0148] f(x) = sign(x) * max(0, |x| - threshold)

[0149] where threshold is the threshold set according to the characteristics of each IMF. For example, for high-frequency IMFs, a higher threshold can be set to remove more noise. This step can effectively remove random noise in each frequency band while retaining the main features of the signal, thereby improving the accuracy of subsequent analysis.

[0150] In one implementation, the variational mode decomposition algorithm is used to decompose the fault traveling wave component to obtain multiple traveling wave mode component signals in different frequency domains, including the following steps:

[0151] Based on the decomposition constraint conditions of the variational mode decomposition algorithm, construct a decomposition constraint model for the fault traveling wave component. The decomposition constraint conditions are as follows: The deviation between the reconstructed signal of the N traveling wave mode component signals obtained by decomposing the fault traveling wave component using the variational mode decomposition algorithm and the fault traveling wave component is minimized, and the sum of the estimated bandwidths of all traveling wave mode component signals is minimized.

[0152] Introduce the Lagrange operator and the penalty factor into the decomposition constraint model to construct the model augmented expression of the decomposition constraint model.

[0153] Based on the model augmented expression and using the alternating direction method of the Lagrange operator, perform alternating iterative optimization in the frequency domain until the Wiener filtering of all traveling wave mode components satisfies the convergence condition. In each round of alternating iterative optimization process, a traveling wave mode component signal in a different frequency domain is decomposed from the fault traveling wave component.

[0154] In this implementation, based on the decomposition constraint conditions of the variational mode decomposition algorithm, construct a decomposition constraint model for the fault traveling wave component, so as to decompose the fault traveling wave component in an optimized way. The decomposition constraint conditions include two key objectives: First, the deviation between the reconstructed signal of the N decomposed traveling wave mode component signals and the original fault traveling wave component should be minimized to ensure the accuracy of the decomposition; Second, the sum of the estimated bandwidths of all traveling wave mode component signals should be minimized, which ensures the simplicity of the decomposition. Specifically, it can be expressed as minimizing the objective function:

[0155]

[0156] Among them, uk(t) is the k-th modal component, ωk is its center frequency, and f(t) is the original signal. This model optimizes simultaneously in the time-frequency domain, which not only ensures the reconstruction accuracy but also minimizes the bandwidth of each component, thus achieving the precise decomposition of complex fault traveling wave signals. Then, the Lagrange operator and penalty factor are introduced into the decomposition constraint model to construct the augmented expression of the decomposition constraint model. The purpose of this step is to transform the constrained optimization problem into an unconstrained optimization problem for easy solution. Specifically, the Lagrange multiplier λ and penalty factor α are introduced to construct the augmented Lagrangian function:

[0157]

[0158] Here, α controls the strength of the bandwidth constraint, and the term λ / 2 ensures the satisfaction of the reconstruction constraint. This expression incorporates the original constraint conditions into the objective function, enabling the optimization process to consider all constraints simultaneously and greatly simplifying the solution process. By adjusting α and λ, a balance can be achieved between the reconstruction accuracy and bandwidth minimization, thereby obtaining the optimal decomposition result.

[0159] Then, based on the augmented expression of the model and using the alternating direction method of the Lagrange operator, alternating iterative optimization is performed in the frequency domain until the Wiener filtering of all traveling wave modal components meets the convergence condition. In each round of alternating iterative optimization process, a traveling wave modal component signal with a different frequency domain is decomposed from the fault traveling wave components. This process is the core of the variational mode decomposition algorithm, and each frequency component is gradually extracted through iterative optimization. The specific process is as follows:

[0160] a) Initialize each modal component uk and center frequency ωk.

[0161] b) For each k, update uk: uk = argminL(ui, ωi, λ), where i ≠ k is fixed.

[0162] c) Update ωk: ωk = argminL(ui, ωi, λ), where i ≠ k is fixed.

[0163] d) Update λ: λ = λ + τ(f - ∑uk).

[0164] e) Repeat b - d until convergence.

[0165] In each iteration, uk is updated through the Wiener filter.

[0166] The expression formula for the convergence condition is as follows:

[0167]

[0168] In the formula: K represents the number of signal of the decomposed traveling wave mode component signals that have been completed, represents the Wiener filter of the traveling wave mode component signals obtained during the s-th round of alternating iterative optimization, represents the Wiener filter of the traveling wave mode component signals obtained during the (s + 1)-th round of alternating iterative optimization, ||·||2 represents the Euclidean norm, and ζ represents the convergence accuracy.

[0169] This process ensures that each mode component is the optimal representation of a specific frequency band in the original signal. Through continuous iteration, the algorithm gradually separates the various frequency components in the fault traveling wave signal until all components meet the convergence conditions. This method can effectively handle non-linear and non-stationary signals and is particularly suitable for analyzing complex fault traveling wave data.

[0170] In one implementation, based on the NTEO energy operator, the signal energy spectra of all traveling wave mode component signals are calculated, and the initial wavefront of the fault traveling wave of the fault traveling wave data is calibrated according to the signal energy spectrum of the highest-frequency traveling wave mode component signal, including the following steps:

[0171] Convert all traveling wave mode component signals into discrete component signals;

[0172] Use the NTEO energy operator to calculate the signal energy spectra of each discrete component signal respectively;

[0173] Select the signal energy spectrum corresponding to the traveling wave mode component signal in the highest frequency domain as the target signal energy spectrum, and calibrate the time point with the highest energy value in the target signal energy spectrum as the initial wavefront of the fault traveling wave of the fault traveling wave data.

[0174] In this implementation, converting all traveling wave mode component signals into discrete component signals is actually converting a continuous analog signal into a series of discrete digital representations. Specifically, during implementation, first determine an appropriate sampling frequency, which should be at least twice the highest frequency component of the signal (following the Nyquist sampling theorem) to avoid information loss and aliasing effects. For example, if the highest frequency component is 5 kHz, the sampling frequency should be at least 10 kHz. Then, sample the continuous signal at equally spaced time points. Suppose there is a traveling wave mode component signal lasting for 1 second with a sampling frequency of 10 kHz, then finally 10,000 discrete data points will be obtained. This conversion enables subsequent digital signal processing while retaining the key features of the original signal. The converted discrete signal can be represented as h(x), where x is the discrete time index.

[0175] Next, the signal energy spectra of each discrete component signal are calculated using the NTEO (Normalized Teager Energy Operator). NTEO is an improved version of the Teager energy operator, with stronger anti-noise ability and higher time resolution. For each discrete component signal, the calculation formula for the signal energy spectrum using NTEO is as follows:

[0176]

[0177] In the formula: Ψ[h i (x)] represents the signal energy spectrum of the i-th discrete component signal h i (x), T represents the resolution parameter, and M represents the number of signals of the discrete component signal. This operator can effectively capture the instantaneous energy changes of the signal, and these energy values constitute the energy spectrum of the signal, intuitively reflecting the energy distribution of the signal at different times. NTEO is particularly suitable for processing non-linear and non-stationary signals because it can accurately locate the mutation and transient features in the signal. Through this method, each discrete component signal will obtain a corresponding energy spectrum, providing an important basis for subsequent fault location.

[0178] Next, select the signal energy spectrum corresponding to the traveling wave mode component signal in the highest frequency domain as the target signal energy spectrum, and calibrate the time point with the highest energy value in the target signal energy spectrum as the initial wavehead of the fault traveling wave data of the fault traveling wave. The core idea of this step is to use the sensitivity of the high-frequency component to the mutation feature to accurately locate the time of the fault occurrence. The component in the highest frequency domain is selected because high-frequency signals usually contain the richest transient information and are the most sensitive to mutations in the system. In the selected energy spectrum, the point with the highest energy value is considered the initial wavehead of the fault traveling wave because a sudden increase in energy usually occurs when a fault occurs. Assume that the energy value of the 500th sampling point in the energy spectrum is the highest, and the sampling frequency is 10 kHz, then it can be determined that the fault occurs at 50 ms after the start of the signal. The advantage of this method is that it can very accurately locate the fault time, and the error is usually at the millisecond level, providing an accurate time reference for subsequent fault type identification and fault location.

[0179] In one implementation, the fault traveling wave components include the fault traveling wave line mode component and the fault traveling wave zero mode component. The fault traveling wave line mode component and the fault traveling wave zero mode component share the initial wavehead of the fault traveling wave. Calculating the fault coefficients of all collector branch lines based on the initial wavehead of the fault traveling wave and according to the collector line topology structure, and combining all the fault coefficients to locate the line fault point in the collector line topology structure includes the following steps:

[0180] Take the fault data detection terminal node closest to the wind farm power transmission system node in the collector line topology as the starting node of the main line, and take the fault data detection terminal node farthest from the wind farm power transmission system node as the ending node of the main line;

[0181] Define the direction from the starting node of the main line to the ending node of the main line as the positive direction of the traveling wave;

[0182] Combined with the initial wavefront of the fault traveling wave and the first propagation speed of the line-mode component of the fault traveling wave, calculate respectively the first propagation time for the line-mode component of the fault traveling wave to propagate to the starting node of the main line, and the second propagation time for the line-mode component of the fault traveling wave to propagate to the ending node of the main line;

[0183] Combined with the initial wavefront of the fault traveling wave and the second propagation speed of the zero-mode component of the fault traveling wave, calculate respectively the third propagation time for the zero-mode component of the fault traveling wave to propagate to the starting node of the main line, and the fourth propagation time for the zero-mode component of the fault traveling wave to propagate to the ending node of the main line;

[0184] Combined with the first propagation time, the second propagation time, the third propagation time, the fourth propagation time and the first propagation speed, calculate respectively the first fault distance between the imaginary fault point and the starting node of the main line, and the second fault distance between the imaginary fault point and the ending node of the main line;

[0185] Combined with the first fault distance, the second fault distance and the node-edge attributes in the collector line topology, and calculate the fault coefficients of all collector branch lines according to the positive direction of the traveling wave;

[0186] If the fault coefficient of any collector branch line is 0, it is determined that the line fault point where the fault traveling wave data is generated is located on the collector branch line with a fault coefficient of 0, and the midpoint of the faulty collector branch line is used as the line fault point in the collector line topology;

[0187] If all fault coefficients are not 0, it is determined that the line fault point is located on the collector main line, calculate by combining the initial wavefront of the fault traveling wave, the propagation speed of the fault traveling wave component and the node-edge attributes, and use the double-ended traveling wave location method to determine the line fault point in the collector line topology.

[0188] In this embodiment, due to the characteristics of numerous branches and short lines in the wind farm collector line, it is difficult for the traditional single-ended method and double-ended method to accurately calculate the fault distance, resulting in a large error in the calculation result and misjudgment of the faulty branch. For this reason, this embodiment uses an improved single-ended traveling wave method to calculate the fault coefficient of the line. Specifically refer to Figure 2, the fault data detection terminal node closest to the wind farm power transmission system node in the collector line topology is used as the starting node Q1 of the main line, and the fault data detection terminal node farthest from the wind farm power transmission system node is used as the ending node Q of the main line. n , determine the collector main line Q1Qn, and then define the direction from the starting node of the main line to the ending node of the main line as the positive direction of the traveling wave. Starting from the starting node Q1 of the main line, along the positive direction of the traveling wave and in ascending order, the substation nodes on the collector main line are sequentially marked as the starting nodes {p1, p2, p3,..., p N} of each collector branch line, where N represents the number of substation nodes. The end of each collector branch line is the last wind turbine node in the collector branch line. The end nodes of the collector branch lines are marked as {q1, q2, q3,..., q N} according to the marking method of the starting nodes of the branch lines.

[0189] Next, determine the initial fault occurrence time t0 according to the initial wavefront of the fault traveling wave. Then, combining the initial fault occurrence time t0 and the first propagation speed of the fault traveling wave line mode component, calculate the first propagation time for the fault traveling wave line mode component to propagate to the starting node of the main line (the propagation time refers to the signal propagation time from the initial fault occurrence time until the signal reaches the corresponding node after propagating through the line. The propagation time described in the subsequent embodiments of this implementation has the same meaning), and the second propagation time for the fault traveling wave line mode component to propagate to the ending node of the main line. Next, combining the initial fault occurrence time t0 and the second propagation speed of the fault traveling wave zero mode component, calculate the third propagation time for the fault traveling wave zero mode component to propagate to the starting node of the main line, and the fourth propagation time for the fault traveling wave zero mode component to propagate to the ending node of the main line. Then, based on the above calculation results, further calculate the first fault distance between the imaginary fault point and the starting node of the main line. The imaginary fault point is the fault occurrence point to be specifically located in the collector line. The calculation formula for the first fault distance is as follows:

[0190]

[0191] In the formula: represents the first fault distance between the imaginary fault point G and the starting node Q1 of the main line, t1 represents the first propagation time, t2 represents the second propagation time, t3 represents the third propagation time, t4 represents the fourth propagation time, and v1 represents the first propagation speed.

[0192] Further calculate the second fault distance between the imaginary fault point and the ending node of the main line. The calculation formula for the second fault distance is as follows:

[0193]

[0194] In the formula: represents the second fault distance between the imaginary fault point G and the termination node Q of the main line n of the main line.

[0195] As Figure 2 shown, the main collector line can be regarded as a special branch composed of branches Q1p1, p1p2, p2p3, p3p4, p4p5, and p5Qn. After a fault occurs, since the node-edge attributes in the collector line topology represent the actual line lengths of each branch, the fault coefficients of each fan branch can be calculated by combining the first fault distance, the second fault distance, and the node-edge attributes in the collector line topology. If the fault occurs in the collector branch line, the fault coefficients of each collector branch line calculated through the above content are shown in Table 1 below.

[0196] Table 1 Fault coefficients of each collector branch line during a fault in the collector branch line

[0197]

[0198] According to the results in Table 1, it can be seen that the fault coefficient of the faulty collector branch line is 0, and the fault coefficients of the non-faulty collector branch lines are not 0. Taking the faulty collector branch line as the reference point, the decision coefficient of the collector branch line approaching the starting node of the main line along the positive direction of the traveling wave is greater than 0, and the decision coefficient of the collector branch line approaching the termination node of the main line is less than 0. Therefore, it can also be further confirmed that if the decision coefficient of a certain collector branch line is 0 and the decision coefficients of the other collector branch lines are not 0, it means that the fault occurs in the collector branch line with a fault coefficient of 0. Since the collector branch line is short, the midpoint of the collector branch line can be simply used as the final line fault location point. If the fault occurs in the main collector line, the fault coefficients of each collector branch line calculated through the above content are shown in Table 2 below.

[0199] Table 2 Fault coefficients of each collector branch line during a fault in the main collector line

[0200]

[0201] According to the results in Table 2, when a fault occurs in the main collector line, the decision coefficients of each collector branch line are all 0, and the value of the decision coefficient is related to the position of the imaginary fault point. The decision coefficient of the collector branch line close to the starting node of the main line along the positive direction of the traveling wave is greater than 0, and the decision coefficient of the collector branch line close to the ending node of the main line is less than 0. Therefore, it can be further confirmed that if all fault coefficients are not 0, the fault point of the line is located on the main collector line. Since the main collector line is very long and the calculated fault distance may correspond to multiple fault positions, it is necessary to further use a more accurate double-ended traveling wave location method for precise positioning. The calculation formula of the double-ended traveling wave location method adopted in this embodiment is as follows:

[0202]

[0203] In the formula: represents the actual length of the main collector line, represents the propagation time of the fault traveling wave generated by the fault point of the line to the starting node of the main line, represents the propagation time of the fault traveling wave generated by the fault point of the line to the ending node of the main line, and v represents the propagation speed of the fault traveling wave generated by the fault point of the line.

[0204] In one of the embodiments, the energy spectrum features of all signal energy spectra are extracted, and the steps for identifying the fault type of the fault point of the line based on the energy spectrum features through a deep learning model are as follows:

[0205] For the signal energy spectrum of each traveling wave mode component signal, energy spectrum segments of different time lengths are intercepted from the signal energy spectrum through time windows of different time scales, and energy time domain features are extracted from each energy spectrum segment respectively. The energy time domain features include energy peak value, energy decay rate, energy rise time, energy duration, and energy phase difference feature;

[0206] Statistical methods are used to extract the energy frequency domain features of the signal energy spectrum in the frequency domain. The energy frequency domain features include energy ratio, energy mean, energy variance, and energy kurtosis;

[0207] The correlation analysis method is used to screen out the energy feature subset with fault type discrimination ability from the energy time domain features and energy frequency domain features;

[0208] The energy feature subset is input into the pre-trained energy feature recognition model, and the fault type of the fault point of the line is output through the energy feature recognition model. The energy feature recognition model is a convolutional neural network model containing a multi-head attention mechanism.

[0209] In this embodiment, performing multi-scale time window analysis on the energy spectrum of each traveling wave mode component signal is a key step in extracting rich feature information. This process involves using time windows of different lengths (such as 10 ms, 50 ms, 100 ms, etc.) to slide through the entire energy spectrum, so as to capture the energy change characteristics at different time scales. For the energy spectrum segments intercepted by each window, five energy time-domain features are extracted: (1) energy peak value, which is the maximum energy value in the segment and reflects the fault intensity; (2) energy decay rate, which calculates the energy decline trend through exponential fitting and characterizes the fault persistence; (3) energy rise time, which is the time from when the energy starts to increase to reach the peak value and indicates the fault development speed; (4) energy duration, which is the duration when the energy exceeds a certain threshold and reflects the fault influence range; (5) energy phase difference feature, which compares the energy differences between different phases and helps to identify fault types such as phase-to-phase short circuits. These features comprehensively characterize the energy distribution characteristics of the fault traveling wave in the time domain and provide an important basis for subsequent fault type discrimination.

[0210] Next, using statistical methods to extract the frequency-domain features of the signal energy spectrum is another important dimension for analyzing fault characteristics. This step mainly focuses on the statistical characteristics of the energy in the frequency distribution, including four key indicators: (1) energy ratio, which calculates the proportion of the energy in a specific frequency band to the total energy and reflects the frequency concentration of the fault signal; (2) energy mean value, which calculates the average energy level of the entire spectrum and characterizes the overall intensity of the fault; (3) energy variance, which measures the degree of dispersion of the energy distribution and indicates the frequency complexity of the fault signal; (4) energy kurtosis, which describes the sharpness of the energy distribution and helps to identify sudden faults. The calculation of these features usually requires first performing a Fourier transform on the energy spectrum to convert the energy spectrum to the frequency domain, and then applying the corresponding statistical formulas to calculate. These frequency-domain features combined with the time-domain features jointly provide a comprehensive description of the fault signal and enhance the accuracy and reliability of subsequent fault type identification.

[0211] Next, using the correlation analysis method to screen out the feature subset with discriminative power for fault types from the energy time-domain features and frequency-domain features is the key step to optimize the feature space. This process aims to reduce redundant information and improve the efficiency and accuracy of fault identification. Specifically, methods such as Pearson correlation coefficient or mutual information can be used to calculate the correlation between each feature and the fault type, as well as the mutual correlation between features. For example, calculate the correlation coefficient between each feature and the fault type, and select the features with the absolute value of the correlation coefficient greater than 0.5; at the same time, for the feature pairs with the correlation coefficient exceeding 0.8 between each other, only retain the one with a higher correlation with the fault type. This step can not only improve the training efficiency of the subsequent model, but also reduce the risk of overfitting and enhance the generalization ability of the model on new data. The selected feature subset includes energy peak value, energy proportion of the main frequency band, and energy decay rate, and these features together constitute a concise and effective fault feature representation.

[0212] Finally, input the selected energy feature subset into the pre-trained energy feature recognition model to output the fault type of the line fault point. This step uses a convolutional neural network model with a multi-head attention mechanism. This structure combines the advantages of CNN in local feature extraction and the ability of the attention mechanism in capturing long-range dependencies. The specific structure of the model is as follows: First, the input layer receives the feature subset; then, extract high-level features through multiple convolutional layers and pooling layers. Specifically, there are 3 convolutional layers, and each layer uses convolutional kernels of different sizes to capture feature patterns of different scales; then, calculate the correlation between different features through the multi-head attention layer, using 3 attention heads, and each head independently learns the association between features; finally, the fully connected layer and the softmax layer output the probability distribution of the fault type. This model structure can effectively utilize the extracted energy features, automatically learn the complex patterns of fault types, and thus achieve high-precision fault type recognition. The training process of the model involves a large amount of labeled historical fault data, and continuously adjusts the network parameters through the backpropagation algorithm to finally achieve the best performance on the validation set. Through the above process, starting from the original traveling wave modal component signal, through multi-scale time-domain analysis, frequency-domain statistical feature extraction, feature selection based on correlation, and finally using an advanced deep learning model for fault type recognition, a comprehensive, efficient, and accurate fault diagnosis system is constructed. This system can quickly process complex fault traveling wave data, extract key features, and give reliable fault type judgments.

[0213] The present invention also discloses a fault monitoring and management system for a 35 kV wind farm overhead collector line, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the fault monitoring and management method for the 35 kV wind farm overhead collector line described in any of the above embodiments.

[0214] Among them, the processor may adopt a central processing unit (CPU). Of course, according to the actual usage, other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. may also be adopted. The general-purpose processor may adopt a microprocessor or any conventional processor, etc. The present application does not limit this.

[0215] Among them, the memory may be an internal storage unit of the computer device. For example, the hard disk or memory of the computer device, or it may also be an external storage device of the computer device. For example, the plug-in hard disk, smart media card (SMC), secure digital card (SD) or flash card (FC) etc. equipped on the computer device. And the memory may also be a combination of the internal storage unit and the external storage device of the computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory may also be used to temporarily store the data that has been output or will be output. The present application does not limit this.

[0216] Those of ordinary skill in the art should understand that: The discussion of any of the above embodiments is only exemplary and is not intended to imply that the protection scope of the present application is limited to these examples; Under the idea of the present application, the technical features in the above embodiments or different embodiments can also be combined, and the steps can be implemented in any order, and there are many other variations in different aspects of one or more embodiments in the present application as above. For the sake of brevity, they are not provided in detail.

[0217] One or more embodiments of the present application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application shall be included within the protection scope of the present application.

Claims

1. A fault monitoring and management method for 35 kV overhead collector lines in a wind farm, characterized in that, Including the following steps: Obtain the collector line diagram of the wind farm collector line; Simplify the collector line diagram into a collector line topology structure, which includes a collector main line directly connected to the wind farm power transmission system, and a plurality of collector branch lines connected to the collector main line; Obtain the fault traveling wave data in the wind farm collector line through the fault data detection terminal deployed in the collector main line; Decouple the fault traveling wave data into fault traveling wave components by using the Karenbauer transform method; Decompose the fault traveling wave components by using a modal decomposition algorithm to obtain a plurality of traveling wave modal component signals in different frequency domains; Calculate the signal energy spectrum of all the traveling wave modal component signals based on the NTEO energy operator, and calibrate the fault traveling wave initial wavehead of the fault traveling wave data according to the signal energy spectrum of the highest frequency traveling wave modal component signal; Calculate the fault coefficients of all the collector branch lines based on the fault traveling wave initial wavehead and according to the collector line topology structure, and locate the line fault point in the collector line topology structure by combining all the fault coefficients; Extract the energy spectrum features of all the signal energy spectra, and identify the fault type of the line fault point based on the energy spectrum features and through a deep learning model; Generate a fault monitoring report by combining the line fault point and the fault type.

2. The fault monitoring and management method for the 35kV wind farm overhead collector line according to claim 1, wherein, The step of simplifying the collector line diagram into a collector line topology structure includes the following steps: Identify the key components in the collector line diagram and the collector lines between the key components; Simplify all the key components into key nodes, and simplify the collector lines into key node edges between the key nodes to form a plurality of topological lines. The key nodes include wind turbine generator nodes, substation nodes, fault data detection terminal nodes, and wind farm power transmission system nodes, and the number of fault data detection terminal nodes is 2; Add node edge attributes to the key node edges according to the actual length of the collector lines, and add key node attributes to the wind turbine generator nodes according to the unit status information of the wind turbine generators corresponding to the wind turbine generator nodes; For any one of the topological lines, if both ends of the topological line are the fault data detection terminal nodes, and all nodes in the topological line except the fault data detection terminal nodes are the substation nodes, then mark the topological line as the collector main line, and all the substation nodes are connected to the wind farm power transmission system nodes through the collector main line; If both ends of the topological line are the substation node and the wind turbine generator node respectively, then mark the topological line as the collector branch line; After all the topological lines are marked, obtain the collector line topology structure after simplifying the collector line diagram.

3. The fault monitoring and management method for the 35 kV overhead collector line of a wind farm according to claim 2, wherein The step of decomposing the fault traveling wave components by using a modal decomposition algorithm to obtain a plurality of traveling wave modal component signals in different frequency domains includes the following steps: Collect the wind farm meteorological information in the same time period as the fault traveling wave data through the weather monitoring module configured in the wind farm power transmission system; Calculate the meteorological severity of the fault traveling wave data acquisition period based on the meteorological information of the wind farm and according to the preset meteorological condition rules; Integrate the key node attributes of all the wind turbine nodes, and calculate the unit state complexity of all the wind turbine nodes according to all the key node attributes; Calculate the signal smoothness of the fault traveling wave data by combining the meteorological severity and the unit state complexity; If the signal smoothness is less than the preset smoothness threshold, decompose the fault traveling wave component by using the empirical mode decomposition algorithm to obtain traveling wave modal component signals in multiple different frequency domains; If the signal smoothness is greater than or equal to the smoothness threshold, decompose the fault traveling wave component by using the variational mode decomposition algorithm to obtain traveling wave modal component signals in multiple different frequency domains.

4. The fault monitoring and management method for the 35 kV overhead collector line of a wind farm according to claim 3, wherein The step of decomposing the fault traveling wave component by using the empirical mode decomposition algorithm to obtain traveling wave modal component signals in multiple different frequency domains includes the following steps: Calculate the signal maximum value and the signal minimum value of the fault traveling wave component, and combine the signal maximum value and the signal minimum value and use the cubic spline interpolation method to obtain the upper envelope line and the lower envelope line of the fault traveling wave component; Perform an average calculation on the upper envelope line and the lower envelope line to obtain the average component of the fault traveling wave component; Take the difference between the fault traveling wave component and the average component as the preliminary IMF component; If the number of zero points and the number of extreme values of the preliminary IMF component are equal, or the average value of the upper and lower envelope lines of the preliminary IMF component at any time is 0, then take the preliminary IMF component as the traveling wave modal component signal in one of the frequency domains; Calculate the remaining signal component after removing the traveling wave modal component signal from the fault traveling wave component; Take the remaining signal component as a new round of fault traveling wave component and repeat the above component decomposition steps until the component decomposition cut-off condition is met; After completing the component decomposition of the fault traveling wave component, use the soft threshold function to filter out the signal noise of all the traveling wave modal component signals in the corresponding frequency domain; The expression formula of the component decomposition cut-off condition is as follows: Where: M represents the number of signal quantities of the traveling wave mode component signals that have been decomposed, represents the traveling wave mode component signal obtained in the m-th round of the component decomposition step, represents the traveling wave mode component signal obtained in the (m - 1)-th round of the component decomposition step, and ε represents the cut-off threshold.

5. The fault monitoring and management method for the 35 kV wind farm overhead collector line according to claim 3, characterized in that, The step of decomposing the fault traveling wave component by using the variational mode decomposition algorithm to obtain traveling wave modal component signals in multiple different frequency domains includes the following steps: Construct a decomposition constraint model for the fault traveling wave component based on the decomposition constraint conditions of the variational mode decomposition algorithm. The decomposition constraint conditions are as follows: After decomposing the fault traveling wave component by using the variational mode decomposition algorithm, the deviation between the reconstruction of the N traveling wave modal component signals obtained and the fault traveling wave component is the smallest, and the sum of the estimated bandwidths of all the traveling wave modal component signals is the smallest; Introduce the Lagrange operator and the penalty factor into the decomposition constraint model to construct the model augmented expression of the decomposition constraint model; Based on the model augmented expression and using the alternating direction method of the Lagrangian operator, perform alternating iterative optimization in the frequency domain until the Wiener filtering of all the traveling wave mode components satisfies the convergence condition. In each round of the alternating iterative optimization process, a traveling wave mode component signal in a different frequency domain is decomposed from the fault traveling wave component; The expression formula of the convergence condition is as follows: where: K represents the number of signals of the traveling wave mode component signals that have been decomposed, represents the Wiener filter of the traveling wave mode component signals obtained during the s-th round of alternating iterative optimization, represents the Wiener filter of the traveling wave mode component signals obtained during the (s + 1)-th round of alternating iterative optimization, ||·||2 represents the Euclidean norm, and ζ represents the convergence accuracy.

6. The fault monitoring and management method for the 35kV wind farm overhead collector line according to claim 1, characterized in that, The steps of calculating the signal energy spectra of all the traveling wave mode component signals based on the NTEO energy operator and calibrating the initial wave head of the fault traveling wave of the fault traveling wave data according to the signal energy spectrum of the highest-frequency traveling wave mode component signal are as follows: Convert all the traveling wave mode component signals into discrete component signals; Use the NTEO energy operator to calculate the signal energy spectra of each of the discrete component signals respectively. The calculation formula of the signal energy spectrum is specifically as follows: where: Ψ[h i (x)] represents the signal energy spectrum of the i-th discrete component signal h i (x), T represents the resolution parameter, and M represents the number of signals of the discrete component signals; Select the signal energy spectrum corresponding to the traveling wave mode component signal in the highest frequency domain as the target signal energy spectrum, and calibrate the time point with the highest energy value in the target signal energy spectrum as the initial wave head of the fault traveling wave of the fault traveling wave data.

7. The fault monitoring and management method for the 35 kV wind farm overhead collector line according to claim 2, characterized in that, The fault traveling wave component includes a fault traveling wave line mode component and a fault traveling wave zero mode component. The fault traveling wave line mode component and the fault traveling wave zero mode component share the initial wave head of the fault traveling wave. The steps of calculating the fault coefficients of all the collector branch lines based on the initial wave head of the fault traveling wave and according to the collector line topology structure, and locating the line fault point in the collector line topology structure are as follows: Take the fault data detection terminal node closest to the wind farm power transmission system node in the collector line topology structure as the starting node of the main line, and take the fault data detection terminal node farthest from the wind farm power transmission system node as the ending node of the main line; Define the direction from the starting node of the main line to the ending node of the main line as the positive direction of the traveling wave; Combined with the initial wave head of the fault traveling wave and the first propagation speed of the fault traveling wave line mode component, calculate respectively the first propagation time when the fault traveling wave line mode component propagates to the starting node of the main line, and the second propagation time when the fault traveling wave line mode component propagates to the ending node of the main line; Combined with the initial wave head of the fault traveling wave and the second propagation speed of the fault traveling wave zero mode component, calculate respectively the third propagation time when the fault traveling wave zero mode component propagates to the starting node of the main line, and the fourth propagation time when the fault traveling wave zero mode component propagates to the ending node of the main line; Combined with the first propagation time, the second propagation time, the third propagation time, the fourth propagation time and the first propagation speed, calculate respectively the first fault distance between the imaginary fault point and the starting node of the main line, and the second fault distance between the imaginary fault point and the ending node of the main line; Combined with the first fault distance, the second fault distance and the node edge attribute in the collector line topology structure, and calculate the fault coefficients of all the collector branch lines according to the positive direction of the traveling wave; If the fault coefficient of any one of the collector branch lines is 0, it is determined that the line fault point generating the fault traveling wave data is located on the fault collector branch line with the fault coefficient of 0, and the midpoint of the fault collector branch line is used as the line fault point in the collector line topology structure; If all the fault coefficients are not 0, it is determined that the line fault point is located on the main collector line, and the line fault point in the collector line topology structure is determined by combining the initial wave head of the fault traveling wave, the propagation speed of the fault traveling wave component and the node edge attribute and using the double - ended traveling wave location method.

8. The fault monitoring and management method for the 35kV overhead collector line of a wind farm according to claim 7, characterized in that The calculation formula of the first fault distance is as follows: In the formula: represents the first fault distance between the imaginary fault point G and the starting node Q1 of the main line, t1 represents the first propagation time, t2 represents the second propagation time, t3 represents the third propagation time, t4 represents the fourth propagation time, and v1 represents the first propagation speed; The calculation formula of the second fault distance is as follows: Wherein: represents the second fault distance between the imaginary fault point G and the main line termination node Q n therebetween.

9. The fault monitoring and management method for the 35 kV wind farm overhead collector line according to claim 1, characterized in that, The steps of extracting the energy spectrum features of all the signal energy spectra and identifying the fault type of the line fault point based on the energy spectrum features through a deep learning model are as follows: For the signal energy spectrum of each traveling wave mode component signal, energy spectrum segments with different time lengths are intercepted from the signal energy spectrum through time windows with different time scales, and energy time - domain features are extracted from each energy spectrum segment respectively. The energy time - domain features include energy peak value, energy decay rate, energy rise time, energy duration and energy phase - to - phase difference feature; Statistical methods are used to extract the energy frequency - domain features of the signal energy spectrum in the frequency domain. The energy frequency - domain features include energy ratio, energy mean, energy variance and energy kurtosis; The correlation analysis method is used to screen out the energy feature subset with fault type discrimination ability from the energy time - domain features and the energy frequency - domain features; The energy feature subset is input into a pre - trained energy feature recognition model, and the fault type of the line fault point is output through the energy feature recognition model. The energy feature recognition model is a convolutional neural network model containing a multi - head attention mechanism.

10. A fault monitoring and management system for a 35 kV overhead collector line in a wind farm, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the fault monitoring and management method of the 35kV wind farm overhead collector line as described in any one of claims 1 to 9.

Images