Intelligent diagnosis and isolation device and method for distribution network line fault
Through high-frequency signal injection and a variety of advanced signal processing technologies, combined with graph theory path simulation and electromagnetic transient simulation, an intelligent diagnostic isolation system is built, which solves the accuracy and dynamic problems of distribution network fault diagnosis and isolation, and improves faults Processing efficiency and system stability.
Patent Information
- Application Number
- CN202411073619.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-08-07
AI Technical Summary
Existing distribution network fault diagnosis and isolation technologies are difficult to achieve accurate fault location and dynamic isolation when facing complex network structures and variable load conditions, resulting in low power supply reliability and recovery efficiency.
Using high-frequency signal injection and a variety of signal processing technologies (such as short-time Fourier transform, wavelet transform, integrated random forest algorithm, graph theory path simulation and electromagnetic transient simulation) to build an intelligent diagnostic isolation system to achieve fault feature extraction , classification and optimal isolation strategy formulation.
It improves the accuracy and speed of fault diagnosis, realizes dynamic optimization of isolation strategies, minimizes the impact of faults, adapts to distribution networks of different sizes and complexities, and improves system stability.
Smart Images

Figure CN118607390B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of electric variable testing and analysis, and in particular to a device and method for intelligently diagnosing and isolating faults in a power distribution network. Background Art
[0002] Existing distribution network line fault diagnosis and isolation technologies mainly rely on traditional protection devices and manual inspections. These methods usually use basic electrical parameters such as current, voltage and frequency for fault detection, and use preset protection strategies to isolate faults. In terms of fault location, commonly used technologies include impedance measurement, traveling wave method and fault indicator. The impedance measurement method uses the change in impedance at the fault point to estimate the fault location, but its accuracy is limited in complex network structures. The traveling wave method locates the fault by analyzing the time difference between the high-frequency transient signal generated by the fault and arriving at different measurement points. It has high accuracy, but has high requirements for equipment and synchronization technology. The fault indicator narrows the fault range by installing multiple indicating devices on the line, but it is difficult to accurately locate. In addition, some emerging technologies such as artificial intelligence and big data analysis have also begun to be applied to the field of fault diagnosis, trying to improve the accuracy of diagnosis through data mining and pattern recognition.
[0003] However, these traditional methods still face many challenges in the face of the complexity of modern distribution networks. First, with the widespread access of distributed energy, the topology and operating characteristics of the distribution network have become more complex, and the traditional fault characteristics have become unclear, increasing the difficulty of diagnosis. Secondly, various types of faults (such as single-phase grounding, phase-to-phase short circuit, high-resistance fault, etc.) have different manifestations under different network conditions, and a single diagnostic method is difficult to adapt to all situations. Furthermore, the access of rapidly changing loads and intermittent energy sources makes the network state dynamic and changeable, which puts higher requirements on the real-time nature of fault diagnosis. Finally, in terms of fault isolation strategy formulation, traditional methods often adopt fixed isolation schemes, which are difficult to dynamically optimize according to the real-time network status and the scope of fault impact, and are prone to excessive power outages or suboptimal recovery schemes. These problems combined have seriously affected the power supply reliability, power quality and system recovery efficiency of the distribution network. Summary of the invention
[0004] In view of this, an embodiment of the present invention provides a device and method for intelligent diagnosis and isolation of distribution network line faults, which are used to improve the efficiency of intelligent diagnosis and isolation of distribution network line faults.
[0005] The present invention provides a distribution network line fault intelligent diagnosis and isolation device, comprising: an acquisition module, used to inject and acquire high-frequency signals into a distribution network key node set of a pre-acquired target distribution network to obtain high-frequency response data, wherein the high-frequency response data includes signal characteristic data within the range of 50kHz-500kHz;
[0006] A processing module, used for synchronously sampling and preprocessing the high frequency response data to obtain reconstructed feature data;
[0007] A transformation module, used for performing short-time Fourier transform and wavelet transform on the reconstructed feature data to obtain a fault feature vector;
[0008] A classification module, used to classify the fault feature vector to obtain a preliminary fault diagnosis result;
[0009] An analysis module, used to perform graph theory path simulation analysis on the preliminary fault diagnosis result to obtain a target isolation strategy;
[0010] The simulation module is used to perform real-time digital simulation on the target isolation strategy, obtain the target isolation operation instruction, and isolate and control the target distribution network through the target isolation operation instruction.
[0011] In the present invention, the acquisition module is specifically used for: performing hierarchical analysis on the electrical parameters corresponding to the key node set of the distribution network to obtain the key node importance ranking; performing node clustering on the key node importance ranking by using the fuzzy C-means algorithm to obtain multiple node groups; dynamically adjusting the signal injection frequency of each node group by using the adaptive particle swarm optimization algorithm to obtain the signal injection parameter set corresponding to each node group; setting the parameters of the high-frequency signal generator of the target distribution network according to the signal injection parameter set corresponding to each node group to obtain the target signal source; performing power amplification processing on the target signal source to obtain the injection signal; performing power line coupling operation on the target distribution network according to the injection signal to obtain the line transmission signal; performing high-speed sampling on the line transmission signal to obtain the original sampling data; performing frequency band limiting on the original sampling data by using a bandpass filter to obtain the response signal of the limited frequency band; performing analog-to-digital conversion on the response signal of the limited frequency band to obtain the digitized high-frequency response data; performing spectrum analysis on the digitized high-frequency response data by fast Fourier transform to obtain the signal characteristic data in the range of 50kHz-500kHz.
[0012] In the present invention, the processing module is specifically used to: perform time stamp alignment on the high-frequency response data to obtain synchronous sampling data; perform noise suppression on the synchronous sampling data through a wavelet threshold denoising algorithm to obtain preliminary noise reduction data; perform Fourier transform on the preliminary noise reduction data to obtain frequency domain representation; perform bandpass filtering on the frequency domain representation through a Butterworth filter to obtain filtered frequency domain data; perform inverse Fourier transform on the filtered frequency domain data to obtain time domain filtered data; perform adaptive decomposition on the time domain filtered data through an empirical mode decomposition algorithm to obtain multiple intrinsic mode functions; perform Hilbert transform on the multiple intrinsic mode functions to obtain instantaneous frequency and instantaneous amplitude; perform trend extraction on the instantaneous frequency and instantaneous amplitude through a singular spectrum analysis algorithm to obtain the main features of the signal; perform principal component analysis on the main features of the signal to obtain target signal feature data; reconstruct the target signal feature data through a support vector regression algorithm to obtain the reconstructed feature data.
[0013] In the present invention, the transformation module is specifically used to: perform window segmentation on the reconstructed feature data to obtain multiple time window data segments; perform windowing processing on the time window data segments to obtain windowed data segments; perform fast Fourier transform on the windowed data segments to obtain a short-time spectrum; perform feature extraction on the short-time spectrum through a peak detection algorithm to obtain a frequency domain feature point set; perform discrete wavelet transform on the frequency domain feature point set to obtain multi-scale wavelet coefficients; perform sparse processing on the multi-scale wavelet coefficients through a threshold method to obtain sparse representation data; perform inverse wavelet transform on the sparse representation data to obtain a reconstructed signal; perform time-frequency analysis on the reconstructed signal through a Hilbert-Huang transform to obtain an instantaneous frequency and an instantaneous amplitude; perform statistical feature extraction on the instantaneous frequency and instantaneous amplitude to obtain statistical features in the time-frequency domain; perform dimensionality reduction processing on the time-frequency domain statistical features through a principal component analysis algorithm to obtain the fault feature vector.
[0014] In the present invention, the classification module is specifically used to: evaluate the feature importance of the fault feature vector to obtain a feature weight distribution; perform dimensionality reduction processing on the fault feature vector through a principal component analysis algorithm to obtain a reduced-dimensional feature vector; perform data normalization processing on the reduced-dimensional feature vector to obtain a standardized feature vector; perform sample weight adjustment on the standardized feature vector through an adaptive boosting algorithm to obtain a weighted training sample; perform random sampling on the weighted training sample to obtain a plurality of sub-training sets; perform classifier training on each sub-training set through a decision tree algorithm to obtain a plurality of basic classifiers; integrate the plurality of basic classifiers to obtain a random forest classifier; perform performance evaluation on the random forest classifier through a cross-validation method to obtain target model parameters; perform parameter correction on the random forest classifier according to the target model parameters to obtain a random forest model; perform predictive classification on the standardized feature vector through the random forest model to obtain a preliminary fault diagnosis result.
[0015] In the present invention, the analysis module is specifically used to: perform spatial mapping on the preliminary fault diagnosis results to obtain the fault point topology data corresponding to the target distribution network; construct a minimum spanning tree for the fault point topology data through a minimum tree generation algorithm to obtain a network backbone structure; perform a depth-first search on the network backbone structure to obtain a path set from the fault point to each key node in the key node set of the distribution network; perform shortest path calculation on the path set through the Floyd algorithm to obtain an optimal isolation path; perform node importance evaluation on the optimal isolation path to obtain a key isolation node sequence; sort the key isolation node sequence to obtain a node priority list; perform cluster analysis on the node priority list to obtain multiple isolation area candidate solutions; optimize the multiple isolation area candidate solutions through a genetic algorithm to obtain a target isolation area solution; perform load balancing analysis on the target isolation area solution to obtain a load redistribution strategy; and perform strategy correction on the load redistribution strategy through Monte Carlo simulation to obtain the target isolation strategy.
[0016] In the present invention, the simulation module is specifically used to: digitally model the target isolation strategy to obtain a digital twin model of the distribution network; perform fault simulation on the digital twin model of the distribution network through an electromagnetic transient simulation algorithm to obtain fault response characteristics; perform timing analysis on the fault response characteristics to obtain a key time node sequence; perform network state prediction on the key time node sequence through a state estimation algorithm to obtain network state changes during the isolation process; perform stability analysis on the network state changes to obtain system stability indicators; weigh the system stability indicators through a multi-objective optimization algorithm to obtain an optimal isolation operation sequence; perform time discretization processing on the optimal isolation operation sequence to obtain discrete time control instructions; perform smoothing processing on the discrete time control instructions through a fuzzy control algorithm to obtain a continuous control curve; perform digital signal processing on the continuous control curve to obtain a digital control signal; encapsulate the digital control signal through a real-time control protocol to obtain the target isolation operation instruction, and isolate and control the target distribution network through the target isolation operation instruction.
[0017] The present invention also provides a method for intelligent diagnosis and isolation of distribution network line faults, comprising:
[0018] Inject and collect high-frequency signals into a distribution network key node set of a pre-acquired target distribution network to obtain high-frequency response data, wherein the high-frequency response data includes signal characteristic data within a range of 50kHz-500kHz;
[0019] Synchronously sampling and preprocessing the high frequency response data to obtain reconstructed feature data;
[0020] Performing short-time Fourier transform and wavelet transform on the reconstructed feature data to obtain a fault feature vector;
[0021] Classifying the fault feature vector to obtain a preliminary fault diagnosis result;
[0022] Performing graph theory path simulation analysis on the preliminary fault diagnosis results to obtain a target isolation strategy;
[0023] The target isolation strategy is subjected to real-time digital simulation to obtain a target isolation operation instruction, and the target distribution network is isolated and controlled by the target isolation operation instruction.
[0024] In the technical solution provided by the present invention, the acquisition module adopts high-frequency signal injection technology to perform signal injection and acquisition on key nodes of the distribution network within the range of 50kHz-500kHz. This frequency range effectively avoids the fundamental frequency and harmonic interference of the power system, improves the signal-to-noise ratio of the signal, and provides high-quality original data for subsequent analysis. The importance of nodes is sorted and grouped by hierarchical analysis and fuzzy C-means clustering algorithm, and the signal injection parameters are dynamically adjusted in combination with the adaptive particle swarm optimization algorithm, so as to realize adaptive signal injection for different network structures and improve the pertinence and effectiveness of data acquisition. The processing module adopts a variety of advanced signal processing technologies, such as wavelet threshold denoising, Butterworth filtering, empirical mode decomposition and Hilbert transform, etc., to perform comprehensive preprocessing and feature extraction on the collected high-frequency response data, effectively remove noise interference, and extract the essential information of fault characteristics. The transformation module performs time-frequency domain analysis on the reconstructed feature data by combining short-time Fourier transform and wavelet transform, fully explores the characteristics of the fault signal in the time and frequency dimensions, and provides rich feature vectors for subsequent fault diagnosis. The classification module uses the integrated random forest algorithm, combined with feature importance evaluation, principal component analysis and adaptive boosting techniques to construct a fault classifier, which can accurately identify various complex fault types and significantly improve the accuracy and robustness of fault diagnosis. The analysis module innovatively introduces graph path simulation analysis, constructs the network backbone structure through the minimum spanning tree algorithm, and combines depth-first search and shortest path algorithms to quickly determine the optimal isolation path. At the same time, the genetic algorithm is used to optimize the isolation scheme, considering load balance and system stability, and formulates the optimal isolation strategy, which effectively reduces the scope of fault impact. Finally, the simulation module constructs a digital twin model of the distribution network, combines electromagnetic transient simulation and multi-objective optimization algorithm, and performs real-time simulation and optimization of the isolation strategy to generate accurate isolation operation instructions. Throughout the process, the modules work closely together to form a closed-loop intelligent diagnosis and isolation system, which not only greatly improves the accuracy and speed of fault diagnosis, but also realizes the dynamic optimization of the isolation strategy, minimizing the impact of faults on the distribution network. In addition, the modular design and scalability of the algorithm of this method enable it to adapt to distribution networks of different scales and complexities, and has a wide range of applicability. Through real-time digital simulation and fuzzy control algorithm, the method can also smoothly perform isolation operations, reduce the impact during the isolation process, and improve the stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0026] Figure 1 It is a schematic diagram of an intelligent diagnosis and isolation device for distribution network line faults in an embodiment of the present invention;
[0027] Figure 2 The present invention is a flowchart of a method for intelligent diagnosis and isolation of distribution network line faults in an embodiment of the present invention. DETAILED DESCRIPTION
[0028] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0029] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.
[0030] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0031] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , Figure 1 Schematic diagram of an intelligent diagnosis and isolation device for distribution network line faults according to an embodiment of the present invention. Figure 1 As shown, including:
[0032] The acquisition module 101 is used to inject and acquire high-frequency signals into a distribution network key node set of a pre-acquired target distribution network to obtain high-frequency response data, wherein the high-frequency response data includes signal characteristic data within a range of 50kHz-500kHz;
[0033] The processing module 102 is used to synchronously sample and preprocess the high frequency response data to obtain reconstructed feature data;
[0034] The transformation module 103 is used to perform short-time Fourier transform and wavelet transform on the reconstructed feature data to obtain a fault feature vector;
[0035] The classification module 104 is used to classify the fault feature vector and obtain a preliminary fault diagnosis result;
[0036] The analysis module 105 is used to perform graph path simulation analysis on the preliminary fault diagnosis results to obtain a target isolation strategy;
[0037] The simulation module 106 is used to perform real-time digital simulation on the target isolation strategy, obtain the target isolation operation instruction, and isolate and control the target distribution network through the target isolation operation instruction.
[0038] It should be noted that the acquisition module 101 is used to inject and collect high-frequency signals on the key node set of the distribution network of the pre-acquired target distribution network. These key nodes are important connection points in the power grid, and faults often have a significant impact on these nodes. In order to detect these effects, a high-frequency signal generator is used to inject a high-frequency signal in the range of 50kHz to 500kHz. The frequency of this signal is selected because it can penetrate the complex structure of the distribution network and is sensitive to different types of faults. For example, assume that several high-frequency signal injectors are arranged at the connection points of a substation and multiple transmission lines in a distribution network. These injectors will inject signals according to the set frequency range (for example, 50kHz, 100kHz, 250kHz and 500kHz). Then, the response nodes in the distribution network will respond to these signals, and the high-speed data collector will collect the change data of the response signal within milliseconds. The key here is to ensure that the collected data covers the entire frequency range, so as to obtain complete high-frequency response data. These high-frequency response data include signal feature data in the range of 50kHz-500kHz, and the data contains information indicating the health status of the line, such as impedance change, reflection coefficient, attenuation, etc. These characteristic data provide the basis for subsequent signal processing and fault feature extraction.
[0039] The processing module 102 performs synchronous sampling and preprocessing on the collected high-frequency response data. Since the collected data may be out of sync, timestamp alignment must be performed first to ensure that all data points are on the same time base. A high-speed timer can be used for time synchronization here to align the data at different sampling points. For example, assuming that a large amount of high-frequency response data is collected over a period of time, there is a slight time deviation in the data due to the different clock accuracy of different collection nodes. By adding a timestamp to each data packet and using an accurate clock synchronization algorithm, such as the Network Time Protocol (NTP) or GPS clock synchronization technology, all data are aligned on the same time base.
[0040] After completing the time alignment, the data is subjected to noise suppression. High-frequency data is usually affected by noise such as electromagnetic interference, so the wavelet threshold denoising method is used. Wavelet threshold denoising is a method of suppressing noise by decomposing the signal into different frequency bands and then applying a specific threshold on each frequency band. This method effectively retains the details of the signal while removing most of the random noise. The ultimate goal of the processing module is to obtain reconstructed feature data, which reflects the main characteristics of the high-frequency response signal and provides a reliable input for subsequent feature extraction and analysis.
[0041] The processed data enters the transformation module 103, which performs short-time Fourier transform (STFT) and wavelet transform on the reconstructed feature data to obtain the fault feature vector. Short-time Fourier transform is a technology that converts time domain signals to frequency domain. It can retain time and frequency information at the same time by windowing the signal. Specifically, the reconstructed feature data is divided into multiple time windows, and the data in each window is Fourier transformed to obtain a short-time spectrum. For example, a high-frequency response signal is divided into 0.1 second windows, and Fourier transform is performed in each window to obtain the spectrum information of each time segment. In this way, not only can the frequency component of the signal at a certain moment be known, but also the change of the signal over time can be observed. At the same time, wavelet transform is also performed. Wavelet transform is a time-frequency analysis tool that decomposes the signal at multiple scales through a series of high and low frequency filters, and can capture the local and global features of the signal at the same time. Through wavelet transform, the multi-scale wavelet coefficients of the signal can be obtained, and these coefficients are sparsely processed to retain important signal components and remove noise and redundant information. Finally, through the inverse wavelet transform, a signal reconstruction containing time and frequency characteristics is obtained. The result of these operations is a multi-dimensional fault feature vector, which contains the main characteristics of the signal in time and frequency and can well reflect the nature of the fault.
[0042] After obtaining the fault feature vector, the classification module 104 classifies it to obtain a preliminary fault diagnosis result. First, the fault feature vector is evaluated for feature importance. Feature importance evaluation is to determine which features are most useful by calculating the contribution of each feature to fault identification. In practical applications, the principal component analysis (PCA) method can be used to reduce the dimension of the fault feature vector, retain the main feature components, and remove minor and redundant features. Next, the feature vector after dimension reduction is normalized to ensure that all features are on the same scale, thereby improving the performance of the classifier. The normalized feature vector is input into an adaptive boosting algorithm for sample weight adjustment to obtain a weighted training sample. The adaptive boosting algorithm is an integrated learning method that improves the classification accuracy by combining multiple basic classifiers. For example, in a distribution network, multiple fault feature vectors are input into an adaptive boosting algorithm for classifier training after dimension reduction by PCA. Multiple sub-training sets are constructed by random sampling of samples, and each sub-training set is classified using a decision tree algorithm, and finally a random forest classifier is obtained. Through the cross-validation method, the performance of the classifier is evaluated and the parameters are adjusted to obtain an optimized classification model. This model can classify new fault feature vectors and give preliminary fault diagnosis results.
[0043] The preliminary fault diagnosis results output by the classification module 104 enter the analysis module 105 for further analysis and optimization. The analysis module first performs spatial mapping on the fault diagnosis results to generate topological data of the fault point. These topological data reflect the location and impact range of the fault in the distribution network. The minimum tree generation algorithm is used to construct the minimum spanning tree of the topological data to form a network backbone structure. The network backbone structure is subjected to a depth-first search to obtain a set of paths from the fault point to each key node. These paths represent possible paths propagating from the fault point to other nodes. The shortest path calculation of the path set is performed using the Floyd algorithm to obtain the optimal isolation path. These paths not only consider the shortest distance, but also comprehensively consider the power load and node importance to ensure the efficiency and safety of isolation. By evaluating the node importance of the optimal isolation path, a priority list of key isolation nodes is obtained, and cluster analysis is performed to form multiple candidate isolation area solutions. These solutions are optimized through a genetic algorithm to ultimately determine an optimal isolation area solution. In order to ensure the rationality of the isolation scheme, load balancing analysis is also carried out, and the Monte Carlo simulation method is used to correct the load redistribution strategy to obtain the final target isolation strategy.
[0044] Finally, the simulation module 106 performs real-time digital simulation on the target isolation strategy to obtain the target isolation operation instructions. The simulation module first digitally models the target isolation strategy and generates a digital twin model of the distribution network. This model is a high-fidelity digital replica of the distribution network that can simulate the real grid behavior. For example, in actual operation, the electromagnetic transient simulation algorithm is used to simulate the fault of the digital twin model to obtain the response characteristics of the fault. These response characteristics include voltage fluctuations, current changes, etc. These data are analyzed in time series to determine the changes in key time nodes. Through the state estimation algorithm, the network state prediction is performed on these key time nodes to obtain the changing trend of the network state during the isolation process.
[0045] The stability analysis of these changing trends is carried out to calculate the stability index of the system. In order to balance the system stability and the efficiency of the isolation operation, a multi-objective optimization algorithm is used to optimize the system stability index, and finally an optimal isolation operation sequence is obtained.
[0046] Through the collaborative work of the above modules, the acquisition module uses high-frequency signal injection technology to inject and collect signals from key nodes of the distribution network within the range of 50kHz-500kHz. This frequency range effectively avoids the fundamental frequency and harmonic interference of the power system, improves the signal-to-noise ratio of the signal, and provides high-quality original data for subsequent analysis. The importance of nodes is sorted and grouped through hierarchical analysis and fuzzy C-means clustering algorithm, and the signal injection parameters are dynamically adjusted in combination with the adaptive particle swarm optimization algorithm, so as to realize adaptive signal injection for different network structures and improve the pertinence and effectiveness of data acquisition. The processing module uses a variety of advanced signal processing technologies, such as wavelet threshold denoising, Butterworth filtering, empirical mode decomposition and Hilbert transform, to perform comprehensive preprocessing and feature extraction on the collected high-frequency response data, effectively remove noise interference, and extract the essential information of fault characteristics. The transformation module uses a combination of short-time Fourier transform and wavelet transform to perform time-frequency domain analysis on the reconstructed feature data, fully explore the characteristics of the fault signal in the time and frequency dimensions, and provide rich feature vectors for subsequent fault diagnosis. The classification module uses the integrated random forest algorithm, combined with feature importance evaluation, principal component analysis and adaptive boosting techniques to construct a fault classifier, which can accurately identify various complex fault types and significantly improve the accuracy and robustness of fault diagnosis. The analysis module innovatively introduces graph path simulation analysis, constructs the network backbone structure through the minimum spanning tree algorithm, and combines depth-first search and shortest path algorithms to quickly determine the optimal isolation path. At the same time, the genetic algorithm is used to optimize the isolation scheme, considering load balance and system stability, and formulates the optimal isolation strategy, which effectively reduces the scope of fault impact. Finally, the simulation module constructs a digital twin model of the distribution network, combines electromagnetic transient simulation and multi-objective optimization algorithm, and performs real-time simulation and optimization of the isolation strategy to generate accurate isolation operation instructions. Throughout the process, the modules work closely together to form a closed-loop intelligent diagnosis and isolation system, which not only greatly improves the accuracy and speed of fault diagnosis, but also realizes the dynamic optimization of the isolation strategy, minimizing the impact of faults on the distribution network. In addition, the modular design and scalability of the algorithm of this method enable it to adapt to distribution networks of different scales and complexities, and has a wide range of applicability. Through real-time digital simulation and fuzzy control algorithm, the method can also smoothly perform isolation operations, reduce the impact during the isolation process, and improve the stability of the system.
[0047] In a specific embodiment, the acquisition module 101 is specifically configured to perform the following steps:
[0048] (1) Perform hierarchical analysis on the electrical parameters corresponding to the key node set of the distribution network to obtain the importance ranking of the key nodes;
[0049] (2) Clustering nodes by sorting the importance of key nodes using the fuzzy C-means algorithm to obtain multiple node groups;
[0050] (3) Dynamically adjust the signal injection frequency of each node group through the adaptive particle swarm optimization algorithm to obtain the signal injection parameter set corresponding to each node group;
[0051] (4) Setting parameters of the high-frequency signal generator of the target distribution network according to the signal injection parameter set corresponding to each node group to obtain the target signal source;
[0052] (5) Power amplify the target signal source to obtain an injected signal;
[0053] (6) Perform power line coupling operation on the target distribution network according to the injected signal to obtain a line transmission signal;
[0054] (7) Perform high-speed sampling on the line transmission signal to obtain the original sampling data;
[0055] (8) The original sampled data is band-limited by a bandpass filter to obtain a response signal with a limited frequency band;
[0056] (9) Performing analog-to-digital conversion on the response signal of the limited frequency band to obtain digital high-frequency response data;
[0057] (10) The digitized high-frequency response data is subjected to spectrum analysis by fast Fourier transform to obtain signal characteristic data in the range of 50 kHz to 500 kHz.
[0058] Specifically, the key nodes of the distribution network refer to the nodes that play an important role in the distribution network, such as substations, important load centers, etc. For the electrical parameters corresponding to these key node sets, hierarchical analysis is required to determine the importance ranking of each node. The so-called hierarchical analysis is a method of hierarchical analysis of multiple decision criteria by constructing a judgment matrix. This method can decompose complex decision problems into multiple levels and analyze them layer by layer to obtain the importance weight of each node. For example, for multiple substations in a distribution network, a judgment matrix can be constructed considering factors such as load, geographical location, and historical fault records, and the relative importance of each node can be calculated through the Analytic Hierarchy Process (AHP).
[0059] These nodes are clustered using the Fuzzy C-Means (FCM) algorithm. The Fuzzy C-Means algorithm is a clustering algorithm that allows each data point to belong to multiple classes and assigns different memberships based on the distance from the data point to the class center. For example, in a distribution network, all key nodes can be clustered based on the node importance calculated in the previous step. Using the FCM algorithm, these nodes can be divided into several groups, each of which contains nodes that are close in importance. The purpose of this is to manage the frequency and parameters of signal injection in a hierarchical manner to improve the accuracy of fault diagnosis.
[0060] According to the characteristics of each node group, the frequency of signal injection is adjusted. This step uses the Adaptive Particle Swarm Optimization (APSO) algorithm. The particle swarm optimization algorithm is an optimization algorithm based on swarm intelligence. It simulates the process of bird flocks foraging and finds the optimal solution to the problem through mutual cooperation and competition between particles. The adaptive particle swarm optimization algorithm has been improved on this basis, so that it can dynamically adjust the speed and direction of particles according to the complexity of the problem and the change of the objective function, thereby improving the convergence speed and accuracy of the algorithm. In the specific scenario, each node group can be regarded as a particle group. Through the APSO algorithm, the signal injection frequency of each group is dynamically adjusted to obtain the optimal signal injection parameter set. With the signal injection parameter set, the parameters of the high-frequency signal generator can be set. The high-frequency signal generator is a device that can generate a specific frequency signal. By setting its parameters, a signal source within the target frequency range can be generated. In practical applications, the high-frequency signal generator can be set to multiple frequency bands according to the characteristics of each node group to ensure that each group can receive a high-frequency signal suitable for its characteristics.
[0061] In order to enhance the transmission effect of the signal, the target signal source needs to be power amplified. A power amplifier is a device that can amplify the power of the input signal. Through gain adjustment, it can amplify the power of the high-frequency signal to a level sufficient to overcome the transmission loss in the distribution network. For example, in the distribution network, long-distance cable transmission will cause signal attenuation. Through the processing of the power amplifier, it can be ensured that the injected signal still has sufficient strength after passing through the transmission path. The injected signal is transmitted in the target distribution network through power line coupling. Power line coupling is a technology that uses power lines as signal transmission media. It can transmit high-frequency signals to remote nodes through power lines. This process can be regarded as the propagation of high-frequency signals on power lines. The signal will be affected by factors such as line impedance and reflection, thereby generating line transmission signals.
[0062] In order to obtain the changes in these line transmission signals, they need to be sampled at high speed. A high-speed sampler is a device that can sample a signal multiple times in a very short time, and it can capture transient changes in the signal. For example, high-speed samplers are installed at multiple nodes of the distribution network to sample the line transmission signals at time intervals of milliseconds or even microseconds to obtain the original sampled data. However, the original sampled data usually contains a lot of useless frequency band information. In order to extract useful signal features, the original sampled data needs to be band-limited, and a bandpass filter is used here. A bandpass filter is a filter that can pass a signal within a specific frequency band. It can effectively filter out other frequency components and retain only the frequency range of interest. Specifically, the bandpass filter can be set in the range of 50kHz-500kHz, and the original sampled data can be filtered to obtain a response signal with a limited frequency band.
[0063] In order to further process these response signals, they need to be converted into digital signals. This step is achieved through an analog-to-digital converter (ADC). An analog-to-digital converter is a device that converts analog signals into digital signals. By sampling and quantizing analog signals, digital high-frequency response data can be obtained.
[0064] Finally, these digitized high-frequency response data are subjected to spectrum analysis, where Fast Fourier Transform (FFT) is used. Fast Fourier Transform is an algorithm that converts time domain signals to frequency domain. It can quickly calculate the amplitude and phase distribution of the signal at each frequency by performing discrete Fourier transform on the signal. For example, in practical applications, the digitized high-frequency response data is input into the FFT algorithm, the signal spectrum in the range of 50kHz-500kHz is calculated, and the characteristic data of the fault signal is extracted from it. These characteristic data include frequency components, amplitude changes, etc.
[0065] In a specific embodiment, the processing module 102 is specifically configured to perform the following steps:
[0066] (1) Perform time stamp alignment on the high-frequency response data to obtain synchronous sampling data;
[0067] (2) Use the wavelet threshold denoising algorithm to suppress the noise of the synchronous sampling data and obtain preliminary denoised data;
[0068] (3) Perform Fourier transform on the preliminary denoised data to obtain frequency domain representation;
[0069] (4) bandpass filtering the frequency domain representation through a Butterworth filter to obtain filtered frequency domain data;
[0070] (5) Perform inverse Fourier transform on the filtered frequency domain data to obtain time domain filtered data;
[0071] (6) Adaptively decompose the time-domain filtering data using the empirical mode decomposition algorithm to obtain multiple eigenmode functions;
[0072] (7) Performing Hilbert transform on multiple eigenmode functions to obtain instantaneous frequency and instantaneous amplitude;
[0073] (8) The trend of instantaneous frequency and instantaneous amplitude is extracted through the singular spectrum analysis algorithm to obtain the main characteristics of the signal;
[0074] (9) Perform principal component analysis on the main features of the signal to obtain target signal feature data;
[0075] (10) Reconstruct the target signal feature data through a support vector regression algorithm to obtain reconstructed feature data.
[0076] Specifically, the high-frequency response data collected from the key nodes of the distribution network are timestamped. The purpose of timestamp alignment is to ensure that the data at different sampling points can be compared and analyzed on the same time basis. In practical applications, high-frequency response data may come from multiple sampling points or multiple sampling devices. Due to the different clock accuracy of each device, there are slight time differences in the collected data. Therefore, it is necessary to synchronize these data. The network time protocol (NTP) or global positioning system (GPS) clock synchronization technology is used to add a timestamp to each data packet and align the data of different devices to a unified time basis to obtain synchronized sampling data. Assume that multiple high-frequency samplers are arranged in a distribution network. These samplers collect data thousands of times per second, but due to the clock error of the device, there is a time deviation in the data at each sampling point. Through timestamp alignment, these data can be adjusted to the same time axis, so that subsequent signal processing can be analyzed based on accurate time series.
[0077] The synchronous sampling data is subjected to noise suppression by using the wavelet threshold denoising algorithm. Wavelet threshold denoising is a method that decomposes the signal into different frequency bands through wavelet transform and then applies a specific threshold on each frequency band to suppress noise. Specifically, the wavelet transform decomposes the original signal into wavelet coefficients of different scales, which reflect the changes of the signal at different frequencies. By setting a threshold, the noise components smaller than the threshold are removed, the main part of the signal is retained, and then the denoised signal is reconstructed back through the inverse wavelet transform.
[0078] For example, the high-frequency response data in a distribution network is subjected to wavelet transform to obtain high-frequency and low-frequency components. In the high-frequency part, due to the presence of noise such as electromagnetic interference, a threshold is set, and the wavelet coefficients less than the threshold are cleared to remove the noise. Then, the noise-reduced data is obtained through inverse wavelet transform, which retains the main characteristics of the signal while eliminating noise interference. The data that has been initially denoised is subjected to Fourier transform to convert it to the frequency domain. Fourier transform is a mathematical tool that converts time domain signals into frequency domain signals, which can analyze the components of the signal at different frequencies. Through Fourier transform, complex signals in the time domain can be represented as a superposition of a series of sine waves to obtain a frequency domain representation. This frequency domain representation can reveal the amplitude and phase changes of the signal at different frequencies, which helps to analyze the frequency characteristics of the signal.
[0079] Suppose that Fourier transform is performed on the high-frequency response data after noise reduction to obtain a spectrum diagram. In this spectrum diagram, the horizontal axis represents frequency and the vertical axis represents amplitude. The intensity distribution of the signal at different frequencies can be clearly seen, such as the signal peaks at frequencies such as 50kHz, 150kHz, and 300kHz, which reflect the main frequency components in the signal. Use the Butterworth filter to bandpass filter the frequency domain representation. The Butterworth filter is a filter with a flat frequency response that can effectively filter out high-frequency and low-frequency noise and retain the effective frequency band of the signal. Specifically, the bandpass range of the Butterworth filter is set between 50kHz and 500kHz, and the frequency domain signal is filtered. In this way, noise and interference signals beyond this range can be removed to obtain filtered frequency domain data.
[0080] In actual operation, a Butterworth filter is applied to high-frequency response data with a frequency range of 0kHz to 1MHz, and the frequency components below 50kHz and above 500kHz are filtered out, retaining the signal in the middle frequency band. Through this processing, most of the low-frequency interference and high-frequency noise can be eliminated, and a relatively clean frequency domain signal can be obtained. Then, the filtered frequency domain data is subjected to an inverse Fourier transform to convert it back to the time domain. The inverse Fourier transform is the inverse process of the Fourier transform. By converting the frequency domain signal back to the time domain, the change of the signal over time can be restored. In this way, the time domain filtered data can be obtained, which reflects the main characteristics and change trends of the signal over time.
[0081] By performing Empirical Mode Decomposition (EMD) on these time-domain filtered data, they can be decomposed into multiple Intrinsic Mode Functions (IMFs). Empirical Mode Decomposition is an adaptive signal decomposition method that can decompose complex time-domain signals into several simple oscillation components, each of which is called an Intrinsic Mode Function. The EMD method does not require a preset basis function and can be decomposed according to the characteristics of the signal itself to obtain a set of Intrinsic Mode Functions, which reflect the changes of the signal on different time scales. For example, the time-domain filtered data in a distribution network is decomposed by EMD to obtain several IMFs. These IMFs correspond to different time scales, for example, the first IMF may reflect high-frequency noise, the second IMF may reflect the changes of the intermediate-frequency signal, and the third IMF may reflect the low-frequency trend. Through this decomposition, the characteristics of the signal on different time scales can be analyzed separately, providing rich information for subsequent analysis.
[0082] Perform Hilbert transform on multiple intrinsic mode functions to obtain instantaneous frequency and instantaneous amplitude. Hilbert transform is a mathematical tool used to analyze the time-varying characteristics of a signal. By performing Hilbert transform on a signal, the envelope and instantaneous frequency of the signal can be obtained. The instantaneous frequency is the frequency of the signal at a certain moment, and the instantaneous amplitude is the amplitude of the signal at a certain moment. This information can reveal the temporal variation of the signal. In practical applications, Hilbert transform is performed on each IMF to calculate its instantaneous frequency and instantaneous amplitude. For example, for an IMF that reflects the change of an intermediate frequency signal, its instantaneous frequency and amplitude at different time points can be obtained. These data can help understand the frequency variation of the signal over time, as well as the trend of the signal strength over time.
[0083] Then, the trend of instantaneous frequency and instantaneous amplitude is extracted through the singular spectrum analysis (SSA) algorithm. Singular spectrum analysis is a time series analysis method based on singular value decomposition. It extracts the main trend and periodic components of the signal by performing singular value decomposition on the embedded matrix of the time series. Specifically, the instantaneous frequency and amplitude data are converted into a matrix, and the matrix is subjected to singular value decomposition to obtain the main characteristics and trends of the signal.
[0084] For example, when performing SSA analysis on the instantaneous frequency data in a distribution network, it is found that its main frequency components are concentrated in several specific frequency ranges, and these frequency components have certain periodic changes in time. Through this analysis, the main trend of the signal can be extracted and possible fault characteristics can be identified.
[0085] Perform principal component analysis (PCA) on the main features of the signal to obtain the target signal feature data. Principal component analysis is a dimensionality reduction technique that finds the main components of the data and removes redundancy and noise by performing linear transformation on multidimensional data. Specifically, the main features obtained from singular spectrum analysis are used as input, and the principal components of the signal are extracted through PCA to obtain a low-dimensional feature vector. For example, PCA analysis is performed on a signal data containing multiple frequency and amplitude features to extract the most important principal components. These principal components reflect the main characteristics of the signal and reduce the dimension of the data, making subsequent analysis and processing more concise and efficient.
[0086] Finally, the target signal feature data is reconstructed through the support vector regression (SVR) algorithm to obtain the reconstructed feature data. Support vector regression is a regression analysis method based on support vector machine. It can predict and reconstruct the input data by building a regression model. Specifically, the principal components extracted by PCA are used as input, and regression analysis is performed through the SVR model to reconstruct the complete signal feature data.
[0087] In practical applications, SVR regression is performed on data containing main frequency and amplitude features, and a support vector regression model is trained to predict the complete signal features. In this way, a reconstructed feature data set can be obtained, which contains the main frequency and amplitude information of the signal, and after noise suppression and feature extraction, the main features of the signal are retained.
[0088] In a specific embodiment, the transformation module 103 is specifically configured to perform the following steps:
[0089] (1) Window segmentation is performed on the reconstructed feature data to obtain multiple time window data segments;
[0090] (2) Performing windowing processing on the time window data fragment to obtain the windowed data fragment;
[0091] (3) Perform fast Fourier transform on the windowed data segment to obtain the short-time spectrum;
[0092] (4) Extract features from the short-time spectrum using a peak detection algorithm to obtain a set of frequency domain feature points;
[0093] (5) Perform discrete wavelet transform on the frequency domain feature point set to obtain multi-scale wavelet coefficients;
[0094] (6) The multi-scale wavelet coefficients are sparsely processed by the threshold method to obtain sparse representation data;
[0095] (7) Perform inverse wavelet transform on the sparse representation data to obtain the reconstructed signal;
[0096] (8) Perform time-frequency analysis on the reconstructed signal through Hilbert-Huang transform to obtain the instantaneous frequency and instantaneous amplitude;
[0097] (9) Extract statistical features of instantaneous frequency and instantaneous amplitude to obtain statistical features in the time-frequency domain;
[0098] (10) The principal component analysis algorithm is used to reduce the dimension of the time-frequency domain statistical features and obtain the fault feature vector.
[0099] Specifically, the purpose of window segmentation of the reconstructed feature data is to divide the long time series data into smaller time segments. Each segment can be regarded as a part of the signal, which is helpful for subsequent analysis. Time window segmentation is a common signal processing method. By setting a fixed time interval, long signal data is cut into multiple shorter segments. Assuming that a distribution network fault signal is analyzed and divided into time windows of 0.1 seconds each, the changes of the signal in different time periods can be effectively captured.
[0100] Next, these time window data fragments are windowed. Windowing is to reduce the boundary effect by multiplying a window function, thereby improving the spectrum resolution of the Fourier transform. Common window functions include Hanning window and Hamming window. The purpose of windowing is to reduce the spectrum leakage caused by data truncation. For example, applying the Hanning window function to each 0.1 second time window data fragment for windowing can gradually reduce the data to zero at the window boundary, thereby reducing the impact of boundary effects on spectrum analysis.
[0101] By performing a fast Fourier transform (FFT) on the windowed data segments, the time domain signal can be converted into a frequency domain signal. Specifically, for each windowed data segment, the FFT algorithm can be used to obtain the spectrum distribution of the signal in the time window, for example, the signal strength and frequency components in the range of 50kHz to 500kHz can be calculated. Through such a transformation, a short-time spectrum can be obtained, which reflects the frequency distribution of the signal in a specific time period. The peak detection algorithm is used to extract features from the short-time spectrum to find the main feature points in the frequency domain. The peak detection algorithm is an algorithm for detecting the peak position in a signal. It identifies the peak points in the spectrum by comparing the local maximum values of the signal strength. For example, a threshold can be set to detect the peak points in the spectrum that exceed the threshold. These points reflect the strongest frequency components in the signal. In this way, a set of frequency domain feature points can be obtained, which represents the main components of the signal in the frequency domain.
[0102] By performing discrete wavelet transform (DWT) on the frequency domain feature point set, the multi-scale characteristics of the signal can be further analyzed. Discrete wavelet transform is a wavelet-based signal decomposition method that can decompose the signal into wavelet coefficients of different scales, thereby analyzing the changes of the signal at different frequencies. By performing DWT on the frequency domain feature point set, multi-scale wavelet coefficients can be obtained, which reflect the characteristics of the signal in different frequency bands. For example, for a signal containing multiple frequency components, high-frequency, wavelet detail coefficients and low-frequency, approximate coefficients can be obtained through DWT decomposition. These coefficients can describe the changing characteristics of the signal at different scales.
[0103] Then, the multi-scale wavelet coefficients are thinned by the threshold method, which can retain the important components in the signal and remove noise and redundant information. The threshold method is a common signal denoising method. By setting a threshold, the wavelet coefficients smaller than the threshold are regarded as noise and set to zero. In this way, the low-energy components in the signal can be effectively removed and the main high-energy components can be retained. For example, a signal containing multiple wavelet coefficients is thinned by setting a threshold, and the wavelet coefficients smaller than the threshold are removed to obtain sparse representation data.
[0104] By performing an inverse wavelet transform on sparsely represented data, the wavelet coefficients can be converted back to the time domain to obtain a reconstructed signal. The inverse wavelet transform is the inverse process of the wavelet transform. By combining the wavelet coefficients, the representation of the signal in the time domain can be restored. For example, for a set of wavelet coefficients that has been sparsely processed, a reconstructed time domain signal can be obtained through an inverse wavelet transform. This signal retains the main features of the original signal while removing noise and redundant information.
[0105] By performing time-frequency analysis on the reconstructed signal using the Hilbert-Huang Transform (HHT), the instantaneous frequency and instantaneous amplitude of the signal can be obtained. The Hilbert-Huang Transform is a time-frequency analysis method for nonlinear and non-stationary signals. It decomposes the signal into several intrinsic mode functions (IMFs) through empirical mode decomposition (EMD), and then performs Hilbert transform on these IMFs to obtain the instantaneous frequency and amplitude of the signal. For example, performing HHT analysis on a reconstructed signal containing complex frequency components can obtain the instantaneous frequency and instantaneous amplitude of the signal at each time point. This information can reflect the frequency change and intensity change of the signal.
[0106] By extracting the statistical features of the instantaneous frequency and instantaneous amplitude, we can obtain the main statistical features of the signal in the time-frequency domain. Statistical feature extraction is a method of describing the characteristics of a signal by calculating the statistical quantities of the signal, including mean, variance, skewness, kurtosis, etc. For example, for a signal data containing multiple instantaneous frequencies and amplitudes, we can calculate the average frequency of the signal, the variance of the frequency change, the skewness and kurtosis of the amplitude distribution, and other statistics to obtain the statistical features in the time-frequency domain. These statistical features can effectively describe the changing patterns and main components of the signal.
[0107] Finally, the fault feature vector can be obtained by reducing the dimension of the statistical features in the time and frequency domains through principal component analysis (PCA). Principal component analysis is a data dimension reduction technology that converts high-dimensional data into low-dimensional data through linear transformation while retaining the main information in the data. For example, PCA analysis of a high-dimensional signal data containing multiple statistical features can extract the main components of the signal and convert the original high-dimensional data into a low-dimensional feature vector, which reflects the main change trend and fault characteristics of the signal.
[0108] For example, suppose a high-frequency fault signal in a distribution network is analyzed. First, the signal is segmented, each time window is 0.1 seconds long, and the Hanning window function is applied to each window for windowing. Then, the windowed data segments are fast Fourier transformed to obtain a spectrum. The peak detection algorithm is used to identify the main peak points in the spectrum, which reflect the main frequency components in the signal. Discrete wavelet transform is performed on these frequency domain feature point sets to obtain multi-scale wavelet coefficients. The wavelet coefficients less than the threshold are removed by the threshold method to retain the important components. Then, the reconstructed signal is obtained by inverse wavelet transform. The reconstructed signal is analyzed in time and frequency by Hilbert-Huang transform to obtain instantaneous frequency and instantaneous amplitude data. Statistical features are extracted from these data to calculate the main statistics of the signal, such as average frequency, variance, etc. Finally, the statistical features are reduced in dimension by principal component analysis to obtain a low-dimensional fault feature vector, which reflects the main fault characteristics of the signal and can be used for subsequent fault diagnosis and isolation decisions.
[0109] In a specific embodiment, the classification module 104 is specifically configured to perform the following steps:
[0110] (1) Evaluate the feature importance of the fault feature vector and obtain the feature weight distribution;
[0111] (2) Perform dimension reduction processing on the fault feature vector through the principal component analysis algorithm to obtain the reduced dimension feature vector;
[0112] (3) Perform data normalization on the reduced-dimensional feature vector to obtain a standardized feature vector;
[0113] (4) The sample weights of the standardized feature vectors are adjusted through an adaptive boosting algorithm to obtain weighted training samples;
[0114] (5) Randomly sample the weighted training samples to obtain multiple sub-training sets;
[0115] (6) Perform classifier training on each sub-training set using a decision tree algorithm to obtain multiple basic classifiers;
[0116] (7) Integrate multiple basic classifiers to obtain a random forest classifier;
[0117] (8) Evaluate the performance of the random forest classifier through the cross-validation method to obtain the target model parameters;
[0118] (9) Modify the parameters of the random forest classifier according to the target model parameters to obtain a random forest model;
[0119] (10) The standardized feature vector is predicted and classified using the random forest model to obtain preliminary fault diagnosis results.
[0120] Specifically, feature importance evaluation refers to evaluating the importance of each feature in fault diagnosis when analyzing the fault feature vector. The purpose of feature importance evaluation is to determine which features contribute most to fault classification, thereby optimizing the performance of the model. In practical applications, feature importance can be evaluated by a variety of methods, such as information gain, Gini index, feature selection algorithm, etc. Assume that in a distribution network, a fault vector including multiple features such as voltage, frequency, and power has been extracted. Through feature importance evaluation, it is found that the voltage fluctuation feature and the frequency fluctuation feature contribute the most to fault classification, while other features such as power change have little effect on fault classification. Based on this evaluation result, a weight can be assigned to each feature to form a feature weight distribution, so that these important features can be focused on in subsequent model training. Principal Component Analysis (PCA) is used to reduce the dimensionality of the fault feature vector. PCA is a statistical technique that maps high-dimensional data to a new low-dimensional space through orthogonal transformation while retaining the main variation information of the data. Through PCA, the number of features can be reduced while retaining the information of the original data to the greatest extent. Assuming there is a fault vector containing 10 features, PCA dimensionality reduction can be used to reduce it to a new feature vector containing 3 main components. These 3 main components reflect the main change trends of the fault features, which can effectively reduce the computational complexity and improve the generalization ability of the model.
[0121] In order to further standardize the data, the feature vector after dimensionality reduction is normalized. Normalization refers to converting data to the same scale range, usually standardizing the data to a standard normal distribution with a mean of 0 and a standard deviation of 1. Normalization helps to eliminate the scale differences between features and improve the convergence speed and classification effect of the model. Assume that in the feature vector after dimensionality reduction, the voltage fluctuation range is [-100, 100], and the frequency fluctuation range is [-0.5, 0.5]. Through normalization, these features can be converted to the range of [-1, 1], so that different features have the same weight during the model training process. The sample weights of the standardized feature vector are adjusted by the adaptive boosting algorithm (Adaptive Boosting, AdaBoost). AdaBoost is an ensemble learning algorithm that iteratively adjusts the weights of samples and strengthens those misclassified samples, so that the model can pay more attention to samples that are difficult to classify. Specifically, in each round of iteration, AdaBoost adjusts the weights of samples according to the error rate of the classifier, increases the weights of those misclassified samples, and retrains the classifier. Suppose that in the distribution network fault diagnosis, it is found that some features perform poorly on certain fault categories. By adjusting the weights of these samples through the AdaBoost algorithm, the model can better learn these difficult-to-classify samples in the next round of training, thereby improving the classification accuracy. The weighted training samples are randomly sampled to obtain multiple sub-training sets. Random sampling refers to randomly extracting samples from the original training set to form multiple sub-training sets, which can be used to train different classifiers. The purpose of random sampling is to increase the diversity of data and reduce the risk of overfitting. Suppose there is a fault diagnosis training set containing 1,000 samples. Through random sampling, 10 sub-training sets containing different samples can be generated, each containing 100 samples.
[0122] The decision tree algorithm is used to train the classifier for each sub-training set to obtain multiple basic classifiers. The decision tree is a classification model based on a tree structure. It recursively splits the data to build a decision tree model for classification. Each node represents the judgment condition of a feature, and the leaf node represents the classification result. Assume that in a sub-training set, a tree model is constructed by the decision tree algorithm. Each branch of the model represents the judgment path of different features, and finally a fault classification result is obtained. By training different decision tree models for each sub-training set, multiple basic classifiers can be obtained. Multiple basic classifiers are integrated to obtain a random forest classifier. Random forest is an ensemble learning method based on decision trees. It integrates multiple decision tree classifiers and obtains the final classification result by voting or averaging. The advantage of random forest is that it can form a strong classifier by integrating multiple weak classifiers, thereby improving the accuracy and stability of classification. Assume that there are 10 decision tree classifiers. These 10 classifiers can be integrated by the random forest algorithm. For each new sample, the random forest will obtain the final classification result by majority voting.
[0123] The performance of the random forest classifier is evaluated by the cross-validation method to obtain the target model parameters. Cross-validation is a model evaluation method that evaluates the performance of the model by dividing the dataset into a training set and a validation set, performing multiple training and validations. The purpose of cross-validation is to reduce the bias of model evaluation and improve the generalization ability of the model. Assume that the dataset is divided into 5 parts, 4 parts are used for training each time, and 1 part is used for validation, and this is repeated 5 times to obtain the average performance of the model. Through cross-validation, the accuracy, precision, recall and other performance indicators of the random forest classifier can be evaluated to obtain the optimal model parameters.
[0124] According to the target model parameters, the random forest classifier is parameterized to obtain a random forest model. Parameter modification refers to adjusting the model's hyperparameters, such as the number of trees and maximum depth, based on the optimal parameters obtained by cross-validation. Assume that in cross-validation, it is found that the optimal parameters of the random forest are 50 decision trees with a maximum depth of 10. Through parameter modification, a random forest model with optimal parameters can be constructed.
[0125] The random forest model is used to predict and classify the standardized feature vectors to obtain preliminary fault diagnosis results. The random forest model can predict and classify new samples. According to the input feature vector, the model will perform multiple votes or averages to obtain a fault classification result. Assuming that a new distribution network fault feature vector is input, a preliminary fault diagnosis result can be obtained through the prediction of the random forest model, such as the fault type of "short circuit" or "disconnection".
[0126] For example, suppose that in a distribution network, a set of high-frequency fault signals are collected by sensors. First, through feature importance evaluation, it is found that voltage fluctuation and frequency fluctuation are the most important features. Then, through principal component analysis, the original 10 features are reduced to 3 main components. Then, the reduced features are normalized to make them on the same scale. Through the adaptive boosting algorithm, the weights of the samples are adjusted to generate weighted training samples. The weighted training samples are randomly sampled to generate 10 sub-training sets, and 10 basic classifiers are trained through the decision tree algorithm. Then, these basic classifiers are integrated through the random forest algorithm to obtain a random forest classifier. The model performance is evaluated through the cross-validation method, the optimal parameters are obtained, and the parameters are corrected. Finally, the new fault feature vector is predicted and classified through the random forest model to obtain preliminary fault diagnosis results, such as judging that the signal belongs to a "short circuit fault".
[0127] In a specific embodiment, the analysis module 105 is specifically configured to perform the following steps:
[0128] (1) Perform spatial mapping on the preliminary fault diagnosis results to obtain the fault point topology data corresponding to the target distribution network;
[0129] (2) Use the minimum tree generation algorithm to construct the minimum spanning tree of the fault point topology data to obtain the network backbone structure;
[0130] (3) Perform a depth-first search on the network backbone structure to obtain a set of paths from the fault point to each key node in the distribution network key node set;
[0131] (4) Using the Floyd algorithm to calculate the shortest path of the path set, we can obtain the optimal isolation path.
[0132] (5) Evaluate the node importance of the optimal isolation path and obtain the key isolation node sequence;
[0133] (6) Sort the key isolation node sequence to obtain a node priority list;
[0134] (7) Perform cluster analysis on the node priority list to obtain multiple candidate isolation area solutions;
[0135] (8) Optimizing multiple isolation area candidate solutions through genetic algorithms to obtain the target isolation area solution;
[0136] (9) Perform load balancing analysis on the target isolation area solution and obtain a load redistribution strategy;
[0137] (10) The load redistribution strategy is modified through Monte Carlo simulation to obtain the target isolation strategy.
[0138] Specifically, the preliminary fault diagnosis results are spatially mapped to map the diagnosed fault information to the actual geographical location and network topology of the distribution network. Spatial mapping is a method of converting data from abstract space to physical space, which can help determine the specific location and impact range of the fault. In the distribution network, spatial mapping can combine the fault diagnosis results with the Geographic Information System (GIS) data of the distribution network to obtain the specific fault point topology data in the target distribution network. Assuming that in a distribution network, the fault diagnosis results show that a certain transmission line has a fault, through spatial mapping, the specific location of the fault can be determined, such as the middle section of a certain transmission line, and combined with the geographical location data of the distribution network to form a topological map of the fault point. The minimum spanning tree (MST) of the fault point topological data is constructed by the minimum tree generation algorithm. The minimum spanning tree is a tree structure that can cover all nodes and has the smallest edge weight. It can minimize the total connectivity cost of the network while ensuring that all nodes are connected. The purpose of constructing the minimum spanning tree is to determine the network backbone structure of the distribution network, so as to find the shortest path between the fault point and other key nodes. Assuming that there are multiple substations and transmission lines in a distribution network, a minimum spanning tree structure covering all substations and transmission lines can be constructed through the minimum spanning tree algorithm. This structure can reflect the trunk path of the distribution network and minimize the total cost of network connectivity.
[0139] By performing a depth-first search (DFS) on the network backbone structure, a set of paths from the fault point to each key node in the distribution network key node set can be obtained. Depth-first search is an algorithm that traverses or searches a tree or graph. It searches along the depth of the tree until a leaf node is found, and then backtracks to search for the next node. Through depth-first search, all paths from the fault point to other important nodes in the distribution network can be found. Assuming that in a distribution network, the fault point is located in the middle of a transmission line, all possible paths from the fault point to each substation can be found through depth-first search. These paths constitute a complete set of paths, reflecting the connectivity relationship between the fault point and other key nodes. The shortest path calculation of the path set can be performed through the Floyd-Warshall Algorithm to obtain the optimal isolation path. The Floyd algorithm is a dynamic programming algorithm that can find the shortest path between all node pairs in the multi-source shortest path problem. The algorithm continuously updates the distance between nodes and finally obtains a shortest path matrix. The Floyd algorithm can be used to calculate the shortest path from the fault point to other key nodes in the distribution network. For example, if there are multiple substations and transmission lines in a distribution network, the Floyd algorithm can be used to calculate the shortest path from the fault point to each substation and determine the optimal isolation path, thereby achieving rapid isolation and recovery of the fault.
[0140] By evaluating the node importance of the optimal isolation path, a sequence of key isolation nodes can be obtained. Node importance evaluation refers to determining which nodes play a key role in the isolation process by calculating the importance of each node in the network. The evaluation method can use indicators such as degree centrality and betweenness centrality. Degree centrality refers to the degree of direct connectivity between a node and other nodes, while betweenness centrality refers to the number of paths of a node in the network. Through node importance evaluation, it can be determined which nodes have the greatest impact on the connectivity of the network during the isolation process. Assume that in a distribution network, it is found through evaluation that some substations and transmission lines play a key role in the fault isolation process. These nodes need to be given priority for isolation and restoration, and a sequence of key isolation nodes is obtained. The sequence of key isolation nodes is sorted to obtain a node priority list. The node priority list refers to sorting the nodes according to their importance to determine the priority of each node in the isolation process. Nodes with high priority need to be isolated and restored first to ensure the stability and connectivity of the network. By sorting the priorities of key isolation nodes, it can be determined which nodes need to be processed first when a fault occurs, thereby improving the efficiency of fault isolation. Assume that in a distribution network, by sorting the key nodes, it is determined that certain substations need to be isolated and restored first, forming a node priority list.
[0141] Cluster analysis of the node priority list can obtain multiple candidate isolation area solutions. Cluster analysis is a data mining technology that groups data and divides similar samples into the same group. Through cluster analysis, the nodes in the node priority list can be divided into multiple groups, each of which represents a candidate isolation area solution. Each isolation area represents a possible fault isolation solution. By analyzing these solutions, the optimal isolation area can be determined. Assuming that in a distribution network, multiple substations and transmission lines can be divided into different isolation areas through cluster analysis, each of which represents an independent fault isolation solution. By optimizing multiple isolation area candidate solutions through genetic algorithms, the target isolation area solution can be obtained. Genetic algorithm is an optimization algorithm that simulates natural selection and genetic mutation. It continuously evolves through operations such as selection, crossover, and mutation to find the optimal solution to the problem. Through genetic algorithms, multiple isolation area candidate solutions can be optimized to find an optimal isolation area solution that can minimize network loss and isolation cost. Assuming that in a distribution network, different isolation solutions are optimized through genetic algorithms, and finally an optimal isolation area solution is obtained, which can maximize the connectivity and stability of the network while minimizing the isolation cost.
[0142] By performing a load balancing analysis on the target isolation area solution, a load redistribution strategy can be obtained. Load balancing analysis refers to determining a reasonable way to distribute the load by analyzing the load conditions in the network to ensure the smooth operation of the network. During the fault isolation process, the load may be transferred from the fault area to other areas, so load redistribution is required to ensure that other areas can bear the transferred load to avoid overload and power outages. Through load balancing analysis, it can be determined how to reasonably distribute the load to other nodes during the isolation process to ensure the stable operation of the network. Assuming that in a distribution network, through load balancing analysis, it can be determined that the load in the fault area needs to be transferred to other substations, and the amount of load that each substation needs to bear is determined to form a load redistribution strategy.
[0143] By modifying the load redistribution strategy through Monte Carlo simulation, a target isolation strategy can be obtained. Monte Carlo simulation is a simulation technology that simulates system behavior through a large number of random samples. It can predict and analyze the behavior of the system under uncertain conditions. Through Monte Carlo simulation, the effect of load redistribution strategy under different conditions can be simulated to find the optimal strategy correction plan. Assuming that in a distribution network, through Monte Carlo simulation, the effect of different load redistribution strategies under various fault conditions can be simulated, the optimal load distribution plan can be found, and the strategy can be modified to finally form a target isolation strategy. This strategy can quickly and effectively perform fault isolation and load redistribution when a fault occurs, ensuring the stable operation of the distribution network.
[0144] In a specific embodiment, the simulation module 106 is specifically configured to perform the following steps:
[0145] (1) Digitally model the target isolation strategy to obtain a digital twin model of the distribution network;
[0146] (2) Use electromagnetic transient simulation algorithm to simulate the fault of the distribution network digital twin model and obtain the fault response characteristics;
[0147] (3) Perform timing analysis on the fault response characteristics to obtain the key time node sequence;
[0148] (4) Use the state estimation algorithm to predict the network state of the key time node sequence and obtain the network state changes during the isolation process;
[0149] (5) Perform stability analysis on network status changes to obtain system stability indicators;
[0150] (6) The system stability indicators are weighed through a multi-objective optimization algorithm to obtain the optimal isolation operation sequence;
[0151] (7) Performing time discretization on the optimal isolation operation sequence to obtain discrete time control instructions;
[0152] (8) The discrete time control instructions are smoothed by fuzzy control algorithm to obtain a continuous control curve;
[0153] (9) Performing digital signal processing on the continuous control curve to obtain a digital control signal;
[0154] (10) The digital control signal is encapsulated through a real-time control protocol to obtain a target isolation operation instruction, and the target distribution network is isolated and controlled through the target isolation operation instruction.
[0155] It should be noted that, for example, suppose a city's distribution network is being modeled, which includes multiple substations, trunk lines, distributed energy resources (such as solar panels and wind turbines), and hundreds of power loads. The digital twin model not only takes into account the static characteristics of these devices, but also dynamically reflects various operations and events that may occur in actual operation. The established digital twin model is fault simulated using an electromagnetic transient simulation algorithm. This simulation technology simulates various possible fault scenarios, such as short circuits, overloads, or equipment damage. For example, if a short circuit occurs in a substation in the model, the simulation algorithm will simulate the rapid increase in current, the sudden change in voltage, and its impact on surrounding nodes and equipment when the short circuit occurs.
[0156] Perform a detailed timing analysis on the fault response characteristics obtained by simulation. Timing analysis helps determine the sequence of key time nodes after the fault occurs, such as the time when the fault is first detected, the time trend of current and voltage changes at each node, and the time when the device responds. This analysis not only helps understand how the fault evolves, but also provides key information for the formulation of subsequent isolation control strategies. Use state estimation algorithms to predict the network status of these key time node sequences. The state estimation algorithm combines real-time collected data with model prediction capabilities to infer the voltage stability, equipment operating status, and possible fault propagation paths of each node in the future time period. For example, by combining model predictions and real-time measurement data, the state estimation algorithm can predict the voltage stability of surrounding nodes and changes in equipment operating status after a fault occurs.
[0157] After predicting the network status changes, conducting system stability analysis becomes a key step. This analysis evaluates the stability of the overall system operation after a fault occurs, taking into account the impact of various possible isolation operation plans on the system. For example, the analysis may include an assessment of voltage, frequency, and load balance to determine under what conditions the system may become unstable or maintain stability.
[0158] The multi-objective optimization algorithm is used to comprehensively weigh the system stability indicators to find the optimal isolation operation sequence. The multi-objective optimization algorithm takes into account various possible isolation schemes and optimizes the isolation operation sequence that can maximize the stability of the system under different circumstances. For example, the algorithm can consider isolating the faulty node in the shortest time while minimizing the impact of the system on other parts to ensure the stability and security of the system.
[0159] The optimal isolation operation sequence is processed in time discretization to generate specific discrete time control instructions. These instructions specify the execution time and operation object of each isolation step to ensure the effective execution of the isolation process. For example, the isolation time window of each node and the order of isolation execution are encoded into specific control commands for execution by the actual automation control system. The discrete time control instructions are smoothed using fuzzy control algorithms to generate continuous control curves. Fuzzy control technology can provide smooth transitions between discrete control instructions to ensure the consistency and stability of the entire isolation operation process. For example, the speed and transition mode of each isolation operation are adjusted through fuzzy logic control to cope with possible changes and uncertainties in the real-time environment. The generated digital control signals are encapsulated into target isolation operation instructions and applied to the actual distribution network to perform the required isolation operations. These instructions are transmitted to the automation control system through real-time control protocols to ensure that each isolation step is executed according to the optimal strategy and safety standards to minimize the impact of faults on system operation.
[0160] like Figure 2 As shown, Figure 2 1 is a flow chart of a method for intelligent diagnosis and isolation of distribution network line faults in an embodiment of the present invention, which specifically includes:
[0161] S201, injecting and collecting high-frequency signals into a distribution network key node set of a pre-acquired target distribution network to obtain high-frequency response data, wherein the high-frequency response data includes signal characteristic data within a range of 50kHz-500kHz;
[0162] S202, synchronously sampling and preprocessing the high frequency response data to obtain reconstructed feature data;
[0163] S203, performing short-time Fourier transform and wavelet transform on the reconstructed feature data to obtain a fault feature vector;
[0164] S204, classifying the fault feature vectors to obtain preliminary fault diagnosis results;
[0165] S205, performing graph theory path simulation analysis on the preliminary fault diagnosis results to obtain a target isolation strategy;
[0166] S206 , performing real-time digital simulation on the target isolation strategy, obtaining a target isolation operation instruction, and performing isolation control on the target distribution network through the target isolation operation instruction.
[0167] By executing the above steps, the acquisition module uses high-frequency signal injection technology to inject and collect signals from key nodes of the distribution network within the range of 50kHz-500kHz. This frequency range effectively avoids the fundamental frequency and harmonic interference of the power system, improves the signal-to-noise ratio of the signal, and provides high-quality original data for subsequent analysis. The importance of nodes is sorted and grouped by hierarchical analysis and fuzzy C-means clustering algorithm, and the signal injection parameters are dynamically adjusted by adaptive particle swarm optimization algorithm, which realizes adaptive signal injection for different network structures and improves the pertinence and effectiveness of data acquisition. The processing module uses a variety of advanced signal processing technologies, such as wavelet threshold denoising, Butterworth filtering, empirical mode decomposition and Hilbert transform, to perform comprehensive preprocessing and feature extraction on the collected high-frequency response data, effectively remove noise interference, and extract the essential information of fault characteristics. The transformation module uses a combination of short-time Fourier transform and wavelet transform to perform time-frequency domain analysis on the reconstructed feature data, fully explores the characteristics of the fault signal in the time and frequency dimensions, and provides rich feature vectors for subsequent fault diagnosis. The classification module uses the integrated random forest algorithm, combined with feature importance evaluation, principal component analysis and adaptive boosting techniques to construct a fault classifier, which can accurately identify various complex fault types and significantly improve the accuracy and robustness of fault diagnosis. The analysis module innovatively introduces graph path simulation analysis, constructs the network backbone structure through the minimum spanning tree algorithm, and combines depth-first search and shortest path algorithms to quickly determine the optimal isolation path. At the same time, the genetic algorithm is used to optimize the isolation scheme, considering load balance and system stability, and formulates the optimal isolation strategy, which effectively reduces the scope of fault impact. Finally, the simulation module constructs a digital twin model of the distribution network, combines electromagnetic transient simulation and multi-objective optimization algorithm, and performs real-time simulation and optimization of the isolation strategy to generate accurate isolation operation instructions. Throughout the process, the modules work closely together to form a closed-loop intelligent diagnosis and isolation system, which not only greatly improves the accuracy and speed of fault diagnosis, but also realizes the dynamic optimization of the isolation strategy, minimizing the impact of faults on the distribution network. In addition, the modular design and scalability of the algorithm of this method enable it to adapt to distribution networks of different scales and complexities, and has a wide range of applicability. Through real-time digital simulation and fuzzy control algorithm, the method can also smoothly perform isolation operations, reduce the impact during the isolation process, and improve the stability of the system.
[0168] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the embodiments, a person skilled in the art should understand that the specific implementation modes of the present invention can still be modified or replaced by equivalents, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.
Claims
1. An intelligent diagnosis and isolation device for distribution network line faults, characterized in that: include: An acquisition module is used to inject and acquire high-frequency signals into a set of key nodes of a distribution network of a target distribution network acquired in advance, and obtain high-frequency response data, wherein the high-frequency response data includes signal characteristic data within a range of 50kHz-500kHz; A processing module, used for synchronously sampling and preprocessing the high frequency response data to obtain reconstructed feature data; A transformation module, used for performing short-time Fourier transform and wavelet transform on the reconstructed feature data to obtain a fault feature vector; A classification module, used to classify the fault feature vector to obtain a preliminary fault diagnosis result; An analysis module, used to perform graph theory path simulation analysis on the preliminary fault diagnosis result to obtain a target isolation strategy; The analysis module is specifically used for: Performing spatial mapping on the preliminary fault diagnosis results to obtain fault point topology data corresponding to the target distribution network; A minimum spanning tree is constructed for the fault point topology data by using a minimum tree generation algorithm to obtain a network backbone structure; Performing a depth-first search on the network backbone structure to obtain a set of paths from the fault point to each key node in the key node set of the distribution network; Performing shortest path calculation on the path set by using the Floyd algorithm to obtain an optimal isolation path; Performing node importance evaluation on the optimal isolation path to obtain a key isolation node sequence; Sorting the key isolation node sequence to obtain a node priority list; Performing cluster analysis on the node priority list to obtain multiple isolation area candidate solutions; Optimizing the plurality of isolation area candidate solutions by using a genetic algorithm to obtain a target isolation area solution; Performing load balancing analysis on the target isolation area solution to obtain a load redistribution strategy; Modifying the load redistribution strategy through Monte Carlo simulation to obtain the target isolation strategy; The simulation module is used to perform real-time digital simulation on the target isolation strategy, obtain the target isolation operation instruction, and isolate and control the target distribution network through the target isolation operation instruction.
2. The intelligent diagnosis and isolation device for power distribution network line fault according to claim 1 is characterized in that: The acquisition module is specifically used for: Performing hierarchical analysis on electrical parameters corresponding to the key node set of the distribution network to obtain the importance ranking of the key nodes; The importance ranking of the key nodes is sorted by a fuzzy C-means algorithm to perform node clustering to obtain multiple node groups; The signal injection frequency of each node group is dynamically adjusted by an adaptive particle swarm optimization algorithm to obtain a signal injection parameter set corresponding to each node group; According to the signal injection parameter set corresponding to each node group, the high-frequency signal generator of the target distribution network is parameterized to obtain a target signal source; Performing power amplification processing on the target signal source to obtain an injection signal; Performing power line coupling operation on the target distribution network according to the injection signal to obtain a line transmission signal; Performing high-speed sampling on the line transmission signal to obtain original sampling data; The original sampling data is subjected to frequency band limitation by a bandpass filter to obtain a response signal of a limited frequency band; Performing analog-to-digital conversion on the response signal of the limited frequency band to obtain digital high-frequency response data; The digitized high-frequency response data is subjected to spectrum analysis by fast Fourier transform to obtain signal characteristic data within the range of 50kHz-500kHz.
3. The intelligent diagnosis and isolation device for power distribution network line fault according to claim 1 is characterized in that: Processing module, specifically used for: Performing time stamp alignment on the high frequency response data to obtain synchronous sampling data; Performing noise suppression on the synchronous sampling data by using a wavelet threshold denoising algorithm to obtain preliminary noise-reduced data; Performing Fourier transformation on the preliminary noise reduction data to obtain a frequency domain representation; Performing bandpass filtering on the frequency domain representation by using a Butterworth filter to obtain filtered frequency domain data; Performing an inverse Fourier transform on the filtered frequency domain data to obtain time domain filtered data; Adaptively decomposing the time domain filtering data by an empirical mode decomposition algorithm to obtain a plurality of eigenmode functions; Performing Hilbert transform on the multiple eigenmode functions to obtain instantaneous frequencies and instantaneous amplitudes; The trend of the instantaneous frequency and instantaneous amplitude is extracted by a singular spectrum analysis algorithm to obtain the main features of the signal; Performing principal component analysis on the main features of the signal to obtain target signal feature data; The target signal feature data is reconstructed by using a support vector regression algorithm to obtain the reconstructed feature data.
4. The intelligent diagnosis and isolation device for power distribution network line fault according to claim 1, characterized in that: Transformation module, specifically used for: Performing window segmentation on the reconstructed feature data to obtain a plurality of time window data segments; Performing windowing processing on the time window data segment to obtain a windowed data segment; Performing a fast Fourier transform on the windowed data segment to obtain a short-time spectrum; Extracting features from the short-time spectrum using a peak detection algorithm to obtain a frequency domain feature point set; Performing discrete wavelet transform on the frequency domain feature point set to obtain multi-scale wavelet coefficients; Performing sparse processing on the multi-scale wavelet coefficients by a threshold method to obtain sparse representation data; Performing inverse wavelet transform on the sparse representation data to obtain a reconstructed signal; Performing time-frequency analysis on the reconstructed signal by Hilbert-Huang transform to obtain instantaneous frequency and instantaneous amplitude; Extracting statistical features of the instantaneous frequency and instantaneous amplitude to obtain statistical features in the time-frequency domain; The time-frequency domain statistical features are subjected to dimensionality reduction processing by a principal component analysis algorithm to obtain the fault feature vector.
5. The intelligent diagnosis and isolation device for power distribution network line fault according to claim 1, characterized in that: Classification module, specifically used for: Performing feature importance evaluation on the fault feature vector to obtain feature weight distribution; Performing dimensionality reduction processing on the fault feature vector by using a principal component analysis algorithm to obtain a reduced-dimensional feature vector; Performing data normalization processing on the dimension-reduced feature vector to obtain a standardized feature vector; The sample weight of the standardized feature vector is adjusted by an adaptive boosting algorithm to obtain a weighted training sample; Randomly sampling the weighted training samples to obtain multiple sub-training sets; Classifier training is performed on each sub-training set through a decision tree algorithm to obtain multiple basic classifiers; Integrating the multiple basic classifiers to obtain a random forest classifier; Performing performance evaluation on the random forest classifier by a cross-validation method to obtain target model parameters; Modify the parameters of the random forest classifier according to the target model parameters to obtain a random forest model; The standardized feature vector is predicted and classified by the random forest model to obtain a preliminary fault diagnosis result.
6. The intelligent diagnosis and isolation device for power distribution network line fault according to claim 1, characterized in that: Simulation module, specifically used for: Digitally modeling the target isolation strategy to obtain a digital twin model of the distribution network; Performing fault simulation on the distribution network digital twin model through an electromagnetic transient simulation algorithm to obtain fault response characteristics; Performing a time series analysis on the fault response characteristics to obtain a key time node sequence; The network state is predicted for the key time node sequence by a state estimation algorithm to obtain the network state change during the isolation process; Performing stability analysis on the network state change to obtain a system stability index; The system stability index is weighed by a multi-objective optimization algorithm to obtain an optimal isolation operation sequence; Performing time discretization processing on the optimal isolation operation sequence to obtain discrete time control instructions; The discrete time control instruction is smoothed by a fuzzy control algorithm to obtain a continuous control curve; Performing digital signal processing on the continuous control curve to obtain a digital control signal; The digital control signal is encapsulated through a real-time control protocol to obtain the target isolation operation instruction, and the target distribution network is isolated and controlled through the target isolation operation instruction.
7. A method for intelligent diagnosis and isolation of distribution network line faults, characterized in that: include: Inject and collect high-frequency signals into a distribution network key node set of a pre-acquired target distribution network to obtain high-frequency response data, wherein the high-frequency response data includes signal characteristic data within a range of 50kHz-500kHz; Synchronously sampling and preprocessing the high frequency response data to obtain reconstructed feature data; Performing short-time Fourier transform and wavelet transform on the reconstructed feature data to obtain a fault feature vector; Classifying the fault feature vector to obtain a preliminary fault diagnosis result; Performing graph theory path simulation analysis on the preliminary fault diagnosis results to obtain a target isolation strategy; The preliminary fault diagnosis results are subjected to graph theory path simulation analysis to obtain a target isolation strategy, including: Performing spatial mapping on the preliminary fault diagnosis results to obtain fault point topology data corresponding to the target distribution network; A minimum spanning tree is constructed for the fault point topology data by using a minimum tree generation algorithm to obtain a network backbone structure; Performing a depth-first search on the network backbone structure to obtain a set of paths from the fault point to each key node in the key node set of the distribution network; Performing shortest path calculation on the path set by using the Floyd algorithm to obtain an optimal isolation path; Performing node importance evaluation on the optimal isolation path to obtain a key isolation node sequence; Sorting the key isolation node sequence to obtain a node priority list; Performing cluster analysis on the node priority list to obtain multiple isolation area candidate solutions; Optimizing the plurality of isolation area candidate solutions by using a genetic algorithm to obtain a target isolation area solution; Performing load balancing analysis on the target isolation area solution to obtain a load redistribution strategy; Modifying the load redistribution strategy through Monte Carlo simulation to obtain the target isolation strategy; The target isolation strategy is subjected to real-time digital simulation to obtain a target isolation operation instruction, and the target distribution network is isolated and controlled by the target isolation operation instruction.
