Power distribution network fault identification and positioning system based on wide area measurement technology

Abstract

The invention relates to the technical field of power system fault diagnosis, and discloses a power distribution network fault identification and positioning system based on a wide area measurement technology, and the system comprises a key measurement point optimization configuration module which determines the arrangement position of an optimal measurement point, and identifies a high-risk weak link in a power distribution network; the virtual-real fusion measurement network construction module adopts a generative adversarial network to generate blind area virtual measurement data conforming to a physical rule, and fuses actual measurement data and the virtual measurement data; the active fault excitation implementation module is used for implementing safe and controllable tiny disturbance injection and actively detecting response characteristics of weak links; the state estimation enhanced fault positioning module is used for constructing a health state evaluation model and calculating the deviation between a transfer function and a health baseline; the toughness fault positioning execution module is used for extracting key information dimensions and ensuring the reliability of fault positioning; through a virtual-real fusion measurement technology, efficient monitoring of the whole area of the power distribution network is realized on the basis of limited actual measurement equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of power system fault diagnosis, and more specifically, it relates to a distribution network fault identification and location system based on wide-area measurement technology. Background Art

[0002] With the continuous expansion of the scale of the distribution network and the increasing complexity of the network topology, traditional fault identification and location technologies are no longer able to meet the requirements of the reliable operation of modern distribution networks. Existing technologies mainly perform fault identification and location based on real-time monitoring or fault recording data. However, in the actual application process, three main technical problems are faced: First of all, although the key weak links of the distribution network (such as important tie lines, substation equipment, etc.) are high-fault areas, due to economic constraints, it is impossible to achieve full-network measurement coverage, resulting in insufficient monitoring data. The traditional distribution network measurement coverage rate is usually only 30% to 40%. A large number of nodes lack real-time monitoring, making it difficult to capture the characteristic information in the initial stage of the fault. There are inevitably measurement blind spots in the distribution network. The fault states in these areas are difficult to directly observe, increasing the uncertainty of fault location. The existence of measurement blind spots can increase the fault location error by 3 to 5 times. Especially for underground cables and complex branch lines, the location error may reach several kilometers. The measurement equipment itself may fail due to faults. In the case of partial loss of measurement data, the accuracy and reliability of traditional fault location methods will be reduced. In the case of partial failure of the measurement equipment (about 30%), the fault location accuracy rate of traditional methods drops below 50%, seriously affecting the fault recovery speed of the distribution network.

[0003] The above problems limit the application effect of existing technologies in complex distribution networks. There is an urgent need for a fault identification and location method that can improve the fault warning ability under the condition of limited measurement resources and still maintain a high accuracy rate in the case of partial failure of the measurement equipment. Summary of the Invention

[0004] The present invention provides a distribution network fault identification and location system based on wide-area measurement technology, which solves the technical problems of insufficient measurement coverage, the existence of blind spots and easy failure of equipment in the distribution network in related technologies, and low fault location accuracy.

[0005] The present invention provides a distribution network fault identification and location system based on wide-area measurement technology, including: A key measurement point optimization configuration module, which determines the best measurement point layout position through a sensitivity evaluation algorithm and identifies the high-risk weak links that need to be key monitored in the distribution network; Virtual-real fusion measurement network construction module, based on the optimal measurement point layout position, uses a generative adversarial network to generate virtual measurement data for blind spots that conform to physical laws, and fuses the actual measurement data and virtual measurement data through a dynamic weight method to form a hybrid measurement matrix covering the entire network; Active fault excitation implementation module, for high-risk weak links, implements safe and controllable small perturbation injection, actively detects the response characteristics of weak links, and obtains equipment health status information; State estimation enhanced fault location module, constructs a health status evaluation model based on equipment health status information, and calculates the deviation between the transfer function and the health baseline; Resilient fault location execution module, based on the hybrid measurement matrix, extracts key information dimensions in the case of uncertain measurement data or partial measurement point failures to ensure the reliability of fault location.

[0006] Furthermore, the determination of the optimal measurement point layout position by the sensitivity evaluation algorithm includes: Apply a multi-dimensional weak link identification algorithm to evaluate the risk degree of each node in the distribution network, calculate the risk score of each node, and the risk score comprehensively considers topological vulnerability, historical fault frequency, electrical stress, and node importance; Analyze the contribution degree of each candidate measurement point to network observability, construct an observation equation and calculate the observation sensitivity matrix of each candidate measurement point to obtain the measurement point sensitivity index; Based on the risk score and sensitivity index, construct an optimization model for measurement point configuration, and solve the optimization model to determine the final measurement point configuration plan.

[0007] Furthermore, the use of a generative adversarial network to generate virtual measurement data for blind spots that conform to physical laws includes: Utilize the electrical topology relationship and power flow constraints to construct a physical association model between measurement points and blind spot nodes; Train the generator and discriminator of the generative adversarial network. The generator receives random noise and electrical quantity information of surrounding measured points as inputs to generate virtual measurement data, and the discriminator discriminates the authenticity of the data; Introduce physical constraint regularization terms into the loss function of the generator, including Kirchhoff's voltage law constraint term, Kirchhoff's current law constraint term, and power flow equation constraint term.

[0008] Furthermore, the implementation of safe and controllable small perturbation injection for high-risk weak links includes: Construct a perturbation signal for active fault excitation. The perturbation signal adopts the form of a decaying sine wave, and the amplitude is set within a safe range that does not affect the normal operation of the system; Inject the perturbation signal into the key weak links of the distribution network safely through the power electronic interface device, and adopt the closed-loop control method to control the injection process; Monitor the key parameters of the system in real time, set the warning threshold and the emergency threshold, and trigger the corresponding protection actions when the monitored parameters exceed the thresholds.

[0009] Furthermore, the construction of the health status evaluation model according to the device health status information includes: Record the response of the device to the perturbation signal and obtain the system response transfer function; Calculate the deviation between the current transfer function and the healthy reference transfer function; Evaluate the health status score through multi-feature fusion. When the health status score is lower than the preset threshold, it is judged that the device has a potential failure risk and enters the warning state; Use time-domain and frequency-domain feature analysis methods to identify the precursor features of faults from the response signal and construct the precursor feature vector of faults.

[0010] Furthermore, the steps of extracting the key information dimension in the case of uncertain measurement data or failure of some measurement points include: Extract the most valuable fault features from the mixed measurement data, perform principal component analysis on the original high-dimensional features, and obtain the reduced-dimensional feature matrix; Evaluate the contribution of each principal component to the discrimination of fault types through the Fisher discrimination criterion; Select several principal components with the most discriminative power according to the contribution value to form the key information dimension set; Combine the credibility weights of multi-source data to evaluate the reliability of each measurement data. Calculate the credibility of the actual measurement data based on the device status and measurement residuals. For virtual measurement data, directly use its generated reliability as the credibility; Furthermore, ensuring the reliability of fault location includes: Construct a resilient fault location algorithm based on the extracted key information dimension and data credibility; The resilient fault location algorithm uses the weighted voting fusion method to identify the fault type, and sums the product of the credibility weight of each evidence and the probability estimate of the fault type based on the evidence; Construct a probability distribution map of the fault location, sum the product of the credibility weight of each evidence and the location probability estimate based on the evidence, and then divide by the sum of the credibility weights; Output the fault type, fault location and the corresponding credibility to support the decision-making of the operation and maintenance personnel.

[0011] Furthermore, the resilient fault location algorithm includes an evidence collection layer, an evidence preprocessing layer, a local inference layer and an evidence weighted fusion layer, specifically including: The evidence collection layer processes five types of evidence sources, namely actual measurement data, virtual measurement data, actively excited response characteristics, historical fault modes, and topological constraint information; The evidence preprocessing layer applies specific preprocessing methods to each type of evidence; The local inference layer independently analyzes each type of preprocessed evidence to generate preliminary estimates of the fault type and location; The evidence weighted fusion layer combines the credibility of each evidence source and uses a Bayesian network framework to fuse the local inference results to form the final fault judgment result.

[0012] Furthermore, the small perturbation injection adopts a safety injection control system, and the safety injection control system includes: A hybrid control architecture based on DSP and FPGA, where the DSP is responsible for upper-layer control strategies and safety monitoring, and the FPGA realizes high-speed waveform generation and real-time control; A closed-loop control parameter adaptive adjustment unit that automatically optimizes the controller parameters according to the system impedance characteristics; A multi-layer safety protection unit that monitors key parameters, including voltage, current, frequency, and harmonic content, and triggers corresponding protection actions when the monitored key parameters exceed the thresholds.

[0013] The present invention provides a storage medium, including a memory and one or more processors. An executable code is stored in the memory, and when the one or more processors execute the executable code, it is used to implement the above-mentioned distribution network fault identification and location system based on wide-area measurement technology.

[0014] The beneficial effects of the present invention are as follows: Through the virtual-real fusion measurement technology, on the basis of limited actual measurement devices, efficient monitoring of the entire distribution network is realized, especially for key weak links and measurement blind spots. The measurement coverage rate is improved compared with traditional methods, and the risk of missed fault detection caused by monitoring blind spots is reduced.

[0015] Discover potential faults in advance: Based on the active fault excitation technology of the present application, abnormal characteristics of equipment can be detected before a fault occurs, the early warning accuracy rate is improved, power outages are avoided by identifying fault precursors in advance, the power supply reliability is improved, and user power outage losses are reduced.

[0016] The fusion and optimization of the measurement point configuration and virtual measurement data improve the fault location accuracy rate under the same measurement point coverage rate. In the case of partial measurement device failures, a relatively high fault location accuracy rate can still be maintained, reducing the fault inspection time and scope.

[0017] Without a large number of additional physical measurement devices, the system performance is improved through software algorithms and perturbation techniques, the monitoring cost of each distribution network node is reduced, and the input-output ratio is high; In the case of measurement device failures, network topology changes, etc., the system of the present application can still maintain efficient operation, the calculation time of the positioning algorithm is reduced, the resilience and real-time response ability of the system are improved, and it can adapt to various complex operating environments. Description of the Drawings

[0018] Figure 1 is a module diagram of a distribution network fault identification and location system based on wide-area measurement technology in the present invention. Detailed Embodiments

[0019] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein, and the functions and arrangements of the elements discussed can be changed without departing from the scope of protection of the content of this specification. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0020] At least one embodiment of the present invention discloses a distribution network fault identification and location system based on wide-area measurement technology, as Figure 1 shown, including: A key measurement point optimization configuration module, which determines the optimal measurement point layout position through a sensitivity evaluation algorithm and identifies high-risk weak links that need to be key monitored in the distribution network; Using the sensitivity evaluation algorithm to analyze the distribution network topology and historical operation data to determine the optimal measurement point layout position, including the following sub-steps: Step 1.1, multi-dimensional weak link identification; The multi-dimensional weak link identification algorithm provided in this application evaluates the risk degree of each node in the distribution network, and calculates the risk score of each node according to the following risk scoring formula : ; Where represents the comprehensive risk score of node , is the topological vulnerability index of node , obtained through electrical topology map analysis; is the historical fault frequency index, statistically analyzed based on historical operation data; is the electrical stress index, calculated by the degree of deviation of electrical parameters from the rated value; is the node importance index, considering the importance degree of the node power supply load; are the weight coefficients of the topological vulnerability index, historical fault frequency index, electrical stress index, and node importance index respectively.

[0021] Optionally, in different distribution network environments, the weight coefficients of each index can be adjusted according to the actual situation. For example, in the core urban area with high requirements for power supply reliability, the weight of the node importance index can be increased to 0.4 - 0.5; while in the area with serious equipment aging, the weight of the electrical stress index can be increased to 0.3 - 0.4.

[0022] The topological vulnerability index can be further divided into multiple sub - indexes: ; Among them is the topological vulnerability index of node , is the node degree index, indicating the number of nodes directly connected to node ; is the node betweenness index, indicating the number of shortest paths passing through node ; is the node closeness index, indicating the reciprocal of the average distance from node to all other nodes; is the node closeness index, indicating the reciprocal of the average distance from node to all other nodes; are the weight coefficients of the node degree index, the node betweenness index, and the node closeness index respectively.

[0023] After the risk score calculation is completed, the system sorts the scores from high to low to identify the high - risk weak links that need to be key - monitored in the distribution network.

[0024] Step 1.2, Measurement point sensitivity assessment; Using the measurement point sensitivity assessment algorithm, analyze the contribution degree of each candidate measurement point to network observability; first, construct the observation equation: ; Among them is the measurement vector, containing the measurement values of all measurement points; is the state variable 's non - linear function; is the measurement error vector, indicating the random error of the measurement value; is the system state variable.

[0025] Then calculate the observation sensitivity matrix of each candidate measurement point : ; where is the observation sensitivity matrix, representing the sensitivity of the measured value to the state variables; is the partial derivative of the measurement function with respect to the state variables; is the initial estimate or operating point of the state variables; represents the calculation of the partial derivative when the state variables are equal to the initial estimate, is the system state variable; is the state variable 's non - linear function.

[0026] By analyzing 's eigenvalues and condition numbers, evaluate the improvement of the system observability by adding measurement points to obtain the measurement point sensitivity index .

[0027] Step 1.3, Optimized layout of measurement points; Based on the risk score and sensitivity index, construct an optimization model for the measurement point configuration: ; Budget constraint: ; High - risk node coverage constraint: ; where represents the set of positions where measurement devices can be installed; represents whether to install a measurement device at the candidate position, which is a 0 - 1 decision variable. 1 means installing a measurement device at position and 0 means not installing; is the sensitivity index of the measurement point ; represents maximizing the objective function; represents the cost of installing a measurement device at position ; represents the maximum budget available for installing measurement devices; represents all measurement points that can monitor the state of node ; is the set of high - risk nodes; represents "for all", indicating that the condition is satisfied for each element in the set; represents the summation operator.

[0028] By solving this optimization model, determine the final measurement point configuration plan to achieve optimal coverage of key weak links with a limited budget.

[0029] The virtual-real fusion measurement network construction module generates virtual measurement data for blind areas that conform to physical laws using a generative adversarial network based on the optimal measurement point layout positions, and fuses the actual measurement data and the virtual measurement data through a dynamic weight method to form a hybrid measurement matrix covering the entire network; It includes the following sub-steps: Step 2.1, physical association model construction; Using the electrical topological relationship and power flow constraints, construct a physical association model between the measurement points and the blind area nodes; based on Kirchhoff's laws of node voltage and branch current, establish the relationship between the node state variables and the surrounding measurement quantities: ; ; where is the branch current between nodes , indicating the current flowing from node to node ; is the corresponding admittance, indicating the admittance value between nodes and ; and respectively represent the complex voltages of nodes and node ; and respectively represent the active power and reactive power injected into nodes ; represents the complex conjugate of the voltage of node ; represents the complex conjugate of the admittance between nodes and ; is the number of nodes connected to node ; is the imaginary unit, satisfying and is used to represent complex numbers; represents the summation operator.

[0030] Based on the typical characteristics of the distribution network, a simplified DC power flow model can be used to establish the physical association: ; where represents the active power flowing from node to node ; respectively represent the voltage magnitudes of nodes and node ; represents the base voltage value of the system; is the reactance value between and ; respectively represent the voltage phase angles of node and node ; represents the sine function; represents approximately equal to.

[0031] Optionally, for a low - voltage distribution network or a case where the line impedance ratio is relatively large, an improved Distflow equation can be adopted: ; ; where and are the active power, reactive power, and voltage amplitude of node respectively; are the active power, reactive power, and voltage amplitude of node respectively; are the resistance and reactance of branch respectively, representing the line parameters connecting node and node ; and are the active and reactive loads of node ; represents the square of the voltage amplitudes of node and node .

[0032] For example, in a 10 kV urban distribution network application scenario, based on 15 actual measurement points, the system needs to estimate the states of 28 nodes in measurement blind areas. For the main line area, a DC power flow model is used to establish physical associations; while for the end - low - voltage branches, the Distflow equation is used to construct a more accurate model. Through this hybrid physical model method, the system provides a reliable physical basis for generating virtual measurement data under normal operating conditions.

[0033] Therefore, through these physical constraint equations, the electrical states of blind - area nodes can be preliminarily estimated, providing a physical basis for subsequent generation of virtual measurement data.

[0034] Step 2.2, Generation of virtual measurement data based on a generative adversarial network; The method provided in this application uses an improved Generative Adversarial Network (GAN) algorithm to generate virtual measurement data in blind areas that conforms to physical laws. The GAN algorithm includes a generator G and a discriminator D, and the optimization objective is: ; where represents the probability distribution of actual measurement values; represents the random noise distribution input to the generator; is the random noise input, representing the random vector input to the generator; represents the input noise is the output after being mapped by the generator; is the discrimination probability of the discriminator for the authenticity of the data; represents the expectation operator; represents obeys the distribution ; represents obeys the distribution represents the natural logarithm function; represents the minimization operation with respect to the generator G; represents the maximization operation with respect to the discriminator D; represents the value function of the generative adversarial network.

[0035] It should be noted that to ensure that the generated virtual measurement data conforms to the physical laws of the power grid, a physical constraint regularization term is introduced into the loss function of the generator G: ; where represents the objective function that needs to be minimized during the training process of the generator; represents the ability of the generator to deceive the discriminator; is the physical constraint loss, used to penalize the generated data that violates the physical laws of the power grid; is the weight coefficient for balancing the GAN loss and the physical constraint, controlling the importance of the physical constraint in the total loss.

[0036] In the implementation of the improved GAN model in this application, the generator G consists of a multi-layer neural network. The input layer receives random noise and the electrical quantity information of surrounding measured points. It includes three fully connected hidden layers in the middle, with 128, 256, and 128 neurons respectively in each layer. The activation function uses LeakyReLU to enhance the model's learning ability for non-linear features. The output layer uses the tanh activation function to ensure that the generated electrical quantities such as voltage and current are within a reasonable range.

[0037] The discriminator D also adopts a multi-layer neural network structure. The input layer receives measurement data (real or generated), includes two hidden layers with 128 neurons in each layer, uses the LeakyReLU activation function, and the output layer is a single neuron that uses the sigmoid activation function to output the probability value of data authenticity.

[0038] Physical constraint loss is the innovative part of this application, and its specific calculation method is as follows: ; where is the physical constraint loss; is the Kirchhoff voltage law constraint term, which ensures that the sum of voltages in a closed loop is zero and represents the degree to which the generated data violates the KVL law; is the Kirchhoff current law constraint term, which ensures that the algebraic sum of node currents is zero and represents the degree to which the generated data violates the KCL law; is the power flow equation constraint term, which ensures that the generated data satisfies the power flow equation and represents the degree to which the generated data violates the power flow equation; are respectively and the weight coefficients of.

[0039] In the virtual measurement scenario of the 10kV feeder blind area in the B area of the distribution network, when applying the improved GAN of this application, first use the historical data of the first six months of normal operation period to train the model, and then use the trained model to generate virtual measurement data for 4 distribution network nodes lacking measurement points. Compared with the traditional interpolation method, the virtual data generated by the model of this application is more in line with the physical characteristics of the power grid. Especially under the conditions of load mutation and small disturbance, the average relative error between the virtual measurement data and the real value is reduced, providing a more reliable data basis for subsequent fault location.

[0040] Step 2.3, fusion of virtual and real measurement data; Through the dynamic weight method, fuse the actual measurement data and the virtual measurement data to form a hybrid measurement matrix covering the whole network: ; where is the hybrid measurement matrix covering the whole network, with the dimension of the number of nodes × the number of time points. Each row represents a node, and each column represents a time point; is the actual measurement data matrix, with the dimension of the number of actual measurement points × the number of time points, and contains the measurement data of all actual measurement points; is the virtual measurement data matrix, with the dimension of the number of virtual measurement points × the number of time points, and contains the generated data of all virtual measurement points.

[0041] In addition, for each virtual measurement point Allocate weights according to their generation reliability

[0042] ; where is the weight of the virtual measurement point , representing the credibility of the data of the virtual measurement point , and its value range is usually [0, 1]; is the distance to the nearest measured point, representing the electrical distance or topological distance, and is used to evaluate the proximity relationship between the virtual measurement point and the actual measurement point; is the local network topology complexity, reflecting the complexity of the local network structure, and is usually calculated through the number of node connections or network structure characteristics; is the consistency index between the historical generated data and the measured data, and its value range is [0, 1]; is the weight calculation function, which determines the weight of the virtual measurement point according to multiple factors.

[0043] It can be seen that the weight will be used in the subsequent fault location algorithm to guide the degree of trust in data from different sources during the decision-making process.

[0044] The active fault excitation implementation module injects a safe and controllable small perturbation into the high-risk weak links, actively detects the response characteristics of the weak links, and obtains the equipment health status information; Implement safe and controllable active fault excitation for the identified key weak links, and detect potential fault characteristics in advance, including the following sub-steps: Step 3.1, perturbation signal construction; In the implementation scheme of this application, a perturbation signal for active fault excitation is constructed. The perturbation signal needs to meet two requirements: one is to be able to effectively excite potential fault characteristics, and the other is not to affect the normal operation of the system. This application adopts the form of a decaying sine wave: ; where is the perturbation signal, representing the perturbation signal injected into the system at time ; is the amplitude of the perturbation signal, representing the maximum amplitude of the perturbation signal; is the frequency of the perturbation signal, and a frequency band that can effectively excite the response characteristics of the equipment is selected; is the phase of the perturbation signal, controlling the initial phase of the signal; is the attenuation coefficient of the perturbation signal, controlling the attenuation rate and duration of the signal amplitude; is the time variable; represents the sine function; is twice the pi, used to convert the frequency to the angular frequency; is an exponential decay term, where is the base of the natural logarithm.

[0045] In addition, for different types of devices and potential failure modes, this application constructs a perturbation signal library: ; where respectively represent the 1st, 2nd, th perturbation signals, is the number of signals in the perturbation signal library, representing the total number of different types of preset perturbation signals; Each signal has the best excitation effect for a specific type of potential failure.

[0046] The specific implementation of the perturbation signal construction algorithm of this application is as follows: First, through historical fault data analysis, identify the precursor feature spectral characteristics of different types of faults, and determine the sensitive frequency band of each fault type Subsequently, construct perturbation signals with different parameters for each fault type, where the amplitude is determined by the binary search method and is as large as possible without causing system anomalies, usually set to 0.5% to 3% of the rated value; the frequency selects the sensitive frequency corresponding to the fault type; the phase is dynamically adjusted according to the current operating state of the system to avoid resonance with the natural oscillation frequency of the system; the attenuation coefficient is set according to the signal duration requirement.

[0047] In the application scenario of potential fault detection of 35kV distribution outgoing line equipment in a 220kV substation, this application constructs perturbation signals for three potential faults: insulation aging, abnormal contact resistance, and partial discharge. For example, for the insulation aging fault, select (0.8% of the rated voltage), construct a perturbation signal, and after injection, it successfully excites a small change in the tangent value of the insulation dielectric loss angle, and an insulation deterioration point is detected in advance, avoiding a possible equipment breakdown fault.

[0048] Step 3.2, safe injection control; It should be noted that through the power electronic interface device, the perturbation signal is safely injected into the key weak links of the distribution network. The injection process adopts a closed-loop control method: ; where represents the signal actually injected into the system at time represents The ideal perturbation signal expected to be injected at a certain moment; and respectively represent the proportional control coefficient and the integral control coefficient; represents the integral operator, which performs time integration on the error signal; represents the time element, the integration variable; is the error signal, that is the ideal perturbation signal expected to be injected at a certain moment and the signal actually injected into the system at a certain moment the difference.

[0049] Meanwhile, this application monitors the key parameters of the system in real time and sets multiple levels of safety thresholds: Warning threshold: Adjust the injection amplitude; Emergency threshold: Immediately stop the injection.

[0050] Among them is the key parameter of the system; is the warning threshold, usually set at about ±3% of the normal value; is the emergency threshold, which is triggered when the system parameter reaches a dangerous level and is usually set at about ±5% of the normal value; represents the absolute value or deviation amplitude of the parameter, which is used for comparison with the threshold.

[0051] The specific implementation of the safe injection control algorithm includes three main parts: Firstly, there is the perturbation signal injection device, which adopts a hybrid control architecture based on DSP and FPGA. DSP is responsible for the upper control strategy and safety monitoring, and FPGA realizes high-speed waveform generation and real-time control. The sampling frequency reaches 20 kHz to ensure the precise control of the perturbation signal; Secondly, there is the adaptive adjustment of the closed-loop control parameters. The controller parameters and are automatically optimized according to the system impedance characteristics, the range is usually 0.05 - 0.2, the range is 0.01 - 0.05, and they are dynamically adjusted according to the tracking error during the injection process. Thirdly, there is the multi-layer safety protection mechanism, which monitors 10 key parameters such as voltage, current, frequency, and harmonic content. The warning threshold is set at ±3% of the normal value, and the emergency threshold is set at ±5%. If any parameter exceeds the threshold, the corresponding protection action will be triggered.

[0052] Step 3.3, response feature extraction; Record the response of the recording device to the perturbation signal, extract the characteristic parameters, and construct the health status fingerprint. First, obtain the system response: ; where is the transfer function of the object under test, describing the relationship between the system input and output, is the complex frequency domain variable, which is directly related to its health status; represents the response output of the system to the perturbation signal at time ; represents the perturbation signal injected into the system at time .

[0053] Subsequently, through the small-signal analysis method, calculate the characteristic parameters of the transfer function: Amplitude-frequency response characteristic represents the gain of the system at different frequencies; Phase-frequency response characteristic represents the phase change of the system at different frequencies; Damping coefficient represents the speed of oscillation decay of the system; Natural frequency represents the inherent oscillation frequency of the system.

[0054] where is the imaginary unit, is the angular frequency, and represents the frequency; represents the magnitude of the transfer function at frequency ; represents the phase angle of the transfer function at frequency .

[0055] It can be seen that these parameters constitute the health status fingerprint of the device for subsequent health status assessment: ; where is the health status fingerprint, including the set of characteristic parameters describing the health status of the device; represents the magnitude of the transfer function at frequency ; represents the phase angle of the transfer function at frequency ; is the imaginary unit; represents the damping coefficient; represents the natural frequency.

[0056] The state estimation enhanced fault location module constructs a health status assessment model based on the device health status information and calculates the deviation between the transfer function and the health baseline; Optimize the configuration of measurement points in combination with virtual measurement data, and use state estimation technology to complete missing information to achieve early identification and accurate positioning of faults, including the following sub-steps: Step 4.1, construction of a hybrid measurement model; The method provided in this application is based on the virtual-real fusion measurement network constructed in the virtual-real fusion measurement network construction module to establish a hybrid measurement model: ; where is the hybrid measurement vector, which includes actual measurement values and virtual measurement values, and the dimension is the total number of measurement points; are the actual measurement value vector and the virtual measurement value vector respectively; and represent the measurement functions of the actual measurement point and the virtual measurement point respectively, mapping the system state to the measurement value; represent the actual measurement error and the virtual measurement error respectively, reflecting the uncertainty of the measurement value; is the system state variable, representing state quantities such as the voltage amplitude and phase angle of the power grid.

[0057] Considering that the reliability of virtual measurement data is lower than that of actual measurement data, this application introduces a weight matrix

[0058] ; where is the weight matrix, which assigns different weights to the measurement values of different measurement points and is a diagonal matrix; and are the weight sub-matrices of actual measurement and virtual measurement respectively; The elements in are set by the weight calculated in step 2.3; represents the zero matrix, indicating that there is no cross weight between actual measurement and virtual measurement.

[0059] Step 4.2, health state assessment; It should be noted that based on the active excitation response characteristics obtained in the active fault excitation implementation module, a device health state assessment model is constructed. First, calculate the deviation between the current transfer function and the health benchmark: ; where is the current transfer function, representing the frequency-domain response characteristics of the device under the current state; is the health state benchmark transfer function, representing the frequency-domain response characteristics of the device in the healthy state; is the L2 norm, the Euclidean norm, used to calculate the vector length or matrix size; The transfer function deviation represents the distance between the current transfer function and the healthy baseline transfer function.

[0060] Optionally, the calculation of the transfer function deviation can select different norm forms according to the specific application scenario. For example, for devices sensitive to changes in specific frequency points in the frequency domain (such as transformers), the L1 norm can be used: ; where is the set of frequency points of interest, which contains frequency points that are particularly important for evaluating the health status of the device; represents the summation operator, which sums all elements in the set; represents the absolute value operator, which calculates the modulus of a complex number or the absolute value of a real number; is the current transfer function, representing the frequency domain response characteristics of the device in the current state; is the healthy state baseline transfer function, representing the frequency domain response characteristics of the device in the healthy state; is the transfer function deviation after adopting the L1 norm, represents the angular frequency.

[0061] For devices that need to comprehensively consider the characteristics of multiple frequency bands, the weighted Euclidean norm can be used: ; where is the weight coefficient of the th frequency point, reflecting the importance of the th frequency point for health status evaluation, and different weights can be set according to the frequency points sensitive to device failures; represents the square root operator, which calculates the square root of the expression; represents the square of the modulus of a complex number or the square of a real number; is the transfer function deviation after adopting the weighted Euclidean norm; is the current transfer function, representing the frequency domain response characteristics of the device in the current state; is the healthy state baseline transfer function, representing the frequency domain response characteristics of the device in the healthy state.

[0062] Subsequently, the health status score is evaluated through multi-feature fusion: ; where is the health status score, representing a quantitative score of the overall health status of the device, usually in the range of [0, 1], and 1 represents complete health; and are the deviations of the damping coefficient and the natural frequency respectively, representing the difference between the current damping coefficient and the healthy state damping coefficient and the difference between the current natural frequency and the healthy state natural frequency; is a health status scoring function, a function that maps the transfer function deviation to a health status score; is the transfer function deviation.

[0063] In some embodiments, the health status scoring function may adopt a fuzzy logic inference system to map each feature deviation to the interval [0, 1], and calculate the final health status score through a set of fuzzy rules. For example, the following fuzzy rules can be established: If is "large" and is "positive large", then is "low"; If is "small" and is "close to zero", then is "high"; Optionally, for complex equipment, the health status assessment may adopt a multi-level scoring strategy. First, calculate the health status scores of each key component of the equipment (such as the iron core, winding, insulation system, etc. of the transformer) respectively , and then obtain the overall health status score by means of weighted average or taking the minimum value: ; Or ; Where is the number of key components, indicating the total number of key components contained in the equipment; is the key component index of the equipment, indicating the number of a specific component; is the weight of each component, indicating the contribution degree of different components to the overall health status; represents the minimum value operator, which returns the minimum value in the parameter list; represents the summation operation from ; respectively represent the health status scores calculated by the minimum value strategy and the weighted average strategy; respectively represent the health status scores of the 1st, 2nd, and the th key components.

[0064] For example, in the application of the health status assessment of 35 kV distribution equipment in a 110 kV substation, after the system conducts an active excitation test on the circuit breaker equipment, it evaluates the status of each component through the above multi-level scoring strategy. The system detects an obvious deviation in the transfer function of the operating mechanism in the frequency band of 55 to 65 Hz. Combining the characteristic of the increase in the damping coefficient, the health status score decreases, and the system generates a warning message. Subsequent maintenance confirms that there is a problem of poor lubrication inside the operating mechanism, and timely maintenance avoids possible refusal-to-operate failures.

[0065] In addition, when , it is determined that the device has a potential failure risk and enters the warning state, where is the warning threshold for the health state. When it is lower than this threshold, the health state warning of the device is triggered.

[0066] Step 4.3, identification of precursors to failure; Using time-domain and frequency-domain feature analysis methods, identify the precursor features of failure from the response signal. For time-domain analysis, calculate the statistical features of the response waveform: ; where is the set of time-domain statistical features, including the statistical features extracted from the time-domain waveform; is the maximum value of the response waveform, the peak value of the time-domain waveform; is the minimum value of the response waveform, the trough value of the time-domain waveform; is the standard deviation, a statistic representing the degree of data dispersion; is the skewness, a statistic representing the degree of asymmetry of the data distribution; is the kurtosis, a statistic representing the degree of peakedness of the data distribution; is the measured response signal.

[0067] For frequency-domain analysis, extract multi-scale features through wavelet transform: ; where is the set of frequency-domain features, including the features extracted from frequency-domain analysis; is the energy of the th frequency band, representing the energy density of the signal within a specific frequency band; is the frequency band index, representing the number of a specific frequency band; is the ratio of the energy of adjacent frequency bands, the ratio of the energy of the current frequency band to the energy of the adjacent frequency band; respectively represent the energy of the 2nd and the th frequency bands; respectively represent the energy ratios of the 2nd and the th frequency bands; is the total number of frequency bands.

[0068] Therefore, combining time-domain and frequency-domain features, the present application constructs a precursor feature vector of failure containing a set of features that may indicate the occurrence of failure, for failure type identification and location.

[0069] The resilient fault location execution module, based on the hybrid measurement matrix, extracts the key information dimensions in the case of uncertain measurement data or failure of some measurement points to ensure the reliability of fault location; It includes the following sub-steps: Step 5.1, key information dimension extraction; The method provided by this application applies the key information dimension extraction algorithm to extract the most valuable fault features from the hybrid measurement data. First, perform principal component analysis on the original high-dimensional features: ; where is the original feature matrix, containing the original features of all samples, with the dimension of the number of samples × the number of features; is the principal component matrix, containing the eigenvectors of the principal components, with the dimension of the number of features × the number of features; is the feature matrix after dimensionality reduction, the low-dimensional representation after principal component analysis, with the dimension of the number of samples × the selected number of principal components; represents the transpose of the matrix .

[0070] Subsequently, through the Fisher discriminant criterion, evaluate the contribution of each principal component to the discrimination of fault types: ; where are the between-class scatter and within-class scatter of the th principal component respectively; is the principal component index, representing the number of a specific principal component; is the Fisher discriminant criterion value, representing the contribution degree of the th principal component to the discrimination of fault types.

[0071] According to the value, this application selects the most discriminative principal components to form the key information dimension set, where is the number of selected principal components, the number of retained principal components.

[0072] Step 5.2, data credibility evaluation; It should be noted that, combined with the credibility weights of multi-source data, evaluate the reliability of each measurement data. For actual measurement data, calculate the credibility based on the device state and measurement residuals: ; where is the credibility of the actual measurement data, representing the reliability of the measurement data of node , with the value range [0, 1]; is the state index of the measurement device, reflecting the node Measure the working state of the measuring device; is the measurement residual, the difference between the measured value and the estimated value; is the credibility calculation function, which determines the credibility of the measurement data according to multiple factors; is the node index, representing the number of a specific node.

[0073] In addition, for virtual measurement data, directly use the weight calculated in step 2.3 as the credibility.

[0074] Step 5.3, Fault type identification and location; Based on the key information dimensions and data credibility extracted in the foregoing steps, this application constructs a resilience fault location algorithm and adopts a weighted voting fusion method: ; wherein is the finally identified fault type, representing the fault type finally determined by the algorithm; is the th credibility weight of the evidence; represents the conditional probability that the fault type is under the condition of the given evidence ; represents the total number of evidences used for fault identification; represents the summation operation from ; represents finding the fault type that makes the expression take the maximum value .

[0075] It can be seen that the fault location also adopts a weighted method to construct a probability distribution map of the fault location: ; wherein is the probability that the fault location is at , representing the probability that the fault location is ; is the location probability estimate based on the evidence , representing the conditional probability that the fault location is under the condition of the given evidence ; is the fault location variable; is a specific location point; is the credibility weight of the th evidence;

[0076] Therefore, the system finally outputs the fault type, fault location and the corresponding credibility to support the decision-making of the operation and maintenance personnel.

[0077] The resilience fault location algorithm of this application adopts a multi-level and multi-modal fault inference structure in implementation. The first layer is the evidence collection layer, which simultaneously processes five types of evidence sources: actual measurement data, virtual measurement data, actively excited response characteristics, historical fault patterns, and topological constraint information. The second layer is the evidence preprocessing layer, which applies specific preprocessing methods to each type of evidence. For example, noise filtering, anomaly detection, and normalization are performed on the measurement data; wavelet transform and spectral analysis are performed on the actively excited characteristics; clustering analysis is performed on the historical fault patterns, etc. The third layer is the local inference layer. For each type of preprocessed evidence, dedicated classifiers (such as support vector machines, random forests, or neural networks) are trained respectively to generate preliminary fault type and location estimates. The fourth layer is the evidence weighted fusion layer, which combines the credibility of each evidence source and uses a Bayesian network framework to fuse the local inference results to form the final fault judgment result.

[0078] The algorithm of this application also adopts a progressive inference strategy, which can adjust the inference accuracy according to the quantity and quality of the available evidence. In the case of insufficient or low-quality evidence, the system first gives a rough range estimate (such as a fault section or area); as more evidence is collected or processed, the system gradually pinpoints the location of the fault point. In addition, this method also introduces an adaptive confidence threshold: the system provides a confidence score for each inference result. When the confidence is lower than the threshold, the system will suggest that the operation and maintenance personnel obtain additional evidence or start an alternative location method to avoid wasting resources caused by misjudgment.

[0079] A storage medium includes a memory and one or more processors. Executable code is stored in the memory. When the one or more processors execute the executable code, it is used to implement the above-mentioned distribution network fault identification and location system based on wide-area measurement technology.

[0080] Here, the present invention provides an implementation example: This implementation mode has been actually applied in the distribution network in the urban-rural fringe of a certain city. This distribution network area includes 2 110 kV substations, 5 35 kV substations, and 23 10 kV distribution substations, with a coverage area of approximately 76 square kilometers and a power supply population of approximately 320,000. The distribution network in this area has the following characteristics: First, the degree of equipment aging varies. Some lines and equipment have been in operation for 15 to 20 years, presenting potential fault risks; second, the measurement coverage is uneven. There are relatively sufficient measurement points in the core area, while the measurement points in the edge area are sparse; third, the fault frequency is relatively high, with an average annual number of faults reaching 26 times, and the faults caused by equipment aging account for more than 40%.

[0081] Before implementing this method, the average fault location time of the regional distribution network was 3.7 hours, and the longest location time reached 7.5 hours. The fault warning ability was weak, and 95% of the faults were only discovered after they occurred, resulting in frequent large-scale power outages. In addition, the coverage rate of existing measurement equipment was only 33%, and most areas lacked real-time monitoring capabilities, increasing the difficulty of fault location.

[0082] Based on the above situation, the goal of this implementation method is to improve the fault warning ability and fault location accuracy of the distribution network. At the same time, under the condition of limited budget, the monitoring coverage is expanded through virtual measurement technology.

[0083] First, the multi-dimensional weak link identification algorithm is applied to analyze the structural characteristics and historical operation data of the distribution network. According to the calculation results, the system identifies 15 high-risk weak links, mainly concentrated in the distribution lines with serious equipment aging and important load-intensive areas.

[0084] According to the measurement point sensitivity assessment and budget constraints, 23 new measurement points are optimized and arranged on the basis of the original 47 measurement points in the whole network. Among them, 12 are arranged at the outgoing ends of the main substations, 8 are arranged at the key nodes of important lines, and 3 are arranged at the access points of high-value users. The optimized configuration of these measurement points improves the monitoring coverage rate of high-risk areas from 51% to 91%.

[0085] For the still existing measurement blind areas, especially 6 10kV distribution substations and 12 branch lines in the edge areas, a physical correlation model is constructed. The DC power flow model is adopted for the main lines, and the improved Distflow equation is adopted for the end branch lines.

[0086] On this basis, an improved generative adversarial network is trained. The network uses the historical operation data of half a year as the training set. The generator contains 3 hidden layers, and the discriminator contains 2 hidden layers. The weight coefficient of the physical constraint regularization term is set to 0.45 to ensure that the generated virtual data conforms to the physical characteristics of the power grid.

[0087] Through this network, virtual measurement data is successfully generated for 43 measurement blind area nodes, increasing the network measurement coverage rate from the original 33% to 92%. Among them, the average error rate of the virtual measurement data is controlled within 1.3% under normal operating conditions.

[0088] For the identified 15 high-risk weak links, especially 5 old lines and 3 distribution transformers with long-term high load operation, safe and controllable active fault excitation is implemented.

[0089] For example, for the switchgear at pole No. 3 of the 10kV Fenghua Line, a perturbation signal with an amplitude of 0.8% of the rated voltage, a frequency of 120Hz, and a duration of 0.5 seconds was injected through a power electronic injection device installed on the line. During the implementation process, the system monitored 10 key parameters such as voltage, current, and harmonic content in real time to ensure the safety of the injection process.

[0090] Through the analysis of the response signal, the system detected that the phase-frequency response characteristics of the switchgear showed abnormal deviation in the range of 55 to 70Hz. At the same time, the damping coefficient increased by 35%, and the health status score dropped to 0.68 (lower than the warning threshold of 0.75). Subsequent maintenance found that there was a problem of poor contact in the switch, and early maintenance avoided possible sectional faults.

[0091] In this example, the system constructed a hybrid measurement model, integrating data from 70 actual measurement points and 43 virtual measurement points. The weights of the virtual measurement data were dynamically adjusted according to their generation reliability, and the weight range was 0.4 - 0.8.

[0092] Based on the device health status fingerprint obtained by active excitation, the system evaluated the health status of all network devices. During the implementation period, the system identified a total of 7 potential fault points, including 3 cases of transformer insulation aging, 2 cases of overheating of cable joints, and 2 cases of poor contact of switches. These potential fault points were marked and entered the warning state, and the maintenance personnel arranged maintenance plans accordingly.

[0093] To verify the resilience of the system, 3 cases of partial failure of measurement points were simulated, randomly making 30% of the measurement points (21) fail. Under this condition, the system still maintained a high fault location accuracy rate through key information dimension extraction and data credibility evaluation.

[0094] In a real fault case, a short-circuit fault occurred in a 10kV branch line due to a fallen tree, and at the same time, communication failures caused data loss at 3 nearby measurement points. The system used a multi-level fault inference structure, combined with the remaining measurement data, virtual measurement data, historical fault patterns, and topological constraint information, and successfully located the fault point within 800 meters of the actual location, reducing the location range by about 78% compared with the traditional method, greatly reducing the workload of the line patrol personnel.

[0095] One year after the implementation of this method in this distribution network, through the statistical analysis of the operation data, the following two key technical effects were verified: Through the active fault excitation and health status evaluation technologies, the system successfully warned of 23 potential faults, accounting for 82% of the total number of faults in that year. The average lead time of the system warning was 18 days, and the longest lead time reached 32 days. The warning accuracy rate was 91.3% (21 out of 23 warnings were subsequently verified).

[0096] Potential fault warning time distribution: There are 4 cases with a warning time of 5 to 10 days, 12 cases with a warning time of 11 to 20 days, 5 cases with a warning time of 21 to 30 days, and 2 cases with a warning time exceeding 30 days. There are differences in the warning effects of different types of faults: The warning of equipment aging faults is advanced by an average of 25 days, the warning of poor contact faults is advanced by an average of 16 days, and the warning of insulation deterioration faults is advanced by an average of 22 days.

[0097] By discovering and handling these potential faults in advance, 18 possible power outages were avoided, the power outage losses were reduced by approximately 2.96 million yuan, and the number of beneficiary users reached 85,000. Compared with the situation before implementation when 95% of the faults were only discovered after they occurred, the early discovery rate of potential faults achieved by this method is 82%, reflecting technological progress.

[0098] Among the 28 faults that occurred throughout the year (including 5 sudden faults that could not be warned and 23 faults that were warned and observed), the system located all the faults. The average location accuracy rate reached 89.3%, and the average location time was shortened to 0.6 hours, a reduction of 83.8% compared with 3.7 hours before implementation.

[0099] Through the virtual-real fusion measurement technology, the system increased the measurement coverage rate from 33% to 92%. Even in areas where no actual measurement points were arranged, the virtual measurement data provided effective support for fault location. For faults occurring near 6 10kV distribution transformer substations in the edge area, the average location error was controlled within 1.2 kilometers, an improvement of 68.4% compared with 3.8 kilometers of the traditional method.

[0100] The resilience of the system was also verified. In the test of simulating the failure of 30% of the measurement points, the fault location accuracy rate of the system remained at 86.7%, while the accuracy rate of the traditional method under the same conditions was only 41.2%. In addition, in 4 real cases of partial faults in the communication system, the system still controlled the average location error within 1.5 kilometers.

[0101] The above technical effect verification data shows that this implementation method improves the fault warning ability and fault location accuracy of the distribution network. Especially under the conditions of insufficient measurement coverage and partial failure of measurement equipment, it can still maintain stable performance, fully reflecting the technical innovation and practical value of this application.

[0102] The embodiments of the present invention are described above, but these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are only illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. A distribution network fault identification and location system based on wide-area measurement technology, characterized in that It includes: A key measurement point optimization configuration module that determines the optimal layout position of measurement points through a sensitivity evaluation algorithm and identifies high-risk weak links that need to be key monitored in the distribution network; A virtual-real fusion measurement network construction module that, based on the optimal layout position of measurement points, uses a generative adversarial network to generate virtual measurement data for blind spots that conform to physical laws, and fuses actual measurement data and virtual measurement data through a dynamic weight method to form a hybrid measurement matrix covering the entire network; An active fault excitation implementation module that injects safe and controllable small perturbations for high-risk weak links, actively detects the response characteristics of weak links, and obtains equipment health status information; A state estimation enhanced fault location module that constructs a health status evaluation model based on equipment health status information and calculates the deviation between the transfer function and the health baseline; A resilient fault location execution module that, based on the hybrid measurement matrix, extracts key information dimensions to ensure the reliability of fault location in the case of uncertain measurement data or partial measurement point failures.

2. The distribution network fault identification and location system based on wide-area measurement technology according to claim 1, characterized in that The determination of the optimal layout position of measurement points through the sensitivity evaluation algorithm includes: Applying a multi-dimensional weak link identification algorithm to evaluate the risk degree of each node in the distribution network, calculating the risk score of each node, and comprehensively considering topological vulnerability, historical fault frequency, electrical stress, and node importance in the risk score; Analyzing the contribution of each candidate measurement point to network observability, constructing an observation equation and calculating the observation sensitivity matrix of each candidate measurement point to obtain the measurement point sensitivity index; Based on the risk score and sensitivity index, constructing an optimization model for measurement point configuration, and solving the optimization model to determine the final measurement point configuration plan.

3. The distribution network fault identification and location system based on wide-area measurement technology according to claim 1, characterized in that The use of a generative adversarial network to generate virtual measurement data for blind spots that conform to physical laws includes: Using the electrical topology relationship and power flow constraints to construct a physical association model between measurement points and blind spot nodes; Training the generator and discriminator of the generative adversarial network. The generator receives random noise and electrical quantity information of surrounding measured points as inputs to generate virtual measurement data, and the discriminator discriminates the authenticity of the data; Introducing physical constraint regularization terms into the loss function of the generator, including Kirchhoff's voltage law constraint terms, Kirchhoff's current law constraint terms, and power flow equation constraint terms.

4. The distribution network fault identification and location system based on wide-area measurement technology according to claim 1, characterized in that The implementation of safe and controllable small perturbation injection for high-risk weak links includes: Constructing a perturbation signal for active fault excitation. The perturbation signal adopts the form of a decaying sine wave, and the amplitude is set within a safe range that does not affect the normal operation of the system; Injecting the perturbation signal safely into the key weak links of the distribution network through a power electronic interface device, and using a closed-loop control method to control the injection process; Real-time monitoring of key system parameters, setting warning thresholds and emergency thresholds, and triggering corresponding protection actions when the monitored parameters exceed the thresholds.

5. The distribution network fault identification and location system based on wide-area measurement technology according to claim 1, characterized in that The construction of a health status evaluation model based on equipment health status information includes: Recording the response of the equipment to the perturbation signal and obtaining the system response transfer function; Calculating the deviation between the current transfer function and the health benchmark transfer function; Evaluating the health status score through multi-feature fusion. When the health status score is lower than the preset threshold, it is judged that the equipment has potential fault risks and enters the warning state. Using time-domain and frequency-domain feature analysis methods, identify the fault precursor features from the response signals and construct a fault precursor feature vector.

6. The distribution network fault identification and location system based on wide-area measurement technology according to claim 1, characterized in that The steps of extracting the key information dimension in the case of uncertainty in measurement data or failure of some measurement points include: Extract the most valuable fault features from the mixed measurement data, perform principal component analysis on the original high-dimensional features, and obtain the feature matrix after dimensionality reduction; Evaluate the contribution of each principal component to the discrimination of fault types through the Fisher discriminant criterion; According to the contribution values, select several principal components with the most discriminative power to form a set of key information dimensions; Combined with the credibility weights of multi-source data, evaluate the reliability of each measurement data, calculate the credibility of the actual measurement data based on the device state and measurement residuals, and directly use its generated reliability as the credibility for virtual measurement data.

7. A fault identification and location system for a distribution network based on wide area measurement technology according to claim 1, characterized in that, The ensuring the reliability of fault location includes: Based on the extracted key information dimensions and data credibility, construct a resilient fault location algorithm; The resilient fault location algorithm uses a weighted voting fusion method to identify the fault type, multiplying the credibility weights of each piece of evidence by the probability estimate of the fault type based on the evidence and then summing them up; Construct a probability distribution map of the fault location, multiply the credibility weights of each piece of evidence by the location probability estimate based on the evidence and then sum them up, and then divide by the sum of the credibility weights; Output the fault type, fault location and the corresponding credibility to support the decision-making of the operation and maintenance personnel.

8. A fault identification and location system for a distribution network based on wide area measurement technology according to claim 7, characterized in that, The resilient fault location algorithm includes an evidence collection layer, an evidence preprocessing layer, a local inference layer and an evidence weighted fusion layer, specifically including: The evidence collection layer processes five types of evidence sources: actual measurement data, virtual measurement data, actively excited response characteristics, historical fault patterns and topological constraint information; The evidence preprocessing layer applies specific preprocessing methods to each type of evidence; The local inference layer independently analyzes each type of preprocessed evidence to generate preliminary fault type and location estimates; The evidence weighted fusion layer combines the credibility of each evidence source and uses a Bayesian network framework to fuse the local inference results to form a final fault judgment result.

9. A fault identification and location system for a distribution network based on wide area measurement technology according to claim 1, characterized in that, The micro-perturbation injection adopts a safety injection control system, and the safety injection control system includes: A hybrid control architecture based on DSP and FPGA, where DSP is responsible for upper-layer control strategies and safety monitoring, and FPGA realizes high-speed waveform generation and real-time control; A closed-loop control parameter adaptive adjustment unit that automatically optimizes the controller parameters according to the system impedance characteristics; A multi-layer safety protection unit that monitors key parameters, including voltage, current, frequency, and harmonic content, and triggers corresponding protection actions when the monitored key parameters exceed the thresholds.

10. A storage medium, characterized in that, Including a memory and one or more processors, where the memory stores executable code, and when the one or more processors execute the executable code, it is used to implement a distribution network fault identification and location system according to any one of claims 1-9.

Images