Weak-signal fault identification of inverter-based microgrid

Abstract

To provide a method and a system for identifying presence, location, and type of a weak-signal fault in an islanded inverter-based microgrid.SOLUTION: Upon receiving voltage and current measurements from an intelligent electronic device installed in a microgrid, a variation mode decomposition algorithm is applied to detect existence of fault based on denoised time series of measurements using discrete wavelet transform algorithm. After detecting the presence of fault, a correlation-based matrix is applied to locate a suspicious fault location, and then K-nearest neighbors model is utilized to identify a faulty branch among those locations using dynamic time warping algorithm to measure a distance between neighbors. Following fault localization, fault classification is done by observing sequence components and phasor measurements and feeding the observation inputs to a fault classification logic circuit model.SELECTED DRAWING: Figure 1A

Description

[Technical field]

[0001] The present invention relates generally to power systems, and more particularly to weak signal fault isolation for inverter-based microgrids. [Background technology]

[0002] Microgrids have attracted more attention in recent years with the development of a renewable energy-conscious society. A microgrid is a localized power grid that can be disconnected from the traditional power grid and operated autonomously, enhancing the reliability of the power grid and mitigating grid faults. The operation of a microgrid is very flexible in that it can be operated in either grid-tied or islanded mode. However, microgrid characteristics, such as the fault levels and control strategies of inverter-based distributed generators, can be significantly different in different operation modes. Therefore, traditional fault protection schemes may not be applicable to islanded microgrids.

[0003] Fault currents in microgrids vary with the type of distributed generators, operating conditions, and network topology. As a result, traditional overcurrent relays become unreliable. If the fault is a weak signal fault, the problem becomes even more complicated. Weak signal faults, such as high impedance faults, inverter DC side short circuit faults, and inverter tripping faults, occurring in the power distribution grid are often too small in magnitude to be detected and bypass traditional relays. Undetected weak signal faults pose security threats to human life and the power grid. Take high impedance faults as an example, live conductors in contact with the ground are dangerous to plants and animals and cause losses. Weak signal faults are common, and the development and need for distributed generation will only increase the occurrence of such low current faults. Therefore, fast and reliable fault identification, including fault detection, location, and classification, is necessary to restore grid power supply and ensure the safety of surrounding areas.

[0004] There are some related works on weak signal fault identification in microgrids. An example of this work can be found in the paper entitled "High Impedance Fault detection in Low Voltage Distribution Systems Using Wavelet and Harmonic Fault Indices" by Vineeth and P. Sreejaya, published in 2020 IEEE International Conference on Power Electronics, Smart Grid and Renewable Energy (PESGRE2020), 2020, pp.1-6. In this paper, the authors used multiresolution wavelet analysis to detect disturbances caused by faults in the neutral current. However, how to select an appropriate mother wavelet function is a key issue that must be resolved for practical application.

[0005] Another example is the detection and location of high impedance faults using power line communication (PLC) devices proposed by AN Milioudis, GT Andreou and DP Labridis in the paper entitled "Detection and Location of High Impedance Faults in Multiconductor Overhead Distribution Lines Using Power Line Communication Devices" published in IEEE Transactions on Smart Grid, vol. 6, no. 2, pp. 894-902, March 2015. The proposed method derives the fault location of the faulty line based on the response of impulse injection of all PLC devices installed along the power line. The main limitation of this method is the availability of PLC devices.

[0006] Yet another example is the use of machine learning such as convolutional autoencoders to identify fault locations in power grids, as proposed by L. Zheng, P. Xu and J. Bai in a paper titled "power grid fault location method based on pretraining of Convolutional Autoencoder" in Proc. IEEE Int. Conf. on Computer Science, Artificial Intelligence and Electronic Engineering (CSAIEE), 2021, pp. 324-327. In this method, a convolutional autoencoder is used to pre-train many samples, and then a classifier is used to fine-tune a small batch of balanced samples. However, such methods are highly dependent on the grid topology and equipment status, and also increase the computational complexity.

[0007] Another example is given in the paper titled "Fault Type Classification in Microgrids Including Photovoltaic DGs," in which the relative behavior of voltage magnitude and phase angle measurements was used to classify weak signal faults in power grids. The paper was written by A. Hooshyar, E F El-Saadany and M. Sanaye-Pasand and published in IEEE Transactions on Smart Grid, vol. 7, no. 5, pp. 2218-2229, Sept. 2016. However, such methods may lead to false positive cases, as voltage magnitude and angle may behave similarly to a fault due to changes in load, line capacitance, or switch operation.

[0008] Therefore, there is a need to develop a more accurate method for identifying weak signal faults in inverter-based microgrids. Summary of the Invention

[0009] The present disclosure provides a combined weak signal fault detection, location, and classification method for islanded inverter-based microgrids. The disclosed method identifies weak signal faults using time series measurements of voltage, current, and power collected from intelligent electronic devices (IEDs) installed in the microgrid. The weak signal faults can be inverter-related faults or high impedance faults. The inverter-related weak signal faults are inverter DC side short circuit faults or inverter tripping faults. The high impedance faults are ground faults with high fault impedance, and the type of ground fault is a three-line to ground fault, a two-line to ground fault, or a single line to ground fault.

[0010] Some embodiments of the present invention provide a two-step method for weak-signal fault detection, in which measurements are first denoised by reconstruction using discrete wavelet transform (DWT), and then the presence of weak-signal faults is detected by analyzing the frequency components generated by variational mode decomposition (VMD) using the denoised measurements. The performance of this two-step method is insensitive to noise because the weak-signal faults are analyzed using the denoised data reconstructed by DWT. This is verified by experimental tests on both noisy and noiseless data. Meanwhile, this two-step fault detection method is scalable to larger systems regardless of the grid configuration.

[0011] Some embodiments of the present invention provide a hybrid method for determining fault location after the presence of a fault is detected, in which the correlation between time series current components between two consecutive IEDs is first utilized to identify a suspected fault location, and then the final fault location is confirmed by identifying the branch containing the fault using a time series K-nearest neighbors (KNN) model with the neighborhood distance measured by dynamic time warping (DTW). Since the disclosed hybrid model focuses on reducing false positive cases, it can achieve near perfect accuracy for localizing weak signal faults after appropriately tuning the hyperparameters of correlation threshold and number of nearest neighbors.

[0012] Some other embodiments of the present invention provide a logic-based fault classification method that utilizes the relationship between current sequence components along with voltage and current phasor measurements with fault type. Testing shows that the disclosed method can classify high impedance faults, inverter DC side short circuits, and inverter tripping with 100% accuracy, with the exception of the uncertainty between two-line to ground faults and one-line to ground faults. This fault classification is also scalable to larger systems, regardless of the grid configuration.

[0013] The presently disclosed embodiments are further described with reference to the accompanying drawings, in which: The drawings shown are not necessarily to scale, with emphasis generally being placed upon illustrating the principles of the presently disclosed embodiments. [Brief description of the drawings]

[0014] [Figure 1A] FIG. 1 is a block diagram illustrating a method for detecting, locating, and classifying weak signal faults in an inverter-based microgrid according to an embodiment of the present disclosure. [Figure 1B] FIG. 1 is a schematic diagram illustrating components and steps for detecting, locating, and classifying weak signal faults in an inverter-based microgrid according to an embodiment of the present disclosure. [Figure 1C] FIG. 1 is a block diagram illustrating a fault isolation and control system for detecting, locating, and classifying weak signal faults in inverter-based microgrids according to some embodiments of the present invention. [Figure 2A] FIG. 1 is a schematic diagram illustrating an inverter-based microgrid according to an embodiment of the present disclosure. [Figure 2B] FIG. 1 is a schematic diagram illustrating a grid-forming inverter for use in a microgrid in accordance with an embodiment of the present disclosure. [Figure 2C] FIG. 1 is a schematic diagram illustrating a grid tracking inverter for use in a microgrid in accordance with an embodiment of the present disclosure. [Diagram 3] FIG. 2 is a schematic diagram illustrating current waveforms for a triple line to ground (TLG) fault with intrinsic mode function (IMF) mode 1, according to some embodiments of the present disclosure. [Figure 4] FIG. 2 is a schematic diagram illustrating current waveforms for a double line to ground (DLG) fault in IMF mode 1 according to some embodiments of the present disclosure. [Diagram 5] FIG. 13 is a schematic diagram illustrating current waveforms for an IMF Mode 2 DLG failure according to some embodiments of the present disclosure. [Figure 6] FIG. 1 is a schematic diagram showing current waveforms of an inverter DC-side short-circuit (DCSC) fault in IMF mode 2 according to some embodiments of the present disclosure. [Figure 7] FIG. 2 is a schematic diagram illustrating current waveforms for an inverter tripping (IT) fault in IMF mode 2 according to some embodiments of the present disclosure. [Figure 8] FIG. 1 is a schematic diagram illustrating current variations before and after a fault according to some embodiments of the present disclosure. [Figure 9] 1 is a table that illustrates a schematic correlation matrix for branch 1, according to some embodiments of the present disclosure. [Figure 10]11 is a table that illustrates a schematic correlation matrix for branch 2, according to some embodiments of the present disclosure. [Figure 11] 13 is a table that illustrates a schematic correlation matrix for branch 3, according to some embodiments of the present disclosure. [Figure 12] 1 is a table outlining the accuracy of a K-nearest neighbor (KNN) model, according to some embodiments of the present disclosure. [Figure 13] 1 is a table outlining the effectiveness of the disclosed method, according to some embodiments of the present disclosure. [Figure 14] FIG. 2 is a schematic diagram illustrating logic circuitry used for fault classification according to some embodiments of the present disclosure. [Figure 15] 15 is a schematic diagram illustrating a comparison of actual fault types and classification results using the disclosed logic circuit shown in FIG. 14 according to some embodiments of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] overview The present disclosure relates generally to power systems, and more particularly, to weak signal fault isolation for inverter-based microgrids.

[0016] The following description provides exemplary embodiments only and is not intended to limit the scope, applicability, or configuration of the present disclosure. Rather, the following description of exemplary embodiments provides those skilled in the art with an enabling description for implementing one or more exemplary embodiments. Various changes may be made in the function and arrangement of elements without departing from the spirit and scope of the disclosed subject matter as defined in the appended claims.

[0017] Specific details are provided in the following description to provide a thorough understanding of the embodiments. However, those skilled in the art will understand that the embodiments may be practiced without these specific details. For example, systems, processes, and other elements in the disclosed subject matter may be shown as components in block diagram form so as not to obscure the embodiments in unnecessary detail. In other instances, well-known processes, structures, and techniques may be shown without unnecessary detail so as not to obscure the embodiments. Additionally, like reference numbers and names in the various drawings indicate like elements.

[0018] The present disclosure is directed to fault identification for islanded inverter-based microgrids that include both inverter-based generators with grid-shaping and grid-following control strategies and conventional synchronous generators, with intelligent electronic devices (IEDs) installed on switchable devices to provide high fidelity waveform measurements to the microgrid operator. Weak signal faults include high impedance ground faults, inverter DC side short circuit faults, and inverter tripping faults. Weak signal fault identification is solved in three stages. In the first stage, a method combining discrete wavelet transform (DWT) and variation mode decomposition (VMD) is used to detect the presence of weak signal faults. In the second stage, the fault location of the detected weak signal fault is determined by incorporating a correlation matrix with a K-nearest neighbor (KNN) model. In the third stage, a logic circuit model defined utilizing the relative behavior of sequence components and phasor measurements is used to determine the fault type.

[0019] FIG. 1A is a block diagram illustrating a method (i.e., a computer-implemented method) 100B for detecting, locating, and classifying weak signal faults in an islanded inverter-based microgrid using an interface 153, a switch control device 157, a hardware processor 155, and a memory storing instructions that cause the hardware processor to perform the steps of the method 100B, in accordance with an embodiment of the disclosure.

[0020] The method 100B includes receiving 125 current and voltage time series data from intelligent electronic devices (IEDs) installed in the microgrid via a communication network using an interface 153.

[0021] The method 100B includes a step 130 of denoising the time series of current and voltage received for each IED using a discrete wavelet transform using a hardware processor 155.

[0022] Still referring to step 132 of FIG. 1A, the hardware processor 155 uses the IMF components generated by VMD from the denoised time series to detect the presence of a weak signal fault and determine the fault location if an inverter related weak signal fault is detected. By step 132, the weak signal fault is further classified as an inverter related fault or a high impedance fault if detected. For the identified inverter related fault, the fault location is determined according to the current and voltage of the AC side of the corresponding inverter, and then proceeds to step 138 to further classify the fault type. For a high impedance fault, proceed to the next step to determine the fault location.

[0023] In step 134, a hardware processor 155 is used to determine a suspected location within the microgrid of a weak signal fault using the current correlation matrix if the detected fault is a high impedance fault.

[0024] In step 136, a hardware processor 155 is used to determine the fault location by applying a time series KNN model using the time series power profile; if not directly measured, the power profile can be derived using time series current and voltage data.

[0025] In step 138, a hardware processor 155 is used to identify the fault type using the logic circuit model and the components and phasors of the current and voltage sequences.

[0026] Still referring to step 140 of FIG. 1A, method 100B includes isolating the faulted line section by activating a switching operation of a switch connected using a computing device 157 via a communication network.

[0027] FIG. 1B is a schematic diagram illustrating components and steps for detecting, locating, and classifying weak signal faults in an islanded inverter-based microgrid according to an embodiment of the present disclosure.

[0028] FIG. 1B may include a hardware processor 155 in communication with the input interface 135, the memory 137, the interface 153, and a computing device 157. The computing device 157 may be connected to a set of switches 110 installed in the microgrid 115. The microgrid may include a set of intelligent electronic devices (IEDs) 145 for monitoring and collecting status information of the microgrid 115. The status information is a time series measurement of current and voltage measured by the IEDs 145. Each IED corresponds to a switchable device 110. The fault localization and control system realized in the method 100B may not only control the set of switches 110, but may also send and receive information. It is contemplated that the hardware processor 155 may include two or more hardware processors depending on the requirements of a particular application. Of course, other components including input interfaces, output interfaces, and transceivers may be incorporated into the method 100B.

[0029] Still referring to FIG. 1B, an embodiment of the system 100B includes receiving 125 current and voltage time series data from the IEDs of the microgrid over a communication network using an interface 153.

[0030] The method 100B includes using a hardware processor 155 to denoise 130 the received time series currents and voltages for each IED in the microgrid using a discrete wavelet transform.

[0031] Step 132 includes detecting the presence of a weak signal fault using the IMF components generated by the VMD and determining the fault location if an inverter-related fault is detected. Step 132 further identifies the detected weak signal fault as an inverter-related fault or a high impedance fault. For the inverter-related fault, the fault location is determined according to the current and voltage of the AC side of the corresponding inverter, then proceed to step 138. Otherwise, if the detected fault is not an inverter-related fault, proceed to next step 134.

[0032] In step 134, if the detected fault is a high impedance fault, the current correlation matrix is ​​used to determine the suspected location of the weak signal fault.

[0033] Step 136 includes determining the fault location by applying the time series KNN model using the time series power profile.

[0034] Step 138 involves identifying the fault type using the logic circuit model and the components of the current and voltage sequences and phasors.

[0035] The method 100B includes isolating 140 the faulty line section by activating a switching operation of a switch connected using a computing device 157 over a communication network.

[0036] FIG. 1C is a block diagram illustrating a fault isolation and control system 100 in an islanded inverter-based microgrid according to some embodiments of the present invention.

[0037] The fault identification and control system 100 includes a human machine interface (HMI) 167 connectable to the keyboard 111 and the pointing device / medium 112, a processor 155, a storage device 154, a memory 137, a network interface controller (NIC) connectable to a network 151 including a local area network and an Internet network, a display interface 161 connected to a display device 165, an input interface 139 connectable to an input device 135, and a printer interface 133 connectable to a print device 131. The memory 137 is configured to load a fault identification and control program 159 by associating with the storage device 154 when performing the method 100B. In some cases, the memory 137 and the storage device 154 may be referred to as memories.

[0038] The fault localization and control system 100 can receive electrical signals 195 indicative of time series measurements of current and voltage of intelligent electronic devices 145 disposed in the inverter-based microgrid 115 via a network 151 connected to the NIC 163. The network 151 is connected to an external system(s) 101 capable of providing control signals to the intelligent electronic devices of the microgrid 115 to perform remote control of the intelligent electronic devices. Furthermore, the fault localization and control system 100 can provide fault localization status data (signals) to the external system 101 via the network 151 so that the external system 101 can control switching operations disposed in the microgrid 115. Furthermore, the fault localization and control system 100 can be controlled from the external system 101 by receiving control data (signals) of the fault localization and control system 100 via the network 151.

[0039] The storage device 154 includes pre-fault topology and parameters 158 for the microgrid 115 and a fault identification and control program module 159. The input device / medium 135 may include a module that reads a program stored on a computer readable recording medium (not shown).

[0040] To identify weak signal faults in the microgrid 115 , the fault isolation and control system 100 may receive microgrid status data from intelligent electronic devices 145 included in the microgrid 115 .

[0041] According to some embodiments of the present invention, the microgrid 115 may include a set of terminal buses connected with the line sections, and a set of inverter-based generators or synchronous generators. The microgrid 115 includes IEDs connected with a microgrid controller (not shown) installed in the microgrid 115. The microgrid controller connected with the IEDs is configured to operate and control the microgrid 115 by isolating and restoring the microgrid using the switches of the IEDs in response to receiving a fault identification command indicating a fault location and a fault type of the weak signal fault. In this case, the fault identification command is transmitted from the fault identification and control system 100 to the microgrid controller. In some cases, the microgrid controller is installed in the external system 101 and controls the microgrid via the external system 101. Furthermore, the fault identification command may be transmitted to a display monitor including a display interface (not shown) installed in the external system 101 to indicate a warning regarding the fault location status and the fault type of the weak signal fault to an operator of the external system 101. The fault identification and control system 100 displays the operation state of the microgrid 115 on the display monitor of the external system 101 by sending the fault identification command to a display interface of the display monitor installed in the external system 101 using the fault identification command. The fault identification and control system 100 receives time series measurement data indicating currents and voltages measured for the intelligent electronic devices via the network 151 (communication network) using the interface 153. The memory 137 can load a computer executable program stored in the storage 154, including a pre-fault (normal state) topology and parameters 158 and a fault identification and control program (module) 159 configured to identify a weak signal fault occurring in the microgrid 115 and determine a control scheme for the fault in the microgrid 115.At least one processor 155 and interface 153 in communication with memory 137 are used to execute fault localization and control program 159 loaded from storage 154. For example, when executed by processor 155, fault localization and control program 159 causes processor 155 to receive voltage and current measurements 195 from intelligent electronic device 145, and processor 155 executes fault localization and control program 159 to determine whether weak signal fault occurs in microgrid 115 by examining eigenmode frequency components generated by variation mode decomposition algorithm based on time series measurements denoised using discrete wavelet transform algorithm. If the presence of fault is detected, fault localization and control program 159 further requests processor 155 to provide normal connection topology and parameters from storage 154, and then program 159 first determines suspected fault location based on correlation matrix of neighboring intelligent electronic devices and narrows down the fault location among those locations by utilizing K-nearest neighbors where distance between neighbors is measured by dynamic time warping algorithm. After the fault location is identified, the fault identification and control program 159 further identifies the fault type by monitoring the sequence components and phasor measurements and providing the observation inputs to the logic circuit model. The processor 155 then outputs a post-fault topology indicating the location of the fault and the associated switch operation required by the determined fault control strategy. Furthermore, the interface (NIC) 163 can receive a measurement signal 195 from the microgrid 115 via the network 151 at every preset period. If the fault identification and control system 100 receives time series measurements of current and voltage indicating a predetermined normal signal range from the IEDs installed in the microgrid 115 but does not determine / detect a fault in the microgrid 115, it can generate a normal state command and send a signal of the normal state command via the network 151 to a display interface of a display monitor installed in the external system 101 to display a normal operating state of the microgrid 115 on the display monitor of the external system 101.Normal state command and fault identification command data generated by the fault identification and control system 100 may be transmitted via network 151 to other maintenance system(s) to enable one or more systems to monitor the operating status of the microgrid 115.

[0042] In some cases, instructions to initiate / perform fault isolation may be sent to the fault isolation and control system 100 using the keyboard 111 or from the external system 101 via the network 151 .

[0043] Defining and Simulating Inverter-Based Microgrids A microgrid is a group of interconnected loads and distributed energy resources that act as a single controllable entity to the main grid. A microgrid can be connected or disconnected from the main grid to operate in grid-tied or islanded mode. The distributed energy resources can be power generation sources such as conventional synchronous generators or power inverter-based generators. The loads are supplied by the generators through distribution lines. Intelligent electronic devices (IEDs) are installed in the microgrid to provide real-time information for the microgrid controller to manage the operation and control of the grid.

[0044] Intelligent electronic devices is a term used to describe microprocessor-based controllers of power system equipment such as protection relay devices, tap changer controllers, circuit breaker controllers, capacitor bank switches, recloser controllers, and voltage regulators. In this disclosure, each IED is associated with a switchable device.

[0045] FIG. 2A shows an exemplary inverter-based microgrid with seven buses. The microgrid is connected to a three-phase 25 kV, 200 MVA synchronous generator at bus 5 and a three-phase equivalent generator representing the main grid at bus 1. The microgrid is operated in island mode, i.e., the equivalent generator at bus 1 is disconnected from the microgrid. Buses 2 and 4 contain two inverters. A grid forming inverter (GFM) is connected to bus 4 and a grid following inverter (GFL) is connected to bus 2, both connected to their respective buses through their respective PCCs. Load L1 is rated at 200 kW and 4.16 kV. Loads L2-L7 are rated at 100 kW and 4.16 kV. Two separate PV panels are connected to each inverter through inverter controls and IGBT switches. The DC output from each PV panel is represented by an 8000 V DC source connected to each inverter. Measurements are collected from 12 IEDs installed at locations as shown in FIG. 2A. The microgrid includes three branches. Branch 1 includes IEDs M1-M4, Branch 2 includes IEDs M5-M8, and Branch 3 includes IEDs M9-M12. Earth faults are applied at different locations of the grid. As shown in FIG. 1C, IEDs M1-M12 include switches configured to isolate and restore the microgrid in response to fault identification commands received from the fault identification and control system 100. Each of the IEDs is configured to provide / transmit via the network 151 a data set / signal regarding the varying states and phasor measurements of the sequence components.

[0046] Any system of three unbalanced phasor vectors can be decomposed into three symmetrical components, including a positive sequence, which represents a balanced three-phase system with the same phase sequence as the original sequence, a negative sequence, which represents a balanced three-phase system with the opposite phase sequence to the original sequence, and a zero sequence, which represents a system in which the three phasor vectors are equal in magnitude and phase.

[0047] As shown in Figure 2A, there are two basic control techniques for utility-based inverters: grid-forming (GFL) inverters (Figure 2B) and grid-following (GFL) inverters (Figure 2C). Compared to large synchronous machines, inverter-based resources can change their output much faster and can block system frequency changes before power dispatch is triggered.

[0048] A GFM inverter is a controllable voltage source behind a coupling reactance, similar to a grid-tied synchronous generator. A voltage source inverter with droop characteristics can directly control voltage and frequency. When an unexpected event occurs, a droop-controlled GFM source instantly increases or decreases its output power to balance the load and maintain the local voltage and frequency. In a droop-controlled GFM inverter, there is no significant delay between the change in output power and the change in output frequency. Therefore, a GFM source responds to unexpected events much faster than the response of a GFM source. A GFM inverter can be considered as a voltage source (PV bus).

[0049] The GFL inverter controls the active and reactive power output by injecting current at a given phase angle. A phase-locked loop (PLL) is used to track the grid phase angle in real time. The GFL inverter cannot directly regulate the grid voltage and frequency. The voltage and frequency references are provided externally by the GFM inverter or the main grid. Essentially, the GFL inverter must shut down if it loses its voltage / frequency source. The GFL inverter can be considered as a current source (PQ bus).

[0050] Weak Signal Fault Detection for Islanded Inverter-Based Microgrids Weak signal faults in microgrids are detected by using a two-step time-domain frequency component analysis based approach including discrete wavelet transform (DWT) and variational mode decomposition (VMD).

[0051] The Discrete Wavelet Transform (DWT) is a transform that decomposes a given signal f into multiple sets, each set being a time series of coefficients that describes the time evolution of the signal in a corresponding frequency band. The DWT can be used to denoise a noisy signal by synthesizing the original signal into different wavelet components with a preset wavelet function, decomposition level and signal expansion mode, and compress the noise components by applying a signal thresholding algorithm with a preset threshold and threshold mode. The basic idea of ​​DWT-based denoising is that the DWT concentrates the signal features into a small number of large wavelet coefficients. Wavelet coefficients with small values ​​are generally noise, and these coefficients can be avoided or removed without affecting the signal quality. After thresholding the coefficients, the data can be reconstructed.

[0052] For any noisy signal, the DWT decomposes the original time series data set into an approximation component with high scale and low frequency and a set of detail components with high frequency and low scale at different resolution levels. The approximation and detail components are defined by the sum of approximation and detail functions weighted by corresponding approximation and detail coefficients, where each approximation and detail function is specific to a base wavelet function. The denoising of a noisy signal can involve the steps of (1) applying a forward discrete wavelet transform to the data using a wavelet function with a preset signal enhancement mode and a preset decomposition level to obtain a coefficient list of the approximation and detail functions, (2) applying a signal thresholding technique to the detail coefficients with a preset threshold and a preset threshold mode, and (3) applying an inverse discrete wavelet transform with the same settings of the base wavelet function, signal enhancement mode, and decomposition level to reconstruct the signal using the thresholded approximation and detail coefficients. The exemplary settings of the above DWT denoising are as follows: the wavelet function is Wavelet Daubechies 4 function, the decomposition level is set to 5, the signal enhancement mode is periodic padding, the threshold mode is soft thresholding, and the threshold is 0.4.

[0053] VMD is an adaptive quasi-orthogonal decomposition algorithm that aims to separate a noisy signal f into k discrete frequency modes, each with an (estimated) center frequency ω k The VMD is compact, with each mode (estimated) u k We obtain a one-sided frequency spectrum by applying a Hilbert transform to , and then shift the frequency spectrum of each mode to baseband by mixing in an exponential component centered on the estimated mode frequency. The final output is a collection of k intrinsic mode function (IMF) components, each at frequency ω k Harmonic components in u kIMFs, as opposed to simple harmonic functions, generally represent simple vibration modes. By definition, an IMF is any function with the same number of extrema and zero crossings and whose envelope is symmetric about zero.

[0054] In the disclosed two-step approach, the first step is a signal reconstruction using a discrete wavelet transform (DWT) technique, which removes noise from the input signal, i.e., the current and voltage data from the measurement device. The purpose of the first noise removal step is to make the disclosed approach insensitive to noise. In practical applications, there are multiple measurement and system noises that cause erroneous inferences from the output. The disclosed two-step approach reduces the dependence of the robustness of the algorithm on such noise. In the second step, the reconstructed signal is taken as the input, and the inputs are synthesized as Intrinsic Mode Function (IMF) components using a VMD algorithm. In particular, each of the IMF components represents an individual harmonic component in the time domain, and during periods of weak signal failure, one or more IMF components are expected to show observable changes. The second step provides time series data for each IMF component, thereby indicating the presence of high frequency, small magnitude components in the measurements.

[0055] One of the major advantages of the disclosed approach is that it is scalable to grid systems of different configurations. Whatever the type of grid configuration and topology, there exists a time relationship between adjacent locations of the grid, so measurements from locations relatively far from the fault source will show some variation over the fault period. Therefore, regardless of the grid topology, the IMF components of measurements from any location close to the actual fault source can detect the occurrence of a weak signal fault.

[0056] To demonstrate the effectiveness of the disclosed method, a typical fault in a microgrid as shown in Figure 2A is simulated using MathWorks SIMULINK.

[0057] Each type of ground fault including three-line to ground (TLG), two-line to ground (DLG), and one-line to ground (SLG) occurs separately near bus 2 and bus 4. TLG is simulated as a phase A to phase B to phase C to ground fault and is therefore represented as an ABCG fault. Similarly, DLG and SLG are simulated as phase A to phase B to ground and phase A to ground faults, respectively, and are therefore represented as ABG and AG faults, respectively. Inverter DC side short circuit is modeled by short circuiting the DC power supply from the PV panels of each inverter separately. Inverter tripping is represented by tripping the three-phase connection near the PCC for each inverter. The fault type at each of the two aforementioned locations occurs at time 2.5 seconds and is cleared at 3.0 seconds. Inverter tripping occurs from 2.5 seconds to 2.6 seconds.

[0058] Current and voltage phasors, current and voltage sequence components, real and reactive power measurements are collected at time intervals between 2.0 and 3.5 seconds. The data set is sampled at a frequency of 1 kHz, resulting in 1500 measurements over a 1.5 second period.

[0059] Figures 3-7 show the current waveforms of different types of weak signal faults simulated in the microgrid shown in Figure 2A. The waveforms are generated by the VMD algorithm and implemented in IMF mode, whose total number is predefined as 5. The current and voltage measurements received from the IEDs are first denoised with DWT and then fed as input to the VMD.

[0060] Figure 3 shows the current waveforms of the ABCG fault in IMF mode 1. Next, current measurements of the ABCG fault (fault impedance 6000 Ω) on the three-phase line near bus 4 are collected by IED M5. As shown in Figure 3, significant abnormal signal fluctuations in the current waveform can be seen at the aforementioned fault duration of 2.5 seconds to 3.0 seconds. Although the fault is located near bus 4, the current measurements in Figure 3 are from IED M5, reflecting the effectiveness of the VMD algorithm in indicating the presence of a high impedance fault using measurements from any location in the system.

[0061] The abnormal signal variation of the measurement signal is identified when the ratio of the absolute value of the difference in length between the normal signal range of the measurement signal and the abnormal signal range of the measurement signal to the length of the normal signal range of the measurement signal is greater than a fault threshold. The fault threshold can be defined as a ratio based on the difference between the normal signal range and the abnormal signal range. The normal signal range and the abnormal signal range are determined based on the variation range of the measurement signal over a given period (e.g., 2.0 seconds to 3.0 seconds) having a preset length, such as about 1.0 seconds.

[0062] The fault threshold is preset, such as to about 0.15. The fault threshold can be stored in storage 154 and can be used to automatically detect faults in the microgrid 115 using the fault identification and control system 100.

[0063] The above-mentioned approach is a static approach. Identification can also be achieved by using a dynamic approach based on a comparison of signal variations between successive monitoring time windows with the same width. An abnormal signal variation of the measurement signal is identified if the ratio of the absolute value of the difference in length between the signal variation ranges of the measurement signal in two successive monitoring time windows to the length of the signal variation range of the measurement signal in the first of the two successive monitoring time windows is greater than a fault threshold. The width of the monitoring time window is preset to, for example, 0.1 seconds. For example, the maximum current range in FIG. 3 changes from around 800 amperes in the first time window (i.e., 2.4 seconds to 2.5 seconds) to around 1000 amperes in the second time window (i.e., 2.5 seconds to 2.6 seconds). If the fault threshold is set to 0.15, the ratio of the absolute difference (around 200 amperes) to around 800 amperes is around 0.25, which is greater than the fault threshold (0.15), and therefore it can be identified that the abnormal signal variation starts at 2.5 seconds.

[0064] Figures 4 and 5 show the current waveforms for IMF mode 1 and mode 2 ABG faults.

[0065] When an ABG fault occurs on the three-phase line near bus 2 with a fault impedance of 3000 Ω, as shown in Figure 4, a similar abnormal signal fluctuation can be identified in the waveform of current IMF mode 1 from 2.5 seconds to 3.0 seconds. However, since fluctuations exist in this waveform at earlier time instances, IMF2, as shown in Figure 5, needs to be included in the analysis to confirm the presence of the fault.

[0066] Figure 6 shows the current waveforms of an inverter DC-side short-circuit (DCSC) fault in IMF mode 2. As shown in Figure 6, for a DCSC fault in the DC supply of the GFM inverter connected to bus 4, the results obtained from the IMF2 component show the presence of the fault between 2.5 seconds and 3.0 seconds.

[0067] Figure 7 shows the current waveforms of an inverter tripping (IT) fault in IMF mode 2. The IT fault is introduced at 2.5 seconds and removed at 2.6 seconds, so the current phasor shows abnormal signal variations in this period. Similar to the ABG fault and inverter DC side short circuit fault (DCSC), the IT fault can be detected in the IMF2 component.

[0068] As shown in Figures 3-7, VMD can effectively determine the presence of a weak signal fault in an islanded inverter-based microgrid.

[0069] Weak Signal Fault Localization in Islanded Inverter-Based Microgrids A common assumption behind low impedance faults is that measurements near the fault location will vary more than measurements at locations far from the fault. However, this assumption is not valid for weak signal faults because the measurement deviations at all locations will be too small to be detected by the relay.

[0070] Location of inverter related faults including inverter tripping and inverter DC side short circuit is relatively straightforward. The current at the PCC of the tripped inverter indicates the presence of inverter tripping at the respective inverter. In case of inverter DC side short circuit, the voltage and current phasors at the output of the particular inverter are expected to show significant changes during the fault period. Therefore, inverter related faults can be easily identified if abnormal signal variations are identified in the currents and voltages measured at the AC side of the inverter.

[0071] The inventors focus on localizing high impedance faults by utilizing Kirchhoff's Current Law (KCL) at the nodes of an electric circuit. Assume that for a particular branch under normal operating conditions, IEDs installed at two different locations report currents I1 and I2 in the same branch. According to KCL, when there is no fault condition between the two measurement devices, I1 and I2 are expected to follow a similar trend. However, when a fault exists between these two measurement points, a conductive path is formed between the fault point and ground, as shown in Figure 8, resulting in I1 and I2 following an opposite trend, which is not possible between other measurement devices in the grid. The absolute value of the correlation calculated between current measurements obtained from two adjacent measurement devices is generally expected to be high and close to 1, but under fault conditions the correlation becomes very small due to the opposite trends in the time domain. Thus, the correlation can be used to identify the fault location.

[0072]

number

[0073]

number

[0074] One advantage of the disclosed method is that the covariance-based matrix is ​​applicable to all types of ground faults, including single line to ground (SLG), double line to ground (DLG), and triple line to ground (TLG). In addition to this, the disclosed method uses data obtained from IEDs already installed at specific locations in the grid, rather than from additional specific devices. That said, the disclosed correlation-based method has some limitations. First, the correlation between two nodes may be less than a threshold τ for fault conditions occurring at locations far away from the fault. Second, the correlation between these two nodes may be less than τ if there is an existing load or slack generator connected between them. Therefore, the correlation matrix does not give only the correct fault location, but rather gives several possible fault locations (i.e., candidate fault locations), one of these locations being the actual fault location. Although the correct location of the fault may not be obtained, the correlation-based approach helps narrow down the multiple locations to only a few suspect locations (i.e., candidate fault locations).

[0075] To further narrow down the suspect locations, a supervised K-nearest neighbor (KNN) based method is developed that incorporates the power consumption profile. The K-nearest neighbor (KNN) model is a non-parametric supervised learning classifier that utilizes proximity to make classifications or predictions regarding groupings of individual data points. In classification problems, class labels are assigned based on majority vote; that is, the label that appears most frequently around a given data point is used. The metric that judges the KNN model is important for accurate classification. Euclidean distance is commonly used as a metric to find the nearest neighbors between input data sets, where the one-to-one distance between two points is calculated simultaneously from two different data sets. For this reason, Euclidean distance does not provide accurate distance information between two time series data sets when the data sets are not perfectly aligned with each other in the time domain. By using dynamic time warping (DTW) to calculate the nearest neighbors of the time series input data sets in the KNN model, the distance between two time series data sets with different time alignments can be more accurately measured. Therefore, considering that the power consumption profile through the microgrid branches is time series data, DTW is adopted as the metric for the KNN model.

[0076]

number

[0077] In contrast to existing studies, we use the power consumption data of each branch for each fault incident as input to KNN, and the output from KNN classifies the branch as faulty / non-faulty. The KNN model identifies the branch containing the faulty node among the suspected faulty nodes provided by the correlation matrix-based approach. Combining the results of these two methods, we can pinpoint the actual fault location in the microgrid. By selecting the optimal number of neighbors K for KNN and selecting the correlation threshold τ, we can achieve a higher accuracy of localization. Because power grid data contains noise, the correlation between two similar time series measurements may be lower than 0.99. Therefore, a value of 0.90 or higher, e.g., 0.90-0.98, is suitable for the threshold τ.

[0078] The KNN model is trained on a set of training cases with denoised branch power consumption data as input and fault-labelled data as output, where the training cases include non-faulty and faulty cases, where the faulty cases include different fault types, different fault impedances, and different fault locations in the branches, and the number of neighbors K of the KNN is preset.

[0079] The differences between the disclosed fault location method and existing works are as follows: (1) Regardless of the system configuration and topology, the current measurements on both sides of the fault location exhibit different behaviors and have relatively small correlations compared to the correlations between current measurements from two different adjacent locations. (2) The KNN model uses power consumption data from all branches of the system. In both steps, the fault location information does not depend on the system configuration and topology, but rather on the general behavior of the current and power consumption profiles in the time domain. By properly tuning the KNN parameters, a near perfect location accuracy is obtained. (3) In the case of inverter tripping and inverter DC side short circuit, we observe the current and voltage behaviors of only the affected inverter and the corresponding PCC. Since the affected inverter and PCC exhibit the expected behaviors during the fault regardless of the system configuration, the disclosed method becomes scalable for DC side short circuit faults and inverter tripping faults. Thus, the disclosed hybrid method eliminates the dependency of weak signal localization on the grid structure, thereby being scalable to any large-scale grid system.

[0080] To use the above method to identify the location of faults in the microgrid shown in Figure 2A, the positive sequence component of the current is used to construct a correlation matrix C. For each fault type at each of the two fault locations, 1.5 seconds (2 seconds to 3.5 seconds) of total power consumption data is extracted. In addition, for each fault case, the fault impedance is also varied. To demonstrate high impedance faults, three very large values ​​of fault impedance are considered: 1200 Ω, 3000 Ω, and 6000 Ω. Since there are 54 combinations of fault impedances, 54 time series power consumption data are provided for each branch and fed into the input of KNN. For three branches, the total size of input samples containing three non-fault states is 165. The output is classified as 1 if the corresponding branch contains a fault and 0 otherwise. The training and test datasets are split in a ratio of 80:20. The number of neighbors K is varied from about 2 to about 6 so that the accuracy of KNN is the highest. The fault classification accuracy τ versus the correlation matrix threshold τ is varied from 0.9 to 0.98 for analysis.

[0081] With reference to Fig. 2A, a 4 × 4 correlation matrix is ​​created in the first step for each branch. Fig. 9-Fig. 11 show the correlation matrices of the three branches of the grid for a TLG fault with the fault occurring near bus 4. For a correlation threshold τ = 0.9, three possible fault locations are obtained: element (1,2) = (2,1) in branch 1, element (2,3) = (3,2) in branch 2, and element (2,3) = (3,2) in branch 3. Since faults are expected only between two consecutive IEDs, with current measurements going in opposite directions, only elements with one index on the right / left of the diagonal are considered. For example, the (2,3)-th element in branch 3 corresponds to a point between M10 and M11, and therefore indicates that the fault may have occurred near bus 4. This result shows that the correlation matrix-based method is effective for fault localization. Although the bus correlation matrix-based method can indicate possible fault locations, including the actual fault location, it gives multiple suspected locations, which leads to the possibility of false positive (FP) cases. FPs are defined as cases where a non-fault location is deemed suspicious by the disclosed fault localization method. After training KNN with denoised branch power consumption data as input and fault-labelled data as output, we test the accuracy of the results on a test dataset. Accuracy is defined as the ratio of correct predictions over the total number of test cases.

[0082] Figure 12 shows the best accuracy when the number of neighbors is K=3, i.e. the best choice for the number of neighbors is 3. For the same experiment with random Gaussian noise added to the power consumption data, the table shows that the KNN accuracy remains unchanged.

[0083] While the correlation matrix alone can provide a set of correct fault location candidates, in combination with KNN, the number of false positives (FPs) can be reduced. The impact of KNN on FP cases is evident from Figure 13. However, the number of FP cases depends on the correlation threshold τ. For a correlation threshold of 90% and K=3, the disclosed fault detection method provides optimal results in locating high impedance weak signal faults.

[0084] Weak Signal Fault Classification for Islanded Inverter-Based Microgrids In general, fault classification can be achieved by using the superimposed sequence components of currents. In a conventional grid system, different earth faults can be classified by analyzing the relative behavior of the superimposed current sequence components. However, in the case of an inverter-based microgrid, the positive and negative sequence currents are controlled by the reference point of the converter control system. Due to this phenomenon, the current injected by the inverter-based microgrid behaves differently than in the case of a conventional grid. The disclosed fault classification method focuses on the behavior of the current sequence components in the time series measurements. Instead of focusing on the sequence component measurements of a single sample, the time series attributes over the fault duration including the instant of initialization and the instant of fault removal behave differently for each fault type.

[0085] The disclosed fault classification method is based on a logic circuit model as shown in FIG. 14. The logic circuit model defines the relationship between IED time series measurements indicative of fault signatures over the fault duration due to weak signal fault types. The fault types include inverter-based faults and high impedance faults. Inverter-based faults include inverter tripping (IT) and inverter direct current side short circuit (DCSC). High impedance faults include three line to ground faults (TLG), two line to ground faults (DLG), and one line to ground faults (SLG). The fault signatures indicative of IED measurements include positive sequence current change, negative sequence current change, negative sequence current transient, phase current change, phase voltage current, and inverter PCC current near zero.

[0086] Fault classification is performed after weak signal fault is detected by analyzing the time series measurements from t0 to t0+T, and the fault location is identified by hybrid correlation matrix and KNN method. Once the fault location is identified between two IEDs in the distribution line, the sequence components of the corresponding time series of current measurements along with the voltage and current phasor measurements from these devices are fed into the logic circuit of Figure 14 to classify the fault type. The observations from the sequence components and phasor measurements are utilized in the logic circuit model as shown in Figure 14. If a particular sequence component or phasor tour measurement shows abnormal signal variation between t1 and t2, the corresponding input in the logic circuit of Figure 14 is Boolean1, otherwise it is Boolean0. The output of only the corresponding fault type is Boolean1, otherwise it is Boolean0.

[0087] During an inverter DC side short circuit (DCSC) fault of a particular inverter, the voltage and current measurements of the AC side of the affected inverter are expected to vary significantly. If the sinusoidal voltage and current measurements of the AC side of a particular inverter show abnormal signal variations during the fault period, the corresponding fault type can be classified as inverter DC side short circuit. A DCSC fault is determined when abnormal signal variations are sensed for the voltage or current phasors and positive sequence components of the current on the AC side of the investigated inverter, as shown in Figure 14.

[0088] In case of inverter tripping, the inverter is disconnected from the line for a short period of time, so the magnitude of the current at the PCC is greatly reduced. Therefore, when the current phasor is close to zero and the signal of the positive sequence component fluctuates abnormally, it indicates an inverter trip fault. The dedicated logic circuit in Figure 14 shows that when the current phasor is close to zero and the signal of the positive sequence component fluctuates abnormally, it is determined that the inverter has tripped.

[0089]

number

[0090] [Number]

[0091] [Number]

[0092] [Number]

[0093] [Number]

[0094] Fault current I F Since the exact value of depends on the fault impedance, system topology, and the machines connected near the fault location, it is difficult to predict the exact value of the fault current during a high-impedance fault. Despite using exact numerical values or ranges of values, a more generalized and topology-independent classification of faults should be obtained from the time-series behavior of different sequence components during a fault. (4)-(6) show the relative behavior of sequence components during different types of faults. The time-series current sequence measurements from t0 to t0+T (T is a positive non-zero value, the fault occurs at t1 and is removed at t2 (t2<T)) contain information about fault types such as TLG, DLG, and SLG.

[0095] The main advantage of the disclosed fault classification method is its low sensitivity to location. The disclosed method first detects and localizes suspect locations, and then applies the logic circuit of FIG. 14 only at the suspect locations. Therefore, the disclosed classification method is less likely to give an incorrect fault type. The second advantage of the disclosed method is its scalability. Since the fault locations are already determined, it is sufficient to observe the current sequence components and phasor measurements only from the suspect locations and input the corresponding Boolean inputs to the logic circuit of FIG. 14. For more complex grid systems, fault location reduces the suspect locations to only one or a few points, so the accuracy of the disclosed logic circuit-based method is expected to be high even for complex large-scale grid systems. Classification of inverter direct current side short circuit (DCSC) faults and inverter tripping (IT) faults is the third advantage of the disclosed method over existing studies, since existing studies have not addressed the classification approach for these two faults.

[0096] Fault classification by the disclosed logic circuit of FIG. 14 requires Boolean inputs. If a particular phasor or sequence measurement shows abnormal signal variation during the fault period, the corresponding input signal is set to 1. The output from the logic circuit indicates the fault type. With earth faults at two different locations and three different impedance levels each, 18 earth faults are possible. Furthermore, inverter DC side short circuits (DCSC) were applied separately at each DC source, the GFL inverter at bus 2 and the GFM inverter at bus 4. Finally, each inverter was tripped individually near the PCC to generate a total of 22 faults.

[0097] In summary, the fault type of the weak signal fault is classified by the fault classification logic circuit, and if abnormal signal variation is identified for the positive sequence component of the current and no abnormal signal variation is identified for the negative sequence component of the current or abnormal signal variation of a negative sequence transient is identified, the fault type is determined as a three line to ground fault (TLG); if abnormal signal variation is identified for the positive and negative sequence components of the current, the fault type is determined as a two line to ground fault (DLG); if abnormal signal variation is identified for the positive and negative sequence components of the current or no abnormal signal variation is identified for the positive and negative sequence components of the current, the fault type is determined as a single line to ground fault (SLG); if abnormal signal variation is identified for the inverter AC side voltage and abnormal signal variation is identified for the inverter AC side current, the fault type is determined as an inverter DC side short circuit (DCSC) fault; if abnormal signal variation is identified for the positive sequence component of the current and the phase current at the inverter PCC is close to zero, the fault type is determined as an inverter tripping (IT) fault.

[0098] FIG. 15 shows a comparison of classification results by the disclosed logic circuit with actual fault types. The disclosed method shows 100% accuracy for SLG, TLG, inverter DC side short circuit (DCSC), and inverter tripping (IT) faults because it locates the fault before running the classification algorithm. However, for DLG, the disclosed logic circuit shows Boolean1 for both DLG and SLG faults. This result means that there is uncertainty between DLG and SLG faults when the actual fault is only DLG. However, this uncertainty is not expected to be a significant issue for system operators because the mitigation and repair techniques for DLG and SLG faults are similar.

[0099] Testing Notes for an Exemplary Islanded Inverter-Based Microgrid From the above results, it is clear that the disclosed fault identification method can detect weak signal faults and accurately locate the fault with properly tuned parameters even in noisy conditions. Furthermore, the classification method can identify the correct fault type, ensuring accurate mitigation and restoration actions to maintain power grid resilience.

[0100] The disclosed fault localization method presents an effective technique for fault detection, localization, and classification in islanded inverter-based microgrids using voltage and current measurements available from conventional IEDs. First, we applied VMD to detect the presence of a fault at a predefined mode number, such as 5. The results revealed that measurements from any location in the grid can indicate the presence of a weak signal fault for the duration of the event. After detecting the presence of a fault, we applied correlation matrix and KNN to localize the fault. The results showed that the correlation matrix gives the fault location accurately in relatively high FP cases. By further combining KNN with the correlation matrix, the number of FP cases can be significantly reduced from a range of 7-10 to a range of 0-3. Notably, KNN yields 100% accuracy at K=3, while the method combining correlation matrix and KNN yields 0 number of FPs at a threshold of 0.9. Furthermore, it is shown that the accuracy of the disclosed method is independent of measurement noise. The last part of the disclosed method is classification using the relative behavior of sequence components and phasor measurements. The results showed 100% accuracy for all types of faults, except for the uncertainty between double line to earth (DLG) and single line to earth (SLG).

Claims

1. A system for identifying the presence, location and type of weak signal faults in a microgrid, wherein the microgrid is connected to a main grid equivalent to a synchronous generator and includes at least one branch, the at least one branch includes at least one line section connected between two buses, at least one of the two buses is connected to a synchronous generator, at least one bus is connected to a grid forming inverter (GFM), at least one bus is connected to a grid following inverter (GFL), the at least one bus is connected to a load, at least two switches are connected to different locations on the at least one branch, each of the at least two switches is provided with an intelligent electronic device (IED) configured to monitor and control at least one switch, the microgrid is operated in island mode, and the system is, The system comprises at least one processor and a memory storing instructions that cause the at least one processor to execute a step, and the step is The steps include receiving time-series measurements of current and voltage from at least two IEDs installed in the microgrid, The steps include: denoising the time-series measurements using a discrete wavelet transform (DWT) algorithm; The process includes the step of detecting whether each of the denoised time-series measurements indicates the occurrence of a weak signal fault by frequency component analysis using a variational mode decomposition (VMD) algorithm, and if the weak signal fault is detected, the at least one processor further: The steps include: determining whether the weak signal fault is an inverter-related fault or a high impedance fault; determining the fault location for the identified inverter-related fault according to the AC current and voltage of the corresponding inverter; identifying the branch containing the weak signal fault using a time-series K-nearest neighbors (KNN) model and a dynamic time warping (DTW) algorithm; determining the fault candidate location for each of the at least one branch and the fault location of the weak signal fault using the correlation between the time-series components of the denoised time-series measurements between the at least two IEDs; and further, The steps include: identifying the type of the weak signal fault determined using a fault classification logic circuit, according to the fluctuation state of the sequence components and phase vector measurements over the fault period; The steps include generating a fault identification command that indicates the fault location and the identified fault type, A system comprising the steps of isolating and restoring the microgrid in accordance with the fault identification command by transmitting a control signal for the fault identification command to the microgrid controller of the microgrid.

2. The system according to claim 1, wherein the inverter-related failure is a short-circuit failure on the DC side of the inverter or an inverter tripping failure.

3. The system according to claim 1, wherein the high-impedance fault is a ground fault with a high fault impedance, and the type of ground fault is a triple-line-to-ground (TLG), a double-line-to-ground (DLG), or a single-line-to-ground (SLG).

4. The fault type of the weak signal fault is classified by the fault classification logic circuit. If an abnormal signal fluctuation is identified for the positive sequence component of the current, and the abnormal signal fluctuation is not identified for the negative sequence component of the current, or if the abnormal signal fluctuation is identified for a negative sequence transient event, the fault type is determined to be a three-line ground fault (TLG). If the abnormal signal fluctuation is identified for both the positive and negative sequence components of the current, the fault type is determined to be a two-line ground fault (DLG). If the abnormal signal fluctuation is identified for both the positive and negative sequence components of the current, or if the abnormal signal fluctuation is not identified for both the positive and negative sequence components of the current, the fault type is determined to be a one-line ground fault (SLG). If the abnormal signal fluctuation is identified for the voltage on the AC side of the inverter, and the abnormal signal fluctuation is identified for the current on the AC side of the inverter, the fault type is determined to be a DC side short circuit (DC short). The system according to claim 1, wherein if a circuit (DCSC) fault is determined, the abnormal signal fluctuation is identified for the positive sequence component of the current, and the phase current in the inverter's PCC is close to zero, the fault type is determined to be an inverter tripping (IT) fault.

5. The system according to claim 4, wherein the abnormal signal fluctuation of the time-series measurement is identified when the ratio of the absolute value of the difference between the length of the normal signal range of the time-series measurement and the abnormal signal range of the time-series measurement to the length of the normal signal range of the time-series measurement is greater than a fault threshold, and is not identified otherwise, the fault threshold is set in advance, and the normal signal range and the abnormal signal range are determined based on the fluctuation range of the time-series measurement over a period having a set length.

6. The system according to claim 4, wherein if the ratio of the absolute value of the difference in the lengths of the signal fluctuation ranges of the time series measurement values ​​in the two consecutive monitoring time windows to the length of the signal fluctuation range of the time series measurement value in the first monitoring time window of two consecutive monitoring time windows is greater than a fault threshold, an abnormal signal fluctuation of the time series measurement value is identified, the monitoring time windows are time intervals having the same width, and the fault threshold is preset.

7. The step of denoising using the DWT algorithm described above is further: First, the time-series measurements are combined into different wavelet components using a preset wavelet function, a preset resolution level, and a preset signal expansion mode. Next, the system according to claim 1, further comprising compressing noise components by applying a signal thresholding algorithm with a preset threshold and threshold mode.

8. The system according to claim 7, wherein the pre-set wavelet function is the Wavelet Daubechies4 function, the pre-set resolution level is set to 5, the pre-set signal expansion mode is periodic padding, the pre-set threshold is 0.4, and the pre-set threshold mode is soft thresholding.

9. The system according to claim 1, wherein a limited number of eigenmode functions (IMFs) of the denoised time-series measurements are used to detect the presence of a fault, and the number of IMFs is preset.

10. If abnormal signal fluctuations are identified in the current and voltage measured on the AC side of the inverter, an inverter-related fault is identified, and the normal signal range of the time-series measurement and the abnormal signal range of the time-series measurement are compared in relation to the length of the normal signal range of the time-series measurement. The system according to claim 1, wherein if the ratio of the absolute values ​​of the difference in length between the normal and abnormal signal ranges is greater than a fault threshold, an abnormal signal fluctuation in the time-series measurement is identified, the fault threshold is set in advance, and the normal signal range and the abnormal signal range are determined based on the fluctuation range of the time-series measurement over a period having a set length.

11. The system according to claim 1, wherein an inverter-related fault is identified when an abnormal signal fluctuation is identified with respect to the current and voltage measured on the AC side of the inverter, and an abnormal signal fluctuation of the time-series measurement is identified when the ratio of the absolute value of the difference in the length of the signal fluctuation range of the time-series measurement in the two consecutive monitoring time windows to the length of the signal fluctuation range of the time-series measurement in the first of two consecutive monitoring time windows is greater than a fault threshold, and the monitoring time windows are time intervals having the same width, and the fault threshold is preset.

12. The system according to claim 1, wherein a correlation matrix is ​​created for each branch using the denoised positive sequence component of the time-series measurement values ​​of the currents of all IEDs installed along the branch, and a candidate fault location is identified between two IEDs at the branch if the corresponding correlation is less than a correlation threshold, the correlation threshold is preset between about 0.9 and about 0.

98.

13. The system according to claim 1, wherein the time-series KNN model is trained on a set of training cases using denoised branch power consumption data as input and fault-labeled data as output, the training cases include non-fault cases and fault cases, the fault cases include different fault types, different fault impedances, and different fault locations for the branches, and the number of neighbors of the KNN is set between about 3 and about 5.

14. A method for identifying the presence, location, and type of a weak-signal fault in a microgrid, wherein the microgrid is connected to a main grid equivalent to a synchronous generator, includes at least one branch, each branch includes at least one line section connected between two buses, at least one bus is connected to a synchronous generator, at least one bus is connected to a grid forming inverter (GFM), at least one bus is connected to a grid following inverter (GFL), at least one bus is connected to a load, at least two switches are installed at different locations on the same branch, each switch is provided with at least one intelligent electronic device (IED) for monitoring and controlling the switch, the microgrid is operated in island mode, the weak-signal fault may be an inverter-related fault or a high-impedance fault, and the method is: The system comprises using at least one processor and a memory containing instructions that cause the at least one processor to execute a step, and the step is The steps include receiving time-series measurements of current and voltage from at least two IEDs installed in the microgrid, The steps include: denoising the time-series measurements using a discrete wavelet transform (DWT) algorithm; The process includes the step of detecting whether each of the denoised time-series measurements indicates the occurrence of a weak signal fault by frequency component analysis using a variational mode decomposition (VMD) algorithm, and if the weak signal fault is detected, the at least one processor further: The step of determining whether the weak signal fault is an inverter-related fault or a high impedance fault is performed, and for the identified inverter-related fault, the corresponding inverter The fault location is determined according to the AC current and voltage, and for the high impedance fault, the branch containing the weak signal fault is identified using a time-series K-nearest neighbors (KNN) model and a dynamic time warping (DTW) algorithm, and based on the fault candidate location, the fault candidate location is determined for each of the at least one branch and the fault location of the weak signal fault using the correlation between the time-series components of the denoised time-series measurements between the at least two IEDs, and the step further, The steps include: identifying the type of the weak signal fault determined using a fault classification logic circuit, according to the fluctuation state of the sequence components and phase vector measurements over the fault period; The steps include generating a fault identification command that indicates the fault location and the identified fault type, A method comprising the step of isolating and restoring the microgrid in accordance with the fault identification command by transmitting a control signal for the fault identification command to the microgrid controller of the microgrid.

15. The method according to claim 14, wherein the inverter-related failure is an inverter DC side short circuit failure or an inverter tripping failure.

16. The method according to claim 14, wherein the high-impedance fault is a ground fault with a high fault impedance, and the type of ground fault is a triple-line-to-ground (TLG), a double-line-to-ground (DLG), or a single-line-to-ground (SLG).

17. The fault type of the weak signal fault is classified by the fault classification logic circuit. If an abnormal signal fluctuation is identified for the positive sequence component of the current, and the abnormal signal fluctuation is not identified for the negative sequence component of the current, or if the abnormal signal fluctuation is identified for a negative sequence transient event, the fault type is determined to be a three-line ground fault (TLG). If the abnormal signal fluctuation is identified for both the positive and negative sequence components of the current, the fault type is determined to be a two-line ground fault (DLG). If the abnormal signal fluctuation is identified for both the positive and negative sequence components of the current, or if the abnormal signal fluctuation is not identified for both the positive and negative sequence components of the current, the fault type is determined to be a one-line ground fault (SLG). If the abnormal signal fluctuation is identified for the voltage on the AC side of the inverter, and the abnormal signal fluctuation is identified for the current on the AC side of the inverter, the fault type is determined to be a DC side short circuit (DC short). The method according to claim 14, wherein a circuit (DCSC) fault is determined, the abnormal signal fluctuation is identified for the positive sequence component of the current, and the phase current in the inverter's PCC is close to zero, the fault type is determined to be an inverter tripping (IT) fault.

18. The method according to claim 17, wherein the abnormal signal fluctuation of the time series measurement is identified when the ratio of the absolute value of the difference between the length of the normal signal range of the time series measurement and the abnormal signal range of the time series measurement to the length of the normal signal range of the time series measurement is greater than a fault threshold, and is not identified otherwise, the fault threshold is set in advance, and the normal signal range and the abnormal signal range are determined based on the fluctuation range of the time series measurement over a period having a set length.

19. If the ratio of the absolute value of the difference in the length of the signal fluctuation range of the time series measurement in the first of two consecutive monitoring time windows to the length of the signal fluctuation range of the time series measurement in the first monitoring time window is greater than the failure threshold, then the time series measurement The method according to claim 17, wherein a constant abnormal signal fluctuation is identified, the monitoring time window is a time interval having the same width, and the fault threshold is preset.

20. The step of denoising using the DWT algorithm described above is further: First, the time-series measurements are combined into different wavelet components using a preset wavelet function, a preset resolution level, and a preset signal expansion mode. The method according to claim 14, further comprising compressing noise components by applying a signal thresholding algorithm to a preset threshold and threshold mode.

21. The system generates a normal state command while receiving time-series measurements of current and voltage indicating a preset normal signal range, or a fault identification command indicating the fault location and identified fault type, from at least two IEDs installed in the microgrid. The method according to claim 14, further comprising transmitting an operating status signal indicating the normal state command or the fault identification command to an external system via a network, thereby causing the operating status of the microgrid to be displayed on a display monitor of the external system.

Images