Active power distribution network grounding fault identification method and system based on transient relay protection

Abstract

The invention provides an active power distribution network grounding fault identification method and system based on transient relay protection, and relates to the field of electrical fault detection, and the method comprises the steps: carrying out the frequency domain conversion of the transient electrical quantity monitored by an active power distribution network in real time, and generating a transient frequency domain feature; matching the transient frequency domain features with a preset artificial intelligence feature library to obtain one or more candidate fault modes; for each candidate fault mode, performing transient simulation calculation based on the topological model of the active power distribution network to obtain a theoretical transient characteristic corresponding to each candidate fault mode; determining the candidate fault mode corresponding to the theoretical transient characteristic with the highest similarity with the transient frequency domain characteristic as a fault identification result; and generating a protection action instruction for executing fault isolation according to the fault identification result. By implementing the method, the defects that steady-state characteristics are easily interfered and the misjudgment rate is high under complex working conditions such as multi-point power supply and bidirectional power flow can be overcome, and the accuracy and the real-time performance of ground fault identification are improved.

Description

Technical Field

[0001] This application relates to the field of electrical fault detection, and in particular to an active distribution network grounding fault identification method and system based on transient relay protection. Background Technology

[0002] With the large-scale integration of new energy power generation and distributed power sources, distribution networks are gradually developing towards active systems. Active distribution networks, due to their bidirectional power flow and multi-point power supply characteristics of distributed power sources, greatly improve the operational flexibility and power supply reliability of the distribution network. However, the complex structure and variable operating conditions of active distribution networks also bring new operational and management challenges. Especially when ground faults occur in the distribution network, how to identify the fault in a timely and accurate manner and take corresponding protection and control measures has become an important issue in ensuring the safe and stable operation of the distribution network. Rapid and accurate identification of ground faults is a key link in achieving fault isolation, fault self-healing, and improving the system's power supply reliability.

[0003] In related technologies, existing techniques for identifying ground faults in active distribution networks mainly employ methods based on steady-state signal analysis. These methods typically collect and analyze the steady-state characteristics of electrical quantities such as current and voltage at various nodes in the distribution network, including amplitude and phase, and utilize traditional relay protection principles, such as zero-sequence current, zero-sequence voltage, and their directionality, to determine whether a ground fault has occurred and its location. Based on steady-state signals, some existing methods also employ threshold-based judgment, phasor comparison, and specific algorithm models to identify and differentiate ground faults. These methods have been widely used in previously unidirectional power supply and relatively simple passive distribution networks, achieving certain identification results.

[0004] However, under new operating conditions such as multi-point power supply and reverse power flow, the steady-state characteristics of electrical quantities in the distribution network are easily affected by distributed generation, making it difficult for traditional methods to accurately distinguish between normal and fault conditions, resulting in misjudgments or omissions. Especially in the context of dynamic response of distributed generation and frequent changes in system operating status, the characteristic information in the steady-state signal may be masked or distorted, thus affecting the accuracy and timeliness of fault identification. Summary of the Invention

[0005] This application provides a method and system for identifying ground faults in active distribution networks based on transient relay protection, which addresses the problem of low accuracy in existing ground fault identification methods based on steady-state signal analysis in active distribution networks.

[0006] In a first aspect, this application provides a method for identifying ground faults in active distribution networks based on transient relay protection, applied to an active distribution network ground fault identification system, the method comprising: The transient electrical quantities monitored in real time in the active power distribution network are converted into frequency domain to generate transient frequency domain characteristics; The transient frequency domain features are matched with a preset artificial intelligence feature library to obtain one or more candidate fault modes; For each of the candidate fault modes, transient simulation calculations are performed based on the topology model of the active distribution network to obtain the theoretical transient characteristics corresponding to each candidate fault mode. The candidate fault mode corresponding to the theoretical transient feature with the highest similarity to the transient frequency domain feature is determined as the fault identification result; Based on the fault identification results, a protection action command is generated to perform fault isolation.

[0007] Through the above embodiments, the system performs frequency domain conversion on transient electrical quantities monitored in real time in the active distribution network to generate transient frequency domain features, which are then matched with an artificial intelligence feature library to achieve accurate identification of fault modes. This method further utilizes the distribution network topology model for transient simulation, and through similarity analysis between theoretical and actual transient features, the final fault mode is determined, thereby generating protection action commands for fault isolation. This effectively overcomes the shortcomings of easily disturbed steady-state characteristics and high misjudgment rates under complex operating conditions such as multi-point power supply and bidirectional power flow, improving the accuracy and real-time performance of ground fault identification.

[0008] In some embodiments, the step of performing frequency domain conversion on transient electrical quantities monitored in real time in the active power distribution network to generate transient frequency domain features specifically includes: The transient electrical quantities at multiple measuring points are collected, and the transient electrical quantities are voltage and current signals at different locations in the distribution network; The transient electrical quantity is decomposed into feature components at different time and frequency resolutions to obtain multi-scale transient features; The multi-scale transient features are fused to generate fused transient frequency domain features.

[0009] Through the above embodiments, the system collects transient electrical quantities from multiple measuring points and decomposes them into feature components at different time and frequency resolutions to obtain multi-scale transient features. Finally, feature fusion is performed to generate fused transient frequency domain features. This technical solution can fully exploit the transient information of the distribution network at different spatial locations and frequency scales, improving the comprehensiveness and representativeness of feature extraction. Through multi-point collaboration and multi-scale analysis, it can effectively capture complex electrical signal changes generated when a fault occurs, greatly enhancing the system's response capability and fault tolerance to abnormal situations, improving the accuracy of fault identification and anti-interference capability, and adapting to the increasingly complex operating environment of active distribution networks.

[0010] In some embodiments, the step of performing transient simulation calculations based on the topology model of the active distribution network for each candidate fault mode to obtain the theoretical transient characteristics corresponding to each candidate fault mode specifically includes: Obtain the current operating data of the distribution network, which is data describing the real-time switch status and node connection relationships of the distribution network; Based on the current operating data, multiple candidate topology models are constructed, and the candidate topology models are structural models that reflect different network connection methods. For each of the candidate fault modes, transient simulations are performed based on each of the candidate topology models to obtain multiple transient simulation results; The similarity between the multiple transient simulation results and transient frequency domain features is calculated to obtain multiple similarity results; The transient simulation result with the highest similarity among the similarity results is determined as the theoretical transient feature.

[0011] Through the above embodiments, the system incorporates current operating data of the distribution network and constructs multiple candidate topology models. For each candidate fault mode, transient simulations are performed under each topology model, and the simulation results are compared with actual frequency domain characteristics for similarity calculation. This approach significantly improves the flexibility and adaptability of the fault identification process, enabling dynamic responses to changes in the switching states and node connection relationships of the distribution network, ensuring a high degree of matching between theoretical characteristics and actual operating conditions. Through multi-model simulation and similarity optimization, the system can effectively distinguish various complex fault modes, accurately pinpoint the actual fault type and location, and reduce the probability of misjudgment and missed judgment.

[0012] In some embodiments, before the step of performing frequency domain transformation on the transient electrical quantities monitored in real time of the active distribution network to generate transient frequency domain characteristics, the method further includes: Collect distributed transient electrical quantities from multiple smart terminals, wherein the distributed transient electrical quantities are voltage and current signals acquired in real time at different measurement points; The distributed transient electrical quantities are spatiotemporally synchronized and calibrated to obtain calibrated transient electrical quantities; The calibration transient electrical quantities are shared collaboratively among the various smart terminals to obtain the collaborative transient electrical quantities.

[0013] Through the above embodiments, the system adds distributed transient electrical quantity acquisition and spatiotemporal synchronization calibration before feature extraction, and supports collaborative sharing among various intelligent terminals. This distributed collaboration and synchronous calibration mechanism can ensure the temporal consistency and spatial correlation of data collected from different measurement points, effectively eliminating errors caused by sampling time differences or data drift. Through data collaboration among intelligent terminals, the information sharing and complementarity capabilities across the entire network are enhanced, greatly improving the global perception capability and response speed of fault detection.

[0014] In some embodiments, the step of determining the candidate fault mode corresponding to the theoretical transient feature with the highest similarity to the transient frequency domain feature as the fault identification result specifically includes: For each candidate fault mode, the similarity between the theoretical transient features and the transient frequency domain features is calculated in multiple preset evaluation dimensions. The similarity results are weighted and fused according to preset weights to obtain a comprehensive similarity score; The candidate fault mode with the highest comprehensive similarity score is determined as the final fault identification result.

[0015] Through the above embodiments, the system calculates the similarity between the theoretical transient features and the actual transient frequency domain features of candidate fault modes across multiple preset evaluation dimensions, and obtains a comprehensive similarity score through weighted fusion, ultimately determining the optimal fault identification result. This multi-dimensional, weighted fusion similarity evaluation mechanism can fully utilize the complementary advantages of different evaluation indicators to improve the comprehensive judgment ability and discrimination accuracy of fault identification.

[0016] In some embodiments, after the step of matching the transient frequency domain features with a preset artificial intelligence feature library to obtain one or more candidate fault modes, the method further includes: Record the matching results between the transient frequency domain features and the one or more candidate fault modes, wherein the matching results include the candidate fault modes and their corresponding similarity scores; If the highest similarity score in the matching results is lower than a preset threshold, the transient frequency domain feature and the fault handling result corresponding to the transient frequency domain feature will be added as a new sample to the artificial intelligence feature library.

[0017] Through the above embodiments, the system adds records of the matching results between transient frequency domain features and candidate fault modes, and automatically feeds back new data and processing results to the artificial intelligence feature library when the similarity is below a threshold. This adaptive learning and dynamic feature library update mechanism enables the fault identification system to continuously improve and evolve. With the continuous accumulation and feedback of actual operating data, the feature library can be continuously enriched and optimized, improving the system's ability to identify new or rare fault types and its robustness.

[0018] In some embodiments, after the step of generating a protection action command for performing fault isolation based on the fault identification result, the method further includes: Send protection action commands to adjacent protection devices, which are other protection devices electrically connected to the fault area; Receive response information from adjacent protection devices, wherein the response information is the status feedback of the adjacent protection devices to the protection action command; Based on the protection action command and the response information, a coordination optimization command is generated, which is used to optimize the fault isolation strategy for multi-point linkage. The coordination and optimization instructions are executed to complete fault isolation across the entire network.

[0019] Through the above embodiments, after generating protection action commands, the system further realizes linkage and response with adjacent protection devices. By coordinating and optimizing commands, it achieves multi-point linkage fault isolation strategy optimization and ultimately completes fault isolation across the entire network. This coordinated linkage mechanism enables information exchange and strategy collaboration among distributed protection devices, breaking the limitations of traditional independent protection devices and achieving unified linkage and optimal response across the entire network. It effectively avoids protection mismatch and malfunctions caused by single-point responses, improving the efficiency and reliability of fault isolation.

[0020] Secondly, this application provides an active power distribution network grounding fault identification system, the active power distribution network grounding fault identification system comprising: one or more processors and a memory; The memory is coupled to the one or more processors. The memory is used to store computer program code, which includes computer instructions. The one or more processors call the computer instructions so that the active distribution network grounding fault identification system can implement the active distribution network grounding fault identification method based on transient relay protection provided in the above embodiments, which will not be described in detail here.

[0021] Thirdly, this application provides a computer-readable storage medium including instructions that, when executed on an active distribution network grounding fault identification system, enable the active distribution network grounding fault identification system to implement the active distribution network grounding fault identification method based on transient relay protection provided in the above embodiments, which will not be elaborated here.

[0022] Fourthly, this application provides a computer program product that, when running on an active distribution network grounding fault identification system, enables the active distribution network grounding fault identification system to implement the active distribution network grounding fault identification method based on transient relay protection provided in the above embodiments, which will not be elaborated here.

[0023] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. By performing frequency domain conversion on transient electrical quantities in active distribution networks and matching them with an artificial intelligence feature library, intelligent identification of grounding fault modes in distribution networks from multiple points, scales, and dimensions is achieved. This mechanism not only combines rich information from transient signals but also introduces the self-learning and dynamic updating capabilities of the artificial intelligence feature library, enabling the identification system to continuously evolve and optimize, improving the accuracy and adaptability of fault identification, and is particularly suitable for the complex and ever-changing new operating environment of active distribution networks.

[0024] 2. By utilizing real-time acquired transient electrical quantities from multiple measurement points and combining them with the current actual topology of the distribution network, multiple candidate models are dynamically constructed for transient simulation and feature comparison. Through spatiotemporal collaboration and data fusion of multiple models and terminals, the system's ability to capture fault characteristics under complex operating conditions is greatly enhanced, ensuring the consistency and integrity of the data source.

[0025] 3. A multi-point coordinated linkage and adaptive optimization strategy for fault protection actions was proposed. By distributing protection action commands across the entire network and coordinating responses among intelligent terminals, combined with comprehensive similarity assessment and real-time feedback mechanisms, rapid fault isolation and optimal control were achieved. This multi-level, network-wide linkage protection system overcomes the limitations of traditional single-point protection and effectively improves the safety, reliability, and self-healing capabilities of the distribution network. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating an active distribution network grounding fault identification method based on transient relay protection in an embodiment of this application. Figure 2 This is another flowchart illustrating an active distribution network grounding fault identification method based on transient relay protection in an embodiment of this application; Figure 3 This is a schematic diagram of the physical device structure of an active power distribution network grounding fault identification system in the embodiments of this application. Detailed Implementation

[0027] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0028] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0029] For ease of understanding, the method provided in this implementation is described in process below. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating an active distribution network grounding fault identification method based on transient relay protection in an embodiment of this application.

[0030] S101. Perform frequency domain conversion on the transient electrical quantities monitored in real time in the active power distribution network to generate transient frequency domain characteristics.

[0031] Among them, active distribution network refers to distribution network that is connected to distributed power sources (such as new energy power generation equipment) and has bidirectional power flow and multi-point power supply characteristics; transient electrical quantity refers to electrical signals such as voltage and current monitored in real time at the moment of fault occurrence or during transient process in distribution network, which has the characteristics of short time and rapid change; frequency domain conversion refers to the processing of converting electrical quantity signals in the time domain (time dimension) to the frequency domain (frequency dimension), and commonly used methods include Fourier transform, wavelet transform, etc.; transient frequency domain characteristics refer to the parameters that can reflect the fault characteristics extracted after frequency domain conversion, such as the amplitude and phase of specific frequency components.

[0032] Specifically, the system uses intelligent monitoring terminals deployed at various nodes of the distribution network to collect voltage and current signals (transient electrical quantities) during fault transient processes in real time. Since transient signals may not have obvious characteristics in the time domain and are susceptible to noise interference, frequency domain conversion can decompose the signal into different frequency components, highlighting the fault-specific frequency characteristics (such as ground faults may be accompanied by specific high-frequency components). Subsequently, key feature parameters are extracted from the converted frequency domain signal to form transient frequency domain features, providing standardized input for fault mode matching.

[0033] Optionally, 1. Collect three-phase voltage and current time-domain signals from multiple measurement points, with the sampling frequency set to 10kHz~50kHz to capture transient details; 2. Perform frequency domain conversion on the time-domain signal of each measurement point using wavelet transform, and decompose it to obtain wavelet coefficients at 10 different scales (frequency ranges); 3. Extract features such as maximum value, energy value, and abrupt change time from the coefficients at each scale, and generate transient frequency domain features by splicing and fusing them.

[0034] Optionally, Fourier transform is performed on the real-time transient electrical quantity to obtain the frequency-amplitude spectrum of the signal; frequency components of the fault-sensitive frequency band (such as 1kHz~20kHz) are screened out; the total energy, peak frequency and amplitude ratio in the frequency band are calculated and combined to form transient frequency domain characteristics.

[0035] It is understandable that other frequency domain conversion methods, such as short-time Fourier transform, can also be used to achieve this, and no limitation is made here.

[0036] S102. Match the transient frequency domain features with the preset artificial intelligence feature library to obtain one or more candidate fault modes.

[0037] The pre-set artificial intelligence feature library refers to a large amount of historical fault case data stored in advance, including transient frequency domain feature samples and labels corresponding to different grounding fault modes (such as single-phase grounding and two-phase grounding); the candidate fault mode refers to the fault type that is initially screened through feature matching and is similar to the current transient frequency domain features, providing fault options to be verified for subsequent accurate identification.

[0038] Specifically, the system inputs the transient frequency domain features generated by S101 into the matching module and compares them with historical fault features stored in the artificial intelligence feature library. The feature library is formed through training machine learning models (such as support vector machines and neural networks) and contains the mapping relationship between features and fault modes. During the matching process, the similarity between the current feature and each sample in the library (such as cosine similarity and Euclidean distance) is calculated, and fault modes with similarity higher than a certain threshold are selected as candidate fault modes (usually 1 to 3), reducing the workload of subsequent simulation calculations.

[0039] Optionally, the transient frequency domain features are standardized (e.g., normalized to the [0,1] range) to match the scale of the feature library samples; the pre-trained K-nearest neighbor (KNN) model in the feature library is called to calculate the distance between the current feature and the samples in the library; the fault modes corresponding to the three samples with the smallest distance are selected as candidate fault modes.

[0040] Optionally, using the decision tree model in the feature library, the transient frequency domain features are input sequentially according to dimensions such as frequency component amplitude and energy proportion; the decision tree branches are used to judge (such as whether the high frequency component amplitude exceeds the threshold) to filter out the fault modes that meet the conditions; the filtering results are sorted according to the matching confidence and the top two are retained as candidate fault modes.

[0041] Understandably, deep learning models (such as CNNs) can also be used for feature matching, and this is not a limitation here.

[0042] Optionally, after matching transient frequency domain features with the artificial intelligence feature library, the system can also store the complete matching results through a log recording module, including the names of all candidate fault modes, their corresponding similarity scores, and matching timestamps, providing data support for subsequent tracing and analysis. Subsequently, the system automatically extracts the highest similarity score from the matching results and compares it with a preset threshold (e.g., 0.7). If the highest similarity score is lower than the threshold, it indicates that the existing feature library lacks historical samples that highly match the current fault features, potentially indicating new fault modes or insufficient coverage of existing samples. In this case, the system integrates the current transient frequency domain features (containing complete frequency domain parameters) with the corresponding final fault handling results (e.g., manually verified fault types, actual isolation strategies, and fault verification information) to form standardized new sample data. Finally, the new samples are written into the artificial intelligence feature library through the feature library update interface, enabling dynamic expansion of the feature library and providing richer reference samples for the identification of similar faults, gradually improving the system's adaptability to complex operating conditions.

[0043] S103. For each candidate fault mode, perform transient simulation calculations based on the topology model of the active distribution network to obtain the theoretical transient characteristics corresponding to each candidate fault mode.

[0044] Among them, transient simulation calculation refers to the calculation of the transient process of a fault based on a topological model and candidate fault modes, using simulation software; theoretical transient characteristics refer to the frequency domain characteristics corresponding to the candidate fault modes obtained by simulation calculation, which are used to compare with the actual transient frequency domain characteristics.

[0045] Specifically, the system first acquires the current operating data of the distribution network (such as switch opening and closing status, distributed generation output), and updates the topology model to reflect the real-time network structure. For each candidate fault mode (such as phase A ground fault), the corresponding fault point and fault type are set in the topology model. The transient process after the fault occurs is simulated using transient simulation software (such as PSCAD / EMTDC), and the simulated voltage and current signals are output. The simulated signals are then subjected to the same frequency domain transformation and feature extraction as in S101 to obtain the theoretical transient characteristics corresponding to the candidate fault mode, providing a basis for subsequent similarity comparison.

[0046] Optionally, based on the real-time switch status and node connection relationship of the distribution network, the line parameters and node topology of the topology model are updated; for each candidate fault mode, the fault location (such as the line midpoint) and fault type (such as high-resistance grounding) are set in the topology model to trigger transient simulation; the voltage and current time-domain signals within 0.1 seconds of the simulation output are extracted by wavelet transform to obtain the theoretical transient features.

[0047] Optionally, a topology simulation platform containing a distributed power source dynamic response model is constructed; for candidate fault modes, different fault initial phases (0°, 90°, 180°) are set to perform multi-scenario simulations and obtain multiple sets of transient signals; Fourier transform is performed on each set of signals, the average amplitude and phase difference of the characteristic frequency band are calculated, and the signals are fused to form theoretical transient characteristics.

[0048] It is understandable that simulation parameters can be optimized by combining real-time power flow data, but this is not a limitation here.

[0049] S104. The candidate fault mode corresponding to the theoretical transient feature with the highest similarity to the transient frequency domain feature is determined as the fault identification result.

[0050] In this context, similarity refers to the degree of matching between transient frequency domain features (actual features) and theoretical transient features (simulation features), which is measured by quantitative indicators (such as correlation coefficient and distance metric); fault identification results refer to the specific type (such as single-phase metallic grounding and two-phase grounding) and location information of the final determined distribution network grounding fault.

[0051] Specifically, the system calculates the similarity between the theoretical transient characteristics of each candidate fault mode and the actual transient frequency domain characteristics generated by S101. The calculation dimensions include amplitude error, phase consistency, and energy distribution similarity in the characteristic frequency bands. A comprehensive similarity score is obtained by weighted fusion of the results from each dimension. Finally, the candidate fault mode with the highest comprehensive similarity is selected, and its corresponding fault type, location, and other information are determined as the fault identification result to ensure the accuracy of the identification.

[0052] Optionally, for each candidate mode, the amplitude similarity between theoretical features and actual features in 5 preset frequency bands (such as 1kHz, 5kHz, 10kHz, etc.) is calculated, and absolute error ratio quantization is used; the cosine similarity of feature vectors is calculated to measure the overall distribution consistency; the amplitude similarity (weight 0.6) and cosine similarity (weight 0.4) are weighted and summed, and the candidate mode with the highest score is taken as the result.

[0053] Optionally, a multi-dimensional evaluation matrix is ​​constructed, including three dimensions: frequency peak deviation, energy proportion deviation, and phase difference; normalized similarity (range 0~1) is calculated for each dimension; the weight of each dimension is determined by the analytic hierarchy process (e.g., the weight of energy proportion deviation is 0.5), and the comprehensive score is calculated. The pattern corresponding to the highest score is the recognition result.

[0054] It is understandable that a dynamic weight adjustment mechanism (such as adapting weights according to fault type) can also be introduced, but this is not limited here.

[0055] S105. Generate protection action instructions for performing fault isolation based on the fault identification results.

[0056] Specifically, based on the fault location (e.g., a certain line segment), type (e.g., permanent grounding), and current grid topology in the fault identification results, the system queries the preset protection strategy library (e.g., the tripping of the nearest circuit breaker upstream of the fault point, and the downstream load transfer path). Combined with the real-time grid load status (e.g., avoiding power outages in non-faulty areas), it generates precise protection action commands, specifying the equipment that needs to be activated (e.g., the circuit breaker numbered CB-12), the activation time (e.g., tripping within 0.05 seconds), and subsequent coordination requirements (e.g., linkage with adjacent equipment), ensuring that the faulty area is quickly and accurately isolated.

[0057] Optionally, the nearest Level 2 protection device (such as a section circuit breaker and a bus circuit breaker) can be located based on the fault location; a preset strategy can be matched based on the fault type (permanent grounding) to determine the section circuit breaker to be tripped first; the instruction content includes the device ID, action type (tripping), delay time (0.03 seconds) and feedback requirements, and is sent to the corresponding protection device.

[0058] Optionally, by combining real-time power flow data, the impact of fault isolation on surrounding loads can be assessed, and the action sequence can be optimized (e.g., load transfer before tripping); a main instruction (target equipment tripping) and an auxiliary instruction (adjacent equipment automatic transfer start-up) can be generated; verification parameters (such as the current voltage threshold) can be added to the instruction to ensure that the equipment action conditions are met before execution.

[0059] It is understandable that global optimization instructions can also be generated in collaboration with the scheduling system, but this is not limited here.

[0060] In the above embodiments, the system performs frequency domain conversion on transient electrical quantities monitored in real time in the active distribution network to generate transient frequency domain features, which are then matched with an artificial intelligence feature library to achieve accurate identification of fault modes. This method further utilizes the distribution network topology model for transient simulation, and through similarity analysis between theoretical and actual transient features, the final fault mode is determined, thereby generating protection action commands for fault isolation. This effectively overcomes the shortcomings of easily disturbed steady-state characteristics and high misjudgment rates under complex operating conditions such as multi-point power supply and bidirectional power flow, improving the accuracy and real-time performance of ground fault identification.

[0061] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 2 This is another flowchart illustrating an active distribution network grounding fault identification method based on transient relay protection in an embodiment of this application.

[0062] S201. Collect distributed transient electrical quantities from multiple smart terminals for spatiotemporal synchronization calibration to obtain calibrated transient electrical quantities.

[0063] Among them, intelligent terminals refer to monitoring devices with data acquisition, processing and communication functions deployed at various nodes of the active power distribution network (such as line segmentation, transformer outlet, distributed power source access point); distributed transient electrical quantities refer to voltage and current signals in the fault transient process collected in real time by multiple intelligent terminals at different spatial measurement points, which have the characteristics of dispersion and real-time; spatiotemporal synchronization calibration refers to the process of unifying the time base (eliminating sampling time difference) and verifying spatial correlation (matching the correspondence of measurement point positions) of transient electrical quantities collected by different intelligent terminals; calibrated transient electrical quantities refer to transient electrical quantity data with consistent timing and accurate spatial correlation after spatiotemporal synchronization calibration.

[0064] Specifically, the system uses a communication network to access smart terminals distributed across different locations in the distribution network to collect real-time voltage and current signals from each measuring point during the transient phase of a fault (within 0.1 to 0.2 seconds after the fault occurs). Because the hardware clocks of each smart terminal may deviate and their deployment locations are dispersed, the directly collected data may suffer from time asynchrony (differences in timestamps recorded by different terminals for signals at the same time) and spatial correspondence issues (such as mismatch between measuring point location and signal attribution). Through spatiotemporal synchronization calibration, the system corrects the sampling timestamps of each terminal based on a unified time reference (such as a GPS clock or Network Time Protocol NTP), and verifies the correspondence between signals and measuring point locations according to the distribution network topology, ultimately outputting a time-consistent and spatially accurate calibrated transient electrical quantity.

[0065] Optionally, the system establishes a digital simulation model of the active distribution network containing accurate topology and component parameters, creating an operational characteristic model for distributed generation (DG) within the network that reflects its grid connection location, capacity, and dynamic response under fault conditions (such as inverter control strategies). Subsequently, an optimization task is constructed: using a set of candidate measurement points encompassing all potential installation locations (such as substation outlets, line segmentation points, DG grid connection points, etc.) as the search space, the optimization objective is to "maximize the distinguishability of transient features under any two different fault scenarios," with the economic constraint of "minimizing the total number of measurement points." To achieve this, intelligent optimization algorithms such as genetic algorithms can be employed. This involves iteratively evaluating the merits of different measurement point combinations (i.e., layout schemes)—the evaluation process relies on transient simulations of numerous preset fault scenarios and calculating the characteristic differences of signals collected under the current measurement point combination—ultimately converging to obtain an optimal set of "key measurement point" layout schemes. This scheme minimizes monitoring costs while meeting fault identification accuracy requirements.

[0066] S202. The calibration transient electrical quantities are shared collaboratively among the various smart terminals to obtain the collaborative transient electrical quantities.

[0067] Among them, collaborative sharing refers to the process by which each intelligent terminal transmits and exchanges the calibration transient electrical quantities it collects through communication networks (such as wireless private networks and fiber optic communication); the collaborative transient electrical quantities refer to the integrated data of calibration transient electrical quantities from multiple measurement points across the entire network obtained by each intelligent terminal after sharing, which has comprehensiveness and complementarity.

[0068] Specifically, the system initiates a collaborative communication mechanism between intelligent terminals. Each terminal uploads its spatiotemporally calibrated transient electrical quantities to neighboring terminals or regional aggregation nodes via preset communication protocols (such as MQTT and DL / T645). During the sharing process, terminals prioritize sharing relevant data (such as measurement point data from adjacent sections) based on the distribution network topology, while also deduplicating duplicate data. Through collaborative sharing, the local data originally scattered across various terminals is integrated into a global dataset covering multiple key measurement points. Each terminal not only possesses its own collected data but also acquires complementary data from other terminals (such as high-frequency component information not detected by itself). Ultimately, this results in a comprehensive and low-redundancy collaborative transient electrical quantity dataset, providing a rich data source for multi-scale feature extraction.

[0069] S203. Collect transient electrical quantities at multiple measurement points and decompose them into characteristic components at different time and frequency resolutions to obtain multi-scale transient characteristics.

[0070] Among them, transient electrical quantities refer to the raw, unprocessed voltage and current time-domain waveform data collected by intelligent terminals within a very short time (usually milliseconds) after a fault occurs; time and frequency resolution refer to two dimensions used in signal analysis to describe the ability to distinguish signal details. High time resolution can capture instantaneous changes, and high frequency resolution can distinguish similar frequency components; feature components refer to the sub-signal sequences belonging to different frequency bands or time scales obtained after the original transient electrical quantities are decomposed; multi-scale transient features refer to the set of statistical values ​​extracted from each feature component that can quantify fault information, and are structured feature data.

[0071] This step is executed after the system has completed the acquisition and spatiotemporal synchronization calibration of raw data from multiple smart terminals across the network (S201) and achieved collaborative data sharing among the terminals (S202). Its core task is to transform the high-quality raw time-domain waveform data obtained after collaborative sharing into a structured feature set containing rich fault information that can be understood and processed by subsequent algorithms.

[0072] Specifically, the system first extracts the voltage and current time-domain signal sequences of all key measurement points from the coordinated transient electrical quantity data pool. Since direct analysis of the original time-domain waveforms is difficult and has low information utilization, this step employs signal decomposition techniques to transform them into a multi-scale domain for analysis. Taking Discrete Wavelet Transform (DWT) as an example, the signal can be decomposed into a series of wavelet coefficients with different frequencies and time resolutions, suitable for analyzing the local characteristics of transient signals. The system performs multi-level decomposition on each phase voltage and current signal at each measurement point using specific wavelet basis functions (such as the Daubechies4 wavelet), for example, a 5-level decomposition, yielding a set of detail components (D1, D2, D3, D4, D5) and a set of approximation components (A5). The D1 component corresponds to the highest frequency part of the signal and has the highest time resolution; as the number of levels increases, the frequencies corresponding to the D2 to D5 components decrease sequentially, and the time resolution also decreases accordingly; the A5 component represents the low-frequency profile of the signal. These components together constitute the observation of the original signal at different "scales". Subsequently, to transform these coefficients into features with clear physical meaning, the system calculates multiple statistical characteristic values ​​such as energy, energy entropy, standard deviation, and kurtosis for each component (D1 to A5). Finally, all statistical characteristic values ​​of all measurement points, all signals (such as phase A voltage, phase B current, etc.), and all decomposition scales are arranged and combined in a preset fixed order to form a high-dimensional feature vector set, which is the "multi-scale transient feature".

[0073] S204. Perform feature fusion on the multi-scale transient features to generate fused transient frequency domain features.

[0074] Specifically, the system first receives multiple feature vector sets output by S203. Considering the potential information overlap (redundancy) in the feature vectors of different measurement points, the system can selectively perform dimensionality reduction on the feature vectors of each measurement point, for example, by using Principal Component Analysis (PCA) to retain principal components with a contribution rate exceeding 95%. Next, the system concatenates the feature vectors (after dimensionality reduction or in their original form) of each measurement point in a predefined, fixed order (e.g., arranged from the beginning to the end according to the physical location of the measurement points on the main line), forming a single long vector with higher dimensionality and richer information. Finally, to eliminate the potential adverse effects of significant differences in dimensions and numerical ranges between different features (such as energy and kurtosis) on the training of machine learning models, the system standardizes this fused long vector (e.g., Z-score standardization). The resulting "fused transient frequency domain feature" is a standardized vector with a mean of 0 and a standard deviation of 1 for each dimension. It retains key information from multiple scales and measurement points while maintaining a unified format and scale.

[0075] Optionally, before splicing, the system can employ a feature selection algorithm based on mutual information to evaluate the correlation between each feature in the feature vector of each measurement point and the historical fault label, eliminating features with correlation below a preset threshold to achieve preliminary dimensionality reduction and noise reduction. Then, during feature splicing, the system assigns a weight to the feature vector of each measurement point based on the proportion of the total transient signal energy collected by each measurement point to the total energy of the entire network. Measurement points with a higher energy proportion (usually closer to the fault point) have a larger weight, and their feature vectors are multiplied by the corresponding weight before splicing to enhance their influence in the fused features. Finally, the fused vector obtained after weighted splicing is processed using the min-max scaling method to linearly scale the values ​​of each feature dimension to an interval, thus obtaining the fused transient frequency domain features.

[0076] It is understandable that other methods can be used to achieve feature fusion and processing, such as learning a deep fusion representation of features through an autoencoder, or using methods such as canonical correlation analysis (CCA) for fusion, which are not limited here.

[0077] S205. Construct multiple candidate topology models based on the current running data.

[0078] Specifically, the system acquires current operating data of the distribution network through real-time monitoring systems (such as SCADA systems and distribution automation systems), focusing on extracting the switch status matrix (recording the open / closed state of each switch) and the node connection list (describing the electrical connections between nodes). Because the distribution network may experience false alarms in switch status and temporary topology adjustments (such as line reconnections during maintenance) in actual operation, a single topology model may not accurately reflect the actual network structure. Therefore, the system constructs a baseline topology model based on current operating data and generates multiple variant models for possible topology fluctuation scenarios (such as slight deviations in the states of critical switches or temporary disconnection of distributed power sources), forming a candidate topology model set. Each candidate topology model includes detailed parameters such as line parameters, node impedance, and power source connection locations.

[0079] S206. For each candidate fault mode, perform transient simulation based on each candidate topology model to obtain multiple transient simulation results.

[0080] Specifically, the system iterates through each candidate fault mode obtained in step S202, and combines them with each candidate topology model constructed in step S205 to form a simulation scenario pair of "fault mode-topology model". For each scenario pair, the system sets the corresponding fault parameters (such as fault type, fault location, fault resistance, fault initial phase, etc.) in the simulation platform to simulate the transient process within 0.1 to 0.2 seconds after the fault occurs. During the simulation, the system focuses on recording the voltage and current time-domain waveforms at each key measurement point, and performs preprocessing on the waveforms consistent with actual data processing (such as filtering and sampling frequency matching), finally outputting the transient simulation results corresponding to each scenario pair.

[0081] Optionally, based on the PSCAD / EMTDC simulation platform, import the line parameters, node impedances, and power supply models of each candidate topology model; for each candidate fault mode, set the fault point (e.g., line midpoint), fault type (e.g., A-phase grounding), and fault resistance (e.g., 10Ω) in the model, and trigger the simulation; the simulation outputs the three-phase voltage and current time-domain data within 0.1 seconds after the fault, with the sampling frequency consistent with the actual acquisition (e.g., 20kHz), to obtain the transient simulation results.

[0082] Optionally, a simulation model containing the dynamic response of distributed power sources can be built using MATLAB / Simulink, and the structural data of the candidate topology model can be loaded. For each candidate fault mode, different initial fault phases (0°, 90°, 180°) can be set to perform multi-condition simulation. The voltage and current components of the fault-sensitive frequency band (1~20kHz) in the simulation data can be extracted to generate transient simulation results in the frequency domain.

[0083] It is understandable that this can also be achieved using real-time simulation tools such as RTDS, and this is not a limitation here.

[0084] S207. Calculate the similarity between multiple transient simulation results and transient frequency domain features to obtain multiple similarity results, and determine the transient simulation result with the highest similarity as the theoretical transient feature.

[0085] Specifically, the system converts all transient simulation results generated in step S206 into frequency domain features consistent with the output format of step S204 (e.g., using the same wavelet transform or Fourier transform method). Then, it calculates the similarity between each simulated frequency domain feature and the actual transient frequency domain feature. The similarity calculation dimensions include the overall distribution consistency of the feature vectors (e.g., cosine similarity), the amplitude deviation of key frequency components (e.g., absolute error ratio), and the energy distribution similarity (e.g., KL divergence). The system summarizes all similarity scores between "simulation results and actual features," and for each candidate fault mode, it selects the simulated frequency domain feature with the highest similarity from its corresponding multiple simulation results, determining it as the theoretical transient feature of that candidate fault mode. This process eliminates the influence of topological model uncertainties on theoretical features, ensuring a high degree of matching between theoretical features and actual operating conditions.

[0086] S208. For each candidate fault mode, calculate the similarity between the theoretical transient features and the transient frequency domain features in multiple preset evaluation dimensions.

[0087] Specifically, the system calculates the similarity score for each candidate fault mode based on its theoretical transient characteristics, according to a pre-defined evaluation dimension system. These pre-defined evaluation dimensions may include: similarity of key frequency component amplitudes (e.g., the degree of amplitude matching between peak frequencies within the 1-20kHz range); phase consistency (e.g., the absolute value of the phase difference between key frequency components); energy distribution similarity (e.g., the degree of matching of energy proportions in each frequency band); and similarity of feature vector sparsity (e.g., the degree of position matching of non-zero feature components). For each dimension, the system uses a corresponding quantization method to calculate the similarity score (e.g., amplitude similarity = 1 - |theoretical amplitude - actual amplitude| / max(theoretical amplitude, actual amplitude)), and records the score for each candidate fault mode in each dimension.

[0088] S209. The candidate fault mode with the highest corresponding comprehensive similarity score is determined as the final fault identification result.

[0089] Specifically, the system performs weighted fusion of the similarity scores obtained from S208 across various dimensions based on a preset weighting system: different weights are assigned to different evaluation dimensions (e.g., energy distribution similarity weight 0.3, peak amplitude similarity weight 0.3, phase consistency weight 0.2, feature coverage weight 0.2), and the comprehensive similarity score for each candidate fault mode is calculated by weighted summation (comprehensive score = Σ(dimensional similarity × dimensional weight)). The system sorts the comprehensive scores of all candidate fault modes, selects the candidate fault mode with the highest score as the final fault identification result, and simultaneously outputs the fault type, theoretical matching degree, and associated key measurement point information corresponding to the mode. If the difference between the highest and second-highest scores is small (e.g., difference <5%), the system can trigger secondary verification (e.g., supplementary simulation of specific working conditions) to ensure decision accuracy.

[0090] S210, Send protection action commands to adjacent protection devices and receive response information from adjacent protection devices.

[0091] Among them, adjacent protection equipment refers to protection devices that are directly electrically connected to the fault area (such as sectional circuit breakers upstream / downstream of the fault point, protection relays of adjacent lines, distributed power supply outlet switches, etc.); response information refers to the status feedback returned by adjacent protection equipment after receiving the protection action command, including command reception confirmation, current equipment status (such as whether it is ready), action preparation status (such as whether the action conditions are met) and abnormal prompts (such as equipment failure and inability to execute).

[0092] Specifically, based on the fault location and distribution network topology in the fault identification results, the system locates adjacent protection devices directly related to the fault area (usually level 2-3 related devices, such as upstream and downstream circuit breakers and tie switches of the faulty line, and outlet protection devices of nearby power sources). Protection action commands are sent to these devices via a reliable communication link (such as a fiber optic private network or industrial Ethernet) according to a preset protocol (such as the IEC61850MMS protocol). The commands include a unique identifier and timestamp to ensure accuracy. After transmission, the system initiates a timeout waiting mechanism (usually set to 50-100 milliseconds) to receive response information from each adjacent device in real time: if the device successfully receives the command and its status is normal (e.g., no blocking signal), it returns "ready" feedback; if the device has an anomaly (e.g., mechanical failure, power failure), it returns "abnormal" and the specific reason. The system summarizes all response information to determine the execution capability of adjacent devices, providing a basis for subsequent coordination and optimization.

[0093] S211. Generate coordinated optimization instructions based on protection action instructions and response information and execute them to complete fault isolation across the entire network.

[0094] Among them, the coordination and optimization instruction refers to the adjustment instruction generated by the system based on the protection action instruction and the response information of adjacent equipment, which is used to optimize the multi-point linkage fault isolation strategy, including the correction of action timing, activation of backup equipment, and expansion of action range; the fault isolation of the entire network refers to the ultimate goal of completely disconnecting the faulty area from the entire distribution network by coordinating the actions of multiple protection devices, so as to ensure the normal power supply of non-faulty areas.

[0095] Specifically, the system analyzes the response information received by S210: if all adjacent protection devices return "ready," the original protection action command can be directly executed; if there is a device "abnormal" or "communication interrupted," its impact on fault isolation is assessed (e.g., the inability of critical equipment to operate may lead to insufficient isolation range). Based on the analysis results, the system generates coordinated optimization commands: for executable devices, the action sequence is corrected (e.g., actions are performed sequentially from closest to furthest from the fault point to avoid arc reignition); for non-executable devices, a backup protection scheme is activated (e.g., switching to the action of an adjacent tie switch); if a potential protection dead zone is found, the action range is expanded (e.g., linkage with a higher-level circuit breaker). After the optimization command is generated, the system sends it to the relevant devices through the original communication link, and monitors the action execution process in real time (e.g., confirming changes in device status through remote signaling signals) until the fault area is electrically disconnected from the entire network and the non-fault area is restored to normal power supply, completing the fault isolation of the entire network.

[0096] The active power distribution network grounding fault identification system of this invention is applied to electronic equipment. Figure 3 A schematic diagram of the architecture of an electronic device suitable for implementing embodiments of the present invention is shown.

[0097] It should be noted that, Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0098] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by instructions (computer programs), or by instructions (computer programs) controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor. The electronic device of this embodiment includes a storage medium and a processor, wherein the storage medium stores multiple instructions that can be loaded by the processor to execute any step of the method provided in the embodiments of the present invention.

[0099] Specifically, the storage medium and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more signal lines. The storage medium stores computer-executable instructions that implement data access control methods, including at least one software functional module that can be stored in the storage medium in the form of software or firmware. The processor executes various functional applications and data processing by running the software program and module stored in the storage medium. The storage medium can be, but is not limited to, Random Access Memory (RAM), Read-Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The storage medium stores the program, and the processor executes the program after receiving the execution instructions.

[0100] Furthermore, the software programs and modules within the aforementioned storage medium may also include an operating system, which may include various software components and / or drivers for managing system tasks (e.g., memory management, storage device control, power management, etc.) and can communicate with various hardware or software components to provide an operating environment for other software components. The processor may be an integrated circuit chip with signal processing capabilities. The aforementioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc., which can implement or execute the methods, steps, and logic flowcharts disclosed in this embodiment. The general-purpose processor may be a microprocessor or any conventional processor.

[0101] Since the instructions stored in the storage medium can execute the steps in any of the methods provided in the embodiments of the present invention, the beneficial effects of any of the methods provided in the embodiments of the present invention can be achieved, as detailed in the preceding embodiments, and will not be repeated here.

[0102] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying grounding faults in active distribution networks based on transient relay protection, applied to an active distribution network grounding fault identification system, characterized in that, The method includes: The transient electrical quantities monitored in real time in the active power distribution network are converted into frequency domain to generate transient frequency domain characteristics; The transient frequency domain features are matched with a preset artificial intelligence feature library to obtain one or more candidate fault modes; For each of the candidate fault modes, transient simulation calculations are performed based on the topology model of the active distribution network to obtain the theoretical transient characteristics corresponding to each candidate fault mode. The candidate fault mode corresponding to the theoretical transient feature with the highest similarity to the transient frequency domain feature is determined as the fault identification result; Based on the fault identification results, a protection action command is generated to perform fault isolation.

2. The method according to claim 1, characterized in that, The step of performing frequency domain conversion on the transient electrical quantities monitored in real time in the active power distribution network to generate transient frequency domain characteristics specifically includes: Transient electrical quantities are collected from multiple measuring points. These transient electrical quantities are voltage and current signals at different locations in the active power distribution network. The locations of the measuring points are selected based on a comprehensive screening of the topology of the active power distribution network, the distribution location of distributed power sources, and their operating characteristics. The transient electrical quantity is decomposed into feature components at different time and frequency resolutions to obtain multi-scale transient features; The multi-scale transient features are fused to generate fused transient frequency domain features.

3. The method according to claim 1, characterized in that, The step of performing transient simulation calculations based on the topology model of the active distribution network for each candidate fault mode to obtain the theoretical transient characteristics corresponding to each candidate fault mode specifically includes: Obtain the current operating data of the distribution network, which is data describing the real-time switch status and node connection relationships of the distribution network; Based on the current operating data, multiple candidate topology models are constructed, and the candidate topology models are structural models that reflect different network connection methods. For each of the candidate fault modes, transient simulations are performed based on each of the candidate topology models to obtain multiple transient simulation results; The similarity between the multiple transient simulation results and transient frequency domain features is calculated to obtain multiple similarity results; The transient simulation result with the highest similarity among the similarity results is determined as the theoretical transient feature.

4. The method according to claim 1, characterized in that, Before the step of performing frequency domain transformation on the transient electrical quantities monitored in real time of the active power distribution network to generate transient frequency domain characteristics, the method further includes: Collect distributed transient electrical quantities from multiple smart terminals, wherein the distributed transient electrical quantities are voltage and current signals acquired in real time at different measurement points; The distributed transient electrical quantities are spatiotemporally synchronized and calibrated to obtain calibrated transient electrical quantities; The calibration transient electrical quantities are shared collaboratively among the various smart terminals to obtain the collaborative transient electrical quantities.

5. The method according to claim 1, characterized in that, The step of determining the candidate fault mode corresponding to the theoretical transient feature with the highest similarity to the transient frequency domain feature as the fault identification result specifically includes: For each candidate fault mode, the similarity between the theoretical transient features and the transient frequency domain features is calculated in multiple preset evaluation dimensions. The similarity results are weighted and fused according to preset weights to obtain a comprehensive similarity score; The candidate fault mode with the highest comprehensive similarity score is determined as the final fault identification result.

6. The method according to claim 1, characterized in that, After the step of matching the transient frequency domain features with a preset artificial intelligence feature library to obtain one or more candidate fault modes, the method further includes: Record the matching results between the transient frequency domain features and the one or more candidate fault modes, wherein the matching results include the candidate fault modes and their corresponding similarity scores; If the highest similarity score in the matching results is lower than a preset threshold, the transient frequency domain feature and the fault handling result corresponding to the transient frequency domain feature will be added as a new sample to the artificial intelligence feature library.

7. The method according to claim 1, characterized in that, After the step of generating a protection action command for performing fault isolation based on the fault identification result, the method further includes: Send protection action commands to adjacent protection devices, which are other protection devices electrically connected to the fault area; Receive response information from adjacent protection devices, wherein the response information is the status feedback of the adjacent protection devices to the protection action command; Based on the protection action command and the response information, a coordination optimization command is generated, which is used to optimize the fault isolation strategy for multi-point linkage. The coordination and optimization instructions are executed to complete fault isolation across the entire network.

8. An active power distribution network grounding fault identification system, characterized in that, The active power distribution network grounding fault identification system includes: one or more processors and a memory; The memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code including computer instructions, the one or more processors calling the computer instructions to cause the active power distribution network ground fault identification system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the active distribution network grounding fault identification system, the active distribution network grounding fault identification system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the active power distribution network grounding fault identification system, the active power distribution network grounding fault identification system performs the method as described in any one of claims 1-7.

Images