PMU Current Image CNNs for Fault Location in Active Distribution Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fault diagnosis methods in power distribution networks struggle with accurately determining fault locations, particularly in large-scale networks with integrated distributed generation (DG) sources, and fail to adequately consider uncertainties associated with load demand and fault information, lacking real-time simulation modeling.
Innovation Solution
A fault management method using deep convolutional neural networks (CNNs) trained with three-phase current signal images from phasor measuring units (PMUs) to classify faults, identify fault sections, and locate fault locations, eliminating the need for feature extraction and accounting for DG uncertainties, load demand fluctuations, and fault information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional fault diagnosis methods (impedance-based, high-frequency, traveling-wave) are used in distribution networks with distributed generation, then implementation simplicity or measurement capability is improved, but measurement precision and reliability deteriorate due to network complexity, bidirectional power flows, and uncertainties
Solution Approach 1:
The patent replaces conventional fault diagnosis methods (impedance-based, high-frequency, traveling-wave) with a deep learning-based system using Convolutional Neural Networks (CNNs). The system processes current signal images directly through automated feature extraction and classification, eliminating the need for manual feature engineering and complex signal processing algorithms. This substitution enables the system to handle the complexity of active distribution networks with bidirectional power flows and distributed generation while maintaining high fault location accuracy.
Solution Approach 2:
The patent transforms conventional electrical signal data into image format (current signal images) as input for the CNN-based fault diagnosis system. This parameter transformation allows the application of image processing techniques to electrical fault analysis, enabling automated feature extraction and improving measurement precision in complex distribution networks with distributed generation and uncertain load conditions.
2Measurement precision
If deep CNN-based fault diagnosis system is implemented, then fault location accuracy and reliability are improved, but device complexity and computational requirements increase
Solution Approach 1:
The patent implements a modular deep learning system with separate CNN models for different fault diagnosis tasks: one CNN for fault classification and another for fault location. This segmentation allows each model to be optimized for its specific function, reducing overall system complexity while maintaining high accuracy. The modular architecture also enables independent training and deployment of each component.
Solution Approach 2:
The CNN-based system performs automated feature extraction and fault diagnosis without requiring manual intervention or complex preprocessing. The deep learning models automatically learn relevant features from current signal images and perform classification and location tasks autonomously, reducing the need for complex external processing systems and simplifying the overall diagnostic infrastructure.
3Loss of time
If real-time fault diagnosis is implemented in active distribution networks, then response time is improved, but computational load and data processing requirements increase
Solution Approach 1:
The patent pre-trains CNN models offline using extensive simulation data that captures various fault conditions, load scenarios, and distributed generation configurations. This preliminary training enables the models to learn robust feature representations and diagnostic patterns beforehand. During real-time operation, the pre-trained models perform rapid inference with minimal computational load, achieving fast fault diagnosis while reducing online energy consumption.
Solution Approach 2:
The patent uses current signal images as input data for the CNN-based fault diagnosis system. By transforming electrical signals into image format, the system leverages efficient image processing algorithms and hardware accelerators (such as GPUs optimized for image processing), enabling real-time fault diagnosis with reduced computational energy consumption compared to processing raw electrical signals directly.
Data Source
AI summary
A device and method for fault management for an electric power distribution network incorporating intermittent generation sources. The method involves configuring multiple hyperparameters for a series of Convolutional Neural Networks (CNNs). A first CNN is trained using current signal imagery from phasor measuring units (PMUs) during fault conditions to classify faults. A second CNN is trained with signal images from PMUs captured during pre-fault and fault cycles for identifying fault sections. Similarly, a third CNN is trained using these images to determine the exact fault location. Once trained, the CNNs are employed sequentially. The first CNN classifies the fault, the second detects the fault section, and the third ascertains the fault location. Subsequently, a comprehensive fault management strategy is deployed.