Power Distribution Fault Location Using Mixed-Mode Wave Recording
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current fault location methods in power distribution networks require manual extraction of waveform features and divide the process into two steps, leading to inaccurate and incomplete fault identification, especially for intermittent faults, as they do not utilize direct end-to-end solutions and are limited by machine learning model processing capacity.
Innovation Solution
A multi-channel deep neural network system is constructed to process synchronized high-frequency and low-frequency component sequences from monitoring points, using convolution layers and Long Short-Term Memory (LSTM) cells, with hyperparameter optimization for improved fault location accuracy, enabling end-to-end fault identification and decision-making.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If manual waveform feature extraction and two-step fault location method are used, then device complexity is reduced, but fault location accuracy and detection completeness deteriorate
Solution Approach 1:
The patent replaces manual waveform feature extraction and two-step fault location methods with an end-to-end deep neural network system. The DNN model directly processes raw waveform data from multiple monitoring points and outputs fault location results, eliminating the need for manual feature extraction and intermediate processing steps. This substitution of mechanical/manual processes with an intelligent system simultaneously improves fault location accuracy while maintaining manageable system complexity through automated processing.
Solution Approach 2:
The deep neural network system performs multiple functions simultaneously: it extracts waveform features, identifies fault types, and locates faults in a single end-to-end process. The multi-channel DNN architecture processes high-frequency and low-frequency components from multiple monitoring points concurrently, providing comprehensive fault diagnosis capabilities that improve detection completeness without requiring separate dedicated systems for each function.
2Productivity
If existing machine learning models are applied with truncated wavelet segments, then processing speed is improved, but information completeness and detection capability deteriorate
Solution Approach 1:
The patent segments the waveform processing into multiple frequency channels (high-frequency and low-frequency components) processed by separate DNN sub-networks. Each channel processes relevant frequency components independently, preserving important waveform characteristics while enabling parallel processing. This segmentation allows the system to maintain information completeness by processing different frequency bands separately rather than truncating the entire waveform, while still achieving fast processing through the efficient DNN architecture.
Solution Approach 2:
The patent transforms the one-dimensional time-series waveform into a multi-dimensional representation by separating high-frequency and low-frequency components into different processing channels. This dimensional transformation allows the DNN model to capture both transient fault characteristics (high-frequency) and progressive fault characteristics (low-frequency) simultaneously, preserving complete waveform information while enabling efficient parallel processing through the multi-channel architecture.
3Reliability
If long-span diagnostic waveforms are used for fault progression detection, then detection capability for intermittent faults is improved, but model processing capacity requirements increase
Solution Approach 1:
The patent implements a dynamic multi-channel DNN architecture that adaptively processes different frequency components. The system dynamically separates and processes high-frequency transients and low-frequency progressive changes through dedicated channels, allowing efficient handling of long-span waveforms. This dynamic processing approach enables the detection of intermittent faults by preserving long-span waveform information while managing model complexity through specialized channel architectures optimized for different temporal characteristics.
Solution Approach 2:
The patent changes the processing parameters by separating waveform analysis into different frequency domains with dedicated DNN channels. The high-frequency channel captures transient fault parameters, while the low-frequency channel captures progressive fault parameters from long-span waveforms. This parameter separation allows the system to process extended waveform data for intermittent fault detection while controlling model complexity through frequency-domain decomposition and specialized channel designs.
Data Source
AI summary
The present disclosure relates to the field of power technology, and in particular to a system for locating fault in power distribution network based on mixed mode wave recording.