Transformer Inrush Current Detection Using Neural Network Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current inrush current detection methods for transformers, such as the second harmonic method, are slow and inaccurate, and fail to distinguish between inrush current and fault current, which can lead to ineffective latch-up protection and increased sampling requirements.
Innovation Solution
A neural network-based method that samples current signals to generate a numerical matrix, which is then input into an inrush current detection neural network for rapid and accurate classification, using convolutional layers and pooling to reduce data complexity and improve feature extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the second harmonic method is used for inrush current detection, then the detection can identify inrush current based on harmonic content, but the detection speed is slow and accuracy is insufficient
Solution Approach 1:
The patent replaces traditional signal processing methods (second harmonic analysis) with a neural network-based intelligent detection system. The neural network model processes current signal features to identify inrush current, achieving both high speed and high accuracy by leveraging machine learning capabilities rather than conventional mathematical transformations.
Solution Approach 2:
The patent transforms the detection approach by changing from analyzing harmonic content parameters to using a neural network that processes multiple signal features simultaneously. This parameter transformation enables the system to achieve rapid detection without sacrificing accuracy, as the neural network can evaluate multiple dimensions of the current signal in parallel.
2Speed
If traditional detection methods are used, then the system can operate with standard sampling rates, but the detection accuracy and speed are insufficient
Solution Approach 1:
The patent substitutes traditional sampling and analysis methods with a neural network-based detection system that can achieve high-speed detection while maintaining high accuracy. The neural network processes feature-extracted data rather than raw sampled data, enabling faster detection without requiring increased sampling rates.
Solution Approach 2:
The patent extracts key features from the current signal before feeding them to the neural network. This feature extraction process isolates the most relevant characteristics of the signal, allowing the neural network to make accurate detections faster by focusing on essential features rather than processing complete raw signal data.
3Reliability
If conventional detection algorithms are applied, then the system structure remains simple, but the detection performance is inadequate
Solution Approach 1:
The patent replaces conventional detection algorithms with a neural network-based system, accepting increased computational complexity in exchange for dramatically improved detection reliability. The neural network, once trained, provides robust and accurate inrush current identification that significantly enhances latch-up protection effectiveness.
Solution Approach 2:
The patent performs preliminary training of the neural network offline using historical data, so that during actual operation, the system only needs to execute the trained model for rapid detection. This preliminary action separates the complex learning phase from the operational phase, making the real-time detection process simpler while maintaining high reliability.
4Measurement precision
If high sampling rates are used to improve detection accuracy, then the detection precision increases, but the data complexity and processing burden increase
Solution Approach 1:
The patent extracts essential features from the current signal at standard sampling rates, obtaining the most relevant information without processing complete high-rate data. This feature extraction reduces data complexity while preserving the information needed for accurate detection, avoiding the need for high sampling rates.
Solution Approach 2:
The patent replaces high-rate sampling with a neural network that processes extracted features. The neural network compensates for the reduced data input by learning complex patterns from the features, achieving high detection precision without the processing burden of high sampling rates.
Data Source
AI summary
An inrush current detection method, an inrush current detection device and a computer-readable storage medium for a transformer are disclosed. The inrush current detection method includes sampling at least a part of a current signal of the transformer to obtain a numerical matrix; providing the numerical matrix as an input to an inrush current detection neural network; and calculating and outputting a label vector corresponding to the numerical matrix by the inrush current detection neural network, wherein the label vector indicates whether the current signal is an inrush current.