DC Arc Fault Detection Using Neural Networks in PV Circuits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting direct-current electric arcs in photovoltaic power generation systems have low accuracy due to the lack of a zero crossing point, making it difficult to extinguish the arc and posing safety hazards, as they rely on manual identification of electric arc characteristics.
Innovation Solution
A direct-current electric arc detection method using neural network models for burning and starting arc detection, where the electrical quantity is input into a burning arc neural network model and a starting arc neural network model after normalization, to improve the accuracy of identifying direct-current electric arc faults.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual identification of electric arc characteristics is used, then the detection method is simple to implement, but the detection accuracy is low
Solution Approach 1:
The patent replaces manual identification methods with automated neural network models (starting arc detection model and burning arc detection model) to analyze electrical quantity data. This substitution of mechanical/manual analysis with automated intelligent algorithms significantly improves detection accuracy while maintaining system implementability through software-based solutions.
Solution Approach 2:
The patent transforms the detection approach by changing from direct manual observation of current waveforms to using normalized electrical quantity parameters processed through neural networks. The normalization of electrical quantity data and the use of different detection models for different arc stages represent parameter changes that enhance detection precision.
2Reliability
If traditional arc detection methods are used, then the implementation is straightforward, but misidentification rate is high
Solution Approach 1:
The patent divides arc detection into two distinct segments: starting arc detection and burning arc detection. Each segment has its own specialized neural network model tailored to the specific characteristics of that arc stage. This segmentation allows each model to focus on specific features, improving reliability by reducing misidentification rates compared to a single general-purpose detection method.
Solution Approach 2:
The patent introduces normalized electrical quantity parameters as intermediaries between the raw current data and the neural network models. This normalization process serves as a mediator that standardizes the input data, making it more suitable for neural network processing and improving the overall reliability of detection while managing complexity through systematic data transformation.
3Productivity
If direct-current electric arc is not detected timely, then the system operation continues normally, but damage to photovoltaic module and transmission line occurs
Solution Approach 1:
The patent implements preliminary detection actions by using the starting arc detection model to identify the initial stage of arc formation before it develops into a harmful burning arc. This preliminary detection allows the system to take preventive measures early, maintaining productivity by avoiding complete system shutdowns while preventing damage through timely intervention.
Solution Approach 2:
The patent establishes a feedback mechanism where the detection models continuously monitor electrical quantity data and provide real-time information about arc conditions. This feedback loop enables the system to respond dynamically to arc development, allowing continuous operation under normal conditions while automatically triggering protective actions when arcs are detected, thus preventing damage while maintaining productivity.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A direct-current electric arc detection method and apparatus, a device, a system, and a storage medium are disclosed. The method includes: obtaining an electrical quantity of a direct current circuit (S101); inputting the electrical quantity into a burning arc neural network model to perform burning arc detection on the direct current circuit, to obtain a burning arc detection result (S102); and if the burning arc detection result is that a burning arc is detected, determining that a direct-current electric arc fault exists in the direct current circuit (S103). Because the electrical quantity is input into the burning arc neural network model to perform burning arc detection on the direct current circuit, a direct-current electric arc fault is detected, and accuracy of detecting a direct-current electric arc fault is further improved.