Feedback-Enhanced Positive Unlabeled Learning for Photovoltaic Fault Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current fault detection methods in photovoltaic systems require large amounts of labeled training data, are not fault-specific, and struggle to accurately classify different types of faults, leading to power loss and potential hazards.
Innovation Solution
A custom Positive Unlabeled learning methodology using a Feedback-enhanced Modified Logistic Regression (MLRf) module that integrates a feedback loop to effectively classify PV faults with a small amount of labeled data, leveraging unlabeled data for enhanced feature engineering and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning algorithms are used for fault detection, then fault classification accuracy can be improved, but large amounts of labeled training data are required which are difficult and expensive to obtain
Solution Approach 1:
The patent implements a feedback loop where the classifier's predictions are fed back into the system to iteratively refine the classification model. The feedback-enhanced modified logistic regression uses prediction results to adjust feature weights and improve subsequent classifications, allowing the system to achieve high accuracy with minimal labeled data by continuously learning from its own predictions and corrections
Solution Approach 2:
The system performs self-training by automatically generating pseudo-labeled data from its own predictions and using this to refine its classification model without requiring external labeled data. The algorithm self-corrects by identifying misclassified samples and retraining on these examples, enabling the system to improve its own performance autonomously with very little initial labeled data
2Device complexity
If general fault detection algorithms are used, then system complexity can be reduced, but the algorithms cannot distinguish among different types of faults
Solution Approach 1:
The patent segments the fault detection process into distinct classification stages, where the modified logistic regression classifier divides faults into specific categories (e.g., soiling, shading, degradation, electrical faults). Each fault type is identified through separate feature analysis and classification rules, enabling the system to distinguish among different fault types while maintaining manageable algorithmic complexity through modular processing
Solution Approach 2:
The system adds a classification dimension by incorporating fault-type-specific features and multiple classification layers. Instead of a single general fault detection output, the algorithm extends the classification space to include multiple fault categories, achieving fault-specific discrimination by operating in an expanded feature and classification dimension while building upon the base logistic regression framework
3Measurement precision
If more labeled data is collected to improve classification accuracy, then fault detection precision can be enhanced, but the cost and time for data collection and labeling increase substantially
Solution Approach 1:
The system performs preliminary classification using the feedback-enhanced modified logistic regression with minimal labeled data before full deployment. The algorithm pre-trains on a small initial labeled dataset and uses feedback loops to pre-refine its classification capabilities, reducing the need for extensive subsequent data collection and labeling by establishing accurate baseline classification performance in advance
Solution Approach 2:
The system creates pseudo-labeled copies of existing labeled data through its feedback mechanism, where predicted classifications are copied and used as additional training examples. This data copying approach generates synthetic training samples from the model's own predictions, effectively multiplying the utility of the small initial labeled dataset without requiring additional field data collection or manual labeling efforts
Data Source
AI summary
Various embodiments of a system and associated method for identifying and classifying faults in a photovoltaic array using relatively little labeled data are described herein. In particular, the system builds on existing PU classification techniques by addition of a feedback loop that enables classification of limited operational data of a photovoltaic array by expanding a plurality of features within the operational data based on a learned importance of each feature.