Quantum Positive Unlabeled Learning Algorithms with Applications to Energy

Background

Positive unlabeled (PU) learning is a semi-supervised machine learning approach to a binary classification problem in which most of the data is unlabeled. PU learning algorithms reduce the computational cost of training machine learning classifiers, because the labeling of data is time-intensive for supervised machine learning algorithms. There has been some recent research in applying customized machine learning algorithms on classical computers for solar array or photovoltaic (PV) fault detection, but these methods typically have limited accuracy, long training times, and heavy computing and memory requirements.

Invention Description

Researchers at Arizona State University have developed a new method for photovoltaic fault detection using quantum positive unlabeled (QPU) learning methods. This method combines the computational advantages of quantum computing with the semi-supervised machine learning paradigm of positive unlabeled learning to improve the efficiency and accuracy of fault detection in solar energy systems. This approach uses quantum-classical hybrid neural network architectures to address challenges posed by quantum noise and computational limitations, with initial results showing reduction in both labeling costs and computational demands.

Potential Applications:

• Solar energy monitoring and fault detection
• Development of quantum-classical hybrid machine-learning models
• Enhanced semi-supervised learning frameworks for applications requiring reduced labeling
• Quantum computing applications (e.g., signal processing, automated monitoring systems)

Benefits and Advantages:

• Reduces computational cost – training machine learning classifiers can be accomplished with reduced labeling costs
• High accuracy – improved photovoltaic fault detection capabilities despite quantum noise and limited computational resources
• Efficient – semi-supervised PU learning approach improves handling of labeling costs

Related Publication: Quantum Positive Unlabeled Learning Algorithms with Applications to Energy