When a solar farm's power output drops, operators must quickly answer one question: is the equipment broken, or is a cloud passing over? The only way to know is by measuring the incoming sunlight. If a weather sensor shows bright sun but a panel produces low power, the equipment is failing. If the sensor shows low light and the power drops proportionally, it is just a cloud. Therefore, accurate sunlight data is the absolute baseline for monitoring a farm's health.
However, physical weather sensors are expensive, spread too sparsely across large arrays, and frequently fail due to dust or bird droppings. When they fail, operators lose their baseline and cannot diagnose underperformance.
This project solves that hardware problem with smart software. You will build a "Virtual Irradiance Sensor" powered by lightweight machine learning. Instead of measuring sunlight directly with a fragile physical instrument, you will train an AI model to reverse-engineer the amount of sunlight hitting a panel based strictly on its electrical output (current, voltage, and temperature).
Once your model is trained on historical data, you will deploy it onto an embedded edge computer (e.g., a Raspberry Pi). This effectively turns the solar panels themselves into a dense, self-calibrating weather network, allowing the system to monitor itself without relying on physical sensors.
This project is designed to bridge the gap between theoretical coursework and real-world engineering, equipping you with highly employable skills in AI and hardware deployment that are heavily sought after in the tech and renewable energy sectors.
School
Photovoltaic and Renewable Energy Engineering
Research Area
Applied machine learning | Data driven engineering | Edge artificial intelligence (Edge-AI) | Renewable energy systems | Sensor fusion
Suitable for recognition of Work Integrated Learning (industrial training)?
No
- Research environment
- Expected outcomes
- Supervisory team
- Reference material/links
The project will be conducted within the ACDC research group at UNSW, providing access to a collaborative research environment and support from peers for discussion, problem-solving, and knowledge exchange.
Software/Tools: Python (Pandas, Scikit-learn, and/or PyTorch).
Hardware: An edge computing device (e.g., Raspberry Pi or NVIDIA Jetson).
Dataset: A small Proprietary utility-scale electrical and meteorological dataset provided by the UNSW ACDC research group.
- Train a baseline predictive model (using PyTorch or Scikit-learn).
- A benchmark report evaluating the hardware's prediction speed and accuracy.
- PV Fundamentals: PVeducation.org – The industry-standard interactive guide to solar physics and how environmental factors impact power generation.
- Machine Learning: Scikit-learn Getting Started – A highly practical, introductory guide to building baseline regression models in Python. https://scikit-learn.org/stable/getting_started.html
- Edge Deployment: PyTorch Edge Tutorials – Official documentation on how to compress and deploy machine learning models onto embedded hardware systems. https://pytorch.org/edge