Intelligent online control and prediction of smart home energy demand with solar integration 

Supervisors 

A/Prof. Hamid Khayyam (RMIT)  

The project 

As more households adopt solar energy and smart appliances, managing real-time energy demand is becoming increasingly complex. Traditional forecasting methods often struggle to keep up with dynamic household behaviour and fluctuating solar generation. 

This project uses machine learning and reinforcement learning to create an intelligent, adaptive energy controller that predicts and manages household energy demand in real time. By integrating battery storage and solar forecasting, the system optimises energy use and cost while maintaining the comfort of residents.  

Using our new cross-validation and PEET methods, the accuracy and robustness of our demand forecasting models have improved by 8% and 34%, respectively, compared to conventional approaches. 

Our online forecasting techniques show a 10% accuracy improvement over traditional methods. 

The approach aims to reduce household electricity costs by up to 5%, to deliver smarter, cleaner, and more flexible residential energy systems. 

This technology empowers households to actively participate in the energy transition—enhancing demand-side flexibility, reducing grid stress, and supporting RACE for 2030’s vision of affordable, reliable, zero-emission energy solutions.