University of Tasmania, Australia

Neuro-Dynamic programming with application to renewable energy system

Renewable energy is of great concern because of limited deposits of traditional energy resources such as oil and coal. Different sources of renewable energy including hydro, wind, solar, biomass, geothermal, and ocean wave have been studied and numerous tests have been conducted to explore their use as an alternative energy supply. However, cost efficiency is the major obstacle for their practical implementation. The aim of this project is to develop a solution to optimise the renewable energy system performance and minimize the cost.

Because of the complexity of the renewable energy system, many factors such as wind power, solar power, power consumption, ocean wave power, etc, cannot be modelled precisely. However progress can be made in understanding the integration of these various renewable sources to provide a reasonable comparative basis. It is viable to use a “grey box” approach to tackle the interesting issues. More specifically, existing models will be carefully selected and modified with “feature-based” artificial neural networks to provide more flexibility to model those un-modelled dynamics. Such model provides the learning capability of the proposed system. Neuro-dynamic programming grown from the celebrated Bellman dynamic programming method and reinforced learning can be used to develop the learning algorithms to adjust the system based on measurement data.