Machine learning approach to understand regional disparity of residential solar adoption in Australia

Abstract

Although Australia has been successful in increasing the total number of residential solar photovoltaic (PV) panels, the disparity of PV adoption among regions has raised concerns about energy justice. To understand the regional difference of PV adoption in relation to the socioeconomic variance, this research introduced a machine learning approach, selected the Conditional Inference Trees algorithm and examined the residential PV installations in 2658 postcode areas covering six states of Australia. The study identified 18 scenarios based on 11 socioeconomic factors that explained the regional difference of residential PV adoption rate. A simple scenario was found for the region with a low density of population where the sparse population distribution is unadventurous for promoting PV among households and the PV adoption rate was reasonably low. The scenario became complex for the region with a high density of population, especially where the high density concurs with a high income; the concurrence was associated with many apartments and consequently a low adoption rate due to the lack of rooftop space. The most complex scenario was found for the region with a medium density of population where more socioeconomic factors interplayed and conditioned each other to explain the PV adoption variance. Generally, a high adoption rate was found for the region with a medium density of population and housing and a middle level of income. The complexity of the socioeconomic factors for explaining the regional difference of PV adoption should be addressed in search of more sophisticated energy policies.