Data-driven analysis and machine learning for energy prediction in distributed photovoltaic generation plants: A case study in Queensland, Australia
MetadataShow full item record
Understanding patterns and energy-related data in photovoltaic systems is one of the key tasks in energy generation and distribution. In fact, the use of data-driven tools and predictive learning models can support the government, power regulatory agencies, and the energy industry in improving their decision-making and operational activities. Bering this in mind, this paper presents a case study of data-driven analysis and machine learning to forecast the energy charge in the distributed photovoltaic power grid of Queensland, in Australia. Our analysis relies on a freely, open energy tracking platform and the design of three Machine Learning approaches built on the basis of Random Forest, Support Vector Machines, and Gradient Boosting methods. Experimental results with real data showed that the trained models allow for very consistent predictions while reaching a high forecasting accuracy (around 95%–93% in Generated - Exported prediction, respectively). Moreover, it was found that the Gradient Boosting-based model ensures robust behavior and low prediction errors, as endorsed by quality validation metrics. Another technical aspect observed is that the variables artificially created to boost the models substantially improve the post-analysis and overall accuracy of the results.