An interpretable classification approach for Solar PV load profiles using decision trees

As penetration of domestic solar PV generation grows there is a need for electricity distribution network operators (DNOs) to have methods for detecting solar PV generators attached to their networks. The cause of this requirement is the need for regulatory compliance, safety of equipment, and protection of workers and consumers. Smart metering is a key component of smart grids and smart metering data creates avenues to address this issue. Algorithmic methods to identify solar PV generation from consumption data will allow DNOs to maintain up to date knowledge of their networks allowing them to address any issues of safety or compliance. In this paper we investigate if classification of smart metering electricity consumption daily load profiles as solar or non-solar is possible with decision tree classifiers. We compare decision trees on our data after different forms of preprocessing are applied. We then apply this approach to smart metering datasets of solar and non-solar customers and show the ability to classify solar PV daily load profiles with up to 92% accuracy. This contributes to the knowledge about smart metering data analytics methods that can be used by smart grid stakeholders.