Francesco Romagnoli, "Electric load time series forecasting and relative predictions on simulation model ", UNIVERSITA’ POLITECNICA DELLE MARCHE, ANCONA, IT, 2021.

Time series forecasting is an important area of machine learning because there are so many prediction problems that involve a time component. A normal machine learning dataset is a collection of observations, while a time series dataset adds an explicit order dependence between observations, represented by time dimension. This additional dimension is both a constraint and a structure that provides a source of additional information. In this topic, electric load time series forecasting represent a crucial task in the next future. The progressive replacement of fossil fuels in a wide energy demand case with the using of electricity, like cooling or heating of buildings and transport with the spreading of electric vehicles, make electric power demand predicitions fundamental in order to develop a smart grid structure at any level, from electric industry to facilities and private houses. Electricity load forecasting allows to make this exponentially growing of using of electricity energy sustainable, with smart management of the energy resources to cover the electric demand and savings costs. The objective of the research is to investigate different time series forecasting algorithms applied to electric consumption of a facility, including the estimation of some confidence bounds, such as minimal and maximal load values as well as the expected load standard deviation.