Predicting future electricity usage from available data

Abstract

This thesis presents the development of a machine learning predictive model for forecasting electricity consumption in residential units in Finland, including traditional houses and seasonal summer cottages. The primary objective was to help power plants and distributors in accurately anticipate the energy demand in the future, contributing to more efficient energy management and planning. This project focused on predicting electricity usage through data preparation and the development of a machine-learning prediction model for future energy consumption in Finland. The objective was to develop a model that accurately predicts household electricity consumption. To create this prediction model, historical energy consumption data were used, which were combined with meteorological datasets from Jyväskylä, Pori, and Vesanto. These datasets were incorporated with important climate-related factors, such as temperature, precipitation, and wind speed, into the prediction model. Additionally, data from all holiday days were integrated to add variations in energy usage patterns during these special dates to the model. To ensure data consistency and relevance, the datasets were cleaned, pre-processed and missing values were handled. Household energy data was merged with weather data as well as holiday data based on the day. This predictive model aimed to assist both power plants and distribution companies in the future. It attempted to increase the accuracy in predicting energy consumption, contributing to more sustainable and adaptive strategies. As a result, planning and anticipating energy demand in homes would lead to more efficient management of electrical resources. This could contribute to an optimized allocation of resources, which could help in a smarter and more sustainable use of electricity.