Predicting Electricity Usage Based on Weather Temperature and Usage History Data Using AI Techniques

Abstract

Building energy demand has increased globally due to the expansion of residential and commercial areas. The study focused on predicting electricity usage based on weather data and past consumption data of residential houses. The case organization, EnCoHub, has faced challenges in their business operations due to inaccurate energy purchases, unexpected demand changes, and future electricity shortages. The study focused on predicting electricity usage based on weather data and past consumption data of residential houses. The case organization, EnCoHub, has faced challenges in their business operations due to inaccurate energy purchases, unexpected demand changes, and future electricity shortages. In order to address these problems, the current analysis focused on predicting electricity consumption using deep learning models, including LSTM and 1D CNN. Additionally, different usage patterns were identified using the KMeans clustering technique. The design science research (DSR) was chosen as the research methodology, and it provides structured steps to execute the research. LSTM and 1D CNN models successfully predicted electricity consumption for the next seven days based on the past thirty days of data. Both models achieved good accuracy. According to the cluster results, the "winter demand group" used the most energy on Fridays. On the other hand, the "summer low demand group" had their highest energy use on Mondays. The "Autumn transition group” utilized the most energy on Thursdays. Finally, the "Spring high consumption group" achieved maximum usage on weekends. Average temperature, day of the week, and season were the main factors to identify these usage groups. Obtained results can be used to optimize the customer contracts according to the usage patterns. Because of accurate predictions, a sufficient amount of energy purchases ensures the effective use of energy sources. These results lead to the financial stability of the case organization and the sustainability of the energy field. The future research is focused on implementing larger data sets integrating consumption data of commercial buildings and large factories with different predicting time frames, including the next 24 hours and next 30 days.