Analysis and forecasting energy consumption for educational institutions
Date Issued
2022-06
Author(s)
Aleksandra Zlatkova
Marija Markovska
Branislav Gerazov
Dimitar Taskovski
Abstract
The energy forecasting is one of the biggest challenges in last years. Nowadays, the problems as energy and economic crises, growth of population, fast development of technology impose the need of energy forecasting. In this paper, three popular deep neural networks: Stacked LSTM, LSTM encoder-decoder and CNN-LSTM encoder-decoder are used for forecasting energy consumption. They are trained and tested on time series that represent the energy consumption of Faculty of Electrical Engineering and Information Technologies in Skopje. The prediction is based on energy consumption in past 24 hours. The proposed models are evaluated with two metrics: RMSE and MAE. All three models show high performance, but LSTM encoder-decoder achieves the highest accuracy.