Univariate Energy Consumption Forecasting in Smart Buildings using RNN & LSTM

Forecasting energy consumption is a significant component of smart grid system stations and smart building management system. Due to significant fluctuations in the supply of power and demand, the forecast of accurate energy usage is a crucial problem for an effective energy management system. Researchers have used various methods of machine learning to address this problem. But due to the changing weather conditions and some other exogenous parameter, the results of the approaches are not good. This paper describes some existing approaches for energy forecasting with their advantages and disadvantages. The paper also proposes an appropriate type and size of dataset to be used for better forecasting of energy usage in smart buildings using two models i.e. simple recurrent neural network (RNN) and long short term memory network (LSTM) model. The findings show that LSTM performs better than RNN in terms of data with increasing trend rather than the stationary data.