Recurrent Neural Networks - Gated Recurrent Unit for Behaviour Analysis and Prediction Modelling to Trigger High Energy-consumption Alert

Nowadays, most people are less concerned about being energy efficient. Being energy efficient means using less electrical power and use energy-saving appliances to reduce energy consumption. In this regard, several measures have been taken by the government to give awareness to Malaysians about the problem. Identifying the need for more intensive energy-efficient usage, this study focuses on exploring the analysis to learn and identify the behaviour of electrical usage and to predict future usage for the user. Due to its well-known capabilities in time-series forecasting, especially on a large dataset, the Recurrent Neural Network (RNN) method is employed in this study for modelling and prediction analysis. RNN has two different models, i.e., Long-Term Short Memory (LSTM) and Gated Recurrent Unit (GRU). GRU is implemented in this case study to predict and forecast future data. Energy profiles in the form of trained models are constructed from the collected energy consumption data, which arc later used for predicting future consumption behaviour. In this study, the accuracy for implementing GRU as prediction modelling gives Explained Variance and R-squared near to 100% and values close to 0 for accuracy of Mean Square Error (MSE). The generated prediction draws a general picture of users' common electricity usage pattern based on their previous records. The actual data that does not fulfil what has been predicted means that the actual data is not following the common trends. In this context, the actual data that exceeds the usual limit can be the main cause of energy wastage. Alerts are sent to users whenever the usual limits are exceeded, especially to warn them about such events, so that the users can take some precautions. In general, such a measure can greatly help the users to reduce their utility bills.