A Novel Methodology to Predict and Detect the Consumption of Power for Smart Commercial Areas Using Stacked GRU and LSTM (Called Deep GRULS Architecture)

In smart commercial environments such as airports, warehouses, and industrial plants, the challenge of balancing power supply and demand often leads to fluctuations in power consumption patterns. This dynamic power consumption behavior is influenced by various factors, making it a complex “multivariate time series problem” These factors include but are not limited to demand variations, seasonal fluctuations, and external influences. Addressing this issue requires accurate prediction of power consumption and the detection of anomalies within these patterns. Such insights are invaluable for the operational teams managing these commercial spaces, as they enable better planning and resource allocation to meet the immediate needs. In light of this, our paper introduces an innovative approach that leverages Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) neural network architectures, integrated through a stacking methodology. This hybrid approach, which combines LSTM and GRU, offers superior accuracy in comparison to conventional methods such as standalone LSTM, Bidirectional LSTM (BLSTM), standalone GRU, Autoregressive Integrated Moving Average (ARIMA), and Auto ARIMA. Through our experimentation and analysis, we demonstrate the effectiveness of this novel technique in addressing the multivariate time series problem of power consumption in smart commercial spaces, ultimately improving operational efficiency and resource management.