A Short-Term Load Forecasting model using Restricted Boltzmann Machines and Bi-directional Gated Recurrent Unit

In recent years, the rate of energy consumption by industries, buildings increased gradually with increased population and daily usage. Thus, Short-Term Load Forecasting (STLF) methods are in demand which are difficult due to uncertainty of load. Integration of Restricted Boltzmann Machines algorithm with Bi-directional Gated Recurrent Unit (RBM-BiGRU) is proposed to learn hierarchical representations of data and perform effective load classification. Initially, Spark based Density Peak Clustering (DPC) with is employed for data partition to provide easy anomaly detection. The Fuzzy C-means clustering (FCMC) algorithm with high interpretability is used for load curve periodicity detection. The Restricted Boltzmann Machines (RBM) algorithm with the ability to learn complex representations is utilized to generate daily load types. The Empirical Mode Decomposition (EMD) further enhances the quality of data by eliminating high frequency noise. Finally, a Bi-directional Gated Recurrent Unit (BiGRU) is utilized for load prediction that handles temporal and long-term dependencies. The proposed RBM-BiGRU model is tested by historical data and achieved less root mean square error of 3.05%, average absolute percentage error of 1.17%. The proposed RBM-BiGRU with less prediction error surpassed other existing models like Convolutional Neural Network (CNN), Long-Short Term Memory (LSTM).