Deep Recurrent Mixer Models for Load Forecasting in Distribution Network

Accurate load forecasting is important for grid security, operation, and planning of the power system. The current grid network witnesses a significant transition into the smart grid in order to improve grid security, reliability and energy management. The processing of the big data generated by various sensor-enabled units requires a variety of advanced methodologies. Deep Learning (DL) is an emerging technology that can be used to optimize operational decision making & generate high intelligence. As a result, DL-based prediction methods achieved promising results. In this paper, we proposed deep recurrent mixer models (LSTM-GRU, GRULSTM) into a unified framework for accurate load prediction. The proposed methodology is based on multi-layered integration of LSTM & GRU networks to take advantage of both the techniques. The proposed models are an effective alternative to existing forecasting models in terms of model loss function & performance evaluation indices with the IEEE-33 bus power distribution network. The simulation results validate that the proposed (LSTM-GRU) mixer model outperforms the existing models. The performance evaluation indices of the proposed model include MSE, RMSE, and MAE are 0.0424, 0.2059, and 0.1106 respectively.