An Advanced Time-Series Forecasting Methodology in Renewable Integrated Grid

To reduce the carbon footprint and decentralize the future power market, bulk renewable energy sources (RESs) are penetrating the electric grid. Although RESs are a great solution to meet the ever-increasing load demand, their dependency on the meteorological conditions is a challenge particularly when thinking about the complete replacement of conventional power generation. To balance the demand and supply, accurate forecasting of load demand and generation from RESs is essential for the suppliers to ensure efficient energy management and precise scheduling. An Adaptive Chirp Mode Decomposition (ACMD) and Attention-Based Bidirectional Long Short-Term Memory (At-BiLSTM)-based short-term forecasting method is proposed in this paper to predict the load demand, and total power generation from RESs. This hybrid approach effectively addresses the uncertainties in the load demand and the RES generation by capturing correlations between the decomposed modes, original power information, and weather variables. For accurate generation forecasting, meteorological data is used separately for both the wind and solar systems. To evaluate the accuracy of the model, performance indicators are used in terms of errors. The proposed method is compared with various existing standalone and hybrid methods to assess the efficacy of the method. The standalone methods include BPNN, RNN, GRU, LSTM, Bi-LSTM, At-BiLSTM, and Transformer learning, whereas the hybrid methods combine ACMD with these standalone methods. The proposed method demonstrates lower errors as compared to the other approaches with optimal performance in terms of mean absolute error (MAE). Specifically, the MAE observed for load, solar power and wind power are 0.078kW, 5.304kW and 0.426kW, respectively. In addition, the proposed method shows an improvement in forecasting accuracy of nearly 18% with respect to the nearest MAE achieved using Attention-BiLSTM method. The forecast model for load forecasting (LF) and RES forecasting is tested on the data collected for Rye microgrid. The proposed forecast model is also tested on the dataset collected from Panama City.