A wind speed forecasting model based on robust local mean decomposition and bidirectional long short term memory

Abstract Wind speed is a key factor in determining wind power generation, but its inherent variability and unpredictability offer significant risks to the stability of power system. Accurate wind speed forecasting is crucial for preserving grid stability, allowing for the smooth integration of wind energy, increasing grid resilience, optimizing power flow, and avoiding disruptions to make energy system sustainable and efficient. To address this, a hybrid model combining robust local mean decomposition (RLMD) and bidirectional long short-term memory (Bi-LSTM) is proposed to forecast the wind speed. RLMD is employed to mitigate the nonstationarity and nonlinearity of wind speed time series by decomposing them into different subseries. Following that, a BiLSTM deep learning model is constructed with optimal hyperparameters. Subsequently the decomposed components and BiLSTM predictor are used to estimate wind speed, and the final result is obtained by ensemble the outputs. To evaluate the performance of the models, the study considered the wind speed dataset from two offshore wind farms: Tamil Nadu and Gujarat. A total of nine models is developed to observe the performance of the proposed models using different statistical metrices. Moreover, the study also tests the null and alternate hypothesis using Diebold–Mariano hypothesis test. The findings reveal that the proposed methodology significantly improves the performance of standalone models and outperforms the others across all datasets.