Short-Term Wind Speed Forecasting Based on Hybrid MODWT-ARIMA-Markov Model

Markov chains (MC) are statistical models used to predict very short short-term wind speed accurately. Such generally trained with a single moving window. However, time series do not possess an equal length of behavior for all horizons. Therefore, window can provide reasonable estimates but is optimal choice. In this study, forecasting model proposed that integrates MCs adjusting dynamic The selects the size based on similar approach leave-one-out method. traditional further optimized by introducing self-adaptive state categorization algorithm. Instead synthetically generating series, modified directly predicts one-step ahead speed. Initial results indicate MC prediction improved performance 50%. Based preliminary findings, novel hybrid integrating maximal overlap discrete wavelet transform (MODWT) auto-regressive integrated average (ARIMA) and MC. It evident from literature suitable predicting residual sequences. were considered as primary decomposition-based in any studies. improvement is, average, 55% deep learning 30% models.