Utilizing physics-based input features within a machine learning model to predict wind speed forecasting error

Abstract. Machine learning is quickly becoming a commonly used technique for wind speed and power forecasting. Many machine methods utilize exogenous variables as input features, but there remains the question of which atmospheric are most beneficial forecasting, especially in handling non-linearities that lead to forecasting error. This addressed via creation hybrid model utilizes an autoregressive integrated moving-average (ARIMA) make initial forecast followed by random forest attempts predict ARIMA error using knowledge variables. Variables conveying information about stability turbulence well inertial forcing found be useful dealing with non-linear prediction. Streamwise speed, time day, intensity, turbulent heat flux, vertical velocity, direction particularly when unison hourly 3 h timescales. The prediction accuracy developed ARIMA–random compared persistence bias-corrected models. shown improve upon latter employed modeling methods, reducing up 5 % below achieving R2 value 0.84 true speed.