This paper presents a new approach allowing Numerical Weather Prediction (NWP) grid model forecasting to be applied to a desired “sub-grid” location. It permits observations from a NWP model using a novel bank of 24 Kalman Filters (KFs) operating simultaneously to accurately predict the wind speed (Zt) 24 hours ahead for a campus based wind turbine at Cork Institute of Technology (CIT) at 20m above sea level (asl) at sub grid location. The NWP model outputs wind speed predictions (mt) for Cork Airport at 152m asl (2.5km distant from CIT) at grid level. The Kalman Filter (KF), acting as a post processing tool with a moving time averaging window, derives a 24 hour ahead predicted wind speed schedule for CIT by applying a wind speed bias model polynomial to map and filter the wind speed bias offset between the two locations. To ensure a robust model, with good modelling and error noise disturbance rejection capabilities inclusive of model offsets [1], the accuracy of the model has been investigated using a particularly turbulent wind data set for December 2013 [2].
It is shown that a 4th order polynomial adaptive wind speed model bias remover is the optimum choice to employ in conjunction with the KF which uses a 3 point a priori moving window averager to adequately eliminate systematic error. The application of a KF to wind speed prediction is implemented in MATLAB software and results are provided in this paper to demonstrate the accuracy and fidelity of the procedure. Hypothesis testing along with statistical analysis has returned wind velocity prediction estimates that demonstrate the accuracy of the KF estimator. This also provides confidence enhancement of the polynomial model choice as a suitable wind velocity bias eliminator. The accuracy of the hourly wind velocity estimate are of strategic importance in wind power prediction where installed wind turbine scheduling is an issue for cost effective electrical network operation with a consequent beneficial economic return on wind generator capital investment.
Short-Term Wind Power Prediction Based on Empirical Mode Decomposition and Extreme Learning Machine