Accurate day ahead temperature prediction using a 24 hour Kalman filter estimator

This paper presents a novel approach to the postprocessing of Numerical Weather Prediction (NWP) grid models to extract ambient air temperature predictions for “sub-grid” locations. This technique estimates air temperature observations from a NWP model using a novel bank of 24 Kalman Filters (KFs) which operate in tandem to accurately predict the air temperature, 24 hours ahead for a desired node point 51.8830N, 8.5353W, for a campus based Photovoltaic (PV) solar energy resource at Cork Institute of Technology (CIT). This accurate prediction resource is also important to meet essential heating system deployment and scheduling for domestic heating conservation and essential services such as fruit growers, hospitals and a range of air temperature sensitive applications. The NWP model outputs predicted air temperature values for Cork Airport 51.8466N, 8.4891W (2.5km from CIT). The 24 KF bank derives a 24 hour look ahead temperature schedule for CIT by applying a polynomial to map and filter the temperature bias offset between the two grid points. A 4^th order polynomial temperature model was employed in the 24 KF bank state estimator and a 3 point moving averager was used in the smoothing of the error in the state and measurement covariance matrices — Qt and Rt respectively. This procedure accurately predicts temperature with significant reduction in the Root Mean Square Error (RMSE) over existing methods [1] for a site specific location through manipulation of a NWP fixed grid model output with fixed grid points. The accurate forecasted day-ahead assignment of PV electricity generation to meet load demand can be confidently met by systems operators via the 24 hour temperature prediction estimates. This is of strategic importance where installed solar scheduling is an issue for cost effective electrical network operation with a consequent beneficial economic return on solar array installation capital investment.