Hydropower output forecast with tank models combined with Kalman Filter estimation

Abstract

In this paper, we studied how to forecast next-day hydropower energy expected output based on the past rainfall record in the reference area. Tank model, which has been conventionally applied for this type of forecast has been improved by employing Kalman Filter stepwise estimation, to accommodate seasonal and yearly variations of rainfall water runoff conditions. Our Kalman Filter Tank Model (KF-Tank Model) yielded a broader choice of parameters while maintaining acceptable level of average forecast errors of about +/-0.05 [PU] per forecast period. Further, it is shown that the error correction mechanism of the KF Tank Model can remain robust for snow-falling and snow-melting seasons, without calling for parameter revisions of the model. © 2008 The Institute of Electrical Engineers of Japan.