When nonlinearity is present, time series prediction becomes a difficult task. The ASHRAE Energy Predictor Shootout II competition problem is no exception; the difficulty is amplified because analytical equations for describing the dynamics are formidable, if not impossible. The problem belongs to a rather interesting class of problems that can arise in many practical situations. A Bayesian approach is taken in performing nonlinear regression on the ASHRAE Predictor Shootout II time series data. The Bayesian framework enables one to perform the regression in a hierarchical manner, i) level 1 - estimation of the parameters; ii) level 2 - estimation of hyperparameters; and iii) level 3 - model comparison. The prediction results appear to be reasonable.

KEYWORDS: year 1996, algorithms, calculating, expert systems, dynamic programming