Different modelling strategies for in situ liquid chillers

This paper presents an overview of various simulation models that may be useful for predicting the in situ performance of vapour compression liquid chillers over a wide range of operating conditions. Five models were considered, namely the regression model, the steady state and transient physical models and the steady state and transient neurale network models. Typical real-time operating data taken under various conditions including start-up, quasi-static and modulating operation were used as input to the five models. The predicted performance was then compared with the experimental data to see the conditions under which each may be suitably applied. It was found that steady state models can give fairly accurate results under mildly dynamic conditions (±5 per cent) although, as the operation becomes more strongly transient, inaccuracies can be as high as 25 per cent. However, the dynamic models perform much better under these conditions (such as start-up and during sudden changes in load condition), predicting performance to within ±10 per cent.