A kernel partial least squares method for gas turbine power plant performance prediction

The change of the performance of gas turbine power plant may be dramatic under off-design conditions. To describe the off-design performance of gas turbine power plant, good prediction tools are essential. The objective of this paper is to asses the feasibility of the kernel partial least squares (KPLS) technique in performance prediction of gas turbine power plant under off-design conditions. Historical data from the real industrial gas-steam combined cycle of a cogeneration plant unit were used to train KPLS regression models and the KPLS parameters, such as the number of latent variables, were determined by a 5-fold cross-validation with the root-mean-squared-error. Results obtained by KPLS models are compared with the measured data. It was shown that, under given off-design conditions, the KPLS tool was able to predict the unit load and gas flow with a high degree of accuracy.