Fed-batch processes are inherently more difficult to model than continuous processes due to the variations under complex operation, drifting and small-sample condition. The classical kernel-based regression (KR) methods, e.g., least squares support vector regression (LSSVR), aim to achieve a universal generalization performance, which may be unsuitable for some local regions. Local LSSVR model which only uses the neighbors of the new sample helps improve the accuracy, but it leads a heavy computation load. Inspired by the idea of universal and local learning simultaneously, an adaptive local weighed kernel learning method (ALW-KR) is proposed here. In ALW-KR, adaptive weights are assigned to corresponding samples based on the similarity measurement, followed by a recursive updating to obtain local model. The proposed ALW-KR framework is applied to the prediction of biomass concentration in the penicillin fed-batch processes. The experimental results show that ALW-KR could predict the biomass concentration more accurate and robust to batch-to- batch variation than traditional KR methods.
Available at: http://works.bepress.com/jun_ji/4/