Output prediction of CMF based on improved hybrid genetic algorithm and support vector machine

Abstract

annealing algorithm;BP neural network;membrane flux Abstract: Through improved select tactics and genetic operators, the accelerating genetic algorithm (AGA) and simulated annealing algorithm (SA) were combined to form a new algorithm called accelerating genetic and simulated annealing algorithm (AGSA). A modified method to develop the flow rate prediction model of the continuous micro-filtration (CMF) system was proposed based on improved hybrid genetic algorithm and support vector machine (SVM). A new self-adapting optimized algorithm was formed and applied to the SVM parameters. The hybrid genetic algorithm was utilized to perform variable selection, and SVM was employed to construct prediction models. The prediction models were verified by a flow rate experiment in a pilot-scale continuous micro-filtration system. Results showed that the proposed model can reveal the rule of flow rate variation in CMF. It produced a small error and exhibited strong correlation (R 2 =0.91, MAE=0.0132, SSE=0.0055, RMSE=0.0155) between predicted and measured values. This result reveals that the model has strong predictability. According to the leave-one-out cross validation of training samples, the model also shows good robustness (R 2 =0.89, MAE=0.0164, SSE=0.0073, RMSE=0.0178). The model developed by AGSA-SVM was compared with the model constructed by a BP neural network. The former exhibited optimal predictive capability and robustness in the comparison and is thus more suitable for the flow rate prediction of CMF.