Title :
An evaluation pattern generation scheme for electric components in hybrid electric vehicles
Author_Institution :
Bio & Meas. Syst. Lab., Hitachi, Ltd., Saitama, Japan
Abstract :
A novel multi-objective optimization scheme (MOEWE) is proposed for the purpose of generating the operation profiles of electric components, even if the control method of the components´ system is unknown. This method continuously receives evaluation feedback from system outputs and updates the scalar weights for each objective to estimate them. A continuous/discrete hybrid radial basis function network (HRBFN) is adopted to describe the values of selected scalar weights. The values are updated by reinforcement learning with feedback rewards generated by an estimator that uses the system outputs. Applying the process sequentially brings the operation profile close to the desired one. The proposed scheme was applied to a hybrid electric vehicle (HEV) simulation using the LA92 driving pattern. The results show that the scheme suitablygenerates the operation profiles of electric components.
Keywords :
hybrid electric vehicles; learning (artificial intelligence); mechanical engineering computing; optimisation; radial basis function networks; LA92 driving pattern; electric components; evaluation pattern generation scheme; feedback rewards; hybrid electric vehicles; hybrid radial basis function network; multiobjective optimization scheme; reinforcement learning; Batteries; Engines; Force; Hybrid electric vehicles; Mathematical model; System-on-a-chip; hybrid electric vehicle; multi-objective optimization; operating profile; radial basis function; reinforcement learning;
Conference_Titel :
Control Applications (CCA), 2010 IEEE International Conference on
Conference_Location :
Yokohama
Print_ISBN :
978-1-4244-5362-7
Electronic_ISBN :
978-1-4244-5363-4
DOI :
10.1109/CCA.2010.5611262
