Improvements of fingerprint localization of hybrid kernel extreme learning machine based on differential evolution
As a possible countermeasure to the problems such as weak generalization performance and poor robustness in fingerprint localization of extreme learning machine(ELM),an improved differential evolution algorithm was proposed to optimize the finger-print localization method of hybrid kernel extreme learning machine.Specifically,the improved Logistic chaos mapping is used to improve the global search of differential evolution algorithm and the dynamic control parameter method is used to avoid falling into the local optimal.The parameters of hybrid extreme learning machine are adjusted adaptively by the improved differential evolution algorithm to improve the training efficiency.At the online stage,mixed kernel function is used to improve the learning performance and generalization performance of extreme learning machine,and L1 penalty function is introduced to prevent overfitting.Compared with the single core extreme learning machine,its generalization ability improves obviously.The positioning error of 92%test points is less than 0.5 m,and the average error is reduced by 32.6%compared with weighted K-nearest neighbor(WKNN).
NETL
NSTL
万方数据
