With improvement of PET quality, the color of PET is required to be controlled. But the mechanism model of PET color is not so clear. In this paper, a modeling method of color values (B*) is introduced. It is based on...
详细信息
With improvement of PET quality, the color of PET is required to be controlled. But the mechanism model of PET color is not so clear. In this paper, a modeling method of color values (B*) is introduced. It is based on sparse pseudo-input gaussian process. Online correction is added to the modeling based on sliding time window. Comparing to support vector regression, the model using gaussian process gives both prediction and predictive variance. The results show that B* value model using gaussian process is more flexible and can be used in real process.
The performance of robust adaptive control system based on traditional revised algorithm is awful in the presence of unmodeled dynamics and the performance will go bad, even be unstable, when the working condition (in...
详细信息
The performance of robust adaptive control system based on traditional revised algorithm is awful in the presence of unmodeled dynamics and the performance will go bad, even be unstable, when the working condition (including the parameter, reference input, and so on) change. This paper designs the multiple models robust adaptive controller to control the discrete single-variable non-minimum phase system. First, the higher part of the system is deem to be the unmodelled dynamic and the normalization is introduced to convert the system unmodeled dynamics to bounded disturbance as well as the dead zone is introduced in parameter update algorithm. And then, the multiple models robust adaptive controllers are designed based on the lower part of the system and the best controller is chosen based on the switch index to control the system. The last simulation illustrates that the proposed method is preferable even the working condition changes.
Linear programming support vector regression shows improved reliability and generates sparse solution, compared with standard support vector regression. We present the v-linear programming support vector regression ap...
Linear programming support vector regression shows improved reliability and generates sparse solution, compared with standard support vector regression. We present the v-linear programming support vector regression approach based on quantum clustering and weighted strategy to solve the multivariable nonlinear regression problem. First, the method applied quantum clustering to variable selection, introduced inertia weight, and took prediction precision of v-linear programming support vector regression as evaluation criteria, which effectively removed redundancy feature attributes and also reduced prediction error and support vectors. Second, it proposed a new weighted strategy due to each data point having different influence on regression model and determined the weighted parameter p in terms of distribution of training error, which greatly improved the generalization approximate ability. Experimental results demonstrated that the proposed algorithm enabled the mean squared error of test sets of Boston housing, Bodyfat, Santa dataset to, respectively, decrease by 23.18, 78.52, and 41.39%, and also made support vectors degrade rapidly, relative to the original v-linear programming support vector regression method. In contrast with other methods exhibited in the relevant literatures, the present algorithm achieved better generalization performance.
Estimation of distribution algorithm (EDA) is a kind of evolutionary algorithm which updates and samples from probabilistic model in evolutionary course. The key of EDA is the construction of probability model suitabl...
详细信息
ISBN:
(纸本)9781424454402
Estimation of distribution algorithm (EDA) is a kind of evolutionary algorithm which updates and samples from probabilistic model in evolutionary course. The key of EDA is the construction of probability model suitable for real distribution. Gaussian distribution is widely used in EDAs but the assumption of normality is not realistic for many real-life problems. In this paper, a new EDA using kernel density estimation (KEDA) is introduced. Adaptive change strategy of kernel width is presented and selection scheme, sampling method are also given cooperated with KEDA. The results of 5 benchmark functions show that results of KEDA outperform PBIL C , UMDA C , EDA G , H-EDA.
暂无评论