RBF neural networks optimization of the control over the class of stochastic nonlinear systems with unknown parameters

Authors:

YingminYi, Faculty of Automation and Information Engineering, Xi’an University of Technology, Xi'an 710048, Shaanxi, China

Xiangru Hu, Faculty of Automation and Information Engineering, Xi’an University of Technology, Xi'an 710048, Shaanxi, China

Abstract:

Purpose. Objective optimization is a very important area in scientific research and practical applications, many problems are related to the objective optimization. The research investigates combinational measures of Particle Swarm Optimization (PSO) and K-means clustering. The dynamic multi-swarm particle swarm optimization based on K-means clustering (KDMPSO) has been obtained, which is a hybrid clustering algorithm integrating PSO and K-means clustering, and it can nicely find global extreme in different problems.

Methodology. The comprehensive and in-depth analysis on PSO and K-means clustering was carried out, and improvement strategies have been found by adopting combinational measures of PSO and K-means clustering. For both continuous and discrete optimization problems, it has strong global search capacity; it effectively reduces the premature convergence of the traditional PSO.

Findings. Combination of the advantages of PSO and K-means clustering solves convergence to local optimum and inefficiency of traditional PSO algorithm in complex optimization problems, ensures that PSO is stable and can maintain the population diversity, avoids prematurity, and enhances the algorithm accuracy.

Originality. The multi-swarm PSO and K-means clustering were studied. In the iteration process, PSO is easy to get trapped in local optimal solution, causing the phenomenon of premature convergence, on the other hand, K-means is extensively used in clustering since it is easy to realize and it is also a highly-efficient algorithm with linear time complexity. For the first time the complementary combinational method of PSO and K-means clustering was considered.

Practical value. Since the optimization measures is widely used in business management, market analysis, engineering design and scientific exploration and other fields, the research results can be applied in in various fields. KDMPSO can effectively make up for the deficiency of the traditional PSO, and have achieved good results.

References:

1. Alpcan, T., Shames, I., Cantoni, M. and Nair, G., 2013. Learning and information for dual control. In: Proc. of the 19th Asian Control Conference, pp. 1−6, Istanbul, June 2013.

2. Alpcan, T. and Shames, I., 2015. An Information-based learning approach to dual control. IEEE Neural Networks and Learning Systems, vol. 26, no.11, pp. 2736−2748.

3. Kronander, K. and Billard, A., 2015. Passive interaction control with dynamical systems. IEEE Robotics and Automation Letters, vol. 1, no.1, pp. 106−113.

4. Zacekova, E., Privara, S., Vana, Z., Cigler, J. and Ferkl, L., 2013. Dual control approach for zone model predictive control. In: IEEE, Proc. of Control Conference, pp. 1398−1403, Europe, July 2013.

5. C. L. P. Chen, Yan-Jun Liu and Guo-Xing Wen, 2014. Fuzzy neural network-based adaptive control for a class of uncertain nonlinear stochastic systems. Cybernetics, vol. 44, pp. 583−593.

6. Tieshan Li, Zifu Li, Dan Wang, Chen, C.L.P., 2015. Output-feedback adaptive neural control for stochastic nonlinear time-varying. Neural Networks and Learning Systems, vol. 26, pp. 1188−1201.

7. Yunpeng Pan and Jun Wang, 2012. Model predictive control of unknown nonlinear dynamical systems based on recurrent neural networks. IEEE Industrial Electronics, vol. 59, no.8, pp. 3089−3101.

8. Talel, B. and Ben Hmida, F., 2013. Fuzzy Kalman filter for nonlinear stochastic systems. In: IEEE, 10th International Multi-Conference on Systems, Signals & Devices, pp.1−7, Hammamet, 18−21 March 2013.

9. Wei-Yu Chiu, Bor-Sen Chen, Poor, H.V., 2013. A multi-objective approach for source estimation in fuzzy networked systems. IEEE Circuits and Systems I: Regular Papers, vol. 60, no.7, pp. 1890−1900.

10. Zhao, J, Wei, H, Zhang, C, Li, W, Guo, W and Zhang, K., 2015. Natural gradient learning algorithms for RBF networks. Neural Computation, vol. 27, no.2, pp. 481−505.

 

2016_01_yi-hu
2016-04-02  549.88 KB  952