1 3 Artif Life Robotics (2014) 19:347–353 DOI 10.1007/s10015-014-0173-x ORIGINAL ARTICLE An improved adaptive switching control based on quasi-ARX neural network for nonlinear systems Imam Sutrisno · Chi Che · Jinglu Hu Received: 7 April 2014 / Accepted: 19 August 2014 / Published online: 15 October 2014 © ISAROB 2014 1 Introduction Adaptive control of nonlinear dynamical systems has attracted much attention and developed significantly dur- ing the last of few decades. Many adaptive control methods have been proposed and the corresponding stability and con- vergence have been proved [1, 2]. Unfortunately, the perfor- mance of linear control models cannot satisfy the require- ment. Hence, some nonlinear prediction models have been developed for nonlinear systems to overcome the difficulty in design of predictor and controller for nonlinear systems. One approach to identify and control nonlinear dynamical systems is neural networks because of its ability to approxi- mate the arbitrary mapping to any desired accuracy [3, 4, 5]. However, there are two major problems on those neu- ral network models. The first, their parameters do not have useful interpretations. The second, they do not have a friendly interface for controller design and system analy- sis. To solve these problems, in the previous work, a quasi- linear auto regressive exogenous (quasi-ARX) neural net- work (QARXNN) modeling scheme has been proposed based on well-developed linear system theory and can be extended to nonlinear systems. The models consist of two parts: a macro model part and a kernel part [6, 7, 8]. The QARXNN model has two properties: the linear property and the nonlinear property. Based on the model character- istics, two controllers can be obtained: one linear controller and one nonlinear controller. The linear controller is used to ensure the control stability and the nonlinear controller is utilized to improve the control accuracy. The 0/1 switching mechanism is proposed between the two controllers. If the switching flag is 0, then a linear controller is employed oth- erwise a nonlinear controller is employed. In such a way, the quasi-ARX prediction model uses only one model to achieve function of two or more models. Abstract In this paper, an improved switching mecha- nism based on quasi-linear auto regressive exogenous (quasi-ARX) neural network (QARXNN) is presented for the adaptive control of nonlinear systems. The proposed switching control is composed of a QARXNN-based pre- diction model and an improved switching mechanism using two new adaptive control laws, first is moving average fil- ter law and second is new switching law. Since the control result of nonlinear predictor is better than the linear predic- tor in most of the time, the adaptive control with a simple switching mechanism has many useless switching during the processing. Hence, the improved smooth switching mechanism is proposed to replace the original switching mechanism; it can improve the performance by reducing the useless switching while guaranteeing stability of the system control. The simulations show that the efficiency of the proposed control method is satisfied in stability, improve control accuracy and robustness. Keywords Adaptive switching control · An improved switching mechanism · Quasi-ARX neural network prediction model This work was presented in part at the 19th International Symposium on Artificial Life and Robotics, Beppu, Oita, January 22–24, 2014. I. Sutrisno · C. Che · J. Hu Graduate School of Information, Production and Systems Waseda University, Hibikino 2-7, Wakamatsu-ku, Kitakyushu 808-0135, Fukuoka, Japan I. Sutrisno (*) Politeknik Perkapalan Negeri Surabaya, Jl. Teknik Kimia, Kampus ITS Sukolilo, Surabaya 60111, Jawa Timur, Indonesia e-mail: imams3jpg@moegi.waseda.jp