Nonlinear system identification using least squares support vector machine tuned by an adaptive particle swarm optimizat
- PDF / 920,473 Bytes
- 12 Pages / 595.276 x 790.866 pts Page_size
- 61 Downloads / 201 Views
ORIGINAL ARTICLE
Nonlinear system identification using least squares support vector machine tuned by an adaptive particle swarm optimization Shuen Wang1,2
•
Zhenzhen Han1,2 • Fucai Liu1,2 • Yinggan Tang1,2
Received: 29 March 2015 / Accepted: 14 July 2015 / Published online: 12 August 2015 Ó Springer-Verlag Berlin Heidelberg 2015
Abstract In this paper, we present a method for nonlinear system identification. The proposed method adopts least squares support vector machine (LSSVM) to approximate a nonlinear autoregressive model with eXogeneous (NARX). First, the orders of NARX model are determined from input–output data via Lipschitz quotient criterion. Then, an LSSVM model is used to approximate the NARX model. To obtain an efficient LSSVM model, a novel particle swarm optimization with adaptive inertia weight is proposed to tune the hyper-parameters of LSSVM. Two experimental results are given to illustrate the effectiveness of the proposed method. Keywords Nonlinear system Identification Least squares support vector machine Adaptive particle swarm optimization
1 Introduction System identification is a process for developing a mathematical model for an unknown system from input and output observations of the system. It is an important issue in many scientific and engineering fields such as control system, communication, signal processing, biological process, etc. Since almost all the real-world systems are
& Yinggan Tang [email protected] 1
Institute of Electrical Engineering, Yanshan University, Qinhuangdao 066004, Hebei, China
2
National Engineering Research Center for Equipment and Technology of Cold Strip Rolling, Qinhuangdao 066004, Hebei, China
inherently nonlinear, there are many limitations in representing these systems using linear models. Therefore, nonlinear system identification should be treated in a way that is different from linear system identification. In the past years, identification of nonlinear system has been studied extensively and many nonlinear modeling techniques have been proposed. Volterra series [25], neural networks (NNs) [1, 11, 12, 37, 39], fuzzy logic systems (FLSs) [2, 7, 8, 10, 14] and block-oriented models [13, 17] are just a few examples. Though many efforts had been paid, there are still some difficult problems associated with existing identification methods for nonlinear systems. Volterra series has many advantages, such as excellent representation ability, linear-in-the-parameters for instance, however, it involves potentially many parameters, which restricts its practical use even for truncated Volterra series [9]. It is commonly believed that NNs and FLSs can provide a fast and efficient solution to complex nonlinear system modeling due to their powerful approximation ability. Nevertheless, it is not an easy work, till now, to determine an optimal structure (number of neurons and hidden layers) of neural networks. Moreover, the learning algorithm has great effect on the performance of neural networks because some algorithms can likely find local optimal solut
Data Loading...
