Identification and Speed Control of Ultrasonic Motors
Based on Modified Immune Algorithm and
Elman Neural Networks
Abstract. An improved artificial immune algorithm with a dynamic threshold is
presented in this paper. Numerical experiments show that compared with the genetic
algorithm and the originally real-valued coding artificial immune algorithm, the
improved algorithm possesses high speed of convergence and good performance of
preventing the premature convergence. The proposed algorithm is employed to train
the network structure, weights, initial inputs of the context units and self-feedback
coefficient of the modified Elman network. A novel identifier and controller are
constructed successively based on the proposed algorithm. A simulated dynamic
system of the ultrasonic motor (USM) is considered as an example of a highly
nonlinear system. The novel identifier and controller are applied to perform the speed
identification and control of the ultrasonic motors. Numerical results show that both
the identifier and controller based on the proposed algorithm possesses not only high
convergent precision but also robustness to the external noise.
Keywords: dynamic threshold; artificial immune algorithm; Elman network;
ultrasonic motor; system identification; control.
1 Introduction
The immune system is the basic and remarkable defense system against bacteria, viruses
and other disease-causing organisms. It can produce millions of antibodies from hundreds
antibody genes and can protect animals which are infected by foreign molecules to survive
[1-4]. The Artificial Immune System (AIS) or Artificial Immune Algorithm (AIA) was
inspired by the immune system. Compared with genetic algorithm (GA), AIA has affinity
calculation function, which could explain the relationship not only between the antigen and
the antibody but also between antibodies. That makes AIA have the unique characteristic to
guarantee the survival of the variant offspring that could match the antigen better. Related
papers [5, 6] show that the algorithms based on AIA have much better performance than
conventional probabilistic optimization algorithms. However, it usually takes long time for
the binary coding AIA to obtain convergence. Furthermore, it is very difficult for AIA to
break away from the local optimal value, which can hold the searching process around this
value and can easily lead to the premature during the evolution.
1