Petr Pivonka

Comparative analysis of fuzzy PI/PD/PID controller based on classical PID controller approach

Fuzzy PID controllers are physically related to classical PID controller. The parameters settings of classical and fuzzy controllers are based on deep common physical background. The original newly introduced method considerably simplifies the setting and realization of fuzzy PI/PD/PID controllers.

Two Level Parallel Grammatical Evolution

This paper describes a two level parallel grammatical evolution (TLPGE) that can evolve complete programs using a variable length linear genome to govern the mapping of a Backus Naur Form grammar definition. To increase the efficiency of grammatical evolution (GE) the influence of backward processing was tested and a second level with differential evolution was added. The significance of backward coding (BC) and the comparison with standard coding of GEs is presented. The new method is based on parallel grammatical evolution (PGE) with a backward processing algorithm, which is further extended with a differential evolution algorithm. Thus a two-level optimization method was formed in attempt to take advantage of the benefits of both original methods and avoid their difficulties. Both methods used are discussed and the architecture of their combination is described. Also application is discussed and results on a real-word application are described.

Adaptive Controllers by Using Neural Network Based Identification for Short Sampling Period

The use of short sampling period in adaptive control has not been described properly when controlling the real process by adaptive controller. The new approach to analysis of on-line identification methods based on one-step-ahead prediction clears up their sensitivity to disturbances in control loop. On one hand faster disturbance rejection due to short sampling period can be an advantage but on the other hand it brings us some practical problems. Particularly, quantization error and finite numerical precision of industrial controller must be considered in the real process control. We concentrate our attention on dealing with adverse effects that work on real-time identification of process, especially quantization. It is shown; that a neural network applied to on-line identification process produces more stable solution in the rapid sampling domain

Numerical Aspects of Inertial Navigation

Abstract This paper presents an investigation of the possibility of using the fixed-point arithmetic in the inertial navigation systems which use the local level navigation frame mechanization equations. Two square root filtering methods, the Potters square root Kalman filter and UD factorized Kalman filter, are compared with respect to the conventional Kalman filter and its Josephs stabilized form. The effect of rounding errors to the Kalman filter optimality is evaluated for various lengths of the fractional part of the fixed-point computational word. Main contribution of this research lies in an evaluation of the minimal fixed-point arithmetic word length for the Phi-angle error model with noise statistics which correspond to the tactical grade inertial measurements units.

The real-time identification of dynamic systems by using the neural net approach

Neural networks play a significant role in the sphere of control of a dynamic systems. It can be appropriately used for continuous identification of parameters of the system as neural controllers, etc. The neural networks must be correctly learnt to meet our requirements. There is a great number of learning algorithms. Among the most used learning algorithms belong the Levenberg-Marquardt and backpropagation.