Name
Lubentsova Elena Valeryevna
Scholastic degree
•
Academic rank
associated professor
Honorary rank
—
Organization, job position
Kuban State Technological University
Web site url
—
Articles count: 2
ANALYSIS OF QUALITY INDICATORS OF AUTOMATIC CONTROL SYSTEMS WITH NONLINEAR APPROXIMATION CONTROL LAW
The subject of research of this work was the study of
the quality of control processes in a nonlinear
automatic control system with an approximating the
control law. In the known published works there are
no results of such studies, which makes it difficult to
synthesis a nonlinear control system for multimode
objects in applied biotechnology, including
technological objects of the agro-industrial complex.
A comparative analysis of the quality of regulation in
the transient and steady-state regimes is carried out. It
is shown that the approximation method used for the
synthesis of the nonlinear control law provides a
linear dependencies in steady-state and close to them
modes in combination with relay modes in transient
regimes, which is a positive factor for improving the
quality of regulation in multimode control systems. It
does not necessary to determine the moments of
switching the dependencies in the control law when
changing modes
The subject of study of this work was learning algorithm of neuro-fuzzy systems with different membership functions. In the prior works there are no published studies of such studies, making it difficult synthesis of neuro-fuzzy control system with new objects in the application of biotechnology, including technological agribusiness entities. A comparative analysis of learning algorithms of neuro-fuzzy system with different membership functions using the method of error back propagation and а hybrid method. For this we used a training sample that contains data of temperature and concentration of dissolved gas in the culture liquid: oxygen (pO2), carbon dioxide (pCO2) of a biotechnological process. It is shown that the hybrid method carries out training of a neural network for the number of periods is 23 times smaller than the algorithm back-propagation errors. The studies found that the two-sided Gaussian membership function provides the smallest learning error of the network δ equal of 3,28•10–3, compared to the other, giving the largest error of training the neural network δ=0,138. Therefore, the task of running the fermentation process effective is the use a hybrid method of education and two-sided Gaussian membership functions. According to the research, we can conclude that for the adaptation of neuro-fuzzy network ANFIS and fuzzy inference system Sugeno zero order to solve biotechnological process control tasks microbiological production efficiency is to use a hybrid method of education and bilateral Gaussian membership functions