Development of methods and facilities vehicle-programmatic realization of neurocontroller on the basis of the programmable logic integrated circuit for the construction of intellectual control system
1. State registration number of the theme - 0113U000223, NTUU "KPI" - 2617-p.
2. Scientific adviser - d.f PhD., Professor. A.Doroshenko
3. The essence of development, the main results.
Neural network systems belong to the class of intelligent systems and are one of possible variants of adaptive and self-adjusting control systems for complex dynamic objects that can operate under structural, parametric and information uncertainty. These systems are able to control objects of any complexity but have problems with learning and adapting them in real time. Solving these problems is possible through the use of neural network structures of optimal complexity, implemented in hardware with a high speed, concurrent computing, and replacement of neural network learning procedures with their adjustment.
Currently, there is a wide range of hardware implementation of neural network elements in control systems, particularly neurosignal, digital signal and systolic processors, ordered neurochips and PLA structures. Each of them have some advantages and disadvantages that are associated with the cost of implementation, availability of development and programming tools, compatibility with other elements of the system, energy consumption and others. At that point, the most promising tools are PLA structures, that in comparison with other are universal, relatively simple and cheap computing tools, allow organizing parallel computing structures, having great speed and well-designed programming tools.
In this paper was developed the concept and methodology of implementation of neural network control system on a hardware platform PLAs, that implement structural elements of adaptive control systems: the object models, controllers, identifiers of states of the object, filters, etc., as well as concurrent computing procedures for these neural elements and the whole system adjustment by using evolutionary optimization technology. Was conducted the research of PLA hardware cost for implementation of neural network structures with different configurations, was designed and manufactured neurocontroller layout based on PLAs and technical documentation for it.
For the development, design, research, adjustment and implementation of neural network elements and control systems based on them in the work was developed specialized software environment that is based on software packages System Generator for DSP Xilinx, MATLAB and LabVIEW.