Investigation of the use of neural networks in active noise control systems with nonlinear processes

The project consisted in carrying out simulation tests related to using neural networks of various structure and training algorithm in active noise control systems. New algorithms for training neural networks - QNBP, QNBPM and PYKQNBPM - were developed on the basis of the results of simulation tests and an analysis of strategies of control and identification of dynamical non-linear systems. The training algorithms were verified with simulation tests in MATLAB. A numerical model of an active noise control system was developed with that aim in mind. In this model the primary path was non-linear. A new control unit/platform of an active noise control system was also developed on the basis of a floating point signal processor. This system was used to verify the developed training algorithms in laboratory conditions. Software was developed for the control system; Software consists of neural networks and training algorithms. The test stand was modified, too; an acoustic waveguide was its principal element. An additional module of the waveguide was developed and produced; it makes non-linear phenomena in the primary path possible.The final stage of the project primarily consisted in testing the developed algorithms and comparing them with existing algorithms regarding their computational complexity and effectiveness.