Neuro-control system for converter based electrical energy source - Test performed in laboratory setup with combustion engine emulator

Abstract

The paper investigates possible advantages from using nonlinear adaptive ANN (Artificial Neural Network)-based controller in a control system of autonomous variable speed electrical energy source with internal combustion engine. The speed is adjusted automatically as a function of load power demand. When the system is in a light or no load condition, the Main Voltage Controller automatically reduces the engine speed in order to reduce the fuel consumption, environmental noise and mechanical wear of engine parts. Optimization of the controller is difficult due to the non-linearity and non-stationarity of the plant. The structure of Main Voltage Controller proposed in this paper employs one hidden layer artificial neural network to estimate the unknown plant nonlinearities on-line. ANN serves as a speed compensator and does not need a process model to predict future performance. To increase the stability and convergence of the algorithm, the Resilient backpropagation (Rprop) adaptive learning scheme has been employed. The presented solution allows maintaining suitable efficiency at steady state and adequate transient performance. The proposed neuro-control system have been widely tested in Matlab/Simulink environment. In addition experimental test has been perform in the laboratory setup where internal combustion engine was emulated by using PMSM drive. Obtained test results have been presented to show effectiveness of proposed neural control system.