Adaptive neural network control of active filters

Abstract

This paper presents a new control design of shunt and series active filters for harmonic compensation in a power distribution system. The topology of the active filter is based on a 3-phase pulse width modulated (PWM) voltage source inverter (VSI) and a control circuit. The control circuit using two neural network controllers is proposed. One adaptive neural network (NN) controller is designed to estimate the harmonic components of the distorted load current and supply voltage. Another neural network controller that uses Levenberg–Marquardt backpropagation (LMBP) for its training is designed to generate accurate switch control signals for the PWM VSI. The resultant series active filter also has the capability to compensate for voltage sags in the distorted supply voltage. A power factor correction function is incorporated in the shunt active filter to achieve a power factor that is near to unity. Simulation results show that the active filters with neural network control can significantly reduce harmonic distortion in the load current and supply voltage and can adapt to variations in system operating conditions.