Study on a novel approach to active power filter control using neural network-based harmonic identification scheme

This paper proposes an effective control scheme for three-phase three-wire (3P3W) active power filter using neural-based harmonic identification scheme. To achieve excellent steady state and dynamic response, the feedback control plus feed-forward control structures is utilized in the proposed control algorithms. The steady-state error minimization is achieved by the feedback loop, where the proportional integral regulators were adopted in d-axis and q-axis of the synchronous rotating reference frame synchronized with grid voltages by using the phase-locked loop. The adaptive linear combiners are utilized in the feed-forward loop, which serves the purpose of load disturbance rejection, and it significantly enhances dynamic performance of active power filter (APF). The modeling of the APF is based on the state-space modeling technique and the decoupled state-space model in synchronous d–q frame is also presented. Moreover, the mathematical formulation of the neural harmonic identification scheme and the selection of learning rate are discussed. It is found that the optimal learning rate is a compromise between the steady-state accuracy and the requirement of dynamic response. To verify the theoretical analysis, extensive simulation results are presented under both the balanced and unbalanced load conditions. The validity and effectiveness of the presented scheme is substantially confirmed by the simulation and experimental results, and it can be easily extended to applications of single-phase and three-phase four-wire APFs.