Diagonal recurrent neural network based predictive control for active power filter

A diagonal recurrent neural network based predictive control strategy for active power filter is presented in this paper. In the strategy, diagonal recurrent neural network is employed to predict future harmonic compensating current. In order to make the predictive model compact and accurate, an adaptive dynamic back propagation algorithm is proposed to obtain the optimum number of hidden layer neurons. Based on the model output, branch-and-bound optimization method is adopted, which generates proper gating patterns of the inverter switches to maintain tracking of reference current without time delay. The model predictive algorithm is used in internal model control scheme to compensate for process disturbances, measurement noise and modeling errors. The proposed control strategy is applied to compensate the harmonic produced by the variable non-linear load. Simulation results show the diagonal recurrent neural network based predictive controller gives better harmonic compensation performance than digital adaptive controller.