A PSO with quantum infusion algorithm for training Simultaneous Recurrent Neural Networks

Simultaneous recurrent neural network (SRN) is one of the most powerful neural network architectures well suited for estimation and control of complex time varying nonlinear dynamic systems. SRN training is a difficult problem especially if multiple inputs and multiple outputs (MIMO) are involved. Particle swarm optimization with quantum infusion (PSO-QI) is introduced in this paper for training such SRNs. In order to illustrate the capability of the PSO-QI training algorithm, a wide area monitor (WAM) for a power system is developed using a multiple inputs multiple outputs Elman SRN. The SRN estimates speed deviations of four generators in a multimachine power system. Since MIMO structured SRNs are hard to train, a two step approach for training is presented with PSO-QI. The performance of PSO-QI is compared to that of the standard PSO algorithm. Results demonstrate that the SRN trained with the PSO-QI in the two step approach tracks the speed deviations of the generators with the minimum error.