Decentralized Predictive Control of Large Scale Systems Using Neuro-Fuzzy Identifiers for their Interconnections

Abstract This paper proposes an approach for the design of discrete-time decentralized control systems covering not only the case of m-step delay sharing information pattern, but also any general nonclassical information pattern where the non-local information is not spread among the subsystems. It employs the model-based predictive control (MBPC) scheme combined with fuzzy prediction for the interconnections among the subsystems. A state space model is used at each control station to predict the corresponding subsystem output over a long-range time period. The interaction trajectories are considered to be non-linear functions of the states of the subsystems. In all cases, the interconnections and the necessary predictions for them are estimated by an appropriate neuro-fuzzy identifier trained on-line using the back-propagation training algorithm. Representative computer simulation results are provided and compared for nontrivial example systems.