Purpose -The paper aims to present a new concept based on a multi-agent approach in the area of nonlinear model predictive control (MPC) for fast systems. Design/methodology/approach -A contribution to decentralized implementation of MPC is made. The control of the nonlinear system subject to constraints is achieved via a set of actions taken from different agents. The actions are based on an analytical solution and a neural network is used to monitor the closed system using a supervisory loop concept. Findings -The high online computational need to solve an optimal control actions in nonlinear MPC, which results in a non-convex optimization, is compared with the new proposed concept. Simulation results show that this approach has very remarkable performances in time computing and target arrival.Research limitations/implications -In practice, each MPC problem of the individual agent in multi-agent MPC can run in parallel at the same time, instead of in serial, one agent after another. A parallel processor can be useful for real time implementation. However, it is estimated that how much time can be gained by performing the computations in parallel instead of in serial. Practical implications -The proposed concept discussed in the paper has the potential to be applied to systems with rapid dynamics. Originality/value -The multi-agent MPC compares favorably with respect to a numerical optimization routine and also offers a solution for non-convex optimization problems in single-input single-output systems.

In this paper, a new approach based on the multi-agent concept is proposed for model predictive control (MAMPC) of a nonlinear fast dynamic system. This concept achieves a transformation of the centralized MBPC in a decentralized and supervised technique. The control of the nonlinear system subject to constraints is achieved via a set of actions taken from different agents. The actions are based on an analytical solution and a fuzzy supervision is used to monitor the closed system using a supervisory loop concept. These strategies of control permits to unlock the obstacle met by the MBPC dedicated to the oscillating and fast dynamic systems and facilitate their real time implantation. The multi-agent compares favorably with respect to a numerical optimization procedure and offers a solution for non convex optimization. A simulation example shows the fast execution of the proposed concept compared to conventional procedure.