Search results

The purpose of this paper is to present a new concept based on a neural network validity approach in the area of multimodel for complex systems.

The multimodel approach was recently developed in order to solve the modeling problems and the control of complex systems. The strategy of this approach coincides with the usual approach of the engineer which consists in subdividing a complex problem to a set of simple, manageable sub‐problems that can be solved separately. However, this approach still faces some problems in design, especially in determining models and in finding the appropriate method of calculating validities.

A novel approach based on neural network validity shows very remarkable performances in multimodel for complex systems.

The validity of each model is based on the convergence of each neural network. For a fast convergence the proposed approach can be online to give a good performance in multimodel representation for system with rapid dynamics.

The proposed concept discussed in the paper has the potential to be applied to complex systems.

The suggested approach is implemented and reviewed with a complex dynamic and fast process compared to the residue approach commonly used in the calculation of validities. The results prove to be satisfactory and show a good accuracy.

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.

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.

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.

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.

The proposed concept discussed in the paper has the potential to be applied to systems with rapid dynamics.

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.