The design of robust MIMO neural network disturbance observer for multi-variable system

Multi-variable systems widely exist in the practical engineering control systems whose performances are always severely interrupted by strong disturbances including unmodeled dynamics, parameter variations, couplings and external disturbances. Disturbance observer (DOB) is known as an effective technique to estimate disturbances and has been extensively applied for feed-forward compensation design in the presence of disturbances. Yet many disturbance observer techniques in previous literature are just used for single-input-single-output (SISO) systems or the DOBs can be applied in the multi-variable systems, but the DOBs are still SISO DOBs. A decoupled robust multi-input-multi-output neural network disturbance observer (MNNDOB) is designed for the multi-input-multi-output (MIMO) systems. Simulation results on the mixing tank show that the proposed method has better disturbance estimation performance when there are severe model mismatches compared with the MIMO linear disturbance observer.