MPC with adaptive-bias RBFNN and Q-reinforcement learning for fault-tolerant control of nonlinear processes

A nonlinear model predictive control (NMPC) with a feedback fault compensator, using an adaptive-bias radial basis function neural network (RBFNN), is proposed to achieve high precision control of nonlinear processes with dynamic modelling deviations or fault disturbances. The stability control is realised by NMPC with prediction optimisation and feedback correction. An RBFNN is constructed to compensate for the lumped deviations caused by modelling errors, fault disturbances and random noise. A particle swarm optimisation algorithm with Q-reinforcement learning (QPSO) is established to obtain the optimal control parameters, and its fitness function is designed by cumulative absolute tracking error (CATE). The closed-loop stability of the proposed algorithm is analysed by constructing a Lyapunov function. The superiority of the proposed control algorithm is verified through comparative testing. The average coupling error (MCE) of the control algorithm proposed is 0.14, which has been reduced by three times compared to other algorithms.