Neuron Adaptive and Neural Network Based on Gradient Descent Searching Algorithm for Diagonalization of Relative Gain Sensitivity Matrix Decouple Control for MIMO System

Some currently used algorithms for decoupling and control based on neural network are very complicated and massive online computations and operations are necessary, which causes problems in real-time control and practical implementation. A kind of fault-tolerant decouple and control algorithms for non-linear and time-varying MIMO systems based on neuron adaptive PID and neural network is proposed. Online control is implemented by neuron adaptive PID controller and online decouple is implemented by two-layered neural network based on gradient descent searching algorithms for diagonalization of relative gain sensitivity matrix. Decouple and control is implemented in parallel. Adaptive and self-organizing functions of neuron adaptive PID control is implemented with tuning of weight factors which are modified online with gradient descend algorithm. Stability analysis for the close loop system with neuron adaptive PID controller, online optimization algorithms for proportional factors and self-learning rates in neuron adaptive PID controller are given. A two-layered neural network decoupler is constructed. Self-learning algorithm based on diagonalization of relative gain sensitivity matrix is used in decouple network and gradient descent algorithm is used in self-learning process Real-time simulation results with the proposed decouple and control algorithm and other strategies, are given, analyzed and compared. Simulation results show the proposed algorithms are effective in improving dynamic performance of system and reducing couples between variables greatly. Meanwhile, strong fault-tolerant performance and satisfactory control effects are achieved.