A class of model reference adaptive decouple control based on RBF neural network in deaerator system

Difficulties in water level control of deaerator are caused by strong couples between water level of deaerator and pressure in outlet of condenser pump, which causes unavailability of automatic control with traditional control algorithms. A kind of general-purposed decoupling and control algorithm for time-variant MIMO system with strong coupling is proposed in this paper. Model reference adaptive control (MRAC) and decouple control are combined together in the proposed control algorithm. Using the arbitrary non-linear approximation ability of RBF neural network, RBF neural network controller (RBF-NNC) is designed. The linking weights between hidden layer and output layer are modified with gradient descent algorithm. Pattern concept and its related learning mechanism in neural network off-line learning is introduced into online self-learning algorithm for RBF neural network and 2 learning methods based on pattern concept are presented. RBF neural network identifier (RBF-NNI) is introduced to acquire the controlled object related information in the online self-learning process of RBF-NNC. Online optimization algorithm for self-learning rate in the modification of linking weights in RBF-NNC and RBF-NNI and implementation process for the complete control algorithm is given. Simulation experiments for the deaerator MIMO system are performed. Comparison of simulation results from the proposed control algorithm with that from other 2 widely used algorithms shows that desirable effects in decouple and control are achieved and much better with the proposed control algorithm. Meantime, relatively good real-time performance is achieved as well.