Self-adapting anti-surge intelligence control and numerical simulation of centrifugal compressors based on RBF neural network

Abstract

Anti-surge control of centrifugal compressors is an essential issue for the operation of long-distance natural gas pipeline systems. A suitable controller can make a centrifugal compressor runs smoothly and stably and improve the economy. This work presents a new intelligence control strategy with self-adapting ability. The strategy includes the proportional integral (PI) control self-tuned by radial basis function neural network (RBF-NN), recycle trip control, special derivative control, surge line correction, and asymmetric output of the controller. A hybrid numerical simulation platform is built to validate the anti-surge strategy, and a real centrifugal compressor is simulated. The results show that the strategy makes the anti-surge valve respond quickly, decreases the surge control line’s margin and backflow rate, and improves the economy. In the controller, the special derivative control can make the anti-surge valve open earlier and effectively reduce the fluctuating of inlet flow rate. Aiming at the problem that the gradient descent method is more sensitive to the initial value when solving RBF-NN, a hybrid algorithm of k-means, recursive least square, and gradient descent (KRG algorithm) is proposed. It is successfully applied in the anti-surge controller. Even if the given RBF-NN initial parameters are not good enough, the KRG algorithm illustrates good learning stability and increases the adaptive ability of RBF-NN.