Deep predictive controller designed for centrifugal compressor system anti-surge

Taking the characteristics of surge triggered in centrifugal compressor system as the analysis object, a class of controller termed as depth predictive controller is proposed in this study for anti-surge control in multiple-operational mode scenes, which involves the modular in terms of Depth interval predictor, operational mode discriminator and Dynamic Matrix Control (DMC) with anti-surge expert experience, etc. Specifically, the centrifugal air compressor model is built on the Apros platform, moreover, Long-Short Term Memory (LSTM) under Bootstrap structure is treated as depth interval predictor for providing the tendency perception of load flow of compressed air as well as the corresponding reliability. An extended DMC method by combining the operational mode discriminator and the expert experience of anti-surge control is designed for realizing the anti-surge control of the centrifugal compressor system in real-time. The experimental results show that the proposed method can provide a safe and reliable result to adapt to the anti-surge control of air compressor under different working conditions.