Decoupling control of double-level dynamic vacuum system based on neural networks and prediction principle

Double-level dynamic vacuum (DDV) systems have great value in meteorology and environment fields. Overcoming the coupling impact is the premise to achieve rapid, precise, and, especially, independent control of the vacuum pressures in two concatenate chambers of the DDV systems. A decoupling method based on neural networks and prediction principle is presented due to the nonlinearity, dead-time, and strong coupling characteristics of the system. With the aid of neural networks and the prediction principle, first, the air flow rate estimation model and the vacuum pressure prediction model are developed respectively; then, the predictive expressions of disturbances between the upstream and downstream vacuum pressures are obtained by the ideal gas equation and by the models mentioned above. Thereby, the controller outputs can be compensated properly in advance. Results show the settling time of the upstream and downstream pressure is reduced to about 87 from 93 and to 60 from 76 s, respectively; most importantly, one vacuum pressure of the DDV system doesn’t fluctuate along the other with time lag. For a typical application, the settling time is reduced to about 86 from 119 and to 42 from 107 s. It implies this method is of great importance to improve the experimental efficiency.

► The air flow rate estimation model is developed based on neural networks.
► The vacuum pressure prediction model is built by the prediction principle.
► The predictive expressions of disturbances are gotten by models and equation.
► The strongly coupled DDV systems are decoupled into two SISO systems.