Identification and Process Control for MISO systems, with Artificial Neural Networks and PID Controller

Industrial processes with multiple input and single manipulated variables are very complex systems to control in automatic models. Such is the case with processes related to gases extraction or transport phenomena. The present research is focused on the development of a control algorithm (automatic control strategy), based on artificial neural networks, to identify an industrial process by using process historical records, as well as knowledge from the operation itself. The output of the identification stage feeds a classic PID controller to perform control actions (hybrid controller). Here, an actuator or final control element is modeled, estimating its space-state dynamic equation. With the estimated model, a local control loop is conformed controlling the main process or manipulated variable. For this, the process of gases transport in a copper smelter plant was chosen, where the necessary data and scenarios for the proposed control algorithm testing was obtained. This application attempts to present a solution to problems inherent to manual control, multiple key variables coexisting in a system, mechanical stress in equipment due to manual actions, etc. The control strategy is based on a computer simulation made with real process data, showing improvement of the transient periods in the final actuators due to control signals, as well as establishing that these kinds of technologies could be implemented in both, an existing plant hardware/software or in a conventional control system.