| In the past two decades, model predictive control (MPC) technology has gained its industrial position in refinery and petrochemical industries and is beginning to attract interest from other process industries. Model development remains by far the most critical and time-consuming step in implementing industrial MPC. Conventionally, models are identified through a series of single variable open loop step tests. Overtime these models change due to a number of reasons such as change in operating conditions, instability of the process, drift in process and environmental conditions. This causes control performance problems. For this reason, multivariable closed loop identification is proposed for MPC Identification techniques based on autoregressive moving average (ARMA), state space and neural network models are investigated. These techniques are tested through computer simulations and ultimately on field data from an industrial process. Direct Identification method is employed for this purpose. |
