Integration of PSO-Based Virtual Sensor and PID to Control Benfield Concentration of a Stripper Unit in a Fertilizer Plant

Production process in a fertilizer plant requires some physical variables, which are called primary variables, to be measured indirectly due to the limitation of measuring conditions and methods. Alternatively, virtual sensing system is used to predict and estimate these primary variables by using information from the secondary variables which are relatively easier to be measured. A development of virtual sensing system or virtual sensor to estimate benfield solution concentration of a stripper unit in a fertilizer plant is introduced in this paper. The virtual sensor consisting of a Diagonal Recurrent Neural Network (DRNN) method is used to model the relationship of primary and secondary variables, coupled with the Extended Kalman Filter (EKF) that will estimates the primary variables based on the measurement made by the secondary variables. During the learning process of DRNN, Particle Swarm Optimization (PSO) is proposed, to determine the optimal weights of the network. Using the estimated primary variable, a PID control strategy is designed and optimized by PSO to obtain an optimized controlled response. Using the real-time operational data acquired from the stripper unit of a fertilizer plant, results show that the proposed control strategy is able to control the benfield solution of the stripper unit satisfactorily.