HYBRID ESTIMATION BETWEEN SLIDING MODE OBSERVER ANDNEURAL NETWORKS ESTIMATOR FOR GAS-PHASE ETHYLENE POLYMERIZATION PROCESS

Abstract

Estimation of an ethylene concentration of gas-phase ethylene polymerization process has been known to be a difficult task because of the complex and high nonlinearity of the reactions. Besides, this process operates in a narrow temperature range, so it is to be prone to unstable behavior and runaway easily. In the past research, estimator usage produced unsatisfactory results, therefore hybrid approached is the best solution. Hybrid estimator is combination of two estimators to enhance the estimator’s performance and overcoming their limitations. This research work proposes a hybrid estimator which combined between sliding mode observer (SMO) and neural network (NN) estimator to estimate the ethylene concentration. Initially, the SMO is provided to estimate of all state variables. However, it has usually been prone to the error of the estimation between state variables and actual data. Then, the NN estimator is used to provide the estimates the ethylene concentration again for reducing the error value of SMO. Performance of the SMO-NN hybrid estimator has been compared with the SMO and NN estimator in normal, noise and various disturbance conditions. Simulation results have shown the SMO-NN hybrid estimator is the best approached in estimating the ethylene concentration and provide good accuracy and able to handle noise compared with single estimator.