Hybrid Artificial Neural Network-Particle Swarm Optimization For Prediction of Dnbp Polymerization Retarder Usage in Industrial Styrene Monomer Plant

The distillation towers of styrene monomer (sm) plants consume a considerable amount of expensive and toxic 2,4-dinitro-6-sec-butyl phenol (dnbp) as a polymerization retarder. the minimization of the operating cost, as well as preventing environmental pollution, is highly desirable to maximize the profit and have a clean technology. however, it is not easy to predict the actual usage of dnbp in the tower because of the nonlinear behavior of the industrial distillation tower in the polymerization process, and also the inaccuracy of experimental results of the dnbp in outlet products. to overcome these difficulties, a prediction model for determining dnbp consumption using a hybrid model in which the ann in combination with particle swarm optimization (pso) is proposed in this study. moreover, all useful parameters (9 parameters) in different years have been gathered from the industrial dcs system for training ann. after combining pso with ann, the main valid parameters have been filtered. from nine proposed settings, five of them have been selected and used for predicting dnbp consumption in the sm plant. the obtained results showed that the proposed ann-pso hybrid model is a powerful tool for predicting dnbp usage with an average relative error of 9% between technical and calculated hybrid ann-pso model data.