Soft sensing modeling via artificial neural network based on PSO-Alopex

In this paper, algorithm of pattern extraction (Alopex) is introduced into the particle swarm optimization (PSO) to train the artificial neural network (ANN), which is used to construct the soft sensing model. PSO has some significant features such as simpler expression, less parameters and easier operation, but it is easily to run into the local optima. Alopex generates ´noises´ randomly to get rid of local optima, which ensures the PSO to converge the global optimum. Two benchmark functions test show that the improved algorithm is effectiveness. At last, the soft sensor is applied to estimate ethane concentration and ethylene concentration in ethylene distillation column. Experimental results show that the soft sensors based on combining ANN and PSO-Alopex overcome the existent problems in actual process application, which on-line estimations are suitable for control purposes.