Modeling study of nonlinear dynamic soft sensors and robust parameter identification using swarm intelligent optimization CS-NLJ

In order to give a sufficient description of system characteristics, this paper proposes a Wiener model for nonlinear dynamic soft sensors. Based on the model, an improved swarm intelligent identification algorithm is proposed by combining the global cuckoo search (CS) and the local new Luus-Jaakola (NLJ) optimization. The Levy flight in the CS helps the search escape from local optima, and the NLJ search achieves the exploitation of local areas. Thus, a good balance is guaranteed between algorithmic exploration and exploitation. Because of the strong search ability of CS-NLJ algorithm, the approximation accuracy of identification can be improved in the presence of stochastic Gaussian or heavy-tailed noises. Further, the robust identification for dual-rate soft sensors is achieved by using the proposed model and the proposed identification algorithm. Finally, the simulation results verify the theoretical findings.