As the key part of a fluid catalytic cracking unit (FCCU), the control of the reactor-regenerator system is very important to the FCCU process. For this purpose, we adopt three multiple inputs/single output Elman neural networks (ENNs) for modeling the reactor-regenerator system. To determine the weight parameters of ENN models optimally, we propose the splicing process inspired cuckoo search algorithm (spCS), in which the new RNA crossover operator is designed to increase the population diversity. And a modified selection strategy based on Euclidean distances and fitness values is presented for the selection of dissimilarity individuals. The searching performance of the spCS is demonstrated by the numeral experiments and comparisons with CS and ACS. Then, spCS and CS are compared to optimize the parameters of ENNs. The simulation results validate that the spCS based ENN model outputs are in better agreement with experimental data in modeling the FCCU reactor-regenerator system.