A self-adaptive evolutionary algorithm called SaAEA algorithm was proposed. SaAEA algorithm evolved between two levels

that is

the individuals evolved by AEA (Alopex-based evolutionary algorithm) algorithm and the parameters evolved by PSO (particle swarm optimization) algorithm

and eventually made the algorithm parameters achieve self-adaptive adjustment with the population evolution. At the same time

the crossover strategy used in differential evolution algorithm was introduced in AEA algorithm

to alleviate the problem of premature convergence and achieve population diversity

and overcome the disadvantage of falling into local optimum in AEA algorithm. The SaAEA algorithm was tested on 14 benchmark functions

and simulation results demonstrated that the performance of the improved algorithm was greatly improved comparing with the basic AEA algorithm. The new algorithm maintained the population diversity effectively. The solution quality and convergence rate were significantly improved. Finally

the new algorithm was used for the neural network soft sensor modeling for ethylene cracking severity

and a good result was achieved.