An ENA-based strategy replacing subobjectives definition in incremental learning

Incremental evolution has proved to be extremely useful in complex process control. The need to define manually the subobjectives at each stage constitutes the method weakest point, hindering its generalization. We propose the application of evolving neural arrays (ENA) in order to implement incremental evolution (without the explicit explanation of subobjectives) applicable to a large set of process control problems. The measures carried out show the advantage of evolving neural arrays over traditional methods handling neural network populations. SANE has been particularly used as comparing reference for its high throughput.