Structure of the attention-based operation layer, comprising operation layer, attention layer, and a concatenation operation. Each layer generates different weights of attention.