Dragonfly visual multipath peak evolutionary neural network for solving constrained optimization problems

Based on the information processing mechanism of the dragonfly visual system and population evolution concepts, this paper explores the Dragonfly Visual Multi-Path Spiking Evolutionary Neural Network (DVMPPENN) for solving strongly nonlinear constrained optimization problems. In the model design, a state matrix corresponding to candidate solutions and a function value matrix are used as inputs. The state update strategy is established based on the neural response characteristics of the Dragonfly Small Target Motion Detection (STMD) neurons and Wide-Field Motion Detection (WFMD) neurons, resulting in a visual evolutionary neural network with only two adjustable parameters. The computational complexity of this network is determined solely by its input resolution and leverages the visual neural network to handle the constraints of optimization problems. Comparative numerical experiments demonstrate that this neural network has a significant advantage in solving strongly nonlinear constrained optimization problems and provides important insights for solving engineering optimization problems.