Game Based Pigeon-Inspired Optimization with Repository Assistance for Stochastic Optimizations with Uncertain Infeasible Search Regions

Evolutionary algorithms have been widely used to solve stochastic optimization problems (SOP) in recent studies and their performances are largely determined by the reliabilty of evaluations as well as the exploration and exploitation behaviors. However, due to the black-box features and physical limitations to the research objects, evaluating most modern systems in real world may be expensive or even suffer from large numbers of infeasible solutions and high proportions of uncertain infeasible search regions (ISR), which arouses the expectations for less trials and errors during the entire optimization process. This article presents two variants of the pigeon-inspired optimization (PIO) algorithm respectively for the stochastic optimizations with and without ISR penalties. Both variants use a repository to maintain the historical individuals with local optima and a multi-strategy evolutionary game mechanism to promote the balances between explorations and exploitations in the corresponding search space. The second variant also uses the repository to reallocate infeasible solutions into feasible regions and motivate the pigeons moving faster to reduce the unneccessary trials during the optimizations. The performances of the first PIO variant have been evaluated on CEC'2017 benchmark functions, while a simple example of vehicular control is provided to demonstrate the effectiveness of the second variant on the SOP solving with minimal ISR touches.