Adaptive hyperoptimization of parameters in population algorithms
Abstract
Adaptive hyperoptimization of parameters in population algorithms
Incoming article date: 21.01.2026The article presents a new architecture for adaptive parameter tuning of population algorithms based on a combination of reinforcement learning, evolutionary selection mechanisms, and a fuzzy logic system. The developed model allows for dynamic optimization of parameters such as mutation rate, crossbreeding probability, and population size, without the need to retrain the algorithm. The practical approbation was carried out on the task of forecasting the time series of urban traffic in Beijing. The results showed an improvement in accuracy compared to modern analogues, confirming the high efficiency of the proposed approach in dynamically changing environments.
Keywords: adaptive parameter setting, reinforcement learning, population algorithms, fuzzy logic, hyperoptimization, time series forecasting