This paper explores the principles of neural network hybridization and immune algorithms, which allows us to create new types of information processing models that have a higher quality of problem solving while reducing computational training costs. The use of hybrid neural networks for solving time series prediction is considered.