Гібридний метод маршрутизації задач у хмарному середовищі з використанням нейромережевого прогнозування та генетичної оптимізації

The article provides a comprehensive study of the problem of effective routing support for cloud testing and proposes a scientifically sound hybrid method that combines predictive capabilities of artificial neural networks with the global search capabilities of genetic algorithms. As part of the development, the method was implemented in a software environment based on the TensorFlow framework for building a model for predicting information flows and the DEAP library for evolutionary optimization of routes, taking into account multi-criteria constraints. The proposed approach synergistically evaluates and minimizes waiting times for tasks, energy consumption, and server thermal loads, which is particularly relevant for modern high-load cloud infrastructures. Experimental modeling conducted in a series of scenarios with different queue profiles and peak loads demonstrated reduction in average request processing time by up to 52%, a decrease in server equipment energy consumption by up to 30%, and optimization of the thermal regime by 40% compared to traditional routing algorithms, con-firming the practical feasibility of hybridizing machine learning methods and evolutionary strategies. The results obtained demonstrate the ability of the proposed method to ensure the stability of server infrastructure under conditions of dynamic load changes, increasing the reliability and quality of cloud services. Practical implementation is recommended for corporate cloud systems and large data centers, for which adaptive resource management and reduction of total operating costs are important. At the same time, the phased integration of the solution with existing routing tools ensures service continuity and simplifies parametric configuration. Further research may focus on the application of more complex neural network architectures, such as LSTM or Transformer, expanding the domain of application to Internet of Things tasks and high-load web platforms, as well as studying the behavioral characteristics of the method in terms of fault tolerance and extreme peak loads. The presented hybrid approach forms the basis for the creation of new generations of intelligent cloud infrastructure management systems and opens up prospects for further improvement of optimization and forecasting technologies in the field of cloud testing.