ENSURING FUNCTIONAL RESILIENCE OF A SOFTWARE-DEFINED COMPUTER NETWORK BASED ON AUTOREGRESSIVE MACHINE LEARNING METHODS

DOI: 10.31673/2412-4338.2025.048904

Abstract

The article studies the problem of ensuring the functional resilience of software-defined computer networks (SDN) in conditions of stochastic fluctuations in traffic and dynamic changes in network flows. Existing approaches to increasing the fault tolerance of SDN are analyzed, in particular, redundancy methods, multi-controller architectures, hybrid protection and recovery models, as well as machine learning algorithms for load forecasting. It is shown that most classical solutions are focused on static controller models and do not take into account short-term traffic fluctuations, which leads to delays and packet loss in critical network sections. The aim of the study is to develop a method for increasing the resilience of SDN based on autoregressive machine learning approaches taking into account the statistical properties of traffic. The paper proposes an improved model for predicting load parameters, which combines classical AR models with the method of decomposition of a random process according to the Karunen-Loev expansion scheme. This approach provides a more accurate reproduction of traffic dynamics and allows you to detect moments of approaching the limit modes of operation of SDN nodes. The simulation results confirm the effectiveness of the developed method: the normalized mean prediction error (NRMSE) does not exceed 2–3% even in the case of sharp load spikes, while for classical methods it reaches 25–40%. The proposed approach can be used for adaptive load balancing between controllers, timely response to overload and increasing the level of quality of service (QoS). Thus, the results obtained create a scientific basis for building intelligent SDN control systems with increased functional resilience.

Keywords: Software-Defined Networking (SDN), functional resilience, autoregression, machine learning, Karunen-Loev Expansion, individual forecasting, traffic anomalies, intelligent analysis methods.