METHOD FOR FORECASTING LOAD AND RESOURCE ALLOCATION USING NEURAL NETWORKS IN TELECOMMUNICATION SYSTEMS

Abstract

This paper presents a neural network–based method for load forecasting and resource allocation in modern telecommunication systems, focusing on improving the accuracy and adaptability of network resource management under dynamic and heterogeneous traffic conditions. The proposed approach integrates convolutional (CNN) and recurrent (LSTM) neural network architectures, enabling the model to capture both local short-term patterns and long-term temporal dependencies in network load behavior. Such hybridization addresses one of the key limitations of traditional statistical methods, which often fail to represent nonlinear and rapidly fluctuating traffic characteristics typical for contemporary 5G/6G and IoT-driven infrastructures.

The mathematical model formalizes the forecasting process as an optimization problem, where predicted load values serve as the basis for adaptive resource allocation. This allows network controllers to proactively mitigate overload risks, balance computational and radio resources, and optimize service quality while reducing delays in decision-making. Experimental evaluation using synthetic and real-like traffic traces demonstrated that the model achieves low prediction error and robust stability, particularly during peak load intervals. Comparative analysis confirmed the advantage of the proposed approach in capturing recurrent traffic patterns, smoothing abrupt fluctuations, and providing accurate multi-step forecasts.

The developed method can be integrated into SDN/NFV-oriented architectures, cloud-fog-edge ecosystems, and network slicing environments. Its application enables autonomous and data-driven resource orchestration, enhancing scalability, reliability, and Quality of Service (QoS). The results indicate strong potential for further expansion of the model through applying transformer-based architectures, multi-modal traffic datasets, and reinforcement learning mechanisms for continuous closed-loop optimization. Overall, the proposed method provides an effective foundation for intelligent resource management in future telecommunication networks.