Deep Learning Traffic Prediction and Resource Management for 5G RAN Slicing

The future development of wireless mobile networks (5G and beyond) aims to create a service-aware network using the network slicing technique. These networks are anticipated to handle substantial amounts of traffic and offer a wide range of services tailored to meet consumer needs. The use cases, such as Enhanced Mobile Broadband (eMBB), Ultra-Reliable Low Latency Communication (uRLLC), and Massive Machine Type Communication (mMTC), have diverse service needs that depend on the applications they support. Based on this premise, the study suggests formulating a multi-objective problem to precisely forecast traffic using the Deep Convolution Neural Network and Long Short-term Memory Networks predictive models. The efficacy of this Deep Reinforcement Learning model is evaluated using learning algorithms such as Adaptive Moment Estimation (ADAM), Stochastic Gradient Descent Momentum, and Root Mean Squared Propagation (RMSProp). The findings display the comparison of evaluation measures such as precision, recall, F1-score, and accuracy. Additionally, this projected data is utilized as the traffic for the subsequent resource allocation model. The second model employs the advanced Tasmanian Devil Optimization-Elite (TDO-E) method to allocate resources effectively. This study presents the initial application of the TDO algorithm in the 5G slicing technique for resource allocation. The TDO-E is an advanced version that utilizes a revolutionary way to achieve optimized results by implementing a strategy of sharing information. This resource allocation technique focuses on two main issues: (i) Creating and assigning slices to new devices. (ii) Implementing slice reconfiguration to optimize resource use. We have shown that the predictive model utilizing the ADAM algorithm has achieved superior outcomes compared to other algorithms, with a corresponding output value of 99.785% in the evaluation metrics. The resource allocation for devices with TDO-E shows a 6.79% improvement compared to TDO. The TDO-E evaluates the techniques in terms of resource efficiency and adaptability. The proposed approach is also evaluated using different numbers of devices. The results demonstrate the efficacy of the planned approach by comparing graphs depicting the percentage of resources provided in relation to the number of devices. The implementation of predictive models with slice reconfiguration will significantly advantage network operators by optimizing resource allocation and enhancing service quality for customers.