Neural network model as Internet of Things congestion control using PID controller and immune-hill-climbing algorithm

Abstract

Congestion control of traffic flow in the Internet of Things (IoT) and fog computing is very essential to achieve certain Quality of Service (QoS) in critical applications. The huge data and the limited resources in IoT and cloud system ascertain the real need for efficient congestion control scheme especially in real-time fog computing applications. In this paper, a Proportional-Integrator-Differentiator (PID) controller is used and tuned in a rate-based congestion control scheme to be implemented for IoT applications data communication. The tuning strategy is based on reformulating the problem to be as an optimization problem with a compact fitness function reflecting the control requirements. The main contribution of this paper is to adopt an enhanced hybrid immune-hill-climbing immune algorithm as the tuning process. Then, an Artificial Neural Network (ANN) is built and trained in order to be used instead of the repetitive and exhausting retuning. The proposed method works in a cascaded manner; the immune algorithm is used for coarse tuning and then the hill climbing algorithm is triggered for fine-tuning. For the purpose of comparison, a simplified network model is built and simulated using two control methods, namely the control scheme and the well-known single-bit indicator scheme. The experimental results prove that the PID controller tuned by the immune-hill-climbing algorithm is superior in terms of the stability of both, buffer occupancy in the switch and the source rate. Both the transient behavior and the steady-state response for the PID controller are shown to be much better than the single-bit indicator scheme. Consequently, the proposed control scheme outperforms the single-bit indicator in terms of link utilization, buffer utilization, packet drop and robustness against any burst change in fog and cloud computing systems. It is shown that using trained ANN instead of the repetitive and exhausting retuning is improving the response time.