Energy Resource Allocation for Green FiWi Network Using Ensemble Learning

In this paper, we utilize a random forest regression based ensemble learning to effectively predict the solar power available for the fiber-wireless (FiWi) network components, such as optical network units (ONUs) and access points (APs) which is collectively known ONU-AP. Thereafter, a joint energy resource allocation framework is proposed to minimize the required number of photovoltaic (PV) panels and batteries. To solve the joint energy resource allocation problem, we divide it into two sub-problems, minimum PV panel allocation for a fixed number of batteries and minimum battery allocation for a fixed number of PV panels. The two sub-problems are further solved using the proposed
MinBatAlloc
and
MinPVAlloc
algorithms. Moreover, we introduce a system parameter
$\alpha $
, that signifies the ratio between solar power supplied to operate ONU-AP and to charge the batteries. The results are shown by varying
$\alpha $
and its impact on the energy resource allocation and battery lifetime. We compare the performance of our proposed approach with non-ML based approaches, such as, maximum, minimum, median, and outage threshold based energy resource allocation. Through the obtained results it has been shown that the proposed approach considerably improves the performance in terms of outage, lifetime, carbon dioxide emissions, and cost.