Adaptive-tuning method based on neural networks for PID controllers applied to Passive Optical Networks (PONs)

In this paper we propose a novel tuning technique for PID controllers with the aim to efficiently control network resources and parameters in Passive Optical Networks (PONs). Contrary to other tuning existing methods applied to these networks (Ziegler-Nichols (ZN), genetic algorithms (GA)), our proposal, based on neural networks, makes a real time and an automatic readjustment of the tuning parameters according to the updated network conditions. Therefore, the neural network makes the tuning process online in contrast to ZN and genetic algorithms, in which the tuning process is made offline just before the PON activates the PID controller. Indeed, the PID controller together with the neural network efficiently guarantees minimum bandwidth levels to users depending on their priority profiles. This new tuning technique has been compared with the manual ZN method and with a tuning technique based on a genetic algorithm. Simulation results demonstrated that the neural network efficiently automates the tuning process, achieving higher accuracy than the other methods when guaranteeing the stipulated bandwidth values to different priority profiles. Furthermore, the integration of neural networks in PID systems to manage network resources in PONs has been never considered before.