Section: New Results

Servicing

Participants : Xu Li, Kalypso Magklara, Nathalie Mitton, Tahiry Razafindralambo, Dimitris Zorbas.

Servicing wireless sensor networks include many primitives. It can range from cloud connection [12] to mobile IPv6 management [29] going through energy prediction [20] and launching mobile robots on request of a specific demand [5] or to reload sensors [23] , [17] .

Node reloading

A critical problem of wireless sensor networks is the network lifetime, due to the device's limited battery lifetime. The nodes are randomly deployed in the field and the system has no previous knowledge of their position. To tackle this problem, in [23] , we use a mobile robot, that discovers the nodes around it and replaces the active nodes, whose energy is drained, by fully charged inactive nodes. We propose two localized algorithms, that can run on the robot and that decide, which nodes to replace. We simulate our algorithms and our findings show that all nodes that fail are replaced in a short period of time.

In [17] we focus on an emerging kind of cooperative networking system in which a small team of robotic agents lies at a base station. Their mission is to service an already-deployed WSN by periodically replacing all damaged sensors in the field with passive, spare ones so as to preserve the existing network coverage. This novel application scenario is here baptized as "multiple-carrier coverage repair" (MC2R) and modeled as a new generalization of the vehicle routing problem. A hybrid metaheuristic algorithm is put forward to derive nearly-optimal sensor replacement trajectories for the robotic fleet in a short running time. The composite scheme relies on a swarm of artificial fireflies in which each individual follows the exploratory principles featured by Harmony Search. Infeasible candidate solutions are gradually driven into feasibility under the influence of a weak Pareto dominance relationship. A repair heuristic is finally applied to yield a full-blown solution. To the best of our knowledge, our scheme is the first one in literature that tackles MC2R instances. Empirical results indicate that promising solutions can be achieved in a limited time span.

Energy prediction

One way to improve energy supply for sensor nodes is through ambient energy harvesting from solar, thermal or vibration energy sources coupled with rechargeable energy storage. Wireless sensors have to adapt to the stochastic nature of the energy harvesting sources. We are convinced that predicting the temporal availability of ambient energy resources is vital to plan the harvesting efficiency, optimum resource utilization and energy conservation within sensor nodes. In [20] we propose a novel two stage Autoregressive Weather conditioned Solar Energy Prediction (AWSEP) model which is characterized by low computational complexity and is used to accurately estimate the amount of solar energy that will be harvested in the near future in a particular region. Our algorithm re-learns the model parameters during the prediction processing situations where the prediction error becomes larger than a predefined prediction error threshold mainly because of the unreliable nature of outdoor solar energy sources caused by changes in weather conditions. The proposed AWSEP model performance is evaluated by varying energy harvesting source prediction intervals, sampling rates, trade-offs in prediction accuracy and computational costs using real solar datasets. We concluded that AWSEP algorithm is more accurate, has reduced computational complexity and memory utilization than other prediction schemes in literature. Our proposed algorithm can assist a node to automatically adapt to the changing weather conditions for effective power management and sensing task scheduling.

Servicing sensor nodes

Due to the robots' potential to unleash a wider set of networking means and thus augment the network performance, WSRNs have rapidly become a hot research area. In [5] , we elaborate on WSRNs from two unique standpoints: robot task allocation and robot task fulfillment. The former deals with robots cooperatively deciding on the set of tasks to be individually carried out to achieve a desired goal; the latter enables robots to fulfill the assigned tasks through intelligent mobility scheduling.