Abstract

The work in this thesis investigates the problem of designing a path for a mobile sink, like an Unpiloted Aerial Vehicle (UAV), to traverse a sensing field and collect data from a network of wireless sensor nodes. This problem is investigated as an optimization problem with a range of different network architectures, different network constraints and different objective functions.The thesis develops a new multi-level optimization framework which provides excellent solutions across a range of different scenarios. This framework starts with a Travelling Salesperson Problem (TSP) tour, generated with a simulated annealing framework. It then uses a gradient-descent style optimization framework to test for new waypoint positions which optimize the objective function, which initially is minimum tour time. Since tour time consists of the travel time (determined by the TSP tour of waypoints) plus additional stopping time for data collection, an inner optimization loop, based on Linear Programming, is used to find a communications schedule for a specific trial tour which minimizes stopping time, and hence minimizes tour time. Unique amongst previously published solutions, this algorithm is able to model radio ranges with data rates that vary with range in arbitrary increments.In particular, three different research questions are posed relating to several different scenarios. High quality, minimum tour time data collection paths are generated for known data loads at sensor nodes, using an algorithm variant called TSP-DC. In comparison with previously published solutions, it is shown to provide superior tour times.Next, a second variant, called TSP-DA is developed to deal with the situation where data loads are unknown when the tour is first developed, and the algorithm needs to recalculate paths when new nodes are encountered. Again results are better than a pre-planned worst case tour, and in most cases are close to the tour time with known loads.The extra cost of re-evaluating the tour is investigated, and if cloud computing support is available, then the recalculation time cost is small compared to the tour time saved.The computational complexity of the algorithms are investigated. Theoretical and experimental analyses show that TSP-DC has O(n2) complexity and TSP-DA has O(n3) complexity.Another algorithm variant, TSP-EM, is developed which minimizes energy consumption of the mobile sink, which gives good results, and is able to automatically adapt to different scenarios, such as the difference between ground robot and UAV energy costs. The final algorithm variant, TSP-ST, was able to optimize the tour path for network architectures where data could be forwarded from leaf nodes to relay nodes to cluster heads. In this case, an added constraint was the amount of energy at each sensor node, which could be different for every node. Again, the algorithm could generate good paths, across a broader range of scenarios than previously published algorithms. For example, with low node energy or low radio range, the sink would visit every node. For high node energy and a fully connected network, the network would forward data through a multi-hop tree to a single cluster head closest to the sink base station.This multi-level optimization framework is a significant new contribution, combining three different optimization methods (simulated annealing, gradient descent, linear programming) to provide a uniquely powerful framework for optimized path planning.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call