In summary, this thesis differs from past works in the following manners:
1. Past works on the MLC problem ensure all targets are monitored by at least one sensor node throughout a WSN’s lifetime. The key limitation is that the network lifetime is restricted by sensor nodes’ finite battery capacity, and sen- sor nodes do not have the ability to recharge their battery. On the other hand, the duty cycling algorithms in energy harvesting WSNs mainly focus on max- imizing events detection probability or network coverage level. They do not consider the significance of complete targets coverage. Moreover, past works do not consider the recharging opportunities of sensor nodes. To fulfill these gaps, this thesis considers the MLCEH problem with an objective to maximize network lifetime using energy harvesting sensor nodes. The key constraints in- clude complete targets coverage, energy and recharging opportunities.
2. Past centralized coverage algorithms assume the base station/sink knows the exact battery level information of sensor nodes. However, this is not valid in practice because sensor nodes have a random recharging rate, and it is impractical for the sink to know the exact battery level of each sensor node. To this end, this thesis proposes to use stochastic programming [71] to cope
with the uncertainty battery levels of sensor nodes.
3. Past works have not considered designing a distributed algorithm for complete targets coverage using sensor nodes equipped with energy harvesting capabil- ity.
4. Similar to the MLC problem, past works on the MLCC problem do not consider the recharging opportunities of sensor nodes. To fulfill this gap, this thesis proposes a MLCCEH sub-problem to ensure all sensor nodes in the active state cover all targets whilst ensuring there is at least one path to the base station/sink.
5. Existing works on conventional WSNs node placement problem do not con- sider the energy harvesting capability of sensor nodes. Moreover, the works on energy harvesting WSNs neglect the importance of energy neutral operation. Thus, to fulfill this gap, this thesis considers the MEHNP-PC and MEHNP- ENCC problem to ensure energy neutral operation and complete targets cov- erage.
Chapter
3
Novel Algorithms for Complete Targets
Coverage
This chapter addresses the MLC problem in the context of energy harvesting WSNs. The goal is to maximize a WSN’s lifetime whilst ensuring all targets are monitored by at least one sensor node at all times. This complete targets coverage problem, however, has only been studied in conventional or non-rechargeable WSNs; see Chap- ter 2. This means they do not consider the energy harvesting capability of sensor nodes. On the other hand, existing works that solve the coverage problem in en- ergy harvesting WSNs have only focused on maximizing the coverage probability of targets by using duty cycling [53] or prediction techniques [26]. Thus, they do not consider continuous monitoring of targets.
To address this research gap, this chapter considers the MLCEH problem where the aim is to schedule the active and sleep time of energy harvesting sensor nodes such that all targets are completely covered at all times for the longest time. To this end, two algorithms are proposed. The first one, called LP-MLCEH, relies on a LP solver. Its objective is to maximize network lifetime subject to complete targets coverage and energy constraints. Numerical results show that LP-MLCEH
doubles network lifetime when compared to similar algorithms developed for finite battery WSNs. However, it incurs high computational cost due to multiple calls to a LP solver. To this end, this chapter proposes the Maximum Utility Algorithm (MUA). Experiment results show that MUA achieves 34 of the network lifetime of LP-MLCEH.
3.1
Network Model
This chapter considers a WSN modeled as a sensor-target bipartite graph (S, Z, E, W ), where S is the set of sensors, Z is the set of targets, and E is the set of edges con- necting a sensor si ∈ S to one or more targets in Z. Note, si and zj index sensors
and targets, where i = 1 . . . |S| and j = 1 . . . |Z|. Lastly, wij ∈ W is an edge weight
that represents the residual active time of sensor node si with respect to target zj.
Let Ei (Joules) be the current energy of sensor node si, which is bounded by
the battery capacity Bmax. In addition, it has a recharging rate of Eir (Joule/s).
Also, each sensor consumes Ec
i (Joule/s) when active. Let Z(si) be a function that
returns the set of targets covered by sensor si; i.e., sensor si covers |Z(si)| targets.
Conversely, S(zj) is a function that returns the set of sensors covering target zj.
Assume that time is divided into unequal time slots. Define Ct ⊆ S to be the set
cover at time slot t that is monitoring at least one target. Let Z(Ct) return the
set of targets covered by sensor nodes in the set cover Ct. With a slight abuse of
notation, let W (Ct) and W∗(Ct) return the set of weights for sensor nodes in and
outside of set cover Ct respectively. Let φ(Ct, zj) be a coverage mapping function
that returns one if target zj is covered by Ct. Otherwise it returns zero. Also, E(Ct)
is an indicator function that returns one if the residual energy of all sensors in Ct is
sufficient to provide cover throughout time slot t.
Here, an epoch, denoted as δt, is a time instant defined as one of the following: (i)
when a target is not monitored by any sensor node, or (ii) when an in-active sensor node has a full battery. In the first case, any developed algorithm needs to activate
another set cover to monitor all targets. In the second case, these algorithms need to take into account sensor nodes with a full battery such that they are used to monitor targets whilst affording other nodes recharging opportunities. This means a time slot t is the duration between epoch δt and δt+1.