2023 IEEE International Conference on Communications (ICC): IoT and Sensor Networks Symposium

A Node Task Assignment Algorithm for Energy

Harvesting Wireless Multimedia Sensor Networks
Chong Han†,‡ , Leilei Ding† , Lijuan Sun†,‡ and Jian Guo†,‡
† College of Computer, Nanjing University of Posts and Telecommunications, Nanjing 210023, China
‡ Jiangsu High Technology Research Key Laboratory for Wireless Sensor Networks, Nanjing 210023, China

Abstract—By using directional sensor nodes such as cameras, Yang et al. proposed a task assignment algorithm based on link
wireless multimedia sensor networks are typically used in traffic quality between nodes to assign tasks to a pair of cooperative
monitoring, target tracking and other fields. However, most sensor nodes [8]. Raee et al. studied the problem of static
sensor nodes powered by batteries are strictly limited in energy
and cannot achieve long-term frontal tracking and monitoring task assignment in virtual WSNs [9], and on this basis,
of moving objects. To solve these problems, a model of directed proposed a DTA task assignment algorithm for dynamic tasks
sensor nodes with solar energy harvesting is introduced in this [10]. Weikert et al. proposed a multiobjective optimization
ICC 2023 - IEEE International Conference on Communications | 978-1-5386-7462-8/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICC45041.2023.10279261

paper, where the energy of such nodes is no longer limited to algorithm [11] that showed comparable performance in terms
batteries and the directed sensing area enables better frontal of network lifetime while alleviating the network latency.
tracking. On this basis, we propose a distributed algorithm for
directional task assignment called EN-DADA. EN-DADA is a Traditional sensor nodes are powered by non-rechargeable
task assignment algorithm with energy harvesting and orientation lithium batteries, so node energy is regarded as a scarce
awareness, including a task classification phase and node bidding resource. Lin et al. proposed a novel hybrid search and
phase. The task is first classified to determine the candidate node removal strategy [12] for the deployment of wireless chargers
set that can execute the task, and then the task assignment is
determined according to the monitoring income of each node and placing wireless chargers reasonably. Adu-Manu et al.
in the candidate node set. Experimental results show that the summarized the existing energy harvesting technology [13]
proposed task assignment algorithm has advantages in terms of of multimedia wireless sensor networks and focused on solar
task revenue and network lifetime when using the same energy energy harvesting technology and solar energy prediction
harvesting model. methods. Li et al. proposed a priority task scheduling algo-
Index Terms—Wireless multimedia sensor network, task as-
signment algorithm, energy harvesting, directed sensing model. rithm [14] suitable for WSNs on the basis of energy harvesting.
Evangeline et al. proposed quantum-based task assignment
(QBTA) [15], and Zhou et al. modeled energy distribution as a
I. I NTRODUCTION cooperative game among multiple servers and derived a Nash
bargaining solution, based on which they designed a reliability-
Wireless multimedia sensor network (WMSN) is a dis-
A tributed perception network composed of a group of
multimedia sensor nodes with computing, storage and commu-
aware task assignment (RATA) heuristic algorithm [16].These
assignment algorithms prolong the network lifetime, but they
do not take into account the problem of network profit after
nication capabilities that senses various media information in distribution.
the surrounding environment. These multimedia sensor nodes Compared with traditional wireless sensor networks, WM-
are also called directional sensor nodes because they have the SNs focus more on the acquisition and processing of media
directivity of field of view (FoV). Compared with traditional such as images. Therefore, unlike the WSNs studied in [5],
sensor networks, WMSNs have broad application prospects we adopt directed multimedia sensor nodes based on solar en-
in the fields of security monitoring, traffic supervision, and ergy harvesting to perform direction-sensitive task assignment,
environmental monitoring [1]. monitor moving objects as well as possible, and maximize
In recent years, regarded as the infrastructure of Internet network profit on the basis of energy savings.
of Things (IoT) [2], wireless sensor network technology has
The main contributions of this paper are as follows:
developed rapidly [3]. For the problem of limited energy,
Fouad et al. [4] proposed a hybrid sensor selection technique • Task classification is performed, and the nodes that do not
to increase the lifetime of wireless sensor networks (WSNs) meet the energy requirements are put into a dormant state,
by using the optimal number of sensors. The energy harvesting which reduces the energy consumption of the wireless
strategy [5] also was considered to solve the problem of node multimedia sensor network.
overreliance on battery energy in WMSNs. • A bidding algorithm is implemented, which can select
To optimize the performance of the network, task assign- the most suitable node in the wireless multimedia sensor
ment in WMSNs has also received extensive attention. Porta network to perform the task of frontal monitoring of
et al. proposed the task assignment algorithm EN-MASSE moving objects.
for wireless sensor networks [6]. Zhu et al. proposed two • The algorithm improves the energy utilization rate and in-
distributed task assignment methods based on TRETA [7]. creases the profit for direction-sensitivemonitoring tasks.

978-1-5386-7462-8/23/$31.00 ©2023 IEEE 931

Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on March 23,2025 at 13:07:51 UTC from IEEE Xplore. Restrictions apply.
 2023 IEEE International Conference on Communications (ICC): IoT and Sensor Networks Symposium

II. P ROBLEM F ORMULATION sensor node; ~v is the speed of the object to be monitored; ~n is
This section presents a relevant model definition and for- the vector pointing from the sensor position to the particle
malization of the problem. We summarize the definitions of position of the object to be monitored; Tj (xj , yj ) is the
the main notation used in the paper in Table I. position of the object to be monitored, and Si (xi , yi ) is the
sensor location.
TABLE I In this paper, the surveillance mission in monitoring area,
D EFINITIONS OF THE MAIN NOTATION USED IN PAPER . for example, tracking a moving object, is called task. Let
T1 , T2 , . . . , Tm be a set of tasks to be performed by the
Notation Definition
network. A task Tj is a tuple (pj , dj , dirj , vj ), where pj is
S a network composed of senor nodes S1 , S2 , . . . , Sn
the profit that the network can obtain by performing the task;
T the tasks in the network T1 , T2 , . . . , Tm
L the radius of the sensor node’s fan-shaped sensing area
dj is the task requirement; dirj is the direction in which the
θ the angle of the sector area task moves; and vj is the speed at which the task moves.
~
L the direction vector of the sensor node The node is usually in a low-power working state, and de-
~v the motion velocity vector of the task tects whether the moving object reaches its monitoring range.
~n the vector pointing from the sensor to the task If the object is within its monitoring range, the activation task
Tj (xj , yj ) the position of the task Tj to be monitored bidding function module starts to work, converting from a
Si (xi , yi ) the location of sensor node Si sleeping node to an active node. The active nodes form a
pj the actual profit that the network obtains from task Tj candidate set, and select the most suitable node to perform
dj the requirements of task Tj the task. The network scene model is shown in Fig. 2.
dirj the direction in which task Tj moves
vj the speed of task Tj B. Sensor-mission assignment problem
Ej the energy consumption required to perform task Tj
We define pij ∗ as the profit obtained by sensor node Si for
Ef the energy harvested by sensor nodes
performing task Tj . When a node Si is assigned to a task Tj ,
EC the supercapacitor energy
EB the battery power
the time at which the task arrives is tsj , and the time at which
gij the profit that node Si can provide for task Tj it leaves the node sensing area is tej . There are two situations
eij the utility that node Si can provide for task Tj during this time.
pij the actual profit of node Si for task Tj The first case is that the energy consumption Ej required
Pj the ideal maximum profit of the task Tj by the current task Tj is less than or equal to the energy Ef
wa , wb , wc the impact weight of utility harvested by the current node Si . In this case, the node directly
pjt the profit of task Tj at time t uses the harvested energy to satisfy the operation of the node
pij ∗ the profit after adding task classification weights and stores the excess energy in the supercapacitor.
ea , er the available node energy and required energy The second case is that the harvested energy is not sufficient
wf ree the impact weight of a free task
to support the operation of the node, and part of the energy
wcapacitor the impact weight of a capacitor task
is provided by the energy EC stored in the supercapacitor of
wbattery the impact weight of a battery task
pj the expected profit of task Tj
the node and the battery energy EB.
Our goal is to maximize the total profit earned by the
network over a given target lifetime. The sensor task assign-
A. Sensor node Models ment problem in energy harvesting wireless sensor networks
is modeled as follows:
A wireless multimedia sensor network consists of sensor XX
nodes S(S1 , S2 , . . . , Sn ), which are predeployed in a fixed max pij ∗ (1)
area. Each node is equipped with a primary lithium battery, i∈S j∈T
a supercapacitor, and a solar photovoltaic panel. The sensor s.t. Tj .Ej ≤ Si .Ef + Si .EC + Si .EB (2)
node perception model is shown in Fig. 1.
0 ≤ Si .EC ≤ Cmax (3)
∗
sleeping node pij ≥ pj (4)
active node
Tj (xj, yj)
tsj > tej (5)
→
L →
v
object to be
monitored
→
n
where Cmax is the capacity of the capacitor, pj is the expected
profit of task Tj . As stated in Eq. (2), the energy required for
Si (xi, yi) the tasks performed in the network will be no greater than the
sum of the energy harvested by the node in the future, the
Fig. 1. The Sensor Node Fig. 2. The Network Scene Model.
Model.
existing energy of the capacitor and the battery capacity. The
supercapacitor storage capacity of the sensor network node
The sensor node-aware range is a sector with a radius of L does not exceed the maximum capacity of the capacitor, and
~ is the direction vector of the
and an angle of θ. In Fig. 1, L Eq. (3) implements this rule. Eq. (4) states that only the nodes

932
Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on March 23,2025 at 13:07:51 UTC from IEEE Xplore. Restrictions apply.
 2023 IEEE International Conference on Communications (ICC): IoT and Sensor Networks Symposium

whose profits are no less than the expected profit of the task Algorithm 1: Task classification in EN-DADA
will participate in the election. Eq. (5) stipulates that the time Input: The future harvested energy Si .Ef of node Si , capacitor
at which the task leaves the node monitoring range is before energy Si .EC, battery energy Si .EB and task energy
Tj .Ej .
the time at which the task enters the node monitoring range. Output: Type of task S(i).type.
1) if Si .Ef ≥ Tj .Ej kSi .EC == Cmax then
III. EN-DADA 2) Si .type = I
According to the introduction of the sensor model and 3) else
4) if Si .Ef < Tj .Ej &&Si .Ef + Si .EC > Tj .Ej then
problems in the previous section, this section introduces the 5) Si .type = II
task allocation algorithm based on energy harvesting. 6) else
7) if Si .Ef + Si .EC + Si .EB ≥ Tj .Ej then
A. Task classification algorithm 8) Si .type = III
9) else
The first step of the task allocation algorithm is to judge 10) Si .type = 0
whether the node can execute the task to be monitored 11) end
according to its situation. Therefore, it is necessary to classify 12) end
13) end
the task and determine whether the task can be executed by
the nodes around the task.
As shown in Algorithm 1, tasks can be divided into four
categories according to energy. 180◦ , the higher the utility; (2) The closer the angle between
~ and ~n is to 0◦ , the higher the utility; (3) The larger ~n is,
L
- For a task in the first category, the sum of the energy
available in the future, the current capacitor energy, and the higher the utility.
the battery energy cannot provide the energy required by According to the above three conclusions, the specific
the task, and the task is of type 0 for this node. The node calculation formula of eij is given as:
does not bid for the task. wa wb
eij = + × wc |~n| (7)
- The second category is that of free tasks: if the energy c + cosα + 1 c + |cosβ|
available in the future is greater than or equal to the ~ · ~v
L
energy required by the task, or the capacitor is full, the cosα = (8)
~ · |~v |
|L|
task is of type I for this node, as shown in lines 1 − 2 of
Algorithm 1. L~ · ~n
- The third category is that of capacitor participation tasks: cosβ = (9)
~ · |~n|
|L|
if the energy available in the future is less than the energy
required by the task, and the sum of the energy available where wa is the influence weight of the angle between L ~
in the future and the current capacitor energy meets the and ~v on the total utility; wb is the influence weight of the
energy requirement of the task, the task is of type II for angle between L ~ and ~n on the total utility; wc is the influence
this node, as shown in lines 4 − 5 of Algorithm 1. weight of the distance between the task and the node on the
- The fourth category is that of battery participation tasks: total utility; c is a constant that prevents the denominator from
if the sum of the energy available in the future and the appearing in the fraction.
current capacitor energy and battery energy can meet the 2) The needs of task dj : Features learned by nodes based
energy requirement of the task, then the task is of type on previous history or acquired at design time based on the
III for this node, as shown in lines 7 − 8 of Algorithm 1. applications. For example, if a task only needs one node at a
If a task is of type I, type II or type III for a node, the node certain time period, then dj = 1.
can participate in bidding on the task. 3) The actual profit pij : When task Tj enters the monitoring
B. Bidding algorithm range of node Si and leaves the monitoring range of node Si ,
the total profit that node Si can provide for it is:
All bidding nodes automatically form a candidate set
N (S1 , S2 , . . . , Sm ), and each node in the candidate set cal- Ptej =tsj + |~vl |
pij = tsj pjt (10)
culates the profit gij that it can provide for the task:
eij pij where l is the path length of the task actually moving through
gij = × (6) the monitoring area of node Si ; tsj is the time at which the
dj Pj
task enters the node monitoring range; tej is the time at which
where eij is the utility that node Si can provide for task Tj . the task leaves the node monitoring range; and pjt is the profit
dj is the needs of task Tj ; pij is the actual profit of node Si of task Tj at time t.
for task Tj , and Pj is the maximum profit of the task. 4) The maximum profit Pj : The profit a node would ideally
1) Utility eij : The geographical location of nodes and tasks provide for the task Tj , that is, the profit in the case where
directly affects the utility eij of the task. To select a node that the task enters in the direction facing the node monitoring and
is better aligned with the direction of the task to monitor the directly leaves the node monitoring range.
task, according to the node perception model, we can obtain Ptej =tsj + |~Lv|
the following: (1) The closer the angle between L ~ and ~v is to Pj = tsj pjt (11)

933
Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on March 23,2025 at 13:07:51 UTC from IEEE Xplore. Restrictions apply.
 2023 IEEE International Conference on Communications (ICC): IoT and Sensor Networks Symposium

where L is the radius of the sector, which is the optimal path The profit of task type II (capacitor task) is (as shown in lines
length in the network. 7 − 8 of Algorithm 2):
In the task classification algorithm, the task classifies each
node, and the node calculates pij ∗ after multiplied by task pij ∗ = gij × wcapacitor (16)
weight w and compares it with the expected profit pj . pj The profit for task type III (battery task) is (as shown in lines
is learned by the node according to the past history or is 9 − 12 of Algorithm 2):
obtained when the task arrives according to the application
characteristics (pj is half of the maximum task benefit of Pj pij ∗capacitor = gij × wcapacitor (17)
in our experiment). The pij ∗ can be expressed as
EBit
pij ∗ = gij × w (12) pij ∗battery = gij × wcapacitor × (18)
er(t)
where w is the relevant weight of task classification, which is ECi ECi
the highest for free tasks, followed by capacitor tasks. pij ∗ = × pij ∗capacitor + (1 − ) × pij ∗battery (19)
Ej Ej
The required battery task is the one with the lowest weight
factor. In fact, we want the network to reach a given target where ECi is the current energy of the capacitor; EBi is the
lifetime, so we choose to use precious battery energy only current energy of the battery; Ej is the energy required by the
for tasks with higher relative profits. If a task is battery task; Ef is the energy that node can obtain in the future; er(t)
intensive, the node evaluates the energy contributions provided is the energy required for the node to run in the expected life
by the supercapacitor and battery separately. We adopt a cycle; wf ree is the free task weight; wcapacitor is the capacitor
factor we more specific to adjust the enthusiasm of nodes task weight; and wbattery is the weight of the battery task.
for participating in new tasks, which makes the nodes more Each node in the candidate set compares the calculated pij ∗
conservative in accepting tasks when the energy is lower than with pj : if pij ∗ is greater than or equal to pj , it bids for the task
that required to reach the target lifetime; nodes become more and sends its own pij ∗ to the surrounding nodes; otherwise,
active in accepting tasks as they approach the target network it does not participate in the bidding. After the nodes in the
lifetime. candidate set exchange the profit pij ∗ of node Si for task Tj
pij ∗ = gij × w × we (13) with each other, the node with the largest pij ∗ activates the
ea monitoring function to execute the task. The above algorithm
we = (14) steps will be repeated until the task leaves the sensing area.
er
where ea is the available energy of the node and er is the C. EN-DADA algorithm flow
energy required by the node to reach the given network target
life. The specific steps of the task allocation algorithm based on
The specific calculation process of the bidding algorithm is the solar energy harvesting are as follows:
illustrated in Algorithm 2. ¬ After the nodes are deployed, each node obtains its own
node information and divides the node energy into two
Algorithm 2: Node Bidding in EN-DADA parts: the current capacitance energy and battery energy.
Input: The type of task Si .type, p̄j . ­ When the target to be monitored enters the monitoring
Output: Whether the node is bidding.
1) if Si .type ! = 0 then range of the node, the node obtains the information of
2) Si .eij = wa / (c + cos (Si · vj + 1) the moving target to be monitored.
+wb / (c + abs (cos (Si .nj ))) × wc × Si .nj ) ® The algorithm calculates the energy that the node can
3) Si .gij = Si .eij /dj × Si .pij /Si .Pj
4) switch Si .type obtain during the task execution time according to the
5) case I solar energy collection prediction model.
6) Si .pij ∗ = Si .gij × wf ree ¯ Each node that can monitor the target compares its current
7) case II
8) Si .pij ∗ = Si .eij × wcapacitor capacitor energy, current battery energy, and expected
9) case III energy available during task execution with the expected
10) pij ∗capacitor = Si .gij × wcapacitor energy required by the monitoring task and classifies the
11) pij ∗battery = Si .eij × wbattery × Si .EB
task.
12) pij ∗ = pij ∗capacitor × Si .EC/Tj .Ej
+ pij ∗battery × (1 − Si .EC/Tj .Ej )
° Each node in the candidate set compares the calculated
13) if pij ∗ ≥ p̄j then pij ∗ with pj : if pij ∗ is greater than or equal to pj , the
14) join the bid node will bid on the task and send its own pij ∗ to the
15) else surrounding nodes; otherwise, it will not participate in
16) refuse to join the bid
17) end the bidding.
18) end ± After the nodes in the candidate set exchange the profit
pij ∗ of node Si for task Tj with each other, the node
The profit of task type I (free task) is (as shown in lines
with the largest pij ∗ activates the monitoring function to
5 − 6 of Algorithm 2):
execute the task.
pij ∗ = gij × wf ree (15) ² Steps ¯-± are repeated until the task is over.

934
Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on March 23,2025 at 13:07:51 UTC from IEEE Xplore. Restrictions apply.
 2023 IEEE International Conference on Communications (ICC): IoT and Sensor Networks Symposium

8 25 90 7
50 50 m2 50 50 m2 50 50 m2 50 nodes
7 80 6 100 nodes
100 100 m2 100 100 m2 100 100 m2
20 150 nodes
150 150 m2 70
Profit per mission round

Profit per mission round

6 150 150 m2 150 150 m2

Number of dead nodes

5

Residual energy
60
5 15
4
50
4
40 3
3 10
30
2
2
5 20
1 1
10
0 0 0 0
0 50 100 150 200 0 50 100 150 200 0 50 100 150 200 0 50 100 150 200
Round Number Round Number Round Number Round Number

Fig. 3. Profit comparison of different Fig. 4. Comparison of residual energy Fig. 5. Comparison of the number Fig. 6. Profit comparison of different
monitoring areas. in different monitoring areas. of dead nodes in different monitoring number of nodes.
areas.

IV. E XPERIMENTAL VALIDATION 35 120

50 nodes
50 nodes
30 100 nodes
A. Experimental setup 100 nodes
150 nodes
100
150 nodes

Number of dead nodes

25
80

Residual energy
In an actual scenario, the real-time collection of solar energy 20
is affected by the weather, so this paper adopts a solar energy 15
60

replenishment model based on the Gaussian kernel density 10

40

function. The function is defined as follows: 5

20

1 1 −(x−x1 )2 1 −(x−xn )2 0 0
f (x) = ( √ e 2h2 + · · · + √ e 2h2 ) (20) 0 50 100
Round Number
150 200 0 50 100
RoundNumber
150 200

n 2πh 2πh
where the kernel density window width is selected as h = 0.1. Fig. 7. Comparison of residual energy Fig. 8. Comparison of the number of
Table II shows the relevant settings of the simulation exper- with different number of nodes. node deaths with different number of
nodes.
iments.
TABLE II
S IMULATION PARAMETER SETTINGS . area is set to 100 × 100 m2 , and the number of nodes is set to
Parameter Parameter Meaning Value
50, 100, and 150 for simulation comparison. The results are
N the number of sensor nodes 100
shown in Figs. 6, 7, and 8.
~
L the direction vector of the sensor node 10m From Fig 6, we can see that the more nodes there are, the
~v the speed of the task 1m/s more tasks can be satisfied, and the higher profits the network
Cmax the capacity of the capacitor 0.02J can obtain. However, when the node monitoring range can
Wmax the number of tasks 200 cover the area to be monitored as well as possible, excessive
wf ree the weighting factor for free tasks 1.2 redundancy is unnecessary. Figs. 7 and 8 show that in the
wcapacitor the weighting factor for a capacitor task 1.05 same monitoring area, the more nodes there are, the greater
wbattery the weighting factor for a battery task 0.95 the residual energy of the network, and more nodes can obtain
and effectively utilize solar energy to maintain the network
B. Analysis of influencing factors life. The fewer nodes there are, the higher the node mortality
(1) Monitoring area: To compare the performance of the rate.
algorithm in different monitoring areas, the number of nodes
is set to 100, and the area sizes are 50 × 50 m2 , 100 × 100 C. Algorithm performance comparison
m2 , and 150 × 150 m2 for simulation comparison. The results This section evaluate the performance of EN-DADA and
are shown in Figs. 3, 4, and 5. compare it with the following three task assignment algo-
In Fig. 3, the profit of the network with an area of 50 × 50 rithms, which use the same energy harvesting Gaussion model
m2 is significantly higher than that of others. In the case of the introduced in section IV.A. In the simulation experiments in
same number of sensor nodes, the smaller the monitoring area this section, the nodes are allocated in a square with an area
is, the greater the node density, and the higher the probability of 50 × 50 m2 and have a node radius of r = 10 m, node
of performing tasks. In Fig. 4 and Fig. 5, after 200 rounds of coverage angle of 60◦ , and 200 rounds of tasks entering. The
tasks, the 150 × 150 m2 area has the largest remaining energy comparison algorithms are EN-MASSE [6], QBTA [15] and
and the least number of dead nodes. The larger the monitoring RATA [16].
area is, the smaller the node density, and there may be no Fig. 9 shows that for direction-sensitive monitoring tasks,
nodes to monitor in the task movement route. Therefore, for the EN-DADA algorithm can obtain higher profits. That is,
the same number of nodes, the larger the monitoring area is, the EN-DADA algorithm is more inclined to select nodes that
the smaller the network profit. monitor specific directions to perform tasks to obtain more
(2) The number of sensor nodes: To compare the perfor- useful information. The other three algorithms do not take
mance of the algorithm under different number of nodes, the into account the task of detecting motion as well as possible,

935
Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on March 23,2025 at 13:07:51 UTC from IEEE Xplore. Restrictions apply.
 2023 IEEE International Conference on Communications (ICC): IoT and Sensor Networks Symposium

12 25 number of remaining surviving nodes, and network life.

EN-DADA EN-DADA
10 EN-MASSE [6] EN-MASSE [6]
QBTA [15] 20 QBTA [15]
V. C ONCLUSION
Profit per mission round

RATA [16] RATA [16]

Residual energy
8
15
This paper proposes a task assignment algorithm with solar
6
10 energy harvesting. The directed multimedia sensing nodes
4

5
used by the algorithm demonstrate a profit advantage when
2
performing direction-dependent monitoring tasks. This not
0
0
0 50 100
Round Number
150 200 0 50 100
Round Number
150 200 only improves the energy utilization rate of the WMSN but
also increases the monitoring profit for direction-sensitive
Fig. 9. Profit comparison of the four Fig. 10. Comparison of residual en- monitoring tasks.
algorithms. ergy of four algorithms.
ACKNOWLEDGMENT
100
This work was supported in part by the National Natural
1200
90 EN-DADA
EN-MASSE [6]
EN-DADA Science Foundation of China under Grant No. 62272242 and
QBTA [15] 1000 EN-MASSE [6]
80
61873131.
Total profit of the first n rounds

RATA [16] QBTA [15]

70 RATA [16]
Number of dead nodes

800
60
50 600 R EFERENCES
40
30 400 [1] H. Yue, Q. Jiang, C. Yin, et al., “Research on data aggregation and
20
200
transmission planning with Internet of Things technology in WSN multi-
10 channel aware network,” The Journal of Supercomputing, vol. 76, no. 5,
0
0 20 40 60 80 100 120 140 160 180 200
0
0 20 40 60 80 100 120 140 160 180 200 pp. 3298–3307, 2020.
Round Number Round Number
[2] M. Min, L. Xiao, Y. Chen, et al., “Learning-Based Computation Of-
floading for IoT Devices With Energy Harvesting,” IEEE Transactions
Fig. 11. Comparison of node deaths Fig. 12. Comparison of total profit of on Vehicular Technology, vol. 68, no. 2, pp. 1930–1941, 2019.
of four algorithms. four algorithms. [3] Y. Su, X. Lu, Y. Zhao, et al., “Cooperative Communications With Relay
Selection Based on Deep Reinforcement Learning in Wireless Sensor
Networks,” IEEE Sensors Journal, vol. 19, no. 20, pp. 9561–9569, 2019.
[4] K. M. Fouad, B. M. Hassan, O. M. Salim, “Hybrid Sensor Selection
so the profits obtained after each task is performed are all less Technique for Lifetime Extension of Wireless Sensor Networks,” Com-
puters, Materials & Continua, vol. 70, n. 3, pp. 4965–4985, 2022.
than those of EN-DADA. [5] Y. Jin, K. S. Kwak, S. J. Yoo, “A Novel Energy Supply Strategy
As illustrated in Fig. 10, with the increase in the number for Stable Sensor Data Delivery in Wireless Sensor Networks,” IEEE
of task rounds entering the network, the node energy is Systems Journal, vol. 14, no. 3, pp. 3418–3429, 2020.
[6] T. L. Porta, C. Petrioli, C. Phillips, and D. Spenza, “Sensor Mission
continuously consumed, and the remaining network energy of Assignment in Rechargeable Wireless Sensor Networks,” ACM Trans-
the four algorithms shows a downward trend as a whole. In the actions on Sensor Networks, vol. 10, no. 4, pp. 1–39, 2014.
EN-DADA algorithm, the residual energy sum of the network [7] X. Zhu, K. C. Li, J. Zhang, et al., “Distributed Reliable and Efficient
Transmission Task Assignment for WSNs,” Sensors, vol. 19, no. 22, pp.
nodes decreases slowly, and the residual node energy sum 5028, 2019.
after 80 rounds is significantly larger than that of the other [8] G. Yang, Z. Zhang, J. Wang, et al., “Task allocation based on node pair
algorithms. With an increasing number of tasks, EN-DADA intimacy in wireless sensor networks,” IET Communications, vol. 14,
no. 22, pp. 1902–1909, 2020.
comprehensively considers the energy situation of nodes. [9] V. M. Raee, D. Naboulsi, and R. Glitho, “Energy Efficient Task
As shown in Fig. 11, the number of nodes whose batteries Assignment in Virtualized Wireless Sensor Networks,” in Proc. of 2018
run out of energy increases continuously for the four algo- IEEE Symposium on Computers and Communications (ISCC), IEEE,
2018, pp. 976–979.
rithms after the execution of tasks increases. Initially, there [10] V. M. Raee, A. Ebrahimzadeh, R. H. Glitho, et al., “Ensuring Energy
is less solar power supply, battery power may be used to Efficiency When Dynamically Assigning Tasks in Virtualized Wireless
perform tasks, and nodes die more quickly. As increasing Sensor Networks,” IEEE Transactions on Green Communications and
Networking, vol. 6, no. 1, pp. 613–628, 2022.
solar energy is supplied, the capacitors store increasing energy, [11] D. Weikert, C. Steup, S. Mostaghim, “Multi-Objective Task Allocation
and the number of node deaths flattens. Compared with the for Wireless Sensor Networks,” in Proc. of 2020 IEEE Symposium Series
other algorithms, the EN-DADA algorithm has a slower death on Computational Intelligence (SSCI), IEEE, 2020.
[12] T. Lin, H. Chang, and Y. Wang, “A Novel Hybrid Search and Remove
rate of nodes. Under the same energy supply conditions, the Strategy for Power Balance Wireless Charger Deployment in Wireless
EN-DADA algorithm shows obvious superiority in terms of Rechargeable Sensor Networks,” Energies, vol. 13, no. 2661, 2020.
the number of node deaths. Fig. 12 shows the total profit [13] K. S. Adu-Manu, N. Adam, C. Tapparello, et al., “Energy-harvesting
wireless sensor networks (EH-WSNs): A review,” ACM Transactions on
of the first n rounds of various assignment algorithms in the Sensor Networks, vol. 14, no. 2, pp. 1–50, 2018.
network, which shows that EN-DADA is most suitable for the [14] W. Li, H. Gao, Y. Liu, et al., “A Priority Task Scheduling Algorithm
application of traffic monitoring and target tracking to obtain based on Residual Energy in EH-WSNs,” in 2020 16th International
Conference on Mobility, Sensing and Networking (MSN), 2020, 43–48.
some specific faces of the monitored target. [15] C. C. Evangeline, C. Kushala, and B. Alluri. “Efficient Task Allocation
The simulation results show that compared with the other Strategies For WSNs,” in Proc. of First International Conference on
algorithms, the EN-DADA algorithm proposed in this paper Secure Cyber Computing and Communication (ICSCCC), 2018.
[16] J. Zhou, K. Cao, X. Zhou, et al., “Throughput-conscious Energy Allo-
can obtain higher profits for direction-sensitive tasks, better cation and Reliability-aware Task Assignment for Renewable Powered
meet the needs of specific tasks, and provide more effective In-Situ Server Systems,” IEEE Transactions on Computer-Aided Design
information. The algorithm has advantages in task profit, the of Integrated Circuits and Systems, vol. 41, no. 3, pp. 516–529, 2022.

936
Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on March 23,2025 at 13:07:51 UTC from IEEE Xplore. Restrictions apply.