Distributed environment control using wireless sensor/actuator networks for lighting applications.

Abstract

We propose a decentralized algorithm to calculate the control signals for lights in wireless sensor/actuator networks. This algorithm uses an appropriate step size in the iterative process used for quickly computing the control signals. We demonstrate the accuracy and efficiency of this approach compared with the penalty method by using Mote-based mesh sensor networks. The estimation error of the new approach is one-eighth as large as that of the penalty method with one-fifth of its computation time. In addition, we describe our sensor/actuator node for distributed lighting control based on the decentralized algorithm and demonstrate its practical efficacy.

Full text

Page 1

Sensors 2009, 9, 8593-8609; doi:10.3390/s91108593

sensors
ISSN 1424-8220
www.mdpi.com/journal/sensors
Article
Distributed E Environment Control Using Wireless
Sensor/Actuator Networks for Lighting Applications
Masayuki Nakamura *, Atsushi Sakurai and Jiro Nakamura
NTT Energy and Environment Systems Laboratories, Nippon Telegraph and Telephone Corporation,
Morinosato Wakamiya, Atsugi-shi, Kanagawa Pref., Japan; E-Mails: a2c@aecl.ntt.co.jp (A.S.);
jnaka@aecl.ntt.co.jp (J.N.)
* Author to whom correspondence should be addressed; E-Mail: masayuki@aecl.ntt.co.jp;
Tel.: +81-46-240-3039; Fax: +81-46-270-2320.
Received: 14 September 2009; in revised form: 22 October 2009 / Accepted: 23 October 2009 /
Published: 28 October 2009

Abstract: We propose a decentralized algorithm to calculate the control signals for lights in
wireless sensor/actuator networks. This algorithm uses an appropriate step size in the
iterative process used for quickly computing the control signals. We demonstrate the
accuracy and efficiency of this approach compared with the penalty method by using
Mote-based mesh sensor networks. The estimation error of the new approach is one-eighth
as large as that of the penalty method with one-fifth of its computation time. In addition, we
describe our sensor/actuator node for distributed lighting control based on the decentralized
algorithm and demonstrate its practical efficacy.
Keywords: sensor/actuator network; distributed control; decentralized system

1. Introduction
Wireless sensor/actuator networks (WSANs) have been investigated in addition to wireless sensor
networks (WSNs) since WSANs have more attractive and useful applications than WSNs alone. In
WSNs, the main objective is to gather raw sensor data or estimate the condition of the environment.
From such a sensing standpoint, Zhao et al. proposed collaborative signal and information processing
(CSIP) to target tracking problems [1]. They showed that collaborative and decentralized sensor
networks are scalable and efficient as regards sensing and communication. Rabbat et al. investigated
distributed algorithms for sensor network data processing [2]. They formulated estimation problems as
optimization problems in distributed WSNs. WSANs have the potential to expand WSN applications,
OPEN ACCESS

Page 2

Sensors 2009, 9


8594
enabling the nodes to perform both sensing and actuation [3-5]. A promising WSAN application is the
control of spatially distributed actuators, such as dimmable lighting ballasts, switches and air
conditioning systems. WSANs can save energy because they can accurately monitor environmental
conditions and thus control the actuators precisely. In this context, Taylor et al. developed WSANs for
energy management in heating, ventilation, and air-conditioning (HVAC) systems [6]. Zhang et al.
proposed a field estimation technique that uses WSANs to control HVAC systems in buildings [7].
Li et al. developed WSANs for lighting control in the home environment [8]. One challenge is to
develop algorithms that save energy without sacrificing user comfort. Sandhu et al. proposed WSANs
for lighting control using a multi-agent system [9]. Singhvi et al. developed a centralized lighting
system to increase user comfort and reduce energy costs by using Motes [10]. Lin et al. proposed a
decentralized algorithm for WSANs for optimal lighting control [11]. We proposed WSANs that can
provide optimal actuator control with respect to energy saving and control signal quality as well as
sensing [12]. The sensor/actuator nodes perform sensing and actuation autonomously. However, the
decentralized algorithm is based on the penalty function method, and it takes a long time to compute
optimal control signals.
In this paper, we introduce an improved collaborative sensing and actuation algorithm in an
optimization framework for controlling lights in workplaces. In our algorithm, an objective function is
defined that balances energy saving against control signal quality. We describe a decentralized
algorithm that is more scalable than the centralized one, and that can autonomously calculate control
signals without a central server. This algorithm uses an appropriate step size in the iterative process for
calculating control signals. We demonstrate its accuracy and efficiency compared with the previously
proposed method by simulations. We also carry out WSAN experiments using Motes to examine the
feasibility of the algorithm. We show that the estimation error of the proposed method is one-eighth as
large as that of the previous method with one-fifth of its computation time. In addition, we describe a
testbed that consists of Motes and infrared (IR) remote controls for distributed lighting control based
on the decentralized algorithm.
2. Distributed WSAN Model
In dense distributed WSANs, collaborative processing is essential for intelligent sensing and for
controlling environments such as shared workplaces. To conserve energy, local sensing usually
determines the local actuation of, for example, a light by using an occupancy sensor. When
sensor/actuator nodes are networked, the quality of the control signals within the WSANs is improved,
resulting in occupant satisfaction. In this paper, we use "control signal" as the signal applied to
actuators. We assume that a control signal, for example a current, corresponds to the controlled
environmental condition, such as the brightness of lights. We focus on the energy saving of the lights.
In addition, the spatial smoothness of brightness is likely to be preferred by users. We propose a
method that balances energy saving and the spatial smoothness of the control signals to improve
control signal quality. Our method is formulated as an optimization problem. Let J be an objective
function defined as:

Page 3

Sensors 2009, 9


8595
),) 1 ()(
2
(
2
1
2
2
1
ij
n
iNj
i
fffJ
i
α
α
−+−= ∑∑
∈=

(1)
where n is the number of sensor nodes, fi is the control signal of sensor node i, fj is the control signal of
sensor node j within the communication range Ni of sensor node i, and α
1
(
2
1
iNj
i
=∈
against ∑
=
i1
2
infrared (PIR) occupancy sensors that detect the presence of people, and that the controlled actuators
are lights. Occupancy sensor response si is binary such that:
) 1 0 (
< <α
is a tradeoff
parameter balancing
))(
2
1
2
j
n
i
ff
−
∑∑
in Equation (1) (the spatial roughness of the control signals)
n
if
2
1
(the consumed energy). For this discussion, let us consider that the sensors are passive

. detect)(not 0
(detect) 1
⎩
⎨
⎧
=
is

(2)
The lights are turned on in areas where people are present. To simplify the problem, we assume
that the control signals are set at 1 where people are present, that is:
. 1 if 1
==
ii
sf

(3)
Minimizing J with respect to fi means that lights must be turned on in areas where people are
present and controlled to balance the spatial smoothness of the control signals and energy saving
otherwise. To obtain fi that minimizes J, the partial derivative of J with respect to fi is calculated as:
. 0 if 0)1 ()(/
==−+−=∂∂
∑
∈
j
sffffJ
iij
N
ii
i
αα

(4)
Equation (4) is a simultaneous equation which represents a centralized algorithm. It can be solved
by a server collecting sensor responses from all sensor nodes. After collecting the sensor responses
},,,,,,{
111niii
sssss
⋅⋅ ⋅⋅
+−
, the server computes control signals
Equation (4) and sends the resulting control signals back to the sensor nodes. The sensor nodes control
their external actuators through their digital output ports depending on the control signals.
It is also possible to use a decentralized algorithm, which can compute each control signal without a
server. Rabbat et al. showed that distributed optimization algorithms are more efficient than
centralized ones in terms of energy and communications [2]. The gradient method can be used to
determine the value of fi that minimizes J [13]. Using the gradient method, fi is incrementally updated
by Δfi as:
},,,,,,{
111niii
fffff
⋅⋅ ⋅⋅
+−
by solving
,0 if )) 1 ()((
/
s
i
fff
fJf
ij
Nj
i
ii
i
=−+−−=
∂∂−=Δ
∑
∈
ααε
ε

(5)
where ε is a positive step size.
Formerly J was redefined to include the condition of Equation (3) using the penalty function
method [12]. This approach was very useful for formulating the objective function without classifying
the values of si. However, the convergence of fi was very slow. In this paper, fi is computed separately
by Equations (3) and (5) depending on the values of si.

Page 4

Sensors 2009, 9


8596
Figure 1. Decentralized sensor/actuator network system and calculated brightness of each node.

Figure 2. Two-dimensional schematic diagram of signal propagation for activation.


Figure 1 shows a decentralized sensor/actuator network system in a shared workplace and an
example of calculated brightness for each node. Each sensor node is arranged at regular intervals and
communicates with neighboring nodes to compute its control signal locally. In this model, there is no
central server collecting sensor data and computing control signals. It is necessary to repeat the

Page 5

Sensors 2009, 9


8597
updates described by Equation (5) and the communications of the control signals between
neighboring nodes.
Figure 2 shows the two-dimensional sensor node arrangement and how the signals propagate in the
network for activation. To simplify the problem, sensor nodes were arranged at regular intervals to
form a mesh network. The dashed circle Ni represents the communication range of sensor node i,
which covers neighboring nodes. First, the nodes detect people and become active. Then they
communicate their signals to neighboring nodes. The nodes that receive the signals become active in
turn so that finally all the nodes are active.
Figure 3 shows how the control signals propagate after activation. They compute their control
signals and communicate them to their neighboring nodes in turn. In this way, the control signals are
computed as they propagate through the network. After some communication and computation
iterations, each node provides the optimal control signal for environment control.
Figure 3. Two-dimensional schematic diagram of control signal computation.

It is very important to reduce the number of iterations in this process. We have to determine a large
ε value that ensures the stability of Equation (5). Suppose that m sensor nodes detect the presence of
people. Then Equation (4) is represented as:
,0 if 0 si==−d Hf

(6)
where H is an (n-m) × (n-m) matrix, and f and d are n-m element column vectors. The elements of H
satisfy the following equation:
. 0-1 | h |
1
≠
i
| h |
ij ii
>≥−∑
−
=
α
mn
j
j

(7)
Then we have:
, | h |
1
≠
i
| h |
ijii
∑
=
j
−
>
mn
j

(8)
and H is strictly diagonally dominant and regular [14].