JAIST Repository

https://dspace.jaist.ac.jp/

Heterogeneous Sensing for Temperature Control in

Title
Cyber-Physical Smart Home Systems

Author(s) Chung, Linh Thuy

Citation

Issue Date 2015-12

Type Thesis or Dissertation

Text version author

URL http://hdl.handle.net/10119/12984

Rights

Supervisor:Lim Yuto, School of Information

Description
Science, Master

Japan Advanced Institute of Science and Technology

Heterogeneous Sensing for Temperature Control in
Cyber-Physical Smart Home Systems

CHUNG, Linh Thuy

School of Information Science
Japan Advanced Institute of Science and Technology
December, 2015
 Master’s Thesis

Heterogeneous Sensing for Temperature Control in

Cyber-Physical Smart Home Systems

1310254 CHUNG, Linh Thuy

Supervisor : Associate Professor Yuto Lim

Main Examiner : Associate Professor Yuto Lim
Examiners : Professor Yasuo Tan
Professor Yoichi Shinoda

School of Information Science

Japan Advanced Institute of Science and Technology

November, 2015
 Abstract

Nowadays, a smart home has been developed to automatically achieve some services using
sensors and actuators with the goal to improve the occupant experience, e.g., comfortable
and easier life environment. Smart home system is one of Cyber-Physical System applica-
tions, which is defined as tight integrations of computation, communication, and control
for active interaction between physical and cyber elements in which embedded devices,
such as sensors and actuators, are wireless or wired networked to sense, monitor and con-
trol the physical world. It is an appropriate and efficient way to design the home control
system. It is believed that in both the academic and industrial communities that CPS
will have great technical, economic and social impacts in the future. CPS environment
contains the di↵erent terms in its own elements e.g, sensors, actuators, communication
media. In real scenario where users need a single result from whole system, handling the
heterogeneity of sensors requires to manage the collaborative nature of sensors, that leads
to difficulty in processing or estimating desired parameters in high accuracy. Heteroge-
neous data from heterogeneous and CPS-based oriented sensor, which are equipped on
di↵erent appliances, have di↵erent sensing performance information(e,g. operating range,
response time, accuracy, setting interval), that might cause by the unpredictable change
of environment
This paper proposes a new framework, the heterogeneous data processing and estimat-
ing system (HDPES) that can provide a highly accurate sensed data and/or estimate
a desired data using the CPS-oriented and heterogeneous sensors in the cyber-physical
smart home environment. The design of HDPES is considered in heterogeneity of sensing
performance and sensing data to increase the reliability and accuracy of the temperature
control system in Smart Home
By using the raw data from experiments, we analyze and evaluate our proposed frame-
work in the home environment by using R software, a useful program for statistical com-
puting and data analysis. Through multiple data estimation methods, simulation results
reveal that our proposed system HDPES is adaptable and feasible for satisfying normali-
sation sensing error and estimation the desired parameter at a particular estimating point
in cyber-physical smart home environment.
Acknowledgements

I wish to express my sincere appreciation to all those who have o↵ered me help during
the whole period of my research.
First and most, I wish to express my sincere gratitude to my supervisor, Associate
Professor Yuto Lim, for all of his assistance and guidance. He helps me to finish my
master study with his knowledge, enthusiasm and patience, not only for my research but
also for many other skills like how to prepare a good slides as well as, how to make a good
presentation. He always take a lot of time to explain and discuss with me when I have
problem in research. I especially treasure the experience gained over the past year from
him.
I would like to thank Professor Yasuo Tan, my sub supervisor for his encouragement and
support throughout my study and experiments. Thanks to his comments and guidance,
not only to my research, but also presentation skill, make me understand my research
better and express views more clearly.
I would like to express my thanks to Professor Yoichi Shinoda for taking time to attend
my defence and review my thesis. His comments given at my mid-term defence have been
very helpful for my research, which make me better understand and give me important
views about my research.
Many thanks to my friends, their helps, encouragements make me through many difficult
periods.
Last but not least, thank to my parents, their great love and fully understanding through
all these years is my strongest motivation to go on my study and pursue my life goals.
Contents

1 Introduction 1
1.1 Research Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Cyber-Physical Systems and their Applications . . . . . . . . . . . 1
1.1.2 Smart Homes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.3 Heterogeneous and Homogeneous Sensor . . . . . . . . . . . . . . . 3
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Sensor Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.2 Sampling Interval and Empty Value . . . . . . . . . . . . . . . . . . 9
1.3 Thesis Objective and Contributions . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Design of Heterogeneous Data Processing and Estimating System 11

2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Related Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.1 Kalman Filterring (KF) Method . . . . . . . . . . . . . . . . . . . . 13
2.2.2 Computational Fluid Dynamic (CFD) Method . . . . . . . . . . . . 13
2.2.3 Bayesian Estimation Method . . . . . . . . . . . . . . . . . . . . . . 13
2.2.4 Fitting Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Design of Heterogeneous Data Processing and Estimating System . . . . . 14
2.3.1 Architecture and Design Framework . . . . . . . . . . . . . . . . . . 14
2.3.2 Data Abstracting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.3 Data Processing and Estimation . . . . . . . . . . . . . . . . . . . . 20

3 Evaluation of Heterogeneous Data Processing and Estimating System 24

3.1 Evaluation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.1.1 Root Mean Square Error (RMSE) . . . . . . . . . . . . . . . . . . . 24
3.1.2 Mean Absolute Error (MAE) . . . . . . . . . . . . . . . . . . . . . 25
3.1.3 Integral of Absolute Error (IAE) . . . . . . . . . . . . . . . . . . . 25
3.2 Simulation and Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2.1 Setup, Scenario and Setting . . . . . . . . . . . . . . . . . . . . . . 26
3.2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

ii
4 Conclusion 35
4.1 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2 Research Challenges and Directions . . . . . . . . . . . . . . . . . . . . . . 35

A Survey of Sensor Operating range 37

B Source code for MEF algorithm 39

C Source code for Data Imputation 40

D Source code for MMEM algorithm 43

E Source code for FM algorithm 44

iii
List of Figures

1.1 Cyber-Physical Systems Environment . . . . . . . . . . . . . . . . . . . . . 1

1.2 SmartHome Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Look-up table Data Collection Method . . . . . . . . . . . . . . . . . . . . 4
1.4 CPS-based Data Collection Method . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Maximal T-tolerance per sensor type . . . . . . . . . . . . . . . . . . . . . 6
1.6 Performance Information of Sensor SHT71 and SHT75 . . . . . . . . . . . 6
1.7 Relative of measured temperature and accuracy . . . . . . . . . . . . . . . 7
1.8 Relative of response time and percentage error . . . . . . . . . . . . . . . . 8
1.9 Data distribution of measured sensor and refered sensor . . . . . . . . . . . 9

2.1 Related works of temperature estimation . . . . . . . . . . . . . . . . . . . 12

2.2 Heterogeneous Data Processing and Estimating System . . . . . . . . . . . 15
2.3 Data Abstracting block diagram . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Error Computation block diagram . . . . . . . . . . . . . . . . . . . . . . . 18
2.5 Data Imputation block diagram . . . . . . . . . . . . . . . . . . . . . . . . 19
2.6 Fitting Method block diagram . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.7 Most Minimum Error Method block diagram . . . . . . . . . . . . . . . . . 22

3.1 Experiment intelligent home environment – iHouse . . . . . . . . . . . . . 26

3.2 Room layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3 Performance of Data Imputation . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Performance of MMEM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.5 Performance of MMEM in di↵erent sampling interval . . . . . . . . . . . . 32
3.6 Performance of MMEM vs FM: RMSE . . . . . . . . . . . . . . . . . . . . 33
3.7 Performance of MMEM vs FM: MAE . . . . . . . . . . . . . . . . . . . . . 33
3.8 Performance of MMEM vs FM: IAE . . . . . . . . . . . . . . . . . . . . . 34
3.9 Performance of MMEM: Elapsed time . . . . . . . . . . . . . . . . . . . . . 34

iv
List of Tables

1.1 Type of sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3.1 Sensor Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.2 Sensor Specification Reference . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Simulation parameters and settings: Part I . . . . . . . . . . . . . . . . . . 29
3.4 Simulation parameters and settings: Part II . . . . . . . . . . . . . . . . . 29

v
Chapter 1

Introduction

1.1 Research Background

1.1.1 Cyber-Physical Systems and their Applications

Figure 1.1: Cyber-Physical Systems Environment

Cyber-Physical Systems (CPS) are integrations of computation, networking, and phys-

1
ical processes. Embedded computers and networks monitor and control the physical pro-
cesses, with feedback loops where physical processes a↵ect computations and vice versa.
CPS integrates the dynamics of the physical processes with those of the software and
networking through a lot of sensors and actuators, providing abstractions and modelling,
design, and analysis techniques for the integrated whole[1]
CPS brings many benefits by merging computation and communication with physi-
cal processes. CPS applications bring advance in many areas such as: health care and
medicine, disaster detection and recovery, energy, robotics, smart transportation, smart
home and another smart structure.

1.1.2 Smart Homes

Figure 1.2: SmartHome Environment

2
Smart Home is a living environment that incorporates the appropriate technology, called
Smart Home technology, to meet the resident goals of comfort living, life safety, security
and efficiency. Smart Home technology started for more than a decade to introduce the
concept of networking devices and equipment in the house. It is a home automation system
that allows for controlling over a home environment, media systems, home security, and
integrates with an easily accessible user interface through home network. The devices
and systems in Smart Home environment can communicate with each other and can be
controlled automatically in order to interact with the household members and improve
the quality of their daily life.
In a Smart Home system, one of CPS applications, many devices and appliances are
equipped with sensors and actuators to meet the demands or preferences of occupants such
as: controllable door, fire alarm, lighting control, human present detector, temperature
control, etc. These di↵erent devices and appliances lead to the presence of heterogeneous
sensor in Smart Home.

1.1.3 Heterogeneous and Homogeneous Sensor

Nowadays, sensors used in CPS environment can be classify into two types: homogeneous
and heterogeneous sensors. Homogeneous sensors are identical in term of energy, hardware
complexity, performance. Heterogeneous sensors consist of multiple physically di↵erence
types of sensor.

Table 1.1: Type of sensors

Heterogeneous Sensors Homogeneous Sensors
They consist of sensors have di↵erences They have the same device characteristics
in processing power, resource, communi- and sensing features from the same manu-
cation module, AC power supply. facturer
They are also di↵erent in sensing perfor-
mance in terms of response time, accuracy,
resolution, unit and operating range

1.2 Motivation
In Cyber-Physical Smart Home environment, to collect data from the domain of hetero-
geneous sensors and actuators, two types of data collection method can be implemented.

3
 Figure 1.3: Look-up table Data Collection Method

Look-up table Data Collection Method is a feasible method to collect the sensed data
from traditional heterogeneous sensor and retrieve the sensing performance. Usually, the
sensor performance data sheet is commonly static from a pre-prepared database, which is
provided by the manufacturer.

4
 Figure 1.4: CPS-based Data Collection Method

CPS-based Data Collection Method: proposed method to be expected to collect not

only the sensed data, but also the dynamic sensing performance from CPS sensor in timely
manner. In this research, CPS sensor is defined as a sensor that is able to send its own
sensing performance that might cause by the unpredictable change of environment using
a CPS communication module.

1.2.1 Sensor Performance

Each of sensor has its own performance parameters (e.g., resolution, accuracy, repeatabil-
ity, operating range, response time), which are di↵erent from others. These performance
parameters can contribute error to the accuracy of sensed data, which are changing ac-

5
cording to dynamic environment. Sensor performance characteristics give a technical
information for certain sensor performance parameter with the specified definition and
meaning. Figure 1.6 and 1.5 shows the sensor performance parameters in data-sheet of
sensor SHT7x (including SHT71 and SHT75), relative humidity and temperature sensors
with pins.

Figure 1.5: Maximal T-tolerance per sensor type

Figure 1.6: Performance Information of Sensor SHT71 and SHT75

6
 In this research, the influence of sensor performance parameter is analyzed, especially
considering in accuracy, operating range and response time

1.2.1.1 Operating Range and Accuracy

Sensor works stable within recommended normal range, after return to normal range it
will slowly return towards calibration state by itself. Changes in the temperature of the
surrounding area can result in significant shifts in the dielectric constant of area, which
introduces inaccuracies in the sensor readings. Accuracy parameter is obtained in sensor
performance to describe generally as the largest expected error between actual and ideal
output signals. Error comes from influence of measured temperature in operating range
on accuracy parameter is described as this equation:
8
>
<dopt , d = dopt
d = ( dmin dopt )/(dopt dmin )· | d dopt | + dopt , d < dopt
>
:
( dmax dopt )/(dmax dopt )· | d dopt | + dopt , d < dopt

Figure 1.7: Relative of measured temperature and accuracy

Figure 1.7 shows the relative of measured temperature and sensor accuracy performance
parameter in percentage unit. Before 12:00 AM the temperature is lower than 10 C,
percentage error is higher than 10% . However, after 12:00, the temperature increases
and the percentage error decreases

7
1.2.1.2 Response Time
Response time, is an expression of how quickly a sensor responds to temperature changes.
Time constant is a particular case of response time, which is defined as the length of time
it takes a sensor to reach 63% of a step temperature change.
The response time depends on heat capacity of and thermal resistance to sensor sub-
strate, that means sensor with di↵erent type of thermocouple will give di↵erent average
response time. A rapid response time is essential for accuracy in a system with sharp
temperature changes [19]. Measured temperature dik of sensor Si at time tk is represent
in the following equation
Iset
dik = d0ik (d0ik di(k 1) ) ·e ⌧ (1.1)
where t is response time of sensor Si at time tk , Iset is the setting time on sensor made by
engineer to give an output sensing value, d’ik is the actual temperature
Iset
) d0ik dik = d0ik di(k 1) ·e ⌧ (1.2)

An error fraction function is defined:

d0ik dik Iset
=e ⌧ (1.3)
(d0ik di(k 1) )

Figure 1.8: Relative of response time and percentage error

Figure 1.8 represent percentage error betweens actual temperature and measured value
of sensor, which is inversely proportional to the ratio of setting interval and response time

8
1.2.2 Sampling Interval and Empty Value
Some definition of time in sampling data are defined:

• Setting interval Iset is the time, which is set by engineer on sensor, to give sensing
data. Iset is greater or equal to the minimum response time of sensor

• Sampling interval Isamp is the period betweens two samples, which is made by request
from user/application

• Sampling time tsamp is the time to observe sensor

When the fraction Isamp

Iset
is not an integer and each sensor starts with di↵erent time although
they have the same sampling interval, empty value presents in sample.
This figure shows the of measured sensor and refer sensor in sampling interval 30 sec-
onds, setting interval of each sensor is 120 seconds. Each sensor starts with di↵erent time
although they have the same sampling interval. This leads to the presence of empty data
when the system read sensors’ reading at each sampling time.

Figure 1.9: Data distribution of measured sensor and refered sensor

9
1.3 Thesis Objective and Contributions
This research aims to develop the heterogeneous data processing and estimating system
(HDPES) that can provide a highly accurate sensed data and/or can estimate a desired
data using the CPS-oriented and heterogeneous sensors in the cyber-physical smart home
environment.
The contribution of this research is divided into 3 folds: (i )Specify a new framework
for using CPS sensors with heterogeneous sensing data from cyber-physical smart home
environments. Emphasis on resolving dynamic total error by selecting only some of input
sensors using the minimum error first (MEF) algorithm; (ii ) To propose a novel estima-
tion method, minimum error method (MEM) to improve the accuracy of the parameter
considered (temperature) at specific location; (iii )To study and analyse the relationship
between total error and the performance of the proposed framework. By using a simulator,
which is written in R language.

1.4 Thesis Outline

• Chapter 1 explains the research background, motivation, objective and contribu-
tions.

• Chapter 2 gives some related work, shows our designed proposed framework, pro-
posed algorithm to reducing error in sensing data and estimating the desired pa-
rameter in CPS smart home environment

• Chapter 3 evaluates our proposed system by conducting some simulated studies and
data analysis.

• Chapter 4 concludes the thesis, points some research challenges and draws our future
works.

10
Chapter 2

Design of Heterogeneous Data

Processing and Estimating System

2.1 Overview
Indoor applications developed intelligent diversification in many areas to bring utilities
to the user in everyday life or toward the maximum saving of energy for electrical equip-
ment in the smart home. The Home Automation field is expanding rapidly as electronic
technologies converge. The home network encompasses communications, entertainment,
security, convenience, and information systems [20]. Advancements in the fields of in-
telligent home systems such as doors control system for safety purposes, healthcare; the
warning systems: fire alarm, gas leak detection; environment adjustment of living space.
This research focuses on the gathering, handling heterogeneous data from sensors of
di↵erent appliances to serve the same purpose is to estimate and control a specific variable.
Specific variables could be as temperature, illumination, humidity, occupants’ locations,
sound, etc. The proposed system can be applied to achieve this aim. However, in this
research temperature, one of importance information for thermal comfort in smart home,
is used as the particular parameter to study and analysis data, system performance.

2.2 Related Works

Temperature is the most basic environmental parameter. Almost the response of appli-
ances in smart home are closely related to the temperature. Temperature has an inter-
connection influences with people’s daily life. Some related works of temperature control
in home environment can be found in [2], [3], [4] and [5] . These works considering in
optimization accuracy of estimated temperature in room based on homogeneous sensors.
We can se the overview through the following history graph:

11
Figure 2.1: Related works of temperature estimation

12
2.2.1 Kalman Filterring (KF) Method
S.Sharifi, proposed method for accurate estimation of temperature at various locations on
a chip considering the inaccuracies in thermal sensor readings due to limitations mainly on
thermal sensor placement and sensor noise. Kalman filter (KF) is ultilized for temperature
estimation and for elimination of sensing inaccuracies as well. This technique typically
reduces the standard deviation and maximum value of temperature estimation errors
by about an order of magnitude. The most important of this technique is efficiently
in order to estimate the temperatures at the locations of interest where no sensor is
available. Model order reduction is used to reduce the size of the model and generate
a much smaller yet accurate linear system. Kalman Filtering estimates the temperature
at di↵erent locations on the chip based on the inaccurate temperature readings at sensor
locations and inaccurate power consumption estimates [2].
The thermal network is represented in state space form with the grid cell temperatures
as states and the power consumption as inputs to this system. The outputs of this state
space model are the temperatures at the sensor locations which can be observed by sensor
readings.

2.2.2 Computational Fluid Dynamic (CFD) Method

F.Yan et al. based on CFD theory and Airpak software, the mathematical model of indoor
air flow was established. CFD method use a large quantity of experimental data, the
influences of miscellaneous factors such as the influences of wind velocity, wind frequency,
and temperature to indoor thermal comfort are analysed. Considering the dynamics of the
simulated wind velocity, indoor air flow is generally incompressible and of low turbulence,
the turbulence model of Airpak.
This paper found when the wind velocity is in wave patterns, the increment of wind
velocity can also bring preferable thermal comfort, even if the temperature reaches to a
high level. Then specific values are given to some typical cases. The research can provide
theoretic reference and beneficial experience to building ventilation design. The RNG
K " model is used, which is established based on the Standard K " model, widely
applied in the condition of turbulence [3]

2.2.3 Bayesian Estimation Method

M.Jing, a new temperature sensor array was setup to measure temperature, which uses
the Bayes estimation to fuse the data measured by the temperature sensor array, and
calculate the accurate temperature measurements system implementation. To measure
temperatures, the new sensor array are used and made a rapid accurate judgment to the
change of the environmental temperature. The sensor of the sensor array is quartz tuning
fork non-contact temperature sensor based on the polymer line, with the high precision
and the low cost and the simple manufacturing process. The system will use the Bayes
estimation to fuse the measured temperature data the array get, and it overcomes that
the precision is not high [4]

13
2.2.4 Fitting Method
Z.Cheng proposes a fitting method to estimate the actual room temperature, which felt by
occupant. By using the design idea of CPS, a hybrid temperature control (HTC) system
was proposed. It enables to monitor and maintain the room temperature in the desired
interval. Through simulations and field experiments, the relationship between control
performance and sensing accuracy was captured. This method improved the sensing ac-
curacy without increasing monetary cost of the system implementation. To increase the
sensing accuracy is crucial to improve the efficiency of the control system, linear regres-
sion method is used to establish the fitting function. With the help of Simulink Design
Optimization toolbox in Matlab software, the mean value of square error is minimized
based on trust region reflective algorithm [5]

2.3 Design of Heterogeneous Data Processing and

Estimating System
2.3.1 Architecture and Design Framework
In this research, the proposed system can work with data come from two sources: tra-
ditional heterogeneous sensors and CPS sensors with di↵erent types of data collection
methods: Look-up table data collection method and CPS-based data collection method.
In CPS-based data collection method, besides sensing data, sensing performance of
CPS sensor is also sent directly to server by using CPS communication module. In the
other hands, server will send a request message to database storage to retrieve static
sensing performance of normal heterogeneous sensor, which does not have any mechanism
to update sensing performance in timely manner

14
Figure 2.2: Heterogeneous Data Processing and Estimating System

15
 Figure 2.2 shows the basic architecture of HDPES. In this architecture, cyber world and
physical world are defined. These two worlds connect together through a communication
media (e.g, WSAN). In particular, the WSAN comprises of two components: sensors
and actuators. The sensors in the physical side send data included the environment
temperature for inside room periodically to data storage in the cyber side.
Subsequently, the sensed data and performance parameter from server is extracted and
sent to Data Abstracting block. Inside Data Abstracting block, total error from the
influence of unpredictable changing environment and performance parameters on sensing
accuracy is appreciated to reduce sensing error before supply input for next block. Based
on the data from previous block, Data Processing and Estimating block use a special
estimate method, which is proposed by the authors, to predict desired parameter at
estimating point and send to controller. The mission of controller is to compute a control
signal to achieve appropriate actuators to perform the corresponding task to influence the
home temperature.

2.3.2 Data Abstracting

Functions of the Data Abstracting are:
1. To reduce sensing error of the input desired parameter by using the Fitting Method
2. To recover empty data in each sampling interval immediately by using a temporal
model, Autoregressive Integrated Moving Average (ARIMA)

Figure 2.3: Data Abstracting block diagram

Data Abstracting consists of Fitting Method and ARIMA model with the goal to supply
better input data for estimating a highly accurate value of the desired parameter in Data
Processing and Estimating block

16
2.3.2.1 Error Computation
To optimise the accuracy of desired parameter, this method also makes an error compu-
tation, which is an adding bounding condition for choosing a good input data suitable
with scenario sampling. To choose better sensors’ readings with lower error for the inputs
by using Minimum Error First (MEF) Algorithm.
The measured value dik of Si at time tk consists of actual value d’ik and random generated
error in range ±Eik :
E is the total expected error representing the overall change in the sensor performance
which is caused by the unpredictable change of the surrounding environment as well as
sensors unreliability. E sometimes is provided by the sensor specification.
In this simulated study, assuming that E of sensor Si at time tk : Eik consist of error
come from influence of operating range to accuracy and response time parameter
T
X
Eik = "o,ik + "r,ik (2.1)
k=1

The block diagram of reducing sensing error mission is shown in figure 2.4 . In this
block, server checks the existing of total error band in sensing performance information,
and figure out Eik if it does not in static prepared database.
Subsequently, Minimum Error First Algorithms (MEF) collects all the generated error
from the total number of sensors, sort in increasing order, then choosing the number of
input sensors (n) based on minimum E and send data output to Data Imputation block.

17
Figure 2.4: Error Computation block diagram

18
2.3.2.2 Autoregressive Integrated Moving Average model (ARIMA)

Figure 2.5: Data Imputation block diagram

19
The time series data that has inexplicable changes in direction, is analysed and build
a temporal model by modelling it in ARIMA model ARIMA(p,q) models are a class of
linear models, that are capable of representing stationary and non-stationary time series.
ARIMA model rely heavily on autocorrelation patterns in data both ACF and PACF are
used to select an initial model.
The model is generally denoted to as an ARIMA(p,d,q) model where, parameters p,
d, and q are non-negative integers used to refer to the order of the auto-regressive, the
amount of di↵erencing, and moving average parts of the model respectively

dc
ik = 0+ 1 d1(k 1) + 2 d2(k 2) +...+ p di(k p) ⇥1 "1(k 1) +⇥2 "2(k 2) +...+⇥q "i(k q) (2.2)

An auto-regressive (AR) model is a simplified version of ARIMA model which describes

random time-varying process. The AR model specifies that the output variable depends
linearly on its own previous values. The AR model of sensor data with order p is defined
as follows
dc
ik = 0 + 1 d1(k 1) + 2 d2(k 2) + ... + p di(k p) (2.3)
where p is the order of auto-regressive terms, F1 , F2 , ...,Fp are the parameter of the model
A q-order moving average model, or MA(q), is a linear regression of the current and
previous error of a random series. A model with autoregressive terms can be combined
with a model having moving average terms to get an ARIMA(p,q) model

dc
ik = ⇥0 + ⇥1 "1(k 1) + ⇥2 "2(k 2) + ... + ⇥q "i(k q) (2.4)

where q is the number of moving average terms, J1 , J2 , ...,Jp , eq is white noise.

To recover empty data, ARIMA model of each sensor selected from the previous block,
is contributed based on history data in o✏ine part. Empty data is recover in real-time in
online part.

2.3.3 Data Processing and Estimation

Functions of the Data Processing estimates the temperature by using Data Estimation
Methods to obtain an accurate value at the considered point. This block uses the history
embedded data output of the previous Data Imputation Block.

2.3.3.1 Average Method (AM)

The commonly used method for control system estimating the actual room temperature
at estimating point is using the average value of data sensed by the equipped sensors [8].
This is the equation of average method:
PN
c dik
dk = i=1
A
(2.5)
N

20
2.3.3.2 Root Mean Square (RMS)
This is the improvement of average method to reach the higher accuracy of estimated
value. In statistics, the root mean square value, also known as the quadratic mean, is a
statistical measure defined as the square root of the arithmetic mean of the squares of a
set of values. The RMS value is always greater than or equal to the average
s
PN 2
cR i=1 dik
dk = (2.6)
N

2.3.3.3 Fitting Method (FM)

Figure 2.6: Fitting Method block diagram

This is the method proposed by Cheng [5]. In this research, fitting function is conducted
by using training data based on history value of measured sensors, selected in previous
block.
Linear regression was the first type of regression analysis to be studied rigorously, and
to be used extensively in practical applications. In linear regression, the relationships are
modelled using linear predictor functions whose unknown model parameters are estimated

21
from the data. The block diagram of FM shows how to build a fitting function in fitting
method.
This is the equation of fitting function:
dc
F
k = 1 d1k + 2 d2k + ... + i dik (2.7)

2.3.3.4 Most Minimum Error Method (MMEM)

Figure 2.7: Most Minimum Error Method block diagram

This is the proposed method, using fitting method (FM), root mean square(RMS), to
obtain an additional representation of the current values of sensors’ readings. Choosing
the readings with lowest error (closest to the estimated value obtained from FM).

22
 According to the correlation coefficients between the measured sensors and refer sensor,
make regression analysis which analyses the relationship between the data fitting. There-
fore, when the mutual amendment is made in the use of a higher correlation coefficient
failure without valid data of relevant measured sensors can be used to supplement it.
Considering in this concept, MMEM is a method to compute the minimum error among
the sensors’ readings and FM,RMS or another estimate method in each sampling interval
and choose the best values for estimated data.
This is the mathematical equation to represent for MMEM estimated value:
0 1
|d1k dc F
k|
B F C
B |d2k dc k| C
C
|dIk dc F
k | = M in B (2.8)
@ ... A
|d ik dc
F
k |

dbk = dIk

23
Chapter 3

Evaluation of Heterogeneous Data

Processing and Estimating System

3.1 Evaluation Methods

To understand how our framework will act in the physical word, simulations by using R
software and simulated studies conducted from experiment data in the intelligent home
environment are adopted to verify the proposed system. R is derived from an original
set of notes describing the S and S-Plus environments written in 1990 by Bill Venables
and David M. Smith when at the University of Adelaide. R is an integrated suite of
software facilities for data manipulation, calculation and graphical display. R software is
the commonly software using in statistical and computing
It is important from the research perspective, as well as from a practical view, to be able
to decide on an algorithm that matches the domain and the task of interest. The standard
way to make such decisions is by comparing a number of algorithms o✏ine using some
evaluation metric. Many evaluation metrics have been used to rank algorithms, some
measuring similar features, but some measuring drastically di↵erent quantities.
To evaluate the performance of the proposed system, three of evaluation metric are
computed: the root mean square error (RMSE), the mean absolute error (MAE) and the
integral of absolute error (IAE)

3.1.1 Root Mean Square Error (RMSE)

The RMSE is a frequently used measure of the di↵erence between values estimated by
an algorithm and the values actually measured from the real environment. An algorithm
estimation with respect to the estimated value, the RMSE value is defined as the square
root of the mean squared error as written as:
s
PT d dV k )2
k=1 (dV k
RM SE = (3.1)
T

24
3.1.2 Mean Absolute Error (MAE)
The MAE is another statistical measurement that used to measure how close the estimated
values are to the measured values. The MAE measures the average magnitude of the errors
in a data set, over the verification sample of the absolute values of the di↵erences between
forecast and the corresponding observation, without considering their direction. In other
words, it measures the accuracy for the continuous variables.
The MAE and the RMSE can be used together to analyse the variation in the errors of
the data set. The value of RMSE will always be greater or equal to the MAE. The value
of RMSE will always be greater or equal to the MAE
T
1X d
M AE = dV k dV k (3.2)
T k=1

3.1.3 Integral of Absolute Error (IAE)

The IAE is a widely used performance metric in control community, which is recorded
to measure the performance of the control application. The IAE is calculated as follows,
where, t denotes total simulation time. In general, the larger the IAE values imply the
worse the performance of the control algorithm
Z k
IAE = dc
V (k) dV (k) dk (3.3)
0

25
3.2 Simulation and Data Analysis
3.2.1 Setup, Scenario and Setting

Figure 3.1: Experiment intelligent home environment – iHouse

In this section, I verify and examine how the proposed Heterogeneous Data Processing
Estimation System (HDPES) will behave in estimated data by making the simulation
conducted with R software tool. In the simulation, I use the raw data from the experiments
that were conducted at the intelligent house environment, iHouse, which is located at Nomi
city, Ishikawa, Japan. Figure 3.1 shows the overview of iHouse. Three parts are included
in the simulations.

26
 Use the measured data of the sensor located at the centre of Bedroom A of the iHouse
as a reference reading. Layout of the bed room A showed in figure 3.2

Figure 3.2: Room layout

Scenarios: Evaluate the proposed system, 8 CPS-sensors from 4 types (A,B,C,D) with
di↵erent sensor performance. Data for those sensors is created based on actual measured
data of the reference sensor with di↵erent generated random error with the maximum
total error band as described in table 3.1

27
 Table 3.1: Sensor Specification
Accuracy C Response Time
Quality Type Operating range C (Optimum Values C) (seconds)
A
2 (High accurate) -40 ! 150 ± 0.25 (25) 2
B
4 (Typical sensor) -40 ! 123.8 ± 0.25 (25) 17.5
C
(Sensor inside
1 wall-clock) -40 ! 70 ± 1.5 (25) 30
D
(Sensor inside
1 air-conditioner) -50 ! 80 ± 2 (25) 10

The reference source to refer the sensor performance is shown in table 3.2

Table 3.2: Sensor Specification Reference

Type Refer Source
A
(High accurate) http : //www.analog.com/media/en/technical-
documentation/datasheets/ADT 7420.pdf
B
http : //www.sensirion.com/f ileadmin/user upload/
(Typical sensor) customers/sensirion/Dokumente/Humidity/
Sensirion Humidity SHT 7x Datasheet V 5.pdf
C
(Sensor inside
wallclock) http : //www.acurite.com/timex115atomicdigitalwallclock-
withtemperature moon phasecalendar75331t.html
D
(Sensor inside
air-conditioner) http : //www.smartclima.com/airconditionertemperature-
sensor.htm
http : //www.vishay.com/docs/29053/ntcintro.pdf

3.2.1.1 Part I: Analysis of Data Imputation Block

In this part, the error computation and data imputation are combined in the system and
evaluate the result . The data I used in simulation is conducted at the iHouse master
room, which is mainly used for work and study. The values and parameters used in the
simulation are shown in Table 3.3 .

28
 Table 3.3: Simulation parameters and settings: Part I
Parameter Value
Vroom (L ⇥ W ⇥ H): volume of room 5.005m ⇥4.095m ⇥2.4m
Iset : setting interval sensors 2mins
Isamp : sampling interval of system 30secs, 1min, 2mins, 3mins, 4mins, 6mins
tsamp : period time to observe data 1day
N : number of measured sensor 2!8
Observation Time for reference sensor 15 and 16 - December 2013

3.2.1.2 Part II: MMEM performance

This part evaluate the performance of proposed algorithm Most Minimum Error method.
The data is used in simulation is conducted at the iHouse master room, which is mainly
used for work and study. The values and parameters, other than those shown in Table
3.3 , used in the simulation are shown in Table 3.4

Table 3.4: Simulation parameters and settings: Part II

Parameter Value
Iset : setting interval sensors 2mins
Isamp : sampling interval of system 3mins
tsamp : period time to observe data 1day

All temperature sensor are used in this simulation have the same unit and meaning
(to measured room’s temperature). However, in practical environment, sensors can be
di↵erent unit (Celsius, Kelvin, Fahrenheit) and di↵erent meaning or di↵erent function
(measure room’s temperature, measure temperature inside an appliance/ device, etc.).
Because of these reasons, before apply MMEM algorithm, the unit of all used sensors
must be synchronised to the standard unit (Celsius) by the proposed system. Besides, a
threshold for inside room’s temperature is necessary to make a boundary for temperature
value in room.
Linear regression, root mean square method, minimum square error method, are the
di↵erent represent of average method. These methods are easy to implement and save
time in computing. The disadvantage of these methods is extremely sensitive to extreme
values. Hence, using FM, RMS, AM for data sets of sensors’ readings containing a few
extreme values is not a good solution. In this case, median value of a large data set can
be a better alternative (Gaussian).
With a few sensors, the present solution is to normalise all sensor’s readings di by
using percentage of relative value between sensor reading and measured temperature of

29
reference sensor dd
ref .
di
di ! d0i = · 100
dd
ref

3.2.2 Results
3.2.2.1 Part I: Analysis of Data Imputation Block
The results we report in figure 3.3 is conducted by using average method on reduced error
data and raw data for estimating the desired parameter at estimating point RMSE is
shown of the system with and without the Data Imputation Block. With data imputing,
RMSE decreases up to 16% (at Isamp = 30s)

Figure 3.3: Performance of Data Imputation

3.2.2.2 Part II: MMEM performance

This simulation is to evaluate the performance of MMEM in estimating accurate temper-
ature. In this simulation, the input sensed data of measured sensor are already imputing
to fulfil all empty value, which caused by di↵erent setting interval and the start time to
sample of each sensors. Apply MMEM in the following 4 cases:

• 8 sensors (4 type A, 2 type B, 1 type C, 1 type D) ! worst case

30
 • 4 sensors (2 type A, 2 type B ) ! not shown

• 3 sensors (2 type A, 1 type B) ! not shown

• 2 sensors(2 type A) ! best case

3.2.2.2.1 Accuracy
After using Minimum Error First algorithm to choose sensors, which has lower error band,
the estimated temperature is closer to the measured temperature of reference sensor. The
di↵erence for 4 cases with di↵erent the number of high error band sensors, the result is
better with the decreasing of randomly generated error E. In figure 3.4 , case of using 2
high accurate sensor is the best case, and the case of using 8 sensors (included large error
band sensor, which are in air-conditioner and wall-clock).

Figure 3.4: Performance of MMEM

The result of MMEM also represents the influence of Data Imputation block in case
of sampling interval is : 30secs, 1min, 2mins, 4mins and 6mins. The results is reported
in figure 3.5 is conducted by using Most Minimum Error method on reduced error data
and raw data for estimating the desired parameter at estimating point. RMSE, MAE is
shown of the system with the Data Imputation Block. With data imputing, RMSE in
case of sampling interval equal to 3 minnutes is the best case in this simulation (0.27 C)

31
 Figure 3.5: Performance of MMEM in di↵erent sampling interval

In figure 3.6, 3.7 and 3.8, the estimated value of two methods: Fitting Method(FM)
and the proposed method Most Minimum Error Method(MMEM) are compared with
temperature from reference sensor. The di↵erence for both RMSE, MAE, IAE between
MEM and FM increases with the average of total error E. However, in this simulation,
the increasing of RMSE, MAE, IAE is not much.
The average of total error is represented in the following equation:
Pn PT
k=1 Eik
j=1 T
E= (3.4)
n
• 8 sensors (4 type A, 2 type B, 1 type C, 1 type D): 40.10%

• 4 sensors (2 type A, 2 type B ): 33.78%

• 3 sensors (2 type A, 1 type B): 32.61%

• 2 sensors(2 type A): 30.26%

• RMSE of MMEM is 87% less than for FM.

• MAE of MMEM is 89% less than for FM.

• IAE of MMEM is 34% less than FM.

32
Figure 3.6: Performance of MMEM vs FM: RMSE

Figure 3.7: Performance of MMEM vs FM: MAE

33
 Figure 3.8: Performance of MMEM vs FM: IAE

3.2.2.2.2 Elapsed Time

These simulations take place on a computer with following specification: CPU speed
(1.7GHz Core i7), Memory (8GB), OS: Macintosh. The result in figure 3.9 is conducted
after taking account in 10 times to get the average value of elapsed time.
Elapsed is directly proportional to the number of sensors. Elapsed time of 2 sensors is
58% less than the elapsed time of 8 sensors

Figure 3.9: Performance of MMEM: Elapsed time

34
Chapter 4

Conclusion

4.1 Concluding Remarks

In this research, a new framework is specified for using CPS sensors with heterogeneous
sensing data from cyber-physical smart home environments. The design of Heterogeneous
Data Processing and Estimating System is presented with expected collection data method
for heterogeneous sensing. Emphasis on resolving dynamic total error by selecting only
some of input sensors using the minimum error first (MEF) algorithm. A novel estimation
method, minimum error method (MEM) is proposed to improve the accuracy of the
parameter considered (temperature) at specific location. The relationship between total
error and the performance of the proposed framework is studied and analysed by using a
simulator, which is written in R language.
By comparison simulation result of di↵erent sensors, which are equipped on di↵erent
appliances, to evaluate and verify HDPES; the algorithm of MMEM inside the proposed
system can obtained to estimate the desired parameter with highly accuracy at estimated
point.
Sensing performance of specified sensors, which is designed for di↵erent purpose, have
various factors (e,g resolution, precision, accuracy, hysteresis, operating range, humidity).
Through data analysis, the a↵ect of environment and sensing performance factors have a
strong impaction on accuracy of sensed data, this is also the reason of di↵erence betweens
estimated and actual value.
For the continuous work, I will make a deeply detail survey about sensor performance
factors of usual sensor types in home appliances and influence of these factors on sensed
value. This survey supports to do a completely error computation to give a better condi-
tion for choosing input data in processing and estimating desired parameter.

4.2 Research Challenges and Directions

The complexity of Cyber-Physical Systems, resulting from their intrinsically distributed
nature, the heterogeneity of physical elements (i.e. sensors and actuators), the lack of
reliability in communications, variability of the environments in which they are employed,

35
makes data analysis, processing and estimating as a complex task[6]. Base on these views,
my future research direction will aim to expand the proposed system HDPES with a fully
API such as an interface to integrate between the proposed system and the other home
application/controller, CPS oriented sensor. This API will be developed in CPS-base
oriented to obtain these below folds:

1. To consider the case when abnormally long response time sensor causing communi-
cation delay longer than the communication cycle

2. Spatial correlation also need to be considered same as temporal correlation has been
studied in this research

3. To formally define the interface between the already in use controllers and its pro-
tocols on one hand and the proposed system on the other hand

I expect that the machine learning technique can be developed for improving the per-
formance of error computation and estimating algorithm in the proposed system. This is
a new and promising research domain for CPS approach especially in smart home envi-
ronment. Since the outcomes of the HDPES system using the co-design framework can be
varied dynamically due to the unpredictable changing of environment factors, the machine
learning technique may be able to improve the predictive variables accuracy of estimating
method, it may leads to the entire system is adaptable to the dynamic change of smart
home environment under di↵erent sensing performance of CPS-based oriented sensors.

36
Appendix A

Survey of Sensor Operating range

Use data from Smart Temperature Sensor Performance Survey [7], this graph show the
a↵ect of measured temperature to sensor accuracy of many type of sensor Relative inac-
curacy corresponds to the slope of an imaginary boo placed around the sensor’s error by
this formula:
PPIA
· 100
Specified temperature range

37
38
Appendix B

Source code for MEF algorithm

function MEF( data , n )

{
% ‘ data ’ i s a m a t r i x i n c l u d e s e n s i n g d a t a from a l l measured
sensor ,
nrow < nrow ( data )
n c o l < n c o l ( data )
%I n i t i a l t h e e r r o r a r r a y
E < matrix ( data = NA, nrow , n c o l )
estimate < c ()
for ( i i n 1 : nrow )
{
for ( j i n 1 : n c o l )
{
%E s t i m a t e T o t a l Error o f s e n s o r s ’ r e a d i n g s
% e . o : t h e e r r o r from o p e r a t i n g range , e . r : t h e e r r o r
from r e s p o n s e time
E[ i , j ] < e . o [ i , j ] + e . r [ i , j ]
}
}
% E s t i m a t e t h e a v e r a g e v a l u e o f t o t a l e r r o r o f each s e n s o r
E . a v e r a g e < colMeans (E)
% s o r t () function return array of average t o t a l error of a l l
measured s e n s o r s i n i n c r e a s i n g o r d e r
E . a v e r a g e . sort < sort ( data . frame (E . a v e r a g e ) , d e c r e a s i n g=
FALSE)
l i s t . s e n s o r s < colnames (E . a v e r a g e . sort [ c ( 1 : n ) ] )

%Return d a t a o f n s e l e c t e d s e n s o r s
return ( data [ , c ( l i s t . s e n s o r s ) ] )
}

39
Appendix C

Source code for Data Imputation

function d a t a i m p u t a t i o n (m, o l d d a t a , o r d e r , samples )

{
%m: d a t a o f s e l e c t e d s e n s o r , o l d d a t a : h i s t o r y d a t a o f
s e l e c t e d s e n s o r s , s a m p l e s : t h e number o f p r e d i c t e d v a l u e
by ARIMA model
model < arimamethod (m, o r d e r )
pdata < m
i < 1, j < 1
avgrow < m[ , n c o l (m) ]
counter < 1

% Offline part
% C o n t r i b u t e arima model f o r each s e n s o r i n s e l e c t e d s e n s o r s
from Error Computation
for ( c i n 1 : n c o l (m) )
{
f i t < arima ( t s ( o l d d a t a [ , j ] ) , o r d e r )
pdata [ , c ] < f o r e c a s t . Arima ( f i t , samples ) $upper
[ ,2]
}
% This can be s e p a r a t e d i n Online p a r t
% Recover empty v a l u e s
for ( i i n 1 : nrow (m) )
{
for ( j i n 1 : n c o l (m) )
{
v a l < m[ i , j ]

% Check which i s t h e empty v a l u e

i f ( ! i s . na ( v a l ) )

40
 next
else
{
i f ( i == 1 | | i == 2 )
{
k < 1
while ( k<=10)
{
i f ( ! i s . na (m[ k+1, j ] ) )
{
m[ i , j ] < m[ k+1,
j]
break
}
k = k+1
}
}

% arg1 , arg2 , arg3 , are c o e f f i c i e n t s o f ARIMA

model
% p v a l [ 1 ] , p v a l [ 2 ] i s t h e 2 p r e v i o u s empty
value of current value
else
{
arg1 < a s . numeric ( model [ 1 , j ] )
arg2 < a s . numeric ( model [ 2 , j ] )
a r g 3 < a s . numeric ( model [ 3 , j ] )

x < 1
pval < c ()
k < i

% Find 2 p r e v i o u s non empty v a l u e

while ( x <= o r d e r [ 1 ] && k>=2)
{
i f ( ! i s . na (m[ k 1, j ] ) )
{
p v a l [ x ] < m[ k 1, j ]
x = x+1
}
k = k 1
}

41
 pval < a s . numeric ( p v a l )

%Apply arima model e q u a t i o n

i f ( ! i s . na ( a r g 1 ) && ! i s . na ( a r g 2 ) )
{
val < arg1 ⇤ pval [ 1 ] + arg2 ⇤ pval [ 2 ] +
a r g 3 ⇤( p v a l [1] p v a l [ 2 ] )
i f ( v a l > 0 . 8 ⇤ rowMeans (m[ i 1 , ] ) )
m[ i , j ] < v a l
else
m[ i , j ] < rowMeans (m[ i
1 ,])
}
else
m[ i , j ] < pdata [ i , j ]

}
}
}
}
return (m)
}

42
Appendix D

Source code for MMEM algorithm

function e s t i m a t e = MMEM( data , r e f e r )

{
% ‘ data ’ i s a m a t r i x i n c l u d e s e n s i n g d a t a from a l l measured
s e n s o r , a v e r a g e v a l u e ( form AM) , f i t t i e d v a l u e ( from FM)
nrow < nrow ( data )
n c o l < n c o l ( data )

%I n i t i a l t h e e r r o r a r r a y
e < matrix ( data = NA, nrow , n c o l )

estimate < c ()
for ( i i n 1 : nrow )
{
for ( j i n 1 : n c o l )
{
%C a l c u l a t e A b s o l u t e Error b e t w e e n s t e m p o r a l e s t i m a t e d
d a t a and r e f e r e n t d a t a
e [ i , j ] < abs ( data [ i , j ] r e f e r )
}
% which . min ( ) f u n c t i o n r e t u r n i n d e x o f minimum v a l u e
min < which . min( e [ i , ] )

estimate [ i ] < data [ i , min ]

}
%Return E s t i m a t e d v a l u e
return ( e s t i m a t e )
}

43
Appendix E

Source code for FM algorithm

function FM( o l d d a t a , data )

{
% C o n t r i b u t e f i t t i n g f u n c t i o n from h i s t o r y d a t a o f n
s e l e c t e d s e n s o r from Data A b s t r a c t i n g b l o c k i n
p r e v i o u s day
formula < ” olddata [ , 9 ] ˜ 0 ”

nrow < nrow ( data )

ncol < n c o l ( data )

m< data

for ( j i n 1 : n c o l )
{
f o r m u l a < s t r c ( formula ,”+ o l d d a t a [ , ” , j , ” ] ” )
}
%f i t : contains f i t t i n g f unc tio n
f i t < lm ( a s . f o r m u l a ( f o r m u l a ) )

%p r e d i c t new t e m p e r a t u r e o f each s e l e c t e d s e n s o r i n
c u r r e n t day
d . f i t < p r e d i c t ( f i t ,m)

%m: i n c l u d e d f i t t e d v a l u e s and s e n s o r s ’ r e a d i n g s o f n
selected sensors
m < c b i n d (m, d . f i t )
colnames (m) < c ( colnames ( data ) , ”FM” )

return (m)
}

44
Bibliography

[1] E.A. Lee, “Cyber-Physical Systems Design Challenge,” International Symposium

on Object/Component/Service-Oriented Real-Time Distributed Computing (ISORC),
pp.363-369, 2008.

[2] S.Sharifi, “Accurate Temperature Estimation for Efficient Thermal Management,”

IEEE 9th Quality Electronic Design (ISQED), pp.137-142, 2008.

[3] F. Yan, “Simulation of Indoor Dynamic Thermal Comfort based on Computational

Fluid Dynamic,” IEEE Intelligent Computation Technology and Automation (ICI-
CTA), pp.860-865, 2011.

[4] M.Jing,“Research of high precision quartz temperature sensor system based on data
fusion technology,” IEEE Measurement, Information and Control (ICMIC), pp.1051-
1054, 2013

[5] Z. Cheng, Y. Tan, and A.O. Lim, “Fitting method for hybrid temperature control in
smart home environment,” IEEE International Conference on Modelling, Identifica-
tion and Control (ICMIC),pp.300-305,2014.

[6] G.Andrea, S.Giandomenicoand, and V.Andrea, “Rainbow: an Intelligent Platform

for Large-Scale Networked Cyber-Physical Systems,” International Workshop on
Networks of Cooperating Objects for Smart Cities (UBICITEC), pp.70-85, 2004.

[7] K.A.A. Makinwa, “Smart Temperature Sensors in Standard CMOS,” Procedia En-
gineering, pp. 930-939, Sept. 2010.

[8] M. Maasoumy, Q. Zhu, C. Li, F. Meggers, and A. Sangiovanni Vincentelli, “Co-design

of control algorithm and embedded platform for building HVAC systems,” In Proc.
of the ACM/IEEE International Conference on Cyber-Physical Systems, pp. 61-70,
2013

[9] H.Sandberg, M.Rabi, M.Skoglund and K.H.Johansson, “Estimation over Heteroge-

neous Sensor Network”, In Proc. of the 47th IEEE Conference on Decision and Con-
trol, pp.4898-4903, 2008

[10] W.Elmenreich, R. Leidenfrost, “Fusion of Heterogeneous Sensors Data,” IEEE Intel-

ligent Solutions in Embedded Systems, pp.1-10, 2008

45
[11] N.Nower, Y.Tan, Y.Lim, “Efficient Temporal and Spatial Data Recovery Scheme
for Stochastic and Incomplete Feedback Data of Cyber-physical Systems,” IEEE 8th
International Symposium on Service Oriented System Engineering (SOSE), pp.192-
197, 2014

[12] R.H.Shumway, S.S.David, “Time Series Analysis and Its Applications,” Springer Text
in Statistics, pp.83-171, 2010

[13] B.G. LipTk, “Instrument Engineer Handbook: Process Control and Optimization,”
CRC Press Taylor and Fracis Group, 4th ed., 2006

[14] R.H.Shumway, S.S.David, “Vector Autoregressive Models for Multivariate Time Se-
ries,” Springer Modeling Financial Time Series with S-PLUS, pp.385-429, 20006

[15] M.C.Vurcan, O.B.Akan, I.F.Akyildiz, “Spatio-temporal correlation: theory and ap-

plications for wireless sensor networks,” Springer Sensor Networks and Configuration,
pp.75-98, 2007

[16] J.Yick, B.Mukherjee, D.Ghosal, “Wireless sensor network survey,” Computer Net-
works, International Journal of Computer and Telecommunications Networking,
ISSN: 1389-1286, pp.2292-2330, 2008

[17] P.Samundiswary, P.Priyadarshini, P.Dananjayan, “Performance Evaluation of Het-

erogeneous Sensor Networks,” IEEE International Conference on Future Computer
and Communication (ICFCC), pp.264 - 267, 2009

[18] Thomas D.McGee, “Principles and methods of temperature measurement,” Willey-

Interscience Publication, 1988

[19] Robin E. Bentley, “Theory and Practice of Thermoelectric Thermometry,” Springer

Handbook of Temperature Measurement Vol.3, 1998

[20] R.J. Robles, T.Kim, “Applications, Systems and Methods in Smart Home Technol-
ogy: A Review,” SERSC International Journal of Advanced Science and Technology
Vol.15, pp.37-47, 2010

[21] Z.Hong-bing, Z.Zhe-zhao, “A fitting Method of the Temperature Characteristic Curve

of Sensor,” IEEE, Wireless Communications, Networking and Mobile Computing, 5th
International Conference on, pp.1-4, 2009

[22] A. Gunawardana, G.Shani, “A Survey of Accuracy Evaluation Metrics of Recommen-

dation Tasks,” Journal of Machine Learning Research (JMLR) Vol.10, pp.2935-2962,
2009

[23] J. Wan, H. Yan, H. Suo and F. Li, “Advances in cyber-physical systems research,?
KSII Transactions on Internet and Information Systems, vol. 5, no. 11, pp.1891-
1908, 2011

46
[24] F. Xia, X. Kong and Z. Xu, “Cyber-physical control over wireless sensor and actuator
networks with packet loss,” Wireless Networking Based Control, Springer, pp. 85-102,
2011

[25] K. Wan, K.L. Man and D. Hughes, “Specification, analyzing challenges and ap-
proaches for cyber-physical systems (CPS),? Engineering Letters, vol. 18, no. 3, pp.
308-305, 2010

[26] E.A. Lee, “CPS foundations,” ACM/IEEE Design Automation Conference (DAC),
Anaheim, USA, pp.737-742, 2010

[27] F.Xia, T.Mukherjee, Y.Zhang and Y.Song, “Sensor Networks for High-Confidence
Cyber-Physical Systems,” Int. J. of Distributed Sensor Networks Vol.201, Article ID
245734, 2011

[28] R.J. Hyndman, “Yule-Walker estimators for continuous-time autoregressive models?,

J. of Time Series Analysis, vol.14, pp.281?296, 1993

[29] A. Azadeh, S.M. Asadzadeh, R.J. Marandi, S.N. Shirkouhi, G.B. Khoshjhou and S.
Talebi, “Optimal estimation of missing values in randomized complete block design
by genetic algorithm,” Knowledge-Based Systems, Elsevier, vol.37, pp.3747, 2013

[30] C.J. Wilmott, K. Matsuura, “Advantages of the mean absolute error (MAE) over root
means square (RMSE) in assessing average model performance,” Climate Research,
vol.30, pp.79-82, 2005

47