An intelligent slope disaster prediction and monitoring system based on WSN and ANP

(1)

An intelligent slope disaster prediction and monitoring system based

on WSN and ANP

Che-I Wu

a,⇑

_{, Hsu-Yang Kung}

b

_{, Chi-Hua Chen}

c,d

_{, Li-Chia Kuo}

b

a

Computer Center, National Pingtung University of Science and Technology, Pingtung 912, Taiwan, ROC

b

Department of Management Information Systems, National Pingtung University of Science and Technology, Pingtung 912, Taiwan, ROC

c

Institute of Information Management, National Chiao Tung University, Hsinchu 300, Taiwan, ROC

d

Telecommunication Laboratories, Chunghwa Telecom Co., Ltd, Taoyuan 326, Taiwan, ROC

a r t i c l e

i n f o

Keywords:

Wireless Sensor Networks Analytic Network Process Disaster prediction model

Portrait-based Disaster Alerting System

a b s t r a c t

Taiwan generally has large-scale landslides and torrential rainfall during the typhoon season. As Wireless Sensor Networks (WSN) and mobile communication technologies advance rapidly, state-of-the-art tech-nologies are adopted to build a model to reliably predict and monitor disasters, as well as accumulate environmental variation-related information. By integrating WSN and Analytic Network Process (ANP), this study evaluates the weight of disaster factors that adopt the consistency index of pair comparisons on hillslopes. The weight estimation and classiﬁcation of disaster factors are based on the K-means model to build the hillslope prediction model. The Portrait-based Disaster Alerting System (PDAS) is designed and implemented using the proposed disaster prediction model. The PDAS adopts Web-GIS to visualize the environmental information. Evaluation results of the system indicate that the proposed prediction model achieves more accurate disaster determination than the conventional method.

1. Introduction

Increasing numbers of natural calamities have occurred in Taiwan in recent years. These disasters often cause serious natural destruction after torrential rainfall or earthquakes, causing heavy losses to people’s lives and property. The KALMAEGI typhoon dam-aged many parts of Taiwan in July 2008, due to the heavy torrential rainfall. Moreover, the survivors did not know how to report the disaster promptly to the authorities, so rescue teams could not locate the survivors. Therefore, wireless hand-held devices are required to transmit the multimedia information of disasters, such as images, sounds and characters.

Taiwan generally has large-scale slope failure and torrential rainfall to cause sediment disaster during the typhoon season. Those disasters often result in the serious nature destruction and create the heavy losses of people’s lives and properties. The kind of hillslopes disaster is numerous, and this mainly discussing topic of thesis choose often appear slope failure for sediment disaster to study in Taiwan. It expects to make discussion with the topic this thesis that can help people and prevention and rescuing units to prevent and alarm creating disaster. It was usually the gold period to prevent and rescue disaster with taking place before and creat-ing at that time. It had taken place relevant disasters that all the

materials were afterwards to collect and study, and judge causing disaster factors in the past, the time has already had no enough to save a critical situation. Therefore, this research study and analyze the past slope failure as basis of consulting, and combining new information technology to propose two major system themes, which are prediction supporting model and awareness monitoring system, to assist and solve problems of disaster. At ﬁrst the mainly causing disaster factors of slope failure must be discussed and se-lected, so survey and examine the trial zone environment in thesis research. Numerous environment causing disaster factors will be chosen, assessed, analyzed, then select seven causing factors which include gradient, soil characteristics, 24-h accumulated rainfall, vegetation index, soil displacement, soil hydrous and temperature, to cause slope failure. Then according to values of selecting disaster factors into designed prediction supporting model, the model sys-tem will assess and analyze the taking place disaster grade and possibility. In order to reach and develop early alarming effect, the prediction supporting model can make sure disaster preventing and alarming functions really, and the study also plans comple-mented monitoring and transmitting tools. The system will utilize these monitoring and transmitting functions to complete effect of pre-warning and informing immediately.

This study also mainly proposes and designs a real-time disas-ter information system, which is important for people to develop PDAS to assist the prevention disaster works, to obtain, inform, and display the disaster situation. In order to achieve forecasting

http://dx.doi.org/10.1016/j.eswa.2013.12.049

⇑Corresponding author. Tel.: +886 87703202. E-mail address:joey@mail.npust.edu.tw(C.-I. Wu).

Contents lists available atScienceDirect

Expert Systems with Applications

(2)

and monitoring disaster functions, PDAS is also implemented using the proposed disaster prediction supporting model, which predic-tion efﬁciency model includes Analytic Network Process (ANP), Back-Propagation Neural Network Analysis (BPN), and Multivariate Statistical Analysis (MSA), to compare the adaptable model.

As Wireless Sensor Networks (WSN) and mobile communica-tion technologies advance rapidly, state-of-the-art technologies are adopted to build a model to reliably predict and monitor disas-ters, as well as accumulate environmental variation-related infor-mation. The PDAS also combines multimedia transmission technology and quality of service (QoS) mechanism to reveal the real disaster situations, for example, the accurate position and the real-time image/video of accident events. Heterogeneous Net-work Users use the handheld devices to transmit and receive mul-timedia information about slope failure via the wireless/mobile and internet communications.

Accordingly, this study adopts Embedded Multimedia Communi-cation technology to design the Portrait-based Disaster Alerting Sys-tem (PDAS) in order to solve the space and time limitations. Users can use the hand-held devices with high mobility via wireless net-work (3G/GPRS/GSM) to obtain disaster multimedia stream service (Castillo-Effer, Quintela, Moreno, Jordan, & Westhoff, 2004). Addi-tionally, this investigation also combines customized services, Loca-tion-Aware Service, Wireless Sensor Network, Multicast, Web GIS, Intelligent Agents and Analytic Network Process (ANP) (Neaupane & Piantanakulchai, 2006). The PDAS transmits the sensing and pre-diction information of the monitored area to the database system to analyze the data. Additionally, geographical information system (GIS) technology combines the analysis system and alarming mech-anisms to operate the model. The detected materials then accede to ANP model to appraise, analyze and process sensing hillslopes disas-ter factor data. Finally, a warning message based on the analytical results is released to mark where the victim stays immediately on the Web-GIS layer. This information would inform the prevention and relief personnel about the disaster area clearly and quickly. 2. Research background and theory discussion

The Portrait-based Disaster Alerting System, (PDAS) is designed to provide (i) mobile user (MU), (ii) Hillslope Monitoring Sensor (HMS), (iii) Integrated Service Server (ISS), and (iv) Intelligent Hill-slope Decision System (IHDS). The PDAS offers sensing and predict-ing information of the disaster area. Required research background and relevant technology for this study are (1) Geographic Informa-tion System (GIS) and (2) Wireless Sensor Networks (WSN). 2.1. Geographic Information System (GIS)

GIS develops geographical coordinate information that assesses space distribution and database management technology, as well

as combines systems such as geographical mathematics and map surveying. GIS has two parts, namely subject and operation. Geo-graphical information systems are adopted to store several differ-ent geographical information, there are two types including raster and vector. Digital geographical materials stored in geo-graphical information databases are classiﬁed as Spatial Data, Geography Data, and Attribute Data (ESRI, 1996, 2000). Fig. 1 shows the operation of a GIS.

2.2. Wireless Sensor Networks (WSN)

Recently developed sensors can not only detect the goal and change of the environment, but also handle the collected data. However, some problems need to be considered. If a base station is far from sensors, then the sensors need to adopt the routing net-work method so that a lot of sensors group a path to transfer mate-rials to the base station (Evans-Pughe, 2003). Additionally, the battery of sensors may not be replaceable, energy that is consid-ered indispensable needs to be controlled when conﬁguring sensor design and network management (Akyildiz, Weilian, Sankarasubr-amaniam, & Cayirci, 2002). The hardware structure of the sensor comprises four major parts, (i) sensing unit, (ii) processing unit, (iii) transceiver unit, and (iv) power unit. Sensors can be adopted in location systems, mobilizers and power generators. The system adopts an MTS420 sensor to accumulate temperature, and an MDA300 sensor to measure the soil water content (ECHO 10) and temperature (108-L) (Ruiz & Loureiro, 2003). A TMOTE SKY dis-placement sensor is adopted to obtain slope materials instantly, to combine and analyze ANP model to obtain the disaster weight, and to judge the probability of disaster at any time, as shown in Fig. 2.

3. System design

The PDAS is a four-tier system as shown inFig. 3. Users can uti-lize various terminal devices that include PC, notebook, Tablet PC, 3G/4G mobile phone and personal digital assistant (PDA) to access PDAS. The Portrait-based Disaster Alerting System, (PDAS) is de-signed to provide (i) mobile user (MU), (ii) Hillslope Monitoring Sen-sor (HMS), (iii) Integrated Service Server (ISS), and (iv) Intelligent Hillslope Decision System (IHDS).

3.1. Mobile user site

Mobile user’ sites provide functions such as location-based ser-vice, customized serser-vice, heterogeneous networks, web-based GIS. mobile users can adopt the terminal servers of hand-held devices to link mobile communication network or internet network, to login and use PDAS systems, and to perform GPS to locate

(3)

Location-Aware Services automatically. PDAS has a user position system to provide customized services at time, including different devices to offer different pictures, and information about the deb-ris-ﬂow and rainfall in the disaster area. Since users adopt different network protocols, the system lets mobile equipment integrate dif-ferent environments (Fujiwara, Makie, & Watanabe, 2004), such as GSM, GPRS, internet and IEEE802.11x. Web GIS is similar to web mapping, with an emphasis on analysis, processing of project spe-ciﬁc geo-data and exploratory aspects.

3.2. Hillslopes monitoring sensor site

The research has a sensing area located on the hillslopes abla-tion test zone at Naabla-tional Pingtung University of Science and Tech-nology. The area has 4 level slopes, and it can be divided into vegetation indexes 10%, 60%, and 80%. We choose vegetation index 10% to observe area A and vegetation index 60% for area B, shown asFig. 4.

Fig. 5shows the deployment diagram of the sensing area. Com-munication platforms were assigned by Tmote Sky (MoteIV corpo-ration) and NPR2400 (Crossbow Technology Micaz). Tmote Sky with KXM52-1050 (Kionix) Displacement was obtained by Tri-Axis Accelerometers; soil temperature was measured by 108-L Temper-ature Probe (Campbell Scientiﬁc, Inc); soil hydrous values were collected by ECH2O (Decagon Devices, Inc) with MDA300CA Sensor

Board, and position was judged by GPS with MTS420 (Crossbow Technology), as shown inFig. 6. The common limitation of most monitoring systems is that sensors that are destroyed by the large-scale disaster cannot send data back to the database server. Therefore, the proposed system sends data back to the base station every second. Moreover, to ensure that the WSN continues

(a) MDA300 (b) ECHO-10 (c) MRP4200 (d) 108-L

Fig. 2. Sensor devices.

Fig. 3. Architecture of PDAS system.

(4)

working successfully, the WSN power system is designed to con-join power and battery (Lindsey & Raghavendra, 2002).

3.3. Integrated service server

The Integrated service server comprises two parts: (i) the web server providing web-based information service based over heter-ogeneous network environments, such as GPRS, GSM, IEEE 802.11b and internet, and (ii) the Intelligent Agents providing mobile cli-ents with customized information. Multimedia server provides functions such as Intelligent Agents, Real-time Disasters Alters, Virtual Reality of Multimedia, Real-time Disaster Monitoring, Disasters Multimedia Management, Multicast Chat Room, Web Service and Web GIS.

3.3.1. Intelligent Agents

The intelligent agent performs functions such as collecting, searching, classifying and processing data, enabling users can ob-tain real-time information (Wooldridge, 1995). The intelligent agent system has the following six parts:

(i) Sensor network monitor agent (SNMA), which accumulates the sensor information for experts to analyze and research; (ii) Position sensation agent (PSA), which selects user

coordi-nates to store on the database server;

(iii) Prediction and notiﬁcation agent (PNA), which predicts disasters, and relays them to users;

(iv) News agent (NA), which catches and ﬁlters news about deb-ris ﬂow news to mobile clients;

(v) Rainfall information agent (RIA), which monitors informa-tion about the precipitainforma-tion stainforma-tion to users at any time, and (vi) Video monitor agent (VMA), which monitors the IP-CAM in sensing zones, and sends this information to users at any time.

3.3.2. Real-time Disaster Monitor

PDAS integrates IP Cameras of road monitoring system of disas-ter areas in Pingtung County to expand the controlling range of the disaster. The system adopts an encoder to compress huge multi-media materials into appropriate sizes, and provides a steady stream service and effective real-time control service (Chang & Guo, 2006).

3.3.3. Disaster Multimedia Management

Mobile users can take pictures and videos of a disaster situation by 3G handheld devices. The users then store these ﬁles, and trans-mit then to the server through the mobile and wireless communi-cation network. Moreover, PDAS establishes the image management agent to safeguard the information management of historical images in the Internet.

3.4. Intelligent hillslopes decision system

The IHDS monitors hillslopes conditions; predicts hillslopes de-gree of hazard; accurately identiﬁes the present condition, and provides suitable emergency measures. IHDS comprises two parts, the (1) Hillslopes Disaster Inference Engine; and the (2) Emergency Action Inference Engine.

3.4.1. Hillslopes Disaster Inference Engine

This engine deﬁnes seven disaster factors, namely including gradient, soil characteristics, 24-h accumulated rainfall, vegetation index, soil displacement, soil hydrous and soil temperature, to evaluate and analyze the probability of hillslopes disaster.Table 1 shows the classiﬁcation of disaster factors.

The hillslope prediction model was built according to the ANP model. The research samples are the monitoring results in Nantou County from 2006 to 2008. To increase the model accuracy, all 20 samples were chosen from three degrees for testing to avoid bias towards a particular degree.

The ANP model compares in pairs with relative weight estima-tion and formaestima-tion of supermatrix, and adopts supermatrix weight of which is two factors in seven disaster factors to choose the best way (Chen, Lin, Chang, Ho, & Lo, 2013; Saaty, 1980, 1996). There-fore, this rule that is the ANP comparing two relative weights is fol-lowed to build the most adaptive model, as shown inFig. 7.

The model process is divided into four steps as follows: Step I: Pair comparison with relative weight estimation

Table 2shows a sample of a questionnaire with cluster compar-ison, which is scored using a sample calculation with relative weight and consistency index. The geography expert ﬁlls in the ta-ble to deﬁne the weight of each disaster factor. The pair compari-son formula is shown below:

Fig. 5. WSN deploying diagram.

(5)

€ A ¼ 1 a12 a1n 1=a12 1 a2n .. . .. . . . . .. . 1=a1n 1=a2n .. . 1 2 6 6 6 6 6 4 3 7 7 7 7 7 5 ¼ w1=w2 w1=w2 w1=wn w2=w1 w2=w2 w2=wn .. . .. . . . . .. . wn=w1 wn=w2 .. . wn=wn 2 6 6 6 6 6 4 3 7 7 7 7 7 5 ð1Þ

Wi: weight of the main element i; i = 1, 2, 3, . . ., n; aij: ratio between

two main elements; i, j = 1, 2, 3, . . ., n

aij¼ 1=aji and aik¼ aij ajk ð2Þ W ¼ ½wk; where wk¼ Xn j¼1 a0 ij=n ð3Þ

While a pair comparative matrix is a positively reciprocal ma-trix, a policymaker that compares the values in the pair matrix can-not easily reach an identical situation of main elements. Therefore, the system must adopt examination of consistency to obtain the consistency index (CI); to ﬁlter the information, and to guarantee that the calculation results reﬂects actual conditions.

The aijvalue in the positive reciprocal matrix changes with even

a very small change in the kmaxvalue. Hence, between two

differ-ence intensity of kmaxand n consistency level of commenting

crite-rion can be determined. The deﬁnition formula is given below:

C:I: ¼kmax n n 1

n : number of assessing element

ð4Þ

The systems policymaker judges consistency as C.I. = 0, but shows inconsistency if C.I. > 0.1.Saaty (1980)suggested that the deviation value is acceptable if C.I. 6 0.1.

Step II: Formation of initial supermatrix

Table 3shows the weight of each disaster factor and eigenvec-tor, which are calculated from the data inTable 2. The consistency index indicates that the true result of calculation is guaranteed to respond to actual conditions. The weight that can be inserted into the order supermatrix forms the first supermatrix.Fig. 8shows the first supermatrix obtained in calculating the weight values and fill-ing them in order, as in Step 1.

Step III: Formation of weight supermatrix

The Step II initial supermatrix is transformed to a matrix in which each of its columns sums to unity.

Step IV: Limiting supermatrix

The ﬁnal supermatrix is obtained when a high stable value of disaster factors to multiply matrix repeatedly (2 (n 1) multiplica-tions each), then we get the terminal supermatrix.Fig. 9shows the process of deriving the weight of the supermatrix from the ele-ment-by-element multiplication of the initial supermatrix and the limiting supermatrix with global priority weights. The weight of the supermatrix is raised to the power weight until its conver-gence using MATLAB.

Through limiting supermatrix table substitutes our training disaster factor samples, all of the samples are convert to weights in the range 0.1255–0.4836. For instance, consider a sample belonging to Gradient level 6, implying sandy soilsand, rainfall above 450 mm, loose vegetation, soil displacement below 40 cm, soil hydrosity above 80%, and soil temperature above 50. The weight of this sample is 0.0596 + 0.0464 + 0.126 + 0.0652 + 0.0093 + 0.0664 + 0.0138 = 0.3867. Various classiﬁcations are

Table 1

Disaster factors classiﬁcation.

Factors Gradient Soil characteristic Rainfall (mm) Vegetation index Soil displacement (cm) Soil hydrous (%) Soil temperature (°C) Class Level 7 Gravel >450 Loose >100 >80 >50

Level 6 Sand 350–450 Medium 70–100 60–80 30–50 Level 5 Silt 250–350 Dense 40–70 20–60 15–30 Level 4 Clay <250 Extremely dense 0–40 0–20 <15 Level 3

(6)

available, including natural breaks, quartile, equal intervals and K-means.

3.4.2. Emergency Action Inference Engine (EAIE)

The EAIE receives the requests of emergency action measures from the Hillslopes Disaster Inference Engine, and checks whether current environmental conditions and hillslopes level correspond to the hillslopes emergency action measures in the knowledge base.Table 4shows the actions of the EAIE (Sharpe, 1938). 4. System evaluation

Most previous researches have adopted Back-Propagation Neu-ral Network (BPN) and Multivariate Statistical Analysis (MSA) to estimate and predict hillslopes disasters. Therefore, this study

compares prediction efﬁciency of the proposed hillslopes predic-tion model derived from Analytic Network Process with a model built by BPN and MSA.

4.1. Back-Propagation Neural Network

The following inference factors significantly influence the learn-ing efficiency and convergence of the BPN algorithm: (a) the num-ber of learning cycles, (b) the learning rate, (c) the numnum-ber of network layers and (d) the number of neurons in each hidden layer (Kung, Chen, & Ku, 2012; Lin & Chen, 2013; Lo et al., 2011; Skapura, 1995). In general, too many learning cycles, network layers and neurons of hidden layers lead over-learning and high error rates. 4.2. Multivariate Statistical Analysis

Multivariate Statistical Analysis evaluates variable value and variable characteristics of inﬂuence factors (Varnes, 1978). The analysis model ﬁrst calculates each greater variable value (V) for each higher hillslopes disaster, as shown in formula(5). The weight of each variable is calculated as in formula(6). The respective de-gree index (DI), which the instability index (Dt), is calculated from the variable weights. The analysis considers the main factors, and then in order to build the MSA, shown as formula(7) and (8).

V ¼

r

X 100%

r

:Standard deviation

X : Destruction percentage average of each factor ð5Þ Wi¼ Vi V1þ V2þ V3þ Vn ;i ¼ 1 . . . n ð6Þ di¼ 9ðXi XminÞ ðXmax XminÞ þ 1 ð7Þ Dt ¼ Ss0:18 Rn0:13 Nd0:15 Ge0:15 Gm0:09 Sw0:15 St0:15 ð8Þ Table 3

Vector of each factor.

Factor G SC RN NC SD SH ST Eigenvector G 1 6 5 3 1/6 3 3 0.213523 SC 1/6 1 1/3 1/3 1/9 1/3 1 0.034842 RN 1/5 3 1 3 1/4 1 2 0.103668 NC 1/3 3 1/3 1 1/6 1/3 2 0.065188 SD 6 9 4 6 1 3 9 0.427613 SH 1/3 3 1 3 1/3 1 2 0.11151 ST 1/3 1 1/2 1/2 1/9 1/2 1 0.043656 kmax= 7.773956 C.R. = 0.097722

Fig. 8. Formation of initial supermatrix. Table 2

ANP factor weight questionnaire.

Scale of relative importance (Saaty, 1980)

Factor-soil slope Evaluation

1 2 3 4 5 6 7 8 9

Soil slope is much more important than soil category h h h h h j h h h Soil slope is much more important than rain h h h h j h h h h Soil slope is much more important than natural discovery h h j h h h h h h Soil displacement is much more important than soil slope h h h h h j h h h Soil slope is much more important than soil hydrous h h j h h h h h h Soil slope is much more important than soil temperature h h j h h h h h h 1: Equal importance; 3: moderate importance; 5: strong importance, 7: very strong or demonstrated importance; 9: extreme importance 2, 4, 6, 8: reciprocals of above.

(7)

4.3. Prediction efﬁciency

To achieve the same conditions for building each modes, the contrastive model was built by adopting the same 60 data that were adopted to train the hillslope prediction model. The predic-tion model presents the complete interacpredic-tion of input variables, and has a high accuracy.

Figs. 10–12show the corresponding trend diagrams of the ac-tual hillslope levels and those inferred from the BPN, MSA and pro-posed ANP hillslope prediction models, respectively. As indicated

inFigs. 10 and 11, the actual changes in hillslope levels did not overlap with the inferred levels from the traditional model. Analyt-ical results indicate that the traditional model had a low accuracy. In contrast, the actual hillslopes levels overlapped well with those inferred by the proposed drought forecast model, as indicated in Fig. 12.

Tables 5–7show the confusion matrices of the inferred hillslope levels from the MSA, BPN and ANP models, respectively. The total correct rate for predicting the hillslopes level of the coming day by the traditional BPN and MSA models were 61.67% and 81.67%, respectively, while that of the proposed model was over 88.33% after training. Thus, these results indicate that the ANP model higher hillslope prediction efﬁciency than the traditional BPN and MSA models.

5. Conclusions and future work

This study designs and develops a ‘‘Portrait-based Disaster Alerting System (PDAS) with Wireless Sensors’’ that can be adopted at any time and any place. The PDAS can be adopted in mobile de-vices, such as PDAs, laptop computers and 3G smart phones. Users of the system can not only adopt the customized services and interface to inquire and view those disaster data easily, but also ob-tain rich real-time multimedia by these devices. In summary, the PDAS has at least the following four advantages: (1) PDAS can transfer real-time multimedia and urgent messages to users

Fig. 9. The limiting supermatrix with global priority weights.

Table 4

Emergency Action Inference.

Slope degree Emergency Action Inference Safe area Conserve natural environment

Dangerous area Continuing to observe this area, and strengthen the draining water facilities and plants to protect slopes

Very dangerous area As well as the above actions, coordinate other approaches, such as building the retaining wall, and setting up the ground anchor

0 1 2 3 4 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 Serial Number of Testing samples

H il ls lope de gr e e

Practical Hillslope Level Inferred Hillslope Level from BPN

Fig. 10. The corresponding trend diagram of practical hillslopes level and inferred hillslopes level from the BNP model.

(8)

through mobile communication. (2) PDAS provides real-time mul-timedia for users by adopting ‘‘Mulmul-timedia Streaming for Embed-ded Applications Technique’’, ‘‘Adaptive Multimedia QoS Technique’’ and ‘‘Image Compression Technique for MPEG-J’’. (3)

PDAS provides the information about disasters, and sends the GPS coordinates of the disaster area to Disaster Prevention and Re-sponse Center to reduce the effect of disasters. (4) Historical disas-ter data is analyzed using the ANP model, enabling users to predict and prevent disasters. (5) An integrated application of new tech-nology is to be used for disaster prevention. (6) A combination of disasters statistical clustering classiﬁcation collects disaster data are divided into several different grades of disaster.

But future research of the study will add (1) planning escape paths, (2) wireless network provisioning technology solutions, and (3) solving power problem for sensors to achieve a more com-plete prediction system.

This study also mainly proposes and designs a real-time disas-ter information system, which is important for people to develop PDAS to assist the prevention disaster works, to obtain, inform, and display the disaster situation. In order to achieve forecasting and monitoring disaster functions, PDAS is also implemented using the proposed disaster prediction supporting model, which predic-tion efﬁciency model includes Analytic Network Process (ANP), Back-Propagation Neural Network Analysis (BPN), and Multivariate Statistical Analysis (MSA), to compare the adaptable model. The PDAS integrates technology and management parts to analyze the factors of slope failure, scheme forecasting model, and estab-lish monitoring and informing multimedia system.

Hopefully, the PDAS will become an integrated disaster infor-mation system for prevention and relief of mudslide disasters. By adopting the Multimedia Recognition technique’’, PDAS can intelli-gently monitoring disasters automatically, reducing the casualty rate in combination with the Medical Information System. Sensors can thus accumulate accurate disaster data for the ANP model. The PDAS has also integration services of prediction supporting model, awareness monitoring system, and establishing real-time disaster informing system, this thesis can reach the comprehension effects of prevention and rescuing information function.

H ills lo p e d eg re e

Practical Hillslope Level Inferred Hillslope Level from MSA

Fig. 11. The corresponding trend diagram of practical hillslopes level and inferred hillslopes level from the MSA model.

H ills lo p e d eg re e

Practical Hillslope Level Inferred Hillslope Level from ANP

Fig. 12. The corresponding trend diagram of practical hillslopes level and inferred hillslopes level from the ANP model.

Table 5

The confusion matrix of inferred hillslope level from the MSA model. Desire Target

Safe area Dangerous area Very dangerous area Total Total error rate

Safe area 11 16 0 27 0.38333

Dangerous area 0 14 0 14 Total correct rate

Very dangerous area 0 7 13 20

Total 11 37 13 0.61667

Table 6

The confusion matrix of inferred hillslopes level from the BPN model. Desire Target

Safe area 25 1 1 27 0.18333

Total 25 7 28 0.81667

Table 7

The confusion matrix of inferred hillslope levels from the ANP model. Desire Target

Safe area 25 0 0 25 0.11667

(9)

References

Akyildiz, I. F., Weilian, S., Sankarasubramaniam, Y., & Cayirci, E. (2002). A survey on

sensor networks. IEEE Communications Magazine, 40(8), 102–114.

Castillo-Effer, M., Quintela, D. H., Moreno, W., Jordan, R., & Westhoff, W. (2004). Wireless sensor networks for flash-flood alerting. In Proceedings of the fifth IEEE international Caracas conference on devices, circuits and systems, 4–5 November (pp. 142–146). Caracas, Venezuela.

Chang, N. B., & Guo, D. H. (2006). Urban ﬂash ﬂood monitoring, mapping, and forecasting via a tailored sensor network system. In Proceedings of the 2006 IEEE international Caracas conference on networking, sensing and control, 23–25 April (pp. 757–761). Florida, USA.

Chen, C. H., Lin, H. F., Chang, H. C., Ho, P. H., & Lo, C. C. (2013). An analytical framework of a deployment strategy for cloud computing services: A case study

of academic websites. Mathematical Problems in Engineering, 2013, 1–14.

ESRI (1996). Using the ArcView spatial analyst (pp. 1–143). USA: ESRI. ESRI (2000). Using ArcMap (pp. 3–487). USA: ESRI.

Evans-Pughe, C. (2003). ZigBee wireless standard. IEEE Review, 49(3), 28–31.

Fujiwara, T., Makie, H., & Watanabe, T. (2004). A framework for data collection system with sensor networks in disaster circumstances. In Proceedings of the 2004 international workshop on wireless Ad-Hoc networks, 31 May–3 June (pp. 94–98). Oulu, Finland.

Kung, H. Y., Chen, C. H., & Ku, H. H. (2012). Designing intelligent disaster prediction models and systems for debris-ﬂow disasters in Taiwan. Expert Systems with

Applications, 39(5), 5838–5856.

Lin, H. F., & Chen, C. H. (2013). An intelligent embedded marketing service system based on TV apps: Design and implementation through product placement in

idol dramas. Expert Systems with Applications, 40(10), 4127–4136.

Lindsey, S., & Raghavendra, C. S. (2002). PEGASIS: Power efﬁcient gathering in sensor information systems. In IEEE aerospace conference proceedings, 9–16 March (pp. 1125–1130). Big Sky, Montana.

Lo, C. C., Kuo, T. H., Kung, H. Y., Kao, H. T., Chen, C. H., Wu, C. I., et al. (2011). Mobile merchandise evaluation service using novel information retrieval and image

recognition technology. Computer Communications, 34(2), 120–128.

Neaupane, K. M., & Piantanakulchai, M. (2006). Analytic network process model for

landslide hazard zonation. Engineering Geology, 85(3–4), 281–294.

Ruiz, L., & Loureiro, A. (2003). MANNA: A management architecture for wireless

sensor networks. IEEE Wireless Communication Magazine, 41(2), 116–125.

Saaty, T. L. (1980). The analytic hierarchy process. New York: McGraw-Hill.

Saaty, T. L. (1996). Decision making with dependence and feedback: The analytic

network process. Pittsburgh, Pennsylvania: RWS Publication.

Sharpe, C. F. S. (1938). Landslides and related phenomena: a study of mass-movements

of soil and rock. New York: Columbia University Press.

Skapura, D. M. (1995). Building neural networks. New York: ACM Press.

Varnes, D. T. (1978). Special report 176: landslides analysis and control. In Transportation and road research board (pp. 11–33). Washington D.C.: National Academy of Science.

Wooldridge, M. (1995). This is MyWorld: The logic of an agent-oriented testbed for DAI. In Intelligent agents: Theories, architectures and languages. Lecture notes in computer science (Vol. 890, no. 1, pp. 160–178).