of 17
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
   1 3 Lazaros IliadisIlias Maglogiannis (Eds.) 12th IFIP WG 12.5 International Conferenceand Workshops, AIAI 2016Thessaloniki, Greece, September 16–18, 2016Proceedings Artificial IntelligenceApplicationsand Innovations IFIP AICT 475  A Hybrid Soft Computing ApproachProducing Robust Forest Fire Risk Indices Vardis-Dimitris Anezakis 1 , Konstantinos Demertzis 1( & ) ,Lazaros Iliadis 1 , and Stefanos Spartalis 2 1 Lab of Forest-Environmental Informatics and Computational Intelligence,Democritus University of Thrace, 193 Pandazidou st., 68200 Orestiada, Greece {danezaki,kdemertz,liliadis} 2 Laboratory of Computational Mathematics, School of Engineering,Department of Production and Management Engineering,Democritus University of Thrace,V.So 󿬁 as 12, Prokat, Building A1, 67100 Xanthi, Greece Abstract.  Forest   󿬁 res are one of the major natural disaster problems of theMediterranean countries. Their prevention - effective  󿬁 ghting and especially thelocal prediction of the forest   󿬁 re risk, requires the rational determination of therelated factors and the development of a  fl exible system incorporating anintelligent inference mechanism. This is an enduring goal of the scienti 󿬁 ccommunity. This paper proposes an Intelligent Soft Computing MultivariableAnalysis system (ISOCOMA) to determine effective wild  󿬁 re risk indices. Morespeci 󿬁 cally it involves a Takagi-Sugeno-Kang rule based fuzzy inferenceapproach, that produces partial risk indices (PRI) per factor and per subject category. These PRI are uni 󿬁 ed by employing fuzzy conjunction T-Norms inorder to develop pairs of risk indices (PARI). Through Chi Squared hypothesistesting, plus classi 󿬁 cation of the PARI and forest   󿬁 re burned areas (in threeclasses) it was determined which PARI are closely related to the actual burnedareas. Actually we have managed to determine which pairs of risk indices areable to determine the actual burned area for each case under study. Wild  󿬁 re datarelated to speci 󿬁 c features of each area in Greece were considered. The Soft computing approach proposed herein, was applied for the cases of Chania, andIlia areas in Southern Greece and for Kefalonia island in the Ionian Sea, for thetemporal period 1984 – 2004. Keywords:  Takagi-Sugeno-Kang    Fuzzy inference system    T-Norms   Chi-Square test     Individual indices    Uni 󿬁 ed index    Forest   󿬁 res 1 Introduction Greece has a very important forest capital, as 50 % of the territory is covered bywoodland. About 25 % of it is characterized by high vegetation coniferous andbroadleaf high biodiversity, the remaining of low trees and shrubs near inhabited areas.Also there are approximately 2 million acres of rangelands. During the last 20 years,the average annual burned areas in the country are higher than 45,000 acres as a result  ©  IFIP International Federation for Information Processing 2016Published by Springer International Publishing Switzerland 2016. All Rights ReservedL. Iliadis and I. Maglogiannis (Eds.): AIAI 2016, IFIP AICT 475, pp. 191 – 203, 2016.DOI: 10.1007/978-3-319-44944-9_17  of 1500 forest   󿬁 res. The determination of the factors that favor ignition and contributeto the spread of wild  󿬁 res (WF) requires a detailed spatiotemporal analysis of thehistorical data for each area under study. Moreover, the speci 󿬁 cation of the correlationsbetween these parameters is absolutely necessary. This research paper proposes aninnovative hybrid forest   󿬁 re modeling system operating on a local basis. The reasoningof the ISOCOMA employs Computational Intelligence approaches in order to producean overall  󿬁 re risk index. 1.1 Literature Review Iliadis and Betsidou [9] have implemented an intelligent rule based fuzzy inferencesystem (FIS) evaluating wild  󿬁 re risk for the forest departments of Greece. The esti-mation of the risk indices was done by using fuzzy triangular membership functionsand Einstein fuzzy conjunction T-Norms. Iliadis and Zigkrika [11] have also developeda FIS that performs and evaluates scenarios (by assigning weights to the involvedfeatures) towards the estimation of a characteristic overall forest   󿬁 re risk index inGreece. Papakonstantinou et al. [17] have proposed a fuzzy rule based system toproduce the drought risk indices vectors for the forest regions of Cyprus under study.Iliadis et al. [10] have developed a fuzzy inference system under the MATLAB plat-form. The system uses three distinct Gaussian distribution fuzzy membership functionsin order to estimate the partial and the overall risk indices due to wild  󿬁 res in thesouthern part of Greece.  Ö zbayo ğ lu and Bozer [16] estimated the potential burnedareas using geographical and meteorological data. Several computational intelligenceapproaches were used namely: Multilayer Perceptron (MLP), Radial Basis FunctionNetworks (RBFN), Support Vector Machines (SVM) and fuzzy logic. Shidik andMustofa [18] used a Back-Propagation Neural Network which was trained based onmeteorological and forest weather indices, so as to classify the burned area in threecategories. Aldrich et al. [1] investigated the effect of variations in land use and climatein the occurrence of forest   󿬁 res. Catry et al. [6] used logistic regression models topredict the relative probability of ignition occurrence, as a function of the resulting  󿬁 resize. 1.2 Innovations of the Proposed Methodology The main innovation of the ISICOMA is the development of four partial risk indices(PRI), which are derived from the respective analysis of separate parameters, creatingand analyzing meaningful relationships and rules of correlations between them. Thisraises the problem of wild  󿬁 res (WF) on an absolutely realistic basis. In addition, it isfor the  󿬁 rst time that an intelligent system combines the use of an adaptive fuzzyinference Takagi-Sugeno-Kang (AFITS) system with the wide use offuzzy conjunctionT-Norms in order to obtain higher   󿬁 tting rates between PRI and TPRI with the actualburned areas. 192 V.-D. Anezakis et al.  1.3 Data The  󿬁 rst step towards the development of an overall wild  󿬁 re risk index model was thedetermination of all factors that affect the behavior of a forest   󿬁 re. The data collectedfrom the forest inspections and from the Hellenic national meteorological service.According to [12] the following factors have been identi 󿬁 ed as playing a key role(Table 1).Utilizing and analyzing in-depth studies in the raw meteorological, topographicaland vegetative data of the areas concerned [3, 4] the following categories were obtained (Table 2). 1.4 Areas of Study Kefalonia (island in the Ionian Sea) Ilia (prefecture in Peloponnese) and Chania(prefecture in Crete island) have been chosen as the areas of interest. They have richvegetation, they have protected areas (under Natura network) and their climate is dryand hot with low rain height. Also both Chania and Kefalonia are characterized by hightouristic development and growth with high land value. On the other hand ancient Olympia is located in Ilia prefecture. Thus, it is an area of high cultural and touristic Table 1.  Factors affecting  󿬁 re behavior Flammability of vegetation Monthly rainfallCanopy density Previous month rainfallVegetation density AltitudeAir temperature SlopeRelative humidity Ground orientationWind speed, daily rainfall Exposure Table 2.  Classi 󿬁 cation of the  󿬁 re parameters Parameters Class1 Class2 Class3 Class4 Class5 Class6Meteorological Wind 0 – 1 bf 1.1 – 4 bf 4.1 – 7 bf 7.1 – 9 bf >9.1 bf Air temperature/relativehumidityLow risk MediumriskHigh riskTopographic Slope 0 – 20 % 21 – 40 % 41 – 60 % 61 – 80 % 81 – 100 % >100 %GroundOrientation ExposureUnspeci 󿬁 ed North South East West Altitude Low Medium HighVegetation Canopy Density Absent Rare FullVegetation Density Absent Rarecanopy < 0.4Densecanopy > 0.4Flammability of vegetationLow risk MediumriskHigh riskDrought Rainfall (Daily, Monthly,Previous Month)Low Medium High A Hybrid Soft Computing Approach Producing Robust Forest Fire Risk Indices 193  value. During the period 1984 – 2014, totally 1397 wild  󿬁 res occurred in Ilia, 857 inChania and 1298 in Kefalonia.The Fire Ignition Indicator (FIGI) which emerges by combining the effect of tem-perature and humidity and the Spread Index which considers the effect of wind and slope(SPRI) have been used to produce signi 󿬁 cant evidence of forest   󿬁 re risk. In a previousresearch effort of our team [2] have found that the SPRI is  “ High ”  in the 30 – 50 % of thecases, whereas the FIGI has shown smaller high and medium hazard rates. 2 Theoretical Framework and Methodology 2.1 Fuzzy Inference Systems The Sugeno Fuzzy implication is the basic modeling approach used by the ISOCOMA.Introduced in 1985 [19], it is similar to the Mamdani method. While Mamdani FIS usesthe technique of defuzzi 󿬁 cation of a fuzzy output, Sugeno FIS uses weighted average tocompute the crisp output. The fuzzy membership functions (FMF) of the output areeither linear ( 󿬁 rst order polynomials or constant crisp values). A typical rule in aSugeno fuzzy model if the outputs are  󿬁 rst-order linear has the form:If Input 1  ¼  x and Input 2  ¼  y then Output isz  ¼  ax  þ by  þ c  ð 1 Þ For a zero-order Sugeno model, the output level z is a constant crisp value c(a = b=0).The output level z i  of each rule is weighted by the  󿬁 ring strength w i  of the rule. For an AND rule with Input 1 = x and Input 2 = y, the  󿬁 ring strength isw i  ¼  AndMethod  ð F 1 ð x Þ ; F 2 ð y ÞÞ ð 2 Þ where F 1,2  are the membership functions for Inputs 1 and 2. The  󿬁 nal output of thesystem is the weighted average of all rule outputs, computed as in (3). Finaloutput   ¼ P  N  i ¼ 1  w i z i P  N  i ¼ 1  w i ð 3 Þ where N is the number of rules. 2.2 T-Norms This paper attempts to calculate the Unique Overall Risk Index (UORI), resulting fromthe cumulative effect of all the related factors, after performing integration operationson all individual fuzzy sets. This task is carried out, by the use of speci 󿬁 c fuzzyconjunction  “ AND ”  operators (CONO) known as T-Norms in the literature. The Min,the Algebraic, the Drastic, the Einstein and the Hamacher Products act as T-Norms[5, 7, 13 – 15]. The T-Norms are the uni 󿬁 ers of partial risk indices and they are quiteoptimistic as they are assigning the minimum risk value to the overall index [8]. 194 V.-D. Anezakis et al.


Aug 12, 2018


Aug 12, 2018
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks