Study On Evidence Theory And Its Application To The Water Inrush Prediction In Mine

Water hazard in coal mine is one of the main disasters in mine construction andproduction. Water inrush disaster does enormous economic losses and casualties tocoal mine enterprises. Water inrush prediction in mine is a complex problem, as thewater inrush results from multiple effects of hydrogeology, engineering geology,mining condition, rock mechanics, etc. For resolving the uncertain and non-linearproblem of mine water inrush, a new model of a water inrush prediction in mine basedon evidence theory is put forward. The further study of the information fusion basedon evidence theory for combining conflict evidence, processing fuzziness object andconstructing basic probability assignment function are described in this dissertation.Based on the improved evidence theory, the dynamic precursory information of minewater inrush are fused, and the multi-field coupling model for water inrush predictionin mine is establish, which lay a foundation both in theory and technique for waterinrush prediction in mine.To remedy the shortcomings in the recognition of evidence conflict by usingtraditional conflict coefficient, the application conditions of Dempster’s combinationrule are discussed by traditional conflict quantification standard and the distanceamong basic probability assignment functions transformed by pignistic. Furthermore,an improved method of evidence combination is presented. In this method, theconflict level is represented by the pignistic probability distance of evidence, after thatthe conflict level is transformed into similarity level and the support degree ofevidence is obtained. In addition, the weight coefficients are determined. Finally, thebasic probability assignments adjusted by weight coefficient are fused by Dempster’scombination rule. The numerical examples prove that the modified method not onlyhandle the evidence conflict efficiently but also solve the one-ballot veto androbustness problem, and has fast convergence speed. This is extraordinarilysignificant to improve the performance of the information fusion system.Determination of the basic probability assignment is the core and key issue inevidence theory, and is the most difficult step in actual application. However, thegeneral way of determining the basic probability assignment has not appeared as theapplication background of evidence theory is complexity and diversity. Based on thegeneralized triangular fuzzy number, a general method for obtaining basic probabilityassignment is proposed in this dissertation. In the proposed method, the triangular fuzzy number described model of singleton proposition is constructed using theminimum, average and maximum values of the sample data. And the generalizedtriangular fuzzy number described model of multielement proposition is representedby the crossing area of the triangular fuzzy number described models of the singletonpropositions. Based on the degree of membership of the generalized triangular fuzzynumber, the basic probability assignment is obtained. This constructing strategy issimple, practical, and easy to calculation, has broad application prospects.As incapable of disposing the fuzziness objects, evidence theory could begeneralized to fuzzy sets and the information about inaccuracy and fuzziness could berepresented and disposed by using the advantages of evidence theory and fuzzy sets. Amethod for defining the fuzzy closeness degree is put forward in this dissertation. Thefuzzy belief function and plausibility function are proposed on the basis of the newcloseness degree and fuzzy decomposition theorem. They are not influenced by somecritical points of the membership degree function, can gain the actual focal elementchange information effectively and are relatively sensitive to the subtle changes. Thefuzzy evidence combination rule based on fuzzy closeness degree can gain the changeinformation of the fuzzy focal elements effectively and dispose uncertain and fuzzyinformation effectively. The combination results are more conducive to objectivedecision. This is extraordinarily significant to expand the application range ofevidence theory.As viewed from the perspective of information fusion, the methods of waterinrush source recognition and water inrush quantity from coal floor prediction areproposed based on evidence theory and its improved methods in this dissertation. Theexperiment results verify the feasibility and effectiveness of the water inrushprediction model. According to the improved methods, intelligent prediction systemof mine water inrush is developed on a Windows7, Visual Studio2010Professionaland SQL Server2008software platform which can provide accurate and credibledisaster prediction, and effectively improves the level of coal mine safetymanagement.