Research on Comprehensive Evaluation of Regional Water Safety Based on Support Vector Machine

Guo-dong QU,Zhang-hua LOU,De-di LIU
DOI: https://doi.org/10.3969/j.issn.1009-6094.2011.02.053
2011-01-01
Abstract:This paper is aimed to define the water security from the perspective of harmonious between people and water.The regional water resources can satisfy the basic demand of the development of economic,social and cultural aspects through scientific and reasonable planning.A water security index system was constructed on the basis of previous researches since the related faces of water safety problems are very wide,which involving both natural and social indexes,static and dynamic indexes as well as qualitative and quantitative indexes.According to the research outputs achieved in water safety critical value,and refer to the water safety evaluation standard of other river basins as well as the actual situation of Zhejiang province,16 representative indexes were selected to establish the water security rating criteria of Zhejiang province.The support vector machine (SVM) is a kind of new current machine learning method,which is based on the statistical learning theory proposed by Vapnik et al.Sample learning problems were realized by SVM as convex quadratic programming problems,therefore the mentioned method is able to acquire global optimal solution and solve the local extremum problems which were unable to be avoided in the neural network method.To note,SVM with its minimization of structural risks as well as the advantage in Zhejiang Province is a ideal case study due to its abundant water resources and presence of complicated water safety problems.The results showed that if compared with the single factor evaluation method or AHP method,SVM exhibited same reliability but capable to avoid overmuch artificial and subjective factors that may result in inaccurate evaluation.Therefore,the regional water security comprehensive assessment based on support vector machine is reasonable and feasible to be implemented.
What problem does this paper attempt to address?