The Design and Implementation of a Fuzzy Inference System, usingFuzzy Clustering and Genetic Algorithm in a GIS Environment(Case study: Site selection of fuel stations)

Spatial data in Geographic Information Systems (GIS) are inherently uncertain; and therefore a system that can handle and infer from such uncertain data is of vital importance. Fuzzy Inference System (FIS) is one of the wellknown inference systems that have been considered by many scientists in the recent decades. The objective ofthis paper is to propose and implement a new framework for extracting fuzzy membership functions and fuzzy rules that can be used for allocating gas stations in the city of Tehran. In this research, some of the fuzzy membership functions and fuzzy rules were extracted by using training data, and the rest were estimated by interviewing experts and loading the results manually. A combination of fuzzy clustering algorithm and Genetic Algorithm (GA) were used to extract fuzzy membership functions and fuzzy rules automatically. In the first step, initial fuzzy membership functions and rules were extracted by clustering data and mapping them onto the axes of the coordinate systems. The fuzzy C-Means (FCM) algorithm was used for data clustering. Although the FCM is an efficient means, but it requires being define of the number of clusters (C) by an expert. Since C has a closerelation with the number of rules and membership functions, so estimating it in advance is important. The mathematical formula used to determine the optimal number of clusters is referred to a cluster validity index. Fukuyama and Sugeno cluster validity index (VFS) were used to determine the optimum number of clusters in this paper. In the second step the GA, which is a member of the family of evolutionary techniques, was employed to find the optimum form of fuzzy membership functions and fuzzy rules. The purpose of this step was to obtain a suitable set of parameter values that described the fuzzy rules and fuzzy membership functions according to the optimization criterion. At the end, the fuzzy inference system was improved by adding experts’ knowledge in the form of fuzzy membership functions and fuzzy rules. The proposed system was used for allocating gas stations in Tehran, capital of Iran. According to the knowledge of safety and fuel experts, 10 governing factors in site selection of fuel stations are identified and organized in 10 GIS layers such as proximity to the main streets, proximity to highways, etc. The results of this research showed that the accuracy of the fuzzy inference system increased 53% by using GA. Also, by examining the histograms of the land suitability map for allocating gas stations, it was inferred that 9% of the region had a high potential for constructing gas stations.