以基因演算法優化最小二乘支持向量機在地籍坐標轉換之研究

A Study of Cadastral Coordinate Transformations Using the Genetic Algorithm Based on the Least Squares Support Vector Machine

本文提出以基因演算法（Genetic Algorithm, GA）優化最小二乘支持向量機（Least Squares Support Vector Machine, LSSVM）系統參數，以提升地籍坐標轉換精度。利用花蓮與台中兩實驗區的真實資料，以TWD67（Taiwan Datum 1967）轉換至TWD97（Taiwan Datum 1997）的地籍坐標轉換為例，驗證以GA優化後之LSSVM在地籍坐標轉換精度提升的效能。根據實驗結果顯示：（1）LSSVM未優化前，三種核函數的坐標轉換精度表現以RBF（Radial Basis Function）最佳，其次為LIN（Linear kernel），最差為POLY（Polynomial kernel）。（2）LSSVM之RBF經GA參數優化後（RBF+GA），其轉換精度優於RBF。（3）進行RBF系統參數優化後，花蓮與台中兩實驗區之RBF+GA相對於RBF的精度提升率，分別為20%及32%。 The least squares support vector machine (LSSVM) is applied to study the cadastral coordinate transformation accuracy performances. Three kernel functions, i.e., polynomial function (POLY), linear kernel (LIN), and radial basis function (RBF), are implemented in LSSVM. The genetic algorithm (GA) is proposed to optimize the system parameters of LSSVM with RBF (designed as RBF+GA). Two data sets for Hualien and Taichung were tested and analyzed. The test results show that: (1) regarding to the coordinate transformation accuracies after applying LSSVM with different kernel functions, RBF is the best, LIN is the second place, and POLY is the worst; (2) if the system parameters of RBF optimized by GA, the coordinate transformation accuracies of RBF+GA are better than that of RBF; and (3) comparing with RBF, the coordinate transformation accuracy improving rate of RBF+GA, for the Hualien and the Taichung data sets are 20% and 32%, respectively.

The least squares support vector machine (LSSVM) is applied to study the cadastral coordinate transformation accuracy performances. Three kernel functions, i.e., polynomial function (POLY), linear kernel (LIN), and radial basis function (RBF), are implemented in LSSVM. The genetic algorithm (GA) is proposed to optimize the system parameters of LSSVM with RBF (designed as RBF+GA). Two data sets for Hualien and Taichung were tested and analyzed. The test results show that: (1) regarding to the coordinate transformation accuracies after applying LSSVM with different kernel functions, RBF is the best, LIN is the second place, and POLY is the worst; (2) if the system parameters of RBF optimized by GA, the coordinate transformation accuracies of RBF+GA are better than that of RBF; and (3) comparing with RBF, the coordinate transformation accuracy improving rate of RBF+GA, for the Hualien and the Taichung data sets are 20% and 32%, respectively.