A graph data modeling program is developed based on various least square methods. The number of independent variable is limited to be less than or equal to 2. Nonlinear least square method as well as linear and weighted least square method is implemented in the program. For nonlinear least square method, a modified version of the BFGS algorithm is utilized. Manual selection of basis function is one of the major drawbacks of various least square methods and in the present work an automatic selection algorithm is developed based on the equivalent differential equation concept and the correlations between candidate functions. Several technique are also implemented in the program to improve the computational efficiency and extend the applicability.
Many graphs are modeled by the developed program and the results demonstrate reliability and efficiency of the present method.