In this paper

a globally optimized modeling method （GOM） is presented based on the characteristics of canonical piecewise-linear modeling. Using the accelerated controlled random search （ACRS） to perform the canonical piecewise-linear modeling

this method overcomes the difficulties of multiminimum in the modeling. The strategy of defining constrained search domain and parallel processing is used to speed up the modeling

and the result is satisfactory. It is shown with experience of modeling that the GOM is reliable and practical.