A Simulation Optimization Algorithm for CTMDPs Based on Randomized Stationary Policies^1）