This paper presents on problems and kinds of multi-point passing as machine tools in a course search of a virtual factory based on a selective learning from multi-directive behavior patterns using PS (Profit Sharing) by an agent. A behavior is selected stochastically from 8 kinds of ones using QL like Boltzmann Distribution with an invalid rule inhibition plan by a reinforcement function of an equal ratio decrease type. Moreover, a variable temperature scheme is adopted in this distribution, where the environmental identification is valued in the first stage of the search and the convergence of learning is shifted to be valuing as time passes.