Aims of traditional planners had been limited to finding a sequence of operators rather than finding an optimal or near-optimal final state. Consequently, the performance improvement systems combined with the planners had only aimed at efficiently finding an arbitrary solution, but not necessarily the optimal solution. However, there are many domains where we call for quality of the final state for each problem. In this paper, we propose an extension of a planner for optimization problems, and another application of EBL to problem solving: learning control knowledge to improve searching performance for an optimal or a near-optimal solution by analyzing reasons a solution is better than another. The proposed method was applied to technology mapping in LSI design, a typical optimization problem. The system with the learned control knowledge synthesizes optimal circuits four times faster than that without the control knowledge.