Abstract: Off line programming has become an important branch in robotic programming because it has many advantages such as wide applicability and not occupying production time.Welding parameters' planning is one of necessary components in Robotic Arc Welding Task-level Off Line System(RAWTOLS).In order to go on welding parameters' planning,a new training algorithm of the feed forward neural networks,Single Parameter Dynamic Search algorithm,is utilized.Single parameter dynamic search is the characteristic of this algorithm and therefore the calculation quantum of the objective function is reduced greatly.And the idea of functional link is used to preprocess the input of neural networks.The result of the calculation shows that the convergence effect of this method is better than BP algorithm in the field of welding parameters' planning.