Abstract: The adjusting process by the welding power which depends on its own control system is also the reborn process of the outer power net interference, if the outer-voltage fluctuates in the system when more than one welding power were used to weld the same workpiece simultaneously.The group-control sy stem of welding power sources with the combination of fuzzy theory and neural networkswas was studied firstly.The membership function of state variables, inference rules in the system, algorithms for fuzzy inference and defuzzification, etc.were analyzed and devised subsequently. Based on those, adaptive neuro-fuzzy inference system (ANFIS) with Tagaki-Sugeno model was achieved.Finally the completed ANFIS was simulated by using Simulink toolbox in Matlab, and used in Membrane-water-wall welding machine for experimentation.The result shows that the design is superior at less regulating time (shortened by 22%, compared with each welding power adjust separately) and less overshoot (reduced by 40%), which reflects its excellent dynamic behavior.