基于BP算法的逆变点焊电源模糊神经网络控制研究  

Research on fuzzy neural network control of inverter spot-welding power supply based on BP arithmetic

作　　者：陈刚[1] 张勇[1] 王瑞[1] 杨思乾[1] 

机构地区：[1]西北工业大学材料学院,陕西西安710072

出　　处：《电焊机》2007年第9期48-51,64,共5页Electric Welding Machine

摘　　要：引入动量因子对常规BP学习算法进行了改进。在分析模糊神经网络控制模型的基础上,针对模糊神经网络规则多、训练时间长的缺点,采用了给模糊控制规则增加阈值,减少网络训练运算量的优化方法。最后将此优化方法和改进的训练算法应用到逆变点焊电源模糊神经网络(FNN)恒电流控制系统中,通过使用MATLAB语言编程,对该系统进行了仿真分析。仿真结果表明,动量因子的引入不但减小了BP算法学习过程的振荡趋势,加快了收敛速度,而且较好解决了BP网络容易陷入局部极小点的缺陷。模糊规则阈值的引入,有效减少了网络的训练时间。In this paper,BP arithmetic is mended by appending momentum factors.Considered the disadvantages of fuzzy neural network such as overmuch rulers, and time-consuming train time, a method that increase threshold quantity of fuzzy control rulers and decrease the calculation to optimize fuzzy neural network controller is proposed, based on analyzing the design frame of fuzzy neural network.Again,the arithmetic and method are simulated with a MATLAB language program in a constant current control system of inverter spot-welding power supply based on FNN.The simulation results show that momentum factors not only decrease the oscillating trend and accelerate the rate of convergence during the BP arithmetic learning,but also preferably solve the disadvantage that BP network is easily plunged into local minimum.The threshold quantity to fuzzy control rulers can efficiently decrease the training time of network.

关 键 词：逆变点焊 FNN BP算法 动量因子 

分 类 号：TG438.2[金属学及工艺—焊接]