遗传算法优化的神经网络熔深预测模型

Genetic Algorithm Optimized Neural Network Prediction Model of Weld Penetration

依据熔化极气体保护焊(GMAW)焊接时对焊缝熔深的影响因素,结合遗传算法,建立以焊接电流、焊接电压、焊接速度和焊缝熔宽为输入、以熔深为输出的BP神经网络焊缝熔深预测模型。仿真试验和结果分析表明:使用遗传算法能得到神经网络较好的初始权值和阈值,能显著提高网络性能;所建模型符合已有的焊接熔深理论研究,并且预测误差较小,具有较好的泛比能力和较高的预测准确性,对工程实际具有一定的指导意义。

According to the influencing factors of weld penetration caused by gas melt arc welding,this paper combined with genetic algorithm and established a predicted model of weld penetration of BP neural network in which the input are welding current,welding voltage,welding velocity and weld width and the output is weld penetration.The simulation test and result analysis show that BP neural network can get better initial weights and thresholds and significantly improve its performance with the using of genetic algorithm;the established model coinciding with theortic research on weld penetraion has smaller prediction error,better comparing capability and higher prediction accuracy.It has certain guiding senses to the practical engineering projects.