Prediction in longeron automatic welding of generalized regression neural network by ameliorated fruit flies optimization algorithm

As longeron automatic welding, the obstacles in welding area such as stiffening plates, partitions and holes must be automatically avoided. The location of the obstacles on the workpiece are random due to the products are varied and the low-precision of the clamping, we hardly predict the obstacles in a single way. To solve this problem, by the use of ultrasonic sensor collecting the date of the obstacles, raising a prediction model in longeron automatic welding based on the ameliorated fruit flies algorithm optimization generalized regression neural network(AFOA-GRNN). The method by introducing the information element and the sensitivity of the two factors in traditional fruit flies algorithm, improved optimization strategies and location updating of fruit flies position, to optimize the parameters of the generalized neural network, to predict the obstacles in longeron automatic welding. Experiments show that correction type AFOA-GRNN prediction model compared to FOA-GRNN, training speed is faster, and the predicting precision is higher.