基于改进遗传算法优化BP神经网络的表面粗糙度误差预测  

Prediction of surface roughness error based on BP neural network optimized by improved genetic algorithm

作　　者：张剑飞[1] 王磊 刘明 王硕 ZHANG Jianfei;WANG Lei;LIU Ming;WANG Shuo(School of Computer and Control Engineering,Qiqihar University,Qiqihar 161006,China)

机构地区：[1]齐齐哈尔大学计算机与控制工程学院,黑龙江齐齐哈尔161006

出　　处：《高师理科学刊》2023年第7期33-40,共8页Journal of Science of Teachers＇College and University

基　　金：齐齐哈尔市科技计划重点项目(ZDGG-202203)。

摘　　要：随着现代制造业的发展,对工件加工质量和精度越来越追求高标准.表面粗糙度作为评价工件质量的重要指标,对工件质量和产品特性具有重要的影响.针对传统BP(Back Propagation)神经网络在训练过程中易陷入局部极小值和收敛速度慢等不足,遗传算法(Genetic Algorithm,GA)存在随机性问题,提出采用遗传算法和混合蛙跳算法(Shuffled Frog Leaping Algorithm,SFLA)结合来改进BP神经网络(记为SFLA-GA-BP)进行工件表面粗糙度误差预测.以工件表面粗糙度与砂轮粒度、砂轮转速、工件速度、进给率四要素之间的相关关系为研究目标,通过正交实验技术,分别以BP神经网络、遗传算法改进BP神经网络(记为GA-BP)和SFLA-GA-BP神经网络进行建模分析.实验结果表明,SFLA-GA-BP的均方根误差(Root Mean Squared Error,RMSE)比BP网络和GA-BP网络分别提高了1.7%和0.7%、平均绝对百分误差(Mean Absolute Percentage Error,MAPE)分别提高了2%和1.1%,平均绝对误差(Mean Absolute Error,MAE)分别提高了1%和0.6%.SFLA-GA-BP模型的预测误差相比于BP神经网络和GA-BP神经网络更加精准.故SFLA-GA-BP模型对于预测工件表面粗糙度具有更高的准确率和良好的稳定性,同时为企业减少成本,对企业智能化发展具有一定的指导意义.With the development of modern manufacturing industry,the workpiece processing quality and precision are increasingly pursuing high standards.Surface roughness is a significant index to estimatethe surface quality of workpiece,it had significant influence on the quality of machined parts and product performance.Aiming at the shortcomings of traditional BP(Back Propagation)neural networks was effortless to sink into the local minimum and slow convergence rate in the training process,genetic algorithm(GA)had the problem of randomness,according to the genetic algorithm and shuffled frog leaping algorithm(SFLA)were combined to improve BP neural network(named SFLA-GA-BP)to predict surface roughness error.The model taken the correlation between artifacts surface roughness and grinding wheel granularity,grinding wheel speed,artifacts speed and feed rate as the research object,and used orthogonal experiment method.BP neural network,genetic algorithm improved BP neural network(named GA-BP)and SFLA-GA-BP neural network were established model analysis respectively.Experimental show that the root mean squared error(RMSE)of SFLA-GA-BP was 1.7%and 0.7%higher than that of BP and GA-BP,respectively.MAPE increased by 2%and 1.1%,mean absolute error(MAE)increased by1%and 0.6%.The prediction error of SFLA-GA-BP model is more accurate than that of BP neural network and GA-BP neural network.So the SFLA-GA-BP model for predicting the workpiece surface roughness had a higher accuracy and good stability.Meanwhile,it can reduce the cost for enterprises,and has certain guiding significance for the development of enterprise intelligence.

关 键 词：表面粗糙度 神经网络 遗传算法 混合蛙跳算法 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]