应用混合种群遗传神经网络的精梳毛纺工艺参数反演模型

Worsted spinning process parameters inversion model using mixed population genetic neural network

针对传统精梳毛纺工艺参数反演模型收敛性和稳定性不理想、反演精度低等问题,以及标准遗传算法(SGA)应用于复杂优化问题时存在早熟收敛等缺点,以BP神经网络为基础,提出一种混合种群遗传人工神经网络(MPG-ANN)反演模型,首先以混合种群遗传算法优化BP神经网络的权值与阈值建立预测模型,在此基础上根据毛纱CV值建立混合种群遗传算法反演模型,用来反演精梳毛纺生产过程工艺参数。以纺纱车间大量现场工艺检测数据为对象进行反演验证,结果表明:MPG-ANN模型反演精度达97%,相比于标准遗传算法人工神经网络(SGA-ANN)模型提高4%,同时反演结果波动幅度相比于SGA-ANN模型降低了6.28%。该方法可为精梳毛纺生产过程质量控制提供有效的理论指导,对纺织企业新产品工艺开发设计的快速决策具有很好的借鉴作用。

Inversion model of conventional worsted spinning process parameters is defective because of unsatisfactory convergence and stability and low inversion accuracy.Besides,standard genetic algorithm shows premature convergence in the application of complicated optimization problem.Therefore,the paper put forward an inversion model of mixed population genetic-artificial neural network based on BP neural network.First the author established the prediction model by optimizing the weights and threshold of BP neural network through mixed population genetic algorithm.On this basis,the author created the inversion model of mixed population genetic algorithm according to the coefficient of variation（CV） value of yarn and used it for the inversion of process parameters during worsted spinning production.Next,the author conducted inversion verification by taking a large quantity of test data of field process in spinning workshops field process parameters.The results show that the inversion accuracy of MPG-ANN model reaches 97% and increases by 4% relative to SGA-ANN model.In addition,the fluctuating margin of the inversion results of MPG-ANN model reduces by 6.28% at most relative to SGA-ANN model.Finally,the author predicted the CV value of yarn on the basis of inversion results.This method not only has an important guiding role in the textile production process quality control,but also has a very good reference for enterprises rapid process development of new product design decision.