基于神经网络识别的复杂工艺参数柔性优化

Flexible optimization of complex process parameters based on neural network recognition

摘要:

为了提高复杂工艺环境下生产合格率,动态的柔性优化工艺参数组合。以现场生产数据为学习样本和控制对象,基于可适应BP神经网络建立识别生产变化的工艺参数柔性优化模型。在此基础上,通过引入惩罚机制改进粒子群算法在神经网络输入端迭代求解最优参数组合。为了验证模型的有效性,验证实例由3条轻化工艺路线生产数据构成,结果表明模型预测误差绝对值在3%以内,优化得到的参数组合提高合格率到85%以上,有效的提高生产效果和生产柔性。

Abstract:

In order to improve the passing rate of production under complex process and realize dynamically flexible optimization for composite parameters. Taking the field production data as training samples and controlling object, complex parameters optimizing model was built according to adapting neural network, which identify variations of process. Based on this model, PSO algorithm combined with punishment was used to calculate the optimized parameters combination. Finally, the validation instance, consisting with the production data of three light chemical routings, was used to identify the effectiveness of optimizing model. The results show that the error decline to below 3% and the passing rates achieved are higher than 85%. This model can effectively improve the production results and realize flexibility.