摘要
吨蔗能耗和蔗糖色值是制糖过程中两个重要的工艺指标,然而目前这两个指标均无法在线获取。此外,当生产状况发生变化时,依靠人工经验进行参数设定难以获得稳定的生产质量。为了解决这两个问题,提出一种融合深度时间卷积网络和核极限学习机的数据驱动建模方法,实现了2个工艺指标的在线预测。此外,在数据驱动模型的基础上,构建了工艺指标优化模型,并采用改进的麻雀搜索算法进行迭代计算,实现了2个工艺指标的优化。计算实验结果表明:所提出模型的预测可决系数均超过0.9,并与其他4种模型相比具有更高的预测精度。此外,通过求解工艺指标优化模型,优化后的吨蔗能耗降低了1.47%,蔗糖色值降低了2.57%,2个指标得到了较大的改善。
Energy consumption per ton of sugarcane and sucrose color value are two important process indicators in sugar production.Unfortunately,these indicators are not currently available in real-time.Additionally,relying on manual experience for parameter settings makes it difficult to achieve stable production quality when production conditions change.To address these issues,a data-driven modeling approach is proposed,which integrates deep temporal convolutional network and kernel extreme learning machine to enable online prediction of the two process indicators.Additionally,a process indicator optimization model is constructed based on the data-driven model,and an improved sparrow search algorithm is employed for iterative computation to optimize the two process indicators.The computational experimental results demonstrate that the proposed model has coefficients of determination exceeding 0.9 and outperforms the other four models in prediction accuracy.Moreover,by solving the process indicator optimization model,the energy consumption per ton of sugarcane is reduced by 1.47%and the color value of sugarcane is reduced by 2.57%,achieving significant improvements in both indicators.
作者
蒙艳玫
李济钦
MENG Yanmei;LI Jiqin(School of Mechanical Engineering,Guangxi University,Nanning 530004,China)
出处
《广西大学学报（自然科学版）》
CAS
北大核心
2023年第3期579-593,共15页
Journal of Guangxi University（Natural Science Edition）
基金
国家自然科学基金项目(61763001)。
关键词
制糖
数据驱动建模
深度时间卷积网络
核极限学习机
改进的麻雀搜索算法
sugar production
data-driven modeling
deep temporal convolutional network
kernel extreme learning machine
improved sparrow search algorithm
