Abstract Sugarcane juice clarification is an important operation in the production process of sugar industry. At present, the two major production indicators (clear juice color value and gravity purity of the mixed juice) in sugarcane juice clarification process still cannot be measured online, and can only be obtained by offline test. In addition, when production condition changes, the setting and adjustment of the key operational variables are made by experienced workers. The delay caused by the offline test and the randomness of the key operational variables caused by manual adjustment will affect the quality of final product. To address these two issues, we firstly constructed a data-driven model based on the deep kernel extreme learning machine (DK-ELM) to predict the key production indicators for clarification process. The model which combined the kernel method and the multi-layer extreme learning machine (ML-ELM) proved successful in prediction of clear juice color value and gravity purity. Then, we proposed a strategy for optimization of operational variables, avoiding the randomness caused by manual setting. A multi -objective function including the purity and color value was constructed. The niche multi-objective particle swarm optimization (PSO) algorithm was used to solve the multi -objective function under typical working conditions in the production process to get optimal set of key operational variables. The results showed that the niche multi-objective particle swarm optimization (PSO) algorithm can provide optimum production indexes for sugarcane juice clarification.

中文翻译：

---

甘蔗汁澄清过程建模与优化

摘要 甘蔗汁澄清是制糖工业生产过程中的一项重要操作。目前，甘蔗汁澄清过程中的两大生产指标（清汁色值和混合汁的比重纯度）仍无法在线测量，只能通过离线测试获得。此外，当生产条件发生变化时，关键操作变量的设置和调整由经验丰富的工人进行。离线测试造成的延迟和人工调整造成的关键操作变量的随机性都会影响最终产品的质量。为了解决这两个问题，我们首先构建了一个基于深度核极限学习机（DK-ELM）的数据驱动模型来预测澄清过程的关键生产指标。结合核方法和多层极限学习机（ML-ELM）的模型在预测清汁色值和重力纯度方面取得了成功。然后，我们提出了一种优化操作变量的策略，避免了手动设置带来的随机性。构建了包括纯度和颜色值的多目标函数。采用生态位多目标粒子群优化(PSO)算法求解生产过程典型工况下的多目标函数，得到关键操作变量的最优集合。结果表明，利基多目标粒子群优化（PSO）算法可为甘蔗汁澄清提供最佳生产指标。