Data‐driven modeling for crystal size distribution parameters in cane sugar crystallization process

Abstract

Crystal size distribution (CSD) is important in evaluating crystal quality in cane sugar crystallization process. Due to the complex non-linearity, time-delay and strong coupling in cane sugar crystallization process, it is difficult to directly modeling in the mechanism of cane sugar crystallization process to obtain CSD parameters. In order to obtain two main CSD parameters so that to achieve better control and production of cane sugar, this article constructs a data-driven model based on least squares support vector regression (LSSVR) and particle swarm optimization (PSO). Based on LSSVR, the model takes the easy to measureable variables (massecuite brix, massecuite level, massecuite temperature, steam pressure, feeding volume, and vacuum degree) as input variables, and outputs sugar CSD parameters (crystal average size, coefficient of variation of crystal size). PSO algorithm is used to optimize the key parameters of primary model to get better modeling performance. Compared with other modeling methods such as back propagation, extreme learning machine, radial basis function, and SVR, the constructed PSO-LSSVR model has obvious advantages over other models in learning speed and predictive effect, generalization ability. This model has potential to be applied to the control system of cane sugar crystallization process and get better product of sugar.

Recent Advances in the Application of Machine Learning to Crystal Behavior and Crystallization Process Control

“…Crystal size distribution (CSD) is an important index for evaluating the crystallization quality. In order to achieve the prediction of CSD for better control and production of the crystallization process, Meng et al 116 constructed a data-driven model based on Least Squares Support Vector Regression (LSSVR) and Particle Swarm Optimization (PSO) based on previous studies. The model is based on LSSVR, which outputs the CSD parameters (crystal average size, coefficient of variation of crystal size) of crystal products by taking six easily measurable variables as input variables.…”