Research On Modeling And Operation Optimization For Clarifying Process Of Carbonation Method Sugar Factory

Clarification of sugar cane juice is an important step in cane sugar factory, and there are many methods to achieve it. The carbonation method has been widely used in most of the sugar factories because its produced sugar is high purity, low color and popular in market. Currently, in the clarifying process of carbonation method sugar factory, each controller settings of operating variables is given by workers with their experiences according to offline assay results of the key production indices. However, since the workers can’t adjust the controller settings timely with the change of working condition because of assay lag, in real clarifying process, the key production indices are often beyond the range of expectations although the control performances can meet the requirements, leading to the produced sugar yield and quality instable in the end. So, in order to enhance production quality and improve factory economic benefits, it is significant to make research on operation optimization to solve the controller optimal setting problem of carbonation method sugar factory.This paper firstly elaborates the present situation of modeling and operation optimization for clarifying process of sugar factory, meanwhile introduces the concept and industrial applications of optimal setting and case-based reasoning (CBR). Secondly, the clarifying process of carbonation method sugar factory is detailed and is divided into two parts:first carbonation and second carbonation, their key production indices like color value, Calcium salt content and the relative operation parameters are defined; meanwhile, the structure and algorithm of generalized dynamic fuzzy neural network (GDFNN) is introduced, thus the production indices predictive models of each part are built based on online and offline data with GDFNN respectively. For model comparison, another two corresponding models using BP networks are also given at the same time. Thirdly, a controller optimal setting strategy which combines optimization mode library with case retrieval and case revise is proposed for clarifying process. After building the optimization model, the mode library of typical conditions is obtained by Non-dominated Genetic Algorithm Ⅱ(NSGAⅡ) or PSO algorithm. For a new working condition, find its corresponding operation parameters’pre-settings firstly by case retrieval in mode library, then correct these values by expert rule-based reasoning based on errors which is between predictive values and expectations of key production indices, until the errors arriving allowed ranges. And these new operation parameters’ values are as the final settings of controller.Simulation results show that the GDFNN predictive model has better generalization ability and more fast convergence than BP network model. And the proposed optimal setting strategy can give the controllers’settings in different working conditions, avoiding subjectivity and lag of human given. Thus the goal of better quality and yield for refined sugar is achieved.