The Development Of Soft Modeling Method And Optimization Systems For Production Process In Process Industry

Due to the characteristics of nonlinearity, time-variant, and large time delay of production processes in process industry such as petrochemical industry and iron and steel industry, it is very difficult to get exact values of some process parameters that are related to process operation and control. Traditional off-line measurement methods needs a long time, which makes it difficult to achieve the real time control of product quality. Therefore, how to make it possible that these process parameters can be on-line measured is very important for process industries because it can improve the production efficiency, decrease energy consumption, and guarantee product quality.With the development of informatization in process industry enterprises, a large amount of production process data have been accumulated. These data contains a lot of useful information, and thus provides a good chance for data-driven soft sensing modeling. This thesis studied the soft sensing modeling method and its applications based on particle swarm optimization (PSO) and least square support vector machine (LSSVM), and developed soft sensing and optimization systems for practical production processes in process industry. The main contents are as follows:(1) The parameter determination in the LSSVM model was taken as an optimization problem, and an improved PSO algorithm was developed to obtain the optimal values of these parameters.(2) The clustering method was adopted to improve the diversity of training data so as to improve the accuracy and robustness of the LSSVM model.(3) The PSO-LSSVM was applied to two practical production processes:the rare earth extraction process and the BOF steelmaking process. Two soft sensing models were obtained to predict the values of those variables that were hard to be measured. The two models were tested based on practical production data, and the test results showed that they have a good prediction accuracy and robustness.(4) The operation optimization problem of BOF steelmaking process is investigated and a mathematical model is formulated to minimize the deviation from the expected quality. A genetic algorithm is developed to solve this problem and obtain the appropriate operation conditions, which are then set as the targets for process control.(5) Based on the developed soft sensing models and algorithms, the on-line constituent content prediction system for the rare earth extraction process and the on-line prediction and operation optimization system for BOF steelmaking process were developed and tested using practical production data.