Research On Soft-sensor And Predictive Control Of Acetic Acid Distillation Column

Due to the characteristics of distillation process, such as complex, time-varying, nonlinear and strong coupling, distillation control has been always a research hotspot in process control field. Along with the development of the industry, predictive control which is on behalf of the advanced control has been widely used to meet the growing requirements of the control quality. The advanced control technology can improve the stable rate of the production equipment to ensure product quality and reduce energy consumption, so as to realize the bounder control of the product quality.With the acetic acid dehydration azeotropic distillation process as the research background, ASPEN process simulation software is adopted to establish the production technology mechanism of acetic acid dehydration azeotropic distillation system model. Firstly simulate the steady-state of the distillation system, calculate the sensitive plate location, on this basis, and simulate the dynamic-state by using ASPEN DYNAMICS software. Then Due to the characteristics of wavelet kernel extreme learning machine, which has a fast training speed and the algorithm is steady, a method of online wavelet kernel extreme learning machine modeling based on online update is presented to solve the problem that model mismatch the reality because of working point shift in the actual operation. Based on this, a soft sensor of acetic acid concentration is set up and it can realize the online detection of the product. Compared with the offline wavelet kernel extreme learning machine soft sensor, the online one has higher prediction precision. The mean squared error of online updating model is52%lower than offline model. Finally for the performance of indirectly temperature control can’t reach a high level, a model predictive control strategy based on wavelet kernel extreme learning machine (KMPC) is proposed. On the basis of acetic acid soft sensor, directly control of products can be easily achieved. In the control system, the prediction model is established using wavelet kernel extreme learning machine (KELM). A predictive controller is used as the master controller for the acetic acid concentration. One PID controller is used as the slave controller for the reboiler vapor flow. The two controllers constitute a cascade control system. At the same time, in order to eliminate the disturbance as sensitive temperature, bottom temperature, inlet temperature and pressure of reboiler, feedforward controllers and nonlinear predictive controllers of complex distillation column are designed. Do the simulation research of the advanced control stratygies with the control object which is set up by calling the ASPEN DYNAMICS process simulation software through the interface in MATLAB SIMULINK. According to the results of the simulation, compared with the traditional Dynamic Matrix Control (DMC) method, the KMPC strategy is efficiency. Furthermore, the new method is easy to apply because of its simple structure.