Research On Extreme Learning Machine And Its Application In Acetic Acid Soft Sensor Modeling

Distillation column is an important part in process petrochemical industry.It is a complex, time-varying, nonlinear, strong coupling process. In the actual process of distillation, the detecting devices are restricted,then a lot of important process parameters in distillation can not directly measured,led to a delay in the detection signal, feedback imprecise, many advanced control can not get the expect effect, making the effectiveness of the entire distillation process is not high. Soft sensor technology is an effective method to solve such problems.With the acetic acid dehydration azeotropic distillation process as background, First, distillation process data have collinear problems,making the low prediction accuracy and poor generalization ability,this paper proposesd a method called ridge regression extreme learning machine with differential evolution(DE_ELMRR),and combined with acetic acid dehydration azeotropic distillation process, extreme learning machine ridge regression based on differential evolution algorithm of soft measurement model of acetic acid, establishing soft measurement model based on DE_ELMRR.Second,support vector machine(SVM) and neural network training speed can not reach the demand of the actual process and the model of standard extreme learning machine is not instable because of selecting the parameters by random selection, thus,proposed a method called the wavelet kernel extreme learning machine algorithm (KELM),then establishing soft measurement model based on KELM. Finally, using process simulation software Aspen established steady and dynamic simulation, then collected acetic acid distillation process data.Soft sensor models based on DE_ELMRR and KELM methods are established for concentration estimation in acetic acid dehydration azeotropic distillation. DE_ELMRR method compared with standard extreme learning machine, the results show that the algorithm accuracy is greatly higher than the standard extreme learning machine, and also increase generalization ability.Then KELM method compared with SVM, simulations show that learning speed has improved significantly with high accuracy and generalization ability...