Optimisation of Soft Sensor Models for Crude Distillation Unit (CROSBI ID 598008)
Prilog sa skupa u zborniku | sažetak izlaganja sa skupa | međunarodna recenzija
Podaci o odgovornosti
Mohler, Ivan ; Novak, Mirjana ; Golob, Marjan ; Ujević Andrijić, Željka ; Bolf, Nenad
engleski
Optimisation of Soft Sensor Models for Crude Distillation Unit
From industrial facilities greater production effectiveness and compliance with the environmental laws are expected. Hence, there is a need for continuous measurements and more effective process control, which imposes the need for monitoring a large number of process variables using appropriate measuring devices. Due to growing fuel quality demands, continuous measurements product quality properties in the crude distillation unit (CDU) are necessary. One of the key diesel fuel properties is cold filter plugging point (CFPP) which is usually determined only by laboratory assays. On the basis of available continuous measurements of the temperatures, flows of relevant process streams and laboratory analysis taken from the CDU, the soft sensor models for the estimation of CFPP have been developed. Cold filter plugging point is defined as the temperature where the fuel filter plugs due to crystallization of mostly higher linear paraffinic hydrocarbons. Process variables were measured continuously and recorded in distributed control system. The laboratory assays of CFPP were carried out four times a day. Data preprocessing included detecting and outlier removal, generating additional outputs by Multivariate Adaptive Regression Splines (MARSplines) algorithm, detrending and filtering. Soft sensors are developed using linear and nonlinear identification methods. Results of the Output Error (OE) model, Hammerstein–Wiener (HW) model and neuro-fuzzy model are shown. Model structures are optimized by Genetic Algorithm (GA) and ANFIS (Adaptive Neuro-Fuzzy Inference System) algorithm. The soft sensor models based on ANFIS were generated in two stages: 1) fuzzy subtractive clustering algorithm was used to extracting a set of rules that models the data behavior ; 2) tuning of FIS parameters was performed using hybrid learning algorithm. For the consequent parameters training Least Squares method (LS) was used. The premise parameters were fixed. After the consequent parameters were adjusted, the approximation error was back- propagated through every layer to update the premise parameters as the second step. This part of the adaptation procedure was based on the gradient descent principle. Developed models were evaluated based on the final prediction error (FPE), root mean square error (RMSE), mean absolute error (AE) and FIT values. Results obtained by the OE and HW models on the validation data set show very good agreement with laboratory assay and splined data. Therefore these models can be used as soft sensors in the real plant. Results obtained by the neuro-fuzzy ANFIS model, validated on independent data set on three different numbers of epochs show excellent agreement. ANFIS model is the best one and it can be used for inferential process control.
soft sensor; process control; nonlinear identification; ANFIS; crude distilation unit; cold filter plugining point
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
Podaci o prilogu
---.
2013.
objavljeno
Podaci o matičnoj publikaciji
ECCE9 / ECAB2
Harmsen, Jan ; Van den Akker, Harry ; Ferreira, Guilherme
Den Haag: European Federation of Chemical Engineering
Podaci o skupu
ECCE9 / ECAB2
poster
21.04.2013-25.04.2013
Den Haag, Nizozemska