Naslov
The Development of Soft Sensors for Continuous Estimation of 95% of the Distillation Point (D95) of a Diesel Product

Autori
Ujević Andrijić, Željka ; Bolf, Nenad ; Herceg, Srečko

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Skup
The 12th international conference Distillation & Absorption 2022

Mjesto i datum
Toulouse, Francuska, 18-21.09.2022

Vrsta sudjelovanja
Poster

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
soft sensor, neural network, random forest

Sažetak
Soft sensor models for continuous estimation of 95% distillation point (D95) of a diesel product in a hydrocracking plant of refinery have been developed. Based on continuously measured data collected from the plant and laboratory analysis, data driven models have been developed using different machine learning methods. Data preprocessing included the selection of adequate data for model development, outlier’s detection, as well as data filtering. After selection of influential input variables, neural network and random forest models have been developed. By altering the number of hidden neurons and transfer function, neural network models results have been analyzed. The statistical performance criteria of developed models on validation data set indicate that the neural networks and random forest models estimate D95 of a diesel product reliably. By applying the present case study results in the refinery information system, a more stable operation of the plant, higher product quality and significant savings are expected.

Izvorni jezik
Engleski

POVEZANOST RADA