Izvorni znanstveni članak
Development of a Soft Sensor in a Propylene Production Plant
Željka Ujević Andrijić
; Department of Measurements and Process Control, University of Zagreb Faculty of Chemical Engineering and Technology, 10000 Zagreb, Croatia
Nikola Rimac
; Department of Measurements and Process Control, University of Zagreb Faculty of Chemical Engineering and Technology, 10000 Zagreb, Croatia
Ines Martić
; Department of Digital transformation, INA - Industrija nafte, d.d., Avenija Većeslava Holjevca 10, 10020 Zagreb, Croatia
Sažetak
Propylene is a crucial intermediate in petrochemical synthesis, which requires high purity. This study details the creation of a soft sensor model for continuous monitoring of propylene levels in a propane/propylene splitter refinery facility. The data obtained from the plant of the input variables are processed using different pre-processing methods. Soft sensor models of neural networks with multilayer perception (MLP) and neural networks with long short-term memory (LSTM) were created using the Python programming language. During the development of the MLP model, various hyperparameters were tested, including the number of neurons in the hidden layer and the impact of the activation function type on the model’s quality. Similarly, when developing the LSTM model, the number of LSTM units and the number of time steps into the past were also examined. A statistical analysis of the findings was performed, which revealed that both model types give strong correlation values between model data and real data of propylene content and that both neural network model types may be used in the refinery information system. The use of developed soft sensors guarantees that propylene content information is always up to date and continuous, allowing for fast responses to changes in propylene content for enhanced process management. The soft sensors improve the end product’s quality and can result in considerable cost savings.
Ključne riječi
soft sensor, neural networks, multi-layer perceptron, long short-term memory networks, propylene
Hrčak ID:
327427
URI
Datum izdavanja:
3.2.2025.
Posjeta: 0 *