Process dynamics modeling of H₂S removal plant at Pt. Saka Indonesia Pangkah limited based on artificial neural network

In modeling the H2S removal plant, Artificial Neural Network (ANN) was used because the sour gas composition input in the plant was unknown, so modeling using chemical reactions was not possible. Feed-forward architecture with the Levenberg - Marquardt training algorithm is used in designing a dynamic model of the plant process, with the number of hidden layers 1. In the input layer, there are five nodes to find out the dynamics of the plant, namely sour gas flow rate, sour gas temperature, amine rate, amine temperature, and the pressure difference between sour gas and amine inputs. While at the output layer, there is one output node, namely ppm H2S on sweet gas. Trial and error methods are used in determining the best model by using a variation of the number of hidden nodes 1-10 in the hidden layer. The parameter used is the smallest RMSE value generated by the ANN model. The simulation is carried out, the best ANN model is obtained, with a 4-7-1 structure. The resulting RMSE value on the structure is 0.4947, with an RMSE value of 0.4947 indicating that the model has an error percentage of 9.95%.