Retrieval of carbon and inorganic phosphorus during hydrothermal carbonization: ANN and RSM modeling.

In this study, modeling and optimization of Hydrothermal Carbonization (HTC) of Poultry litter were conducted to convert it into high-value materials. The aim was to understand the process and predict the effect of the influencing parameters on the product properties. The recovery of Inorganic Phosphorous (IP) and Carbon (C) was regarded as the model's response, although temperature and reaction time were thought to be important variables. Response Surface Methodology (RSM) was used along with temperature and time data sets ranging from 150 to 300C and 30-480 min, respectively, to identify the parameters influencing the response, three-dimensional networks, and optimization. Next, Multilayer Perceptron (MLP) and Radial Basis Function (RBF) were used to compare the results and improve the model fit. For these two neural networks, 7 neurons in two layers and 14 neurons in one layer were the ideal numbers. With fewer neurons and better accuracy and efficiency, the MLP model beat RBF with lower Mean Squared Error (MSE) values for both C (0.0015812 vs. 0.0037103) and IP (0.0014376 vs. 0.00623011) recovery and a higher R