AI-driven optimization of renewable steam gasification products using coal bottom ash and CaO adsorbents for efficient CO2 capture

This paper models the steam gasification of palm kernel shells with coal bottom ash and CaO adsorbent as a catalyst using the Multi-Layer Perceptron Neural Network (MLP-NN). The model considers the effects of CaO/biomass ratio, temperature, and coal bottom ash wt.% at a fixed steam/biomass ratio. The product gas composition (PGCs), syngas yield, HHVgas, and LHVgas of gas are utilized as inputs to the network, while the influence of parameters is employed as an output. Recent studies have focused on improving the accuracy of MLP-NN models by incorporating PGC when selecting parameters for the model's MLP-NN model inputs. The main contribution of this study is the development of an improved MLP-NN model by a novel primary difficulties chimp optimization algorithm (ChOA) and five benchmark metaheuristic-based methods to address the two main limitations of gradient descent learning algorithms, namely, trapping in local minima and poor convergence rate. An experimental gasification system (GS) at a pilot scale is used to compare the PGC estimated by the MLP-NN model with actual experimental data. For nearly all scenarios, the MLP-NN predictions showed excellent agreement with experimental data, achieving a coefficient of determination R2 = 0.999. The error metrics were also very low, with RMSE <0.06, MAD <0.04, and AARE <1%, underscoring the robustness of the model. Both anticipated and experimental values have been shown to be relatively negligible in terms of RMSE, MAD, and AARE.