**6. A computational approach for conversion: Density functional theory and artificial neural network**

Machine learning is one of the most powerful too of artificial intelligence and has been utilized for numerical prediction, classification, efficiency and pattern recognition. Among ML tools, artificial neural network (ANN) has become the popular nonlinear algorithm, adaptive structure, tunable and easy to train for various

#### **Figure 9.**

*Typical example of feed-forward approach using artificial neural network architecture for input/output variables, output representing system efficiency, (b) schematic cartoon illustrating the possible adsorption site (charge depletion, and charge accumulation), (c) partial PDOS showing ORR. Reprinted from the literature [139].*

applications such as catalyst, biology and energy [138]. In general, ANN architecture is comprised of at least three (3) layers, i.e. input layer, hidden layer and output layer as depicted in **Figure 8**. Each layer contains numerous neurons that connect to the next layer, where they connect it represented by weights. As depicted in **Figure 9**, ANN is composed of single layers (Hidden layers) that represent the parameters such as current and cell potential. The output is represented by the efficiency of the system. ML utilizing the artificial neural network has attracted considerable interest in energy conversion and electrocatalytic reaction to predict the throughput using the algorithms. ML has attracted many researchers in the field of catalysis. Lu et al. leverage the use of neural networks and density functional theory (DFT) for predicting the surface defects for oxygen reduction reaction. Mehiritz et al. report the first model of electrooxidation of ethanol using an artificial neural network (ANN) utilizing the differential evolution (DE) algorithm. The best results (Model) were obtained with a single hidden layer [140].

In addition, theoretical techniques, such as quantum-chemical modeling employing density functional theory (DFT, Vienna ab initio simulation package (VASP) methodologies, are used to expedite the process of finding the ideal material. This route enables us to evaluate different catalytic materials without using experimental analysis, as well as determine the catalyst's adsorption qualities and the impact of the material's composition and structure on the kinetics of the catalytic process. Nrskov and co-authors first use DFT-based techniques to compute the adsorption energy of reactants and intermediates. In this section, the efforts to identify the highly active catalysts and relevant electrocatalysis reaction mechanisms have been studied, particularly focusing on the d-band centre location, the electronegativity of the central atom to the neighboring, along with density of state (DOS). These theoretical properties are utilized in electrochemical systems to determine the appropriate descriptor for catalytic activity, specifically towards the ORR and OER. As illustrated in **Figure 9**, this is the typical schematic diagram depicting the DOS pattern and d-band centre level. There has been extensive research on the theoretical modeling of energy conversion and storage to screen the catalyst process. For example, Sunday et al. reported reactive molecular dynamics on surface PtNiFe heteroatoms to model the catalyst mechanism. Based on the energy barrier, it was found that NiO efficient than monometallic Ni and Pt, the HOads and Hads are likely easier on NiO during water
