Efficient neural network models of chemical kinetics using a latent asinh rate transformation

Abstract

We propose a new modeling strategy to build efficient neural network representations of chemical kinetics. Instead of fitting the logarithm of rates, we embed the hyperbolic sine function into neural...

Goal‐Oriented Two‐Layered Kernel Models as Automated Surrogates for Surface Kinetics in Reactor Simulations

“…In contrast, surrogate models like splines or the (error-based modified) Shepard interpolation approach map precomputed solutions to accelerate reactor simulations [2,15,16,18,20,[27][28][29][30][31] or even spatial subsystems of the reactor [32,33] and breakthrough curves [34]. Lately, primarily machine learning techniques like random forests [35,36] or neural networks [2,3,37] have been used for accurate predictions of steady-state surface kinetics because they can overcome the so-called curse of dimensionality [38], i.e., the exponentially increasing difficulty to learn high-dimensional data. A promising alternative are kernel methods because their training is deterministic and data efficient even for highdimensional problems.…”

“…Training data for sampling-based surrogate models are usually generated on a simple grid without considering further information about the system's behavior [2,3,15,16,20,29,35]. However, the input-output behavior typically varies strongly over the input domain, e.g., due to sharp transitions between regions of high and low catalytic activity [18].…”

“…The rate constants for adsorption reactions k ads;i j of species i and the rate constants for all other surface reactions k surf j are calculated using Eqs. ( 5) and ( 6), respectively, with the site density G (2.49081 10 -5 mol m -2 ), the universal gas constant R (J mol -1 K -1 ), the temperature T (K), the molecular mass M i (kg mol -1 ), the reference temperature T 0 (300 K), the temperature exponent b (unitless), the sticking coefficient s 0,i (unitless), the preexponential factor A j (s -1 ), and the activation energy E A,j (J mol -1 ) [2,3,44].…”

“…Machine learning methods such as neural networks find increasing application in chemical engineering [1], e.g., to model complex reaction kinetics [2,3], predict reactor behavior [4], and assist catalyst design and discovery [5]. A promising alternative to neural networks are kernel methods, especially the Vectorial Kernel Orthogonal Greedy Algorithm (VKOGA) [6].…”

Increasing Computational Efficiency of CFD Simulations of Reactive Flows at Catalyst Surfaces through Dynamic Load Balancing

“…Moreover, their accuracy relies on many tunable parameters whose definition is empirical or requires several trial-and-error analyses . On the other side, machine learning is driving attention to off-the-fly tabulation of chemical source terms based on polynomial approximation, , ensemble methods, , or artificial neural networks. − Such approaches revealed to be very effective in reducing the computational costs and enabled the inclusion of extremely expensive kMC simulations into CFD analysis . These approaches have been employed to speed-up the computations of the reactive source terms in the context of heterogeneous chemistry, whereas their use for the approximation of the entire chemical substep in the operator-splitting approach is mainly confined to homogeneous chemistry simulations. − …”