An internal shock model calibrated with real gamma-ray burst light curves using a genetic algorithm

The origin of gamma-ray burst (GRB) prompt emission remains an open question. The internal shock (IS) model is a leading scenario for converting relativistic ejecta kinetic energy into gamma rays, but its parameters have not yet been fully calibrated against observed GRB light curves (LCs) to reproduce their diversity. We adopt a machine-learning framework to optimise the IS model by comparing simulated and observed LC properties from three GRB catalogues (Swift/BAT, Fermi/GBM, CGRO/BATSE). Assuming a redshift-dependent GRB formation rate, we employ a genetic algorithm to minimise a loss function based on six independent metrics capturing both average behaviours and statistical distributions. The optimised model reproduces several key observational properties, including the average post-peak temporal profile, autocorrelation function, and the distributions of duration, signal-to-noise ratio, number of peaks, peak flux, and fluence. We also derive constraints on the central engine activity: (i) the number of emitted shells is well described by a generalised Zipf distribution, analogous to the Gutenberg-Richter law for earthquakes, and (ii) the rest-frame shell-emission times follow a negative exponential distribution, indicating a stochastic process with a constant ejection probability. This calibrated IS model provides a physically grounded framework for interpreting GRB variability and predicting GRB populations detectable by future missions.