2025MNRAS.540..582H

We present a comprehensive analysis of the time-resolved spectral energy distributions (SEDs) of the blazar OJ 287 over a 15-yr period (2008-2023), using multiwavelength data. In the $\gamma$-ray band, multiple flaring episodes were observed, with the strongest flare reaching a peak flux of $(5.60\pm 1.11)\times 10^{-7}\:{\rm photons\:cm^{-2}\:s^{-1}}$ on MJD 55869.03 (2011 November 4). In the optical/UV band, the source was in an active state between MJD 57360 (2015 December 4) and 57960 (2017 July 26), during which the highest flux of $(1.07\pm 0.02)\times 10^{-10}\:{\rm erg\:cm^{-2}\:s^{-1}}$ was observed on MJD 57681.23 (2016 October 20). In the X-ray band, both the flux and spectral index exhibit variability. To investigate the origin of the broad-band emission from OJ 287, we systematically modelled 739 quasi-simultaneous SEDs using a leptonic model that self-consistently accounts for particle injection and cooling. This analysis is possible due to the recent development of a surrogate neural-network-based model, trained on kinetic simulations. This innovative time-resolved neural network-based approach overcomes the limitations of traditional single-epoch SED modelling, enabling to explore the temporal evolution of key model parameters, such as the magnetic field strength, Doppler factor, and electron injection distribution, across different states of the source. We identified distinct emission states characterized by unique combinations of magnetic field $B$, electron index $p$, and Doppler boost $\delta$, associated with different underlying mechanisms such as varying acceleration processes (e.g. shocks, turbulence) and magnetic confinement. The analysis provides insights into the jet physics processes, including particle acceleration mechanisms and dynamic changes in the jet structure.