The MELCOR code as a severe accident simulation tool does not have the capability to capture the quench process of a debris bed which may form in the wet cavity during a severe accident of light water reactors (LWRs). Although the coupled MELCOR/COCOMO simulation could overcome the limitation (Chen et al., 2022), the calculation time was explosively escalated due to mechanistic modeling of debris bed thermal-hydraulics in COCOMO. To suppress the computational cost, a surrogate model (SM) was developed in our previous study (Wang et al., 2023), and its coupling with MELCOR could realize a quick estimation of the quench process of one-dimensional debris beds. The present study is an extension of the previous work, aiming at the development of a new surrogate model for the quench process of two-dimensional conical debris beds. The new surrogate model (SM) was based on artificial neural networks (ANNs) and trained by the database from COCOMO calculations of various conical debris beds quenched in the reactor cavity of a Nordic boiling water reactor (BWR). The MELCOR was then coupled with the new SM to simulate a postulated station blackout (SBO) scenario in the BWR. The results show that the coupled MELCOR/SM simulation could provide similar ex-vessel debris bed quench period and containment pressure/temperature trends as the coupled MELCOR/COCOMO. Compared with the MELCOR standalone calculation, the coupled calculations predicted earlier points of time for water pool saturation and containment venting, since the heat transfer from conical debris bed to water pool is faster in the coupled simulations.