The nanoparticle aggregation has significant influence on heat transfer properties. The contradictions between different researches exist and the underlying mechanisms of heat conduction remain unclear. To address these issues, the thermal conductivity of solar salt-SiO2 nanofluids in the non-aggregated and aggregated states was investigated, based on the combination of the advantages of molecular dynamics (MD) in revealing physical insights and lattice Boltzmann method (LBM) in simulating properties for complex structures. LBM models were simplified on the basic of MD results. Different heat transfer physics in nanofluids at nanoscale and mesoscale were compared. The results show that the thermal conductivity of aggregated nanofluid is higher than the non-aggregated nanofluid. By analyzing the diffusion coefficient, number density and vibrational density of states, we found that the previous hypotheses (such as Brownian motion, interfacial layer and interfacial thermal resistance) cannot explain the enhanced thermal conductivity induced by the nanoparticle agglomeration. The heat transfer mechanisms of aggregated nanofluid were further analyzed from new perspectives, i.e., the contributions of different material components and fluctuation modes to heat flux. It is found that the nanoparticle aggregation leads to the nanoparticle contribution increase, indicating the formation of low thermal resistance channels within the aggregates. The heat conduction contributed by collision decreases in the aggregated state, while that contributed by potential energy increases. Moreover, the thermal conductivity of aggregated nanofluid is negatively correlated with temperature, aggregate size and fractal dimension of agglomerates, and positively correlated with mass fraction of nanoparticles, degree of agglomeration, and backbone length of each aggregate.