The measurement of high-energy cosmic ray spectra is important for understanding extreme astrophysical processes and the origins of cosmic rays, representing one of the core scientific objectives of the LHAASO-KM2A experiment. This study employs deep learning algorithms to directly extract event features from extensive raw data. Within the energy range of the "knee" region. This research employs ParticleNet, a graph-based neural network model that achieves markedly improved energy resolution and reduced bias compared to traditional parametric methods.