Speaker
Description
Energy reconstruction is a key component of gamma-ray spectral measurements with LHAASO-WCDA. While conventional methods use the number of triggered detector units, Nhit, as a proxy for shower size and primary energy, the spatial, timing, and charge information of individual hits can provide additional constraints. We develop a ParticleNet-based approach to event-level energy regression, using a graph neural network to learn correlations among detector hits and incorporating event geometry into the reconstruction. Reconstruction bias and energy resolution are evaluated with simulated events across different energy, Nhit, and zenith-angle ranges. To explore its application to observations, we construct a two-dimensional analysis in Nhit and predicted energy, together with the corresponding detector response, and apply a forward-folding approach to the Crab Nebula spectrum. This poster presents the model design, performance evaluation, and preliminary Crab analysis. We discuss the effects of background estimation, the point-spread function, and simulation–data differences, as well as comparisons within a common analysis framework to assess the additional information provided by the predicted energy beyond Nhit.
