The　heating,　ventilation,　and　air　conditioning(HVAC)　system　consumes　a　large　amount　of　energy　in　buildings.　Accurate　modeling　of　the　HVAC　refrigeration　room　system　is　crucial　for　building　temperature　control　and　optimization　of　energy　consumption.　In　this　paper,　Levenberg　Marquardt　(LM)　algorithm　and　particle　swarm　optimization　(PSO)　algorithm　are　used　to　establish　a　nonlinear　autoregressive　neural　network　model　(PSO-NARX)　for　the　modeling　of　water　chillers　in　HVAC　systems.　NARX　is　a　model　used　to　describe　nonlinear　discrete　systems.　At　the　same　time,　using　particle　swarm　optimization　algorithm　can　improve　the　accuracy　of　the　prediction　model.　The　experimental　results　show　that　the　PSO-NARX　model　can　effectively　model　and　predict　the　chiller　model,　and　its　performance　is　better　compared　to　traditional　DNN　neural　networks.