In　order　to　address　the　challenge　of　electric　vehicle　charging　decisions　under　multiple　uncertain　factors　in　the　power-transportation　coupling　system,　this　paper　proposes　a　graph　multi-agent　reinforcement　learning　algorithm　to　optimize　the　electric　vehicle　charging　guidance　strategy.　Firstly,　second-order　cone　optimization　is　employed　to　solve　the　optimal　power　flow　of　the　distribution　network,　deriving　the　marginal　node　prices　of　the　distribution　network　for　updating　the　charging　prices.　Then,　based　on　the　graph　theory,　the　information　of　vehicles,　charge　stations　(CSs),　traffic　roads,　and　the　distribution　network　is　transformed　into　a　dynamic　graph.　An　attention-based　graph　neural　network　multi-agent　reinforcement　learning　algorithm　is　adopted　to　optimize　the　electric　vehicle　charging　strategy.　Finally,　the　proposed　algorithm　is　simulated　and　validated　in　a　transportation　network.　The　simulation　results　indicate　that　the　proposed　graph　multi-agent　reinforcement　learning　algorithm　can　effectively　reduce　the　time　and　economic　costs　of　electric　vehicle　users,　promote　balanced　operation　of　CSs,　and　exhibit　good　adaptability　and　scalability　of　the　model.　©　Beijing　Paike　Culture　Commu.　Co.,　Ltd.　2025.