Thermal　management　of　fuel　cells　is　crucial　for　maintaining　their　efficient　operation,　as　improper　temperature　regulation　can　significantly　reduce　fuel　cell　output　performance　and　lifespan.　With　the　growing　global　energy　demand　and　the　pursuit　of　sustainable　energy,　fuel　cells　play　a　key　role　in　achieving　renewable　energy　goals,　particularly　in　the　context　of　the　United　Nations　Sustainable　Development　Goal　(SDG-7).　Therefore,　ensuring　the　high　efficiency　and　stability　of　fuel　cells　is　essential　for　driving　the　transition　to　clean　energy.　However,　existing　thermal　management　strategies　are　primarily　based　on　steady-state　models　and　traditional　control　methods,　which　are　inherently　limited　by　slow　response　times,　low　control　accuracy,　and　insufficient　utilization　of　parasitic　cooling　power.　These　methods　struggle　to　address　the　nonlinear　dynamic　characteristics　and　multivariable　coupling　issues　of　fuel　cell　systems.　This　study　addresses　these　challenges　by　proposing　an　innovative　control　strategy　that　integrates　adaptive　learning　and　exploration　mechanisms　through　the　use　of　the　twin　delay　deep　deterministic　policy　gradient　(ALEDE-TD3)　algorithm　based　on　dynamic　models.　The　proposed　method　aims　to　minimize　parasitic　cooling　power　while　overcoming　the　limitations　of　traditional　approaches.　By　introducing　dynamic　exploration　and　learning　rate　decay　mechanisms,　the　strategy　significantly　enhances　the　system＇s　response　speed　and　control　stability,　successfully　mitigating　the　response　delays　caused　by　multivariable　coupling.　Simulation　results　show　that　the　ALEDE-TD3　strategy　outperforms　TD3,　DDPG,　and　PID　algorithms　across　all　key　performance　indicators,　with　temperature　overshoot　reduced　by　63.1　%,　87.4　%,　and　88.9　%,　respectively,　and　average　settling　time　reduced　by　46.7　%,　56.7　%,　and　59.1　%,　while　consistently　maintaining　the　lowest　parasitic　cooling　power.　This　approach　not　only　improves　the　overall　performance　of　the　thermal　management　system　but　also　holds　significant　implications　for　advancing　renewable　energy　goals,　demonstrating　the　immense　potential　of　reinforcement　learning　in　fuel　cell　control.　©　2025