The　effective　evaluation　and　improvement　of　agricultural　green　total　factor　energy　efficiency　(AGTFEE)　are　crucial　for　guiding　sustainable　agricultural　development.　The　directional　distance　function　(DDF),　which　can　evaluate　efficiency　values　and　provide　efficiency　improvement　paths,　has　attracted　widespread　attention.　However,　most　existing　research　on　DDF　is　based　on　the　farthest　target　principle,　often　resulting　in　costly　efficiency　improvement　paths.　To　address　this　issue,　this　study　proposes　a　novel　cross-DDF　based　on　a　learning　network　under　the　closest　target　principle.　The　proposed　model　is　applied　to　dynamically　analyze　AGTFEE　in　China　from　2013　to　2022　at　different　levels.　Compared　with　existing　research,　the　proposed　model　offers　more　feasible　and　cost-effective　quantitative　paths　for　improving　AGTFEE.　Moreover,　the　proposed　model　constructs　a　learning　network　based　on　the　interactions　among　decision-making　units　for　peer　evaluation,　avoiding　inflated　efficiency　values.　The　empirical　results　highlight　three　main　findings.　First,　over　the　decade　from　2013　to　2022,　China＇s　AGTFEE　has　exhibited　a　positive　trend,　achieving　significant　progress.　Second,　during　the　same　period,　the　balance　and　consistency　of　AGTFEE　development　have　improved.　However,　differences　remain　among　regions　and　provinces,　with　the　distribution　pattern　showing　＂best　in　the　east,　followed　by　the　west,　and　relatively　poor　in　the　center.＂　Third,　there　are　differences　in　the　improvement　paths　for　AGTFEE　among　provinces.　For　instance,　to　improve　AGTFEE　in　Hebei　Province　in　2022,　it　is　necessary　to　significantly　reduce　the　amount　of　pesticides　used　in　the　agricultural　production　process.