Rapid urbanization has intensified ecosystem service supply–demand mismatches, severely constraining regional sustainable development. Ecological security patterns represent a critical strategy to address these challenges, yet existing studies predominantly focus on natural supply while neglecting human ecological demands. This study develops an integrated framework combining XGBoost machine learning and ecological network theory to assess and optimize ecological security patterns from a supply–demand perspective, using the Chengdu–Chongqing Urban Agglomeration in the upper Yangtze River Basin as a case study. Results reveal a distinct spatial pattern of “high supply in peripheries and high demand in urban cores”. Ecological demand sources and corridors have experienced explosive growth, expanding significantly faster than supply components. While the ecological supply–demand network has become more structurally complex with improved structural connectivity overall, the supply–demand balance has progressively deteriorated. Targeted optimization through supplementing key ecological sources and refining network structure enhanced connectivity by 136.46% and improved supply–demand balance by 35.4%. A “three zones, four corridors, six nodes” optimization scheme is proposed. This research provides valuable insights for mitigating ecological conflicts in rapidly urbanizing regions and informs evidence-based decision-making for territorial ecological restoration and public policy design to balance conservation and socioeconomic development.