This paper presents a comprehensive study on wa-ter level management in the Great Lakes, integrat-ing SARIMA, Bayesian networks, SVR, and PIDmodels. A network model is constructed by divid-ing the system into rivers (dynamic prediction viaBayesian networks) and lakes (dynamic balancevia water balance equations). Eight stakeholder-related factors are analyzed using Pearson corre-lation and entropy methods, with NSGA-II algo-rithm deriving optimal water levels . The SSIFM-PID control model, combining SARIMA-SVRprediction and PID regulation, achieves optimallevels within fewer iterations. Sensitivity testsshow robust stability under dam outflow perturba-tions and environmental changes (moderate dis-turbances show decreasing index variation). ForLake Ontario, weighted models and TOPSIS areintroduced to address stakeholder priorities andsocial opinion.