[Objective] To address water quality deterioration in Dianchi Lake caused by combined agricultural and urban non-point source pollution as well as confluence of rainwater and polluted water， this study constructed a Bayesian Network （BN） risk assessment framework. This framework integrates pollution identification， spatial tracing， and probabilistic inference to support “zoning management and targeted restoration” strategies in Dianchi Lake. [Method] Water quality data from 10 national control sections （2021-2024） in Dianchi Lake were pre-processed by eliminating outliers and interpolating missing values. Total Phosphorus （TP） was identified as the typical pollutant via Spearman correlation analysis. A BN structure was constructed based on TP concentrations using a hybrid algorithm： a Most Weight Supported Tree （MWST） derived from the Mutual Information matrix， node ordering via Breadth-First Search （BFS）， and structural optimization using the K2 score function. Parameters were learned using the Expected Maximization （EM） algorithm after discretizing data according to the Surface Water Environmental Quality Standard （GB3838-2002）. Model performance was evaluated using Leave-One-Out cross-validation. The validated BN was applied to assess water quality risk through forward prediction， sensitivity analysis， reverse tracing， and causal chain identification. [Result] The model achieved an average validation accuracy exceeding 80%. In 2025， the probability of lake-wide water quality exceeding Class IV standards is <15%. By 2035， the probabilities of exceeding Class III standards are projected as follows： DQ （66%）， LJY （59%）， GYSD （42%）， CHZX （41%）， DCN （38%）， HKX （29%）， BKY （26%）， GYSZ （22%）， and GYSX （14%）. Sensitivity analysis indicated that water quality at HKX is most significantly influenced by HWZ （0.469 2）， BYK （0.367 5）， and GYSZ （0.186 7）. When HKX water quality is Class V， the probability of Class V conditions at BYK， GYSZ， GYSD， LJY， and HWZ exceeds 40%； when HKX is Class IV， the probability of Class IV conditions at BYK， GYSZ， GYSD， LJY， HWZ， and DCN exceeds 80%. The identified causal pollution chain is HWZ→BYK→GYSZ→GYSD→LJY. [Conclusion] The MWST-BFS-K2 hybrid algorithm effectively quantified water quality risk in Dianchi Lake. In 2025， the risk of exceeding Class IV standards is extremely low lake-wide. By 2035， DQ faces high risk； LJY， GYSD， and CHZX face medium risk； while DCN， HKX， BYK， and GYSZ face low risk. The risk of exceeding Class III standards at GYSX is negligible. The pollutant migration path follows “HWZ→BYK→GYSZ→GYSD→LJY.” Management should prioritize blocking pollution transmission from GYSZ to GYSD. Water quality fluctuations at HWZ， BYK， and GYSZ significantly impact HKX. Key pollution sources for Haikou River include BYK， GYSZ， GYSD， LJY， HWZ， and GYSX in 2025， with DCN replacing GYSX as a priority area by 2035. This framework provides a scientific basis for the ecological management of plateau lakes.