Physics-guided mixture density network architecture and predictive performance. (IMAGE)
Caption
Physics-guided mixture density network architecture and predictive performance. The PgMDN integrates hydraulic variables (ΔHu, ΔHd, Qu, Qd, Hd) into LSTM layers and enforces two physical constraints into the loss function: local mass balance and a coupling between mean variation and uncertainty. The model outputs a conditional probability distribution p(qt|xt) of lateral offtake discharge. Evaluated against a standard mixture density network, the PgMDN achieves lower prediction errors (MAE, RMSE) and substantially improved reliability and skill scores. SHAP analysis identifies key hydraulic drivers of predictive uncertainty, supporting the model's interpretability for real-world canal operations.
Credit
Environmental Science and Ecotechnology
Usage Restrictions
Credit must be given to the creator. Only noncommercial uses of the work are permitted. No derivatives or adaptations of the work are permitted.
License
CC BY-NC-ND