Data-Driven Open-Loop Simulation for Digital-Twin Operator Decision Support in Wastewater Treatment
A new controlled continuous-time state-space model (CCSS-RS) has been developed for digital-twin-style decision support in wastewater treatment plants (WWTPs). The model separates historical state inference from future control and exogenous rollout, using typed context encoding, gain-weighted forcing, semigroup-consistent rollouts, and Student-t plus hurdle outputs to handle heavy-tailed and zero-inflated sensor data. Tested on the public Avedøre full-scale benchmark with 906,815 timesteps, 43% missingness, and irregular sampling intervals of 1-20 minutes, CCSS-RS achieved an RMSE of 0.696 and CRPS of 0.349 at a horizon of H=1000 across 10,000 test windows. This represents a 40-46% reduction in RMSE compared to baseline methods. The model addresses the open engineering-AI challenge of simulating plant response under prescribed control plans while tolerating irregular and missing sensing over 12-36 hour planning horizons.
Key facts
- CCSS-RS is a controlled continuous-time state-space model for WWTP digital-twin decision support.
- The model separates historical state inference from future control and exogenous rollout.
- It uses typed context encoding, gain-weighted forcing, semigroup-consistent rollouts, and Student-t plus hurdle outputs.
- Tested on the public Avedøre full-scale benchmark with 906,815 timesteps.
- 43% missingness and irregular sampling intervals of 1-20 minutes.
- Achieved RMSE 0.696 and CRPS 0.349 at H=1000 across 10,000 test windows.
- Reduces RMSE by 40-46% relative to baselines.
- Supports 12-36 hour planning horizons.
Entities
Locations
- Avedøre