AI for Waste Water Treatment
summary
Only one of seven airflow sensors works, so a TensorFlow neural net is trained on one lane plus one year of one-minute samples to predict per-lane air flow, pressure and energy use; that model is packaged with Flask/Docker on Azure and served as an API and validated by domain experts. Control uses DDPG reinforcement learning with the learned simulator model, pretrained on current control behavior, and augmented data to prevent impossible actions.
AI-written summary · about
chapters (18)
- 0:00Intro
- 0:35Who am I
- 1:14About Waternet
- 2:04About the wastewater treatment plant
- 4:34How wastewater treatment works
- 6:20Aeration process
- 6:33Schematic
- 9:59Energy Consumption
- 11:19Model
- 16:03Results
- 16:26Deployment
- 18:00reinforcement learning
- 20:22DDP key algorithm
- 22:21Implementation
- 23:00Actor Critic
- 24:14Monitoring Tool
- 24:48Learnings
- 28:32Questions
related talks
17mAI Assisted Load Sharing for Improved Energy Efficiency
36mAI-Powered Real-Time Quality Control
21mApplication of AI technologies in nuclear power plants
28mAI vs Climate Change: Insights from the Cutting Edge
26mFrom Data to Decision: Monitoring and Improving Water Networks using AI/ML
18mAI Owes its Rebirth to Open Source and the Cloud
41mAI in Web: TensorFlow js
38mAutomatic retraining of TensorFlow models