Optimization of Waste-to-Resource Plants with Machine Learning and Multi- Objective Reinforcement Learning

In waste-to-resource processes, variations in operating conditions are common, to accommodate varying process objectives and ambient disturbances. In these cases, well-tuned operating conditions are disrupted, and correlations among process and quality variables are broken, leading to unexpected outcomes. Due to large number of involved variables and their complex relations, it is challenging to re-tune system parameters. Further, process optimization focus may change, or multiple objectives need to be optimized concurrently. Therefore, the goal of this project is to optimize the waste-to-resource plants owned by Anaergia by developing a set of advanced time series, machine learning and reinforcement learning algorithms, aimed at improving the existing parameter tuning and optimization mechanism. The proposed research program addresses the aforementioned challenges through 3 projects: a) Develop machine learning models to detect key correlations from the data for further process modeling; b) Design multi-objective reinforcement learning based recommendation system for multiple optimization objectives; c) Develop a process monitoring and diagnosis framework for real-time anomaly detection and root cause analysis.

Faculty Supervisor:

Qinqin Zhu;Hector Budman

Student:

Partner:

Anaergia Inc.

Discipline:

Engineering

Sector:

Professional, scientific and technical services

University:

University of Waterloo

Program: