.jpg?RenditionID=8)
Industrial processes from waste water treatment to factory packing lines, are controlled by manual or semi-automated rule-based approaches. Machine learning offers the opportunity to build control systems which can adapt to changing conditions and optimise themselves to meet business requirements – for example reducing energy use, cost, or improving output quality.
By building simulation models, or by learning how a process behaves from operational data, we can provide recommendations as to how process operations can be improved. Contact us to discuss
collaborative R&D projects to model and improve your business processes.
To see our AI controller in action and explore how it could be applied in your industry – contact us
Applications
Optimisation and control of industrial processes including:
- Production and packing lines
- Water treatment and distribution
Benefits
- Knowledge capture by training AI using expert input
- Improved efficiency of existing capital infrastructure
Conferences
The following poster was presented at the European Geosciences Union General Assembly, April 2018:
Deep Reinforcement Learning for wastewater treatment control: An
application
Lan Ngoc Hoang, Edward Pyzer-Knapp, and Christopher Thompson
IBM Research, STFC Daresbury Laboratory, Warrington, UK
Advances in modelling and control have helped improved resilience and performance of environmental systems.
Recently, Deep Reinforcement Learning has emerged as a promising technique in providing flexible and evolving
strategies in various domains. We present an application of Deep Reinforcement Learning implemented on two
controllers of a wastewater treatment model. In each of these controllers, a deep neural network was trained to
predict the value function Q which represents the objective function of energy operation cost, effluent quality
and environmental fines. The controllers learn to operate and adapt the treatment plant’s controller setting
under different weather and inflow conditions. The application demonstrates a potential for applying Deep
Reinforcement Learning on environmental systems to improve management and control.
