loading page

Approximately Optimal Fixed-Structure Controllers Using Neural Networks
  • +3
  • Daniel McClement,
  • Nathan P. Lawrence,
  • Philip Loewen,
  • Michael Forbes,
  • Johan Backstrom,
  • Bhushan Gopaluni
Daniel McClement
The University of British Columbia

Corresponding Author:[email protected]

Author Profile
Nathan P. Lawrence
University of British Columbia
Author Profile
Philip Loewen
University of British Columbia
Author Profile
Michael Forbes
Honeywell International Inc
Author Profile
Johan Backstrom
Backstrom Systems Engineering Ltd.
Author Profile
Bhushan Gopaluni
University of British Columbia
Author Profile


Fixed structure controllers (such as proportional-integral-derivative controllers) are used extensively in industry. Finding a practical and versatile method to tune these controllers, particularly with imprecise process models and limited online computational resources, is an industrially relevant problem which could improve the efficiency of many plants. In this paper, we present two flexible neural network-based approaches capable of tuning any fixed structure controller for any control objective and process model and compare their advantages and disadvantages. The first approach is derived from supervised learning and classical optimization techniques, while the second approach applies techniques used in deep reinforcement learning. Both approaches incorporate model uncertainties when selecting controller parameters, reducing the need for costly experiments to precisely estimate model parameters in a plant. Both methods are also computationally efficient online, enabling their widespread usage.