Reliable Predictive Control Exploiting Operational Data with Real-time Applications

Model predictive control (MPC) is widely used in process industries to control constrained systems with multiple input and outputs.

Traditionally, the MPC is used in a two layer architecture where the upper layer gives the economic optimal operating point and the MPC is used in the lower layer tracks the optimal operating point.

Recent studies shows that a significant improvement in the economic performance of the plant can be obtained if both the layers are combined together.

One of the pressing issues preventing the process industries from adopting the aforementioned control scheme is the presence of plant-model mismatch.

The work on this project uses ideas form the linear control theory to handle the structural plant-model mismatch in a robust NMPC framework, efficiently. We develop a model-error model (MEM) which uses plant measurements to improve the knowledge of the plant.

We focus on developing a systematic way of choosing the MEM structure based on the data collected from an industrial production plant and use them for monitoring and control purposes.

We develop an algorithm which works in parallel with the commercially available advanced process control solutions and makes them robust to plant model mismatch. Our project builds a computationally tractable scheme for model-based NMPC robust against the plant-model mismatch. As a result, a safe, reliable and resource-efficient operation is established.

The theoretical developments of the project are implemented into a software package and released as an open-source project such that the collaboration with academia and industrial stakeholders is fostered.

A demonstration on an industrial production plant and a laboratory pilot plant is also planned to showcase the benefits of the developed techniques in the real-world environment. A sound dissemination plan of the project ensures that the project reaches its target audience.