Work

Global surrogate process model adaptation using self-adaptive characteristic maps for cost-optimal production scheduling on the example of a power-to-methanol process

Year: 2026

Type: article

Abstract: This paper describes a framework for the flexible scheduling of a novel power-to-methanol process, which uses self-adaptive characteristic maps as surrogate process models. A newly developed global adaptation algorithm adjusts the process model during operation using new measurements to account for plant-model mismatch due to sudden or incremental drifts in process behavior. Through parallel drift detection, structural errors in the characteristic map are identified at the operating points in use, and the characteristic map is then adjusted across the entire operating range using Gaussian process regression (GPR). This ensures an accurate characteristic map across the entire operating range. The adapted characteristic maps are then used to calculate a cost-optimal schedule, taking into account various production constraints and dynamic electricity prices, which is used in a plant emulation to calculate the outputs. The results show a rapid adaptation of the characteristic maps to changing process behavior using the newly developed algorithm. On average, the reduction in plant-model mismatch using the root mean squared error (RMSE) between the ground truth and the model characteristic map, considering a sudden drift of the process behavior, ranges between 76.7 % and 97.6 % using the developed global algorithm, while the local algorithm used as a benchmark has a significantly lower reduction of the RMSE of about 5 % to 12.7 %. The smaller plant-model mismatch reduces production costs by an average of 11.8 % compared to local adaptation and 21.4 % compared to no adaptation, considering various test cases. At the same time, the required production target is met more accurately and production constraints are met more often when using the global algorithm. In the case of incremental drift of the process behavior, production costs and the plant-model mismatch are also significantly lower compared to local or no adaptation.

Source: Digital Chemical Engineering

DOI: 10.1016/j.dche.2026.100353

Authors: Max Kollmer, Mücahit Terzi, Akasha Nonavinakere Venkatesh, Mohit Singh, Jörg Matthes

Institution: Karlsruhe Institute of Technology

Language: English

Topic: Advanced Control Systems Optimization

Subfield: Control and Systems Engineering

Field: Engineering

Domain: Physical Sciences

SDGs: Affordable and clean energy, Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $1,650