Dynamic Simulation Data for CO2 Purification Process Optimization

Synthetic time-series data generated from a detailed dynamic process model of a CO2 purification plant, used to design and test a real-time optimization and Kalman filter-based control framework. It captures process variables, estimated states, and controller setpoints under various operational scenarios like ramps, load changes, and disturbances. The data was created to estimate unmeasured mass flows and compute optimal solvent flow setpoints to minimize usage while maintaining food-grade CO2 purity.

Use Cases

• Train and compare first- and second-order Kalman filters to estimate unmeasured mass flows like CO2 inlet gas flow and rich solvent flow from noisy measurements.
• Benchmark optimization algorithms like fmincon and genetic algorithms against baseline LG-ratio control to compute optimal lean solvent flow setpoints that minimize solvent usage.
• Analyze process variables such as flows, temperatures, pressures, and compositions under capacity ramps and disturbance tests to evaluate control system stability.
• Calculate key performance indicators including CO2 purity (>99.8%), lean solvent reduction (~24–25%), and production increase (~6.3%) to quantify the economic impact of an RTO strategy.

Strengths

• Data is generated from a detailed dynamic process model implemented in K-Spice, reflecting realistic sensor variance and process noise configured via covariance matrices.
• Covers operation from turndown to full capacity, including specific disturbance tests on inlet temperature, pump discharge pressure, and condenser pressure.
• Supports evaluation of concrete economic impacts, such as an implied annual revenue impact for a 55,000 t/y bio-CO2 plant.

Limitations

• The dataset is synthetic, generated from a simulation model, and may not fully capture all real-world process complexities and unmodeled dynamics.
• Scope is focused on a single-unit operation (CO2 scrubber) within a larger purification train, limiting broader plant-wide analysis.
• Key metadata such as row count, column count, and specific file formats are unknown, which complicates assessing the dataset's scale and structure.

Provenance

Source

Hvidsten, Jan Inge via DataverseNO Harvested Dataverse.

Collection Method

Synthetic/virtual dataset generated from a dynamic process model implemented in K-Spice, with Kalman filters and optimizers implemented in MATLAB and coupled via OPC.

Time Range

Covers operational scenarios including capacity ramps and disturbance tests, but no specific calendar date range is provided.

Freshness

The dataset was last updated on 2026-03-01.