Model predictive control of an LNG liquefaction process using Jmodelica.org
Master thesis
Permanent lenke
http://hdl.handle.net/11250/2561561Utgivelsesdato
2018Metadata
Vis full innførselSamlinger
Sammendrag
In this thesis Model Predictive Control (MPC) theory has been used to develop a controller for a Liquefied Natural Gas (LNG) liquefaction process using the Jmodelica.org framework. Jmodelica.org is an open source platform that combines Modelica models with the possibility of running dynamic optimization problems directly on these models. The model used is a cascade liquefaction process model written in Modelica, with the corresponding Optimal Control Problem (OCP) used by the MPC written in Optimica, where Optimica is an extension to the base Modelica language. The possibilities of using Jmodelica.org for the implementation of a Robust MPC are also briefly looked into.
During initial simulations the MPC is found to control the process to an optimal operating point under a given set of constraints on the inputs and system variables. The controllers objective is mainly to minimize the energy used by the three compressors for a given natural gas load. The MPC controls the process to this point, which is previously known optimal point, both with and without measurement noise on the dynamic states of the model. The MPC's performance is tested and compared to a Proportional Integral (PI) controller for different changes to environmental variables. For these tests the MPC controller performs better with regards to minimizing the total energy used, especially for changes that require new set-points to be used by the PI controller. However due to the PI controller not being fully optimized, and as there are clear improvements that can be made to the PI controller, no final decision can be made on the actual improvements introduced by the MPC. The robust MPC variant is also tested and compared to the standard MPC for a small unnoticed change. The robust controllers ability to withstand expected disturbances built in to its implementation, is shown at the cost of operating in a more conservative manner, with regards to total energy used by the three compressors in the process.
The nature of the Jmodelica.org platform lends itself in a practical manner to the development and implementation of MPC control structures. The benefits introduced by doing the process modeling work in Modelica, are combined with the wide range of options for solving dynamic optimization problems in Jmodelica. These problems can then be solved using efficient numerical solvers and algorithms already embedded within the platform. Some specific areas where more work can be done to improving the model and improve the MPC's performance are also discussed. These areas are most notably by either modifying the model, or adding more accurate process model units to better describe the dynamics present in each refrigeration cycles. As for the MPC controller itself, some changes that can be made to the transcription process are also discussed, which can potentially help further reduce the computational time of the controller.