Neural-network-based control of flow and temperature in the forcing vaporizer system of an LNG carrier

Abstract

The purpose of this thesis is to improve the forcing vaporizer system control performance of Liquefied Natural Gas Carrier (LNGC) which is constructed in DSME (Daewoo Shipbuilding & Maritime Engineering). During voyage, an LNG carrier requires fuel gas from the cargo tanks loading liquefied natural gas (LNG) and forcing vaporizer evaporates LNG to meet fuel demand of boiler. However unstable flow and temperature control of forcing vaporizer system which is controlled by Distributed Control System (DCS) based PID controls decrease overall vessel performance and stability. The heat exchanger of forcing vaporizer system cannot be precisely considered as a dynamic modeling because its model has two phase change streams and complex structure and thermal state to analyze. Therefore a feed-forward neural network controller with no a priori knowledge regarding process dynamics is suggested to improve forcing vaporizer system control and then is configured as a controller to the forcing vaporizer system process. Actual input/output data are obtained by an LNG carrier test to learn the neural network and make a HYSYS modeling. The neural network controller is tested with a commercial program HYSYS to verify the control performance because of controller performance test environment difficulty and heavy test cost in an actual LNG carrier. The proposed neural network controller has an existing controller such as a Distributed Control System (DCS) or Programmable Logic Controller (PLC) and to get neural network training data and to control the system when the neural network controller is not used. The neural network controller shows very fast recovery under the effect of load changes. The neural network is better controller than a PID controller for flow and temperature control in the forcing vaporizer system. If the well trained neural network is applied to forcing vaporizer system as researched in this thesis, the advanced control performance is expected.