(568m) A Genetic Algorithm-Based Distributed Sensor Placement for Water Gas Shift Reactor in IGCC Plant Using Kalman Filter
Integrated gasification combined cycle (IGCC) with CO2 capture is an attracting technology for power generation due to its high efficiency and near-zero emission. A particular unit that is crucial to the performance of IGCC is the water gas shift reactor (WGSR), which is responsible for adjusting the syngas H2/CO ratio to meet power production and carbon capture targets. The efficacy of the WGSR is greatly affected by number of disturbances, such as upstream feed disturbance, and faults, such as catalyst deactivation due to poisoning and fouling. An early detection of disturbances and faults can facilitate the preventive actions and ensure safe and efficient operation of the reactor. However, faults and disturbances are not essentially observable by measurements and require a state estimation of the internal state of the reactor for them to be detected. This can be done by using a first-principle model of the reactor and series of measurements distributed throughout the reactor. It is therefore necessary to find the optimal number and location of the sensors for efficient estimation of faults and disturbances.
A high-fidelity model of the WGS reactor was developed in our previous work. The model is presented by a system of non-linear differential and algebraic equations (DAE). The discretized equations form a space of grid-points that are available for state measurement. An extended Kalman filter was implemented to estimate the faults using the measurements chosen from the space. A genetic algorithm is then employed to search the space for obtaining an optimal location of the sensors. However, the approach is inefficient due to large search space and computational burden for non-linear state estimation. To address this, a reduced order model of the WGSR is developed by linearization of the non-linear model. The genetic algorithm uses the linearized model and a linear Kalman filter to find the optimal location of the sensors. This approach significantly reduces the computational work compared to our previous method.