Abstract
This work aims to study the feasibility of using an online feedforward artificial neural network (ANN) to control various actuators in a hybrid fuel cell gas turbine (FC-GT) simulation plant. This unique facility known as Hybrid Performance, or HYPER, is housed at the US Department of Energy’s National Energy Technology Laboratory in Morgantown, WV. Using a cyber-physical approach, HYPER incorporates a high-fidelity FC model in software, which interacts with a gas turbine and corresponding balance of plant components in hardware, in real time. This methodology allows research of FC-GT operational issues as well as control application studies for such systems in a safe manner. An open loop perturbation of the FC model load current is used to retrieve target data from load bank and bypass airflow valve actuators which control turbine speed and FC cathode airflow respectively. The steady state FC anodic side fuel flow is also fed to a supervised ANN which learns the pattern of actuator response to the given FC perturbations. By mimicking the manually operated actuators, the FC solid temperature gradient is maintained within safe operating bounds. The feedforward ANN is useful for its simplicity and flexibility in controlling a variety of desired actuator responses based on input combinations. The benefits and drawbacks of using ANN’s are discussed, as well as suggestions for improvement.
Palmvein Recognition using Block-Based WLD Histogram of Gabor Feature Maps and Deep Neural Network with Bayesian Optimization
