Gas Turbine Modeling by Using Neural Networks Trained on Field Operating Data

In the paper, Neural Networks (NNs) for the simulation of gas turbines are studied and developed in order to reproduce gas turbine behavior. The data used for both training and testing the NNs are collected from a FIAT Avio 701F single shaft gas turbine during normal operation over a period of about seven months. The paper mainly focuses on the choice of the data to be used for NN training and testing by evaluating their number and significance. The selection of data used for the NN training phase is performed according to an optimization procedure, which iteratively considers new data patterns. The performance of the best NN model is then tested over all the available field data. NN model response is also compared with that of a cycle program calibrated on the same gas turbine