Simulation of power plant transients with artificial neural networks: Application to an existing combined cycle

Abstract

To maintain the high performance of gas-turbine-based combined cycles, transients must be properly taken into account in the design phase and efficiently monitored in the operational phase, because they are not negligible time intervals. The use of artificial intelligence techniques such as expert systems, fuzzy sets and neural networks (NNs), coupled with advanced measurement and monitoring devices, can provide a reliable and efficient monitoring system. An existing two-pressure-level combined cycle has been simulated by dividing its simplified model into blocks representative of the main elements. An NN is associated with each of these blocks. Once the training and testing of the NN are complete, using data from a simulator, the blocks are put either in a cascade arrangement or in a parallel arrangement, providing reliable systems that can predict the load-change transient behaviour of the entire plant. The parallel approach was then tested on data from the real plant. The excessive simplification introduced with the simulator required the addition of selected real cases to the training set that are able to fit the NN response to reality. The results obtained are encouraging for use in an on-line monitoring system which evaluates the difference between the measured data and the predicted data.