Due to techno-economic assets, the demand of combined cycles (CC) is currently growing. Nowadays, in a diversified electricity mix, these plants are often used on a load cycling duty or in the intermediate load range. The ability to start quickly and reliably may be a decisional criterion for the selection of the plant, in addition to the design performance, the cost and the pollutant emissions. Therefore, together with the simulation of CC transients, a proper monitoring system aimed at keeping high plant performance during the transients is required.
With the help of advanced measurement and monitoring devices, artificial intelligence (AI) techniques as expert systems (ES) and neural networks (NN) can fulfill this duty.
The goal of this paper is to show that a NN technique can be used reliably to obtain the response of a complex energetic system, such as CCs, during a slow transient and consequently as part of an on-line monitoring system.
In this work, a CC power plant is simulated by dividing it into three blocks, which are representative of the three main elements of the CC: namely the gas turbine (GT), the heat recovery steam generator (HRSG) and the steam turbine (ST). To each of them a NN is associated. Once the training and testing of the NNs is carried out, the blocks are then arranged in a series cascade, the output of a block being the input of the subsequent one. With this solution, the NN-based system is able to produce the transient response of a CC plant when the input information are the GT inlet parameters.
The transient data, not easy to obtain from measurements on existing plants, are provided by the CCDYN simulator (Dechamps, 1995). The performance obtained by the NN based system are observed to be in good agreement with those given by CCDYN, the latter being validated on the basis of measurements in an existing plant. The NN code, providing the departures of the measured data from the predicted ones, can be considered as a proper system for on-line monitoring and diagnosis.