Conference article

Hybrid Model for Fast Solution of Thermal Synchronous Generator With Heat Exchanger

Khaled Aleikish
University of South-Eastern Norway, Porsgrunn, Norway

Madhusudhan Pandey
University of South-Eastern Norway, Porsgrunn, Norway

Thomas Øyvang
University of South-Eastern Norway, Porsgrunn, Norway

Bernt Lie
University of South-Eastern Norway, Porsgrunn, Norway

Download article

Published in: Proceedings of The 61st SIMS Conference on Simulation and Modelling SIMS 2020, September 22-24, Virtual Conference, Finland

Linköping Electronic Conference Proceedings 176:13, p. 91-98

Show more +

Published: 2021-03-03

ISBN: 978-91-7929-731-2

ISSN: 1650-3686 (print), 1650-3740 (online)


Overheating of synchronous generators may lead to a shortened generator lifespan, thus strict constraints are imposed on their operation. A common constraint is to restrict the power factor of the generator to lie below, say, 0.86 overexcited. In some recent work, a dynamic thermal model of the generator with cooling heat exchanger has been developed; the idea is that this allows for better monitoring of generator temperature, while relaxing on the power factor constraint. The current model is only valid for an ideal case of constant heat capacity. In this work, the generator model is extended to allow for temperature dependence in heat capacity of water and air in the heat exchanger model. The consequence of this more realistic model, is that it is no longer possible to find an explicit, analytic solution of the heat exchanger model, and it is now necessary to instead solve numerically a two point boundary value problem for each time step. It is shown that the effect of temperature dependence in the heat capacities has a noticeable effect on the solution of the model. The inclusion of on-line numeric solution of the heat exchanger model does, however increase the computation time of the thermal generator model by a factor of several thousand. Here, we instead study the possibility to solve the numeric heat exchanger model multiple times off-line, and then fit a regression model that gives a correction to the analytic solution. Both linear regression and nonlinear regression (neural network) is considered. Both types of regression models allow for a speed-up in the computation time of the thermal generator model of a factor of around 2000. In the computations, computer language Julia was used, with the DifferentialEquations and the Flux packages.


linear regression, nonlinear regression, thermal model, machine learning, surrogate model, hybrid model


No references available

Citations in Crossref