Evolutionary Computing Methods for Parameter Optimization of a Thermal Power Plant Operator Training Simulator

J. Alam Jan and B. Šulc (Czech Republic)


evolutionary computing, genetic algorithm, simulated annealing, optimisation techniques, thermal power plant training simulator


For most of models a good knowledge of parameters is important. However, not in all cases, it is possible to use records of real process data as in case of operator training simulators. In operator training simulators designed in [1], such data are relatively easily available. Usually in processes like electric energy generation, there is no chance to perform any special experiments. It means, that only special evaluation methods using data from a normal plant operation can be used to produce information about model parameters that are very difficult to compute analytically. Exactness of these parameters needn’t be extremely high but it must ensure a good likeness of simulated behaviour by the training simulator. Two global optimisation techniques - “Simulated Annealing (SA)” and “Adaptive Genetic Algorithm (AGA)” have been used in these tests of suitable parameter optimisation methods. A fuzzy logic controller (FLC) has been applied to adjust the parameters of AGA. The root mean squared error (rmse) method has been used as an objective function for testing achievement of optimal parameter values. In this parameter optimisation procedure the parameters of model has been changed either by SA or AGA techniques to minimize the objective function rmse.

Important Links:

Go Back