The Study Of Steam Turbine Exhaust Wetness Based On Soft-sensing Technique

The last few stages of condensing turbines in the large-scale power plants andmost stages of nuclear turbines are often in the wet steam region. The wet steam maylead to moisture loss which decreases the efficiency of steam turbine and the erosionon blade which may threaten the safety of the steam turbines. Thus the calculationand measurement of the steam turbine exhaust wetness is particularly important. Thecurrent calculation and measurement methods of steam turbine exhaust wetnessmainly focus on thermodynamic methods and optical methods. But these methodshave some defects in the algorithm and require complex measurement devices nearthe steam turbine outlet generally, which not only cause interference with the mainsteam flow, but also lead to harmful consequences once the measurement devicesparts fall into the low pressure cylinder. To the problems of the thermodynamicmethods and optical methods, a new method for the steam turbine exhaust wetnessmeasurement and calculation based on soft-sensing technique is presented in thisthesis.First of all, this thesis summarizes the development of soft-sensing techniqueand researches the theory of the soft-sensing technique deep. Also this thesis analyzesthe influencing factors of steam turbine exhaust wetness to select auxiliary variables.Then the thesis establishes the sample set for the soft-sensing model by thethermodynamic performance test data of a thermal power plant.Furthermore, BP artificial neural network, RBF artificial neural network andsupport vector machine are applied to establish soft-sensing models of steam turbineexhaust wetness. In order to overcome the disadvantage of local minimum points ofBP neural network, the thesis uses the genetic algorithm and particle swarm algorithmto optimize the initial weights of BP neural network; similarly, for the support vectormachine model, this thesis also adopts genetic algorithm and particle swarm algorithm to optimize some parameters; and for the RBF artificial nuural network, thethesis uses OLS algorithm to optimize the center vector. By the simulation, theoptimized forecast models improve the prediction accuracy significantly. Then thisthesis selects the optimal forecast model as the final model after the analysis andcomparison for all the forecast models.Finally, although the above methods could have been able to forecast the exhaustwetness accurately, it is necessary to obtain the equation between exhaust wetnessand various influential factors. So the improved high dimension nonlinear partial leastsquares method is applied to establish the independent equation between exhaustwetness and various influential factors.