A New Processing Method Combined with BP Neural Network for Francis Turbine Synthetic Characteristic Curve Research

Abstract EN

A BP (backpropagation) neural network method is employed to solve the problems existing in the synthetic characteristic curve processing of hydroturbine at present that most studies are only concerned with data in the high efficiency and large guide vane opening area, which can hardly meet the research requirements of transition process especially in large fluctuation situation.

The principle of the proposed method is to convert the nonlinear characteristics of turbine to torque and flow characteristics, which can be used for real-time simulation directly based on neural network.

Results show that obtained sample data can be extended successfully to cover working areas wider under different operation conditions.

Another major contribution of this paper is the resampling technique proposed in the paper to overcome the limitation to sample period simulation.

In addition, a detailed analysis for improvements of iteration convergence of the pressure loop is proposed, leading to a better iterative convergence during the head pressure calculation.

Actual applications verify that methods proposed in this paper have better simulation results which are closer to the field and provide a new perspective for hydroturbine synthetic characteristic curve fitting and modeling.