Abstract:

Abstract:Elman neural network has the adaptability to time varying characteristics, the sensitivity to historical data and the advantage of self learning, it can approximate to any nonlinear mapping by an arbitrary accuracy. Meanwhile, the gradient descent method can make function has properties of monotone decreasing and gradient converges to 0. Compared with the traditional Back Propagation neural network, the method which combining the gradient descent method and Elman neural network has better prediction precision for the short term reservoir inflow forecasting in hydropower station.

Key words: Keywords:Elman, neural network, gradient descent method, forecasting, reservoir inflow