Abstract: BP neural network is  widely used in hydrological forecasting.
   Aiming at the handicaps in present methods such as slow convergence
  and easily getting into local optimization, this paper presents a novel hydrological forecasting
  method based on the tunneling function which is one of the effective deterministic methods.  The tunneling
   function method can find a lower minimizer by leaving the minimizer previously found. By repeating these
   processes, a global minimizer can be obtained at last. Experiments show the proposed method
  not only can obtain high accuracy in flood forecasting, but also has longer effective real-time.
  The hydrological forecasting
  method can be used in operational hydrological forecasting.

Key words: nonlinear programming, hydrological forecasting, tunneling function, BP neural network