A Month Ramfall Forecast Model Research In Hongze Lake Based On The BP Algorithm

Long term rainfall forecast is a rainfall prospect for the next month or quarter which play an improtant role in many ways such as agricultural irrigation or flood disater prevention。 Especially in agricultural meteorology, there will be enormous economic benefits if we improve the precision of long term rainfall forecast. The traditional long term rainfall forcast methods are based on the statistical analysis of historical data.Artificial neural network (ANN) is a model imitating the biological nervous sysyem. Theoretically a ANN can approximate to any non-linear function. The impact factors of long-term rainfall are very complex, so far there is no comprehensive theoretical basis for long-term rainfall forecast, but we can use ANN to do this even if we don’t know the principles. The error back propagation ANN known as BP network has some defects when to solve problems such as slow convergence speed, local minimum or difficult to decide the structure of a network. According to some research some improved algorithms can make up these shortcomes. These algorithms include momentum&adaptive learn rate backpropagation, genetic algorithm and simulated annealing algorithm.A improved algorithm based on BP algorithm is raised in this article. This algorithm can overcome the BP algorithm’s defects. The procedureof new algorithm is aslo given. This new algorithm can avoid local minimum and improve convergence speed.The improved momentum&adaptive learn rate backpropagation algorithm is used to the long term rainfall forecast in the area of Hongze lake. Simulative experiments show that the new algorithm has not only faster convergence speed but aslo higher precision.