Abstract:
The interpretation and application of numerical forecast products is the direction of the modernization of the weather forecast and the key to improving the level of weather forecast.Using principle component analysis to choose the precipitation predictors in data of T213 Model.The predictors are used to daily categorical forecast of precipitation in Quzhou by Extreme Learning Machine(ELM).The results show that the method of principle component analysis can simplify the matrix of predictors and get new predictors.The new predictors which are mutually orthogonal contain most of the information of original predictors.ELM neural network has a valid prediction ability to training sample,and the TS score of the rainfall forecast for testing sample is improved from 0.3 to 0.8.Particularly for a heavy rain,ELM neural network shows strong prediction ability.To contrast with the method of Back Propagation (BP),ELM has more simple network setting and more fast computation speed.The prediction ability to training sample and testing sample of the ELM neural network are also better than BP neural network.