Study on short-range precipitation forecasting method based on genetic algorithm neural network

In order to establish the forecasting model of the genetic-neural network, the genetic algorithm was used to optimize the connection weight and structure of the neural network through application of retaining the best individual in the genetic evolution process. This method overcomes the randomicity of the initial weight value, and, avoids the network oscillation as well as its being trapped into the local solution in the determination of the NN structure. Taking Guangxi short-range precipitation forecast as the example, this paper conducts the processing of dimensions cut on the massive forecast factors selected from the T213 numerical weather forecast products and achieves the effect of the concentration of the effective information, and establishing the forecasting model of daily rainfall grades during May-June in Guangxi by combining Japanese fine-mesh numerical weather forecast (NWF) predictor. The result indicates that the accurate rate of 24 hours forecast for the average precipitation equal to or even greater than 10mm within Guangxi with this model is 0.57, 0.50, 0.30, which is 7-15% over the present routine operational T213 and Japanese numerical weather forecast products.