Fog Forecast Experiment of Single Station Based on LVQ Neural Network

采集大连某机场2004-2007年大雾、轻雾和无雾天气事件共186例,选取雾天气事件前期(前一日08:00,14:00,20:00(北京时)实测资料)的温、压、湿、风等要素指标为预报因子,基于学习向量量化神经网络(learning vector quantization,LVQ),采用逐级预报思想建立起某机场雾天气事件的预报模型.在网络训练过程中,动态调整网络神经元比例参数,提高模型的预报能力;采用根据检验准确率适时终止训练的"先停止"技术,有效提高了模型的泛化能力.预报试验表明:无论是拟合率还是独立预报准确率,模型均已达到较高水准,具有实际应用意义.The generating and dissolving of fogs are too complex for empirical and linear systems methods to forecast and these methods cannot meet the needs of flight training.To meet this end,a new fog predicting model is proposed based on learning vector quantization neural network.The forecasting model of fog weather events is established using sequential forecast idea,adopting principal component analysis (PCA) and learning vector quantization network too.186 cases of heavy fog,mist or fog-free weather events on a certain airport is studied.Temperature,pressure,moisture,wind and other elements observed at 08:00,14:00,and 20:00 the day before the foggy weather are selected as prediction factors.Based on Learning Vector Quantization neural network,the prediction model of airport foggy weather events is established using sequential forecast idea (fog versus fog-free,heavy fog versus mist),and the prediction factors can be simplified using the principal component analysis. In the network training process,the model forecasting capability is improved in accordance with fitting accuracy to dynamically adjust neurons sealing parameters of the network.Adopting"to stop"technology of the timely termination training in accordance with testing the accuracy,generalization ability of the model is effectively improved.Forecasting experiments show that,the proposed model can effectively distinguish fog,mist and fog.Both the fitting rate and the forecasting accuracy are satisfactory so the model is practical.中文核心期刊要目总览(PKU)中国科技核心期刊(ISTIC)中国科学引文数据库(CSCD)01110-1142