Research On Fog Weather Forecast Based On Machine Learning Method

Heavy fog is a sudden common natural disaster weather,which leads to characteristics of low visibility when it occurs.It exerts a negative influence on people's travel.If we can forecast the fog weather in advance,then it would effectively avoid the financial loss and reduce the minimally risk.The state-of-the-art researches on fog mainly involve Radar Real-Time Monitoring of fog and Numerical Analysis methods,but both of them hinge severely on the forecast experience of experts.Also,their calculation steps may cost huge time complexity.With the development of artificial intelligence,the tricks of machine learning have been well applied to such Aerospace industry,System security as other fields of predictive tasks.For foggy research,,this approach mainly more focused on recognising the current state of fog weather than predicting the future situation of fog.As a consequent,Applying this to short-term fog research will be of great significance for early coping with foggy occurance.Our paper,with the help of the meteorological elements data of the China Integrated Meteorological Information Sharing System(CIMISS),is to use the machine learning method to analyse the future trend of heavy fog in Liaoning Province.According to the requirements of the Northeast Regional Meteorological Center,our partner,our short-term fog research contents include the following two aspects related to fog weather: short-term fog weather classification predictions and short-term fog visibility prediction.For the former,we solve the problem of how to predict the occurance of fog and analyse the main meteorological characteristics that effect the result of fog classification.Based on these issues,we find a method that solve the main feature analysis and design the structure of machine learning that combines Support Vector Machine(SVM)and feature select Wrapper method.Specifically,this combined method is a feature select method that relies on Machine Learning methods.It randomly combines different features of the data to obtain a subset of column attributes and inputs them to the SVM for training.Based on the training accuracy of each combined subset of data in the SVM,the optimal group of attributes is selected to achieve good prediction performances and feature analysis of short-term fog classification.For the later,The study of short-term fog visibility prediction primarily involves thelearning the relation between meteorological features and fog visibility,calculating the importance degree of features and ranking them..Based on above problem,we adapt distribute XGBoost model,Gradient Boosting Decision Tree(GBDT)algorithm which is based on boosting learning strategy of Ensemble Method,,to establish this relation and analyse these meteorological elements.Eventually,the ranking of meteorological elements participating in visibility prediction.is obtained in corresponding short-term time.Based on the research results of the abovementioned foggy aspects,a fog weather prediction system is developed by us,which is applied to the disaster weather forecasting system of the Northeast Meteorological Center to realize the prediction of fog weather.Its predictive capability is estimated using the meteorological standard called TS Score.The results show that this system can accurately predict the occurrence of fog and matches meteorologically forecast standard.