Font Size: a A A

Research On Short Term Forecast Of Fog Based On Deep-Learning

Posted on:2020-01-13Degree:MasterType:Thesis
Country:ChinaCandidate:T T HanFull Text:PDF
GTID:2370330575465132Subject:Control engineering
Abstract/Summary:PDF Full Text Request
Fog is one of the main weather phenomena affecting visibility,which is closely related to human life.The accurate forecast of fog plays an extremely important role in people’s life and production.Since the selection process of meteorological factors is complicated and the features need to be designed manually,the traditional statistical method can not obtain a good generalization performance.In recent years,with the rapid development of deep learning technology and the powerful feature expression ability of the deep neural network model,the relevant features of the data can be to be extracted and learned automatically,the classification and prediction performance of the model are increased greatly.In this disseration,deep learning will be used to utilized for short-term forecasting research of fog,this is the first attempt of deep learning in fog forecasting.The main research effort and contributions on this dissertation are introduced as follows:(1)Convolutional neural network(CNN)is used to construct the short-term fog forecasting model.By classifying time series of meteorological elements,the short-term fog forecasting for the next 1-4 hours is realized.Firstly,the original meteorological element data are normalized and the default values and outliers are removed to construct the time series data of different lengths.In order to solve the problem of unbalanced samples in fog,a random under-sampling method is used to construct a balanced sample data set.By optimizing the network on the dataset,the optimal network structure of CNN for short-term fog forecasting model is constructed and the network model is verified by many experiments.(2)In order to further improve the accuracy of forecasting,the Long Short-Term Memory(LSTM)is employed to construct short-term fog forecasting model.At the same time,the synthetic Minority Oversampling Technology(SMOTE)algorithm is used to enhance the original meteorological element time series data.Dropout layer is added to optimization model to prevent the over-fitting during the training process and the optimal parameters of the LSTM model are determined by experiments.In further experiments,CNN-LSTM forecasting model is realized by combining convolutional neural network with LSTM network.The experimental results show the proposed LSTM model obtain better forecasting performance.Finally,PyQT5 is used to compile a software for short-term fog forecast.
Keywords/Search Tags:Short-term fog prediction, Data preprocessing, Convolutional neural network, Long and Short-Term memory network, CNN-LSTM forecast model, Unbalanced sample, SMOTE algorithm
PDF Full Text Request
Related items