Font Size: a A A

Application Of Long Short-term Memory Network In Short-term Rainfall

Posted on:2022-07-30Degree:MasterType:Thesis
Country:ChinaCandidate:J B WangFull Text:PDF
GTID:2510306539453334Subject:Applied Statistics
Abstract/Summary:PDF Full Text Request
With the improvement of the economic level,people’s requirements for weather forecasts are gradually increasing,and short-term rainfall may threaten the normal economic life of the people.In this case,short-term forecasting came into being.Short-term rain is a kind of shortterm forecast.However,most of the current short-term rain forecasts are based on numerical models,using the principles of synoptics,dynamic meteorology,and statistics.Rain conditions at a time.In recent years,the number of ground observation stations in my country has exceeded50,000,and the observational data has become increasingly abundant,from hour-level data in the past to minute-level data now.In this era of big data,effectively digging out hidden information from a large amount of data has become the development direction of the industry.In this context,deep learning technology is booming.It can learn the abstract characteristics and laws of the data,and thus the rainfall situation.Make predictions.Therefore,this article will use the data from the ground meteorological observation station and use the deep learning technology to predict the short-term rainfall situation.The main research contents are as follows:(1)Use the combination of convolutional neural network and long-short-term memory network to construct short-term and imminent rainfall prediction models,which can make full use of the convolutional network’s ability to capture local features and the long-short-term memory network’s ability to capture time series features.First of all,the original meteorological data is preprocessed,including standardization,removing outliers and missing values in the data,constructing long time series data that is not synchronized,and using random undersampling to balance the sample data.Secondly,model training is optimized on this data set,and the optimal step size corresponding to different moments and the optimal ratio of positive and negative samples in the training set are determined respectively.Finally,through the analysis of the time series data of meteorological elements,the forecast of short-term and impending rainfall in the next 1-3 hours is realized.(2)Use the long and short-term memory network to process and predict the same data and step size,and also use traditional machine learning algorithms such as K-nearest neighbors and support vector machines to compare with hybrid models.The results show that in the shortterm rainfall prediction,the classification effect of the neural network is significantly higher than that of the traditional machine learning algorithm,and in the 1-2 hours prediction,the performance ability of the hybrid model is slightly higher than that of the long-term short-term memory network.When it is small,the appearance ability of the two is equivalent.
Keywords/Search Tags:Weather forecast, Short-term and impending rainfall, Machine learning, convolutional neural network, Long and short-term memory neural network
PDF Full Text Request
Related items