Study on typhoon quantitative rainfall prediction based on big data Hadoop Spark parallel framework calculus

107

About 80 typhoons are generated every year in the world, and the typhoon generated in the northwestern Pacific Ocean is the strongest and strongest. Taiwan is happening on the main path of the typhoon in the northwestern Pacific. The typhoon brings abundant rainwater to fill the reservoir, and it also causes loss of life. Such as the reduction of agricultural production, the closure of industrial and commercial activities, flooding in some areas, and the landslides and landslides. The purpose of this study is to predict the typhoon precipitation forecast by predicting the typhoon precipitation forecast and warning the people of all walks of life in Taiwan to prevent rainstorms and assess whether they are on holiday, thereby reducing the loss of people and the economy. This study used four stations in northern Taiwan to establish a typhoon precipitation prediction model to predict the amount of precipitation due to typhoons. This study uses Deep Neural Network (DNN) to establish precipitation prediction model and combines the big data technology Hadoop Spark decentralized framework acceleration mode to establish and calculate speed, and design four different cases according to different data volume (Case0) – Case3) to compare the accuracy of precipitation predictions. Sources are meteorological data from the Central Meteorological Administration from 1961 to 2017 on typhoon warning and ground stations. This study screens relevant attribute data based on individual cases and station design. According to different cases, the data is divided into three sets of data sets, training data sets, verification data sets, and test data sets. The training data sets are used in each case to establish a deep neural network model, and the verification data set is used to find the depth. The best hyperparameters of the Neural Network (DNN), the test data set is treated as a set of independent data sets to test whether the mode can be used in actual situations. This study divides the method into two phases. The first phase uses the Deep Neural Network (DNN) to establish a precipitation prediction model, which records the total time used in the process of establishing the mode, while the second phase will mode. Put into the big data (Big Data) technology Hadoop Spark to establish a parallel calculus framework, and record the total time used for parallel calculus and its prediction accuracy. The research results show that the position of the target station in the first stage will affect the prediction of typhoon precipitation and increase the error of the predicted value. The data of the station used in the establishment mode will affect the accuracy of the model and the amount of data will affect the stability of the model. The second phase uses big data technology Hadoop Spark decentralized framework combined with deep learning (Deep Learning) to establish a mode to establish a typhoon precipitation prediction model with similar accuracy in a shorter time. The Deep Neural Network (DNN) does outperform Multiple Linear Regression (MLR) in most cases. The Big Data technology Hadoop Spark parallel calculus framework can help more complex depths. The Deep Neural Network (DNN) saves mode setup time.