Typhoon Path Prediction Based On Recurrent Neural Network

After the typhoon landed,it was often accompanied by severe weather conditions such as strong winds and heavy rains,which brought serious life threats and property losses to the residents in the coastal areas.China has a long coastline and is located on the west coast of the Pacific Ocean where the typhoons are relatively active.It has long been affected by typhoon disasters.Therefore,it is important to predict the information of the typhoon path after the typhoon are formed.This is related to the people's transfer and disaster prevention in the affected areas.In recent years,Artificial neural network technology has been widely applied to various fields such as weather data forecasting,etc.Compared with traditional dynamic numerical forecasting technology and statistical dynamic forecasting technology,Artificial neural networks technology has the characteristics of strong nonlinearity,low consumption of computer resources and strong robustness.This paper considers the sequence of typhoon path data and applies recurrent neural network to typhoon path prediction.At last Improved recurrent neural network structure was found.Firstly,this paper discusses the calculation of the similarity of sequence data.By analyzing the characteristics and scope of the Euclidean distance,the cosine distance and the dynamic time warping distance,Finally the dynamic warping distance is selected,and a new similarity function is defined,This similarity can effectively selcet typhoons with similar motion characteristics to the target typhoon.So the target typhoon can get a better training set,which can improve the prediction accuracy.Then,this paper discusses the simple recurrent neural network and its two variant long-short term memory networks and the gated unit network model,and analyzes the characteristics and advantages of each structure one by one.The typhoon occurred between 1949 and 2015 collected by the tropical cyclone best path dataset were selected as the experimental data.The latitude and longitude information of the typhoon every 6 hours was used to predict the typhoon's future location with the simple recurrent neural network,long-short term memory network and the gated unit network.The results show that when typhoons' path approximate straight line,the prediction effects of the three models are similar.For typhoons with relatively large volatility such as path approximation curve type,the simple recurrent neural network has the worst prediction result.Considering the complexity and prediction effect,the gated recurrent unit model is optimal.Finally,an improved recurrent network is proposed in this paper.The network structure model has lower complexity and have good effect.