基于循环神经网络的西北太平洋台风路径预测

西北太平洋是全球台风高发区域之一，我国每年受登录台风影响，造成沿海地区发生狂风、暴雨和风暴潮等灾害性天气，准确预测西北太平洋台风路径对我国防震减灾意义重大。 基于循环神经网络的多元时间序列预测模型,以处理好的包含中国气象局（CMA）台风最佳路径数据集和欧洲中期天气预报中心（ERA5 ）数据集两种资料的多元时间序列数据为样本，训练3种循环神经网络并预测未来 6、12、24 h 的 热带气旋（TC）中心位置。结果表明预测失效为24 h时利用长短期记忆神经网络（LSTM）模型预测E值为173.15 km,误差要远小于门控循环神经网络（GRU）模型和循环神经网络（RNN）模型，认为利用 LSTM 网络对未来 24 h TC 中心位置的预测具有可行性和参考价值。

Northwest Pacific Typhoon Path Prediction Based on Recurrent Neural Network

The Northwest Pacific is one of the regions with high frequency of typhoons in the world. Every year, China is affected by landings of typhoons, which causes severe weather such as wind, rain and storm surge in coastal areas. It is of great significance to accurately predict the typhoon path in northwest Pacific region for disaster prevention and mitigation in China. The multi-component time series prediction model based on recurrent neural network takes the multi-component time series data including the best typhoon track dataset of China Meteorological Administration (CMA) and the European Centre for Medium Range Weather Forecasts (ERA5) as samples. Training three kinds of recurrent neural networks to predict the center location of tropical cyclones (TC) in the future 6, 12, and 24 hours, The results show that when the prediction failure is 24 hours, the E value predicted by long short term memory( LSTM )model is 173.15 km, and the error is much smaller than that of gated recurrent unit(GRU )model and recurrent neural network(RNN) model. It is considered that it is feasible and valuable to predict the location of TC center in the next 24 hours by using LSTM network.