Rainfall Forecasting Using Recurrent Neural Network and LSTM in Central Taiwan

Slope sliding often occurs in the mountainous areas of Taiwan, and it often causes casualties. One of the main causes of the damages is rainfall. Therefore, if it is possible to estimate the rainfall in the frequently collapsed place in a few hours before the slope collapse event may occur. The people can be evacuated or prohibited from entering the road or areas in advance to avoid the disaster caused by the collapse. The amount of accumulated rainfall and duration are correlated to many factors, including seasonality, extreme climate, typhoon path, terrain, etc. These factors may affect the accumulated rainfall and rainfall intensity. Therefore, this study used a large amount of past rainfall data to try to estimate rainfall and intensity. Under the recent rapid development of science and technology, artificial intelligence (Artificial Intelligence, AI) capabilities have been strengthened. Artificial intelligence technology has been introduced in various fields. Artificial intelligence has the ability to achieve deep learning. In the field of geotechnical engineering, the technology of artificial intelligence is beginning to be widely valued, such as related geodetic data analysis and image recognition. This research uses the existing research and application of artificial intelligence in the field of geotechnical engineering. Recurrent neural network (RNN) and long short-term memory (LSTM) models are used to predict the future rainfall trends of potential sliding slopes in mountainous areas of Taiwan. The hourly rainfall information obtained from the rainfall station far away from the potential sliding location were used to simulate and predict the rainfall duration and accumulated rainfall at the location. The predict rainfall information can be used to determine whether sliding damage may occur. It can also be used to calculate the evacuation time before the critical rainfall, and there is a buffer time before the collapse and damage may occur, and countermeasures and warning measures can be made.