Deep Learning-based Hourly Temperature Prediction

Abstract

Accurate prediction of temperature is an important part of fine weather forecast services (such as heating energy consumption in winter, Winter Olympic Games, etc.). Therefore, the accurate prediction of hourly temperature is very significant in the management of human health and the decision-making of government. In this study, a long short term (LSTM) memory model was proposed and used to predict the next hour's temperature in mega-cites in North China. It was fully considered for the historic temperature and meteorological condition. As a result, the predictor secured a fast and accurate prediction performance by fully reflecting the long-term historic process of input time series data through LSTM. The meteorological data from Beijing, Tianjin, Shijiazhuang and Taiyuan which represents the mega-cites of North China during October 1 to December 31 in 2016-2018 were used to verify the validity of the proposed method. In conclusion, the proposed method was proved to have a good prediction performance in cooling and turning warming processes, making up for the poor performance of turning weather prediction in the previous research. It confirmed that the forward supplement LSTM model has the best prediction ability for hourly temperature prediction in Beijing among mega-cites in North China. The results also indicate great potential of the machine learning method in improving local weather forecast and the potential to serve the 2022 Winter Olympics.