PM10 density forecast model using long short term memory

Jung-Hwan Park, Seong-Joon Yoo, Kyung-Joong Kim, Yeong Hyeon Gu, Keon-Hoon Lee, U.-Hyon Son

Abstract: This paper suggests a PM10 forecast model using Long Short Term Memory (LSTM). Data used for the study are collected from Seoul, Korea for the period of January 2005 up to March 2016. As the collected data has a lot of noise, the moving average technique is used to preprocess data for smoothing. Time series data of PM10 was converted into 30-day sequence data to use it as the input data for LSTM. LSTM learns through the sliding window process where sequence data moves to the space adjacent to it. The linear regression and recurrent neural network models are compared to evaluate the performance of LSTM. From the result, the suggested model showed a 500% improvement over linear regression and 100% over the recurrent neural network for its performance level.