Temperature time series prediction based on autoregressive integrated moving average model

Abstract

This paper establishes a prediction model for land and ocean temperature time series based on the improved autoregressive integrated moving average (ARIMA) model. First, the temperature time series was normalized and differenced before passing the stationarity test by augmented Dickey-Fuller (ADF) method, while the model parameters were determined by the autocorrelation coefficient and the partial autocorrelation coefficient. After that, the model was trained by the historical temperature data series, and applied to predict the temperatures in future. To validate the model, several experiments were conducted using the average land and ocean temperature data of Lawrence Berkeley National Laboratory. The results of the ARIMAbased model were contrasted against those of the support vector regression (SVR) and the random forest (RF). The comparison shows that the ARIMA-based model was 10%~30% smaller than the SVR and the RF in the values of RMSE and MAE, and 1%~10% higher in the value of R 2 . This means our model outperformed the two benchmark algorithms. KEYWORDS: autoregressive integrated moving average (ARIMA) model, temperature prediction, time series analysis, difference, stationarity test. MOTS-CLÉS: modè le à moyenne mobile inté gré et autoré gressif (ARIMA), pré vision de la tempé rature, analyse des sé ries temporelles, diffé rence, test de stationnarité .