Short-term air temperature prediction by adaptive neuro-fuzzy inference system (ANFIS) and long short-term memory (LSTM) network

Abstract

Air Temperature (AT) is a crucial parameter for many disciplines such as hydrology, irrigation, ecology and agriculture. In this respect, accurate AT prediction is required for applications related to agricultural operations, energy generation, traveling, human and recreational activities. In this study, four different machine learning approaches such as Adaptive Neuro-Fuzzy Inference System (ANFIS) with Fuzzy C-Means (FCM), ANFIS with Subtractive Clustering (SC) and ANFIS with Grid Partition (GP) and Long Short-Term Memory (LSTM) neural network were used to make one-hour ahead and one-day ahead short-term AT predictions. Concerning the test site, the measured AT data were obtained from a solar power plant installed in the city of Tarsus, Turkey. Correlation coefficient (R), Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) were used as quality metrics for prediction. Predicted values of the LSTM, ANFIS-FCM, ANFIS-SC and ANFIS-GP models were compared with the observed values by evaluating their prediction errors. According to the hourly AT prediction, the RMSE values in the testing process were found to be 0.644 (degrees C), 0.721 (degrees C), 0.722 (degrees C) and 0.830 (degrees C) for the LSTM, ANFIS-FCM, ANFIS-SC and ANFIS-GP models, respectively. On the other hand, the RMSE values of the corresponding methods for daily AT prediction were obtained as 1.360 (degrees C), 1.366 (degrees C), 1.405 (degrees C) and 1.905 (degrees C), respectively. The comparison of hourly and daily prediction results revealed that the LSTM neural network provided the highest accuracy results in both one-hour ahead and one-day ahead short-term AT predictions, and mainly presented higher performance than all ANFIS models.