PURPOSES : This aim of this study is to develop a model for predicting road surface temperature using an LSTM network to predict road surface temperature associated with road icing. METHODS : A long short-term memory (LSTM) neural network suitable for time-series data with time correlation is used in the analysis. Moreover, an optimal neural network architecture is designed via hyperparameter search and verification using learning and validation data. Finally, the generalization performance is evaluated based on the RMSE using unseen data as test data. RESULTS : The results show that the predicted data are similar to the actual road surface temperature patterns , and that the network appears to be generalized. CONCLUSIONS : The LSTM model improves the accuracy and generalization of road surface temperature prediction, as compared with other machine learning models.

ABSTRACT
1. 서론
2. 문헌고찰
3. 분석데이터
4. 장단기 기억(LSTM) 개념 및 모형 설계
    4.1. 장단기 기억(LSTM) 개념
    4.2. LSTM 모형 설계
5. 분석결과
    5.1. 모형 성능 결과
    5.2. 모형 일반성 성능 평가
6. 결론
REFERENCES

• 문재필(한국건설기술연구원 수석연구원) | Moon, Jae-pil Corresponding author