Application of Machine Learning in Daily Reservoir Inflow Prediction of the Bhumibol Dam, Thailand

Abstract

To mitigate floods and droughts in Thailand, the reservoir operations need accurate and reliable hydro-parameter information, e.g., inflow, to support decision making. In this paper, we explore and develop the predictive models for predicting the next-day inflow of the Bhumibol Dam, one of the major reservoirs of Thailand. We applied the machine learning techniques including decision tree, support vector regression, random forest, and extreme gradient boosting (XGBoost). Daily reservoir and climate Data from 2000 to 2021 were used in the analysis. After the series of experiments of model development, we finalize the model with the random forest algorithm having the best performance of MAE=4.232, MSE=83.823, and R2=0.867. However, we believe that the models and the feature sets can be further explored and developed to achieve the better accuracy. As a result, we could practically incorporate the inflow prediction model to aid decision making in the reservoir operation.