Refined machine learning modeling of reservoir discharge water temperature

Abstract

Water temperature is a controlling factor for physical, biological, and chemical processes in rivers, and is closely related to hydrological factors. The construction of reservoirs interferes with natural water temperature fluctuations. Hence constructing a model to accurately and efficiently predict the reservoir discharge water temperature (DWT) is helpful for the protection of river water ecology. Although there have been studies on constructing efficient and accurate machine learning prediction models for DWT, to our knowledge, there is currently no research focused on hourly scales. The study proposed in this paper is based on high-frequency monitoring data of vertical water temperature in front of a dam, water level, discharge flow, and DWT. In this study, six types of machine learning algorithms, namely, support vector regression, linear regression, k-nearest neighbor, random forest regressor, gradient boosting regression tree, and multilayer perceptron neural network, were used to construct a refined prediction model for DWT. The results indicated that the SVR model using the radial basis function as the kernel function had the best modeling performance. Based on the SVR model, we constructed a 1–24 h early warning model and optimized the scheduling of DWT based on changing discharge flow. In summary, a machine learning model for DWT that can provide short-term forecasting and decision support for reservoir managers was refined in this study.