Prediction of outflow temperature of reservoir based on theory-guided machine learning models and optimization of operation for improving outflow temperature

Abstract

The proportion of hydropower within the energy-resource structure is gradually increasing. However, the construction of reservoirs inevitably leads to ecological and environmental issues, especially the release of low-temperature water in summer and its detrimental effects on fish spawning and reproduction. These issues can be alleviated by optimizing the reservoir operation scheme, but the large amount of time required for predicting the reservoir outflow temperature emerges as the primary constraint. In this study, we considered Pubugou (PBG) Reservoir and proposed an efficient and high-precision prediction method for reservoir outflow temperature based on theory-guided machine learning (TGML) algorithms. This method was applied to the multi-objective optimization of the reservoir operations, and an optimization operation plan was proposed to improve the outflow temperature. The research results indicate that the LightGBM machine learning model shows good predictive performance, with a maximum deviation of no more than 1 °C in predicting the outflow temperature. There is a significant competitive relationship between reservoir power generation and the outflow temperature. Compared with the current operational scheme, the maximum power operation plan results in a 4.4% increase in total power generation but causes the average outflow temperature during the fish spawning period to decrease by 0.1 °C. The total power generation of the improved outflow temperature operation scheme was 0.8% lower than that of the actual scheme, but the average outflow temperature during fish spawning increased by 0.2 °C. Properly increasing the discharge flow from the reservoir and lowering the water level can thus help alleviate the issue of outflow temperature.