Deriving Operating Rules of Hydropower Reservoirs Using Gaussian Process Regression

Abstract

Operating rules have been widely used to decide reservoir operations because they can help operators make an approximately optimal decision with limited runoff forecast information. As an effective alternative to explicit stochastic optimization (ESO) for considering hydrologic uncertainty, the implicit stochastic optimization (ISO) has been widely used to derive operating rules for the long-term operation of hydropower reservoirs. Within an ISO framework, operating rules extraction is a typical regression problem. In the past decades, various regression methods have been applied to derive operating rules, including artificial neural network (ANN), support vector regression (SVR) and so on, but these methods almost all are parametric regression method and there are few publications applying Bayesian regression method to derive operating rules. Therefore, Gaussian process regression (GPR), which is the representative Bayesian regression method, is introduced to derive operating rules for the first time in this paper and compared with ANN, SVR and conventional scheduling graph (CSG). China’s Three Gorges Reservoir (TGR) is selected as a case study, and four performance indexes are defined to evaluate different methods. The results show that (1) GPR, ANN and SVR can provide better performance than CSG method and are more practical than deterministic optimization operation; (2) GPR method can provide greater power generation benefits and higher reliability than ANN, SVR and CSG methods, and the average annual power generation increases from 88.481 billion kWh, 88.559 billion kWh and 87.563 billion kWh to 88.586 billion kWh, while the generation guarantee rate increase from 86.88%, 87.07% and 81.99% to 87.45%.