Planning and management of hydropower systems is a complex task that involves solving many optimization problems and advanced statistical modeling. The nonlinear and stochastic nature of the problem is poses important challenges when developing efficient and robust decision-making tools for industrial applications.

On a short-term timescale, the optimal unit dispatching among all available units of the powerhouses must be computed in order to consider the real time physical parameters and operating constraints such as the load balance, electrical network constraints and so on. This optimization problem is highly nonlinear and requires a fast and robust solution to be implemented in a real industrial environment. The unit commitment problem has been addressed by several optimization methods such as successive linear programming, nonlinear programming and dynamic programming.

On a mid-term timescale, decisions regarding water releases and reservoir trajectories must be made in order to ensure safe and efficient operating policies with respect to flood control and load supply. This hydro-scheduling problem is an optimal control problem that aims to find an operating policy that computes the optimal amount of water releases at each reservoir given the state of the system and uncertainties of the inputs. These kinds of problems have been addressed in many applications by the stochastic dynamic programming algorithm. Recently, with the recent and increasing interest in machine learning, the reinforcement learning algorithm has become an efficient alternative to the stochastic dynamic programming algorithm. The performance of these solution methods is largely influenced by the quality of the inflow forecasts, which in itself poses particular difficulties as a modeling problem, especially for snow dominated systems.

Rio Tinto Aluminum owns and operates two hydropower systems in Canada with a total installed capacity of 4100MW. This presentation will focus on the Kemano system in British-Columbia, Canada. It houses eight generators, each capable of delivering 125 MW. It provides all the power for the company's Kitimat aluminium smelter located 75 km northwest or the generating station. The presentation will cover the historical and operational developments related to managing this system by covering all numerical methods and modeling tools developed during the last 8 years. Details on current optimization methods, modeling tools and future developments regarding the planning and management of the Kemano system will be presented.