Developing a stochastic optimization model for operating the Manitoba Hydro multi-reservoir hydroelectric power system

• Main Author: Snell, Jacob
• Other Authors: Asadzadeh, Masoud (Civil Engineering), Stadnyk, Tricia (Civil Engineering) Cote, Pascal (Rio Tinto Alcan)
• Format: Master Thesis
• Language: English
• Published: 2021
• Subjects: Hydropower; Water resource management; Stochastic optimization; Churchill; Nelson River
• Online Access: http://hdl.handle.net/1993/35458

The province of Manitoba generates more than 90% of its electric power from hydroelectric generating stations located in the Nelson-Churchill Rivers basins. Prudent management of the major reservoirs in the system is essential for providing value through economic and reliable electricity. Reservoir managers are challenged by the variability of reservoir inflows, the misalignment of electrical energy demands and seasonality of reservoir inflows, and travel time lags between reservoirs and major generating stations on the Lower Nelson River. This thesis examines some of the challenges of current hydroelectric system management and applies a Sampling Stochastic Dynamic Programming algorithm to the operation of the Manitoba Hydro electric system. Direct consideration of variability and uncertainty of inflows are incorporated in the algorithm by generating a water value function based policy that considers multiple inflow scenarios and inflow scenario transition probabilities derived from conditional probability distributions based on a regression relationship between sequential periods of system inflow. The travel time lag is incorporated into the algorithm directly as a lagged inflow state variable to bridge reservoir release decisions between time periods. Storage values and penalties are incorporated to reflect operating license requirements and to prevent depletion of reserve storage that can lead to infeasibilities in the algorithm. The water value function policy is simulated over 38 historical years of inflow scenarios and compared against historical reservoir operation decisions. Model results show significant improvement in economic values and reductions in energy deficits over historical scenarios but are highly sensitive to the calibration of storage benefit and penalty values. Extensions to the model to use hydrological based inflow models, improving the transition matrix by evaluating alternative hydrological variables, and alternative approaches to the storage benefit and penalty method are discussed. ...