Full Program »
Planning For Hydropower Plant Upgrades and Reinforcement Learning
Hydropower plants provide flexibility and storage to support the penetration of renewable energy sources needed to meet climate goals. Investments to upgrade their capacity dependon accurate valuation. Such strategic valuation in principle depends on long-term market price movements, tactical capacity allocation, and capacity bids that respond to short-term price fluctuations. Given the complexity of this holistic problem, hierarchical planning is commonplace, where investment models simplify tactical capacity allocation decisions and ignore the value ofshort-term production flexibility. We formulate a novel investment model that accounts for these aspects. While our problem is complex, we show how a combination of price modeling, informed by empirical analysis, and the use of reinforcement learning to solve for capacity allocation can lead to insightful semi-analytical investment policies. In particular, these policies highlight that capacity investment is supported at lower power prices when the short-term variability of these prices increases, that is, when the value of short-term production flexibility is higher. A numerical study based on real operational and market data shows that valuations from our model can be computed efficiently. Our findings suggest that investment models enabled by reinforcement learning that value the operational flexibility of production assets at long and short time scales can significantly help promote additional capacity in hydropower. The tools we develop are potentially relevant for analogous valuation of investments in other renewable energy production assets.