Baseline hydropower generation offer curves

Jon Pearce, Arash Khojaste, Golbon Zakeri, Geoffrey Pritchard · 2025

Read Baseline hydropower generation offer curves by Jon Pearce, Arash Khojaste, Golbon Zakeri, Geoffrey Pritchard (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-04854.txt.

Introduction

Jonathan Pearce and co-authors set out a pricing model for hydropower in a 2025 paper. Hydro plants store water and choose when to release it for power. Their offers to the power market shape both income and grid supply. Water inflows change with the seasons, so timing choices carry real risk.

The model treats the choice as a step-by-step decision process with random inflows. It draws on past inflow records to capture the seasonal pattern. From that, it builds offer curves that say how much power to sell at each price. The method is fast to run and easy to read. It aims to give plant operators a clear, grounded way to bid stored water.

Catalog record

CREATORJon Pearce, Arash Khojaste, Golbon Zakeri, Geoffrey Pritchard
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE21 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-04854

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Baseline hydropower generation offer curves Jon Pearce, Arash Khojaste, Golbon Zakeri, Geoffrey Pritchard Abstract We outline a mathematical model for pricing hydropower generation. The model involves a Markov decision process that reflects the seasonal variation in historical time series of water inflows. The procedure is computationally efficient and easy to interpret. Nomenclature Sets and their elements r\in R Inflow regimes (quantile sub-intervals) t\in T Time points f\in F Inflows f^{\prime} Successor of inflow f \ell\in L Water level in the reservoir s\in S System states, S=L\times R\times T r^{\prime}\in R Successor of regime r s^{\prime}\in S Successor of state s a\in A Actions, reservoir outflow Fourier regression models \vec{\phi}(t) Fourier basis functions \omega Frequency in trigonometric polynomials \vec{\beta},\vec{\gamma} Regression parameters Inflow Markov chain p_{r^{\prime}|rt} Probability of transition r\rightarrow r^{\prime} at time t p_{f|rt} Probability of inflow f in regime r at time t \sim Sampling from a distribution Operating Markov decision model p_{s^{\prime}|sa} Probability of transition s\rightarrow s^{\prime} under a c_{sa} Instantaneous cost of action a in state s y^{\ast}_{sa} state-action distribution of water release policy Pricing Markov model u Expected annual cost v^{\ast}_{s} Value to the system of state s I Introduction Many countries and jurisdictions around the world, including most of the United States, meet their electricity needs through organized electricity markets. In these markets, generators submit supply offers that reflect their marginal cost of production. Calculating this cost is relatively straightforward for traditional power sources, such as fossil fuel-fired generation units. For example, natural gas turbines can base their offers on known fuel contract prices and turbine power curves, which describe power output as a function of fuel input. Additionally, the costs of operations and maintenance can be estimated with reasonable accuracy at the time of dispatch. In contrast, estimating the marginal cost of electricity production from hydropower resources is substantially more complex. Offer stacks for hydro reservoirs must reflect the trade-off between immediate power production and the value of conserving water for future use (the so-called opportunity cost), especially given uncertainties in future inflows, fluctuating demand, and volatile electricity prices. Accurately estimating and incorporating this water value into the electricity market is critical for balancing the flexibility of thermal generation against the strategic use of limited hydro resources, ultimately enhancing both reliability and economic efficiency. A reasonable estimate of water values can also serve as a benchmark to assess whether a generator is acting competitively in the market or exerting market power. Furthermore, as renewable generation sources such as wind and solar achieve greater penetration, and as demand and transportation sectors become increasingly electrified, the overall system becomes more stochastic and volatile. This heightened uncertainty further complicates the task of valuing water accurately, making water value estimation increasingly important for efficient market participation and system planning. One possible way to evaluate water over a time horizon is to use stochastic dynamic programming (SDP). This technique explicitly models the sequential and uncertain nature of water inflows and operational decisions over time. By breaking the problem into stages and optimizing expected future profits at each stage, SDP helps identify water release and storage strategies that balance immediate generation with future value. Pereira and Pinto [ 1 ] introduced Stochastic Dual Dynamic Programming (SDDP) as a scalable method to value hydropower systems under uncertainty. Their approach uses dual information to build piecewise linear approximations of future cost functions, enabling efficient planning of water use across large, complex hydro systems with uncertain inflows. In order to utilize SDDP-like approaches, we need to construct a scenario tree to model inflows over the time horizon, which can prove an arduous task. Furthermore, convergence of SDDP can stall; for instance, gaps of nearly 22% are reported for an instance of the long-term planning problem of the Brazilian power system [ 2 ] . In what follows, we will describe a Markov decision process approach to obtaining water values for a reservoir. Our methodology uses historical inflow data over decades to build a Markov model of inflow following [ 3 ] . To account for the periodic (seasonal) nature of inflow, we fit quantile Fourier regressions that capture the stochastic dynamics of inflows over time. Specifically, we develop a time-inhomogeneous Markov model that predicts the inflow volume (more precisely, its corresponding quantiles) for each week of the year, conditioned on the inflow observed in the preceding week. We then utilize this model to find a plan of operations of a reservoir that minimizes the total average long-run cost of the system. Coupled with such an optimized plan is a set of water value curves that provide the value of water in a reservoir for each week of the year. Our main contributions in this study include: • Formulating a Markov decision process-based procedure for pricing hydropower generation, reflecting the seasonal variation in historical time series. The procedure is computationally efficient and provides a principled means of generating price scenarios over long-term investment planning horizons. • Implementing this methodology on a realistic example based on New Zealand, utilizing 74 years of historical inflow data from that country (1948–2021). II Related work Catalao et al. ( [ 4 ] , [ 5 ] ) propose a non-linear approach for scheduling releases in Portugal’s reservoir cascade system that accounts for head effects and the size of the dam and they further develop the model to account for stochastic prices. Garcia-Gonzalez et al. [ 6 ] and Sauhats et al. ( [ 7 ] ) develop a stochastic optimization procedure to plan the operation of three hydropower plants. Their proposed stochastic optimization algorithm is based on time-average revenue maximization and utilizes neural networks. Garcia-Gonzalez et al. ( [ 6 ] ) develop a mixed-integer linear program that takes account of a detailed representation of the generating units. They maximize the expected profit of the generator responding to stochastic prices that are modeled using hidden Markov models. Kleiven et al. ( [ 8 ] ) develop a price model that considers the joint dynamics of forward prices and inflows to assess the frequently used assumption of independence of inflows from reservoir levels. They show that ignoring this dependence leads to underestimation of water values by the producers. Ilak et al. [ 9 ] , [ 10 ] ) develop models for the self-scheduling problem of a hydro producer formulated and solved as a mixed integer linear programming problem. Their model is deterministic but accounts for head effects of the reservoir and considers participation in the ancillary services market. Daglish et al. ( [ 11 ] ) analyze the Brazilian forward market. The Brazilian electricity market is a hydro-dominated electricity market. They assessed generator entry before and after the forward market was introduced. Zhu et al. ( [ 12 ] ) use MDPs to manage a portfolio optimization for hydro power contracts within a two-stage model. The operational model (in the second stage) is the conventional SDDP model. The above papers either respond to electricity market prices or focus on contract markets. In contrast, our approach is to find the true value of water/reservoir storage to the electricity system. Perhaps the most relevant comparison to our work is the work of [ 13 ] . That paper develops a simplified

Read the complete text