Optimizing Hydropower Operations under Uncertainty Sue Nee Tan st542@cornell.edu February 17, 2015
Question: How do modeling parameters affect results from our model? Takeaway: There are tradeoffs between modeling something exactly and taking shortcuts. Question: How does uncertainty in the future affect our operations today? Takeaway: Depending on how we model uncertainty, we could be making decisions that are too conservative These tools help the stakeholder to define what those thresholds are and if they find these assumptions acceptable
3 Short-Term Hydropower Optimization and Uncertainty Analysis Laboratory (SHOAL) Sue Nee Tan, Jonathan R. Lamontagne, Jery R. Stedinger, Christine A. Shoemaker, and Steven B. Barton
4 A Learning Tool Short-term Hydropower Optimization and Uncertainty Analysis Laboratory (SHOAL) is a tool for experimenting with short-term optimization models with various time steps (allows sequences of time steps: 4, 8, 24 hrs) with appropriate flow routing and transitions. representations of turbines and powerhouses representations of uncertainty in meteorology and prices representations of operating constraints representations of power markets and economic objective
Columbia and Snake River Reservoirs 5
Flow Routing Total travel time from GCL to BON is 24.5 hours CHJ 1.5 hr GCL LWG to BON is 11 hours 17 hr 1 hr 1 hr 1 hr 2 hr 1 hr 3 hr IHR 2 hr LMN LGS LWG BON TDA JDA MCN 6
7 Optimal Operation SHOAL model gives optimal storages, powerhouse releases, and spill releases for each project in system (or subsystem) for each time step over the planning horizon considering uncertainty. Consider a 10-day planning horizon for the 10-project system in August 2012. Compare 3 models: 1. M8: 8-hr time steps 2. M24-1: 24-hr time steps w/ 1 PH release 3. M24-2: 24-hr time steps w/ 2 PH releases Consider the resulting system generation characteristics under the resulting optimal policy.
Net Energy Generation Sales (MW) Optimal Energy Sales 8 3000 2500 2000 1500 1000 500 0 M8 0 1 2 3 Days
Energy Net Generation Sales (MW) 9 Optimal Energy Sales 3000 2500 2000 The 1500 model can no longer peak during high price times 1000 500 0 M8 M24-1 0 1 2 3 Days
Energy Net Generation Sales (MW) 10 Optimal Energy Sales 3000 2500 2000 1500 1000 500 0 M8 M24-1 M24-2 0 1 2 3 Days
11 Computational Savings Model CPU time (s) System benefit M8 931 5.11 M24-1 155 4.01 M24-2 268 5.11 Same objective value as M8, but 1/3 the effort! Caveat: exact release decisions may deviate from M8 releases with 8 hour routing.
12 Optimizing Hydropower Operations with Wind Generation Uncertainty Sue Nee Tan & Christine Shoemaker
Our Goal 13 Develop a general framework for optimizing hydropower operations when there is stochastic wind generation in the system. Applicable to multi-reservoir systems Take advantage of the day-ahead market to hedge for wind uncertainty Computationally efficient for practical use Provides an optimal policy for many different conditions, even extreme events Model the sequential adaptive behavior of the system in response to outcomes of the random forcings
14 The Big Idea Use the power market to hedge for wind power production uncertainty: (1) Having made a day-ahead commitment, how much does the wind uncertainty affect the hour-to-hour operations? (2) How can we best adapt for wind generation uncertainty by buying or selling on the day-ahead market?
Day-ahead base price Wind forecasts affect day-ahead wholesale electricity prices We built a model for price as a function of day-ahead wind forecast 50.00 40.00 30.00 20.00 10.00 15 0.00 No wind Low = 4% of windmedium = 33% of capacity wind capacity Wind scenario High = 69% of wind capacity off-peak DA price on-peak DA price
Consider 4 different wind forecast scenarios: We run 4 different models for a 7-day horizon, and look at how the day-ahead commitment for the first day changes Scenario DA energy price Wind generation across 7-day horizon Wind generation within the day Computation time rank (1 = fastest, 4 = slowest) No wind Baseline None None 1 Fixed wind for horizon Fixed wind with withinday deviations Markov wind with withinday deviations Lower than baseline price Lower than baseline price Lower than baseline price, varies based on the wind scenario Fixed None 2 Fixed 4 different possible transitions, each with its own probability 10 different deviation scenarios 10 different deviation scenarios 3 4
Stage 1 Policies for dealing with different representations of uncertainty when NOT marketing wind 17 storage = 4 storage = 1 markov wind, with within-day deviations fixed wind with within-day deviations fixed wind for horizon no wind Day-ahead commitment
Question: How do modeling parameters affect results from our model? Takeaway: There are tradeoffs between modeling something exactly and taking shortcuts. Question: How does uncertainty in the future affect our operations today? Takeaway: Depending on how we model uncertainty, we could be making decisions that are too conservative These tools help the stakeholder to define what those thresholds are and if they find these assumptions acceptable