Stochastic Optimization in Energy Systems DIMACS Workshop on Algorithmic Decision Theory Rutgers University October 27, 2010 Warren Powell CASTLE Laboratory Princeton University http://www.castlelab.princeton.edu © 2010 Warren B. Powell, Princeton University Slide 1 Lecture outline The problem of uncertainty Modeling stochastic optimization problems Energy storage portfolios The unit commitment problem for PJM Long-term energy resource planning © 2010 Warren B. Powell Slide 2 Challenges in models and algorithms Strategic questions in energy policy and economics » How do we design market mechanisms to control green house gases? » How do we design policies to achieve energy goals (e.g. 30% renewables by 2030) with a given probability? » How does the imposition of a carbon tax or investment tax credit change the likelihood of meeting this goal? » We need models of competitive equilibria and cooperative games in the presence of uncertainties about technology, policy and climate change. » How do we quantify the risks associated with the pairing of energy from wind with dispatchable energy from hydroelectric facilities in the presence of climate change? Challenges in models and algorithms Design and control of energy systems » How do we balance storage of energy from the grid in electric vehicles in the presence of variability of energy from wind, volatile prices, and uncertainty in usage patterns for vehicles? » How do we control the storage of energy across a portfolio of energy storage devices? » How should a utility work with urban building managers to balance energy consumption from the grid, storage devices and backup generators? » How can grid operators modify their unit commitment models to capture significant contributions of energy from wind and solar? » How can companies design efficient verification policies for forests being used as carbon offsets? Challenges in models and algorithms All of these problems can be formulated as some type of stochastic optimization problem. in st ra a c e For For dy .3 e ga m dule Sche Cancel game $2300 -$200 6 y. nn we ath er lou -$1400 -$200 su Use st c me le ga u d e Sch Canc el g am e .1 ast rep ort eca rec Fo Simple problems can be solved as decision trees… Most problems are hard. se rt t u po no er re Do eath w e gam dule e h c S Cancel game $3500 -$200 e gam dule Sche Cancel game $2400 -$200 Decision making under uncertainty Mixing optimization and uncertainty Energy sources Time ? ? ? ? ? “large optimization model (e.g. NEMS, MARKAL, …)” Decision making under uncertainty Mixing optimization and uncertainty Energy sources Time 12.334 Solar No 6.2 142 Scenario 1 Decision making under uncertainty Mixing optimization and uncertainty Energy sources Time 18.917 Wind No 3.6 89.1 Scenario 2 Decision making under uncertainty Mixing optimization and uncertainty Energy sources Time 22.314 Solar Yes 5.9 117 Scenario 3 Decision making under uncertainty 22.314 Solar Yes 18.917 Wind 5.9 No 12.334 Solar 89.1 3.6 No 142 6.2 117 Scenario 3 Scenario 2 Scenario 1 Now we have to combine the results of these three optimizations to make decisions. Decision making under uncertainty Don’t gamble; take all your savings and buy some good stock and hold it till it goes up, then sell it. If it don’t go up, don’t buy it. Will Rogers It is not enough to mix “optimization” (intelligent decision making) and uncertainty. You have to be sure that each decision has access to the information available at the time. Decision making under uncertainty Socrates – used by Pacific, Gas and Electric Wetter Known 15% exceedence Unknown 50% exceedence 0.3 85% exceedence 0.4 probability (weight) 15% exceedence 0.3 50% exceedence Now Jan Average over 30 Flow forecast scenarios Feb Mar Drier Apr 85% exceedence May… Sep… Stochastic programming does not “cheat”, but it does not scale. It is best designed for coarse-grained sources of uncertainty, but will not handle finegrained temporal resolution. Jan Lecture outline The problem of uncertainty Modeling stochastic optimization problems Energy storage portfolios The unit commitment problem for PJM Long-term energy resource planning © 2010 Warren B. Powell Slide 13 Modeling stochastic optimization problems Attribute vectors: a Commodity type Type Location Location Age Location Dam type Water level Generators Slide 14 Modeling stochastic optimization problems Modeling resources: » The attributes of a single resource: a The attributes of a single resource a A The attribute space » The resource state vector: Rta The number of resources with attribute a Rt Rta aA The resource state vector » The information process: Rˆta The change in the number of resources with attribute a. Slide 15 Modeling stochastic optimization problems The system state: St Rt , Dt , t System state, where: Rt Resource state (how much capacity, reserves) Dt Market demands t "system parameters" State of the technology (costs, performance) Climate, weather (temperature, rainfall, wind) Government policies (tax rebates on solar panels) Market prices (oil, coal) Slide 16 Modeling stochastic optimization problems Making decisions: xt A decision to buy, sell, move, repair, price, or control X ( St ) Decision function (or "policy"); maps states to decisions (also known as actions or controls). We generally have a family of policies/functions so we write Slide 17 Modeling stochastic optimization problems Exogenous information: Wt New information = Rˆt , Dˆ t , ˆt Rˆt Exogenous changes in capacity, reserves Dˆ t New demands for energy from each source ˆt Exogenous changes in parameters. Slide 18 Modeling stochastic optimization problems The transition function St 1 S (St , xt ,Wt 1 ) M Known as the: “Transition function” “Transfer function” “State transition model” “System model” “Plant model” “Model” Slide 19 Modeling stochastic optimization problems Resources Demands Slide 20 Modeling stochastic optimization problems t t+1 t+2 Slide 21 Modeling stochastic optimization problems Optimizing over time t t+1 t+2 Optimizing at a point in time Slide 22 Modeling stochastic optimization problems The objective function t max E C St , X ( St ) t Expectation over allState variable Contribution function Decision function (policy) Finding the best policy random outcomes Modeling stochastic optimization problems Policies » 1) Myopic policies • Take the action that maximizes contribution (or minimizes cost) for just the current time period: X M (St ) argmax xt C(St , xt ) » 2) Lookahead policies • Plan over the next T periods, but implement only the action it tells you to do now. X ( St ) arg max xt , xt 1 ,..., xt T M T C(S , x ) t ' t t' t' Modeling stochastic optimization problems Policies (cont’d) » 3) Policies based on value function approximations Let Vt ( St ) be an approximation of the value of being in state St X M ( St ) arg max xt C ( St , xt ) EVt 1 ( St 1 ) » 4) Policy function approximations Let X ( St ) be a function that directly tells you an action given that you are in a state St . Modeling stochastic optimization problems Ways of approximating functions (policies or value functions) » 1) Lookup tables • When in a (discrete) state, returns an action • When in a (discrete) state, returns the value of being in that state. • There is one value (parameter) to determine for each state. » 2) Parametric models • A closed form, analytic function determined by one or more parameters. 1 (sell) If St =pt M X ( St | ) 0 (hold) Otherwise Vt ( St ) 0 1St 2 ( St ) 2 Modeling stochastic optimization problems Ways of approximating functions (policies or value functions) » 3) Nonparametric models n V n ( s) i i ˆ v k ( s , S h ) i 1 n k ( s, S ) i i 1 h i 2 ( s S ) kh ( s, S i ) exp h Nonparametric methods approximate a function by using a weighted sum of observations, where the weight declines with the distance between the observed state and the state we are trying to estimate. Lecture outline The problem of uncertainty Modeling stochastic optimization problems Energy storage portfolios The unit commitment problem for PJM Long-term energy resource planning © 2010 Warren B. Powell Slide 28 Energy storage portfolios Premise » The optimal mix of energy resources depends on the entire portfolio. » The performance of any single source of energy (including storage) depends on the marginal difference between demand and all other forms of energy in the portfolio. This is what determines the residual demand. » It is not just what is happening (the physical process) but also how well you could predict what was going to happen (the information process). © 2010 Warren B. Powell Slide 29 Energy storage portfolios Wind » Varies with multiple freqeuencies (seconds, hours, days, seasonal). » Spatially uneven, generally not aligned with population centers. Solar » Shines primarily during the day (when it is needed), but not entirely reliably. » Strongest in south/southwest. © 2010 Warren B. Powell Slide 30 Energy storage portfolios 30 days 1 year © 2010 Warren B. Powell Slide 31 Energy storage portfolios Hydroelectric Flywheels Ultracapacitors Batteries © 2010 Warren B. Powell Slide 32 Energy storage portfolios Storage technologies Name Charge rate Discharge rate Power rating Capacity Loss Efficiency Batteries sodium-sulfur 5 hours 5 hours 10 MW 20 Wh/kg < lithium 80-90% lead-acid 10 hours 10 mins.-hour 100 kW 20 Wh/kg 10%/month 75% lithium 1 hour 1 hour 10-100 kW 150 Wh/kg10%/month 90-95% Ultracapacitor seconds few seconds 10-100 kW 5 Wh/kg Flow battery Pumped hydro storage (PHS) 5 hours 5 hours hours Flywheel SMES (supercond. mag. ) 5%/day 95-100% 200 kW-1MW minimal 70-78%? 10 hours 1 GW evaporation 75%-80% minutes minutes 10kW-10MW 10 Wh/kg frictional 95% seconds seconds 10-100 MW .2 Wh/kg large 95% above ground hours 2 hours 15 MW minimal 48% below ground hours 10 hours 100-300MW minimal 48% Comp. air energy storage (CAES) » Storage technologies different in terms of: • • • • • Charge rates Discharge rates Power rating Capacity Loss © 2010 Warren B. Powell Slide 33 Optimal control of wind and storage Controlling the storage process » Imagine that we would like to use storage to reduce demand when electricity prices are high. » We use a simple policy controlled by two parameters. Price Withdraw Store Energy storage portfolios We now have to optimize over policies » This means finding the best values for withdraw , store » The policy refers to the structure of the decision function, and the parameters withdraw , store » This means we have to solve T max E tC ( St , X ( St )) t 0 » We can write T F ( withdraw , store | ) tC ( St ( ), X ( St ())) t 0 © 2010 Warren B. Powell Slide 35 Optimizing storage policy Store Withdraw 36 Optimizing storage policy Initially we think the concentration is the same everywhere: Estimated contribution Knowledge gradient » We want to measure the value where the knowledge gradient is the highest. This is the measurement that teaches us the most. Optimizing storage policy After four measurements: Estimated contribution Measurement Knowledge gradient Value of another measurement New optimum at same location. » Whenever we measure at a point, the value of another measurement at the same point goes down. The knowledge gradient guides us to measuring areas of high uncertainty. Optimizing storage policy After five measurements: Estimated contribution Knowledge gradient Optimizing storage policy After six samples Estimated contribution Knowledge gradient Optimizing storage policy After seven samples Estimated contribution Knowledge gradient Optimizing storage policy After eight samples Estimated contribution Knowledge gradient Optimizing storage policy After nine samples Estimated contribution Knowledge gradient Optimizing storage policy After ten samples Estimated contribution Knowledge gradient Optimizing storage policy After 10 measurements, our estimate of the surface: Estimated contribution True concentration Optimizing storage policy After 10 measurements, our estimate of the surface: Estimated contribution Knowledge gradient Knowledge Gradient Estimation of Annual Profit 1.26 0 -20 1.24 -40 -60 log(KG) Dollars x 107 1.22 1.2 1.18 -80 -100 -120 -140 1.16 -160 1.14 -180 60 60 60 60 62 64 65 55 55 66 68 70 70 72 50 theta0 50 74 75 76 80 45 78 theta1 theta0 80 45 theta1 Lecture outline The problem of uncertainty Modeling stochastic optimization problems Energy storage portfolios The unit commitment problem for PJM Long-term energy resource planning © 2010 Warren B. Powell Slide 47 The unit commitment problem Designing energy portfolios…. » … is like building a stone wall, where energy comes from sources which vary in terms of capital cost, operating cost and flexibility. © 2010 Warren B. Powell Slide 48 The unit commitment problem The day-ahead problem » Determines which coal/nuclear/natural gas/…plants to turn on/off and when. » Will use wind/solar when available and if needed. » Requires point forecast of wind/solar/demand. 24 min x1 ,..., x24 C( x ) t t 1 ….subject to numerous constraints, including integrality. » The problem is solved using two adjustment parameters • Fraction of generator capacity assumed (e.g. 93 percent) • q Quantile of wind forecast assumed for advance commitments. The unit commitment problem The hour-ahead problem » Each hour, we can make modest adjustments. Plants that are “on” can be adjusted up and down. » Cannot turn coal plants on and off. » Actual may differ from forecast: • Wind may be higher or lower than forecast. If higher, may not be able to use it because of inability to scale back other sources. • Demand may exceed forecast. Wind/solar may fall below forecast. In this case, we find the least cost unit that can be scaled up quickly enough. The unit commitment problem Modeling » A deterministic model 24 min x1 ,..., x24 C( x ) t 1 t » Stochastic formulation – I S min s s x1 ,..., x24 24 s s p C ( x t ) s "scenario" s 1 t 1 • Ambiguous whether decisions are deterministic or stochastic. • Day ahead decisions are deterministic, but hour ahead decisions are stochastic. The unit commitment problem Modeling » Stochastic formulation – II 24 min E C ( xt ,t ' , yt ',t ' ) x t ,t ' xt ,t ' t '1,...,24 y t ',t ' t '1 ( yt ',t ' )t ' 1,...,24 • xt ,t ' is determined at time t, to be implemented at time t’ • y t ',t ' is determined at time t’, to be implemented at time t’+1 » Important to recognize information content • At time t, xt ,t ' is deterministic. • At time t, y t ',t ' is stochastic. The unit commitment problem The unit commitment problem » Rolling forward with perfect forecast of actual wind, demand, … hour 0-24 hour 25-48 hour 49-72 The unit commitment problem The unit commitment problem » Rolling forward with perfect forecast of actual wind, demand, … hour 0-24 hour 25-48 hour 49-72 The unit commitment problem The unit commitment problem » Using forecast of future wind. hour 0-24 hour 25-48 hour 49-72 The unit commitment problem The unit commitment problem » Using forecast of future wind. hour 0-24 hour 25-48 hour 49-72 The unit commitment problem The unit commitment problem » Stepping through the next day. hour 0-24 xt ,t ' t ' The unit commitment problem The unit commitment problem » Stepping forward observing actual wind, making small adjustments hour 0-24 yt ',t ' The unit commitment problem The unit commitment problem » Stepping forward observing actual wind, making small adjustments hour 0-24 The unit commitment problem The unit commitment problem » Stepping forward observing actual wind, making small adjustments hour 0-24 The unit commitment problem The unit commitment problem » Stepping forward observing actual wind, making small adjustments hour 0-24 The unit commitment problem The unit commitment problem » Stepping forward observing actual wind, making small adjustments hour 0-24 The unit commitment problem The unit commitment problem » Stepping forward observing actual wind, making small adjustments hour 0-24 The unit commitment problem The unit commitment problem » Stepping forward observing actual wind, making small adjustments hour 0-24 The unit commitment problem Papers can be divided into four categories with respect to the handling of information: » The model is wrong, and the paper is implementing an incorrect algorithm. • “Incorrect” means cheating by using information from the future. » The model is wrong (or imprecise), but the experimental work is correct. • Many people know how to do a proper simulation, but don’t know how to model it. » The model is correct, but the experimental work is wrong. • The programmer cheated by using information from the future; the modeler used proper mathematics that did not match the code. » The model is correct, and the experimental work is correct. The unit commitment problem Safest way to write out the objective function t max E C St , X ( St ) t » Our policy combines: • Lookahead policy to solve unit commitment problem • Myopic policy to solve hourly adjustment problem. Myopic policy exploits tunable parameters (p,q) in the lookahead policy. • We can write the dependence on (p,q) using F S 0 | ( p , q ) t C S t , X S t | ( p, q ) t The unit commitment problem The value of optimizing (p,q) The unit commitment problem Energy profile with 2 percent from wind Feb 15 Feb 16 Feb 17 Feb 18 Feb 19 Day-ahead generation 5000 4000 3000 2000 1000 Real-time adjusted generation 0 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 5000 4000 3000 2000 1000 0 © 2010 Warren B. Powell Slide 68 The unit commitment problem Forecast from unit commitment model Feb 18 Day-ahead Feb 18 Real-time 5000 Real-time adjusted generation Day-ahead scheduled generation 5000 4000 3000 2000 1000 1 1 6 12 Hour 18 24 4000 3000 2000 1000 1 1 © 2010 Warren B. Powell 6 12 Hour 18 24 Slide 69 The unit commitment problem Energy profile with 20 percent from wind Jan 8 Jan 9 Jan 10 Jan 11 Jan 12 Day-ahead generation 5000 4000 3000 2000 1000 Real-time adjusted generation 0 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 1 6 12 18 Hour 24 5000 4000 3000 2000 1000 0 © 2010 Warren B. Powell Slide 70 The unit commitment problem The effect of modeling uncertainty in wind The unit commitment problem With better wind forecasts 2009 Wind Data - Forecast using 2009 past Windcorresponding Data hour of past 7 days Wind Generation MW 3000 2009 Wind Data 0.3 quantile of corresponding hour from past 7 days 2500 2000 1500 1000 500 0 -500 1000 2000 3000 4000 5000 Hour 6000 7000 8000 2009 Wind Data - Forecast using past 7 hours 3000 Wind Generation MW 2009 Wind Data 0.3 quantile of previous 7 hours 2500 2000 1500 1000 500 0 -500 1000 2000 3000 4000 5000 Hour © 2010 Warren B. Powell 6000 7000 8000 Slide 72 The unit commitment problem With better wind forecasts: ($ millions) Total Day-ahead system cost Total Real-time system cost Total Generator dispatch cost Total Generator turndown savings 20% Wind – forecast with past 7 days at corresponding hour Winter 69.240 79.584 11.109 0.766 Summer 83.886 90.222 7.606 1.269 20% Wind – forecast with past 7 hour Winter 61.389 68.764 7.730 0.354 Summer 78.299 85.000 7.608 0.907 » Using a stochastic model that properly captures the flow of information, we can quantify the value of better forecasts. © 2010 Warren B. Powell Slide 73 Lecture outline The problem of uncertainty Modeling stochastic optimization problems Energy storage portfolios The unit commitment problem for PJM Long-term energy resource planning © 2010 Warren B. Powell Slide 74 Goals for an energy policy model Potential questions » Policy questions • How do we design policies to achieve energy goals (e.g. 20% renewables by 2015) with a given probability? • How does the imposition of a carbon tax change the likelihood of meeting this goal? • What might happen if ethanol subsidies are reduced or eliminated? • What is the impact of a breakthrough in batteries? » Energy economics • What is the best mix of energy generation technologies? • How is the economic value of wind affected by the presence of storage? • What is the best mix of storage technologies? • How would climate change impact our ability to use hydroelectric reservoirs as a regulating source of power? © 2010 Warren B. Powell Slide 75 Intermittent energy sources Wind speed Solar energy © 2010 Warren B. Powell Slide 76 Long term uncertainties…. Tax policy 2010 2015 Solar panels Batteries Price of oil 2020 2025 Carbon capture and sequestration © 2010 Warren B. Powell 2030 Climate change Slide 77 SMART-Stochastic, multiscale model SMART: A Stochastic, Multiscale Allocation model for energy Resources, Technology and policy » Stochastic – able to handle different types of uncertainty: • Fine-grained – Daily fluctuations in wind, solar, demand, prices, … • Coarse-grained – Major climate variations, new government policies, technology breakthroughs » Multiscale – able to handle different levels of detail: • Time scales – Hourly to yearly • Spatial scales – Aggregate to fine-grained disaggregate • Activities – Different types of demand patterns » Decisions • Hourly dispatch decisions • Yearly investment decisions • Takes as input parameters characterizing government policies, performance of technologies, assumptions about climate © 2010 Warren B. Powell Slide 78 The annual investment problem 2008 New information 2009 New information oil oil oil ˆ oil ˆ oil oil oil ˆ ˆ ˆ ˆ R x R Dt t Rt 1 xt 1 Rt 1Dt 1 t 1 oil t oil oil t t windˆ wind wind wind windˆ windˆ wind wind ˆ Rtwindxtwind Rt Dt ˆt Rt 1 xt 1Rt 1 Dt 1 ˆt 1 R x Rˆ Dˆ ˆ R x Rˆ Dˆ ˆ nat gasnat gas nat gasnat gasnat gas nat gasnat gas nat gasnat gasnat gas t t t t 1 t 1 t t t t 1 t 1 R x Rˆ Dˆ ˆ coal coal coal coal coal t t t t t R x Rˆ Dˆ ˆ coal coal coal coal coal t 1 t 1 t 1 t 1 t 1 © 2010 Warren B. Powell Slide 79 The hourly dispatch problem Hourly electricity “dispatch” problem © 2010 Warren B. Powell Slide 80 The hourly dispatch problem Hourly model » Decisions at time t impact t+1 through the amount of water held in the reservoir. Hour t Hour t+1 © 2010 Warren B. Powell Slide 81 The hourly dispatch problem Hourly model » Decisions at time t impact t+1 through the amount of water held in the reservoir. Hour t Value of holding water in the reservoir for future time periods. © 2010 Warren B. Powell Slide 82 The hourly dispatch problem © 2010 Warren B. Powell Slide 83 The hourly dispatch problem Hour 2008 1 2 3 4 © 2010 Warren B. Powell 8760 2009 1 2 Slide 84 The hourly dispatch problem Hour 2008 1 2 3 4 © 2010 Warren B. Powell 8760 2009 1 2 Slide 85 SMART-Stochastic, multiscale model 2008 2009 © 2010 Warren B. Powell Slide 86 SMART-Stochastic, multiscale model 2008 2009 oil 2009 wind 2009 nat gas 2009 coal 2009 © 2010 Warren B. Powell Slide 87 SMART-Stochastic, multiscale model 2008 2009 2010 © 2010 Warren B. Powell 2011 2038 Slide 88 SMART-Stochastic, multiscale model 2008 2009 2010 © 2010 Warren B. Powell 2011 2038 Slide 89 SMART-Stochastic, multiscale model 2008 2010 2009 ~5 seconds ~5 seconds ~5 seconds © 2010 Warren B. Powell 2011 ~5 seconds 2038 ~5 seconds Slide 90 Approximate dynamic programming Step 1: Start with a pre-decision state Stn Step 2: Solve the deterministic optimization using Deterministic an approximate value function: optimization n n n 1 M ,x n vˆt max x Ct ( St , xt ) Vt (S ( St , xt )) to obtain xtn. Step 3: Update the value function approximation Vt n1 (Stx,1n ) (1 n1 )Vt n11 (Stx,1n ) n1vˆtn Recursive statistics Step 4: Obtain Monte Carlo sample of Wt (n ) and Simulation compute the next pre-decision state: Stn1 S M (Stn , xtn ,Wt 1 ( n )) Step 5: Return to step 1. © 2010 Warren B. Powell Slide 91 SMART-Stochastic, multiscale model Use statistical methods to learn the value of resources in the future. Resources may be: Vt ( Rt ) » Stored energy • Hydro • Flywheel energy • … » Storage capacity Value • Batteries • Flywheels • Compressed air » Energy transmission capacity • Transmission lines • Gas lines • Shipping capacity » Energy production sources Amount of resource • Wind mills • Solar panels • Nuclear power plants © 2010 Warren B. Powell Slide 92 Benchmarking on hourly dispatch ADP objective function relative to optimal LP 2.50 Percentage error from optimal 2.50% 2.00% 2.00 1.50% 1.50 1.00% 1.00 0.50% 0.50 0.06% over optimal 0.00% 0.00 0 50 100 150 200 250 Iterations 300 © 2010 Warren B. Powell 350 400 450 Slide 93 500 Benchmarking on hourly dispatch Optimal from linear program Optimal from linear program Reservoir level Rainfall Demand © 2010 Warren B. Powell Slide 94 Benchmarking on hourly dispatch Approximate dynamic programming ADP solution Reservoir level Rainfall Demand © 2010 Warren B. Powell Slide 95 Benchmarking on hourly dispatch Optimal from linear program Optimal from linear program Reservoir level Rainfall Demand © 2010 Warren B. Powell Slide 96 Benchmarking on hourly dispatch Approximate dynamic programming ADP solution Reservoir level Rainfall Demand © 2010 Warren B. Powell Slide 97 Multidecade energy model Optimal vs. ADP – daily model over 20 years 40.00% 35.00% Percent over optimal 30.00% 25.00% 20.00% 15.00% 10.00% 0.24% over optimal 5.00% 0.00% 0 100 200 300 400 500 600 Iterations © 2010 Warren B. Powell Slide 98 Energy policy modeling Traditional optimization models tend to produce all-or-nothing solutions Investment in IGCC Traditional optimization IGCC is cheaper Pulverized coal is cheaper Approximate dynamic programming Cost differential: IGCC - Pulverized coal © 2010 Warren B. Powell Slide 99 Stochastic rainfall 700 600 Precipitation Sample paths 500 400 300 200 100 0 0 100 200 300 400 500 600 700 800 Time period © 2010 Warren B. Powell Slide 100 Stochastic rainfall 9000 8000 ADP Reservoir level 7000 Optimal for individual scenarios 6000 5000 4000 3000 2000 1000 0 0 100 200 300 400 500 600 700 800 Time period © 2010 Warren B. Powell Slide 101