Académique Documents
Professionnel Documents
Culture Documents
Fengqi You
Chemical and Biological Engineering Northwestern University Evanston / Chicago, IL 60208
Motivation
Other objectives
Uncertainty & risk, responsiveness Energy efficiency and energy payback time
* Grossmann, I. E., & Guilln G, G. (2010) Scope for the application of mathematical programming techniques in the synthesis and planning of sustainable processes. Computers & Chemical Engineering, 34 (9), 1365-1376.
Motivation
36 32
Production Volume
28 24 20 16 12 8 4 0
Biodiesel Other Renewables Cellulosic Biofuels
Ethanol Biodiesel
Corn Biofuels
20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 15 20 16 20 17 20 18 20 19 20 20 20 21 20 22
(Energy Independence and Security Act of 2007; Biomass Multi Year Program Plan, EERE, U.S. DOE, 2012 )
Year
Feedstock Production
Feedstock Logistics
Biofuels Production
Biofuels Distribution
Harvesting sites
Collection Facilities
Biorefineries
Demand Zones
7
Challenges
Major Challenges:
How to capture the characteristics of biofuel supply chains?
Biomass: deterioration, seasonality, preprocessing and storage Biofuels: various conversion pathways/technologies, intermodal network
How to effectively integrate all the elements of the biofuel supply chain across temporal and spatial scales How to quantify the environmental impacts and social performance? How to tradeoff the economic, environmental and social objectives?
2012 CAPD Annual Meeting, Carnegie Mellon University, Pittsburgh, PA 8
Integrated Approach
Techniques
Process/SC Design Planning & Scheduling Inventory Control Multi-Objective
Coordinate the supply, production and distribution of biomass & biofuel For seasonal supply of biomass and uncertain demand of biofuels Economic, environmental (LCA: field-to-wheel), social (EIO: jobyear)
Production Inventory
Harvest
9 10 11 12
Uncertainty
Constraints:
Minimize: Total annualized cost Minimize: Total GHG emission (life cycle assessment: GREET model @ ANL) Maximize: Total job year created (economic input-out analysis: JEDI model @ NREL) Flow /inventory balance constraints
Flow balance at the harvesting sites Inventory balance at collection facilities Inventory balance at the biorefineries Flow balance at the demand zones
PHASE IV Interpretation
12
13
Major output
Feedstock availability, geographical distribution and seasonality Harvesting site locations, harvest capacity, weather variability Transportation network and modes, distance, intermodal transportation Biomass density, weight and volume capacity, preprocessing and storage When, where, which biomass should be harvested? How, when, how much to transport the feedstocks? Where, how much and how long should the biomass be stored? When, where and what type should the feedstocks be preprocessed? What should be the optimal capacity of collection/storage facilities?
2012 CAPD Annual Meeting, Carnegie Mellon University, Pittsburgh, PA 14
Integrated Biorefineries
Potential locations and conversion pathways, transportation connectivity Production capacity, techno-economics, government incentives and policy Feedstock handling, water usage and availability, byproducts ASPEN models for feedstocks and technologies with different capacities
Sugar/Starch Biomass
Major output
C5/C6 Fermentation
Biochemical Conversion
Number, size, location and technologies Amounts of ethanol and byproducts Biomass Feedstock and water usage Production and inventory levels
Pretreatment
Thermochemical Conversion
WGS Gasoline/Diesel
15
Major output
When, how much to transport the biofuels from biorefineries to blending facilities and demand zones? Which transportation mode to be used for the deliveries? What is the maximum optimal distribution distance for different transportation mode (truck vs. dedicated ethanol pipeline)?
2012 CAPD Annual Meeting, Carnegie Mellon University, Pittsburgh, PA 16
- constraint Method
Cost
Max Optimal Cost
Minimize: Emission
Impossible!
Smallest Emission Largest Emission
Emission
17
IL Population density
102 Counties
102 harvesting sites 102 potential collection facilities 102 possible biorefinery site locations 102 blending facilities/demand zones
138 MGY
Population density
30%
Feedstock
Cost Breakdown
Biochemical Thermochemical
Feedstock Inventory
19
Suboptimal Solutions
Population density
Agricultural residues: corn stover, etc. Energy crops: switchgrass, miscanthus, etc. Wood residues : forest and mill residue, urban wood Two major conversion technologies (Biochem. and thermochem.) Three major transportation modes (Truck, train, & water) 102 Counties (harvesting sites, plant locations, demand zones) 12 time periods per year (for 20 years)
2012 CAPD Annual Meeting, Carnegie Mellon University, Pittsburgh, PA
20
Pareto Curve
Good Choice
Population density
Feedstock Inventory
10% 3% 19%
Resource of wood residue
Investment
39%
29%
Feedstock
Cost Breakdown
22
14000 12000 10000 8000 6000 4000 100000 150000 200000 250000 300000 350000
Remarks
Unit cost reduce from $3.663/gal in Case 1 to $3.225 in Case 2
Large scale production (near term vs. Yr 2022)
Economy of scale, shorter average transportation
Feedstock diversity
hedge the seasonality, lower inventory cost, reduce deterioration
6
25
Problem Solution
pdf for x
0.3
0.2
0.1
0 -10
-5
10
26
Problem Statement
Minimizing Cost & Risk for Biofuel SC Design under Uncertainty
Given: time periods, cost data, potential locations and technologies, production & transportation capacity, incentives, uncertainty distributions of supply and demand Decisions: network design, facility location, technology selection, capital investment, production levels, inventory control, and logistics management Objective: Minimizing Cost & Risks
Harvesting Sites
Hydrocarbon Biorefineries
2012 CAPD Annual Meeting, Carnegie Mellon University, Pittsburgh, PA
Demand Zones
27
IL Population density
102 Counties for harvesting sites, potential biorefinery plant locations, and demand zones Three Types of Feedstocks
Agricultural residues, energy crops, & wood residues
Two-stage Decisions
First stage decisions (here-and-now)
Network design, technology selection, capital investment
Result of SP Model
E[Cost] = $ 2,822.6 15.6 MM
(95% confidence interval)
N = 1,000 scenarios
30
Stochastic Programming Model 100 scenarios 408 65,118,126 2,939,130 1,000 scenarios 408 651,171,126 29,379,330
Impossible to solve directly takes >10 hours by using standard L-shaped only 1.5 hours with multi-cut version
Computational Performance
31
Risk Management
SP model: minimize expected cost (risk-neutral objective) A few risk measures
Variance Upper partial mean
Probabilistic financial risk (Barbaro et al., 2002) Downside risk (Eppen et al., 1988) CVaR (Rockafellar and Uryasev, 2000)
0.3
0.2
0.1
0 -10
-5
10
Downside Risk
32
min : CVaR ( x, ) =
sS
ps s
+ VaR
) b, x n f ( x,=
s CostsOperations VaR, s S s 0, s S
VaR 0
2012 CAPD Annual Meeting, Carnegie Mellon University, Pittsburgh, PA 33
VaR Constraints
34
CVaR
35
Downside Risk
36
C5/C6 Fermentation
Biochemical Conversion
Pretreatment
Thermochemical Conversion
WGS Gasoline/Diesel
Triglycerides Source
Oil Seeds
Objective: Integration of biorefinery process design with biofuel supply chain optimization
Representation of detailed process models and operational logistics Multi-scale and multi-site modeling geospatially distributed production facilities and supply chain infrastructure Focusing on advanced infrastructurecompatible biofuels, i.e. drop-in fuel
37
Update LB
Fixed 0-1 variables
39
Motivation
year1
Hundreds of oil spills (>10,000 gallons) per Planning the response operations is important but non-trivial
The case of Deepwater Horizon/BP Oil Spill
Costs up to $40 billion2 for cleanup and coastal protection Many thousands of people and equipments involved
1970
1980
1990
2000
41
Literature Review
Very few on oil spill response planning
Most modeling papers on oil spill are for oil weathering process Psaraftis & Ziogas (1985), Wilhelm & Srinivasa (1996, 1997), Ornitz & Champ (2003), Gkonis et al. (2007), etc.
Coastal protection operations have not been taken in account in response planning it may cost more to protect the coast than cleanup Only single objective is used minimizing either time or cost
Multi-objective optimization
2012 CAPD Annual Meeting, Carnegie Mellon University, Pittsburgh, PA 42
wind
Spreading
Sedimentation
2012 CAPD Annual Meeting, Carnegie Mellon University, Pittsburgh, PA 43
45
Mechanical (skimming)
Water content Slick thickness Weather condition Hydrodynamics
In-situ Burning
Slick thickness Oil viscosity Parent oil density Weather condition
Chemical Dispersant
Emulsification degree Dispersant availability Weather & sea condition Regulation
Coastal Protection
Oil slick area Boom availability Staging area location Sea & weather condition
46
Problem Statement
Given:
An oil spill
location & spill amount, oil physical / chemical properties, cleanup target
Major Decisions:
Oil spill cleanup Coastal protection Oil transport & weathering
Challenges
Modeling Challenge
Coastal protection, spill cleanup, oil transport and weathering process
Time-dependent oil physical and chemical properties, hydrodynamics, weather conditions, facility availability, performance degradation, cleanup operational window, and government regulations Different time representation: discrete (planning) vs. continuous (weathering) Account for the complex interactions between them (spreading, evaporation, dispersion, and emulsification v.s. cleanup and boom protection)
Multi-Objective Challenge
Measure of responsiveness Tradeoff between economic and responsiveness
Computational Challenge
Multi-Objective mixed-integer dynamic optimization (MIDO) problem Non-convex MINLP after discretization based on orthogonal collocation on FEs
2,052 discrete variables, 11,482 continuous variables, 14,006 constraints
2012 CAPD Annual Meeting, Carnegie Mellon University, Pittsburgh, PA 48
Constraints:
Cleanup planning constraints
Availability of mechanical systems Cleanup rate of skimmers Availability of burning systems Operational window of burning sys. Availability of dispersant systems Performance of dispersant systems Chemical dispersant balance Dispersant availability Regulation on dispersant application
(bilinear terms)
49
Challenge: Initialization
Resulting model is a large-scale non-convex MINLP
EX1: 2,052 discrete var., 11,482 continuous var., 14,006 constraints EX1: Solving the RMINLP directly with any NLP solver leads to infeasibility
* Biegler et al. (2002);
Cuthrell & Biegler (1987)
50 2012 CAPD Annual Meeting, Carnegie Mellon University, Pittsburgh, PA
MILP
Cleanup rate Volume, area, thickness, viscosity, water content, evaporation rate, dispersion amount, etc.
(Oil Weathering)
ODE
Step 1: Solve the ODE with zero cleanup rate (eq. to natural weathering process) Step 2: Construct the approximate MILP model for initialization
Fix state variables based on the ODE solution, except volume and area Compute the percentage of oil removed by natural weathering at time t (t) Add the following volume balance constraints to the MILP:
51
S3
Solution
Direct solution: infeasible for any solver Proposed approach: 139CPUs/instance (CPLEX + KNITRO + DICOPT)
52
A
1000
B C
150
170
Conclusion
Remarks
Summary
Various objectives for energy-environmental system optimization
Economic, environment, social, risk, responsiveness
Key component: finding a suitable quantitative measure Computational challenges lie in:
Large-scale optimization problems Handling uncertainties and risks
Extensions
Algae for CCS and biodiesel production Organic photovoltaic systems ($$$, LCA, EPBT) New material and process development for CCS
2012 CAPD Annual Meeting, Carnegie Mellon University, Pittsburgh, PA 54
Fengqi You
Chemical and Biological Engineering Northwestern University
you@northwestern.edu http://you.mccormick.northwestern.edu
2012 CAPD Annual Meeting, Carnegie Mellon University, Pittsburgh, PA 55