Optimal Planning of Urban Infrastructure Networks
for Multiple Energy Carriers
Operations research in engineering
Johannes Dorfner
Institute for Renewable and Sustainable Energy Systems
Technische Universität München
5th MSE Colloquium
9 July 2015
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
1 / 22
Table of Contents
Introduction
Methods
Case study
Results
Conclusion
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
2 / 22
Section 1
Introduction
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
3 / 22
Introduction
Modelling and optimisation
Modelling
Translating the abstract rules concerning the decision space of a certain topic
into mathematical equations. Here, these equations then form an optimisation
problem.
Energy infrastructure
Every technical component between
É
É
transformer substation and the power outlet and
gas pipeline and a warm house,
in other words: energy conversion (local and plants), distribution networks for
providing electricity and heat.
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
4 / 22
Introduction
Conceptual problem formulation
Main question
In an ideal sandbox world (greenfield planning),
É
É
É
which energy carriers
energy distribution networks and
energy conversion process
should be employed where to satisfy our energy demands?
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
5 / 22
Introduction
Motivation
Anwendungsbilanzen 2011 und 2012 für den Sektor Private Haushalte by RWI Essen, 2013
Johannes Dorfner
TUM EI ENS
http://ag-energiebilanzen.de/
Infrastructure Optimisation
07/2015
6 / 22
Section 2
Methods
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
7 / 22
Methods
rivus: Optimal infrastructure network planning
Input
É
Energy demand
É
Street network graph
É
Technology portfolio
É
É
Output
É
Infrastructure network
É
Conversion capacities
É
Total costs
Cost and technical parameters
É
Commodity usage
Commodity source points
É
CO2 emissions
Johannes Dorfner
TUM EI ENS
rivus
Infrastructure Optimisation
07/2015
8 / 22
Methods
Mathematical modelling: Fixed cost formulation
z = cfix  + cvar P
(1)
P ≤  Pmax
(2)
P ≥  Pmin
(3)
z (e/a)
One binary decision variable  ∈ {0, 1} allows to approximate economics of
scale: Higher capacities P (kW) lead to lower specific costs z (e/a):
cvar
cfix
0
0
Pmin
Pmax
P (kW)
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
9 / 22
Methods
Mathematical modelling: Fixed cost formulation
One binary decision variable  ∈ {0, 1} allows to approximate economics of
scale: Higher capacities P (kW) lead to lower specific costs z (e/a):
z = cfix  + cvar P
(1)
P ≤  Pmax
(2)
P ≥  Pmin
(3)
z (e/a)
MILP
cvar
cfix
0
0
Pmin
Pmax
P (kW)
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
9 / 22
Methods
Mathematical modelling: Fixed cost formulation
One binary decision variable  ∈ {0, 1} allows to approximate economics of
scale: Higher capacities P (kW) lead to lower specific costs z (e/a):
z = cfix  + cvar P
(1)
P ≤  Pmax
(2)
P ≥  Pmin
(3)
z (e/a)
MILP
cvar
cfix
0
0
LP
Pmin
Pmax
P (kW)
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
9 / 22
Methods
Mathematical modelling: Fixed cost formulation
One binary decision variable  ∈ {0, 1} allows to approximate economics of
scale: Higher capacities P (kW) lead to lower specific costs z (e/a):
z = cfix  + cvar P
(1)
P ≤  Pmax
(2)
P ≥  Pmin
(3)
z (e/a)
MILP
cvar
cfix
0
0
LP
Pmin
Pmax
P (kW)
Drawback: binary variables can make problem much harder to solve.
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
9 / 22
Methods
Schema of modelling and optimisation toolchains
Research
question
Assumptions
Input
Model
Data I/O
Solver
É
Spreadsheet
É
MATLAB
É
GAMS
É
CPLEX
É
Geographic
É
Python
É
Pyomo
É
Gurobi
É
Database
É
Excel
É
TOMLAB
É
GLPK
Raw result data
Output
Feedback
Johannes Dorfner
TUM EI ENS
É
Reports
É
Diagrams
É
Maps
Interpretation
Infrastructure Optimisation
07/2015
10 / 22
Methods
Reproducibility
All models are public. Results can be reproduced, verified and criticised.
https://github.com/tum-ens/urbs
https://github.com/ojdo/rivus
Johannes Dorfner
TUM EI ENS
(capacity planning, unit committment)
(infrastructure networks)
Infrastructure Optimisation
07/2015
11 / 22
Methods
Automation by software stack
Heavily relying on existing packages for diverse tasks:
É
Python as scripting language, for automation and code structure
É
Pandas for data pre- and post-processing
É
Pyomo for mathematical modelling, solver interfaces
É
SciPy stack for general purpose algorithms and visualisation
É
IPython (and Notebook) for exploratory programming
É
Geographic packages for crafting custom geoprocessing algorithms
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
12 / 22
Section 3
Case study
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
13 / 22
Bilder © 2015 DigitalGlobe,GeoBasis-DE/BKG,Kartendaten © 2015 GeoBasis-DE/BKG (©2009), Google
Spatial demand distribution
Aggregated peak demand (kW) per street segment
Electricity
Heat
Aggregation
from 1738 buildings
to 203 edges
for smaller problem
Total peak demand
Electricity 26.7 MW
Heat 47.6 MW
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
16 / 22
Input data overview
Data
Geographic 1738 buildings, 203 edges, 153 vertices, 2 source vertices
Buildings mainly residential, some commercial, few big industrial
Peak demand 26.7 MW electric, 47.6 MW heat
Full-load hours 3620 h electric, 2700 h heat
Time resolution 5 time steps (2 for peak loads, 3 represent steady state)
Scenarios
Both: green-field planning, i.e. no existing infrastructure assumed
Scenario
Description
Today
Future
cogeneration of heat & power from natural gas
heat pumps (effective efficiency > 100 %)
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
17 / 22
Section 4
Results
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
18 / 22
Capacities in scenario Base
Network capacities and conversion unit capacities (both kW)
Electricity
Heat
Natural gas
Full electric grid, full gas grid; heat generated locally from gas
https://github.com/ojdo/rivus/runhg15.py:scenario_no_electric_heating()
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
19 / 22
Capacities in scenario Future
Network capacities and conversion unit capacities (both kW)
Electricity
Heat
Natural gas
Fuller electric grid, no gas grid; heat generated locally in heat pumps
https://github.com/ojdo/rivus/runhg15.py:scenario_renovation()
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
20 / 22
Section 5
Conclusion
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
21 / 22
Conclusion
What to remember?
Modelling is useful tool to explore sensitivities in unknown decision space
(a.k.a. operations research)
Reproducibility browse or try it out for yourself
https://github.com/tum-ens/urbs (stable)
https://github.com/ojdo/rivus
(fresh beta)
Infrastructure coupling of sectors electricity, heat (and others) will play key role in
future energy system (yes, now it sounds obvious)
The conclusions of most good operations research studies are obvious.
– Robert E. Machol (1917–1998)
Johannes Dorfner
TUM EI ENS
Infrastructure Optimisation
07/2015
22 / 22