Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
CE 191: Civl and Environmental Engineering Systems Analysis
LEC 00 : Course Introduction
Professor Scott MouraCivl & Environmental EngineeringUniversity of California, Berkeley
Fall 2013
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 1
Why take CE 191?
Learn to abstract mathematical programs
from physical systems to “optimally” design
a civil engineered system.
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 2
Why take CE 191?
Learn to abstract mathematical programs
from physical systems to “optimally” design
a civil engineered system.
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 2
You graduated from Berkeley, you just found your dream job in your dream field, in your dream city.
One day, your boss calls you in his office, and asks you to
solve a civil and environmental engineering problem.
You spend three weeks trying to find a solution. You find « some kind of a solution, but you are not sure ».
The week after, your boss calls you again.
What do you do?
One year from today
One answer (Garey and Johnson)
[Computers and intractability, Garey and Johnson, 1979]
I can't find an efficient algorithm, I guess I'm just too dumb
[Computers and intractability, Garey and Johnson, 1979]
Answer 1 (if you did not take CE 191)
I can't find an efficient algorithm, because no such algorithm is possible
[Computers and intractability, Garey and Johnson, 1979]
Answer 2 (in your wildest dreams)
I can't find an efficient algorithm, but neither can all these famous people
[Computers and intractability, Garey and Johnson, 1979]
Answer 3 (after taking CE 191)
Your boss wants the cheapest solution: - Can you find a solution? à feasibility - Can you find a cheaper solution? à optimality - Can you find the cheapest solution? à uniqueness
- Can you find a cheap solution? à suboptimal - How cheap is your solution? à degree of suboptimality
- Why can’t you find a cheap solution? à hardness
Problem 1: feasability, uniqueness, optimality
Your boss gives you a huge data file He does not care if your solution is cheap, as long as you can tell
him how cheap it is He wants your algorithm to find the solution of the problem in 5
minutes - Is your algorithm fast? à polynomial time - Is your algorithm slow? à non polynomial time
Problem 2: computational complexity
Your boss gives you a huge data file every day She is happy if you can guarantee her that on average you will give
her the optimal solution every other day in less than 5 minutes - Does you algorithm always converge? à deterministic/random - Does it do well on average? à expected sense
Problem 3: deterministic or not
Your boss wants to know how many trucks she needs to send to Sacramento next Friday.
Your boss wants to know how many pounds of sand she needs to
send to Sacramento next Friday. Which problem is easiest to solve for you? - Your algorithm says: 223276.25 pounds à continuous - Your algorithm says: 25 trucks à discrete - Your algorithm say: 24.6 trucks à LP-rounding?
Problem 4: discrete / continuous
One of the people you supervise (from Stanford) tells you he just found the perfect model for your problem, it is very precise, but it involves the cosine of the square root of the quantity of fuel burned by your trucks.
The other person you supervise (from MIT) tells you she just found a
not so precise model, but it is proportional to the quantity of fuel burned by your trucks plus a constant.
Which one should you put in your algorithm to give the best answer
to your boss in a reasonable time? - Tractable models à linear/affine - Harder models à nonlinear/nonconvex
Problem 5: linear / nonlinear
Class Format
Lectures: MW 8-9am, 406 Davis Hall
Lab Section: Th 2-5pm or 5-8pm, 345 Davis Hall
Website: http://bspace.berkeley.edu
Professor Scott [email protected]
Office Hours: M,Tu,W 9-10am@ 625 Davis Hall
GSI Xiaofei [email protected]
Office Hours: Tu 10-11am, F 10-11am@ 345 Davis Hall
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 3
Technical Content
Optimization TopicsLinear programming (LP)
Quadratic programming (QP)
Integer programming (IP)
Dynamic programming (DP)
Nonlinear programming (NLP)
CEE TopicsWater resource management
Planning future energy supplies
Scheduling in a construction project
Investment portfolio optimization
Bicycle sharing
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 4
Textbooks
No textbooks are required,
the following is officially recommended for additional background:Civil and Environmental Systems Engineering; C. Revelle, E. Whitlatch, R.Wright; Pearson Prentice Hall, 2004.
The following textbooks are also useful resources:Convex Optimization; S. Boyd and L. Vandenberghe; Cambridge UniversityPress, 2004.Principles of Optimal Design; P. Papalambros and D. Wilde; CambridgeUniversity Press, 2000.
All textbooks have been placed on 2 hr. reserve in the library.
Lecture format:Mostly blackboard w/ some slidesSupplemental slides available online
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 5
Software
MATLAB used for lab assignments.
UCB IST License: You can request a license athttp://ist.berkeley.edu/software-central/matlab
For purchase from Mathworks for $99:http://www.mathworks.com/academia/student_version/
Computer Access: A CEE Computer Lab Account is required to use the computersin 345 Davis. Use the link below to request an account. http://www.ce.berkeley.edu/resources/computing/create_lab_account
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 6
Lab 0 - Matlab Review
Familiarize yourself with computer lab in 345 Davis
Use class website bspace.berkeley.edu
Complete survey on background
Review basic Matlab tasks
Rehearse electronic submission to bSpace
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 7
Grading
Straight scale (no curve):
Labs: 50pts - 5 lab assignments, 10pts eachMidterm: 20pts - Mon Oct 21, 8-9am
Final: 30pts - Exam Group 4: Mon Dec 16, 2013 7-10pm
A total of 100pts are possible.
Philosophy: Consistency and transparency
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 8
Policies
Late Submissions: One point is subtracted for each 24 hours submitted late (roundedup to nearest integer). Two free late days are allowed on any lab of your choice.
Regrade Policy: If you feel a problem was graded incorrectly, you may submit aregrade request to the GSI. This request MUST be submitted within one week ofreceiving the graded assignment, with a short paragraph justifying the regrade. Anyregrade request is subject to a full regrade, i.e. points may be lost.
Planned Absences: You may request to submit assignments early or late. E-mail meyour request two weeks prior to the assignment due date. Requests due to extendedholidays will not be granted. Requests due to emergencies will be handledcase-by-case.
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 9
How to Succeed
Ask questions in class
Form a study group
See instructor after class
See instructor during OH
See GSI during OH
Send us an e-mail. Use [CE 191] in subject.
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 10
Flowchart of Methods-based Courses
CE 191 Op(miza(on
E 7 Matlab Intro
CE 155 Transporta(on
Systems
CE 186 Cyber Physical
Systems
CE 268E Civil Systems & Environment
CE 271 Sensors & Signals
CE 2XX Energy Systems
& Control
CE 290I Sensors & Signals
CE 291F Control of DPS
EE 127A Op(miza(on Models & Apps
IEOR 262A/B, 263A/B, 264 Math Programming
EECS 227 Convex
Op(miza(on
EE 120, C128 Control Systems
ME C134 Control Systems
ME C23X, EE 220-‐3 Control Systems
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 11
Example 1: Barcelona Water Network
There are 67 tanks, 10 water sources, 111 valves / pumps, 88 points of water consumptionand 15 complex nodes. [Trnka, Pekar, Havlena, IFAC2011]
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 12
Example 2: Shortest Path (routing)
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 13
Example 3: Electrified Vehicle Energy Management
Optimized PHEV charging based on variableelectricity price via quadratic programming.
SUPERVISORY
CONTROLLER
M/G1
M/G2
PLANETARY
GEAR SET
BATTERY PACK
DRIVE
CYCLE
ENGINE
VEHICLE
Battery State of Charge
Acceleration
Engine
Speed
Engine
Torque
M/G1 Torque
M/G2
Torque
Vehicle
Speed
Fuel Consumption
Cost
Grid Electricity
Consumption Cost
Anode-Side Film
Growth Penalty
Optimal energy management given statisticaldriving behavior via stochastic dynamic
programming.
Prof. Moura | UC Berkeley CE 191 | LEC 00 - Intro Slide 14