MAEG4070 Engineering Optimization

Lecture 1 Introduction: What is Optimization?

Yue Chen
MAE, CUHK
email: [email protected]
Sep 7, 2021

1
2
Course Setup

Welcome to the course on Engineering Optimization, with an introduction

of its application in energy systems!

Basic administrative details:

• Instructor:
- Prof. Yue CHEN [email protected] (FYB 604C)
- Office hour: 14:00-15:00, every Friday
- English, Cantonese, Mandarin

• Teaching assistants:
- LIU Zhaocong [email protected] (ERB 411)
- TAO Hanzhi [email protected] (ERB 113)
- YAN Junyan [email protected] (AB1 2/F)

3
4
Course Grading Scheme

• Due days for assignments will be specified.

• Assignments will carry a 50 % penalty if
handed in late.
Assignments • No credit will be given to assignments
25% which are more than three days late.
Final exam • Copying is strictly prohibited.
50%

Mid-term exam • No make-up exam is given unless you get

25% approval from the Director of Registry Services
for permission for absence.
• If not permitted, a zero grade in that exam will
be given.

5
Perquisite and reference

Assuming work with knowledge of:

• Calculus, Linear Algebra
• Formal mathematical thinking
• Programming (Matlab, Python, …)

If you fall short on any one of these things, don’t worry!

Reference:
• S. Boyd, L. Vandenberghe. Convex optimization. Cambridge university
press, 2004.
• R. Sioshansi, A. J. Conejo. Optimization in Engineering: Models and
Algorithms. Springer, 2017.
• S. S. Rao. Engineering optimization: theory and practice. John Wiley &
Sons, 2019.
6
Orientation of this course

Basic models: Algorithms & Solvers: Engineering problems:

• Linear programming • Lagrange / Dual Methods • Scheduling
• Convex optimization • Linearization approaches • Machine Learning
• Integer programming • CPLEX, MOSEK, … • Market Analysis
• …… • …..

A course bridges mathematical models and engineering applications

A course combining math and programming

7
Pictures from Google image
What’s optimization?

Different people may have different understanding about optimization

• Listing all possibilities and pick up the best?
• qualitative suggestions that lead to a better product design?
• trial and error?

Translate into 𝑃𝑃: min 𝑓𝑓(𝑥𝑥)

𝑥𝑥∈𝒳𝒳

Conceptual idea Optimization problem

A quantitative and systematic methodology to search for the best design

among numerous possibilities while satisfying given constraints.
8
Where can we see optimization?

As a matter of fact, any rational human behavior could be modelled as an

optimization problem (intentionally or subconsciously) almost all the time.

Demand & supply

Minimize
of goods
Travel time

warehouse placement Location of warehouses

9
Pictures from Google image
Where can we see optimization?

Parameters, Plan Maximize

Load Bearing

bridge construction Detailed Design

Historical stock prices, Maximize

economy Returns

stock market Stock Portfolio

10
Pictures from Google image
Where can we see optimization?

Parameters, Use plan Minimize

Weight

Airplane design Design

List of items Maximize

Total value

knapsack problem Number of items

11
Pictures from Google image
Successful Applications in Real-world

Organization Application Cost Saving

AT&T Location planning of $ 406 million
telephone selling centers
79 electric utilities Purchasing and energy $ 125 million
management
Delta Air Line Aircraft scheduling $ 100 million
SNCF Train scheduling $ 15 million

• Not all engineering problems can be quantitatively modelled as an

optimization and solved with current technologies.
• With the development of more advanced mathematics and computational
tools, we can model and simulate more sophisticated process, behaviors, etc.
• The application of optimization is increasingly practical, significant, and
worthwhile!
12
History of optimization

Greek mathematicians considers geometrical optimization problems.

Euclid • the minimal distance between a point and a line.

• the greatest area among the rectangles with given total length of edges

Zenodorus • A sphere encloses the greatest volume for a given surface area

Heron • light travels between two points through the path with shortest traveling
time when reflected at or refracted through a boundary

13
Pictures from Google image
History of optimization

The Earliest Optimization Approach: Calculus

𝑑𝑑𝑑𝑑 𝑥𝑥
• where = 0 gives a maximum or minimum.
𝑑𝑑𝑑𝑑
• Pierre De Fermat & Joseph-Louis Lagrange: calculus-based formulae for identifying optima
• Isaac Newton & Johann C.F. Gauss: iterative methods to search for an optimum.

Classic Optimization approaches

• Leonid Kantorovich: Linear Programming, 1939
• George Dantzig: the Simplex Method, 1947
• John von Neumann: the Theory of Duality, 1947
• The Karush-Kuhn-Tucker (KKT) condition, 1939 and 1951 by two separate groups

• Models: Unconstrained/Constrained optimization, single-/multi-objective optimization, etc.

• Algorithms: the steepest descent method, Newton’s method, the penalty method, etc.

14
History of optimization

First Generation Optimization: Local, Iterative, and Gradient-based

• Considers a local optimal
• Iterative methods: based on the location of the previously explored point
• Reliance on gradients or higher order derivatives
• Cons: hard to explain why optimal is optimal, limited computing power

15
Pictures from Google image
History of optimization

Second Generation Optimization: The Metaheuristic Approach

• Genetic Algorithms, “survival of the fittest”; Particle Swarm Optimization
• Pros: global optimization approaches; do not require gradients; support parallel computation.
• Cons: an enormous amount of trial points; the computation time could be hours or days.

Genetic algorithms Particle swarm optimization

16
Pictures from Google image
History of optimization

Third Generation Optimization: The Al-Based Approach

• Also called surrogate-based optimization
• Pros: support parallel computation, do not need gradients, use fewer design trials than before,
offer insight and knowledge about the design problem
• Cons: curse-of-dimensionality (Richard E. Bellman)

17
Pictures from Google image
Basic concept

𝑃𝑃: min 𝑓𝑓(𝑥𝑥)

𝑥𝑥∈𝒳𝒳

• Variable: 𝑥𝑥 - entities and parameters that a designer can change

• Objective: 𝑓𝑓 𝑥𝑥 - the goals that a designer wants to achieve
• Constraints: 𝒳𝒳 – technical restrictions that a product or process must satisfy; or
can be called feasible region; 𝑥𝑥 is a feasible point if 𝑥𝑥 ∈ 𝒳𝒳.

An optimization is either:
• Infeasible, where 𝒳𝒳 = Φ
• Unbounded, where 𝒳𝒳 ≠ Φ, but 𝑓𝑓(𝑥𝑥) can go to infinitely negative (unbounded
below, also called "optimum/minimum not attained") or infinitely positive
(unbounded above, usually treated as an infeasible point) when 𝑥𝑥 varies in 𝒳𝒳
• Feasible, where 𝒳𝒳 ≠ Φ, and the value 𝑓𝑓(𝑥𝑥) is finite.
18
Basic concept

Global optimum. Let 𝑓𝑓(𝑥𝑥) be the objective function, 𝒳𝒳 be the feasible region, and
𝑥𝑥0 ∈ 𝒳𝒳. Then 𝑥𝑥0 is the global optimum if and only if 𝑓𝑓 𝑥𝑥 ≥ 𝑓𝑓 𝑥𝑥0 , ∀𝑥𝑥 ∈ 𝒳𝒳.

Local optimum. Let 𝑓𝑓(𝑥𝑥) be the objective function, 𝒳𝒳 be the feasible region , and
𝑥𝑥0 ∈ 𝒳𝒳. If there is a neighborhood of 𝑥𝑥0 with radius 𝜀𝜀 > 0:

Such that ∀𝑥𝑥 ∈ 𝜒𝜒 ∩ 𝑁𝑁𝜀𝜀 (𝑥𝑥0 ), we have 𝑓𝑓 𝑥𝑥 ≥ 𝑓𝑓 𝑥𝑥0 . Then 𝑥𝑥0 is a local optimum.

19
Classification of optimization problems

Problem:
• Constrained v.s. nonconstrained
• Single-objective v.s. multi-objective
• Single-level v.s. multi-level
• Deterministic v.s. uncertain

Response:
• Linear v.s. nonlinear
• Convex v.s. nonconvex
• Smooth v.s. nonsmooth

Variable:
• Continuous v.s. discrete (integer/binary)

20
Content of this course (tentative)
Linear programming
Lecture 2
Linear
Dual Theory – Part I
Lecture 3
Linearization
technique Programming
Lecture 7 Unconstrained optimization Lecture 13
Lecture 5 & 6

Optimization Non-Linear Constrained optimization

Lecture 8 & 9
Lecture 1 Lecture 4
Dual Theory – Part II Engineering examples
Lecture 10
Lecture 14

Distributed optimization Lecture 11

Multi-objective optimization Lecture 12

Theory
21
Procedure of applying optimization in engineering

In general, applying optimization in engineering follows five steps:

Step 1: Propose optimization problems, collect relevant data and information

Step 2: Establish mathematical model: determine the variables, the objective

function and constraints

Step 3: Analyze the model and select the appropriate optimization method

Step 4: Solve: obtain the optimal solution via programming

Step 5: Test and implementation of optimal solution

22
Pictures from Google image
Example-1 Parametric Regression

Given a data set 𝑦𝑦𝑖𝑖 , 𝑥𝑥𝑖𝑖 , ∀𝑖𝑖 = 1, … , 𝑁𝑁

we assume that the relationship between the
dependent variable 𝑦𝑦 and the regressor 𝑥𝑥 is
linear.
𝑦𝑦 = 𝑏𝑏0 + 𝑏𝑏1 𝑥𝑥 + 𝜀𝜀
We want to minimize the sum of square error.

𝑁𝑁
2
min � 𝑦𝑦𝑖𝑖 − 𝑏𝑏0 − 𝑏𝑏1 𝑥𝑥𝑖𝑖
𝑏𝑏0 ,𝑏𝑏1 𝑖𝑖=1

23
Example-2 Production planning
A company has some resources to produce three products (denoted as A, B, C). Each
product consumes a different mix of resources, and there will be a profit from selling
the product. The endowment of resources and its relationship with products are:

A B C Endowment
Steel 3 4 2 600
Wood 2 1 2 400
Label 1 3 3 300
Machine 1 4 4 200
Profit 2 4 3

Question: How to maximize the total profit?

24
Example-2 Production planning

Variables: Denote 𝑥𝑥1 , 𝑥𝑥2 , 𝑥𝑥3 be the production of products A, B, C, respectively.

Objective: How to maximize the total profit?

Constraints: do not violate the resource endowment

Endowment limits

Production must larger than 0

25
Example-3 Transportation

A building materials company has three cement factories 𝐴𝐴1 , 𝐴𝐴2 , 𝐴𝐴3 and four dealers
𝐵𝐵1 , 𝐵𝐵2 , 𝐵𝐵3 , 𝐵𝐵4 . Its output, sales volume and freight are shown in the table below

𝐵𝐵1 𝐵𝐵2 𝐵𝐵3 𝐵𝐵4 Production

𝐴𝐴1 8 7 3 2 2000
𝐴𝐴2 4 7 5 1 10000
𝐴𝐴3 2 4 9 6 4000
Sales 3000 2000 4000 5000

Question: How to minimize the freight?

26
Example-3 Transportation
Variables: Denote 𝑥𝑥𝑖𝑖𝑖𝑖 as the quantity from factory 𝐴𝐴𝑖𝑖 to dealer 𝐵𝐵𝑗𝑗

Constraints: Production ability limits; selling requirement

Objective: Minimize the total freight

Production ability

Sells requirement

Quantity must larger than 0

27
Example-4 Location problem

Suppose there are 𝑗𝑗 = 1, … , 𝑛𝑛 markets whose location is (𝑎𝑎𝑗𝑗 , 𝑏𝑏𝑗𝑗 ). Market 𝑗𝑗 needs 𝑞𝑞𝑗𝑗
product. We plan to build 𝑚𝑚 warehouse whose capacities are 𝑐𝑐𝑖𝑖 , 𝑖𝑖 = 1, … , 𝑚𝑚. How to
choose the location of these warehouses so that the total cost is minimized?

28
Thanks！

29