Back to Table of Contents

People use “optimize” rather frequently these days. This is a quote from BBC's website about its sleep profiling tool, “Whether you're having trouble sleeping or not, use the Sleep Profiler and get tailored advice on how best to * optimise* your sleep.” For someone doing research using optimization (see definition below), it is quite hard to grapple with the concept of optimizing one's sleep. What it really means is using this profiler can make you sleep better, which sounds a little less impressive.

Interestingly Oxford Dictionary Online and Merriam-Webster Learner's Dictionary define “optimize” slightly differently:

- “make the best or most effective use of (a situation or resource)” - Oxford Dictionary Online
- “to make (something) as good or as effective as possible” - Merriam-Webster Learner's Dictionary

Although both sound reasonable, read literally the action of “optimize” requires achieving or aiming to achieve superlative qualities (the best, the most or as … as possible), which is rarely attainable in real life. Daily use of the word in fact means “make *most* of a situation”, “make *more* effective use of resource”, or “to make (something) as good as one *would*”. This is very different from what we do in *optimization* studies.

**optimization** (*òp´te-mî-zâ´shen*) – The procedure or procedures used to make a system or design as effective or functional as possible, especially the mathematical techniques involved. (The American Heritage® Dictionary of the English Language, Third Edition copyright © 1992 by Houghton Mifflin Company)

Wikipedia's entry on Mathematical Optimization gives a nice overview of optimization theory and techniques, with plenty of resources for further reading. It serves well as an entrance to the wonderful world of optimization studies. On the other hand, the sheer amount of information can be off-putting - who has the time to become an optimization expert, in order to solve the problem in the project due next week? Luckily, as long as you understand the basic concept of optimization and can learn to use a tool, you can use these techniques in your work. Most of the optimization studies aim for the superlatives. The question **ADOPT** intends to address, however, is how users can take advantage of this technology in their practice in designing buildings. As a result, we are not aiming to provide you a tool that gives you the best solution. Instead, our optimization tools will help you identify better solutions yourself. In this context, “optimize” here does mean “improve”.

There are really thousands and thousands of different optimization algorithms (and their flavors). They fall broadly into two camps (see Wiki): iterative methods (many are gradient-based) and heuristics (mainly direct-search methods, many of which are stochastic). And then there are hybrid methods that combine two or more techniques from either camp. So the total number of possible algorithms is practically infinite. I like the diagram below very much. It is again from Wikipedia, on Heuristic algorithms.

You can see how complicated the maze of algorithms is. For example, for Evolutionary Algorithms, there are five groups. Each of these groups contains different algorithms for constrained/unconstrained and/or single/multiple-objective problems. Take multiple-objective Genetic Algorithms (MOGAs) for instance, there are:

- FFGA - Fonseca and Fleming’s multiobjective EA
- NPGA - Niched Pareto Genetic Algorithm
- HLGA - Hajela and Lin’s weighted-sum based approach
- VEGA - Vector Evaluated Genetic Algorithm
- NSGA - Deb and Srinivas's Nondominated Sorting Genetic Algorithm
- PAES - Pareto Archived Evolution Strategy
- PESA - Pareto Envelope-based Selection Algorithm
- SOEA - Single-objective evolutionary algorithm using weighted-sum aggregation
- SPEA - Strength Pareto Evolutionary Algorithm
- NSGA2, SPEA2 and NPGA2, which are substantial upgrades to the original algorithms
- and more.

NSGA2 and SPEA2 are probably the most widely used MOGAs to date. Researchers keep making improvements to them, for example, the pNSGA2 and aNSGA2 (NSGA2 with passive or active archives) methods reported at BSO12 recently.

“*What is the best optimization algorithm*” is one of the most frequently asked questions among “newbies”. I know a mathematician, who, when he receives this question, will stare at the inquirer and blurt out “what is your problem?!” (The exclamation mark was my addition ) His answer is quite literally “it depends on the problem you are trying to solve”. In building design, however, many optimization problems are complex and “noisy”, on which gradient-based methods will not work. Evolutionary Algorithms are the most popular choice because of their “robustness”, i.e. a correctly implemented EA tends to solve the problem, no matter what kind of problem it is.

Evolution does work! Just think that this very mechanism has led to the development of humans and other complex organisms from the humble beginnings of the first single cell organism; using it to improve building designs is really a trivial task. There is a bit of an initial learning curve, true; especially hard is getting used to the jargon, such as binary encoding, constraints, crossover rate, tournament selection, elitism, … A large part of the aims of this project is to develop tools that incorporate our understanding of optimization problems in building design, so you, as users of the tools, do not have to worry too much about the details in the algorithms. Instead, you can focus on solving your design problem.

What are the design options and parameters, and which operation strategies and parameters do people optimise when designing a building? Again, we are in the process of collecting information from literature. From our experience, though, there are three basic types of variables:

- Design options – this type of variables are normally discrete and nominal. They can be encoded as integers, which may or may not represent an ordinal rank. Optimisation problems involve options that are often combinatorial, therefore harder to solve. Examples of design options include wall construction type, glazing options and building layout.
- Parameters – can be continuous or discrete, but at least ordinal. Examples of parameters include insulation thickness or U value, ventilation rate and window-to-wall ratio.
- Control signals – a control signal is a timed option or parameter. Building/system operation is defined by a sequence of control signals. Control signals are often treated as a set of options/parameters by fixing the time sequence. HVAC operation and timing is an example of a control signal.

A building design problem often involves (hierarchical) combinations of options, parameters and control signals. It will be interesting to see how people choose and treat variables in their optimisation studies.

Corresponding to the choice of variables, there are some typical optimisation problems:

- Optimal control problems – control signals only
- Sizing problems – parameters only
- Configuration problems – a small number of design option variables plus parameters. Control signals may or may not be considered. All the variables form a single-root hierarchy (a tree).
- Whole building/system optimisation – design option variables, parameters, and control signals. They may comprise of any number of structurally linked or disconnected sub-problems. It is a forest.

What are the objectives of building optimisation? Or, for what purposes has optimization been used in building design and operation? Lots of information can be found in literature, though there is as yet no systematic conclusion. We will carry on working to collect and archive research in this area throughout the ADOPT project. The section “Optimisation case studies” gives a template of the kind of information we are collecting.

From our own experience, optimisation objectives for building design and operation can include the following:

- Minimising energy demand: lighting, heating, cooling, auxiliary (fans and pumps)
- Minimising primary energy consumption
- Minimising embodied energy
- Minimising operational carbon emissions
- Minimising embodied carbon emissions
- Minimising a more inclusive objective of “environmental impact” (for which various evaluation metrics have been used)
- Minimising operational cost
- Minimising capital cost
- Some forms of a margin
- Maximising indoor environmental qualities such as daylight, air quality and thermal comfort
- Maximising adaptability (ability to adapt to future changes)

Often more than one of these objectives may be of interest, and they are also frequently in opposition to each other, for example minimising energy demand often involves increasing capital cost. We are investigating how objectives are selected and treated, whether a single objective or a multi-objective approach is more desirable, and how to deal with constrained problems (e.g. comfort criteria, costs etc., these may be treated as a constraint or an objective).

- National Renewable Energy Laboratory: BEopt
- Lawrence Berkeley National Laboratory: GenOpt
- VTT Technical Research Centre of Finland and AaltoUniversity: MOBO
- MATLAB Optimization and Global Optimization Toolboxes
- Rhino/Grasshopper plugins such as Galapagos and Octopus
- Libraries in many languages including Python, R, …
- and many more