Articles, Blog

Interactive optimization for planning of urban energy systems using parallel coordinates

September 15, 2019


Today, both climate change, and high urbanization rates are encouraging people around the world to rethink the way they consume natural resources. This challenges in particular urban planners,
who must develop cities which meet not only social goals, such as providing enough dwellings
in places with high quality of life, but also places which achieve environmental and energy
goals, such as promoting renewable energies, and reducing greenhouse gas emissions. Both of these goals must be achieved while
minimizing public expenses, and ensuring economic prosperity. If you consider a blank piece of land on which
buildings and energy technologies must be located, there are many different ways in
which one could plan it, but only few of which might actually satisfy all stakeholders in
a sustainable way. Here is where interactive optimization methods
can help planners take more informed decisions in early stage urban planning. The search for a preferred neighborhood configuration follows an iterative process, consisting of 3 repeating steps. It begins by generating optimized solutions
according to predefined constraints. In this step, we’re basically asking the
computer to solve the planning question from before. We ask it questions like “Please find the
cheapest urban configuration which has an 85% share of renewable energy” and so on, for
hundreds of cases reflecting the users’ preferences. Second, the user explores the generated data
using 3D maps of the neighborhood, as well as parallel coordinates. By doing so, they can learn about the tradeoffs
between objectives, and define personal constraints or preferences. Third, these preferences are included in the
model, and additional solutions can be generated by the optimization algorithm. This three-step process repeats, until a satisfying
solution is identified, and a decision can be made. Let’s now look a bit closer at the exploration
part in step b, to understand how parallel coordinates work. The following chart is called a parallel coordinates
chart. It allows to easily explore different aspects
about a given object. In our case, it will help us explore the characteristics
of a neighborhood which is being planned. Parallel coordinates consist mainly of two
things: First, polylines – each one representing
a single object, like a neighborhood configuration. And second, the axes – which represent the
different characteristics of the objects. Here we are considering for example the costs
for constructing the neighborhood, its density, how many buildings have a nice view, and so on. The polylines then flow through each axis
and indicate each neighborhood’s characteristics. Let’s use parallel coordinates to learn
about tradeoffs when planning for a sustainable neighborhood, and identify a favorite alternative. In a first step, 15 alternative neighborhood
configurations were generated by optimizing a model of the neighborhood. Each configuration represents the cheapest
solution, for different densities and shares of renewables. Note that the solutions with the higher renewable
energy shares are displayed in green, whereas the less renewable ones are displayed in red. From this broad set of alternatives, we can
narrow down the solution space by selecting only the preferred ranges of interest, for
which additional solutions can be generated. For example, I would like to keep only densities
between 0.75 and 1.75, which are the most realistic for my case. I can do so by “brushing” the corresponding
area on the axis. As we see, only the chosen solutions are displayed,
while the lower density solutions, and the higher ones, are filtered out. This new constraint for density was input
in the model, and 18 new configurations have been calculated for this preferred range of
densities. We notice how new lines have now appeared
on the chart. Exploring the parallel coordinates, the different
indicators can be compared for each scenario. For example, I am interested in the scenarios
that have the highest Landmark View factor, which indicates how many dwellings enjoy a
nice view on the surrounding landmarks. As I select the upper area of the axis, I
quickly realize that a good view means compromising with high density, another criteria I would
like to keep high. So I brush down the view factor, until I reach
an acceptable level of density . I identify two interesting scenarios, which
are acceptable both in view and in density. My preferences for view factors will now be
included in the model. From now on, the optimization process will
only find solutions where at least 28% of building floors have a nice view on the surrounding
scenery. In this third iteration, the optimization found 20 new solutions while respecting the
new constraint on view quality. Adjusting the density level to the satisfying
solutions found previously, I can now also filter the solutions that present the highest
shares of renewable energy, until two interesting solutions remain. I notice that the top solution in terms of
renewable energy, also performs better in density, but is more costly and contains less
open space for green parks than the other. At this stage, if the user is satisfied with
one of the solutions, the decision can be made. Otherwise, the process can continue to generate
more solutions according to new constraints or objectives. At any point, the user can also explore a
given urban configuration in 3D, to get a more realistic feeling of the layout and form,
displaying any spatial information of his or her choice. In this video, we saw how an interactive optimization
method could be used to learn how urban goals interact. Parallel coordinates were in particular useful
to interactively explore the data, and input our preferences into the optimization process. In this way, not only did we become more mindful
about our preferences and existing tradeoffs, but we also identified a cost-optimal solution
that can be included in urban planning processes.

No Comments

Leave a Reply