Talk A meta-heuristic search platform for evolutionary algorithm optimization of building energy performance

Abstract

Buildings are a major consumer of energy, and building energy savings are a major concern for policy makers, developers, owners, and operators. Given the complexity of building design, simulation tools are becoming common in energy efficient design of buildings in the domains of envelope, systems, and controls. Within such simulation based studies, it is often desirable to investigate the effect of many variables in performance metric objectives. This situation leads naturally to the field of multi-objective optimization. Furthermore, given the nonlinearity, sparsity, and representation of the variables, population based meta-heuristics show promise in efficiently and robustly searching the mapping from design space 'D' to objective space 'Z'.

The overall objective of this work is to investigate and compare competing algorithms in order to recommend the best performing method in building energy optimization.

In this context, Python and the many extension libraries is well positioned to perform key steps in a simulation based optimization;
1. Pre-processing and generation of new simulation runs
2. Execution and resource handling of populations of simultaneous simulations
3. Post-processing and presentation of results

A cursory review of several related Python projects will be presented including; Pyevolve [1], Dexen [2], and Deap [3], as well as more domain specific but

non-python projects such as GenOpt [4]. An object oriented framework developed by the for handling Evolutionary Algorithms and meta-heuristics for simulation based optimization will be presented. This framework will lead to a platform fulfilling several performance objectives;

1. Scaling to large population sizes, and parallel execution on the cloud
2. Generalization to alternative optimization methods, from standard hill climbing to particle swarm optimization, etc.
3. Generalization to alternative domain simulation software packages (TRNSYS, EnergyPlus, etc.)

Initial case study results will be presented to illustrate the methodology and the overall performance metrics. Finally, a roadmap will be sketched, in order to answer the following question faced by green building designers; "Given a building design and a large set of possible building variations, can we find an optimally performing building by simulation in a reasonably amount of time?".

[1] http://pypi.python.org/pypi/Pyevolve/0.5
[2] http://pypi.python.org/pypi/deap/0.8.1
[3] http://pypi.python.org/pypi/dexen/0.3
[4] http://simulationresearch.lbl.gov/GO/

tagged by
no related entity