Previous (The version handling system)   |   Start   |   Next (Graphical scripting)

Supporting tools for experts

Parametric runs and optimization

Performing a simulation with a given, fixed model of a building can give valuable insights concerning the operation and performance of a building. However, there may also be a need to compare different variants of the same model, for example to examine the impact of using different building materials, heating units, or plant control strategies. Such analysis is often done manually by the engineer, but in some situations, it is beneficial to automate this process. IDA Parametric runs is a tool in IDA ICE developed for this purpose, which gives the user an interface to flexibly define permissible model modifications and systematically analyze the effects of those modifications in a customizable way. When the number of possible models increases, optimization can guide this analysis and identify the model configurations that are most relevant based on user-defined objectives. There are two stages in the building design and operation process when this can be particularly useful.

In the Pre-design phase, when the design is not fully defined, parametric runs and optimization can automate the exploration of different design option and lets the user investigate how different parameters affects building performance. Here the goal is to optimize the building.

In the Operational phase, comparisons between simulation results and measured signals can be used to investigate differences between the model and the actual building. Here parameter estimation lets the user calibrate the model to fit the data, i.e. the goal is to optimize the model.

Parametric runs

IDA Parametric runs consists of the following tools, which all provide information on how selected outputs are affected by changes to user-defined model parameters by running a series of simulations:

• Sensitivity analysis: lets the user inspect the sensitivity with respect to each parameter individually.
• Monte-Carlo: runs a series of simulations where the parameter values are chosen randomly.
• Brute force: runs all possible parameter combinations. This is only feasible if both the number of parameters and the number of values they can attain is low, as the number of simulations otherwise is prohibitively high.
• Optimization: uses the tool AutoMOO to iteratively run simulations and calculate parameter values with the goal of optimizing user-defined objectives.

AutoMOO

The optimization problems occurring in the pre-design and operational phases outlined above share the main characteristic that the function evaluations are computationally expensive, since each evaluation consists of running a full simulation. Apart from this, the characteristics vary:

• There can be one or multiple (generally conflicting) objectives, for example energy use versus indoor comfort,
• There can be both continuous parameters (for example window size), and discrete ones (for example window type),
• There may or may not be constraints, for example, ensuring that daylight demands are satisfied.

AutoMOO is designed with these properties in mind. To handle the varying characteristics of the optimization problems, AutoMOO uses a collection of different algorithms, which excel under different conditions and are selected dynamically based on the performance on the given problem. To handle the expensive function evaluations, these algorithms are primarily based on Bayesian optimization, which is the go-to method for optimization of expensive-to-evaluate black-box functions in engineering applications today. Bayesian optimization is based on optimizing surrogate models which approximate the true objective functions probabilistically but are much faster to evaluate.

Previous (The version handling system)   |   Start   |   Next (Graphical scripting)