Uncertainty and Sensitivity Analysis - Tips and Tricks

Speeding up Uncertainty and Sensitivity Analysis

Uncertainty and sensitivity analysis studies require hundreds or even thousands of simulations to be carried out and so can be extremely time/calculation intensive. To help speed simulations, you could consider using one or more of these techniques:

1. Analyse a single representative zone in the building and apply findings to whole building.
2. Use a representative simulation period instead of the whole annual simulation. For example, in many climates using the first or second 6 months of the year would be a good approximation to use of the full 12-month simulation. Alternatively, if the analysis is mainly related to summer overheating then you might choose a shorter representative summer period for the simulations.
3. In most cases, the more variables that are defined in the analysis, the more runs will be required to achieve more acceptance results. Therefore, instead of trying a multitude of runs, a preliminary analysis using a more efficient sampling method like LHS can be used to eliminate the least influential variables and reduce the parameter space before doing a detailed run.
4. Use JEPlus or a network simulation server to maximise the number of simulations that are run in parallel at one time. If you have a simulation server with many cores at your disposal then you might consider increasing both the population size and Number of job per batch which allows more cases to be analysed at once.

See also Working with Large Models.

Running Sensitivity Analysis as a Precursor to Optimisation

Complex optimisation with a large number of design variables and options makes it difficult to find clearly distinguishable trends. Also, some variables might not have a significant influence on the output can make the assessment of the relative importance of other important variables difficult. Sensitivity Analysis determines which are the most influential Design Variables, and what their relationship to the output is.

Sensitivity Analysis can be used as a precursor to optimisation to identify and remove the less influential variables from the optimisation runs, thereby reducing the complexity and overall optimisation runtime. Also, If the variable is influential for only one of the objectives, its value can be predetermined with SA and removed from the optimisation. With a focus only on the important variables, the optimisation results are cleaner, quicker and easily understandable. Another advantage of using SA, is that it helps you to understand any trends associated with design variables and makes it easier to see these in the results. For example, SA can tell us the relationship of each Design variables with the Objectives and whether they are directly or inversely correlated, i.e. whether an increase in the design variable leads to an increase or decrease in the objective’s value.

However, there are certain considerations should be kept in mind because unlike optimisation, the idea of Sensitivity Analysis is to understand the relationship between the Design Variables and the Objectives and not to find the best set of designs. In Sensitivity Analysis two important things need to be done in variable definition which will typically be different from optimisation:

1. The order in which design variable options are set up is very important. The design variable options need to be ordered in an in increasing or decreasing value. For example, in a case where wall constructions with different U-Values is one of the design variables and heating energy and capital cost are the two outputs, the overall U-value of the wall affects heating energy, and the wall construction cost affects building capital cost. Therefore, if the U-value of a wall is changed by using different materials, then the wall options sequence when ordered in decreasing U-value, might not correspond with the one created that will be created with increasing capital cost. In this case, the Sensitivity Analysis must be done separately for both the outputs.
   
   
2. The design options selected for the variable should not contain options that have the same value (i.e. duplicates) with respect to the target objectives. For example, in the same case as above there might be two wall constructions that have the same U-value or overall cost. In that case wall construction options that have the same U-value need to be removed from Sensitivity Analysis for heating energy, and wall construction options with the same Cost need to be removed from the Sensitivity Analysis for capital cost.

Therefore, despite the theoretical possibility of doing Sensitivity Analysis for more than one output at a time, DesignBuilder recommends, as a good practice, that sensitivity analysis of each output is set up separately and as a precursor to Optimisation. In each output it is important that the variable options in list type variables have a correct sequence so that SA can find the impact of a progressive change of one option to another on the SA output.

This process is demonstrated in more detail in our Uncertainty and Sensitivity Analysis: Simplify and Speed up Optimisation Studies webinar.

Ensure inheritance path

It is important to ensure that the variations defined by the uncertainty and sensitivity analysis Design Variables will have the intended effect. Uncertainty and sensitivity analysis works by applying variations to a copy of the base model and these changes are applied to the object defined in the Variable. You must take into account how this data will be inherited to down to any child blocks, zones etc to ensure you are testing the design variants you think you are. For example, if the Window to wall % was hard set at block level then changes in this data at building level would not influence the surfaces of the design variants in that block and the results will indicate that the Window to wall % does not influence results.

Switching on Shading when Shading System Type is a Variable

Design variables such as Local shading type and Window shading will only take effect if the corresponding checkbox activating these options is checked. A common mistake for new users is to request these variables without switching on the option.