The difficulty in meeting the CIBSE TM52 criteria using the new enhanced DSYs

MJ Wood and ME Eames

2016-10-10

Introduction

Preventing summertime overheating within buildings is important for the comfort of the occupants and to minimise any cooling requirements. There are many different heating criteria and many building designs that have the potential to meet them. Since buildings are typically designed using computer models, the final designs are affected by the weather inputs and the overheating criteria.

Aims

The aims of this research are to %%%%%%.

Background

Overheating Criteria

CIBSE TM52 (Nicol 2013Nicol, F. 2013. “The Limits of Thermal Comfort: Avoiding Overheating in European Buildings: CIBSE TM52, 2013.” CIBSE.) was released in 2013 and introduces three overheating criteria:

  1. The number of hours where the internal operative temperature is above the maximum acceptable temperature (He)1 Where He stands for hours of exceedance
  2. The daily weighted exceedance (We).
  3. The maximum operative temperature (Tupper)

These criteria are intended to minimise overheating over a broad range of possible overheating events.

The derivation of \(\Delta T\)

He stands for hours of exceendance, where the exceedance is measured by \(\Delta T\), rounded to the nearest degree2 note that \(\Delta T\) is measured on each timestep, rather than over a whole year, as He is.. The exceedance \(\Delta T\) is the number of hours where the internal operative temperature Top is above the maximum acceptable temperature Tcomf. The internal operative temperature Top, is defined as:

\[ T_{op} = \frac{T_a - T_r}{2} \]

Ta is the mean air temperature; and Tr is the mean radiant temperature

The difference between the maximum acceptable temperature and the operative temperature is defined as \(\Delta T\), where:

\[ \Delta T = T_{op} - T_{max}\]

Tmax is maximum acceptable internal temperature.

This equation requires a further definition, that of \(T_{max}\).

\(T_{max}\) is the definition of the maximum acceptable temperature. However, it has been shown that the ‘accpetable temperatue’ is dependent on recent external temperature trends. This is due to temperature adaptation. So, in order to define, \(T_{max}\), the running mean temperature is used:

\[ T_{max} = 0.33 T_{rm} + 21.8 \]

The running mean Trm is defined as:

\[ T_{rm} = (1-\alpha) T_{od-1} + \alpha T_{rm-1} \]

Tod − 1 is the outdoor daily mean temperature for the previous day; Trm − 1 is the running mean temperature for the previous day; and α is an empirically derived coefficient which typically takes the value 0.8.

These equations can be used to derive \(\Delta T\) for each timestep. Once \(\Delta T\) is known, the CIBSE criteria can then be calculated.

Definitions of the three CIBSE Criteria

The hours of exceedance He is defined as the number of hours where the \(\Delta T\) is greater than 1 degree. He should be less than 3% of occupied hours for the period between 1st May until the 31st September.

The second criteria is the daily weighted exceedance, We. Although this is defined as the weighted exceedance, it best thought of as the cumulative exceedance. This is best explained visually (figure 1).

MPG vs horsepower, colored by transmission. MPG vs horsepower, colored by transmission.

The weighted exceedance for a given day is the equivalent to total of the grey areas shown in figure 1. To comply with the second criteria, the total daily weighted exceedance \(W_e\) should not exceed 6.

The third criteria is simple; the maximum value of \(\Delta T\) should be no greater than 4.

Aims of our assessment

Our aim is to examine how easy it is to meet each of the CIBSE TM 52 criteria, both individually and as a set. We do this by looking at the percentage of buildings that pass each criteria, and the percentage of those that pass all the criteria.

Method

We do this by considering the input space of a building model. A simple example of the input space is shown in figure 2.

MPG vs horsepower, colored by transmission. MPG vs horsepower, colored by transmission.

Each of the two variables \(x_1\) and \(x_2\) represent an input to the computer model3 For mathematical ease, these value have been normalised to between 0 and 1. The input space is the whole area where both \(x_1\) and \(x_2\) are between 0 and 1. (i.e. within the area of interest.) In most computer experiments, we do not know where, or how big, the pass area is. One method of determining its size is to use Monte Carlo (MC) methods.

Using the MC approach, we take random samples across the input space. For each of these samples, we run the computer model. If we have a pass, we count this as a pass, if we have a fail, we count it as a fail. The percentage pass rate \(P\) is therefore approximated as \(P^\prime\):

\[ P^\prime_{\text{rate}} \% = \frac{\text{Count pass}}{\text{Count pass} + \text{Count fail}} \]

As the number of MC sample increases, \(P\) will tend towards \(P^\prime\). The number of MC samples required to achieve a good apprximation of \(P\) is dependend on the both the size of the pass space relative to the input space4 A small pass spaces will need more samples to generate the required accuracy and the number of input dimensions in the model. We therefore encouter a number of challenges in using the MC approach.

The main challenge for the MC approach is simply the number of samples it requires. In building optimisation problems, there are usually a lot of building parameters that can be changed. This means that the input space to the model is large, and therefore requires a lot of MC samples. It can typically take many hours and even days to conduct this type of analysis. A typical MC analysis might require 10,000 samples. For a building simulator that takes 1 minute to run, this MC analysis would take nearly 7 days to complete. If we want to perform this analysis on 100s of buildings under different weather conditions, the simulation time could quickly balloon to many years. We therefore need an alternative approach.

Our approach to MC analysis

Our approach to the MC analysis is two-fold. First we reduce the number of inputs to the model using a sensitivity analysis to remove inputs that are not important. We then create an emulator of the building model on which we peform the sensitivity analysis.

We use an emulator in both the sensitivity analysis and the subsequent MC analysis. An emulator is a surrogate for the building model. The emulator we create is based on a Kriging model of the simulator (i.e. building model) output.

The building model being considered

However, there are a number of challenges with this approach.

out solution to this…

Terminology

MC - Monte Carlo (sampling method)