Performing numerous simulations of a building component, for example to assess its hygrothermal performance with consideration of multiple uncertain input parameters, can easily become computationally inhibitive. To solve this issue, the hygrothermal model can be replaced by a metamodel, a much simpler mathematical model which mimics the original model with a strongly reduced calculation time. In this paper, convolutional neural networks predicting the hygrothermal time series (e.g., temperature, relative humidity, moisture content) are used to that aim. A strategy is presented to optimise the networks’ hyper-parameters, using the Grey-Wolf Optimiser algorithm. Based on this optimisation, some hyper-parameters were found to have a significant impact on the prediction performance, whereas others were less important. In this paper, this approach is applied to the hygrothermal response of a massive masonry wall, for which the prediction performance and the training time were evaluated. The outcomes show that, with well-tuned hyper-parameter settings, convolutional neural networks are able to capture the complex patterns of the hygrothermal response accurately and are thus well-suited to replace time-consuming standard hygrothermal models.

When simulating the hygrothermal behaviour of a building component, there are many inherently uncertain parameters. A probabilistic evaluation takes these uncertainties into account, allowing a more dependable assessment of the hygrothermal behaviour. However, this often necessitates many Monte Carlo simulations, which easily become computationally inhibitive. To overcome this time-expense problem, the hygrothermal model can be replaced by a metamodel, a much simpler mathematical model which aims at mimicking the original model with a strongly reduced calculation time. In this paper, a metamodel is developed to directly predict hygrothermal time series (e.g. temperature, relative humidity, moisture content), rather than singlevalued derived performance indicators (e.g. maximum mould index), as these hygrothermal time series yield more information, and also allow the user to post-process the output as desired. So far, no metamodelling strategies able to tackle time series are available in the field of building physics. Because the hygrothermal response of a building component is highly non-linear and transient, this paper focuses on neural networks for time series, as they have proven successful in many other fields. The performance and training time of three popular types of networks (multilayer perceptron, recurrent neural network, convolutional neural network) is evaluated based on an application example of a massive masonry wall. The results indicate that only the recurrent and convolutional networks are able to capture the complex patterns of the hygrothermal response. Additionally, the convolutional network performed significantly better and was 10 times faster to train for the current application example, compared to the recurrent network.

No abstract

Performing a probabilistic assessment of a building component can easily become computationally inhibitive.To solve this issue, the hygrothermal model can be replaced by a metamodel, which mimics the original model with a strongly reduced calculation time. In this paper, convolutional neural networks are used to predict hygrothermal performance. Because neural networks do not extrapolate well outside their training subspace, it is important to select the training data wisely so that the network can be used to predict for a wide variety of cases, while keeping training time as low as possible. The impact of a reduced training subspace is investigated by training a network on a limited number of wall types or exterior climates and evaluate its prediction accuracy for different wall geometries or other climates. The results showed that is indeed possible to train on a well-considered reduced subspace, while maintaining high accuracy, though it does not necessarily save training time.

No abstract