Refine
Document Type
- Article (2)
Language
- English (2)
Has Fulltext
- yes (2)
Faculty / Organisational entity
Daylight is important for the well-being of humans. Therefore, many office buildings use
large windows and glass facades to let more daylight into office spaces. However, this increases the
chance of glare in office spaces, which results in visual discomfort. Shading systems in buildings
can prevent glare but are not effectively adapted to changing sky conditions and sun position,
thus losing valuable daylight. Moreover, many shading systems are also aesthetically unappealing.
Electrochromic (EC) glass in this regard might be a better alternative, due to its light transmission
properties that can be altered when a voltage is applied. EC glass facilitates zoning and also supports
control of each zone separately. This allows the right amount of daylight at any time of the day.
However, an effective control strategy is still required to efficiently control EC glass. Reinforcement
learning (RL) is a promising control strategy that can learn from rewards and penalties and use this
feedback to adapt to user inputs. We trained a Deep Q learning (DQN) agent on a set of weather data
and visual comfort data, where the agent tries to adapt to the occupant’s feedback while observing
the sun position and radiation at given intervals. The trained DQN agent can avoid bright daylight
and glare scenarios in 97% of the cases and increases the amount of useful daylight up to 90%, thus
significantly reducing the need for artificial lighting.
Solar radiation data is essential for the development of many solar energy applications ranging from thermal collectors to building simulation tools, but its availability is limited, especially the diffuse radiation component. There are several studies aimed at predicting this value, but very few studies cover the generalizability of such models on varying climates. Our study investigates how well these models generalize and also show how to enhance their generalizability on different climates. Since machine learning approaches are known to generalize well, we apply them to truly understand how well they perform on different climates than they are originally trained. Therefore, we trained them on datasets from the U.S. and tested on several European climates. The machine learning model that is developed for U.S. climates not only showed low mean absolute error (MAE) of 23 W/m2, but also generalized very well on European climates with MAE in the range of 20 to 27 W/m2. Further investigation into the factors influencing the generalizability revealed that careful selection of the training data can improve the results significantly