Advancing Thermal Management in Buildings Globally: Machine Learning-Optimized Dynamic Window Materials

Dynamic window materials represent a significant advancement in energy-efficient building technology, offering a promising approach to intelligent thermal management and energy conversion in architectural applications. Leading areas of interest in dynamic window research include hydrogels, passive radiative cooling materials, dynamic devices based on reversible metal electrodeposition. These cutting-edge materials are capable of dynamically altering their optical and thermal properties in response to external environmental variations, thereby efficiently managing light and heat transfer within buildings. However, dynamic window materials face significant challenges in their development and application. First and foremost, these materials are often optimized based on specific weather conditions rather than a comprehensive, global spectrum. Achieving continuous optimization on a global scale proves to be difficult due to the immense volume of calculations required. This lack of universal optimization may limit their efficiency across diverse climate zones. Furthermore, there is a noticeable gap in the presence of robust indicators to quantify the dynamic performance of these materials. This void in performance metrics poses a challenge, as without clear quantification, it becomes difficult to determine the true value of these windows, especially since their dynamic features may be superfluous in certain climatic conditions.<br/>To address these issues, we proposed novel indicators to evaluate the necessity level of using dynamic windows in buildings for different climates. We use thermochromic (TC) materials as examples to demonstrate this concept for passive dynamic windows. Through an unprecedented global-scale analysis, we executed more than 2.8 million simulations across over two thousand global locations covering most of the Köppen climate classifications in the world. A typical commercial office space, which is adapted from Department of Energy's prototype building model, is used for EnergyPlus simulation in this study. Beyond typical office space, whole commercial and residential buildings are also used for analysis. World heatmaps of indicators for dynamic TC windows have been obtained by artificial neural networks (ANNs) trained by data from various conditions: differing material properties, window configurations, building and environmental circumstances. The ANN inputs include geographic (latitude) and weather (solar radiation and air temperature) data that can be extracted from weather files and is publicly available as the information in a world-map format. The good linear regressions between target and prediction indicate that the performance indicators can be well predicted from the inputs through ANNs, which are optimized by a hyperparameter tuning process. The heatmaps of indicators obtained from ANNs reveal that TC windows with optimal transition temperatures lack dynamism in most of low-latitude tropical regions, functioning similarly to static windows in terms of energy savings. Seasonal dynamics of TC windows appear more important for energy saving compared with daily fluctuations. Surprisingly, optimal transition temperatures correlate more with clear-state solar transmittance than with dark-state. This insight is unprecedented in existing literature.<br/>To assist researchers in optimizing and evaluating TC materials and windows, we provide an open-source Python tool (https://github.com/LBNL-ETA/PyDynamicWindow, a private repository, will be published soon) to quickly obtain a world heatmap of desired indices or optimal properties within seconds by simply inputting the fixed material properties. Besides thermochromic, other dynamic window materials will also be included in this Python tool in the future. With this powerful tool, researchers can identify climate conditions for dynamic windows and improve their materials and products for target climates.