Development of Comprehensive Solar Radiation and Daylight Illuminance Datasets by Ameliorating the Prediction Models and Categorizing Standard Skies Using Machine Learning Techniques

Description

In most cities, electricity is largely used to create thermally and visually comfortable built environments. Air-conditioning and electric lighting represent about 80% of the total electricity used in commercial buildings, while solar radiation through fenestration provides the majority of a building envelope’s cooling load. Solar irradiance data are a prerequisite for designing solar electric and solar thermal facilities. Data about daylight illuminance and solar irradiance on vertical surfaces are in great demand, particularly for high-rise curtain wall buildings. Long-term data measurement is the most effective and accurate way of setting up such databases, whereas short time-interval data would be more appropriate for examining dynamic variations. However, most places do not measure solar data, or offer only daily global components covering limited periods. Locally, the Hong Kong Observatory routinely records hourly horizontal global solar irradiance data, while hourly diffuse and direct components measurements only began in August 2008 with no measurements of outdoor illuminance. Because it is impracticable to install pyranometers and illuminance meters at every orientation and tilted angle to collect the required data, necessary components are commonly estimated from latitude–related data, such as solar position and climate-related parameters including air-temperature, cloud cover and sunshine hours. Regression techniques are usually adopted to form the empirical models, but these underperform for modelling nonlinear correlations and may provide inaccurate outcomes. To overcome some of these difficulties, the International Commission on Illumination (CIE) has adopted a list of 15 standard skies. Each standard sky denotes a unique, well-defined sky radiance and luminance pattern expressed by mathematical equations that can use to compute solar irradiance and daylight illuminance on inclined surfaces and variously oriented vertical planes. At issue is whether this approach can correctly classify sky conditions. In recent years, machine learning (ML) algorithms have been used for similar purposes. Such approaches extract knowledge from databases and build up prediction models that correlate target output with the accessible inputs. These techniques can be complicated to express without a physical basis, but more accurate than those using regression analysis; moreover, skies can be interpreted by appropriate meteorological data. Another benefit is that individual input variables in the prediction models can be identified for modelling multivariable regressions. The proposed project will use ML techniques for analysing active solar energy applications and passive building designs to improve upon solar irradiance and daylight illuminance prediction models and the categorisation of standard skies.