Although buildings are vital in achieving environmental sustainability, they contribute largely to the increase in global energy consumption and carbon emissions. In Hong Kong, a minimum of 44% of the energy consumed is spent on lighting and cooling. Consequently, exploiting solar energy resources is one of the bioclimatic strategies to curb the increasing energy demand. Thus, solar radiation and daylight illuminance data are necessary for cooling load determinations, daylighting evaluations and energy-efficient building designs. Nevertheless, estimation models have become a necessary alternative due to the unavailability of measured solar radiation and daylight data.

Empirical models have been previously proposed for solar radiation and daylight. Similarly, previous works offered classification approaches for the CIE standard skies since solar irradiance and illuminance can be derived from these standard sky types. However, these empirical methods tend to underperform, especially when the model described is non-linear. The present study ameliorated the estimating and classification models for solar radiation, daylight and CIE standard skies classification using Machine learning (ML) techniques. While the scope of this research was limited to Hong Kong.

After the introduction, the first part of this thesis presents a review of bioclimatic architecture, solar radiation, luminous efficacy, ML models and statistical indicators for model evaluations. Findings show that ML is required in luminous efficacy, vertical irradiance modelling, and CIE standard skies classification. Other key findings were the need for further research on irradiance and luminous efficacies of vertical surfaces and adopting the 15 CIE standard sky classification in luminous efficacy modelling.

The next part of the thesis presented findings from analysed recent solar radiation and daylight data measured between 2019 to 2020. The analysed solar radiation data were used to determine the solar-energy potential of BIPV for the building skin (BS) and the four principal building orientations. The findings showed that the annual average horizontal global, diffuse, and direct irradiance values were 291.8, 164.3, and 127.5 W/m²/day, respectively. Similarly, 120, 72, and 107 klux were obtained as the peak global, diffuse, and direct illuminance, respectively. The analysis also revealed a potential for using BIPV on the entire BS in Hong Kong.

The third part of this thesis identified the 15 CIE standard sky types using available meteorological data and SVM. Generally, using the sky luminance distributions to identify the 15 CIE standard skies is the most effective method, but these are sparingly measured. The study addressed the lack of luminance distribution measurement by classifying the 15 CIE standard skies using measured climatic data. Findings showed that the SVM could classify the skies with an accuracy of 72.4% on the training data and 71.4% on the test data.

In the fourth part of this thesis, ML models for estimating vertical global solar radiation (I_vg) were proposed. The vertical direct and reflected irradiance, clearness index (K_t) and scattering angle (χ) were used to develop these models. ML algorithms were also used for variable importance identification. Findings showed that the ratio of direct normal irradiance to global horizontal irradiance, K_t and χ, are essential variables for modelling along sunlit and shaded vertical surfaces. Also, most of the proposed models offered a good estimate of I_vg up to a relative root mean square error (%RMSE) of 20%.

The fifth part of this thesis presents the findings from measured luminous efficacy and luminous efficacy modelling under the 15 CIE standard skies. The luminous efficacy approach provides a method of deriving daylight illuminance from solar irradiance. The study assessed the horizontal luminous efficacy of the global, direct, and diffuse components for the 15 CIE standard skies in Hong Kong. By using an established vertical luminous efficacy model, it also estimated vertical illuminance on the four principal vertical surfaces. The findings of this study showed that constant luminous efficacies could be used for deriving illuminance data. Furthermore, horizontal luminous efficacy ranged from 40 to 190lm/W, indicating that daylight can provide sufficient visibility during working hours. Lastly, the vertical luminous efficacy model offered reasonable estimations of vertical illuminance data.

In the sixth part of the thesis, ML was used in horizontal and vertical luminous efficacy modelling. The methodology explored ML, sensitivity analysis and empirical approaches. For the horizontal model, twelve (12) diffuse and global luminous efficacy models were proposed. These models comprised six artificial neural networks (ANN) and six empirical models. Similarly, ANN, support vector machines (SVM), and empirical correlations were proposed for vertical efficacy. Findings show that the diffuse fraction and scattering angle are crucial in horizontal and vertical luminous efficacy, respectively. Also, all proposed models could offer acceptable predictions of daylight with peak %RMSE not exceeding 20%.

The last part of this thesis presents the building energy applications of the analysed solar radiation and daylight data in terms of lighting, cooling and semi-transparent BIPV output. The findings showed that the increase in WWR and SC increased the solar-heat gain and cooling load of the analysed case building. Furthermore, the use of daylighting control caused an increase in energy savings. Importantly, semi-transparent BIPV façades with a large window-to-wall ratio (WWR) of 80% can provide an overall energy benefit of up to 7126 kWh pa. The analysis established that semi-transparent BIPV is an alternative fenestration system for energy-efficient building designs.