The TUDelft spectrally resolved albedometer is a measurement device whose design began in 2019 and aims to accurately reconstruct spectral irradiance and spectral albedo using reconstruction techniques which enable this device to be produced at a lower cost than its competitors.
The measurement of spectral albedo using an albedometer device is a crucial component in predicting energy yield for bifacial PV panels, expected to become the dominant photovoltaic technology by market share in 2030 [13]. Building on the work of previous thesis projects at TUDelft, this thesis focuses on the following three topics of improvement:

Accuracy improvement in spectral reconstruction: The albedometer device is recalibrated according to the recently recalibrated EKO device, succeeding in reducing the error uncertainty in the first and last wavelength ranges of the spectral irradiance reconstruction. Whilst the average errors lie outside of acceptable uncertainty bounds, the systematic errors in the first and last wavelength bands are identified and proposed solutions involve adjusting the PSO algorithm and using machine learning to improve the prediction of atmospheric absorption parameters such as total precipitable water.
Spectral albedo reconstruction using Machine Learning: Machine learning techniques are employed to reconstruct down-facing spectral irradiance, achieving errors below ±5% for various sky classes and demonstrating the method’s potential for spectral albedo reconstruction in future work.
Improving device usability: The Albedometer App centralizes and automates data processing code, simplifying spectral irradiance and albedo reconstruction processes and greatly enhancing user experience.

This research is key for the development for the albedometer’s accuracy, functionality and usability. By integrating the model for spectral albedo reconstruction this thesis advances the overall development of the albedometer device, bring it one step closer to realising its full potential as a high accuracy, low cost measurement device, making it a valuable tool for spectral albedo reconstructions and precise energy yield predictions in the bifacial PV sector.

Bifacial modules are expected to have a 60% share in the PV market, which require spectral irradiance data incident on both sides to optimize the design. Sensors that provide spectrally resolved data are expensive and thus a cost-effective spectrally resolved albedometer was designed in two previous PVMD Master Theses. This albedometer measures irradiance in three wavelength bands. The focus of this research is on developing a model that reconstructs the spectral irradiance incident on the top part of the albedometer.

The spectral simulation program SMARTS was converted to MATLAB, after which a Particle Swarm Opimization Algorithm was added to determine values for unknown parameters. The model was first tested against spectra generated by SMARTS. After which both the original model and an model with additional modification coefficients was tested with clear-sky measurements from the albedometer and validated against an EKO spectroradiometer.

The original model proved to fail in the first wavelength band, which caused the other wavelength bands to be reconstructed poorly as well. The adjusted model performed close to the performance of the EKO for clear-sky irradiances above 400 J/Z2 in the total albedometer region, 320-1100 nm.

Short-term solar forecasting is crucial for large scale implementation of solar energy and plays an important role in grid balancing, energy trading, and power plant operation. Cloud movement is the main source of unpredictability within solar forecasting and can be recorded using All-Sky Imagers. Conventional cloud modelling methods using image analysis techniques are unable to extract the spatial configuration and the temporal dynamics of clouds, resulting in poor predictions of the interaction with solar radiation. The goal of this study is to create a deep learning model for short-term irradiance forecasting between 0 and 21 minutes into the future using all sky images combined with auxiliary data. The model performance was assessed by comparing the deep learning model with the persistence model and showed that the deep learning model outperforms the persistence model with 24.8%. A sensitivity analysis to data usage is performed showing that besides using more data, also the variation of using multiple years of data results in better performance. Furthermore, the sensitivity of the model to input variables is assessed, showing that using the clear sky irradiance as input improves model performance with 16% and that meteorological data does not improve performance. Additionally, the model performance was evaluated during different sky conditions showing that the deep learning model outperforms the persistence model for all sky conditions, except overcast conditions. An example of the model behavior is extensively described, showing that the deep learning model tends to predict the trend of the irradiance fluctuations rather than the actual fluctuations. Next to that is in this study shown that the current deep learning model occasional miss important weather events, like obscuration of the Sun, resulting in large irradiance prediction errors. A pathway for future improvements for deep learning models to forecast the short-term irradiance is provided.