Optimizing Machine Learning Model Selection for Perovskite Solar Cells Datasets

Written by Eugene R.

Introduction

In recent years, the energy industry has witnessed a significant shift on the global platform. With climate change affecting different geographic territories, this shift has been driven by the urgent need to address the adverse effects of climate change. To achieve sustainable energy development goals, several countries are now changing their global energy landscape and finding ways to utilize natural resources such as solar energy (Djellouli et al., 2022). This need to reduce dependency on fossil fuels and greenhouse gas emissions comes with multiple opportunities and challenges. For example, solar energy is considered one of the most efficient renewable energy resources. However, to achieve its complete potential, it is important to overcome challenges associated with efficiency, reliability, and integration of solar systems (Zhao et al., 2022).

With the advancement in technology, machine learning (ML) has emerged as a powerful tool that can be used to solve challenges associated with the implementation of solar energy systems. ML is known for its ability to evaluate large datasets, identify patterns, and make accurate predictions (Jha et al., 2017). Its use may suggest unique strategies that can enhance the efficiency of solar energy systems. Similarly, perovskite solar cells have emerged as a promising technology in the field of photovoltaics. These cells are not only flexible but also known for their potential to offer higher efficiency at lower manufacturing costs. Several studies have identified perovskite solar cells as a potential game-changer in the solar industry (Shiogai et al., 2020; Tao et al., 2021). However, the development and optimization of perovskite solar cells remains challenging due to the complexities involved in their performance, material composition, fabrication processes, and device architecture. Salah et al. (2023) noted that optimizing perovskite solar cells can be a time-consuming and resource-intensive process. We propose that ML can be a powerful tool in accelerating the optimization of perovskite solar cells.

Research Aim and Objectives

The role of different ML models in enhancing the efficiency of perovskite solar cell datasets has been studied by both academic and industry experts (Shiogai et al., 2020; Tao et al., 2021). However, there is a lack of comprehensive studies that provide an in-depth comparison of the performance of different ML algorithms across various perovskite datasets. As a result, this research aims to address this gap by systematically evaluating and comparing different ML models used for predicting perovskite solar cell performance. The research also focuses on understanding the difficulties encountered in integrating ML models in perovskite solar cells.

The research has the following objectives:

• To compile and process different perovskite solar cell datasets from literature and experimental sources.
• To implement and evaluate various ML models, such as regressions, decision trees, random forests, support vector machines (SVM), and artificial neural networks.
• To identify the challenges associated with integrating ML models with perovskite solar cells.
• To develop a framework for selecting the most appropriate ML model based on dataset characteristics and prediction tasks.

By leveraging important ML techniques, this research can contribute to the existing pool of knowledge. The outcomes of this study can further help renewable energy experts in making more efficient and data-driven decisions for the global adoption of solar energy systems.

Literature Review

Recently, there has been a significant rise in the generation and use of solar energy all over the world. Enhancing solar energy infrastructure can have multiple advantages for countries, as well as the environment. The application of ML in the solar energy sector has gained considerable attention in recent years. Gandhi et al. (2023) noted that several countries are now looking for optimized approaches to generate solar energy through intelligent techniques. As forecasting the future need of solar energy is one of the first steps in developing a solar energy system, experts are experimenting with technology to meet their requirements. With ML being one of the proven technologies used for data analysis and prediction, it can play a crucial role in the solar energy transformation (Subramanian et al., 2023). The study noted that ML can be used for continuous monitoring of solar cells, modules, and panels. With a combination of deep learning and big data analytics, data can be collected and interpreted to make accurate predictions. Vélez et al. (2024) demonstrated that ML algorithms can be used to quantify linear correlation between synthesis descriptors of perovskite solar cells.

Perovskite materials are widely preferred in multiple scientific fields for their composition diversity. These materials are easily available and possess multiple beneficial properties for the solar application (Shiogai et al., 2020). Li et al. (2019) explored the role of ML in optimizing material composition, developing design strategies, and predicting the performance of perovskite solar cells. Based on 333 data points, new perovskite compositions were synthesized to test the predictability of the model. The study observed that the ML model can be used to predict the underlying phenomena and the performance of perovskite solar cells. In another study conducted by Eibeck et al. (2021), it was observed that baseline ML models perform better on the computational dataset as compared to neural-based models. The study recommended the need for further research to enhance the ability of ML models to predict power conversion efficiency of perovskite cells in well-controlled conditions.

Furthermore, the use of ML in using perovskite material for developing solar cells was studied by Tao et al. (2021). While the study highlighted the potential of using ML in perovskite material for the development of solar cells, it also indicated the need for further research. On the other hand, Salah et al. (2023) provided a comprehensive analysis of the application of ML models for studying complex datasets in the field of solar cell power conversion efficiency. The study indicated the need for further analysis to identify appropriate model selection based on different requirements and selection criteria. Similarly, Li et al. (2021) noted that ML methods have shown remarkable achievements in predicting the basic performance of perovskite materials. However, the unique structural diversity and compositional flexibility of perovskites make it challenging to construct a comprehensive model. This limits the prediction accuracy of ML, and the model may miss out on some fundamental physical information.

Apart from studying the composition and performance prediction of perovskite cells, Zhang et al. (2023) examined the process optimization of perovskite cells using ML. The study used a combination of genetic algorithms and neural networks to optimize the spin-coating parameters for perovskite film deposition. This resulted in a 15% improvement in the film quality of perovskite cells. In another study, Chen et al. (2023) noted that the development of perovskite solar cells is a complex and slow process. To get an ideal cell, manufacturers need to experiment with multiple artisanal samples. This process can be improved by automating the number of consecutive steps in the research process. The study also identified lead halide perovskites as a favourable material for solar cell applications. However, Asghar et al. (2017) argued about the toxicity of lead that can lead to long-term operational instability of solar cells. As a result, it is important to find material that can be advantageous for perovskite solar applications. With appropriate ML models, large datasets demonstrating the use of different materials used in perovskite cells can be analyzed, and a suitable material can be found based on other conditions.

Although these studies demonstrate a successful implementation of ML models in perovskite solar cells, several gaps have been identified in the existing literature. For example, the above studies focus on a single ML algorithm and have limited data for comparing different ML models. This makes it difficult to evaluate the comprehensive application of ML on various models and datasets. Similarly, there is limited information on the interpretability of ML models in connection to perovskite solar cells. Detailed research is essential to facilitate the adoption of ML in the development of perovskite solar cells. Additionally, the impact of different dataset features such as size, diversity, and noise levels needs to be systematically studied to understand model performance for perovskite solar cell applications. This research aims to address these gaps by providing a comprehensive framework for selecting and interpreting ML models for perovskite solar cell datasets.

Research Methodology

Research methods in ML play a crucial role, as the accuracy and reliability of the results are influenced by the research methods used. The nature of the study indicates that the majority of the data used for this research will be quantitative. Thus, a standardized approach will be used for an experimental research design. This includes data collection and preprocessing, model selection, model testing, and model evaluation.

Data collection and preprocessing

To get a better understanding of perovskite solar cell performance, data will be collected from academic journals, books, and online articles. This will provide diversity in dataset characteristics and also give an insight into the different features of perovskite solar cells, such as material composition, fabrication methods, and device architectures. Furthermore, the collected data will be cleaned to handle missing values and avoid any inconsistencies. Preprocessing allows for the identification of key features of datasets and ensure in proper labelling of data (Kamiri & Mariga, 2021). This will further normalize data features to ensure comparability across different datasets. Data analysis is another important aspect of studying the use of ML in perovskite solar cells. It will involve analyzing data to identify patterns, trends, and relationships between earlier studies and their application (Sharifani & Amini 2023).

Model selection

The optimal use of model evaluation, model selection, and algorithm selection plays an important role in developing an appropriate ML model for optimizing perovskite cell datasets (Raschka, 2018). Common methods for model selection include linear regression, decision trees, random forests, support vector machines, gradient boosting machines, and artificial neural networks. After the model section, k-fold cross-validation will be employed to ensure robust performance estimation. In addition to this, ML models will be evaluated using different metrics, such as mean absolute error (MAE), root mean square error (RMSE), R², and domain-specific metrics like perovskite charging efficiency prediction accuracy. The impact of different dataset characteristics will be further analyzed based on model performance.

Model testing and evaluation

A comprehensive analysis of model performance across different datasets will be conducted. This will help in identifying patterns and trends in model performance related to dataset characteristics. Furthermore, a flowchart will be developed to guide model selection based on dataset properties and prediction tasks. The developed framework will later be tested on different perovskite solar cell datasets. The framework will be further refined based on the validation results and expert feedback (Kamiri & Mariga, 2021). Various tools, including programming languages such as Python and R, ML libraries such as TensorFlow and Keras, and statistical packages will be used for accurate model evaluation and selection.

Conclusion

This research proposal provides a comprehensive plan to address challenges faced by researchers in selecting appropriate ML models for perovskite solar cell datasets. This research will systematically evaluate various ML algorithms across diverse datasets. Based on the collected data, a robust framework for model selection and interpretation will be developed. The research aims to offer a comprehensive comparison of different ML models across perovskite solar cell datasets. It will provide an insight into the strengths and limitations of different ML algorithms used for optimizing perovskite solar cell datasets. The research will also enhance the understanding of key factors influencing the performance of perovskite solar cell. By analyzing multiple ML models and datasets, it will facilitate an efficient use of ML in perovskite research. Apart from filling the gaps in the existing literature, this research has the potential to significantly advance the application of ML in perovskite solar cell research. The outcomes of this study can further help in establishing best practices for applying ML in material science, especially in complex systems like perovskite solar cells.

References

Asghar, M. I., Zhang, J., Wang, H., & Lund, P. D. (2017). Device stability of perovskite solar cells–A review. Renewable and Sustainable Energy Reviews, 77, 131-146 https://doi.org/10.1016/j.rser.2017.04.003

Chen, C., Maqsood, A., & Jacobsson, T. J. (2023). The role of machine learning in perovskite solar cell research. Journal of Alloys and Compounds, 960, 170824 https://doi.org/10.1016/j.jallcom.2023.170824

Djellouli, N., Abdelli, L., Elheddad, M., Ahmed, R., & Mahmood, H. (2022). The effects of non-renewable energy, renewable energy, economic growth, and foreign direct investment on the sustainability of African countries. Renewable Energy, 183, 676-686 https://doi.org/10.1016/j.renene.2021.10.066

Eibeck, A., Nurkowski, D., Menon, A., Bai, J., Wu, J., Zhou, L., & Kraft, M. (2021). Predicting power conversion efficiency of organic photovoltaics: models and data analysis. ACS omega, 6(37), 23764-23775 https://doi.org/10.1021%2Facsomega.1c02156

Gandhi, R. R., Kathirvel, C., Kumar, R. M., & Ramkumar, M. S. (2023). Solar energy forecasting architecture using deep learning models. In Machine Learning and the Internet of Things in Solar Power Generation (pp. 105-121). CRC Press.

Jha, S. K., Bilalovic, J., Jha, A., Patel, N., & Zhang, H. (2017). Renewable energy: Present research and future scope of Artificial Intelligence. Renewable and Sustainable Energy Reviews, 77, 297-317 https://doi.org/10.1016/j.rser.2017.04.018

Kamiri, J., & Mariga, G. (2021). Research methods in machine learning: A content analysis. International Journal of Computer and Information Technology (2279-0764), 10(2) https://doi.org/10.24203/ijcit.v10i2.79

Li, C., Hao, H., Xu, B., Shen, Z., Zhou, E., Jiang, D., & Liu, H. (2021). Improved physics-based structural descriptors of perovskite materials enable higher accuracy of machine learning. Computational Materials Science, 198, 11071 https://doi.org/10.1016/j.commatsci.2021.110714

Liu, Y., Yan, W., Han, S., Zhu, H., Tu, Y., Guan, L., & Tan, X. (2022). How machine learning predicts and explains the performance of perovskite solar cells. Solar RRL, 6(6), 2101100 https://doi.org/10.1002/solr.202101100

Raschka, S. (2018). Model evaluation, model selection, and algorithm selection in machine learning. arXiv preprint arXiv:1811.12808 https://doi.org/10.48550/arXiv.1811.12808

Salah, M. M., Ismail, Z., & Abdellatif, S. (2023). Selecting an appropriate machine-learning model for perovskite solar cell datasets. Materials for Renewable and Sustainable Energy, 12(3), 187-198 https://doi.org/10.1007/s40243-023-00239-2

Sharifani, K., & Amini, M. (2023). Machine learning and deep learning: A review of methods and applications. World Information Technology and Engineering Journal, 10(07), 3897-3904 https://ssrn.com/abstract=4458723

Shiogai, J., Chida, T., Hashimoto, K., Fujiwara, K., Sasaki, T., & Tsukazaki, A. (2020). Signature of band inversion in the perovskite thin-film alloys BaS n 1–x P bx O 3. Physical Review B, 101(12), 125125 https://doi.org/10.1103/PhysRevB.101.125125

Subramanian, E., Karthik, M. M., Krishna, G. P., Prasath, D. V., & Kumar, V. S. (2023). Solar power prediction using Machine learning. arXiv preprint https://doi.org/10.48550/arXiv.2303.07875

Tao, Q., Xu, P., Li, M., & Lu, W. (2021). Machine learning for perovskite materials design and discovery. Npj computational materials, 7(1), 23 https://doi.org/10.1038/s41524-021-00495-8

Vélez, J., Botero L, M. A., & Sepulveda, A. (2024). Measurement of information content of Perovskite solar cell’s synthesis descriptors related to performance parameters. Emergent Materials, 1-8. Emergent Materials https://doi.org/10.1007/s42247-024-00667-4

Zhang, J., Liu, B., Liu, Z., Wu, J., Arnold, S., Shi, H., & Brabec, C. J. (2023). Optimizing Perovskite Thin‐Film Parameter Spaces with Machine Learning‐Guided Robotic Platform for High‐Performance Perovskite Solar Cells. Advanced Energy Materials, 13(48), 2302594 http://dx.doi.org/10.1002/aenm.202302594

Zhao, J., Dong, K., Dong, X., & Shahbaz, M. (2022). How renewable energy alleviate energy poverty? A global analysis. Renewable Energy, 186, 299-311 https://doi.org/10.1016/j.renene.2022.01.005