Deep Learning Framework for Assessing Rooftop Photovoltaic Potential in Tianjin, China

Deep-Learning-Based Evaluation of Rooftop Photovoltaic Deployment in Tianjin, China

Rooftop photovoltaics (RPVs) play a critical role in mitigating energy shortages and addressing environmental issues stemming from fossil fuel usage. To enhance the deployment of RPVs in Tianjin, a region characterized by its high solar potential and dense urban infrastructure, this study presents a framework that combines building vector data with a deep learning model to extract information on currently installed RPVs from remote sensing images and assess their future development potential.

The analysis led to the extraction of a total of 86,363 RPV polygons, covering an area of 10.34 km². Notably, over 70% of these RPVs are located on large and low-rise buildings, with a similar proportion found in industrial structures, which are favorable for installation. By integrating solar radiation data with land development planning, we identified potential deployment zones for RPVs that encompass roughly 13% of Tianjin’s land area, translating to an estimated annual power generation capacity of 31.31 TWh. Future RPV installations should prioritize large and low-rise buildings or industrial sites within these zones to maximize power generation and support environmental emission reduction targets. The research framework developed here can also be applied to other urban areas.

Introduction

The historical dependence on fossil fuels for economic growth has led to energy shortages and significant environmental issues. In response, there is an increasing focus on renewable energy sources, particularly solar energy. Photovoltaic (PV) technology has rapidly expanded, with global installed capacity rising from 136 GW in 2013 to 1411 GW in 2023. Within this growth, rooftop PV systems have also seen considerable installation, reaching an installed capacity of 95 GW by 2022. RPVs address space limitations in densely populated urban areas and provide decentralized energy solutions that are directly connected to consumers.

To effectively harness the potential of RPVs, accurate assessments of their deployment capabilities are essential. Most existing studies typically employ a two-step approach: first, estimating available installation space, followed by calculating deployment potential based on solar radiation and PV efficiency. While large-scale studies often utilize empirical regression models, smaller studies tend to leverage deep learning techniques and remote sensing imagery for more precise evaluations. There is a growing need for detailed mapping of installed RPVs, especially given the rapid growth in distributed PV systems in China.

Data and Methods

The research focuses on Tianjin, situated in the North China Plain with abundant solar resources. The study utilizes high-resolution remote sensing images from Google Earth and incorporates solar radiation data from WorldClim. The methodology comprises three main components: extracting installed RPV locations and sizes from remote sensing images using a modified U-Net deep learning model, determining potential deployment zones by considering solar radiation and land use planning, and estimating the power generation and environmental impact of RPVs.

Results and Discussion

The adapted U-Net model demonstrated effectiveness in accurately identifying RPVs, achieving high scores in Precision, Recall, F1-score, and Intersection over Union (IoU). The extraction results revealed significant potential for RPV deployment, particularly in districts with high population density and economic activity. The findings indicate that the majority of RPVs are concentrated in areas with favorable installation conditions, such as industrial buildings.

The analysis of potential deployment zones identified areas with high solar radiation that lack installed RPVs. This strategic identification aligns with future land-use planning, emphasizing the importance of government policies in optimizing RPV deployment.

The environmental benefits of RPV installations are substantial, with the potential for significant reductions in carbon and air pollutants. The economic implications are also notable, with potential savings on electricity bills and favorable return on investments for users.

Conclusion

This study provides a framework for effectively evaluating the deployment of rooftop photovoltaics in Tianjin, revealing key insights into current installations and future potential zones. The results underscore the importance of prioritizing RPVs in suitable buildings to maximize energy generation and contribute to sustainability goals. Future research should address the challenges of RPV deployment in various building types and explore the broader impacts of RPVs on urban climates.