Siemens-EPSRC: Cloud-based solar forecasting for improved grid management

Lead Research Organisation: University of Nottingham
Department Name: Faculty of Engineering

Abstract

The contribution of PV energy to the electric grid continues to grow. Installed capacity in the UK in 2020 was 13.4 GW, (4.1% of total electricity generation compared with only 0.01% in 2010) and is expected to increase to 40 GW by 2030. Accelerating adoption of solar energy will present significant challenges to the electricity transmission and distribution system, as solar power is not dispatchable and therefore its incorporation as a major element of the generation mix requires the accurate estimation of solar energy production. The accurate estimation/prediction of solar energy generation is a significant challenge, especially in countries with widely varying weather patterns such as the UK, due to a poor understanding of the complex distribution of solar energy in the sky. Solar radiation is intermittent and the solar source at any given position on the plane of a PV array is highly dependent on the position of the sun, atmospheric aerosol levels, cloud cover and motion, etc. This inherent variability in the solar source directly affects solar-derived energy fed into power grids and can create severe imbalances between demand and the capacity/transport/distribution/storage of the grid, which can significantly impair grid reliability.

To counter these issues, the long-term aim is to develop a comprehensive digital platform for forecasting solar production (from very short to long term solar radiation forecasting) to significantly improve the prediction accuracy of meteorological parameters, reducing the power mismatch caused by solar forecast errors, and also reducing the continuing requirement for fossil fuel-based generation. To achieve this, the aim for this project is to build on our existing outdoor solar testing facility to significantly improve the prediction accuracy for intra-hour solar forecasting by developing and demonstrating a 'cloud'-based solar measurement and modelling platform to support multiple data sources and intensive prediction algorithms. The target is to achieve a prediction horizon of 20s to 1 hour with temporal resolution of 10s.
 
Description This project aims to develop a comprehensive digital platform for forecasting meteorological parameters. This will improve the prediction of solar power generation, significantly reducing power mismatch caused by forecast errors.

With the support from Siemens and EPSRC, a long-term sky data collection platform has been established. It includes direct and global solar spectrum and irradiance measurement devices, equipment for ambient temperature, relative humidity, and hemispheric and DSLR cameras with fisheye lens for sky image collections. In addition, a few PV panels have been installed close to the platform provide power measurement for independent pattern comparison. Meanwhile, a deep learning solar forecasting model based on image-numerical fusion has been developed in this project using the data (sky image and weather data) provided by the platform. It was found that compared to the classical CNN model based on late feature-level fusion, the transformer framework model based on early feature-level prediction developed in this project improves the balanced accuracy of Ramp Event by 9.43% and 3.91% on the 2-minute and 6-minute scales, respectively. However, based on the results, it can be concluded that for the single picture-digital bimodal model, the spatial information validity of a single picture is difficult to achieve beyond 10 minutes. This work has the potential to contribute to the interpretability and iterability of deep learning models based on sky images.
Exploitation Route Industrial partners have been invited joining the current project meetings and also project advisory board to directly contribute to project development and provide guidance/suggestions for the proposed system development.
Sectors Digital/Communication/Information Technologies (including Software),Energy
 
Description Presentation at EU PVSEC conference 2022 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Professional Practitioners
Results and Impact We made a presentation about the current outcome of our project at the EU PVSEC conference 2022. Over 500 professionals attended the conference online, good feedback were received and audiences have better understanding the importance for PV prediction.
Year(s) Of Engagement Activity 2022