Turning sunlight into smarter energy with AI

Emvelo had already achieved something remarkable with Ilanga CSP-1: a 100MW solar power plant capable of powering more than 100,000 homes. But their ambition didn’t stop there. To make the biggest possible impact, both financially and environmentally, the plant needed to deliver maximum output during the evening, when South Africa’s grid is under the greatest demand. Achieving that goal was far from simple. Running a concentrated solar power plant is a delicate balancing act, with operators constantly making complex decisions in real time. They needed a smarter way to manage the trade-offs between energy storage, generation and unpredictable weather. The main challenges came down to three critical questions: when to use energy instantly and when to store it in molten salt tanks for after sunset; how to forecast solar input accurately enough to make confident operating decisions; and how to meet peak evening demand, when grid exports are most valuable but the sun is no longer shining.

We began by tackling the plant’s vast data challenge. Every day, Ilanga CSP-1 generates high-resolution readings from more than a thousand sensors spread across the site. This information was originally stored in a proprietary format, making it difficult to query and analyse. Our first step was to transform this raw data into a streamlined, cloud-based data lake on Amazon S3, making the information accessible, performant and ready for AI modelling. With the data re-engineered, we created a digital twin of the plant. Using machine learning, we trained a statistical model that could accurately mirror the behaviour of Ilanga CSP-1. This allowed us to run safe experiments, testing different operating strategies and seeing how changes in control decisions would play out in the real world. The final stage was to put optimisation techniques to the test. From simple brute-force searches to advanced genetic algorithms, we explored a range of approaches to discover the most effective way of balancing turbine generation with molten salt storage. Early results were promising, showing that AI-driven control could significantly increase evening energy output, exactly when the grid needs it most.

The research phase showed how AI could improve the operation of concentrated solar power plants. Digital twin modelling and optimisation techniques demonstrated how smarter control strategies could increase evening output, with potential benefits for revenue and environmental impact. Alongside these broader outcomes, the research uncovered promising indications of what AI could achieve in this context: £3m+ potential annual uplift suggested by early optimisation models, increased CO2 displacement potential if evening generation can be improved, and a validated digital twin providing a reliable foundation for future experimentation.

The research phase was only the beginning. With the potential of AI now proven, Emvelo’s focus has shifted to turning these insights into everyday operational tools. Together with Emvelo, we are developing a bespoke web application that will put optimisation directly into the hands of plant operators, enabling them to make faster, smarter decisions with confidence. This next stage ensures that the impact of the project is not just theoretical but transformative, driving long-term value for Emvelo, their customers and the climate.