Artificial Intelligence-enabled Exploration Takes The Guesswork Out Of Searching for Gas

Leveraging Google’s Artificial Intelligence (AI) functionality, Deep Learning Café is helping Renergen use the data they have to streamline the process of searching for natural gas.

At the start of 2022, a report on the role of gas in South Africa’s path to net-zero carbon emissions explained that liquefied natural gas (LNG) is an important part of South Africa’s transition away from more emissions-heavy traditional fossil fuels. But only if it is affordably supplied. The report argues in favour of the importation of gas, suggesting that the possible exploration and development of domestic gas fields is simply too complex and capital intensive. Well, until now.

Renergen and Deep Learning Café have partnered to address this issue. A producer of helium and LNG, Renergen knows a thing or two about the costs associated with gas exploration and development. This is why they are working with Deep Learning Café, a Google Partner and solutions company that uses artificial intelligence (AI) -– Google’s Vertex AI solution in the case of Renergen -– to empower brands to better understand their business data and to help them make their operations more efficient by finding unique ways to solve the problems they face.

The devil is in the data

According to Khalid Patel, head of exploration at Renergen Limited, the search for any natural resource deposit almost always follows a conventional method, and this conventional approach entails significant capital expenditure upfront and occurs at risk. While these conventional methods are true and tested to determine where resources are located and estimate their quantities, this strategy is risky due to the large capital investment and risk of finding something not suitable for extraction. “Obviously, we are really interested in positive confirmations, but negative confirmations are equally as important for us,” says Patel, because this information allows them to focus their energy in areas that could bear more fruit.

Luckily the Renergen team inherited a large database of information but, as Simmone Du Plessis, hydrogeologist at Renergen Limited points out, data is only powerful if it is used and applied in the correct way. Rather than physically analysing all of this information themselves, Renergen enlisted the help of Deep Learning Café to create a robust set of machine learning tools that they could utilise to analyse this wealth of data and make predictions. This essentially is taking the guesswork out of their gas exploration efforts.

“It’s not only about knowing where to look, it’s also about understanding what components and bits and pieces of data are most important in determining where we needed to look,” notes Patel. Highlighting that the industry is typically quite set in its ways, especially in South Africa, he adds that there really was very little history of the use of technology in this field. For Renergen, however, this approach made sense. Having such a large and comprehensive dataset at their disposal and having a limited amount of physical resources to go out and do the exploration and analysis, it was logical for them to leverage machine learning to make more data-driven decisions. With this approach, Renergen can improve their exploration drilling – the process of drilling to solve the unknown or gather information in areas where there’s a lack of information – and their production drilling, which is the drilling they are currently doing in areas where they know there is gas; enabling them to know, with increasing confidence, what to expect in terms of volumes and composition. And they can even combine the two processes to cut costs.

For Dries Cronje, founder of Deep Learning Café, the challenge was to transform this geological and historic data into a clean and easy to access data set that is relevant to what Renergen is trying to achieve. The Deep Learning Café leveraged Google Cloud Platform’s Vertex AI solution to solve Renergen’s business problem in a way that is very similar to how companies like YouTube or Spotify make video and music recommendations. “Based on what they know about you, these kinds of platforms try to predict what videos and music you like or want. They categorise users with similar people with similar interests and similar habits. Like YouTube and Spotify, we are continuously learning and then using these learnings to more accurately predict where the next best areas are for exploration.”

Describing themselves as ‘sceptics’ around what people are selling as ‘artificial intelligence’, Patel and Du Plessis admit that they actually ran a bit of a sneaky test on the Deep Learning Café team to see if they could deliver on their promises. Patel tells the story that they sent over the database of information without telling them about any known sites. “We asked Deep Learning Café to create a preliminary model and then use this model to give us generalised and localised areas of where they think there should be gas. Based on the data they crunched, we also wanted them to show us what were the principal components that led to the determination of an area being a gas hotspot or not. And if I’m not mistaken, they were around 75% to 80% accurate on the first shot.” Du Plessis continued that given the amount of information that they provided the Deep Learning Café team with, anyone would expect them to be a little intimidated but they weren’t intimidated at all:. “This speaks volumes.”

For the Renergen team, it was all about de-risking their work. Du Plessis outlined that previously their strategy exposed them to human error and increased the likelihood that they might overlook patterns that AI is easily able to pick up. “Because most of our information is historical information, there was a lot of inconsistency in the data. This approach definitely shed some light on the kind of information we need to focus on moving forward and what information is most appropriate for the modelling approach we choose,” she concludes. “It has been so refreshing to see what knowledge can still be extracted from the data we’ve always had at our disposal; information that we might have overlooked, because we never really fully understood the relevance. We’ve only had a glimpse of what artificial intelligence is capable of and it’s definitely going to be a continuous learning process as the project evolves.”


Fill out this form and one of our experts will be in touch to discuss how we can help you.

    Sign up to our news