May 24, 2019 Scaling Machine Learning in Energy, Natural Resources, and Utilities
For projects in the energy, natural resources, and utilities sectors, machine learning can enable solutions that push the boundaries of what has traditionally been possible. For many challenges, solutions using machine learning are the only way to tackle the complex data and shifting variables common to these sectors. With the scope of data these companies are handling, and the importance of the predictions needed, machine learning solutions need to be robust and scalable.
Topcoder has a strong history of providing solutions to the energy, natural resources, and utilities companies who need to reduce risks, identify project sites, and react to the ever-changing factors of working in the natural world.
Machine Learning and the Energy Sector
Many projects in the energy sector need to analyze data and then use the results of that analysis to predict future events. For example, a company might need to analyze seismic activity logs to predict future fault lines better or interpret weather data to site a wind farm. Machine learning can lead to artificial intelligence solutions that “understand” the data and learn to make accurate predictions. For the energy sector, this kind of solution will become more and more critical as the industry continues to evolve.
Scaling Machine Learning
Machine learning solutions for the energy sector need to be scalable—and Topcoder projects provide solutions that scale. By using large datasets in challenges and running longer competitions like Marathon Match challenges, the Topcoder community is better able to offer robust algorithms for energy projects. For some projects, the community first generates a new dataset to work with from existing input. And then a follow-up challenge allows coders to use that data to create machine learning solutions.
Machine Learning Case Studies with Topcoder
Aquifer Boundary Identification
Identifying aquifer boundaries is necessary to protect human and environmental safety near oil wells. But accurately mapping those boundaries across hundreds of wells is cumbersome. A Topcoder challenge resulted in an aquifer mapping solution that combined several data sets to estimate aquifer boundaries with very high levels of accuracy.
Topcoder helped a customer digitize a large volume of mud logs. Then, they helped analyze them to help with their future decision making. Because of the complexity of the logs, the digitization process was human-assisted. They used automated OCR to reduce the amount of human labor. Once the logs were digitized, machine learning algorithms were able to extract useful information from the records.
This hybrid approach will be a hallmark of future machine learning projects, experts say. Rather than simply replacing humans, artificial intelligence will be used as a tool to allow human workers to spend more time on higher-level tasks.
A customer needed a tool capable of analyzing video and detecting hazards. It needed to improve worker safety in a variety of scenarios. This type of project was ideally suited for a machine learning (computer vision) solution. The volume of video input would be high, and the particular hazards would be variable. The resulting tool can detect potential hazards by identifying vehicle, machinery, and worker movement and dangerous worker behavior.
Another worker safety project used computer vision to detect whether workers were using protective gear properly. The Topcoder community produced a labeled dataset from existing video footage. Then, they used that dataset to generate a solution to detect issues with how workers are using their protective gear.
For energy, natural resources, and utilities projects, as with all Topcoder solutions, clients’ data and details are secure. Topcoder uses multiple layers of protection throughout each competition. In conclusion, Topcoder is continually working to make data more secure. So, that initiative includes an upcoming Differential Privacy competition this fall that will focus on new ways to obscure client data.