By Nnasom David

A Nigerian researcher has uncovered critical vulnerabilities in artificial intelligence (AI) systems used to forecast green hydrogen production, a breakthrough that could shape Africa’s race toward clean energy independence and highlight new risks in the global shift away from fossil fuels.

Dr. Chiagoziem C. Ukwuoma, a researcher and lecturer at Chengdu University of Technology, Oxford Brookes College, China, revealed in a new study that adversarial attacks could compromise the accuracy of AI models essential to producing hydrogen from biomass and plastic waste.

His findings were presented at the 15th International Conference on Applied Energy (ICAE 2023) in Doha, Qatar.

The study, titled “Security Concerns for Using Deep Learning Models in Predicting Hydrogen Production: A Comparative Study on Adversarial Attack,” demonstrated how small, carefully designed data manipulations known as adversarial attacks could drastically distort predictive outcomes.

Such disruptions, experts warn, could derail industrial hydrogen projects critical to replacing fossil fuels and cutting carbon emissions.

Ukwuoma and his co-authors tested four widely used machine learning models, Random Forest Regressor, XGBoost, Support Vector Regressor, and K-Nearest Neighbours alongside a novel deep learning model on a co-gasification dataset.

Their results showed that adversarial techniques such as the Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD) caused significant drops in accuracy when applied at varying levels of data perturbation (0.1 to 0.8).

At higher perturbation levels, prediction errors spiked dramatically, threatening the reliability of hydrogen production systems.

Ukwuoma said: “This vulnerability is especially alarming for Nigeria and Africa, where hydrogen from local biomass could harness 110 EJ of potential by 2050 to combat energy poverty and black soot pollution from oil flaring. In Port Harcourt, toxic soot has spiked respiratory diseases by 30–50 per cent, contaminating communities reliant on fossil fuels.

“From the Niger Delta’s polluted skies to Africa’s untapped biomass riches, I’ve dedicated my research to secure, sustainable technology that empowers our continent. Adversarial attacks could undermine hydrogen’s promise as a clean alternative to oil, but by spotlighting these flaws, we pave the way for fortified models that protect jobs, health, and the environment.”

Ukwuoma’s findings underscore the need for robust defenses to ensure AI-driven hydrogen optimization doesn’t falter under cyber threats, enabling safer, more efficient production from agricultural waste like RSS, abundant in Nigeria’s rubber belt.

Ukwuoma’s analysis showed that FGSM and PGD attacks drastically reduce hydrogen prediction accuracy, that is RMSE spikes at 0.8 perturbation, exposing AI risks in Africa’s green energy shift.

Co-authored with experts, including Prof. Dongsheng Cai, and funded by the Sichuan Engineering Technology Research Center, the study calls for integrated security protocols in AI for energy sectors.

It aligns with Nigeria’s Energy Transition Plan and the African Union’s green agenda, potentially reducing emissions by 15 per cent through reliable predictions while creating opportunities in waste-to-fuel innovation.

The Society of Technology and Energy Professionals (STEP) applauded the research for bolstering accountability in Nigeria’s oil-dependent economy. Ukwuoma advocates for collaborations with African tech hubs and regulators to implement these safeguards.