Researchers at SLIIT have built an artificial intelligence framework that forecasts power output from biomass-fuelled steam turbines, reporting roughly 25% better accuracy than existing benchmark methods when tested against eight years of real factory data.
The work was published in Energy Conversion and Management and is set out in an announcement from the institute, the Colombo Gazette reported. It addresses power generation in palm oil production, where factories burn agricultural by-products — fibres, shells and husks — as turbine fuel rather than buying electricity.
The problem it targets
The forecasting difficulty is specific and has a cost on both sides of an error. A factory that overestimates the power it will need burns biomass it did not have to burn, consuming a fuel stock that is finite and seasonal. A factory that underestimates falls short and has to draw more electricity from the grid, which is the expensive option it installed the turbines to avoid.
What sits between those two outcomes is a prediction, and the prediction is what the SLIIT model is for. The system learns from industrial operating patterns so that operators can anticipate power requirements in advance and match turbine output more closely to actual demand. The institute frames the practical gains as fuel planning, cost control and production scheduling — the argument being that in an energy-intensive plant, even a modest improvement in forecast accuracy changes how much fuel is wasted.
Who did the work
The research was led by Himaya Perera, a SLIIT Electrical and Electronic Engineering graduate, who according to the institute is now reading for a Master of Engineering in Electronics at La Trobe University in Australia.
The team includes SLIIT’s Eranga Wijesinghe and Bhagya Nathali Silva, working with Shalitha Jayasekara and Honnyong Cha of Kyungpook National University in South Korea.
SLIIT links the work to wider efforts on industrial energy efficiency and renewable resource use, and says the approach has potential applications beyond the sector studied.
Not reported
The announcement does not say where the eight-year dataset came from or whether the plants studied are in Sri Lanka — a material gap, since commercial oil palm cultivation here has been subject to government restriction, and the applicability of the model to local industry depends on it. The specific AI method used, the benchmark methods the 25% figure is measured against, and whether the system has been deployed in a working plant rather than tested on historical data are all unstated. The announcement comes from the institute rather than from an independent assessment of the published paper, and no second verified newsroom had covered the research at the time of writing.
