An Ensemble Model Being Used in CLP's Load Forecasting - Story of CSE Alumni, Mr. Binnie Yiu

"I do hope there would be an opportunity for me to work on a real-life project after graduated from the MSc of Mathematics programme." says Mr. Binnie Yiu, a MPhil in Computer Science and Engineering graduate (Class of 2022) and an Engineer in the Technical Services Department of CLP Power Hong Kong Limited (CLP) since November 2020. One of the directions would be using a machine learning approach in resolving practical problems, as AI is a solution trend in the world. Therefore, Binnie was considering studying a computer science MPhil degree. In particular, the Dual Master's Programme which was organized by CLP Power Academy headed by its Vice Chancellor Ir Paul Poon provided him an opportunity to complete such a research project under Prof. Tong ZHANG of HKUST and CLP industry supervisor Dr. Cheuk Wing LEE. This programme brings together two leaders in their respective fields, and the research results provide quantifiable benefits to the industry as well as giving Binnie better insights on how to deal with new challenges and capitalizing on opportunities. Binnie said, "It perfectly fits my aims and desires to pursue my dream, and it is my pleasure to learn from the programme and the two supervisors."

Ensemble Approach for Short-Term Load Forecasting Using Regularized Greedy Forest

The research team led by Prof. Tong ZHANG has developed an ensemble model to assist CLP in system short-term load forecasting (STLF) for its supply territory, for accelerating its transition toward a digitalized and greener utility of the future. The ensemble model is based on Prof. ZHANG's algorithm - Regularized Greedy Forest (RGF - paper, GitHub), which provides a unique insight in the tree-based machine learning model for the supervised learning problem, by considering the forest structure when forming the decision forest. The primary forecasting task is to predict the hourly electricity load demand for the next 24 hours using historical time series demand data, weather, and relevant calendar-related inputs.

Impact of the Model

The ensemble model is being used in the daily operations of CLP's load forecasting. The forecasting result enables system operators to handle demand responses more effectively and optimize generation resources. It also makes equipment servicing decision-making easier and more effective while guaranteeing sufficient energy supply to the customers and for the society. In addition, the model improves generation scheduling, reduces carbon emissions, and supports the city's ongoing development.