The launch of ChatGPT has made artificial intelligence (AI) one of the hottest topics of debate on the planet, generating excitement and concern from the global scientific community in equal measures.

What is beyond debate is the immense transformative potential of such AI systems which not only provide instant answers and solutions but learn continuously and refine their own ability to meet users’ needs.

The introduction of AI into business operations worldwide means vast amounts of data can now be collected, processed, and analysed in real time, offering valuable insights to businesses in making decisions and forecasts.

A Breakthrough in Forecasting

An ensemble machine learning model devised by Binnie was introduced in 2022, further strengthening the short-term load forecasting accuracy of the company’s power systems.

His technique blends the predictions of several machine learning-based algorithms to make more accurate predictions and has brought numerous benefits, the most important of which is the optimisation of power to reduce carbon emissions.

Binnie’s pivotal role in developing the ensemble model to manage load forecasting started with a keen interest in science, especially AI and machine learning.

The 34-year-old holder of a bachelor’s degree in mechanical engineering and a master’s degree in mathematics, Binnie works in the Technical Services Department at CLP Power’s Power System Business Group and spends his free time learning about new technologies and researching them online.

In 2019, Binnie enrolled in the academic-and industry-run dual master’s degree programme in power engineering to find out more about the industry and trends in science and engineering, as well as to get a fuller picture of the power business.

The part-time dual master’s degree programme is offered by CLP Power Academy in partnership with the Hong Kong University of Science and Technology (HKUST) and the University of Strathclyde in Scotland.

Theoretical Solutions to Real-World Problems

Binnie believes AI is a major game-changer and that machine learning can help resolve practical problems at work. His thesis project for his degree was to develop an ensemble model to improve a power system’s short-term load forecasting performance.

Load forecasting is a crucial aspect of energy management and planning. Every day at noon, engineers at System Operation Department predict the demand for 24 hours from midnight onwards to formulate a day-ahead generation plan.

“The use of electricity varies daily,” says Binnie. “As excess power cannot be stored, an overestimate of demand wastes energy and can be costly. An underestimate, by contrast, can lead to energy shortages or blackouts. So, an accurate forecast is extremely important to generate the optimal amount of power,” he explains.

The forecasts help engineers make operational planning decisions to optimise generation resources effectively, ensuring a reliable power supply for homes and businesses. A better forecasting system also saves fuel and supports Hong Kong’s move towards carbon neutrality.

Making accurate predictions is a challenging task, however. “There are many factors that influence electricity consumption, such as weather, calendar effects, social activities and special events,” says Binnie.

In recent years, drastic changes brought by climate change – including more frequent typhoons and hot weather – make it ever harder to predict electricity demand based on historical data. The wider use of renewable energy and the growing popularity of electric vehicles also make forecasting harder.

Building a Bigger Bank of Intelligence

Algorithms have been used in the past for short-term load forecasting, but there is no universal model that works effectively for every power utility company.

The previous model used by CLP Power drew on historical data such as temperature, relative humidity and seasonality. Although generally reliable, it still had an average prediction error of 2%, which may sound small but is considered significant in the cost-sensitive power industry.

Binnie wanted to develop a more robust model for short-term load forecasting in support of CLP Group’s digitalisation push. Instead of relying on a single algorithm, Binnie used an ensemble approach and created a hybrid model to further improve the model’s forecasting performance.

Supervised by Professor Tong Zhang at the HKUST and Dr Lee Cheuk-wing, Binnie’s thesis proposed combining Professor Zhang’s machine learning algorithm, Regularised Greedy Forest, with two other machine learning applications, creating an ensemble model that delivered improved stability and greater accuracy of short-term forecasting.

With cross departmental support, Binnie tested and implemented the ensemble model from 2021. The model worked well and reduced errors in forecasting by an average of 20% compared with the previous model.

From early 2022, the model was applied to CLP Power’s daily operations where it quickly established itself as an effective way to reduce fuel waste, increase operational performance and support decarbonisation.

A Lifelong Learning Curve

Accurate load forecasting not only saves costs and reduces emissions but also boosts energy efficiency and improves the overall reliability of the power system. Similarly, Binnie’s work has not only benefitted CLP Power but has also brought valuable insights for the industry.

Binnie last summer completed his dual degree programme in less than three years, becoming the first graduate to receive a Master of Philosophy in computer science and engineering from the HKUST, together with a master’s degree with distinction in future energy and power system smart operation and management from the University of Strathclyde, an achievement that brought him the Best MSc Student Award from the university in Scotland.

Binnie says he was motivated by the ability to apply his knowledge to tackle real-world problems. His dedication to his profession, together with the encouragement and support of his parents, colleagues and supervisors throughout the course, ultimately resulted in a breakthrough that has led to a more reliable power supply.

Perseverance is the key both to his own achievement and the machine learning model he has created. Both are constantly seeking to improve themselves to perform as effectively as possible, he adds.

Looking ahead, he says he would like to pursue lifelong learning and focus on his research, finding more ways to apply big data and machine learning to improve business operations.

“The trick to lifelong learning is to stay curious and put what I have learned into practice,” Binnie reflects. “I hope that by combining theories with practice and creating the right tools and models, I can contribute to the power engineering field and leave my own little mark on it.”

He also hopes his model can eventually be expanded from power system operations to other areas of the energy network, widening its reach and encouraging the smart use of energy.

The AI revolution is just beginning, and CLP Group is determined to continue embracing innovative technologies to improve its business performance and play a part in creating a more sustainable energy future.