Using Machine Learning to Optimize Electricity Usage in a Smart City

Starlake City - A data-driven smart city

Starlake City, spanning 187 hectares in Hanoi, Vietnam, situated between the green space of West Lake and Noi Bai International Airport, aims to revolutionize the living experience for its 24,000 residents.

Developed in collaboration with Hanoi City, the vision was to create a smart city that uses AI and machine learning (ML) to improve the living experience for residents.

The AI capabilities were developed through collaboration between Daewoo E&C’s Vietnamese subsidiary THT Development, Bespin Global’s Vietnamese subsidiary, Bespin Global Vietnam, and Australian deep-tech startup PI.EXCHANGE and their flagship product, the AI & Analytics Engine (the Engine).

“The whole idea of the smart city is really to provide the best service to our residents. It’s a lot of little things, that from a resident perspective, makes the difference. Whether that's the electricity, water, or parking. The goal is to provide the best way.”

- John Lee, Bespin Global CTO

The first application developed was a data-driven smart grid, which uses ML to forecast electricity consumption, critically ensuring reliable supply to the city. Bespin Global Vietnam created a software platform to manage the Smart City and utilized the Engine to add machine learning capability to generate electricity consumption forecasts.

Challenges to overcome

Providing a stable supply of electricity to a smart city requires careful planning and management. By forecasting and understanding future electricity consumption the following challenges can be overcome:

1. Electricity demand must always be met, which can be ensured by understanding fluctuations and predicting peak times.

2. Power providers need to be able to efficiently allocate their resources, such as distributing power from different sources and managing energy storage systems.

3. It is crucial to prevent blackouts, and forecasts allow providers to take proactive measures such as initiating load-shedding programs or managing voltage levels.

4. Consumption forecasts can help power providers shift to more sustainable sources of energy by allowing them to predict when renewable sources, such as wind and solar power, will be most effective.

Requirements for a successful project

Bespin Global Vietnam needed an ML solution that could fulfill a number of requirements for the project to be successful:

1. Accessible to engineers without data science experience.

2. Preparing datasets into a format that aligns with business real-world requirements.

3. Produce ML models with high prediction quality.

4. Consuming predictions on an ongoing basis.

The machine learning solution

The AI & Analytics Engine is a complete ML platform that simplifies the process of building, deploying, and managing ML models. The team at Bespin Global Vietnam swiftly constructed the electricity forecasting solution using the Engine.

“I think the AI & Analytics Engine is a perfect fit for the smart city development. We can get a lot of value out of it quickly and easily and doesn’t require a lot of knowledge of AI, so it’s usable by data people or engineers. I think that’s one of the biggest advantages of the Engine.”

The Engine was a perfect fit for Bespin Global Vietnam’s requirements. The user-friendly approach was particularly beneficial for their engineering team, leveraging its capabilities without the need to hire additional data science experts.

“For example, we looked at competitors, but it’s hard to use, and you need a whole staff of people dedicated to the implementation to make it work. The AI & Analytics Engine’s value proposition was quick and straightforward. It’s a great fit for us.”

The data wrangling feature was used to develop a reusable recipe of data transformation actions, greatly speeding up the data preparation process. The model built had a prediction quality, measured by F1-score, of over 77% which was sufficiently high. Finally, the ML pipeline in combination with the SDK allowed them to automatically schedule periodic predictions on an ongoing basis. These aspects resulted in a streamlined project, that delivered value quickly.

Tangible benefits delivered

The ML-powered electricity consumption forecasts were developed to provide a way to plan and manage electricity supply, in order to provide value to residents. Some of the specific advantages that resulted include:

1. More informed decisions on energy generation and usage, leading to improved energy efficiency and reduced waste.

2. Resources are allocated optimally, ensuring that energy is used efficiently and sustainably.

3. Avoidance of over-generation of electricity, resulting in cost savings associated with energy production.

4. Improved grid stability, reducing the likelihood of blackouts.

5. Data enabling informed investments in energy infrastructure.

Wrapping up

The revolutionary smart city project in Starlake City, Hanoi, was designed to provide a first-class experience for its residents in a multitude of ways. The first application, optimizing electricity supply required ML to forecast future electricity consumption, and the AI & Analytics Engine was successfully leveraged, providing value quickly and ensuring a successful project.