23rd January

Weekend Build: Optimising Singapore’s EV Network with K-Means Clustering

The Singapore Green Plan 2030 has a massive target: 60,000 EV charging points by 2030. That is a lot of chargers, and placing them isn’t as simple as throwing a dart at a map. You need to balance grid load, accessibility, and coverage.

This weekend, instead of just reading about it, I decided to build a tool to solve it.

Introducing VoltVantage, a web-based simulation that uses K-Means Clustering to determine the optimal placement of EV charging stations across Singapore’s existing carparks.

The Problem: The Charging Station Placement Problem (EVCPP)

In computer science terms, placing these chargers is an optimisation problem. We have a set of constraints (boundaries, available carparks) and a goal (maximum coverage with minimum overlap).

I grabbed the official dataset of Singapore’s carparks (carparks.json) and the country’s geospatial boundaries (singapore-boundary.json) to create a realistic testing ground. The goal? To see if an algorithm could mathematically determine the most efficient spots to serve the most people.

Under the Hood: The Tech Stack

I built this using a modern, lightweight stack designed for speed and interactivity:

• Frontend: React / Next.js
• Visualisation: react-leaflet for the interactive map layer.
• Geospatial Analysis: Turf.js. This was the workhorse for the spatial logic.
• Data: Real-world HDB/URA carpark data.

The “AI” Sauce: K-Means Clustering 🧠

The coolest part of VoltVantage is the logic in EVPlanner.tsx. Instead of randomly guessing locations, I implemented K-Means Clustering.

For those unfamiliar, K-Means is an unsupervised machine learning algorithm commonly used for grouping data. Here is how VoltVantage applies it to infrastructure:

1. Data Ingestion: It takes the coordinates of thousands of carparks across Singapore.
2. Clustering: The algorithm groups these carparks into k distinct clusters based on their geographical proximity.
3. Centroid Calculation: For each cluster, it calculates the “centroid”, the geometric centre of gravity.
4. Optimisation: This centroid becomes the proposed location for a charging station.

By doing this, the system ensures that the charging station is placed in the “middle” of a high-density area, minimising the average distance any driver in that cluster has to travel to get a charge.

The Result

Watching the simulation run is satisfying. You start with a raw map of carparks, and the algorithm instantly brings order to the chaos. It highlights natural hotspots of vehicle density and places markers exactly where they are needed most.

It’s a functional proof-of-concept that infrastructure planning doesn’t have to be based on gut feeling, it can be solved with geometry and code.

Why I Built It

Coming off the back of my internship at Accenture (H&PS) and the recent Kaggle competition, I’ve been obsessed with how data can influence public infrastructure. VoltVantage was a way to combine my frontend skills with the data science concepts I learned in university.

Plus, it was a great excuse to finally play around with Turf.js.

The code is still a weekend prototype, but the potential is there. If we’re going to hit that 2030 target, we’re going to need tools like this to make sure every charger counts.

  [·                   ]