EV Infrastructure Planning - BuzzOnEarth India Hackathon

Project Overview

This project focuses on Electric Vehicle (EV) Infrastructure Planning in India, aiming to address the growing adoption of EVs. As India expects to have over 1 million EVs on the road by 2024, with the market projected to grow at a CAGR of 23% from 2021-2030, there is an urgent need to optimize the placement of EV charging stations.

Team Members

• Ramesh Babu (22BCS208)
• Tanvi Koranne (22BEC123)
• K.Charan Teja Reddy (22BCS132)
• Arun Jatothu (22BCS113)
• Dasari Venkata Sri Sai Akshay Kumar (22BCS075)

Problem Statement

With the rise in EV adoption, there is a critical need for well-distributed charging stations to ensure:

• Convenience for EV users
• Minimized traffic congestion and delays
• Support for green energy initiatives

Current issues include:

• Rapid urbanization
• Traffic congestion
• Pollution and inadequate waste management

Approach: EV Station Placement

The solution uses a combination of Reinforcement Learning (RL) and Machine Learning (ML) classification to optimize EV station placement in cities. The key methodology involves:

1. Dividing cities into grids for manageable analysis.
2. Applying a machine / deep learning classifier to each grid to predict the suitability of EV stations.
3. Reinforcement Learning selects and refines the grid selection process until optimal placements are identified.

Dataset

The dataset consists of road networks, points of interest (POIs), population density, and other infrastructure details (e.g., civic buildings). The data is collected from 9 cities in Germany to predict EV station placements.

Tech Stack

Frontend:

• React.js
• Tailwind CSS
• Deck.gl (for mapping)

Backend:

• Python
• FastAPI
• Uvicorn

Machine/Deep Learning:

• PyTorch
• Scikit-learn

Intel Optimization Tools:

• Intel® oneAPI AI Analytics Toolkit
• Scikit Learn Intelex
• Intel Extension for PyTorch
• OneDAL, OneDNN, Modin
• Intel® Xeon® Platinum 8468V

Results

The performance of the models is as follows:

• Big City Model: AUC = 0.6893, Accuracy = 81.93%
• Small City Model: AUC = 0.7221, Accuracy = 88.77%
• Combined Cities Model: AUC = 0.7319, Accuracy = 86.06%

The solution demonstrates a predictive accuracy of 70-80%, with potential for improvement as more charging infrastructure is developed.

Future Scope

• Real-Time Data Integration: Enhance predictions by incorporating dynamic data.
• Advanced Algorithms: Explore hybrid models to improve accuracy.
• User Behavior Analysis: Predict future demand by analyzing EV driver behavior patterns.

License

This project is part of the BuzzOnEarth India Hackathon.