Real-time EV Range Prediction

Overview

This project aims to predict the range of an Electric Vehicle (EV) based on real-time parameters. By analyzing factors such as speed, battery status, temperature, and driving patterns, we provide a more accurate range estimate than simple percentage-based calculations.

Features used

The models are trained on the following synthetic features:

• Speed (km/h): The current speed of the vehicle.
• State of Charge (SOC) %: The current battery level.
• Ambient Temperature (°C): The outside temperature.
• Motor RPM: Rotations per minute of the motor.
• Motor Temperature (°C): Temperature of the motor.
• Battery Temperature (°C): Temperature of the battery pack (Critical for efficiency).
• Acceleration (m/s²): Driving aggressiveness.

Project Structure

Real time Range prediction/
├── data/                   # Generated data and plots
├── models/                 # Saved models (XGBoost, MLP, Scaler)
├── src/
│   ├── data_simulator.py   # Generates synthetic data
│   ├── eda.py              # Exploratory Data Analysis
│   ├── train_model.py      # Trains XGBoost model
│   ├── train_advanced_model.py # Trains MLP (Neural Network)
│   └── compare_models.py   # Compares both models
├── requirements.txt        # Dependencies
└── README.md               # Project documentation

Installation

1. Clone the repository.
2. Install the required dependencies:

   pip install -r requirements.txt

Usage

1. Generate Data

If you don't have the dataset, generate synthetic data:

python src/data_simulator.py

This creates data/ev_range_data.csv.

2. Exploratory Data Analysis (EDA)

Visualize the data correlations and distributions:

python src/eda.py

Correlation Matrix

Pairplot

3. Train Baseline Model (XGBoost)

Train the XGBoost regressor:

python src/train_model.py

Feature Importance

Predictions vs Actual (XGBoost)

4. Train Advanced Model (MLP)

Train the Deep Neural Network:

python src/train_advanced_model.py

Training Loss

5. Compare Models

Compare the performance of XGBoost and MLP:

python src/compare_models.py

Results & Comparison

Model	MSE	MAE	R2 Score
XGBoost	118.58	8.59	0.9828
MLP (Neural Network)	106.03	8.18	0.9846

The MLP model slightly outperforms the XGBoost model, achieving a higher R2 score and lower error rates.

R2 Score Comparison

Predicted vs Actual (Comparison)

Conclusion

This project successfully demonstrates that machine learning can accurately predict EV range using real-time vehicle data.

• Key Findings:
  • Speed and SOC are the most dominant factors.
  • Battery Temperature plays a significant role in efficiency; extreme temperatures reduce range.
  • The Neural Network (MLP) captures complex non-linear relationships slightly better than XGBoost, though both models perform exceptionally well (R2 > 0.98).
• Future Work:
  • Integrate real-world driving data.
  • Implement a time-series model (LSTM) if sequential data becomes available.
  • Deploy the model as a real-time API.