Skip to content

rbvwolf/volt-guardian

Repository files navigation

Volt Guardian XAI

Volt-Guardian XAI is an Explainable AI (XAI) project designed to monitor and predict the State of Health (SoH) for Electric Vehicle batteries. This project was developed as part of the 3rd Year Computer Engineering curriculum.

The project combines:

  • Genetic Algorithm (GA) Hyperparameter Optimization
  • XGBoost Regression
  • SHAP (SHapley Additive exPlanations)
  • Feature Engineering
  • Comparative Model Benchmarking

The primary objective is not only to predict battery degradation with high accuracy but also to provide transparent explanations for each prediction.


Key Features

Hybrid GA-XGBoost Architecture

A Genetic Algorithm is used to optimize XGBoost hyperparameters automatically, resulting in improved predictive performance.

Explainable AI (XAI)

SHAP is integrated to:

  • Explain global feature importance
  • Analyze local predictions
  • Generate waterfall and beeswarm visualizations
  • Increase model transparency

Advanced Feature Engineering

The framework introduces custom battery stress indicators:

  • Thermal Stress Index
  • Monthly Charge Intensity
  • Power Stress

These engineered features help the model better capture battery degradation behavior.


Dataset

The dataset contains EV battery operational parameters such as:

  • Battery Capacity
  • Vehicle Age
  • Charging Cycles
  • Fast Charging Ratio
  • Average Temperature
  • Internal Resistance
  • Driving Style
  • Battery Chemistry

Target Variable:

State of Health (SoH %)

Model Performance

GA-XGBoost (Champion Model)

Metric Score
0.9891
Adjusted R² 0.9890
MAE 0.2650
RMSE 0.3404
MAPE 0.28%

Cross Validation

Metric Value
Mean R² 98.83%
R² Std ±0.08
Mean MAE 0.27
MAE Std ±0.01

Model Comparison

The framework evaluates multiple machine learning models:

  1. GA-XGBoost (Hybrid Champion)
  2. Random Forest
  3. Standard XGBoost
  4. Gradient Boosting
  5. Decision Tree
  6. Linear Regression

GA-XGBoost achieved the best overall performance.


Explainability Visualizations

The project generates:

  • SHAP Global Beeswarm Plot
  • SHAP Waterfall Plot
  • Actual vs Predicted Analysis
  • Residual Distribution Analysis
  • Feature Correlation Heatmap

These visualizations help researchers understand the decision-making process of the model.


Technologies Used

  • Python
  • XGBoost
  • SHAP
  • Scikit-Learn
  • sklearn-genetic-opt
  • Pandas
  • NumPy
  • Matplotlib
  • Seaborn
  • Google Colab

Future Work

  • Real-time EV battery monitoring
  • Web-based dashboard deployment
  • Integration with Battery Management Systems (BMS)
  • Deep Learning based SoH estimation
  • Edge AI deployment for embedded systems

Türkçe

Volt-Guardian XAI

Volt-Guardian XAI, elektrikli araç (EV) bataryalarının Sağlık Durumu (State of Health - SoH) tahmini ve yorumlanması amacıyla geliştirilmiş bir Açıklanabilir Yapay Zeka (XAI) sistemidir.

Bu proje;

  • Genetik Algoritma (GA)
  • XGBoost
  • SHAP
  • Özellik Mühendisliği
  • Model Karşılaştırma Analizleri

yaklaşımlarını bir araya getirerek yüksek doğrulukta tahminler üretmektedir.

Öne Çıkan Özellikler

  • Hibrit GA-XGBoost mimarisi
  • SHAP tabanlı açıklanabilir yapay zeka
  • Gelişmiş özellik mühendisliği
  • Çoklu model karşılaştırmaları
  • Akademik düzeyde performans değerlendirmesi

Performans

  • R²: 0.9891
  • MAE: 0.2650
  • RMSE: 0.3404
  • MAPE: %0.28

Kullanılan Teknolojiler

  • Python
  • XGBoost
  • SHAP
  • Scikit-Learn
  • sklearn-genetic-opt
  • Pandas
  • NumPy
  • Matplotlib
  • Seaborn
  • Google Colab

Contributors

Name
Eren OĞAN
Enes GÜZEL
Esengül VELET
Yiğit Mert YILMAZ

Citation

If you use this work in academic research, please cite the associated publication and repository.


Volt-Guardian XAI — Making EV Battery Intelligence Explainable

About

Repository for "Volt-Guardian XAI" (Accepted at ICAIA 2026). A hybrid GA-XGBoost and SHAP framework for predicting and explaining Electric Vehicle (EV) battery State of Health (SoH).

Topics

Resources

License

Stars

Watchers

Forks

Contributors