☀️ InverterShield — AI-Driven Solar Inverter Failure Prediction

Predictive maintenance platform for solar inverters using XGBoost, SHAP explainability, and Generative AI insights.

Architecture

┌──────────────────┐       ┌──────────────────┐       ┌──────────────────┐
│   React Frontend │──────▶│   FastAPI Backend │──────▶│    ML Models     │
│   (Port 5173)    │  API  │   (Port 8000)    │       │  XGBoost + SHAP  │
│                  │       │                  │       │  Isolation Forest │
│  • Dashboard     │       │  • /predict      │       └──────────────────┘
│  • Predictions   │       │  • /dashboard    │                │
│  • Model Metrics │       │  • /chat         │       ┌────────▼─────────┐
│  • AI Copilot    │       │  • /health       │       │   GenAI Layer    │
└──────────────────┘       └──────────────────┘       │  Gemini / Ollama │
                                                      └──────────────────┘

Model Performance

Model	Metric	Score
Binary (Failure/No Fail)	CV Mean F1	0.918
Binary	CV Mean AUC	0.901
Binary	CV Precision	0.924
Binary	Holdout F1	0.908
Multi-Class (3-way)	CV F1 (macro)	0.789
Anomaly Detection	Isolation Forest	✅

Top SHAP Features: month, inv_temp_7d_mean, meter_kwh_import, inv_temp_24h_mean, smu_string_mean_7d_std

Dataset

• 3 solar plants · 6 CSV files · 32 inverters · ~6M raw rows
• 2-year span: March 2024 — March 2026
• Aggregated to hourly intervals → 509,128 rows × 140 features

Quick Start

Prerequisites

• Python 3.13+, Node.js 18+
• ~8GB RAM for training; inference < 2GB

Install & Run

# 1. Clone
git clone https://github.com/Megh-Rana/InverterShield_Solar_Inverter_Failure_Prediction.git
cd InverterShield_Solar_Inverter_Failure_Prediction

# 2. Train models (optional — pre-trained models included via LFS)
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt
python run_pipeline.py --dataset-dir dataset --output-dir models --save-processed

# 3. Start API server
PYTHONPATH=. python -m uvicorn src.api.main:app --host 0.0.0.0 --port 8000

# 4. Start React frontend (separate terminal)
cd frontend && npm install && npm run dev

Docker

docker-compose up --build
# API:      http://localhost:8000
# Frontend: http://localhost:5173

GenAI Setup (Optional)

export GEMINI_API_KEY=your_key_here    # For Gemini
ollama pull llama3                      # For local Ollama

API Endpoints

Endpoint	Method	Description
/health	GET	Health check + model status
/predict	POST	Single inverter prediction
/predict/batch	POST	Batch predictions
/dashboard	GET	Dashboard summary + SHAP
/chat	POST	GenAI chat (Gemini/Ollama)
/model/info	GET	Model metrics & feature count

Project Structure

├── frontend/                 # React + Vite dashboard
│   └── src/
│       ├── pages/            # Dashboard, Prediction, Performance, Chat
│       ├── api.js            # FastAPI client
│       ├── App.jsx           # Layout + routing
│       └── index.css         # Design system
├── src/
│   ├── api/main.py           # FastAPI backend
│   ├── genai/llm.py          # Gemini/Ollama integration
│   ├── ml/trainer.py         # XGBoost + SHAP + Isolation Forest
│   └── data_processing/      # CSV loader + feature engineering
├── models/                   # Trained models (Git LFS)
├── Dockerfile
├── docker-compose.yml
└── requirements.txt

ML Pipeline

1. Load — Normalize heterogeneous CSV schemas → unified per-inverter rows
2. Aggregate — 6M rows → 509K hourly rows (mean continuous, max alarms)
3. Engineer — 140 features: rolling stats, alarm features, KPIs, time features
4. Detect — Isolation Forest anomaly detection → scores as features
5. Validate — 5-fold walk-forward cross-validation (TimeSeriesSplit)
6. Train — XGBoost binary + multi-class classifiers
7. Explain — SHAP TreeExplainer for feature importance

Hackathon Checklist

Requirement	Status	Implementation
Failure prediction model	✅	XGBoost binary + multi-class
7-10 day prediction window	✅	7-day forward-looking target
Feature engineering	✅	140 features
Cross-validation	✅	Walk-forward 5-fold
Explainability (SHAP/LIME)	✅	SHAP TreeExplainer
Anomaly detection	✅	Isolation Forest
GenAI summaries	✅	Gemini 2.0 Flash + Ollama fallback
REST API	✅	FastAPI (6 endpoints)
Interactive dashboard	✅	React + Recharts
Docker containerization	✅	Dockerfile + docker-compose

Tech Stack

ML: XGBoost · scikit-learn · SHAP · pandas · numpy
API: FastAPI · Uvicorn
Frontend: React · Vite · Recharts · Framer Motion · Lucide
GenAI: Google Gemini 2.0 Flash · Ollama
Infra: Docker · docker-compose · Git LFS