📘 TriTrackNet: A Dual-Channel Time Series Forecasting Model

TriTrackNet is a dual-channel time series forecasting model with multi-path interaction and perturbation-based optimization.
It integrates long-range temporal dependencies (main channel) and cross-dimensional interactions (auxiliary channel), while using PerturbOpt to enhance robustness.

This repository provides the implementation and benchmark experiments on several widely used datasets (ETT, Weather, Traffic, Electricity).

---

🚀 Features

• Dual-channel architecture (main + auxiliary path)
• Multi-path attention for temporal and feature dependencies
• PerturbOpt adversarial optimizer for robustness
• Extensive experiments on 7 benchmark datasets

---

📂 Project Structure

TriTrackNet/
│
├── dataset/                # Benchmark datasets (.csv format)
│   ├── ETTh1.csv
│   ├── ETTh2.csv
│   ├── ETTm1.csv
│   ├── ETTm2.csv
│   ├── WTH.csv
│   ├── ECL.csv
│   ├── traffic.csv
│   └── exchange.csv
│
├── TriTrackNet/            # Model code
│   ├── TriTrackNet.py      # Main model
│   └── utils/
│       ├── attention.py    # Attention modules
│       ├── dataset.py      # Data loader
│       ├── perturbopt.py   # PerturbOpt optimizer
│       ├── revin.py        # RevIN module
│       └── __init__.py
│
├── run_ECL.py              # Run script for Electricity dataset
├── run_Traffic.py          # Run script for Traffic dataset
├── run_WTH.py              # Run script for Weather dataset
├── run_ETTm1.py            # Run script for ETTm1 dataset
├── run_ETTm2.py            # Run script for ETTm2 dataset
├── run_ETTh1.py            # Run script for ETTh1 dataset
├── run_ETTh2.py            # Run script for ETTh2 dataset
├── run_exchange.py         # Run script for Exchange dataset
│
├── requirements.txt        # Dependencies
└── README.md               # Project documentation

---

⚙️ Installation

# Clone the repository
git clone https://github.com/yourname/TriTrackNet.git
cd TriTrackNet

# Install dependencies
pip install -r requirements.txt

---

📊 Datasets

The following benchmark datasets are included in the dataset/ folder:

• ETT (ETTh1, ETTh2, ETTm1, ETTm2) – Oil temperature related time series
• Weather – Meteorological variables
• Electricity (ECL) – Regional load series
• Traffic – Road traffic data
• Exchange – Exchange rates

---

▶️ Running Experiments

Each dataset has its own run script. For example:

# Run on ETTh1 dataset
python run_ETTh1.py

# Run on Weather dataset
python run_WTH.py

# Run on Traffic dataset
python run_Traffic.py

All scripts will automatically load the corresponding dataset from dataset/ and train/test TriTrackNet.

---

📈 Results

Our experimental results demonstrate that TriTrackNet consistently outperforms baseline models such as Informer, Autoformer, iTransformer, PatchTST, and DLinear, especially on high-dimensional datasets (Traffic, Electricity).
See the paper for full benchmark results.