MixedLinear: A DLinear and TimeMixer-Based Ensemble Model for Efficient Electricity Demand Forecasting
An innovative ensemble model combining DLinear and TimeMixer components for high-accuracy, computationally efficient electricity demand forecasting.
This project introduces MixedLinear, a novel ensemble approach that leverages the strengths of both DLinear and TimeMixer architectures for enhanced electricity demand forecasting performance.
-
DLinear: "Are Transformers Effective for Time Series Forecasting?" (AAAI 2023)
-
TimeMixer: "TimeMixer: Decomposable Multiscale Mixing for Time Series Forecasting" (ICLR 2024)
MixedLinear combines the computational efficiency of DLinear with TimeMixer's ability to handle complex patterns. The model demonstrates superior performance in electricity demand forecasting while maintaining resource efficiency.
Mixed_Linear/
│
├── models/ # Model definitions
│ ├── Mixed_Linear.py # Main Mixed Linear model
│ ├── DLinear.py # DLinear model
│ ├── NLinear.py # NLinear model
│ ├── Linear.py # Linear model
│ ├── TimeMixer.py # TimeMixer model
│ └── Transformer.py # Transformer model
│
├── layers/ # Model layers
│ ├── Autoformer_EncDec.py
│ ├── Embed.py
│ ├── SelfAttention_Family.py
│ ├── StandardNorm.py
│ └── Transformer_EncDec.py
│
├── exp/ # Experiment framework
│ ├── exp_main.py # Main experiment class
│ └── exp_basic.py # Base experiment class
│
├── data_provider/ # Data loaders
│ ├── data_factory.py
│ └── data_loader.py
│
├── utils/ # Model training/evaluation utilities
│ ├── tools.py # Learning rate adjustment, EarlyStopping
│ ├── metrics.py # Evaluation metrics (MAE, MSE, RMSE, etc.)
│ ├── timefeatures.py # Time feature extraction
│ └── masking.py # Masking utilities
│
├── utils2/ # Project-specific utilities
│ ├── Data_run.py # Unified model execution script
│ └── Date_Processor.py # Date filtering, preprocessing, PCA
│
├── notebooks/
│ ├── 01_Data_Collection.ipynb # Data collection
│ ├── 02_Data_Preprocessing.ipynb # Data preprocessing
│ ├── 03_Model_Training.ipynb # Model training (main)
│ ├── 04_Additional_Data_Processing.ipynb # Additional processing
│ ├── 05_Result_Visualization.ipynb # Result visualization
│ ├── EDA.ipynb # Exploratory data analysis
│ └── log_reader.ipynb # Log file reader
│
├── README.md # This file
├── requirements.txt # Python dependencies
└── LICENSE # MIT License
# Clone the repository
git clone https://github.com/jinsoo96/MixedLinear_Time_Series_Forecasting.git
cd MixedLinear_Time_Series_Forecasting
# Create virtual environment (recommended)
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
# Install dependencies
pip install -r requirements.txtUsing Jupyter Notebook:
# Open notebooks in sequence:
# 01_Data_Collection.ipynb # Step 1: Data collection
# 02_Data_Preprocessing.ipynb # Step 2: Data preprocessing
# 03_Model_Training.ipynb # Step 3: Model training (main)
# 04_Additional_Data_Processing.ipynb # Step 4: Additional processing
# 05_Result_Visualization.ipynb # Step 5: Result visualizationUsing Python Script:
from utils2.Data_run import arg_set, model_run
# Configure experiment
args = arg_set(
folder_path='./Data_Final/1_year',
data='your_data.csv',
model_name='Mixed_Linear_experiment',
pred_len=72, # Prediction horizon
label_len=72, # Label length
num_workers=10 # Data loader workers
)
# Run model
model_run(args)- Python 3.7+
- PyTorch 1.8+
- NumPy
- Pandas
- scikit-learn
- matplotlib
See requirements.txt for full dependencies.
This project is licensed under the MIT License - see the LICENSE file for details.
- Author: Jin Soo Kim
- GitHub: @jinsoo96
Copyright © 2025 Jin Soo Kim All rights reserved.