A TCN-Linear Hybrid Model for Chaotic Time Series Forecasting.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Deep learning, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), is crucial for time series prediction.
• Transformer networks, while popular, face challenges with self-attention mechanisms in long time series forecasting (LTSF).
• Existing models struggle to effectively address the complexities of LTSF, necessitating innovative solutions.

Purpose of the Study:

• To introduce a novel hybrid network, Temporal Convolutional Network-Linear (TCN-Linear), for enhanced long time series forecasting.
• To address the limitations of current deep learning models in LTSF tasks.
• To evaluate the performance of TCN-Linear against established and hybrid models.

Main Methods:

• Developed a hybrid network combining Temporal Convolutional Network (TCN) for temporal prediction and a linear component.
• Utilized TCN's predictive capabilities to enhance the LSTF-Linear model.
• Conducted experiments on time series data from three chaotic systems (Lorenz, Mackey-Glass, Rossler) and real-world stock data.

Main Results:

• The TCN-Linear model achieved the lowest Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Squared Error (MSE).
• The proposed model demonstrated superior performance compared to classical networks and other novel hybrid models.
• TCN-Linear achieved the best R-squared (R²) value, closest to 1, indicating high prediction accuracy.
• The model requires fewer training parameters while delivering enhanced forecasting capabilities.

Conclusions:

• The TCN-Linear hybrid network represents a significant advancement in long time series forecasting.
• This novel approach effectively overcomes the limitations of existing deep learning models for LTSF.
• The TCN-Linear model offers a more accurate and efficient solution for complex time series prediction tasks.