Fuzzy inference-based LSTM for long-term time series prediction.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Long Short-Term Memory (LSTM) networks face challenges in time series forecasting, including accumulated errors, reduced temporal correlation, and poor interpretability.
• These limitations hinder the performance of LSTM models for long-term predictions.

Purpose of the Study:

• To enhance the accuracy and interpretability of LSTM for long-term time series forecasting.
• To address the inherent limitations of traditional LSTM models.

Main Methods:

• A novel fuzzy inference-based LSTM is proposed, integrating a fuzzy system into the LSTM network.
• A fast and complete fuzzy rule construction method based on Wang-Mendel (WM) is introduced for efficient fuzzy rule simplification and complementation.
• Key components include fuzzy prediction fusion, a strengthening memory layer, and a parameter segmentation sharing strategy.

Main Results:

• The proposed fuzzy inference-based LSTM demonstrates improved prediction performance compared to existing models.
• The method effectively enhances the reasoning capability and interpretability of the LSTM network.
• Long-term memory is strengthened, and the gradient dispersion problem is alleviated.

Conclusions:

• The fuzzy inference-based LSTM offers a superior approach for long-term time series forecasting.
• The integration of fuzzy logic significantly boosts accuracy and interpretability.
• The model provides a more robust and understandable solution for complex time series prediction tasks.