HOES: an efficient multi-evolutionary expert system for deep learning model optimization in time series prediction.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Deep learning excels at time series prediction but requires effective training strategies.
• Traditional optimization methods often lack efficiency for complex deep learning models.
• Optimizing deep learning models is crucial for accurate time series forecasting.

Purpose of the Study:

• To develop a novel Hybrid Optimization Expert System (HOES) for optimizing deep learning models in time series prediction.
• To enhance the efficiency and effectiveness of deep learning model training through advanced optimization techniques.
• To improve the predictive accuracy and convergence rates of time series forecasting models.

Main Methods:

• Designed a Hybrid Optimization Expert System (HOES) integrating six evolutionary algorithms.
• Implemented a transmission mechanism for enhanced global search capability.
• Incorporated a memory system to preserve optimal solutions and a punishment system to eliminate ineffective strategies.
• Utilized SJ-LSTM as a representative model for validating HOES performance on diverse datasets.

Main Results:

• HOES demonstrated significant improvements in predictive accuracy and convergence rates across six public datasets.
• Optimized SJ-LSTM using HOES achieved a 24% RMSE and 30% MAE reduction on the Solar Power dataset.
• The system effectively mitigated the risk of local optima, enhancing global optimization capabilities.

Conclusions:

• HOES significantly boosts the global optimization capability of deep learning models for time series prediction.
• The proposed system offers an efficient and effective solution for complex time series forecasting tasks.
• HOES provides a robust framework for improving the performance of deep learning models through advanced evolutionary optimization.