Autoreservoir computing for multistep ahead prediction based on the spatiotemporal information transformation

Area of Science:

• Artificial Intelligence
• Machine Learning
• Dynamical Systems

Background:

• Traditional reservoir computing uses external dynamical systems.
• High-dimensional time series prediction is challenging.
• Accurate multi-step-ahead forecasting is crucial for many applications.

Purpose of the Study:

• To develop an efficient and accurate framework for multi-step-ahead predictions.
• To address limitations of traditional reservoir computing.
• To leverage high-dimensional time series data effectively.

Main Methods:

• Developed the Auto-Reservoir Neural Network (ARNN) framework.
• ARNN directly uses observed high-dimensional dynamics as its reservoir.
• Employed a spatiotemporal information (STI) transformation for mapping data.

Main Results:

• ARNN achieved satisfactory performance in multi-step-ahead predictions.
• The framework demonstrated robustness against noise and time-varying systems.
• Successful application to both model datasets and real-world data.

Conclusions:

• ARNN provides an accurate and computationally efficient method for time series prediction.
• The ARNN transformation effectively expands sample size, enhancing model performance.
• ARNN shows great potential for practical applications in AI and machine learning.