Prediction intervals for a noisy nonlinear time series based on a bootstrapping reservoir computing network ensemble.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Time Series Analysis

Background:

• Accurate prediction intervals are crucial for time series forecasting, but challenging for noisy data.
• Existing methods struggle to provide reliable reliability estimates for nonlinear time series.

Purpose of the Study:

• To propose a novel method for constructing reliable prediction intervals in nonlinear time series forecasting.
• To address the challenge of generating accurate reliability estimates for noisy time series data.

Main Methods:

• A bootstrapping reservoir computing network ensemble (BRCNE) was developed.
• A simultaneous training method using Bayesian linear regression was employed.
• Structural parameters were optimized using 0.632 bootstrap cross-validation.

Main Results:

• The BRCNE demonstrated satisfactory performance in generating prediction intervals.
• The method proved effective for both synthetic noisy data and a practical industrial gas flow dataset.

Conclusions:

• The proposed BRCNE with Bayesian linear regression offers a robust solution for reliable prediction intervals in nonlinear time series.
• This approach shows significant potential for practical applications requiring accurate forecasting with reliability estimates.