ECG-Based Energy Expenditure Prediction Using BiLSTM With Improved Snow Ablation Optimizer.

Area of Science:

• Biomedical Engineering
• Human-Machine Interface (HMI)
• Wearable Technology

Background:

• Accurate energy expenditure (EE) estimation is crucial for optimizing motion control in human-in-the-loop (HIL) systems.
• Current methods may lack accuracy across diverse movement conditions.
• Low-cost wearable devices offer potential for real-time EE monitoring.

Purpose of the Study:

• To propose a novel method for accurate EE estimation using a low-cost wearable electrocardiogram (ECG) device.
• To analyze the impact of various factors on EE estimation accuracy.
• To optimize hyperparameters and input window length for EE prediction in diverse motion conditions.

Main Methods:

• Developed an improved snow ablation optimizer (ISAO) to optimize BiLSTM network hyperparameters and input time series window length.
• Utilized ECG signals and breath-by-breath EE data, optimizing window length for both high-frequency and low-frequency components.
• Evaluated the model using test sets with multiple time intervals, simulating real online testing scenarios.

Main Results:

• The proposed ISAO-BiLSTM network achieved accurate EE prediction across various motion paradigms (RMSE: 56 W; R²: 0.93).
• ECG-based deep learning framework demonstrated superior EE prediction accuracy compared to heart rate (HR) based methods.
• Optimized input data window length is suitable for real-time EE predictions in HIL HMI applications.

Conclusions:

• The ISAO-BiLSTM framework provides a robust and accurate method for estimating energy expenditure using ECG data.
• Optimized input window length enhances the applicability of real-time EE prediction in human-machine interface systems.
• This approach offers a valuable tool for improving motion control strategies in HIL applications.