Energy consumption analysis and prediction in exercise training based on accelerometer sensors and deep learning

This study aims to enhance the accuracy and efficiency of energy consumption prediction during exercise training and address the limitations of existing methods in terms of data feature extraction, model complexity, and adaptability to practical applications. This study proposes an optimized energy consumption prediction model based on accelerometer sensor data and deep learning techniques. In this study, a model architecture integrating Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (Bi-LSTM) network, and an attention mechanism is constructed, with a focus on optimizing local feature extraction, temporal modeling, and dynamic weight allocation capabilities. Additionally, by analyzing the relationship between the X, Y, and Z-axis accelerations, overall magnitude, and energy consumption, a multidimensional feature analysis framework is proposed to enhance the model's comprehensive understanding of motion data. To verify the performance of the model, performance comparison experiments and ablation experiments are designed. The experimental results demonstrate that the optimized model achieves a Mean Squared Error (MSE) of 0.273, an R² of 0.887, and a standard deviation of 0.046 on acceleration data, significantly outperforming comparison models such as Temporal Convolutional Network (TCN), Gated Recurrent Unit with Attention Mechanism (GRU-ATT), and Self-Supervised Transformer (SST). Furthermore, ablation experiments reveal that the synergistic effects of the convolutional network, Bi-LSTM, and attention mechanism significantly improve prediction accuracy and model robustness. Further analysis shows that the optimized model achieves a correlation of 0.829 between overall magnitude and energy consumption, validating its ability to capture complex motion features. Therefore, this study provides an efficient, accurate, and highly adaptable solution for the field of energy consumption prediction in exercise, contributing to research on intelligent motion monitoring, health management, and personalized training program development.