Employing an LSTM model with a self-attention mechanism to model variations in running pace

During the process of running, the pace varies with changes in time. Experienced runners adjust their pace by modifying stride length and frequency. To simulate the relationship between stride length, frequency, and pacing throughout a run using neural network methods, we can leverage the temporal patterns present in the data. The unique memory capabilities of Long Short-Term Memory (LSTM) networks make them well-suited for this task. To enhance the predictive accuracy of the LSTM model, we incorporate a self-attention mechanism that improves the model's ability to associate information from different positions within the input data. This self-attention mechanism enables the model to identify which parts of the data are more significant and thus allocate greater focus during predictions, ultimately enhancing performance. The LSTM method augmented with a self-attention mechanism can accurately fit changes in running pace based on historical exercise data, providing valuable insights for practitioners in running sports. Experimental results indicate that this model achieves a fitting error on the order of 10-4.