Biomechanical data-driven prediction and analysis based on transformer model

Article ID: 1235

Abstract

With the development of high-precision sensors and data acquisition equipment, biomechanical data presents high-dimensional, strong time-series dependence and nonlinear characteristics, and it is difficult for traditional physical modeling and statistical methods to process such data efficiently and accurately. The purpose of this paper is to build a biomechanical data-driven prediction framework based on Transformer model, and realize high-precision prediction by deeply mining the time series characteristics of data, which provides theoretical support and practical application value for medical diagnosis, rehabilitation monitoring and sports science. In terms of methods, this paper preprocesses biomechanical data such as joint angle, electromyography (EMG) and joint stress, and designs a time series prediction framework based on the self-attention mechanism of Transformer model. Through the simulation experiment, five indexes, namely mean square error (MSE), mean absolute error (MAE), determination coefficient (R²), prediction time and Pearson correlation coefficient, are selected to evaluate and compare the performance of the model. The experimental results show that the Transformer model is superior to the traditional LSTM, GRU and ARIMA models in all kinds of biomechanical data prediction tasks: MSE is 0.0152, R² is as high as 0.982, and the prediction time is only 0.76 s. In addition, Pearson correlation coefficient is close to 1 in different data types, which verifies the high consistency between the predicted results of the model and the real values. The conclusion of this paper shows that the Transformer model can effectively capture the global spatio-temporal characteristics of biomechanical data, and has high precision, high efficiency and strong generalization ability, which provides new technical means and theoretical support for biomechanical data-driven analysis and application.