Improve The LSTM Model For Predicting Road Surface Friction Coefficient

The variation in road surface friction coefficient has a crucial impact on safe driving. Existing methods for measuring friction coefficient, such as Long Short-Term Memory (LSTM) networks, suffer from slow convergence and low accuracy. To address these issues, a multi-head attention mechanism algorithm is introduced to optimize and enhance the LSTM network. The final model is a network with 8 layers and 100 neurons, designed for predicting road surface friction coefficients. Experimental simulations comparing the LSTM prediction algorithm show that the LSTM-MHAM model, after 100 training iterations, reduces the average absolute error by 49.4%, decreases the mean squared error by 51%, and lowers the average percentage prediction error by 29.6%. In comparison to the LSTM algorithm, this approach significantly improves convergence speed and prediction accuracy.