Sensor-based Training Optimization in Professional Cycling by Model Predictive Control
Sensorbasierte Trainingsoptimierung im Profi-Radsport mittels einer modellprädiktiven Regelung
This work aims to develop a control solution for an Assisted Bicycle Trainer system which demonstrates the application of Ambient Intelligence (AmI) technology in the area of sports and recreation prototypically by means of the cycling group training effect optimization. Based on mathematical modeling, a cycling intensity model is derived first to predict the cyclist’s power demand according to his position within the group. Second, according to the cyclist’s individual performance level a heart rate model is derived to predict his physiological stress under the physical workload. Based on these models, a non-linear multivariable and multi-objective Model Predictive Control (MPC) strategy is developed. While developing the control strategy, particular focus is set on the typical AmI constraints such as wireless network induced data transmission problems and human in the loop. The performance of the developed MPC strategy is evaluated by comparing cycling training simulations and real-life outdoor trainings to a rule-based strategy which is derived from the standard training methods used presently by cycling coaches. The evaluation results illustrate that the developed MPC can significantly improve the cycling group training effect.