Original Article Proceedings of 7 th I SEA CONFERENCE 2008 Biarritz, June 2-6, 2008 A Dynamic Heart Rate Prediction Model for Training Optimization in Cycling Ankang Le 1* , Thomas Jaitner 2 , Frank Tobias 3 , Lothar Litz 4 (1), (4) : Institute of Automatic Control, University of Kaiserslautern, Germany +49-(0)631-205-4459 / +49-(0)631-205-4462 E-mail: {le, litz}@eit.uni-kl.de (2), (3) : Department of Social Sciences, University of Kaiserslautern, Germany +49-(0)631-205-4967 E-mail: jaitner@sowi.uni-kl.de Topics: Bicycle, Modelling. ABSTRACT: Heart rate can be considered as a reliable indicator of the physiological load both for immediate training monitoring and for post-training analysis in cycling. The aim of this paper is to present a dynamic heart rate prediction model which will be used by a model predictive controller to optimize the cycling training. This model predicts the heart rate of a cyclist online during training or competition based on the physical dynamics of the heart rate to exercise work load. It uses eight parameters to calculate the future heart rate from current values, exercise duration and exercise work load by taking consideration of other effects such as fatigue, exhaustion and recovery. These parameters are identified from training data of nine well-trained cyclists by the least squares method. Each cyclist performed first a stepwise incremental test on a bicycle ergometer to determine his individual anaerobic threshold. Afterwards, they executed two interval tests on the same bicycle ergometer according to their individual anaerobic thresholds for the identification and evaluation of the model parameters. For all subjects, the mean absolute error and standard deviation between the measured and modelled heart rate values without updating are 3.06 and 3.95 bpm respectively. The mean correlation coefficient is 0.9686. If the model output is updated with the measured values every 20 seconds, then the mean absolute error is 1.31 bpm, the standard deviation is 1.92 bpm and the mean correlation coefficient is 0.9907. The result indicates that this model is able to predict the heart rate of cyclists accurately and can be used by a model predictive controller for training optimization. Key words: heart rate prediction model, training optimization, cycling, exercise intensity. 1- I ntroduction The exercise intensity is the most crucial factor to improve a cyclist’s performance. If the exercise intensity is too low, performance will not be improved or even reach a lower level. Excessive training at high intensities may lead to illness or overtraining (Kuipers and Keizer, 1988). Therefore, it is important to monitor the exercise intensity during training. When determining the best way to monitor exercise intensity, a balance has to be found between the validity of the parameter and practicality of using that parameter for intensity monitoring (Achten and Jeukendrup, 2003, Jeukendrup and Van Diemen, 1998). Recent developments of sensor technologies allow to monitor the exercise intensity by measuring the heart rate (HR) as well as the power output of the cyclist during the training or competition for immediate feedback to the athlete and for later analysis. Compared with other indicators of exercise intensity such as speed, cadence or percentage of maximal oxygen uptake, HR and power output are reliable and easy to monitor, and can be used in most situations (Achten and Jeukendrup, 2003, Faria et al. 2005, Gilman 1996). Furthermore, HR plays an important role in the detection and prevention of overtraining (Jeukendrup and Van Diemen, 1998). Some previous studies have modelled the HR with monoexponential or biexponential equations. HR was expressed as the sum of a baseline value plus one or two first order e-function with time delay (Bearden and Moffatt, 2001, Mavrommataki et al. 2006, Linnarsson 1974). However, in those models the work load values were not directly considered by the calculation of the heart rates. Those studies were conducted with constant work loads, moderate or heavy exercise intensities. They did not describe the dynamic responses of HR to the varying workloads quantitatively. Furthermore, the cardiac drift effect was not taken into consideration. Heart rates have been shown to drift upwards up to 20 beats per minute (bpm) during exercises lasting 20-60 minutes, despite unchanged work loads and steady or decreasing plasma lactate concentrations (Kindermann et al. 1979, Mognoni et al. 1990). Another study has shown that while exercising at a HR which is 5% below the HR at the ventilatory threshold, the work load has to be reduced by approximately 17% (from about 220 to 183 watt) over time, although the HR was kept relatively stable (176-180 bpm) (Boulay et al. 1997). Cardiac drift is accentuated by numerous factors such as dehydration and head stress and therefore an important factor for the HR modelling (Achten and Jeukendrup, 2003). ISEA2008_P83 - 1- Copyright of ISEA 2008 CONFERENCE