The aim of this study is to model the relations between final time and kinematic variables for the men's 400 medley. We used the 156-performance analysis of 16 international events. The fourth frequency stroke and laps times were studied using a specific competition analysis program. Modelling performance is an operation requiring the use of mathematical models being able to take into account the several interactions between laps times and stroke rates variables and the non-linear evolution of these variables. Neural networks (NN) are a technique with a great capacity for searching complex non-linear functions and variable interactions. They give us the possibility to understand hidden relations, usually non-linear, between parameters. Woolly inference system (WIS) is an expert knowledge model that contains vagueness information and functioning using if-so type algorithms. In the first experience plan, clustering method was used to increase intra-group homogeneity. In the second experience plan, we separated the data into three groups. Two-thirds were used for the learning process and the last third for the simulation. The comparisons of mean error during the learning phase are presented in Table 1. WIS results for the simulation seems to be better than those from NN. The predicting maximal error is 0.6 s, for the lap times (figure 1), and 1.6 c/min, for the stroke frequencies (figure 2). Once the models have been validated in terms of their coherence and quality, the functions modelling the response of swimmers could be used to simulate, in advance, the lap times and stroke frequencies corresponding to a predicted performance.
© Copyright 2001 Alle Rechte vorbehalten.