The technologies of algorithmic composition have made a good progress in the recent days, and the public is more and more paying attentions on exercise and health. This thesis develops a sport music style algorithmic composition system in order to reduce the fees of copyright and fatigues of exercise. The system contains a chord progression generator fulfilled with first-order Markov chain, an upper register part generator based on recurrent neural network model, a corpus-based drum set rhythm pattern generator, and a structure form generator performed with first-order Markov chain model with formulized accompaniment. The 9 subjects listening to music made by the sport music style algorithmic composition system feels more relaxed than listening to white noise during the aerobic exercise (p<0.05).