Fuzzy theory and machine learning are applied in this paper. Through fuzzy clustering the number of arriving cars, the schemes of signal control are put into knowledge-database in the form of rule-set under different conditions of cars' arriving. The set of traffic control rules is divided into the set of fixed-rule and the set of variable-rule. The genetic algorithm is used to improve the set of variable-rule during the process of traffic signal control. The genetic algorithm is a part of the signal control process instead of calculating the optimal scheme of signal control. A self-learning traffic signal control model based on fuzzy clustering and genetic algorithm is established. The instance of simulation is an isolated intersection controlled by signal. After simulating, the control effects of this self-learning approach，the fixed-time control method and the actuated control method are compared. The result of simulating illustrates that the effect of this self-learning approach is better than the traditional ones.