The invention proposes an online diagnosis method of OLTC contact terminal faults based on the chaos theory and GA-Kmeans. Firstly, in view of the nonlinearity and chaos of OLTC signals, the P G method and mutual information method are used to calculate the embedding dimension and delay time, an embedded dimension and a delay time are calculated by using a P-G method and a mutual information method, chaotic feature judgment is performed on OLTC by using the Kolmogorov entropy, and OLTC phase space reconstruction is performed to obtain phase space characteristic distribution of an OLTC vibration signal. In order to improve the accuracy of initial clustering of Kmeans, the genetic algorithm (Genetic Algorithm) is introduced into the Kmeans algorithm to perform optimization selection on an initial clustering center thereof, then the distribution of a characteristic cluster center of the phase space distribution of the OLTC vibration signal in a phase space is obtained based on the GA-Kmeans algorithm, and quantitative analysis is performed on the contact terminal faults through characteristic quantity. By adopting the diagnosis method, a basis can be provided for realizing quantitative analysis and diagnosis of the contact terminal faults.