In view of the low classification accuracy in the identification of heavy duty locomotive adhesion state, a genetic algorithm based on cuckoo has been proposed to optimize the parameters of least squares support vector machines, with the cross validation method adopted to improve the overall generalization performance of the model. First, the cuckoo algorithm is used to find the initial values of penalty parameters and kernel functions. Next, the genetic algorithm is used to train the least squares support vector machines (SVM), thus obtaining the best parameters of the least squares support vector machines (SVM) model. Under this classification model, the adhesion states of heavy duty locomotive can be divided into four categories: normal condition, fault symptom state, minor fault state and serious fault state. Experimental results show that the classification accuracy of the proposed least squares support vector machine model can reach as high as 94.59%, much higher than that of the limit learning machines with its classification accuracy only being 84.61%. Therefore it is proved that the genetic algorithm can effectively improve the classification accuracy of the least squares support vector machines.