Aiming at the problem that it is difficult to extract fault features from bearing vibration signals, a fault classification method based on the improved whale algorithm for optimizing the least square support vector machine(LSSVM) model is proposed. Firstly, the original signal is decomposed by variational modal decomposition, and the center frequency method is used to solve the problem of selecting the decomposition parameter $ K $ in VMD. Then, we calculate the multi-scale permutation entropy value of each IMF component and extract signal fault characteristics. Furthermore, aiming at the slow convergence speed and low accuracy of the whale optimization algorithm(WOA), the von-neumann and adaptive weights are introduced to improve the whale optimization algorithm, which can appropriately adjust the balance between global search ability and local search ability. Finally, by using an improved whale optimization algorithm, the penalty factor and kernel parameter of the LSSVM are optimized to establish the fault diagnosis model of rolling bearing, and the bearing data set provided by Case Western Reserve University is used to perform simulation experiments. The results show that the proposed method has better fault classification performance and higher accuracy.