In order to realize the fault diagnosis of AGV gyroscope, a fault diagnosis method of AGV based on VMD and LSTM neural network model is proposed. Firstly, krill swarm algorithm is used to optimize the parameters of penalty factor α and modal component K in VMD. Then the optimized VMD decomposes the gyro fault parameters; Finally, the decomposed modal component K is input into the LSTM network for feature learning, to realize the fault identification of AGV gyroscope. Experiments show that this method has achieved the accuracy of 98.8% in identifying two common gyro faults: bias fault and drift fault, and effectively improved the accuracy of AGV gyro fault identification.