A Bayesian learning-based actuator fault estimation method, where modeling is performed on an actuator fault on the basis of a random walk model, and a joint posterior distribution of a system state variable and an actuator fault signal are represented using two mutually independent hypothetical distributions on the basis of variational Bayesian theory; at time k-1, a state and a fault for a system at time k are predicted; at time k, iterative updating is performed on the predicted system state and system fault according to Bayesian theory, an estimate value for the system state at time k and and a variance of the estimate value are output, and an estimate value for the system fault at time k and a variance of the estimate value are output. The present invention fully utilizes Bayesian learning adapted for an online estimation structure, providing an actuator fault signal estimation method under a stochastic system by means of decoupling the system state and fault signal having a mutually coupled variable, which can amply perform estimation on the actuator fault signal.
展开▼