Abstract: In practical array systems, challenges such as array errors, grid mismatch, and limited snapshots are prevalent. This study investigates a passive target direction-of-arrival (DOA) estimation method based on sparse Bayesian learning in such complex environments. By characterizing the array error model and the sparse Bayesian framework, a gradient descent-based grid variational Bayesian algorithm incorporating an adaptive strategy at the covariance estimation level is proposed. Theoretical analysis demonstrates the global convergence of the algorithm. Simulation experiments further evaluate its performance from multiple perspectives, including estimation accuracy, resolution capability, and computational efficiency, with an analysis of its robustness under varying signal-to-noise ratios, snapshot counts, and array error conditions.