Successful detection of false data injection(FDI) attacks are essential for ensuring secure power grids operation. However, most work simulates real intrusion behaviors by establishing FDI attack models, and the simulated data obtained is often different from the real data, resulting in poor learning effects based on machine learning detection methods. Motivated by this fact, considering the large amount of simulated sample data in the source domain and a small number of labeled real samples in the target domain, a detection algorithm based on the deep belief network(DBN) and transfer learning is proposed. The restrict boltzmann machine(RBM) in the DBN can automatically extract features from a large number of unlabeled samples in the target domain, and the model-based transfer learning method overcomes the differences between data and solves the problem of the scarcity of labeled real samples. Finally, the IEEE 14-bus power system is employed to show the advantages and effectiveness of the proposed method.