Abstract: False data injection attack（FDIA） has become a serious threat to the safe and stable operation of power grids due to its high concealment and destructiveness. The imbalance characteristics between attack samples and normal samples affect the attack detection accuracy of the model. Meanwhile, the emergence of multi-type FDIA makes existing algorithms have limitations in identifying attack types. Aiming at the above problems, this paper proposes a multi-type FDIA identification method for power grids based on oversampling and cascade machine learning. First, the FDIA attack path in the power grid coupling interaction process is explored, and multi-type attack behaviors are analyzed. Then, the attack pseudo-samples are generated through the process of clustering, filtering and linear interpolation, and the measurement data balance algorithm based on K-means-synthetic minority oversampling technique（K-means-Smote） is designed. Finally, a multi-type FDIA identification model with improved cascade machine learning is constructed by combining fine-grained feature scanning and an ensemble learning strategy of multiple classifiers. Simulation experiments show that the proposed identification method can effectively identify a variety of FDIA types,with high identification accuracy, low false alarm rate, stable and excellent performance, even for small samples.