With the improvement of the network technology, power dispatching systems (PDS) are increasingly attacked from the Internet. Due to the limitations and regulations in laws and local management requirements, it is difficult for such systems to form a joint force to detect the security status while ensuring the privacy of each system's data. Aiming at this issue, this paper proposes a blockchain based federal abnormal detection method for power dispatching systems. The method provides a tree based federal-privacy-training scheme to aggregate data from various power scheduling systems, and generates a model with higher detection accuracy on the basis of not leaking the plaintext of these data. Moreover, the blockchain is used to connect dispatching systems across areas to enable the secure transfer of data between these systems. The experimental results show that this method not only improves the accuracy for detecting attacks, but also ensures the privacy of the data among PDSs.