Abstract: Advanced persistent threat (APT) has become one of the main threats to the network security of the new type power systems. Because of the features like strong concealment, destructive power and long duration, the existing traditional detection methods can not meet the security requirements of the new type power systems. Therefore, an APT attack detection method using parallel channel and spatial attention mechanism based convolutional neural network (PCSA-CNN) is proposed. The parallel channel and spatial attention mechanism is introduced to highlight the characteristics of APT attack data and generate the corresponding eigenvector matrix, and then a convolutional neural network model is used to detect APT attack. The experiment results indicate that PCSA-CNN model can reach 99.87% accuracy, which is significantly better than the existing mainstream neural network model.