The existing intrusion detection methods for the power Internet of Things have problems such as high false alarm rates and inaccurate detection results. This paper proposes a reinforcement learning based intrusion detection method for the power Internet of Things. Using a sniffer to capture traffic data in the power Internet of Things, normalizing it using the average absolute deviation change method, implementing traffic data reduction using rough set theory, removing redundant data using HAMPLE, extracting traffic data features using reinforcement learning algorithms, and training to complete intrusion behavior detection. Experimental results have shown that the average F1 score value of the proposed method is 93.15%, with an average false positive rate of 0.58%. Compared with traditional methods, the detection accuracy of the proposed method is higher, providing a reference for intrusion detection in the power Internet of Things.