Data collection, analysis and processing is one of the most important problems in power systems. There are various requirements for data business processing in the power internet of things (IoT) systems. In particular, there are strict requirements on delay for stream data processing. However, the existing data processing methods are single in form and cannot well meet the requirements of low-delay processing. Therefore, a stream data processing method of power IoT based on Storm architecture is proposed. First, a distributed stream data processing framework is proposed based on Storm topology. Then the pipeline-based processing method is adopted to shorten the processing time. Then data access, a circular queue and a flow operator are adopted. Then, through the edge computing processing of massive data, efficient collaborative work is achieved. In addition, the support vector machine prediction algorithm is used to predict the development trend of data, and the dirty data is processed by cleaning technology to solve the pollution in the process of data transmission. The simulation results show that, compared with the traditional data processing methods, the proposed method can greatly shorten data processing time and meet the needs of processing a large number of data streams. This work is supported by the National Natural Science Foundation of China (No. 61971190) and the Science and Technology Project of State Grid Hebei Electric Power Co., Ltd. (No. SGHEXT00GCJS2000167).