With the rapid development of the Power Internet of Things, under the guidance of the "cloud-pipe-edge-terminal" construction system, this paper presents a general technical framework for operation state evaluation and trend prediction of distribution transformer. The framework is deployed in cloud center and edge nodes respectively, and use the cloud edge collaboration mechanism to analyze and process massive power data, so as to complete the operation management of large-scale distribution transformer cluster. The specific process includes extracting multi-dimensional characteristics of distribution transformer, such as basic state, real-time state and cumulative state, constructing evaluation index system, realizing real-time portrait description of distribution transformer operation state through dynamic evaluation model. According to the time order and change trend of the characteristic data stream, Long Short-Term Memory network (LSTM) is used to analysis regulation of characteristic data, and Support Vector Regression model (SVR) for its prediction. The future characteristic data flow is obtained, which is input into the dynamic evaluation model to realize the future operation trend prediction of the distribution transformer. Finally, examples are given to illustrate the advanced nature and applicability of the technology framework.