The complexity of management of modern datacenter is rapidly growing. It is emergent that the datacenter management system is able to automatically analyze datacenter status and provide intelligence for workload orchestration and infrastructure management. The state of the art algorithms in machine leaning technology makes possible to conduct real-time data analysis from huge quantities of telemetries or sensor data from datacenter. This paper introduces one supervised machine learning approach to analyze and predict group level power variation trend with minute granularity. The minute granularity prediction is meaningful for workload migration or datacenter cooling management. Two power variation trend models are defined in this paper, consistent direction model and abrupt shift model. The support vector machine is adopted to classify the power relevant feature sampling points into these two models. For each model, a different regression algorithm is used to calculate the coefficients of each regression equation and to estimate the power variation trend accordingly. At the end of this paper, we have evaluated the power variation trend prediction model with real data from the production environment.