Currently, provincial grid companies lack effective methods of predicting the demand of electric energy meters, which may increase the storage and dispatching cost. To solve this problem, this paper proposes a demand prediction model of electric energy meter base on service feature. This model classifies electric energy meters by installation types, then use stationarity test to determine the type of main factors that influence the electric energy meter demand of each installation type, thus adaptively select between ARIMA or LSTM model to analyze and predict the demand. The result of case analysis shows that this model can predict with higher accuracy than existing models.