The invention provides a terminal energy consumption prediction method and device. The method comprises the steps of acquiring energy consumption data; utilizing the energy consumption data for prediction through an electric energy substitution prediction model to acquire a first energy consumption prediction value and a first total energy consumption prediction value; utilizing the energy consumption data for prediction through a grey combined wavelet neural network model and a nonlinear regression model to acquire a second energy consumption prediction value, a second total energy consumption prediction value, a third energy consumption prediction value and a third total energy consumption prediction value; determining a virtual prediction value of a target year through terminal energy increments in target historical years; constructing a cross entropy model based on all the prediction values and determining a target energy consumption prediction value and a target total energy consumption prediction value through the cross entropy model. Through the terminal energy consumption prediction method and device, accurate prediction of the terminal electric energy consumption and total energy consumption in the prediction year is achieved.