Urban building energy consumption is one important part of the total social energy consumption. How to predict building energy consumption of urban development trends and master the changes of urban building energy consumption are important link of building energy-saving. In order to forecast urban building energy consumption from both the macro and micro aspects, multi-layer feed forward artificial neural network based on BP algorithm is proposed in the paper and used MATLAB implements to design and train of improving BP neural network. A macro urban building energy consumption prediction model and a micro urban building energy consumption prediction model are established in the paper, they are the city electricity demand prediction model and air-conditioning system energy consumption prediction model. The accuracy and feasibility of the prediction models were studied by calculation the difference between practical data and simulation model predictions.