Electrification in buildings is an inevitable trend under carbon neutrality targets and this increases the penetration rate of renewable energy sources in power grids. Understanding the key characteristics of electricity use profiles is the basis for advanced planning and design of integrated building energy systems. Electricity use profiles of buildings are driven by various external factors, and exhibit significant randomness. Current statistical, clustering and generative studies have not fully revealed the key aspects of building electricity use profiles. This study proposes a novel physical-feature-based building electricity use profile model. Five physical feature parameters were extracted from annual daily and diurnal (sub-)hourly time-scales, including the power-temperature curves, weekend ratio, variation factor, weekday schedule and weekend schedule. The proposed model was validated using two engineering-informed metrics: peak demand and load duration curves. In a case study of 4,168 non-residential buildings, the average relative simulation error of 99% peak demand was reduced from 8.53% in the clustering model to 1.44% in the proposed model, the average p-value for the K-S test of the simulated peak hour distributions increased from 0.00059 in the clustering model to 0.28 in the proposed model, and the average CV-MAE of the load duration curve was reduced from 7.24% in the clustering model to 5.36% in the proposed model. The proposed methodology and results present fundamental references of typical building electricity use profiles, and to provide useful support for advanced building energy system planning and design.