A non-intrusive multi-nodal load forecasting method for areas with big data collected by smart meters is proposed. The aggregation effects on modelling and forecasting of load are studied, indicating the supremacy and necessity of multi-nodal forecasting. Feature selection of dataset is based on the analysis of factors of electricity demand. Top-down and bottom-up forecasting approach are compared through the multiple linear regression model. Experimental results on real-world dataset shows the advantage of the bottom-up approach supported by smart meter big measurement data in short-term multi-nodal load forecasting task.