The heterogeneous data collected by the new smart metering devices reveal the interaction between components within the power grid system and other smart city systems. To address the scenario of correlating multiple data sources, a data fusion method for heterogeneous data sources is proposed. This method is based on a spatio-temporal Gaussian process model and utilizes a hierarchical Bayesian approach for simplified inference. By combining power data with external data, the model supplements missing data in the power grid, thereby improving the performance of intelligent tasks in the power grid system. The application potential of this method in practical scenarios is demonstrated using the example of load forecasting in the power grid.