We propose a spatial-temporal inference model, which uses a large amount of spatial and temporal data in the city to help governments or enterprises predict future long- and short-term important urban indicator values, such as traffic flow, human mobility, pollution level, number of criminal cases or even commercial profitability. The SIM model exploits IoT to integrate multiple real-time geospatial big data, including population, the flow of people, geographical data, Traffic, and real-time sensor values. SIM can make effective predictions and provide explainability for making decisions.
The past works mainly focused on the statistical analysis of various geospatial data. However, it is hard to effectively model complex relationships between features from massive data, resulting in high time cost and low accuracy. Recently, some studies switched to use deep learning models. However, pure deep learning has low accuracy and cannot provide interpretability in a large number of heterogeneous numerical data. Thus, existing solutions cannot effectively combine dynamics and statics. Experiments confirm that our prediction effectiveness is beyond all current deep learning approaches.
As long as part of the urban historical data of the forecast target can be obtained, together with crawling other time-space dynamic and static information in real-time, our SIM technology can be applied to forecast future indicators of industry or governments. Enterprises can earn profit from SIM＇s inference, and the government can also use resources properly and make good decisions for the future. The work has been successfully applied to traffic flow, human mobility, disease risk, environmental pollution, mass transportation demand, bank branch indicators, illegal parking, crime rate, etc.