Flooding disaster prediction is an important part of emergency management; How to get disaster data in real time and accurately and carry out correlation analysis is a difficult point in disaster investigation. This paper is committed to the research on the perception and fusion of multi-source data of flood and geological disasters. Through the fusion of multi-source data, data analysis engine and algorithm model, data assimilation technology, core professional model and artificial intelligence analysis and other key technologies, a lightweight big data platform is built. The platform includes a flood prediction model and a small visual disaster information system, which can flexibly present the monitoring and analysis results. The data analysis system and algorithm model of this platform form virtual monitoring data through numerical calculation. Entropy method, hot spot technology and machine learning method are used to build up statistical models for water level, flow rate and other data. And the multi-source data could be the parallel verification method, which is helpful to greatly reduce the monitoring points and operation and maintenance costs. This platform can also connect the disaster data association analysis platform with the manual command and decision-making system, and provide the basis for prediction, decision-making and command.