Abstract: ‍ ‍Early warning system uses past experiences and current observations to detect anomalies and sends out warning signals before the danger happens. Traditional early warning systems often rely on experts to analyze data and make decisions. However， it is difficult for the experts to deal with a large amount of information in time， and the decision made is subjective. The effectiveness of early warning is doubtful. To achieve automatic early warning， some researchers try to establish physical model based on causal relationship between warning signs and their dangerous outcomes. Nevertheless， it is still subjective in selecting models， determining indicators and the weights of each indicator， especially in complex scenes. The early warning method relying on physical model is inefficient. Mathematical statistics technology can reduce the subjectivity and inefficiency mentioned above to some extent. Statistical methods need to make assumptions about data characteristics， such as the distribution of data， while it is difficult to make appropriate assumptions in complex scenarios. Unreasonable hypothesis will lead to poor warning performance. With the rapid development of artificial intelligence technology， increasingly artificial intelligence methods have been used in early warning systems. With the help of machine learning technology， the early warning systems can mine and learn particular patterns from a large amount of data， which help the selection of models， indicators， and the weights of indicators； With the help of expert system， the early warning systems can automatically make logical reasoning according to the rules to make decisions professionally and efficiently； With the help of information fusion technology， the early warning system can make use of multi-source observation information efficiently and draw a more comprehensive conclusion. Artificial intelligence technology can overcome the disadvantages of traditional early warning systems in some degree， and improves the accuracy， flexibility， and generality of early warning systems. In this paper， the intelligent methods in early warning systems are reviewed. Firstly， the basic concept of early warning systems， the traditional early warning systems and their limitations are introduced. Then， the early warning system based on machine learning method， expert system and information fusion technology are elaborated. Finally， the conclusions are given， and some issues and future development directions of intelligent methods are discussed.