The invention discloses a channel knowledge map construction method based on an expectation maximization algorithm. The method comprises the following steps: obtaining channel knowledge data of a local communication environment through any mode of offline ray tracing simulation, offline field measurement or online real-time measurement; carrying out statistical modeling on related channel knowledge according to expert knowledge, estimating K-class parameters of a mixed statistical model by using an EM algorithm, and constructing a channel knowledge map reflecting a local signal propagation environment; and when a user needs to communicate, obtaining real-time position information through GPS, Beidou, cellular positioning, laser radar and self-sensor positioning modes, and obtaining channel knowledge of a target position by using the previously constructed channel knowledge map and a mode based on inverse distance weighting. Therefore, the method is used for environment perception adaptive communication. According to the invention, the problems of low accuracy, large data storage capacity demand and high training complexity of a channel prediction method based on a pure model or pure data are solved, so that the cost for acquiring real-time channel state information is reduced.