The high spatial and temporal resolution of geostationary satellite has significant advantages in monitoring and forecasting disastrous weather with high impact. A typhoon precipitation retrieval study based on infrared spectrum bright temperatures of the advanced geosynchronous radiation imager (AGRI) in FY-4A satellite is conducted. The influences on precipitation retrieval accuracy of different distance measurements of K-nearest neighbor (KNN) in the inverse problem of regularization method are discussed. The precipitation retrieval consists of two steps: （1） the identification of precipitation field of view, which uses KNN to identify the “precipitation” and “non-precipitation” signals of brightness temperatures to be retrieved mainly based on the training dictionary samples；（2） the retrieval of the precipitation field of view, which uses inverse problem of regularization method to retrieve the infrared brightness temperatures precipitation on the basis of identifying the precipitation field of view. KNN distance measurement adopts Euclidean, normalized Euclidean, Mahalanobis and Cityblock respectively. Taking typhoon “Ampil (2018)”as an example, the precipitation retrieval experiment is carried out. The experiment shows that the retrieval result has a high similarity with GPM, and different distance measurements have their own advantages in the retrieval of "extreme precipitation".