In order to easily and effectively detect the common problem of drainage pipeline blockage in cities, a device of drainage pipeline was designed and a test of blockage was completed, water level data was monitored online, after optimizing machine learning algorithms, a diagnostic program for pipeline blockage was developed based on the best algorithm in Python code. The results show that when the drainage pipeline is blocked meanwhile the flow rate is the same, the more severe the blockage is, the higher the upstream water level line rises. The closer the blockage is to the position of upstream monitoring point, the higher the upstream water level rises. The downstream water level is almost only related to the flow rate. Through comparing machine learning algorithms, it is found that the K-nearest neighbor algorithm exhibites high accuracy in classifying data set. After optimized by the grid search crossing validation method, the model accuracy is around 98%. It is concluded that this diagnostic method can realize rapid diagnosis of pipeline blockage, optimize the process of dispatch scheduling, and reduce the costs of pipeline operation and maintenance.