The invention provides a method for cleaning veihcle GPS data based on road network data, and relates to the field of traffic big data processing. According to the method, HDFS is adopted to store data, and a MapReduce calculation framework is adopted to process the data, so that abnormal records in the GPS data are eliminated. The abnormal records comprise drift points with overlarge distance deviation under continuous timestamps, track points of multiple records under the same timestamp, namely repeated data, and jump points with overlarge time difference and distance between two adjacent records. For the duplicate data, the duplicate data is removed through a time stamp-based duplicate removal algorithm. The speed of two adjacent points in the GPS track is calculated according to the time and the distance, and if the speed is greater than a set threshold value, the speed is regarded as a drift point, and the data is rejected. For the jump points, a shortest weighted path is calculated between the jump points by using a Dijkstra algorithm in combination with road network data, the shortest weighted path is taken as a supplementary trajectory between the jump points, and road nodes passing through the path are sequentially inserted between the jump points.