Determining the pigging cycle by field manual experience has some blindness.To avoid excessive pigging, based on the slow timing of the wax formation, this paper studies the relationship between production data parameters and the degree of wax formation. Then, the paper applies the improved K-means algorithm to determine the degree of wax formation in different time series and introduces the data into the convolutional neural network (CNN). With the feature extraction and adaptive learning ability of the CNN model, it updates online different durations of wax formation degrees.The pigging effect evaluation index model is built according to the change in wax formation degree and pigging cycle to realize the early warning of pigging operation. The results show that different pipelines have various grade time and cycle lengths in the complete wax formation cycle, which reflects the difference in wax formation degree caused by the difference in input parameters. The improved K-means algorithm divides the wax formation degree into 4 grades. The overall average error of the proposed model is 0.781 d, which is smaller than 2.025 d and 1.225 d of the RNN model and LSTM model. The pigging effect evaluation index of the pipeline to be evaluated is 0.828, indicating a sound pigging effect. The research results can provide a practical reference for improving the management level of pipeline integrity.