The Metro system is indispensable in modern city transportation. Peak times at stations vary depending on the location of the station, as traffic is affected by close stations with time characteristics. Larger crowds during peak hours can cause accidents. Therefore, an accurate prediction of passenger flow is crucial for traffic management. Gramian Angular Field (GAF) uses the angles and differences between polar coordinates to process the autocorrelation time series data. We used GAF and converted time series data with the original data to construct a Convolutional Neural Network (CNN) model. The station traffic data of the Taipei MRT system from 2017 to 2022 was used. The CNN model was used to build a passenger flow level model in Metro stations. By using the traffic data, passenger flow over the next hours was predicted. When compared with the untransformed data, GAF transformation improved the model learning accuracy of a single station by 10%. For the validation data set, the model accuracy increased by 5%.