The accuracy of data-driven flow pattern identification in gas-liquid two-phase flow depends on the quality of training data. If a high similarity of input parameters exists between the training and the application conditions, then, the trained identification model would be applicable. Aiming at pipeline-riser systems in offshore oil and gas fields, a comparative study of different laboratorial two-phase flow loops with different geometric parameters is performed. Statistical parameters commonly used for flow pattern identification are examined. Similarities and differences of parameter distribution in the feature domain are discussed. By analyzing the mean and amplitude scatter distributions, the feature region divisions for identification are determined. First through conditional judgment, and then through LS-SVM recognition, the identification model is constructed. The shortest sample length for effective identification of each flow pattern is analyzed through the relationship between the identification accuracy and the sample length. The typical severe slugging could be easily identified under different pipe geometry conditions even when the sample length is 1 s. The identification accuracies of oscillation, irregular and stable flow patterns vary under different application conditions.