多源数据跨国人口迁移预测

针对跨国人口迁移预测所面临的数据代表性问题，利用多源数据分别构建3个预测模型：线性拟合模型、乘法分量模型和带有时间序列预测（WTSP）的线性拟合模型. 线性拟合模型用于刻画1年内的移民规律；乘法分量模型利用时间序列预测算法对未来迁移模式进行预测；WTSP线性拟合模型利用迁移模式的变化预测跨国人口迁移数量的未来趋势. 对比3个模型的预测结果可知，WTSP线性拟合模型可以有效预测未来的移民规律，相比经典线性拟合模型，WTSP线性拟合模型能体现迁移模式随时间变化的规律，预测准确率可至少提升3%；相比乘法分量模型，WTSP线性拟合模型能呈现更完整的迁移模式，有更强的可解释性.

Transnational population migration forecast with multi-source data

Three prediction models were constructed by using multi-source data, including linear fitting model, multiplicative component model and WTSP (with time series prediction) linear fitting model, aiming at the problem of data representativeness in the prediction of transnational migration. A linear fitting model was constructed to describe the migration rule within one year; a multiplier component model was introduced to predict the future migration pattern by using time series prediction algorithm; and a WTSP linear fitting model was proposed to predict the future trend of transnational migration by using the change of the migration pattern. Compared with the results of the three models, WTSP linear fitting model can effectively predict future migration patterns. Compared with the classical linear fitting model, the WTSP linear fitting model can reflect the law of migration pattern changing with time, and the prediction accuracy can be improved by at least 3%. Compared with the multiplier component model, the WTSP linear fitting model can present a more complete migration model and has stronger interpretability.