In recent years, tourism demand forecasting has attracted more interests not only in tourism area but in data science field. In this study, we follow the previous relevant data science literatures and propose a new neural network enhanced hidden Markovian structural time series model (NehM-STSM). This model takes a multiplicative error structure of a trend and a seasonal element. The trend is modelled by an artificial neural network while the seasonal element is captured by a tailor-made hidden Markovian model with four components: a persistence replicative cycle, a jump component capturing an unexpected event, an amplitude component reflecting the event strength and a random error term. The empirical research is conducted using US incoming tourism data from twelve major source countries across January 1996–September 2017. The proposed NehM-STSM achieves a better performance than the chosen benchmark models for two error measures and most forecasting horizons.

近年来，旅游需求预测不仅在旅游领域而且在数据科学领域都吸引了更多的兴趣。在这项研究中，我们遵循了以前的相关数据科学文献，并提出了一种新的神经网络增强型隐马尔可夫结构时间序列模型（NehM-STSM）。该模型采用趋势和季节要素的乘法误差结构。趋势是通过人工神经网络建模的，而季节性元素是通过量身定制的隐马尔可夫模型捕获的，其中包含四个分量：持久性复制周期，捕获意外事件的跳跃分量，反映事件强度的振幅分量和随机分量错误项。实证研究是使用1996年1月至2017年9月期间来自十二个主要来源国家的美国传入旅游数据进行的。