中文摘要：

由于旅游突发事件的突然爆发性、危害性及信息不对称性，导致旅游客流量在短时间内发生急剧变化，原有模式被打破，非线性趋势和线性特征交织的随机性趋势明显，为旅游客流量正常预测带来极大的难度。本文提出一种面向旅游突发事件客流量混合预测方法，即支持向量回归（Support Vector Regression，SVR）和自回归求和移动平均模型（Autoregressive Integrated Moving Average，ARIMA）结合的混合预测方法：首先通过SVR预测旅游突发事件时期客流量，然后再用ARIMA预测SVR预测值的残差部分，最后将两者预测结果相加；同时针对客流量复杂特征，采用一种混沌粒子群算法（Chaos Particle swarm optimization，CPSO）实现对SVR参数选择。来自黄山风景区汶川地震时期客流量相关数据验证表明，混合预测模型优于单一预测方法，为旅游突发事件时期客流量预测提供了一种有效选择。

英文摘要：

Because of sudden explosiveness and destructiveness as well as information asymmetry caused by tourism emergency events, the tourist flow deviates from original patterns and presents nonlinear and linear features, which causes a great difficulty to tourist flow foreeasting. Traditional forecasting methods cannot solve this complicated problem. The article proposes a kind of tourist flow hybrid forecasting model for tourism emergency events which include two methods. One method is Support Vector Regression （SVR）. It has good ability to deal with nonlinear and small sample problems and has been successfully used in many forecasting fields by researchers. The other method is Autoregressive Integrated Moving Average （ARIMA） which can deal with linear problem easily. At same time, the three parameters C,ε,σ of SVR affect the accuracy of forecast. A kind of Chaos Particle Swarm Optimization （CPSO） is used in the article. By the local search ability of Chaotic Local Search（CLS） as well as global search ability of Adaptive Inertia Weight Factor （AIWF） in CPSO, the optimal parameters C,ε,σ of SVR can be found effectively. The detail process of tourist flow hybrid forecasting model is as follow. Firstly SVR is used to forecast tourist flow during emergencies. Meanwhile, CPSO is implemented to select the SVR parameters; Secondly ARIMA model is provided to forecast residual sequence of forecasting values. Finally two predic ted values will be added, which leads to the final predicted values. Data set from Mount Huangshan during Wenchuan Earthquakes period are used to validate the effectiveness of the hybrid models. The number of the data is from February 12, 2008 to June 12, 2008, including the daily tourist flow and daily tourist flow before eight o ＇clock. The results show that the hybrid ap proaehes are signifieantly higher in accuracy than CPSO-SVR and PSO-SVR. , which provide an effective choice to tourism emergency events flow forecasting as well as similar industries facing the same situati