The Research On Short-term Tourist Flow Forecasting Based On Support Vector Regression

With the rapid development of tourism industry, the number of tourists grows sharply and the short-term tourist flow has made great impact on tourism scenic spots. In recent years, more and more safety accidents have occurred because of the crowded tourists, overload and other problems which caused negative influences on scenic spots. Accurate short-term tourist flow forecasting can provide tourism managers some direct information in advance, avoiding the occurrence of such events. However, in our country, due to the influences of natural climate factors, holiday system, tourism emergencies etc., tourist flow reveals characteristic of nonlinearity, seasonality, and randomness, the traditional forecasting methods can not achieve accurate prediction results successfully. Therefore, it is of great importance to establish scientific and reasonable short-term tourist flow forecasting models, and realize the short-term tourist flow forecasting in different periods, which have great significance on scenic spots especially popular scenic spots, and even the whole tourism industry.As a kind of new machine regression analysis method based on statistical learning theory, Support Vector Regression (SVR) has good ability to deal with nonlinear and small sample problems, and it can also solve the characteristic of nonlinearity, seasonality and randomness in short-term tourist flow. SVR has provided a new choice for complex short-term tourist flow forecasting.Within the context of the tourism scenic spots, this dissertation divideds tourist flow into three different types:ordinary day, holiday and emergencies period respectively.The dissertation will analyse different types of short-term tourist flow forecasting problems.The main contents of this dissertation are as follows:1) On the basis of analyzing reviews of tourist flow forecasting methods both domestic and abroad, the dissertation points out the limitations of methods and measure, then gives the research contens.2) By analyzing of short-term tourist flow main influence factors, the dissertation studies characteristics of short-term tourist flow from different periods. According to different characteristics, the dissertation divides short-term tourist flow into ordinary day, holiday and emergencies period three different types respectively.3) In view of prominent nonlinear characteristics in ordinary day tourist flow, the dissertation puts forward a forecasting model based on Genetic Algorithm and Support Vector Regression, namely GA-SVR model. In this model, GA is used to optimize SVR freedom parameters choice, and then GA-SVR model is compared with Back Propagation Neutral Network (BPNN) model. The representative ordinary day data set from Mount Huangshan shows that GA-SVR model has smaller prediction error and higher accuracy than that of BPNN model.4) For obvious seasonal characteristic in holiday tourist flow, the dissertation combines Adaptive Genetic Algorithm (AGA), Support Vector Regression with two different seasonal adjustment approaches, one is Seasonal Exponential Adjustment (SEA) AGA-SVR model, namely SEA-AGA-SVR, and the other is Seasonal Index Adjustment AGA-SVR model, namely AGA-SSVR. The SEA-AGA-SVR mainly aims at seasonal adjustment of original data before forecasting, while the AGA-SSVR adjusts seasonality of prediction values after forecasting. Holiday daily tourist flow from Mount Huangshan scenic spot in years2008-2012is taken as an example. The experiment results show that both kinds of seasonal adjustment methods can effectively remove seasonal ingredients, and prediction effects are superior to AGA-SVR method. However, the SEA-AGA-SVR model is superior to AGA-SSVR model with higher accuracy and less prediction time.5) Due to uncertainty and randomness characteristic caused by sudden and unpredictability tourism emergencies, a hybrid model is proposed, which hybrids Chaos Particle swarm Optimization (CPSO), support vector regression and Autoregressive Integrated Moving Average (ARIMA) approaches, namely CPSO-SVR-ARIMA model. Firstly CPSO-SVR model is used to forecast tourist flow during emergencies, secondly ARIMA model is provided to predict residual sequence of forecasting values, finally two predicted values will be added, which is predicted values. Data set from Mount Huangshan scenic spots during Wenchuan Earthquake period shows that the hybrid model is significantly higher in accuracy than single methods CPSO-SVR and PSO-SVR.