旅游学刊
旅遊學刊
여유학간
Tourism Tribune
2013年
4期
75~82
,共null页
余向洋 胡善风 朱国兴 李德明
餘嚮洋 鬍善風 硃國興 李德明
여향양 호선풍 주국흥 리덕명
景区客流 最小二乘支持向量机 中期预测 黄山 风景区
景區客流 最小二乘支持嚮量機 中期預測 黃山 風景區
경구객류 최소이승지지향량궤 중기예측 황산 풍경구
tourist arrivals; least squares support vector machines; medium-term prediction; Huangshan scenic
受到季节性、外部冲击和经济周期等因素作用的景区客流波动幅度大,其预测一直是旅游学研究中的难题,尤其是中长期预测。文章采用当前使用极为广泛的最小二乘支持向量机方法(1eastsquaressupportvectormachines,LS.SVM)对黄山风景区客流月度数据(1987年1月~2010年12月)进行了2年时间尺度的预测,结果表明:采用LS—SVM方法进行景区客流中期预测,其预测的各项性能指标均明显优于BP神经网络(BackPropagationNeuralNetwork)、X-12-ARIMA(AutoregressiveIntegratedMovingAverageModel)、经验模态分解(empiricalmodedecomposition,EMD)与LS—SVM组合预测方法,即EMD—LSSVM方法,预测效果理想,并且具有训练时间短、精度高等优点。其较为准确的预报能力能够为景区规划管理和战略决策提供科学依据,具有较高的实用价值。
受到季節性、外部遲擊和經濟週期等因素作用的景區客流波動幅度大,其預測一直是旅遊學研究中的難題,尤其是中長期預測。文章採用噹前使用極為廣汎的最小二乘支持嚮量機方法(1eastsquaressupportvectormachines,LS.SVM)對黃山風景區客流月度數據(1987年1月~2010年12月)進行瞭2年時間呎度的預測,結果錶明:採用LS—SVM方法進行景區客流中期預測,其預測的各項性能指標均明顯優于BP神經網絡(BackPropagationNeuralNetwork)、X-12-ARIMA(AutoregressiveIntegratedMovingAverageModel)、經驗模態分解(empiricalmodedecomposition,EMD)與LS—SVM組閤預測方法,即EMD—LSSVM方法,預測效果理想,併且具有訓練時間短、精度高等優點。其較為準確的預報能力能夠為景區規劃管理和戰略決策提供科學依據,具有較高的實用價值。
수도계절성、외부충격화경제주기등인소작용적경구객류파동폭도대,기예측일직시여유학연구중적난제,우기시중장기예측。문장채용당전사용겁위엄범적최소이승지지향량궤방법(1eastsquaressupportvectormachines,LS.SVM)대황산풍경구객류월도수거(1987년1월~2010년12월)진행료2년시간척도적예측,결과표명:채용LS—SVM방법진행경구객류중기예측,기예측적각항성능지표균명현우우BP신경망락(BackPropagationNeuralNetwork)、X-12-ARIMA(AutoregressiveIntegratedMovingAverageModel)、경험모태분해(empiricalmodedecomposition,EMD)여LS—SVM조합예측방법,즉EMD—LSSVM방법,예측효과이상,병차구유훈련시간단、정도고등우점。기교위준학적예보능력능구위경구규화관리화전략결책제공과학의거,구유교고적실용개치。
Forecasting tourism demand plays an important role in formulating national tourism development policy and strategic planning. It is also important in optimizing the allocation of tourism market resources and drawing up strategic plans and decision making for tourism businesses. As a result, forecasting tourist arrivals has been a main focus in tourism research. However, making such forecasts is a difficult task for tourism researchers. This is particularly the case with widely fluctuating numbers of tourist arrivals in scenic areas, which are impacted by seasonality, external shocks and economic cycles. Making such medium- and long-term forecasts is one of the most challenging areas in tourism research. In this paper, the widely used method of least squares support vector machines was applied in forecasting two- year tourist arrivals in Huangshan scenic areas using monthly data (from January 1987 to December 2010). The analysis results showed that forecast performance indexes using least squares support vector machine algorithms are better than with the BP neural network, ARIMA forecasting model, and combined forecasting using empirical mode decomposition and least squares support vector machines. Moreover, least squares support vector machinealgorithms are superior in both prediction accuracy and computing speed. The ability of least squares support vector machines to forecast tourist arrivals has high practical value and can provide a scientific basis for planning and strategy management of scenic areas. In addition, nine forecasting models were assessed: the time-series regression prediction model, the HoltWinters addition and multiplication model, the TRAMO/SEATS model, a combination of the wavelet and ARIMA forecasting models, the wavelet neural network prediction method, and the four methods cited in the previous paragraph. We investigated the forecasting ability of these nine models in predicting tourist arrivals over one- and two-year time scales in Huangshan scenic areas. Owing to restrictions of space, however, we are unable to detail all the analysis procedures in this paper. The results showed that least squares support vector machines was superior in performance. It was better than the other single forecasting methods and also exceeded the combination of the wavelet and ARIMA forecasting models, the wavelet neural network, and the combined empirical mode decomposition and least squares support vector machines over the two-year time scale; however, these advantages were not evident over the one-year time scale. The effectiveness of least squares support vector machines in forecasting other tourist arrival data requires further examination. The main difficulties in forecasting monthly tourist arrival data lie in the large fluctuations in the numbers of tourists and the lack of data relating to tourist arrivals. These data include such economic figures as GDP, quarterly unit statistics, tourist arrival data other than annual statistics of inbound tourists, the vulnerability to external shocks, and diversity in the data from different regions. In all, the diversity, complexity, and fluctuation in different types of tourist arrivals make it difficult to find a satisfactory tourist arrival forecasting method that is universally accurate. Thus, greater efforts need to be made to develop better methods for accurately forecasting tourist arrival data for different scenic areas.