基于支持向量回归的旅游短期客流量预测模型研究

发布时间：2018-06-08 04:07

  本文选题：旅游短期客流量 + 预测模型　； 参考：《合肥工业大学》2014年博士论文

【摘要】：随着旅游业的快速发展,居民外出旅游人数急剧增长,给旅游景区造成极大冲击。近年来由于游客拥挤、超载等问题造成的安全事故频发,给旅游景区造成了巨大的负面影响。准确的旅游短期客流量预测能够为旅游管理者提前决策提供直接信息,最大限度的避免这种情况的发生。然而在我国,由于受到自然气候、特有的休假制度、旅游突发事件等诸多外部因素影响,旅游短期客流量表现出非线性、季节性、随机性等复杂特点,传统的预测方法往往难以实现准确预测,因此建立科学合理的短期客流量预测模型,实现对旅游景区不同时期的短期客流量预测,对旅游景区尤其是热门景区乃至整个旅游行业意义重大。 支持向量回归(Support Vector Regression, SVR)作为一种基于统计学习理论的新的机器回归分析方法,由于具有处理非线性、小样本等问题能力,能较好地解决旅游短期客流量的非线性、季节性和随机性等问题,为复杂的短期客流量预测提供了一种新的选择。 本文以旅游景区为研究对象,以科学准确预测旅游景区短期客流量为目标,根据旅游短期客流量在不同时期表现出的特点,将其分成平常日客流量、节假日客流量、旅游突发事件时期客流量三种不同类型,分别研究这三种不同类型的短期客流量预测问题。 本文的主要研究内容如下： 1)对国内外旅游客流量预测方法进行了系统综述,指出目前国内外在旅游客流量预测研究上取得成果及在方法、尺度等方面存在的一些局限性,以此为基础,提出本文研究的研究内容。 2)对短期客流量的主要影响因素进行系统分析,进一步分析旅游短期客流量在不同时期的客流量特点,通过对旅游短期客流量不同时期客流量特点的分析,将短期客流量的研究分成平常日客流量、节假日客流量以及突发事件时期客流量。 3)针对平常日客流量非线性突出的特点,提出基于遗传算法(Genetic Algorithm,GA)的支持向量回归模型GA-SVR模型,利用GA对SVR自由参数进行选择,并将该方法与BPNN模型进行对比。基于黄山风景区的有代表性的平常日短期客流量等相关数据验证表明：(GA-SVR模型较BPNN模型预测误差更小,准确性更高。 4)针对每年节假日客流量呈现的明显季节性特点,提出基于季节调整的自适应遗传算法(Adaptive Genetic Algorithm, AGA)支持向量回归模型,即季节指数调整(Seasonal Exponential Adjustment, SEA)的AGA-SVR预测模型(SEA-AGA-SVR)和季节因子调整(Seasonal index Adjustment,SI)的AGA-SVR预测模型(AGA-SSVR)。其中SEA-AGA-SVR主要对短期客流量的季节性进行事前调整后再进行预测；而AGA-SSVR重在事后对预测值进行季节因子调整。来自黄山风景区2008-2012年节假日的客流量数据的实验结果表明,两种季节调整方法均能有效的去除季节性成分,预测效果均优于AGA-SVR方法,但是由于SEA-AGA.SR直接对原始时间序列数据进行季节性处理,预测效果优于事后调整的AGA-SSVR模型,同时预测时间也大大缩短. 5)针对旅游突发事件的突发性、无法预见性而导致的客流量高度不确定性、随机性特点,提出基于混沌粒子群(Chaos Partic]e swarm optimization,CPSO)的SVR和自回归移动求和平均(Autoregressive Integrated Moving Average,ARIMA)相结合的混合模型即CPSO.SVR.A RIMA模型。先通过CPSO对SVR模型进行寻优,再用SVR对突发事件时期客流量进行预测,然后使用ARIMA模型对SVR残差序列预测,最后将两者预测值相加,即为所求预测值。来自黄山风景区汶川地震时期客流量数据的实验结果表明CPSO-SVR-ARIMA混合模型能够很好的抓住突发事件客流量的波动及变化轨迹,预测精度明显高于单一的预测方法CPSO-SVR和PSO-SVR。
[Abstract]:With the rapid development of tourism , the number of people out of travel has increased sharply , causing great impact on tourist attractions . In recent years , due to many external factors such as crowded and overloaded tourists , the forecast of short - term passenger flow of tourism can provide direct information for the advance decision - making of tourism managers . Therefore , it is often difficult to predict the short - term passenger flow due to natural climate , special leave system , tourism emergency and so on .

As a new method of machine regression analysis based on statistical learning theory , support vector regression ( Support Vector Regression ) can be used to solve the problems of non - linearity , seasonal and randomness of short - term passenger flow and provide a new choice for complex short - term passenger flow prediction .

Based on the characteristics of short - term passenger flow in different periods , three different types of short - term passenger flow forecasting are studied in this paper based on the characteristics of short - term passenger flow in different periods according to the characteristics of short - term passenger flow in tourist areas .

The main contents of this paper are as follows :

1 ) A systematic review is made on the domestic and foreign tourist flow forecasting methods , and points out some limitations in the research of tourism passenger flow forecast at home and abroad , and some limitations exist in the methods , scales and so on , and the research contents of this paper are put forward .

2 ) analyzing the main influencing factors of short - term passenger flow , further analyzing the passenger flow characteristics of short - term passenger flow in different periods , analyzing the characteristics of passenger flow in different periods of short - term passenger flow , dividing the study of short - term passenger flow into normal daily passenger flow , holiday passenger flow and passenger flow during emergency period .

3 ) Based on genetic algorithm ( GA ) , a support vector regression model GA - SVM model is proposed , which is based on genetic algorithm ( GA ) . The method is compared with the BPNN model .

4)閽堝�规瘡骞磋妭鍋囨棩瀹㈡祦閲忓憟鐜扮殑鏄庢樉瀛ｈ妭妫�

本文编号：1994380