基于大数据的酒店微观市场的预测与分析

发布时间：2018-03-18 01:01

本文选题：收益管理　切入点：酒店　出处：《太原理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着我国经济实力的增强,中国经济正在由工业主导向服务业主导加快转变。酒店住宿业作为服务业的典型代表,其客容能力过剩和投资回收的压力与日俱增,有效的管理对酒店业变的越来越重要。为了高效的管理酒店,提高酒店的收益,这里引入了收益管理的概念。它主要通过建立实时预测模型和对以市场细分为基础的需求行为分析,确定最佳的销售和服务价格。酒店的收益管理主要包括以下四个部分:需求预测、超量预订、客房分配和定价系统,其中,客户需求预测是收益管理的基础与核心。传统的预测模型一般为历史同期和时间序列,时间序列又分为移动平均(Moving Average),指数平滑(Exponential Smoothing),卡尔曼滤波(Kalman Filters),自适应滤波(Adaptive Filters)和自回归积分滑动平均模型(Autoregressive Integrated Moving Average Model),但酒店收益管理的预测和传统的预测不同,一般有两个时间变量,分别为预定时间和消费时间。在大数据的背景下,如果能够很好的针对酒店行业的数据特点,对酒店需求预测进行研究分析,就可以帮助酒店管理者更好的作出决策,从而提升酒店收益。本文提出了一套基于大数据的针对酒店微观市场的预测方法,从数据预处理、数据筛选到最后的训练与预测。在数据的预处理阶段,本文通过对数据的分析,将原始数据进行了转换,转换后的数据经过KS检验符合高斯分布模型,因此采用了基于统计的异常点检测方法,找出了间夜量特殊的日期。本文还设计了一种纠偏函数,在保持了异常数据的相对关系同时,可以令间夜量特殊的日期得到很好的处理。在数据筛选阶段,本文对比了几种经典的分类方法,设计了一套对于每类单独训练再将整体结果相结合的训练思路,这种思路在可以进一步将模型精度提高的同时,还有利于分析某类预测精度不佳的原因,具有很好的可解释性。最后本文结合前人的研究成果,对经典的方法进行了改进,将经典的基于时间序列的预测算法转化成了更普适的机器学习方法,在与经典预测算法的比较中,效果良好。随着机器学习算法的不断改进,机器学习理论的不断完善,结合本文提供的思路,未来可以使用更多的机器学习模型对该问题进行分析、训练和预测。
[Abstract]:With the strengthening of China's economic strength, China's economy is undergoing a rapid transformation from industry-led to service-oriented. As a typical representative of the service industry, hotel accommodation industry, as a typical representative of the service industry, is under increasing pressure of overcapacity and investment return. Effective management is becoming more and more important to the hotel industry. The concept of revenue management is introduced here. It is mainly through the establishment of real-time forecasting model and the analysis of demand behavior based on market segmentation. Determine the best price for sales and services. The revenue management of the hotel mainly consists of the following four parts: demand forecasting, overbooking, room allocation and pricing system, among which, Customer demand forecasting is the basis and core of revenue management. The time series are divided into moving average, exponential smoothing, Kalman filters, adaptive filters and autoregressive Integrated Moving Average models. However, the prediction of hotel revenue management is different from the traditional one. Generally, there are two time variables, one is the reservation time and the other is the consumption time. Under the background of big data, if we can do research and analysis on the hotel demand forecast according to the characteristics of the hotel industry data, This paper puts forward a set of forecasting methods based on big data for the hotel micro market, which can help the hotel managers to make better decisions and improve the hotel income. From the perspective of data preprocessing, this paper puts forward a set of forecasting methods for the hotel micro market. Data filter to the final training and prediction. In the data preprocessing stage, through the analysis of the data, the original data are converted, the converted data after KS test accord with Gao Si distribution model. Therefore, the outlier detection method based on statistics is used to find out the special date of the night volume. A correction function is designed, which keeps the relative relation of the abnormal data at the same time. In the stage of data screening, this paper compares several classical classification methods, and designs a set of training ideas for each kind of individual training and combining the overall results. This method can further improve the accuracy of the model, at the same time, it is helpful to analyze the causes of the poor prediction accuracy, and it has good interpretability. Finally, this paper improves the classical method combined with the previous research results. The classical prediction algorithm based on time series is transformed into a more general machine learning method. Compared with the classical prediction algorithm, the effect is good. With the continuous improvement of the machine learning algorithm, the machine learning theory is constantly improved. Combined with the ideas provided in this paper, more machine learning models can be used to analyze, train and predict the problem in the future.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F719.2;TP181

【参考文献】