基于指数平滑法和马尔科夫模型的公路客运量预测方法

Prediction method of highway passenger transportation volume based on exponential smoothing method and Markov model

分析了常用的客运量预测方法，提出了一种新的基于指数平滑法和马尔科夫模型的公路客运量预测方法。基于公路客运量的实际值、线性拟合值与二次曲线拟合值，采用二次曲线拟合的方法计算了初始值与平滑系数。以安徽省2000～2009年相关数据为基础，应用指数平滑法预测了2010、2011年的公路客运量。以-11％、-5％、0、5％、119／6为划分阈值，将指数平滑法预测结果的相对误差划分为4个状态区间，应用马尔科夫模型对指数平滑法的预测结果进行修正，并与模糊线性回归模型、指数平滑法的预测结果进行比较。分析结果表明：应用提出的方法，2010、2011年安徽省公路客运量的预测结果分别为14．209、15．712亿人，相对误差分别为1．195％、0．492％；应用指数平滑法，预测结果分别为13．468、14．893亿人，相对误差分别为-3．399％、-4．746％；应用模糊线性回归模型，预测结果分别为13．573、15．325亿人，相对误差分别为-2．647％、-1．983％。提出的方法精度较高，满足实际需求。

The usual prediction methods of passenger transportation volume were analyzed, a new prediction method of highway passenger transportation volume based on exponential smoothing method and Markov model was put out. Based on the actual value, linear fitting value and quadratic curve fitting value of highway passenger transportation volume, the initial value and smoothing coefficient were calculated by using quadratic curve fitting method. According to the related data of Anhui Province in 2000-2009, the highway passenger transportation volumes in 2010, 2011 were predicted by using exponential smoothing method. Taking -11%, -5%, 0, 5%, 11% as division threshold values, the elative errors of prediction results by using exponential smoothing method were divided into four state intervals, the prediction results of exponential smoothing method were modified by using Markov model, and the prediction results among the proposed method, fuzzy linear regession model and exponential smoothing method were compared. Analysis result shows that by using the proposed method, the prediction results of highway passenger transportation volumes in 2010, 2011 are 1. 420 9 × 10^10 , 1. 571 2 × 10^10 persons, relative errors are 1. 195% and 0. 492% respectively. By using exponential smoothing method, prediction results are 1. 346 8 ×10^10, 1.489 3 × 10^10 persons, relative errors are -3. 399% and -4. 746% respectively. By using fuzzy linear regession model, prediction results are 1. 357 3×10^10, 1. 532 5×10^10 persons, relative errors are -2. 647% and -1. 983% respectively. The proposed method has higher precision to meet the actical demands. 7 tabs, 1 fig, 23 refs.