基于时间序列和神经网络的货运收入预测方法研究与实现

发布时间：2018-06-24 06:57

本文选题：预测 + 时间序列　；参考：《北京工业大学》2015年硕士论文

【摘要】：随着企业信息化和数字化的不断深入,对于数据的分析要求也越来越高。企业对于财务数据已经不再满足于计算准确的要求。随着企业清算周期的不断缩短,关注点现在已经逐渐放在了财务数据的计算及时性上。针对企业的这一需求,对企业的货运收入数据预测方法进行了分析及研究。由于实际生产过程中出现的数据分布规律并不统一,因此采用不同的方法分别进行分析处理。此课题主要做了三个工作。第一,针对可以呈线性分布的货运收入数据,使用基于时间序列的方法进行预测。在这个工作中,除了论述当前货运收入预测中经常使用的移动平均法之外,还详细论述了使用指数平滑法和一次线性回归法的预测过程,并且详细论述了这些方法中使用的重要参数的取值原因。第二,对于呈曲线分布的货运收入数据,使用多元函数回归法、BP神经网络法进行预测。本文在这个工作的论述中,除了给出通过计算得到的预测值,还对预测值的误差进行了分析,解释了文中BP神经网络各层神经元个数的取值原因。第三,结合前面两个部分的内容,使用BP神经网络与时间序列结合的预测方法,对货运收入数据进行预测。在这个工作中,除了论述了这种预测方法之外,还指出了这种预测方法同前面预测方法的不同之处,分析了这种方法相比前面两种方法的优势。通过这篇论文的论述,实现了对于大多数情况下公司货运收入数据的预测方法。对于自主开发的软件系统来说,使用商业软件工具进行的预测,无法直接移植到系统中。这篇论文针对不同的货运收入数据的实际案例,详细分析了预测的过程。对于每种预测技术,详细论述了具体的算法。通过这篇论文的论述,可以将这些技术在系统中得到实现。对于公司来说,可以通过收入预测对运营情况做出调整,对于公司的管理有着十分重要的意义。
[Abstract]:With the deepening of enterprise informatization and digitization, the requirement of data analysis is becoming higher and higher. Enterprises are no longer satisfied with the requirements of accurate calculation for financial data. As the liquidation cycle continues to shorten, the focus is now on the timeliness of financial data calculation. According to the demand of enterprises, this paper analyzes and studies the forecasting method of freight revenue data. Because the data distribution law in the actual production process is not uniform, different methods are used to analyze and process the data. This subject has done three main work. Firstly, the time series method is used to predict the freight revenue data which can be linearly distributed. In this work, in addition to discussing the moving average method, which is often used in the current freight revenue forecasting, the prediction process using exponential smoothing method and primary linear regression method is also discussed in detail. The reasons for the important parameters used in these methods are discussed in detail. Secondly, for the freight revenue data with curve distribution, the multivariate function regression method and BP neural network method are used to forecast the data. In this paper, in addition to the calculated prediction value, the error of the prediction value is analyzed, and the reason of the number of neurons in each layer of BP neural network is explained. Thirdly, combining the content of the former two parts, using BP neural network and time series forecasting method, the freight revenue data are forecasted. In this work, in addition to the discussion of this forecasting method, it also points out the difference between this prediction method and the previous prediction method, and analyzes the advantages of this method compared with the former two methods. Through the discussion of this paper, the forecasting method of the company freight revenue data is realized in most cases. For the self-developed software system, the prediction using commercial software tools can not be directly transplanted to the system. This paper analyzes the forecasting process in detail according to the actual cases of different freight revenue data. For each prediction technique, the specific algorithm is discussed in detail. These technologies can be realized in the system through the discussion of this paper. For the company, it is very important for the management of the company to adjust the operating situation through the revenue forecast.
【学位授予单位】：北京工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：F259.2;TP183

【参考文献】