预测方法在粮食行业的应用
摘要
粮食行业作为一个特殊的经济领域，其生产经营行为符合经济运行规律，如何利用这些因素，估计未来某些经济指标的运行趋势，为粮食主管部门作出正确的决策提供辅助决策成为粮食企业的迫切需要。
已有的预测工具由于价格昂贵，专业性很强，不适于一般用户使用。而一些专门的预测软件只针对某一特定领域或某一指定项目，而且大多集成在MIS或ERP系统中，具有很强的运行环境依赖性。因此开发一个专门针对粮食行业，供粮食主管部门的普通决策者使用的预测系统作为需求被提出。
本文对预测模型方法进行了研究后，将经济预测技术应用于粮食智能决策支持系统，将数据仓库的优势和特点与预测技术相结合，实现了基于数据仓库的粮食预测系统。
本文研究和比较了几种常用的预测模型：指数平滑模型，季节调整模型，线性随机模型，线性回归模型，神经网络模型。其中指数平滑模型，季节调整模型，线性随机模型用于时间序列分析，线性回归模型用于因果分析，神经网络模型既可用于因果分析又可用于时间序列分析。
将上述几种预测模型应用于粮食行业，以易理解，实用性为目标，以粮食智能决策支持系统为背景实现了一个面向一般计划管理人员的粮食预测系统。
以粮食数据仓库为基础，建立了数据库、模型库和方法库三库合一的预测数据仓库。数据仓库实现对预测主题和信息的存储与综合。模型库和方法库为决策问题提供定量分析(模型计算)和辅助决策信息。数据库的构成包括各种基础数据，时序数据和相关的数据字典；模型库的内容有预测实例字典，模型参数接口表，模型存储表，预测结果存储表；方法库由预测方法字典和预测算法类库构成。
设计实现了一个预测集成环境，直接支持用户自定义预测实例，对粮食产量和价格进行预测。用户可以选择多种预测方法，得到易于理解的预测结果，对模型的均方误差和曲线拟合情况进行判断比较，综合考虑模型的取舍。实现包括预测实例定义，预测方法选择，模型建立，模型检验，模型存储，多方案预测结果分析在内的完整预测过程。提供数据管理，预测应用，数据查询，操作引导和容错处理等全方面的功能，用友好的图形用户界面引导用户进行操作，并以易于理
解的方式将结果反馈给用户，用户不需要具备很多预测的专业知识，就可以轻松地控制和调整预测系统的运行，使预测这一门复杂的综合科学能够为一般用户服务，具有很强的实用性。
数据管理模块由数据生成、数据操作和数据维护三个子功能组成，用于数据字典和时序数据的生成、录入、审核和维护，是预测建模和预测执行的基础。
数据查询为用户提供数据字典查询、数据信息查询、预测模型查询功能，使用户对各种数据和模型有详尽的了解，并能对数据进行分析，作出粗略的判断和估计。
容错处理及操作引导捕捉操作过程中出现的错误和异常，以友好的对话框和文字方式给用户以提示，并给出处理建议。
预测应用模块运用数据库、模型库和方法库中存储的信息和知识进行定量分析完成预测。
本系统还处在第一期研发和试用阶段，基本满足了实用性，但还有许多工作值得进一步完善和发展，以便更好地满足用户的业务需求。
The grain trade is a special trade in economy，and its production and management is in accord with economic rules. The requirement how to estimate the future trend of some economic indexes in order to provied assistant decision for grain management department become imperious.
The existing forecasting tools are not fit for generic users by reason of their high price and speciality.On the other hand,some special forecasting software just aim at the given domain or the appointed project,and most of them are integrated into MIS or ERP systems,which strongly depend on the running environment.Accordingly,the requirement to produce a forecasting system for grain trade whose users are decisionmakers in grain management department is brought forward.
After studied forecasting models and methods,we make use of economic forecasting technology in "Intelligent DSS of Grain"(IDSSG),and combined the advantage and specialities of data warehouse with forecasting technology to realized the grain forecasting system based on data warehouse.
This paper studies and compares several kinds of frequently used forecasting models:Exponential Smoothing Model,Season Adjustment Model,Linear Random Model,Linear Regression Model,Artificial Neural Network Model.In these models,Exponential Smoothing Model,Season Adjustment Model and Linear Random Model are used in time series analysis,while Linear Regression Model is fit for cause and effect analysis,and Neural Network Model is able to both time series analyze and cause and effect analyze.
With the target of understandability and practicability,we apply those forecasting models mentioned in grain trade and realized a Grain Forcasting and Decision System(GFDS) for common program managers,which take the IDSSG as background.
Based on data warehouse of grain,the GFDS integrates database,model base and method base into one system.The data warehouse stores and synthesizes the forecast topic and information.The model base and method base provide quantitative analysis(model computation) and assistant decision information.The database include basic data,time series data and pertinent data dictionary;The content of model base is forecasting instance dictionary,model parameter interface tables,model storage tables
and forecasting result storage tables;The method base is made up of forecasting method dictionary and algorithm class base.
The integrated forecast environment is Designed and realized,which supports customizing forcast instance.There are several kinds of forcast method to choose and users will get understandable results.After compare the RMS(Rooted Means of Square Errors) and the curve's nicety,it will be decided which model will be chosen. The whole process including forecast instance definition,forecast method choosing,model creating,model proving,model storage and results analysing of multi-scenarios is realized.The system supplies many functions such as data management,forecast application,data query,operation guidance and error disposal.The friendly graph user interface directs users to work and feed the results back in a understandable way.It's no need for users to know much about the forecast technology to control and adjust the system easily,which allows of forecast the complex and synthetical science to serv for common users.So the system is much utilizable.
The data management module comprises three sub-functions which are data produce,data operation and data maintenance.It is used to produce,input,audit and maintain the data dictionaries and time series data,so it is the base of forecast model building and forecast running.
The data query module is able to query data dictionaries,data information and forecast models in order to make users know clearly about all kinds of data and model,and then user can analys the data and give some judgement and estimation.
The operation guidance and error disposal module catchs the errors and abnomities occured during the process of operation and prompts the user by dialog and text in a friendly way.Meanwhile,some advice about how to dea
已有的预测工具由于价格昂贵，专业性很强，不适于一般用户使用。而一些专门的预测软件只针对某一特定领域或某一指定项目，而且大多集成在MIS或ERP系统中，具有很强的运行环境依赖性。因此开发一个专门针对粮食行业，供粮食主管部门的普通决策者使用的预测系统作为需求被提出。
本文对预测模型方法进行了研究后，将经济预测技术应用于粮食智能决策支持系统，将数据仓库的优势和特点与预测技术相结合，实现了基于数据仓库的粮食预测系统。
本文研究和比较了几种常用的预测模型：指数平滑模型，季节调整模型，线性随机模型，线性回归模型，神经网络模型。其中指数平滑模型，季节调整模型，线性随机模型用于时间序列分析，线性回归模型用于因果分析，神经网络模型既可用于因果分析又可用于时间序列分析。
将上述几种预测模型应用于粮食行业，以易理解，实用性为目标，以粮食智能决策支持系统为背景实现了一个面向一般计划管理人员的粮食预测系统。
以粮食数据仓库为基础，建立了数据库、模型库和方法库三库合一的预测数据仓库。数据仓库实现对预测主题和信息的存储与综合。模型库和方法库为决策问题提供定量分析(模型计算)和辅助决策信息。数据库的构成包括各种基础数据，时序数据和相关的数据字典；模型库的内容有预测实例字典，模型参数接口表，模型存储表，预测结果存储表；方法库由预测方法字典和预测算法类库构成。
设计实现了一个预测集成环境，直接支持用户自定义预测实例，对粮食产量和价格进行预测。用户可以选择多种预测方法，得到易于理解的预测结果，对模型的均方误差和曲线拟合情况进行判断比较，综合考虑模型的取舍。实现包括预测实例定义，预测方法选择，模型建立，模型检验，模型存储，多方案预测结果分析在内的完整预测过程。提供数据管理，预测应用，数据查询，操作引导和容错处理等全方面的功能，用友好的图形用户界面引导用户进行操作，并以易于理
解的方式将结果反馈给用户，用户不需要具备很多预测的专业知识，就可以轻松地控制和调整预测系统的运行，使预测这一门复杂的综合科学能够为一般用户服务，具有很强的实用性。
数据管理模块由数据生成、数据操作和数据维护三个子功能组成，用于数据字典和时序数据的生成、录入、审核和维护，是预测建模和预测执行的基础。
数据查询为用户提供数据字典查询、数据信息查询、预测模型查询功能，使用户对各种数据和模型有详尽的了解，并能对数据进行分析，作出粗略的判断和估计。
容错处理及操作引导捕捉操作过程中出现的错误和异常，以友好的对话框和文字方式给用户以提示，并给出处理建议。
预测应用模块运用数据库、模型库和方法库中存储的信息和知识进行定量分析完成预测。
本系统还处在第一期研发和试用阶段，基本满足了实用性，但还有许多工作值得进一步完善和发展，以便更好地满足用户的业务需求。
The grain trade is a special trade in economy，and its production and management is in accord with economic rules. The requirement how to estimate the future trend of some economic indexes in order to provied assistant decision for grain management department become imperious.
The existing forecasting tools are not fit for generic users by reason of their high price and speciality.On the other hand,some special forecasting software just aim at the given domain or the appointed project,and most of them are integrated into MIS or ERP systems,which strongly depend on the running environment.Accordingly,the requirement to produce a forecasting system for grain trade whose users are decisionmakers in grain management department is brought forward.
After studied forecasting models and methods,we make use of economic forecasting technology in "Intelligent DSS of Grain"(IDSSG),and combined the advantage and specialities of data warehouse with forecasting technology to realized the grain forecasting system based on data warehouse.
This paper studies and compares several kinds of frequently used forecasting models:Exponential Smoothing Model,Season Adjustment Model,Linear Random Model,Linear Regression Model,Artificial Neural Network Model.In these models,Exponential Smoothing Model,Season Adjustment Model and Linear Random Model are used in time series analysis,while Linear Regression Model is fit for cause and effect analysis,and Neural Network Model is able to both time series analyze and cause and effect analyze.
With the target of understandability and practicability,we apply those forecasting models mentioned in grain trade and realized a Grain Forcasting and Decision System(GFDS) for common program managers,which take the IDSSG as background.
Based on data warehouse of grain,the GFDS integrates database,model base and method base into one system.The data warehouse stores and synthesizes the forecast topic and information.The model base and method base provide quantitative analysis(model computation) and assistant decision information.The database include basic data,time series data and pertinent data dictionary;The content of model base is forecasting instance dictionary,model parameter interface tables,model storage tables
and forecasting result storage tables;The method base is made up of forecasting method dictionary and algorithm class base.
The integrated forecast environment is Designed and realized,which supports customizing forcast instance.There are several kinds of forcast method to choose and users will get understandable results.After compare the RMS(Rooted Means of Square Errors) and the curve's nicety,it will be decided which model will be chosen. The whole process including forecast instance definition,forecast method choosing,model creating,model proving,model storage and results analysing of multi-scenarios is realized.The system supplies many functions such as data management,forecast application,data query,operation guidance and error disposal.The friendly graph user interface directs users to work and feed the results back in a understandable way.It's no need for users to know much about the forecast technology to control and adjust the system easily,which allows of forecast the complex and synthetical science to serv for common users.So the system is much utilizable.
The data management module comprises three sub-functions which are data produce,data operation and data maintenance.It is used to produce,input,audit and maintain the data dictionaries and time series data,so it is the base of forecast model building and forecast running.
The data query module is able to query data dictionaries,data information and forecast models in order to make users know clearly about all kinds of data and model,and then user can analys the data and give some judgement and estimation.
The operation guidance and error disposal module catchs the errors and abnomities occured during the process of operation and prompts the user by dialog and text in a friendly way.Meanwhile,some advice about how to dea
引文
[1] 张有为.预测的数学方法.国防工业出版社.1991.10.
[2] 周复恭等.应用数理统计学.中国人民大学出版社.1989.9.
[3] 熊大国.随机过程理论与应用.国防工业出版社.1991.1.
[4] 黎子良.统计、推断、决策.南开大学出版社.1988.1.
[5] 冯文权.经济预测与决策技术.武汉大学出版社.1994.4.
[6] 魏权龄等.数量经济学.中国人民大学出版社.1998.9.
[7] 华伯泉.经济预测的统计方法.中国统计出版社.1988.2.
[8] 杨建刚.人工神经网络实用教程.浙江大学出版社.2001.1.
[9] 周春光，梁艳春.计算智能.吉林大学出版社.2001.11.
[10] Box D E P, Jenkins G. Time Series Analysis, Forecasting and Control. San Francisco, CA: Holden Day，1976.
[11] Michael C. Mozer.Neural net architectures for temporal sequence processing，Redwood City CA:Addison-Wesley Publishing，1994.
[12] Spryros Makridakis，Michele Hibon.ARMA Models and the Box-Jenkins Methodology. Journal of Forecasting, 1997,16: 147～163.
[13] Inmon W H..Building Data Warehouse.Second Edition.John Wiley,1996.
[14] D.E. Rumelhart, G.E. Hinton, R.J. Williams. Learning Internal Representations by Error Propagation. Parallel Distributed Processing, Volume 1:MIT Press, Cambridge, MA: 318～362. 1986.
[15] 张维明，数据仓库原理与应用，电子工业出版社，2001.
[16] 黄梯云，智能决策支持系统，电子工业出版社，2001.
[17] 寿志勤，潘皖印，朱少民.集成环境下的预测应用网络系统的设计与实现. 软件学报,第6卷第9期：551～559. 1995.9.
[18] 林海，孙吉贵.预测及其主要模型的实现系统.吉林大学学报信息科学版第20卷第4期:58～63.2002.11.
[19] 白洪涛，孙吉贵，莫旭，杨凤杰. 一个专用ETL程序的实现.计算机应用.2004年第2期:101～105.2004.2.
[20] 莫旭，杨凤杰，白洪涛，孙吉贵.基于数据仓库的粮食预测系统.已投稿.
[2] 周复恭等.应用数理统计学.中国人民大学出版社.1989.9.
[3] 熊大国.随机过程理论与应用.国防工业出版社.1991.1.
[4] 黎子良.统计、推断、决策.南开大学出版社.1988.1.
[5] 冯文权.经济预测与决策技术.武汉大学出版社.1994.4.
[6] 魏权龄等.数量经济学.中国人民大学出版社.1998.9.
[7] 华伯泉.经济预测的统计方法.中国统计出版社.1988.2.
[8] 杨建刚.人工神经网络实用教程.浙江大学出版社.2001.1.
[9] 周春光，梁艳春.计算智能.吉林大学出版社.2001.11.
[10] Box D E P, Jenkins G. Time Series Analysis, Forecasting and Control. San Francisco, CA: Holden Day，1976.
[11] Michael C. Mozer.Neural net architectures for temporal sequence processing，Redwood City CA:Addison-Wesley Publishing，1994.
[12] Spryros Makridakis，Michele Hibon.ARMA Models and the Box-Jenkins Methodology. Journal of Forecasting, 1997,16: 147～163.
[13] Inmon W H..Building Data Warehouse.Second Edition.John Wiley,1996.
[14] D.E. Rumelhart, G.E. Hinton, R.J. Williams. Learning Internal Representations by Error Propagation. Parallel Distributed Processing, Volume 1:MIT Press, Cambridge, MA: 318～362. 1986.
[15] 张维明，数据仓库原理与应用，电子工业出版社，2001.
[16] 黄梯云，智能决策支持系统，电子工业出版社，2001.
[17] 寿志勤，潘皖印，朱少民.集成环境下的预测应用网络系统的设计与实现. 软件学报,第6卷第9期：551～559. 1995.9.
[18] 林海，孙吉贵.预测及其主要模型的实现系统.吉林大学学报信息科学版第20卷第4期:58～63.2002.11.
[19] 白洪涛，孙吉贵，莫旭，杨凤杰. 一个专用ETL程序的实现.计算机应用.2004年第2期:101～105.2004.2.
[20] 莫旭，杨凤杰，白洪涛，孙吉贵.基于数据仓库的粮食预测系统.已投稿.