《基于支持向量机模型的城市天然气需求量中长期预测研究》由会员分享，可在线阅读，更多相关《基于支持向量机模型的城市天然气需求量中长期预测研究（59页珍藏版）》请在金锄头文库上搜索。
1、摘 要 基于支持向量机模型的城市天然气需求量基于支持向量机模型的城市天然气需求量 中长期预测研究中长期预测研究 作者简介： 刘鑫， 男， 1984 年 10 月生， 师从成都理工大学匡建超教授， 2009年 06 月毕业于成都理工大学管理科学与工程专业，获得管理学学硕士学位。 摘 要 摘 要 预测学是一门高集成性的复杂科学， 是人类根据历史经验结合严密的逻辑推理对未来的一种估计，它能够指导计划的制定，并对最终的决策活动产生直接影响。同时，随着现代预测科学的发展，人类的决策过程变得更加理性，尤其是对于不确定性问题定量化的探求越来越重视，于是产生了数量众多的预测数学模型。自上世纪 90 年代以来，
2、支持向量机模型成为各国学者关注的焦点，它较好地解决了传统模型通常存在的过学习、维数灾难、局部极小等问题，改变了人们单纯追求经验风险最小的思维牢笼，在社会的各个方面有了越来越多的应用。 在工业化的滚滚车轮下，煤炭石油的消耗量快速增长，环境问题业已成为人类不得不面对的痛楚。在这种背景下，天然气以其丰富的储量、优良的品质、较低的污染被世界各国大力推崇，资源争夺的大幕就此拉开。在我国，天然气正在经历着由推广到普及的过程，以气代油、以气发电和城市气化将进一步提升天然气的战略地位，供需矛盾问题逐渐显现。在城市天然气的供给环节，通常气源供应部门要与下游用气单位之间签订“照付不议” （Take or Pay）
3、协议，这就需要对燃气公司对于整个城市的需求量有个较为准确的估算。 基于此，本文在充分搜集和整理现有研究资料的基础上，首先简单阐述了常用天然气需求量预测方法的核心思想与适用范围， 分析了现有指标体系构建及变量提取环节的不足，较为系统的总结了影响天然气需求量中长期预测的主要因素，尝试建立起城市天然气需求量中长期预测的指标体系，并运用最佳子集法提取了五维输入变量，即：人均 GDP、天然气占能源消费比重、电力占能源消费比重、天然气供气量和人均可支配收入，以此构建了基于粒子群最小二乘支持向量机（PSO-LSSVM）的预测模型。对于该回归模型输入变量的预测而言，通过多次的建模实验，选定 GM（1，1）模型
4、预测人均 GDP 和人均可支配收入指标， 运用 LSSVM 递推滚动时间序列预测天然气占能源消费比重和电力占能源消费比重指标，通过拟合对数方程得到天然气供气量的预测值，将以上五个模型的输出结果带入 PSO-LSSVM 模型中，即可得到天然气需求量的预测值。最后I 成都理工大学硕士学位论文 以 CD 市 1990 至 2007 年的实际数据为样本，运用由 MATLAB 语言编写的 PSO程序和 LS-SVMlab Toolbox 软件进行模型训练与模型计算， 最终确定该市未来五年的天然气消费量将保持年 3.4%左右的增速，供需缺口有进一步放大的可能。对此，本文提出该市应在现有基础上进一步加强天然
5、气消费，适当提高天然气价格，加快城市天然气配套管网建设，拓宽天然气项目建设的融资渠道，努力争取多气源供应，做好能源消费的原始数据搜集和统计工作六项对策建议。 论文取得的进步有： 第一，从外部环境、内部环境及用户消费三个角度出发，较为系统的总结了影响天然气需求量中长期预测的主要因素，并运用最佳子集法提取输入变量，最终确定了五个核心指标：人均 GDP、天然气占能源消费比重、电力占能源消费比重、天然气供气量和人均可支配收入； 第二，首次将 PSO-LS-SVM 模型引入城市天然气需求量中长期预测中，该模型吸取了 PSO 算法快速全局寻优的特点与 LSSVM 模型解决非线性问题的先天优势， 求解速度快
6、， 泛化性较好， 通过对 CD 市天然气需求量的实际预测证实，与 BP 神经网络相比，该模型的平均相对误差（MAPE）较小，结果的合理性更高，这不仅丰富了天然气需求量预测方法，也可为同类研究提供参考借鉴。 关键词：支持向量机 城市天然气 需求量 中长期预测 II Abstract Study on Urban Natural Gas Demand Middle-Long Term Forecasting Base on SVM Model Introduction of the author: Liu Xin, male, was born in October, 1984 whose tut
7、or was Professor Kuang Jianchao . He graduated from Chengdu University of Technology in Management Science and Engineering major and was granted the Master Degree in June, 2009. Abstract Theory of Forecasting is a highly integrated and complex science. It uses long historical experience and strict l
8、ogical reasoning to estimate the future trends. Forecasting can instruct how to make plans, and have a direct impact on decision- making activities. At the same time, with the development of modern prediction science, human decision-making process become more rational, they pay more and more attenti
9、on to hunt after the answers of uncertainty questions. So a large number of forecasting mathematical models have been invented. Since the 1990s, SVM has become the focus of scholars all over the world, it has solved the problems of over learning, dimension disaster and local minimum effectively and
10、has changed the traditional thinking mode. Therefore, SVM will be more and more popular in all aspects of society. With the process of industrialization, the consumption of coal and oil is growing faster; environmental problem has become the pain of all the people. In the background of this situatio
11、n, natural gas is greatly rewarded around the world for its rich reserves, excellent quality and low pollution. So the natural gas resource competition is very intense. In China, Natural gas is undergoing the process of widespread from expanding. With the process of substituting oil with gas, using
12、gas for power generation and urban gasification, the strategic position will be improved a lot, and also the supply-demand contradiction question is appeared gradually. Normally, in supply link, gas resource supply department have to make “Take or Pay” protocol to downstream gas demand. Urban Gas Co
13、mpany needs to have a comparatively accurate estimation of the future demand. Firstly, based on finishing process of existing research data, this article demonstrated the commonly used methods of natural gas demand forecasting. III 成都理工大学硕士学位论文 Secondly, this paper analyzed the deficiency of existin
14、g index system and variable extraction. Then summarized the main influence factor of natural gas demand middle-long forecasting, established an index system for Urban Natural Gas Demand Middle-Long Term Forecasting. Thirdly, the Best Subset Regression method was used to extract input variables. Four
15、thly, based on that, this thesis built the PSO-LSSVM forecasting model. The GM (1, 1) model was chose to predict the input factors of regression model. Then LSSVM rolling time series model was used to forecast the consumption proportion of natural gas and electric power. We can get the natural gas s
16、upply quantum by fitting logarithmic equation. And then, we can obtain the predictive value by putting these results to the PSO-LSSVM model. Finally, this paper took the real data of CD city from 1990 to 2007 as an example. The results showed that the natural gas demand of this city will maintain an annual growth rate about 3.4 percent in the next five years; demand-supply gap may be further enlarged. So, we proposed six countermeasures for CD as follows: further strengthening the consu
