基于粒子群优化的最小二乘支持向量机天然气负荷预测方法研究

发布时间：2018-03-09 23:33

本文选题：天然气负荷预测　切入点：支持向量机　出处：《华东理工大学》2013年硕士论文　论文类型：学位论文

【摘要】：天然气是一种绿色、经济、高效、便捷、安全环保的能源,提高天然气在能源消费结构中的比例,不仅有利于促进节能减排,又能够促进经济与社会的可持续发展,同时天然气作为一种城市燃气,与人们生活和工业生产息息相关,是反映人们生活消费水平的一个重要指标。因此,科学的天然气负荷预测不仅仅关系着天然气公司的利益问题,同时还关系着整个天然气行业的建设和发展,与广大的人民生活也息息相关。本文基于天然气负荷预测的背景、国内外研究现状以及课题意义的研究,分析天然气负荷的特点、规律及与各种影响因素之间的非线性关系,根据支持向量机(SVM)能够较好地解决非线性、高维数以及局部极小点等实际问题的优势,本文提出采用支持向量回归(SVR)方法建立天然气负荷预测模型并进入深入研究。首先基于SVR的研究,采用训练误差的平方代替松弛变量,将不等式约束改进为等式约束,进而提出最小二乘支持向量回归(LS-SVR)的天然气负荷预测模型,从而避免求解二次规划问题,提高模型训练的速度,但是LS-SVR丢失了SVR的松散性和鲁棒性,导致模型精确度产生一定影响,为解决这些问题,又提出基于加权最小二乘支持向量回归(WLS-SVR)的天然气负荷预测模型。同时,考虑到模型参数对于预测结果精度的影响也至关重要,提出利用粒子群算法(PSO)来优化WLS-SVR中的参数,得到基于PSO-WLS-SVR的预测模型,以进一步提高预测精度。最后,本文使用天然气1980-2011年的相关数据进行研究,分别对基于SVR、WLS-SVR、PSO-WLS-SVR的预测模型进行仿真并比较。仿真结果表明：WLS-SVR的模型预测结果优于SVR模型,且PSO-WLS-SVR的预测模型预测误差更小,从而说明PSO进行参数优化对提高预测精度有一定的优势,也说明PSO-WLS-SVR方法具有一定的有效性和优越性,由此可以说明基于粒子群算法的最小二乘支持向量回归预测方法具有一定的研究价值和社会意义。
[Abstract]:Natural gas is a kind of green, economical, efficient, convenient, safe and environmentally friendly energy. Increasing the proportion of natural gas in energy consumption structure is not only conducive to energy saving and emission reduction, but also to the sustainable development of economy and society. At the same time, as a kind of city gas, natural gas is closely related to people's life and industrial production, and is an important index to reflect people's living consumption level. Scientific natural gas load forecasting is not only related to the interests of natural gas companies, but also related to the construction and development of the entire natural gas industry, and is closely related to the lives of the broad masses of people. Based on the background of natural gas load forecasting, the research status at home and abroad and the significance of the subject, the characteristics, laws and nonlinear relationship between natural gas load and various influencing factors are analyzed in this paper. According to the advantages of support vector machine (SVM), it can solve nonlinear, high dimension and local minima problems. In this paper, the support vector regression (SVR) method is used to establish the natural gas load forecasting model and it is studied deeply. Firstly, based on the research of SVR, the square of the training error is used to replace the relaxation variable, and the inequality constraint is improved to equality constraint. Furthermore, a natural gas load forecasting model based on least squares support vector regression (LS-SVR) is proposed, which avoids solving quadratic programming problems and improves the speed of model training, but LS-SVR loses the looseness and robustness of SVR. In order to solve these problems, a natural gas load forecasting model based on weighted least squares support vector regression (WLS-SVR) is proposed. Considering that the model parameters are also very important to the precision of prediction results, a particle swarm optimization algorithm (PSO) is proposed to optimize the parameters in WLS-SVR and obtain the prediction model based on PSO-WLS-SVR to further improve the prediction accuracy. Finally, using the data of natural gas from 1980 to 2011, this paper simulates and compares the prediction model based on SVR / WLS-SVR / PSO-WLS-SVR respectively. The simulation results show that the prediction result of WLS-SVR is better than that of SVR model, and the prediction error of PSO-WLS-SVR model is smaller. It shows that the parameter optimization of PSO has some advantages to improve the prediction accuracy, and that the PSO-WLS-SVR method has certain effectiveness and superiority. It can be concluded that the least squares support vector regression prediction method based on particle swarm optimization algorithm has certain research value and social significance.
【学位授予单位】：华东理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TU996.3;TP18

【参考文献】