短期电力负荷混合预测模型研究

发布时间：2018-08-13 13:25

【摘要】：近年来,随着我国的电力工业体制不断改革,电力市场竞争机制已经逐步形成,在电力市场竞争机制下,如何准确的预测短期电力负荷需求变得越来越重要。负荷预测精度的提高,有利于发电厂商及时调整供电策略,从而使利润最大化；同时也有利于普通用户根据需求来制定购电计划,使成本最小化。然而,由于受到气象、经济、季节、政策等众多因素的影响,使得准确预测电力负荷的工作变的十分困难和复杂。为此,本文在研究了国内外短期电力负荷预测的现状以及电力负荷的影响因素的基础上,提出了一种基于聚类分析和最小二乘支持向量机的混合预测模型。本文首先对澳大利亚新南威尔士州和昆士兰州的电力负荷特性进行了深入地分析,然后利用SOM神经网络和K-means两种聚类算法选择相似日,构造相似日的训练集合和测试集合并利用LSSVM模型进行预测。针对LSSVM核函数参数与正则化参数的优化问题,本文结合MATLAB并行工具箱,基于先大后小的原则提出了一种新的并行二次网格优化算法,该算法与传统的网格优化算法相比,可以显著降低寻优时间,同时预测精度损失也不大。最后,通过仿真实验纵向对比表明,本文提出的混合预测模型中,SOM-LSSVM的预测精度高于K-means-LSSVM和单一的LSSVM模型；横向对比表明,本文混合预测模型的预测结果同样优于单一的ARIMA模型.ARIMA-BP模型以及WT-LSSVM模型,说明本文提出的混合模型是一个比较理想的短期电力负荷预测方法。
[Abstract]:In recent years, with the continuous reform of the electric power industry system in our country, the competitive mechanism of the electricity market has been gradually formed. Under the competition mechanism of the electricity market, how to accurately predict the short-term power load demand becomes more and more important. The improvement of load forecasting accuracy is helpful for power generation manufacturers to adjust their power supply strategy in time to maximize profit, and to make purchase plan according to the demand of ordinary customers, so as to minimize the cost. However, due to the influence of weather, economy, season, policy and so on, it is very difficult and complicated to predict the power load accurately. In this paper, a hybrid forecasting model based on clustering analysis and least squares support vector machine (LS-SVM) is proposed on the basis of studying the current situation of short-term power load forecasting at home and abroad and the influencing factors of power load. In this paper, the power load characteristics of New South Wales and Queensland are analyzed, and then the similar days are selected by using SOM neural network and K-means clustering algorithm. The training set and test set of similar days are constructed and predicted by LSSVM model. Aiming at the optimization of LSSVM kernel function parameters and regularization parameters, this paper proposes a new parallel quadratic mesh optimization algorithm based on the principle of large first and then small, combined with MATLAB parallel toolbox. The algorithm is compared with the traditional mesh optimization algorithm. The optimization time can be significantly reduced, and the loss of prediction accuracy is small. Finally, the longitudinal comparison of simulation experiments shows that the prediction accuracy of SOM-LSSVM is higher than that of K-means-LSSVM and single LSSVM model, and the horizontal comparison shows that the prediction accuracy of the hybrid prediction model is higher than that of the K-means-LSSVM model and the single LSSVM model. The results of the hybrid forecasting model are also superior to those of the single ARIMA model, the ARIMA-BP model and the WT-LSSVM model, which shows that the hybrid model is an ideal short-term power load forecasting method.
【学位授予单位】：兰州大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TM715

【相似文献】