支持向量机模型改进及在短期边际电价预测中的应用

发布时间：2018-08-02 13:40

【摘要】：支持向量机预测模型虽然已被证实为一类优于神经网络等其他智能算法模型的新兴预测技术，但其本身依旧不够完美，仍然存在尚需改进的地方。本文选取了支持向量机家族中的最小二乘支持向量机预测模型进行研究，并发现在使用该模型进行训练时，由于其对训练样本的各个输入向量的惩罚度是一样的，即在训练过程中对所有输入点都等同看待，因此，一旦在训练样本集中出现离群点时，该点就会对预测模型产生一定的不良影响。这样一来，势必会降低系统的鲁棒性，引起过学习，降低了预测模型的推广能力和预测效果。本文关注了此项问题，，尝试对原模型进行改进，从而使得改进后的预测模型能够很好地克服上述理论缺陷，优化预测过程与结果，达到管理科学理论创新的目的。为了达到上述目的，作者在经过了大量的研究与参考后，运用模糊数学中有关隶属度问题的相关理论和方法与原预测模型相结合，建立了一套新的预测模型。为了考察该模型是否能达到对原预测模型的改进效果，并使得预测结果更佳，文章选取了美国PJM电力市场在2012年内的一些有关日前交易的负荷与边际电价数据进行预测。首先，运用新模型对多个测试日的边际电价进行预测，通过Matlab软件加以实现，并得出了多个测试日的相关预测结果。之后，又运用原模型在相同条件下对相同日期的边际电价进行预测并得出预测结果。最后，通过将两种模型在多个测试日的预测值与真实值之间的误差进行比较，发现新模型的预测误差明显低于原模型。这一点证实了新模型确实起到了对原模型明显的改进效果，所得出的预测结果对未来发电企业的管理者作进一步分析与决策更加具有参考意义。另外，改进后的新模型在算法实现方面较之改进前的原模型并没有明显增加计算的复杂度，这又从另一个方面证实了该改进方案的有效性。综上所述，本文通过改进而得到的新模型可以被广泛推广，而且范围不仅局限在电价预测领域之内。这对于管理科学理论创新与应用等领域的研究都具有一定的贡献与启发意义。
[Abstract]:Support vector machine (SVM) prediction model has been proved to be a kind of new prediction technology which is superior to other intelligent algorithm models such as neural network, but it itself is still not perfect, and there are still some problems to be improved. In this paper, the prediction model of least squares support vector machine (LS-SVM) in the family of support vector machines (SVM) is selected, and it is found that the penalty degree of each input vector of the training sample is the same when the model is used for training. That is, all input points are treated equally in the training process, so once the outliers appear in the training sample set, the outliers will have a certain adverse effect on the prediction model. In this way, the robustness of the system will be reduced, and the overlearning will be caused, and the generalization ability and prediction effect of the prediction model will be reduced. This paper focuses on this problem and attempts to improve the original model, so that the improved prediction model can overcome the above theoretical defects, optimize the prediction process and results, and achieve the purpose of theoretical innovation in management science. In order to achieve the above purpose, after a lot of research and reference, the author combines the theory and method of membership degree in fuzzy mathematics with the original prediction model, and establishes a new prediction model. In order to investigate whether the model can improve the original forecasting model and make the forecast result better, this paper selects some data of pre-day trading load and marginal electricity price of PJM electricity market in the United States in 2012 to forecast. Firstly, the new model is used to predict the marginal electricity price of multiple test days, which is realized by Matlab software, and the related prediction results of multiple test days are obtained. After that, the original model is used to predict the marginal electricity price of the same date under the same conditions and the prediction results are obtained. Finally, it is found that the prediction error of the new model is obviously lower than that of the original model by comparing the errors between the predicted values and the real values of the two models on multiple test days. This proves that the new model does improve the original model obviously, and the predicted results are more useful for the future power generation enterprise managers to make further analysis and decision making. In addition, the algorithm implementation of the improved new model does not significantly increase the computational complexity compared with the original model, which proves the effectiveness of the improved scheme from another aspect. To sum up, the new model can be extended widely, and the scope is not only limited to the field of electricity price prediction. It has a certain contribution and enlightening significance to the research of management science theory innovation and application.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F224;F416.61

【参考文献】