小样本条件下基于VMD-SVM的电网技改项目科技投资预测研究

Transmission Project Technology Annual Investment Forecasting Under the Small Sample Conditions Based on VMD-SVM

摘要:

随着我国电力改革体制的不断推进,未来输配分离的市场环境下,电网的盈利模式将以输配电服务为主,电网在电力市场中的地位将发生较大的变化。新环境下要求电网企业对自身的投资、运行以及成本需要更精确的控制。电网技改工作作为电网科技更新换代的日常工作,对电网运营成本具有较大的影响,因此,科学合理地预测电网技改项目每年的科技投资额就尤为重要。电网技改科技项目的年度投资历史数据由于样本数量少,难以构建基于统计数据的常规预测模型对技改项目的年度科技投资进行科学和客观的预测。本文针对电网技改项目科技投资年度预测属于小样本预测的特点,选取了支持向量机SVM模型作为预测模型,并且在预测开始之前进行变分模态分解(VMD),通过将技改科技投资数据分解为特征各异的子序列,将每一个子序列数据通过不同的SVM模型进行预测,得到各子序列的预测结果后,对子序列的预测结果进行叠加,从而得到最终预测结果。通过某区域电网的实际数据验证,本文提出的VMD-SVM的电网技改项目科技投资预测方法能够有效改进预测精度,预测精度为1.51%,而单采用同参数的SVM模型的预测精度仅有2.02%,本文提出的模型具有更高的精度,证实了本文提出方法的有效性。

Abstract:

The electricity market will realize the separation of transmission and distribution, in the new environment, the main profit model of the power grid company will be dominated by transmission and distribution services, it will require the grid company to control its investment and cost accuracy. As a routine maintenance work to ensure the safe and stable transportation of electric energy, the grid technical transformation project has a great impact of the grid operation cost. Due to the lack of the history data, it is difficult to forecast the project’s investment, and the current investment always relies on experience. It needs to construct a quantitative forecasting model to make scientific and objective predictions of investment in technological transformation projects. In this paper, the variational mode decomposition (VMD) is employed to decomposed the history data into subsequences, then the different subsequences are used to predict each subsequence data through different SVM models. After obtaining the prediction results of each subsequence, the prediction results of the subsequences are added to obtain the final prediction result. Finally, from a case study of a certain grid company, the proposed VTD-SVM forecasting model has better performance on the grid technology transformation project investment prediction than the default SVM, the accuracy is 1.51% and 2.02% perspectivity, it confirms the effectiveness of the proposed method.