基于贝叶斯证据框架优化的LS-SVM预测模型在空间电力负荷预测方法中的应用研究

发布时间：2018-07-24 17:35

【摘要】：因短期电力系统负荷具有明显的随机性,使得其负荷预测工作不容易找到内在发展规律,故很不容易提高短期负荷预测的准确性。因依据简单基础理论的传统负荷预测没有充分考虑影响电力系统负荷发展的关键性因素被研究学者们逐渐淘汰。近年来,因为国家政策的完善使得城市发展越来越规范合理,城市土地性质也越来越明确,这就为电力负荷预测提供了一个优越的有基础的更广阔的研究空间。使得很多研究学者也对空间负荷预测方法进行了系统深入地研究以得到较高的短期负荷预测精度。其中短期负荷预测中发展潜力较大的预测方法当属最小二乘支持向量机(LS-SVM)。随着对其研究的扩展,研究学者们对LS-SVM预测方法的探究越来越深刻。基于深刻了解和意识到短期负荷预测的本质理论和深刻意义,本研究论文选择了一种优化的LS-SVM预测模型来对短期负荷预测进行一个更深广的研究。本文多方面综合考虑了外部影响因素,如负荷功能区土地使用性质和天气等。该预测模型集多种因素所校正并得以自学习适应,使其更加智能化,也更加符合实际环境,但却使模型更加复杂且不易模拟。复杂性的提高还有各种因素的考虑使得该预测模型的鲁棒性下降,使得该预测模型的稳定性被一些影响因素破坏,故本文特别应用由各种因素下误差得出的权系数因子修改其预测模型以提高鲁棒性和稳定性。另外,本文还为更快更准确地搭建预测模型采用了贝叶斯证据框架优化模式。通过此模式优化,本文预测模型更具有实用性和高效性。本文第一章先着重有条理地介绍了电力负荷预测研究原理,第二章研究了进行负荷数据采集的重要平台-地理信息系统(GIS),而且也分析了数据前期处理对空间负荷分类分区的重要性。接下来,第三章则为构建预测模型奠定数据样本基础提供了理论依据。第四章则统筹考虑各种外界的影响因素,研究了一种加权处理的最小二乘支持向量机,并应用贝叶斯证据框架对其模型进行优化。本文最后,完善了整体预测流程,并在电力GIS图层上选取一个行政功能负荷区,采用本文构建的预测模型进行为期10日的负荷预测,与优化前模型输出的负荷作对比,证明本文研究的优化后预测模型的先进可靠性,具有实用价值。可在社会上推广以适应电力发展。
[Abstract]:It is difficult to improve the accuracy of short-term load forecasting because of the obvious randomness of short-term power system load, which makes it difficult to find the inherent development law in load forecasting work. Because the traditional load forecasting based on simple basic theory does not fully consider the key factors that affect the development of power system load, it is gradually eliminated by researchers. In recent years, because of the perfection of national policy, the urban development is more and more standardized and reasonable, and the nature of urban land is becoming more and more clear, which provides a superior and broader research space for power load forecasting. As a result, many researchers have also carried out systematic and in-depth research on spatial load forecasting methods in order to obtain higher short-term load forecasting accuracy. The least square support vector machine (LS-SVM) is the most promising method in short term load forecasting. With the expansion of the research, the research on LS-SVM prediction methods is more and more profound. Based on the deep understanding and realization of the essential theory and significance of short-term load forecasting, this paper chooses an optimized LS-SVM forecasting model to study short-term load forecasting more deeply. In this paper, external factors, such as the nature of land use and weather in the load function area, are considered comprehensively. The prediction model is calibrated by many factors and adapts to self-learning to make it more intelligent and more suitable to the actual environment, but it makes the model more complex and difficult to simulate. The increase of complexity and the consideration of various factors make the robustness of the prediction model decrease, and the stability of the prediction model is destroyed by some influential factors. Therefore, in this paper, the prediction model is modified by the weight factor derived from the error under various factors to improve the robustness and stability. In addition, this paper adopts Bayesian evidence framework optimization model to build prediction model more quickly and accurately. Through the optimization of this model, the prediction model in this paper is more practical and efficient. In the first chapter of this paper, the research principle of power load forecasting is introduced systematically. In the second chapter, the importance of data pre-processing to spatial load classification and partition is analyzed, which is an important platform for load data acquisition, which is called Geographic Information system (GIS),). Then, the third chapter provides the theoretical basis for building the prediction model. In chapter 4, considering all kinds of external factors, we study a weighted least-squares support vector machine (LS-SVM), and apply Bayesian evidence framework to optimize its model. Finally, the whole forecasting process is perfected, and a load area of administrative function is selected on the GIS layer of electric power. The forecasting model constructed in this paper is used to carry out the load forecasting for 10 days, which is compared with the load output from the model before optimization. It is proved that the advanced reliability of the optimized prediction model studied in this paper is of practical value. Can be promoted in the community to adapt to the development of electricity.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM715

【相似文献】