基于改进数据流和小波包分析的超短期负荷预测方法研究

发布时间：2019-02-20 20:00

【摘要】：随着现代电力系统建设规模的不断扩大，全国电网互联工程的逐步开展，越来越复杂的电网结构和运行方式将给电力系统的安全运行和电能质量带来较大的威胁，向电力负荷预测的实时性和精确度提出了更大的挑战。超短期负荷预测利用最新负荷信息，实时跟踪电力系统负荷变化，是动态电网安全检测、自动发电控制和紧急状态处理等的基本前提和重要依据。快速、精确的预测结果能够指导电力部门及时维护电网频率平衡，保证电网运行的安全性和经济性。因此研究兼顾预测实时性和准确性的超短期负荷预测实用方法具有重要意义。本文从负荷构成的基本模型入手，研究了负荷特性及其与相关影响因素之间的关系；针对各影响因素，研究其对负荷变化的具体影响作用，结果表明：时间因素影响较大，使负荷呈现出明显的周期性；天气因素与负荷变化之间有一定的相关性；不确定性因素的影响使负荷表现出较强的波动性，其规律难以把握。根据不同的负荷变化规律，探讨了相应的预测思路和方法。提出了基于改进数据流在线分割的超短期负荷预测模型：利用数据流实时处理技术进行超短期负荷预测，，其快速分段预测能力避免了重复建模，增强了实时性；静态提取蕴含天气因素和负荷周期特性作用的短期负荷预测结果，对分割点进行实时修正，有效地增加了历史信息利用率，提高了分割点预测精度，同时保证了预测实时性；经实际算例检验，结果表明该模型的预测准确性和实时性均优于几种常规超短期预测算法，解决了预测精度与预测速度相互制约的矛盾，同时降低了拐点预测误差，并在天气突变时也具有稳定的适应性。进一步考虑负荷随机波动分量的影响作用，建立基于小波包分析的超短期负荷预测方法：通过小波包分析对负荷随机分量进一步分解，便于深入分析随机分量特性；对分解后的各小波包空间信号进行单支重构，根据各组负荷子序列分量特性，分别建立预测模型，并将各子序列分量预测值相加获得负荷预测结果。经算例分析表明，该算法具有较高的预测精度和稳定的预测效果。
[Abstract]:With the continuous expansion of the scale of modern power system construction and the gradual development of the national power network interconnection project, the more and more complex power network structure and operation mode will bring a great threat to the safe operation and power quality of the power system. It presents a greater challenge to the real-time and accuracy of power load forecasting. The ultra-short-term load forecasting uses the latest load information to track the load change of the power system in real time, which is the basic premise and important basis of dynamic power network safety detection, automatic generation control and emergency processing. The fast and accurate prediction results can guide the power sector to maintain the power network frequency balance in time and ensure the security and economy of the power network operation. Therefore, it is of great significance to study the practical method of super-short-term load forecasting which takes into account the real-time and accuracy of forecasting. Starting with the basic model of load composition, this paper studies the load characteristics and the relationship between load characteristics and related factors. In view of the influence factors, the concrete effect of the factors on the load change is studied. The results show that the time factor has a great influence on the load, which makes the load appear obvious periodicity, and the weather factor has certain correlation with the load change. The influence of uncertain factors makes the load exhibit strong volatility, and its law is difficult to grasp. According to the different law of load change, the corresponding forecasting ideas and methods are discussed. The ultra-short-term load forecasting model based on the improved on-line segmentation of data flow is proposed. The fast segment forecasting ability avoids repeated modeling and enhances the real-time performance of ultra-short-term load forecasting by using data stream real-time processing technology. Static extraction of short-term load forecasting results containing weather factors and load cycle characteristics can effectively increase the utilization rate of historical information and improve the accuracy of forecasting points, and ensure the real-time prediction. The results of practical examples show that the prediction accuracy and real-time performance of the model are better than those of several conventional ultra-short term prediction algorithms, and the contradiction between prediction accuracy and prediction speed is solved, and the error of inflection point prediction is reduced at the same time. It also has a stable adaptability in the event of sudden change in the weather. Considering the influence of load random fluctuation component further, an ultra-short-term load forecasting method based on wavelet packet analysis is established: the random load component is further decomposed by wavelet packet analysis, which is convenient for further analysis of stochastic component characteristics; The decomposed wavelet packet spatial signal is reconstructed by single branch. According to the characteristics of each group of load sub-sequence components, the prediction model is established, and the load forecasting results are obtained by adding the predicted values of each sub-sequence component. An example shows that the algorithm has high prediction accuracy and stable prediction effect.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM715

【参考文献】