基于大数据的分布式短期负荷预测方法

发布时间：2018-03-17 11:00

本文选题：电力系统　切入点：短期负荷预测　出处：《湖南大学》2014年硕士论文　论文类型：学位论文

【摘要】：短期负荷预测作为能量管理系统的重要组成部分,预测误差的大小直接影响电网运行的安全性和经济性。在智能电网建设的大背景下,大规模电网互联以及数据采集成本的降低,可获取的数据类型愈加丰富,智能电网中的大数据趋势日益明显。针对供电区域辽阔的大电网,不同区域负荷特征各异,受外部因素影响规律不同。集中式负荷预测方法难以准确把握各区域的负荷变化规律,且对海量数据挖掘分析能力有限。因此,本文提出基于大数据的分布式短期负荷预测方法,综合利用大数据和人工智能方法的优势,提高负荷预测精度。以此为出发点,本文开展的研究工作如下：首先,提出基于大数据的负荷变化规律分析方法。在成本许可的情况下,针对不同类型的电力用户,应尽量全面收集用户以及环境信息进行负荷变化规律分析,以把握不同区域负荷特征与气象条件的差异。并以中国南方某省的A市和B市的实际负荷数据为基础,分别对两地的年、月、日负荷变化规律进行分析,重点研究夏季典型日负荷变化规律。由此得出不同地区负荷特征各异,受外部因素影响规律不同,在负荷预测工作中需建立有针对性的负荷预测模型。然后,提出基于大数据的分布式短期负荷预测方法。在行政区域划分和气象区域分布的基础上,对220kV以及以上的网架结构进行子网划分,通过评价子网负荷曲线相似度实现子网分类与归并。基于上述子网划分,分别建立子网负荷预测模型(ARIMA和BP神经网络),并提取相似度较高的影响因素向量和对应时刻的负荷作为子网负荷预测模型的输入量,对各个子网分别进行预测。根据子网的负荷预测值和各时刻比例系数的预测值,建立全网负荷预测模型,综合形成大电网的负荷预测结果。最后,以中国南方某省的实际负荷数据为基础,采用本文所提方法分别对工作日和休息日进行短期负荷预测,并与集中式负荷预测结果进行对比分析,以验证本文所提负荷预测方法的有效性和稳定性。算例分析表明本文所提方法在大电网中具有较好的预测精度,并且由于各个子网并行运行,有效降低了负荷预测所需的时间,在实际应用中具备可行性。
[Abstract]:Short-term load forecasting is an important part of energy management system. The magnitude of forecasting error directly affects the security and economy of power grid operation. With the large-scale interconnection of power grid and the reduction of data acquisition cost, the types of data available are becoming more and more abundant, and the trend of big data in smart grid is becoming more and more obvious. In view of the large power grid with a vast power supply area, the load characteristics of different regions vary from region to region. Influenced by external factors, centralized load forecasting method is difficult to accurately grasp the law of load change in various regions, and has limited ability to analyze mass data mining. In this paper, a distributed short-term load forecasting method based on big data is proposed to improve the accuracy of load forecasting by synthetically utilizing the advantages of big data and artificial intelligence. As a starting point, the research work carried out in this paper is as follows:. First of all, put forward the analysis method of load change law based on big data. Under the condition of cost permission, we should collect user and environment information to analyze load change law as far as possible for different types of electric power users. Based on the actual load data of A city and B city in a certain province of southern China, this paper analyzes the variation law of annual, monthly and daily load in two places, based on grasping the difference of load characteristics and meteorological conditions in different regions, and taking the actual load data of A city and B city in southern China as the basis, respectively. Based on the study of the typical daily load variation in summer, it is concluded that the load characteristics of different areas are different, and the influence of external factors is different, so it is necessary to establish a targeted load forecasting model in load forecasting work. Then, a distributed short-term load forecasting method based on big data is proposed. On the basis of administrative division and meteorological area distribution, the network structure of 220 kV and above is subdivided. The subnet classification and merging are realized by evaluating the similarity of the load curve of the subnet. The subnet load forecasting model Arima and BP neural network are established, and the factor vectors with high similarity and the load at the corresponding time are extracted as the input of the subnet load forecasting model. The load forecasting model of the whole network is established according to the load forecast value of the subnet and the forecast value of the ratio coefficient at each time, and the load forecasting results of the large power network are synthesized. Finally, based on the actual load data of a certain province in southern China, the method proposed in this paper is used to carry out short-term load forecasting on weekdays and rest days respectively, and the results are compared with those of centralized load forecasting. In order to verify the validity and stability of the load forecasting method proposed in this paper, the example analysis shows that the proposed method has good prediction accuracy in large power networks, and the time required for load forecasting is effectively reduced because of the parallel operation of each subnet. It is feasible in practical application.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM715

【相似文献】