基于大数据的居民用电行为分析与负荷预测

发布时间：2018-04-02 20:00

本文选题：居民用电行为　切入点：负荷预测　出处：《华北电力大学》2017年硕士论文

【摘要】：随着互联网、物联网、无线传感器等新一代技术的发展,加快了智能电网建设的速度。在智能电网的建设过程中,先进的计量设备、智能终端设备被大量安装与使用,居民的用电方式也趋于多元化。深入感知居民的实际功耗模式对提高负荷预测的精度、保障电力系统的正常运行、能量管理和规划至关重要。本文首先分析居民用电大数据的来源,针对居民用电大数据体量大、类型复杂、速度快与交互性强等特点,指出居民用电大数据在存储、处理等操作所面临的挑战。然后提出了一种基于大数据的居民用电行为分析与负荷预测模型,该模型将智能电表、气象、节假日等数据作为输入,并使用基于内存计算的大数据处理框架Spark对居民用电大数据进行挖掘与分析。最后设计并开发基于大数据的居民用电行为分析与负荷预测原型系统,系统包括Spark集群管理、负荷数据管理、算法分析、预测结果展示等模块。将基于Spark的K-Means聚类算法应用于居民用户的用电模式聚类实验,实验结果表明对居民用户分类具有较高的正确率。并将其运行效果与传统K-Means算法运行效果进行对比,实验结果表明随着数据集规模的不断扩大,基于Spark的K-Means算法表现出良好的性能,减少了聚类执行的时间和提高了聚类的准确性。并针对不同类别的居民用户进行用电行为的分析。基于上述实验,针对每类居民用户建立负荷预测模型,分别使用基于Spark的多层感知器神经网络算法(MLP-NN)与基于Spark的SVM算法实现每类居民用户的负荷预测,实验结果表明MLP-NN具有较高的预测精度。使用两个不同的数据集验证该模型的可行性,并扩大两种数据集的规模,从每类用户的智能电表数据、气象数据和节假日数据中抽取20个特征向量作为算法的输入层数据,实验结果表明该预测方法在一定程度上提高了负荷预测的精度,大数据环境下表现出更好的预测效果。
[Abstract]:With the Internet, the Internet of things, the development of a new generation of wireless sensor technology, accelerate the construction of smart grid speed. In the process of construction of smart grid, advanced metering equipment, intelligent terminal equipment by a large number of installation and use, residents of the electricity also tends to be diversified. The actual power consumption mode of residents' perception of depth to improve the accuracy of load forecasting, guarantee the normal operation of the power system, vital energy management and planning. This paper first analyzes the source of large data by residents, for residents with large data volume, complex types, characteristics of high speed and strong interaction, pointed out that the residential electricity data in the storage, processing and other operations of the challenges faced by the and then an electrical behavior analysis and load forecasting model for large data based on the model of residents, smart meters, weather, holidays and other data as input, and use the base Big data in memory computing processing framework using Spark data mining and analysis of large residents. Finally the design and analysis of behavior and electric load forecasting prototype system with large data systems including Spark residents based on development, cluster management, load data management, algorithm analysis, the prediction results display module. The application based on K-Means clustering algorithm Spark to the residents' electricity pattern clustering experiments, experimental results show that the correct rate is higher for residential users classification. And compared its effect with the traditional K-Means algorithm running effect, the experimental results show that with the continuous expansion of the scale of data set, the K-Means Spark algorithm shows good performance on reducing clustering of execution time and improves the accuracy of clustering. And analyze the consumption behavior of the residents according to different categories of users. Based on the above experiments, for each Resident users to establish the forecasting model, using multilayer perceptron neural network algorithm based on Spark (MLP-NN) and SVM Spark algorithm for each type of load forecasting based on residential users, the experimental results show that MLP-NN has higher prediction accuracy. Using two different data sets to verify the feasibility of the model, and expand the two kinds of data from the data set size, the smart meter of each user, extraction of meteorological data and data in the 20 holiday feature vector as the input data of the algorithm, the experimental results show that the prediction method can improve the accuracy of load forecasting in a certain extent, performance under the big data environment a better prediction effect.

【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13;TM715

【参考文献】