云计算和机器学习算法在电力负荷预测中的研究与应用

发布时间：2018-09-04 14:55

【摘要】：短期负荷预测是实现电力系统优化运行的基础，对于电力系统的安全性、可靠性和经济性都有着显著影响。随着电网智能化程度的加深，一线城市在用电高峰期间，面临数百万条记录的电力数据采集规模，一年的数据存储规模将从目前的GB级增长到TB级，甚至PB级，同时，电力负荷数据维度也从几十向上百过渡。在如此海量高维的数据上进行传统负荷预测，将遭遇单机计算资源不足的瓶颈。云计算技术对于海量高维数据的处理和实时求解具有很强的适应性，可为消耗大量资源的算法提供实时可靠相对廉价的计算资源。智能电网的云计算存储模型虽已经取得一定的发展，但是基于云计算的电力负荷预测的并行算法却鲜有人研究。在此背景下，本文对上述问题展开研究。首先，为了提高负荷分类的精确性和有效性，为电力负荷预测的数据预处理阶段提供有效参考，本文提出一种基于云计算的并行量子粒子群优化模糊C均值聚类算法。将量子粒子群群体智能算法（QPSO）引入到传统模糊C均值聚类算法（FCM）中，利用QPSO较强的全局搜索能力，克服传统FCM算法易陷入局部最优以及其对初始聚类中心过于敏感的缺陷。其次，针对传统支持向量机负荷预测算法执行效率较低的不足，提出将序列极小优化算法引入到电力负荷支持向量机预测算法（ε SVR）中，实现对ε SVR算法的快速训练。此外，针对电力负荷预测的实际应用场景，还提出一种在线序列优化的极限学习机的短期电力负荷预测模型。针对电力负荷数据特性，将极限学习机算法引入到负荷预测领域，并对该算法进行在线序列优化；引入分布式和multi-agent思想，提升负荷预测算法预测准确率。最后，采用云计算技术中的MapReduce编程框架以及HBase分布式数据库等技术，对提出的三种改进算法进行并行化改进，并设计相应的Map和Reduce函数，以提高其处理海量高维数据的能力。最后，进行实验测试与算例分析。选用数据集UCI标准测试数据集与欧洲智能技术网络（European Network on Intelligent Technologies，简称EUNITE）提供的真实短期负荷数据，在实验室搭建的9节点以上的云计算集群上对提出的算法进行性能测试，并与传统负荷预测算法进行对比实验。实验结果表明提出算法的负荷预测精度均优于传统算法，可为电力负荷预测提供有效依据，，且具有较好的并行性能。
[Abstract]:Short-term load forecasting is the basis for the optimal operation of power system, which has a significant impact on the security, reliability and economy of power system. With the deepening of the intelligence of the power grid, during the peak period of power consumption, the first-tier cities will face the power data acquisition scale of millions of records. The scale of data storage in one year will grow from the current GB level to the TB level, or even the PB level, and at the same time, Power load data dimension also from dozens to hundreds of transition. Traditional load forecasting based on such a large amount of high dimensional data will meet the bottleneck of lack of single computing resources. Cloud computing technology has a strong adaptability to the processing and real-time solution of massive high-dimensional data, and can provide real-time, reliable and relatively cheap computing resources for algorithms that consume a large amount of resources. Although the cloud computing storage model of smart grid has made some progress, the parallel algorithm of power load forecasting based on cloud computing is seldom studied. In this context, the above problems are studied in this paper. Firstly, in order to improve the accuracy and effectiveness of load classification and provide an effective reference for the data preprocessing stage of power load forecasting, a parallel quantum particle swarm optimization fuzzy C-means clustering algorithm based on cloud computing is proposed in this paper. The Quantum Particle Swarm Optimization (QPSO) intelligent algorithm (QPSO) is introduced into the traditional fuzzy C-means clustering algorithm (FCM). By using the strong global search ability of QPSO, the traditional FCM algorithm is easy to fall into local optimum and is too sensitive to the initial clustering center. Secondly, aiming at the low efficiency of the traditional support vector machine (SVM) load forecasting algorithm, this paper introduces the sequence minimization algorithm into the power load support vector machine forecasting algorithm (蔚 SVR), and realizes the fast training of the 蔚 SVR algorithm. In addition, in view of the practical application of power load forecasting, a short-term power load forecasting model based on on-line sequential optimization is proposed. According to the characteristics of power load data, this paper introduces the extreme learning machine algorithm into the field of load forecasting, and optimizes the on-line sequence of the algorithm, and introduces the idea of distributed and multi-agent to improve the forecasting accuracy of load forecasting algorithm. Finally, using the MapReduce programming framework of cloud computing technology and HBase distributed database technology, the three improved algorithms are improved in parallel, and the corresponding Map and Reduce functions are designed to improve their ability to deal with massive high-dimensional data. Finally, the experimental test and example analysis are carried out. Using the real short-term load data provided by UCI standard test data set and European Intelligent Technology Network (EUNITE), the proposed algorithm is tested on the cloud computing cluster with 9 nodes or more built in the laboratory. And compared with the traditional load forecasting algorithm. The experimental results show that the accuracy of the proposed algorithm is better than that of the traditional algorithm, which can provide an effective basis for power load forecasting and has good parallel performance.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM715;TP181

【相似文献】