基于Spark和支持向量回归的微电网短期负荷预测研究

发布时间：2018-04-23 19:19

本文选题：短期负荷预测 + 微电网　；参考：《华北电力大学》2017年硕士论文

【摘要】：随着能源和环境问题的日趋严重,新能源的开发和利用受到越来越多的关注,随之带来了微电网的兴起和发展。为了保证微电网高效率的经济运行,准确地负荷预测就变得尤为重要。关于微电网短期负荷预测的研究还相对较少,因此,本文对微电网的负荷预测展开研究,对微电网安全、节能、高效运行具有重要的理论意义和实用价值。首先针对人工蜂群算法易陷入局部最优及收敛速度慢的缺点,引入当前最优蜜源和惯性权重函数,对该算法的食物源更新方式进行改进;然后针对支持向量回归机的参数选择和优化问题,转化成组合优化的问题,并使用改进的人工蜂群算法进行优化求解,进而得到人工蜂群算法优化SVR的预测模型。以微电网短期负荷预测数据为例,将该模型的预测结果与蚁群算法优化的支持向量回归机(ACO-SVR)、粒子群算法优化的支持向量回归机(PSO-SVR)和未改进的蜂群算法优化的支持向量回归机(ABC-SVR)进行对比分析,实验结果表明该模型具有最优的预测效果并且运行时间最短,相比其他模型具有更好的学习和推广能力。近年来电力系统智能化的发展导致了负荷数据的海量化和高维化,负荷预测面临着单机计算资源不足、预测实时性差的挑战。针对电力系统的海量历史负荷数据,对上面模型在Spark平台下进行了并行化设计。利用实验室设备搭建了含一个主节点、七个数据节点的Spark计算集群,在Spark云平台实现了分布式支持向量回归预测算法。测试了本文设计的并行化算法的并行性能,实验分析表明本文设计的并行算法在进行海量数据处理时相对于常用的方法具有更快的处理速度。
[Abstract]:With the increasingly serious problems of energy and environment, more and more attention has been paid to the development and utilization of new energy, which has brought about the rise and development of microgrid. In order to ensure high efficiency of economic operation of microgrid, accurate load forecasting becomes more and more important. The research on short-term load forecasting of micro-grid is relatively few. Therefore, the research on micro-grid load forecasting in this paper has important theoretical significance and practical value for micro-grid safety, energy saving and efficient operation. Firstly, aiming at the shortcomings of artificial bee colony algorithm, which is easy to fall into local optimization and slow convergence speed, the current optimal honey source and inertial weight function are introduced to improve the food source updating method of the algorithm. Then, the problem of parameter selection and optimization of support vector regression machine is transformed into a combinatorial optimization problem, and the improved artificial bee colony algorithm is used to solve the optimization problem, and then the prediction model of artificial bee colony algorithm to optimize SVR is obtained. Taking the short-term load forecasting data of microgrid as an example, The prediction results of the model are compared with the support vector regression machine (ACO-SVRV) optimized by ant colony algorithm, the support vector regression machine (PvR) optimized by particle swarm optimization (PSO) and the support vector regression machine (ABC-SVR) optimized by the unimproved bee colony algorithm. The experimental results show that the model has the best prediction effect and the shortest running time, and it has better learning and generalization ability than other models. In recent years, the development of intelligent power system has led to the sea quantization and high dimensional load data. Load forecasting is faced with the challenge of insufficient computing resources and poor real-time forecasting. Aiming at the massive historical load data of power system, parallel design of the above model is carried out on Spark platform. A Spark computing cluster with one master node and seven data nodes is constructed by using laboratory equipment. The distributed support vector regression prediction algorithm is implemented on the Spark cloud platform. The parallel performance of the parallel algorithm designed in this paper is tested. The experimental analysis shows that the parallel algorithm designed in this paper has a faster processing speed than the common methods in mass data processing.
【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】：TP18;TM715

【参考文献】