基于云计算和机器学习算法的微电网负荷预测

发布时间：2018-05-06 09:25

本文选题：负荷预测 + 微电网　；参考：《华北电力大学》2017年硕士论文

【摘要】：随着能源问题和环境污染问题的日益严峻,综合开发与合理利用新能源势在必行,而微电网的建设可以充分消纳新能源并且优化能源结构。准确地进行负荷预测不仅可以为微电网优化运行和能量管理决策提供重要依据,还可以保证微电网高效率的经济运行。因此,本文针对微电网的短期负荷预测问题展开研究,对微电网系统优化运行具有重要的理论意义和实用价值。本文首先对微电网负荷预测的特点及其影响因素进行了分析,采用改进的混合蛙跳算法对核函数极限学习机的组合参数进行优化(ISFLA_KELM),同时引入Spark on YARN平台,将算法进行并行化改进,在确保预测精度的同时通过并行计算来应对大数据带来的挑战,并采用某微电网真实负荷数据验证预测准确度以及执行效率。本文主要进行以下几个方面的工作。(1)分析了微电网负荷预测面临的问题,并研究了不同预测方法的优缺点。针对微电网负荷预测的影响因素及特点,选择适合的智能优化算法——混合蛙跳算法,并对其存在的缺点进行针对性改进,给出了改进的混合蛙跳算法。(2)研究分析了混合蛙跳优化算法的原理其特点,确定了其相对其他优化算法的优势。并将改进后的混合蛙跳优化算法与核函数极限学习机结合,给出一种新型的微电网负荷预测算法(ISFLA_KELM)。将核函数极限学习机的组合参数作为蛙群优化算法的青蛙个体进行优化。(3)给出了基于Spark的ISFLA_KELM微电网负荷预测算法,针对电力大数据下单机计算资源不足的问题,分别对KELM中的耗时运算和ISFLA算法进行并行化设计,并结合Spark机器学习库以及分布式文件系统,提高算法的执行效率。(4)进行实验测试与算例分析。选用UCI标准数据集提供的真实负荷数据集,在实验室搭建了8个节点的Spark on yarn内存计算平台,然后对提出的算法进行性能测试,并将其与现有的负荷预测方法进行对比。实验结果表明提出算法的负荷预测精度均优于已有算法,且具有较好的并行性能,可为微电网负荷预测提供有效依据。
[Abstract]:With the increasingly serious problem of energy and environmental pollution, it is imperative to develop and utilize new energy rationally, and the construction of micro-grid can fully absorb new energy and optimize the energy structure. Accurate load forecasting can not only provide an important basis for optimal operation and energy management decision of microgrid, but also ensure the economic operation of micro-grid with high efficiency. Therefore, this paper focuses on the short-term load forecasting of microgrid, which has important theoretical significance and practical value for the optimal operation of micro-grid system. In this paper, the characteristics of load forecasting in microgrid and its influencing factors are analyzed, and the improved hybrid leapfrog algorithm is used to optimize the combined parameters of the kernel function extreme learning machine. At the same time, the Spark on YARN platform is introduced to optimize the combined parameters of the kernel function extreme learning machine. The algorithm is parallelized and improved to meet the challenge brought by big data by parallel computation while ensuring the prediction accuracy. The forecasting accuracy and execution efficiency are verified by the real load data of a microgrid. The main work of this paper is as follows: 1) the problems of load forecasting in microgrid are analyzed, and the advantages and disadvantages of different forecasting methods are studied. According to the influence factors and characteristics of load forecasting in microgrid, a suitable intelligent optimization algorithm, hybrid leapfrog algorithm, is selected, and its shortcomings are improved. This paper presents an improved hybrid leapfrog algorithm. The principle and characteristics of the hybrid leapfrog optimization algorithm are analyzed and its advantages compared with other optimization algorithms are determined. By combining the improved hybrid leapfrog optimization algorithm with the kernel function extreme learning machine, a new micro-grid load forecasting algorithm is presented. This paper presents a load forecasting algorithm of ISFLA_KELM microgrid based on Spark, which takes the combination parameter of kernel function extreme learning machine as the frog individual of frog swarm optimization algorithm. It aims at the problem of insufficient computing resources in single machine under power big data. The time-consuming operation and the ISFLA algorithm in KELM are designed in parallel, and the experimental test and the example analysis are carried out by combining the Spark machine learning library and the distributed file system to improve the execution efficiency of the algorithm. Using the real load data set provided by the UCI standard data set, an 8-node Spark on yarn memory computing platform is built in the laboratory, and then the proposed algorithm is tested and compared with the existing load forecasting methods. The experimental results show that the proposed algorithm has better load forecasting accuracy than the existing algorithms and has better parallel performance, which can provide an effective basis for load forecasting of micro-grid.
【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】：TM715

【参考文献】