基于云计算与机器学习的短期光伏发电功率预测研究
发布时间：2018-08-15 11:28
【摘要】：随着全球化地能源应用变革,可再生能源在全球能源结构中的比例迅速增大。光伏发电作为一种高效、清洁能源,正成为可再生能源发电中的新增长点。近年来,全球光伏产业市场强劲增长,各国新增装机容量快速提高;但光伏发电功率受太阳辐照度、温度、湿度等气象因素影响较大,具有间歇性、波动性、周期性特点。大规模光伏接入会拉大电网峰谷差距,造成调峰困难,影响电能质量和电网的安全稳定运行。因此,结合历史数据与未来气象数据有效预测光伏输出功率,帮助调度人员合理的规划电网调度,管理运行,对于电力系统的安全稳定运行具有非常重要的意义。本文选择短期(一天)光伏发电功率预测为主要研究内容。由于影响光伏发电的因素较为复杂,利用Pearson相关系数与Spearman秩相关系数对其进行分析,设计基于功率的相似日聚类方法。之后,为提高光伏发电功率预测的精度,针对所得到的各聚类簇,提出了一种基于自适应烟花算法(Adaptive FireWorks Algorithm,AFWA)优化径向基函数神经网络(Radial Bais Function Neural Network,RBFNN)的预测模型,利用自适应烟花算法的种群协同搜索优势,优化网络参数进而实现更加精确地光伏出力预测。同时,光伏电站在长期运行后积累了大量的历史数据,随着电站的运行,数据量也将越来越大,单机环境下使用大量历史数据计算耗时较长,影响电网的快速调度。本文搭建基于内存的Spark云计算平台,并对所提算法进行并行化改进实现。在Spark平台上运行算法,提高计算效率。在单机和多节点Spark云平台下分别与传统单一RBFNN及粒子群算法(PSO)优化RBFNN对比实验,验证所提算法提高了预测精度,且算法并行化后大大缩短了计算时间。
[Abstract]:With the transformation of global energy application, the proportion of renewable energy in the global energy structure is increasing rapidly. Photovoltaic power generation, as a kind of efficient and clean energy, is becoming a new growth point in renewable energy generation. In recent years, the global photovoltaic industry market has grown strongly, and the installed capacity of various countries has been increased rapidly. However, the photovoltaic power generation power is greatly affected by meteorological factors such as solar irradiance, temperature, humidity and so on, and has the characteristics of intermittent, volatility and periodicity. Large-scale photovoltaic access will widen the gap between peak and valley of power grid, cause difficulty of peak shaving, and affect the power quality and the safe and stable operation of power grid. Therefore, combining historical data with future meteorological data to effectively predict photovoltaic output power, help dispatchers to plan power grid dispatching reasonably, management operation, for the safe and stable operation of the power system has a very important significance. In this paper, short-term (one-day) photovoltaic power prediction is chosen as the main research content. Because the factors affecting photovoltaic power generation are complex, the similar day clustering method based on power is designed by using Pearson correlation coefficient and Spearman rank correlation coefficient. Then, in order to improve the accuracy of photovoltaic power prediction, a prediction model based on adaptive fireworks algorithm (Adaptive FireWorks algorithm) is proposed to optimize the radial basis function neural network (Radial Bais Function Neural). In order to achieve more accurate photovoltaic force prediction, the network parameters are optimized by using the population cooperative search advantage of adaptive fireworks algorithm. At the same time, photovoltaic power station has accumulated a large amount of historical data after long-term operation. With the operation of the power plant, the amount of data will be more and more large, and it will take a long time to use a large number of historical data in a single machine environment, which will affect the rapid dispatch of power grid. In this paper, Spark cloud computing platform based on memory is built, and the proposed algorithm is implemented in parallel. The algorithm is run on the Spark platform to improve the computational efficiency. Compared with the traditional single RBFNN and Particle Swarm Optimization (PSO) optimization RBFNN on single and multi-node Spark cloud platform, the proposed algorithm improves the prediction accuracy and greatly reduces the computation time after parallelization.
【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】：TM615;TP393.09;TP18
[Abstract]:With the transformation of global energy application, the proportion of renewable energy in the global energy structure is increasing rapidly. Photovoltaic power generation, as a kind of efficient and clean energy, is becoming a new growth point in renewable energy generation. In recent years, the global photovoltaic industry market has grown strongly, and the installed capacity of various countries has been increased rapidly. However, the photovoltaic power generation power is greatly affected by meteorological factors such as solar irradiance, temperature, humidity and so on, and has the characteristics of intermittent, volatility and periodicity. Large-scale photovoltaic access will widen the gap between peak and valley of power grid, cause difficulty of peak shaving, and affect the power quality and the safe and stable operation of power grid. Therefore, combining historical data with future meteorological data to effectively predict photovoltaic output power, help dispatchers to plan power grid dispatching reasonably, management operation, for the safe and stable operation of the power system has a very important significance. In this paper, short-term (one-day) photovoltaic power prediction is chosen as the main research content. Because the factors affecting photovoltaic power generation are complex, the similar day clustering method based on power is designed by using Pearson correlation coefficient and Spearman rank correlation coefficient. Then, in order to improve the accuracy of photovoltaic power prediction, a prediction model based on adaptive fireworks algorithm (Adaptive FireWorks algorithm) is proposed to optimize the radial basis function neural network (Radial Bais Function Neural). In order to achieve more accurate photovoltaic force prediction, the network parameters are optimized by using the population cooperative search advantage of adaptive fireworks algorithm. At the same time, photovoltaic power station has accumulated a large amount of historical data after long-term operation. With the operation of the power plant, the amount of data will be more and more large, and it will take a long time to use a large number of historical data in a single machine environment, which will affect the rapid dispatch of power grid. In this paper, Spark cloud computing platform based on memory is built, and the proposed algorithm is implemented in parallel. The algorithm is run on the Spark platform to improve the computational efficiency. Compared with the traditional single RBFNN and Particle Swarm Optimization (PSO) optimization RBFNN on single and multi-node Spark cloud platform, the proposed algorithm improves the prediction accuracy and greatly reduces the computation time after parallelization.
【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】：TM615;TP393.09;TP18
【参考文献】
相关期刊论文 前10条
1 刘泽q，
本文编号：2184064
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2184064.html
