光伏电站短期功率预测方法研究

发布时间：2018-01-19 05:07

本文关键词： 光伏发电短期功率预测超短期功率预测相关系数天气类型　出处：《江苏大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,随着相关科技的进步,光伏发电的单位造价持续下降,因而得到了快速发展。太阳能作为一种清洁可再生能源,利用太阳能进行光伏发电能够大幅缓解能源危机,减轻由传统化石燃料带来的一系列环境问题。然而,地表太阳能属于间歇性能源,使得光伏电站的发电功率呈现出波动性和间歇性特征。当并网光伏电站装机容量较大时,电力系统的安全性与稳定性将受到影响。因此,需要准确预测光伏电站的发电功率以配合电力部门进行合理的计划和调度。本文依托两座并网光伏电站的实际采集数据,分析总结了近年来国内外相关领域的研究进展,对光伏电站短期和超短期功率预测进行了较为详细的分析研究,论文主要包括以下几方面内容:(1)根据相关理论设计了一种地外辐照度计算器,将获取到的原始数据通过异常值监测、有效时间区间判定、插补缺失数据和归一化等步骤进行数据预处理后,建立了用于光伏发电功率预测的数据库;(2)提出了一种基于ELM-SVM的短期功率预测模型。首先,根据天气预报给出的不同天气类型中地外辐照度与发电功率间的相关系数,将各天气类型合并成晴天、多云、雨天三种典型天气类型,并分别建立子预测模型。之后,利用皮尔逊相关系数根据各典型天气类型的特征,选取针对性较强的参数作为子预测模型的输入。最后利用“积分竞争制”回归模型选取法,选取ELM作为晴天条件下的回归模型,SVM作为多云和雨天条件下的回归模型。结果表明ELM-SVM混合预测模型能够发挥不同回归模型的优势,相比使用单一模型预测方法,该混合预测模型具有更强的适应能力和更好的预测效果;(3)使用历史发电功率作为模型输入,本文提出了基于ELM的超短期功率预测模型。相比BP神经网络,ELM具有更好的预测效果。最后,根据ELM模型在不同时间区间内的误差分布特征,将历史发电功率分时段训练并建立子预测模型,实验结果表明,基于ELM的分段式功率预测模型在天气波动较大的环境中表现更佳。
[Abstract]:In recent years, with the progress of related science and technology, the unit cost of photovoltaic power has been continuously reduced, so it has been rapidly developed. Solar energy as a clean and renewable energy. Solar photovoltaic power generation can significantly alleviate the energy crisis and alleviate a series of environmental problems caused by traditional fossil fuels. However, surface solar energy is an intermittent energy source. When the installed capacity of grid-connected photovoltaic power station is large, the security and stability of power system will be affected. It is necessary to accurately predict the generation power of photovoltaic power station in order to cooperate with the power department to plan and dispatch reasonably. This paper relies on the actual acquisition data of two grid-connected photovoltaic power stations. This paper analyzes and summarizes the research progress in the related fields at home and abroad in recent years, and makes a more detailed analysis and research on the short-term and ultra-short-term power prediction of photovoltaic power plants. This paper mainly includes the following aspects: 1) according to the relevant theory, a kind of extraterrestrial irradiance calculator is designed. The original data is monitored by outliers and the effective time interval is determined. After preprocessing the missing data and normalized data, the database of photovoltaic power prediction is established. In this paper, a short-term power prediction model based on ELM-SVM is proposed. Firstly, the correlation coefficient between external irradiance and generation power in different weather types is given. The weather types are combined into three typical weather types: sunny, cloudy and rainy, and sub-prediction models are established respectively. After that, Pearson correlation coefficient is used according to the characteristics of each typical weather type. The parameters are selected as the input of the sub-prediction model. Finally, the "integral competition system" regression model selection method is used to select ELM as the regression model under sunny conditions. SVM is a regression model under cloudy and rainy conditions. The results show that the ELM-SVM hybrid prediction model can play the advantages of different regression models, compared with the single model prediction method. The hybrid prediction model has stronger adaptability and better prediction effect. Using the historical generation power as the input of the model, this paper presents an ultra-short-term power prediction model based on ELM. It has better prediction effect than BP neural network. Finally. According to the error distribution characteristics of the ELM model in different time intervals, the historical generation power is trained in different periods and the sub-prediction model is established. The experimental results show that. The segmented power prediction model based on ELM performs better in the fluctuating weather environment.
【学位授予单位】：江苏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM615

【参考文献】