【作者】 姜强鑫；

【导师】 杨胡萍；

【作者基本信息】 南昌大学 ， 电力系统及其自动化， 2012， 硕士

【摘要】 光伏出力预测是电网规划的重要组成部分,其预测精度关系到电力系统的安全、稳定及运行。同时对电力系统调度、电力市场营销以及发电公司竞价上网都具有很大的影响。由于光伏阵列输出功率受环境影响很大,且具有随机性、波动性和间歇性,因此建立合适的预测模型,提高光伏电站出力预测的精度是本文的主要研究内容。在分析现有的BP神经网络、小波神经网络预测方法的基础上,本文提出了基于改进小波神经网络算法的预测模型,该模型结合小波神经网络的优良性能同时采用Levenberg-Marquardt算法进行数值优化。在训练样本选取上,采用数据挖掘技术,应用改进余弦相似性度量方法和欧式距离相结合的方法,对样本数据库进行分类。应用BP神经网络、小波神经网络及改进小波神经网络三种模型对晴天、阴天、多云转阴、中雨四个样本集分别进行了预测并与实际值进行分析比较,通过相对误差和平均相对误差两个指标来评价本文所用的预测模型。结果显示采用改进小波神经网络算法的预测模型对晴天的预测效果比对阴天、多云转阴、中雨的精度都要高,对于阴天、多云转阴和中雨样本采用改进小波神经网络算法的预测模型能大致的反应光伏出力的变化趋势,对于光伏发电系统配合电力系统制定发电计划仍具有较高的参考价值,证明本文基于改进小波神经网络光伏出力预测模型的正确性及有效性。同时,本文把自抗扰控制技术应用于光伏的DC-DC的控制中,通过仿真结果显示,自抗扰控制技术能提高光伏在外部条件变化时的输出特性,有利于光伏的最大功率输出和提高光伏的发电效率。更多还原

【Abstract】 PV output prediction is an important part of the planning of the grid. The PV array output power is affected by the environment. The output power has randomness, Intermittent and instability. The forecast accuracy of PV output power that related to the power system security, stability, operation. It have a great impact to network planning, power system scheduling, power marketing and power generation companies bidding. Therefore, improve PV output prediction accuracy is the main contents of this thesis.On the basis of the analysis of existing BP neural network, wavelet neural network prediction method, this paper proposed a prediction model that based on improved wavelet neural network algorithm, the model combines the excellent performance of the wavelet neural network and Levenberg-Marquardt algorithm for numerical optimization. On the selection of training samples, this paper using data mining techniques, the use of improved combination of similarity and Euclidean distance to classify the sample database. BP neural network, wavelet neural network and improved wavelet neural network of the three models on the sunny, cloudy, cloudy to overcast, moderate set of four samples were carried out to predict and the actual value of the analysis and comparison, by the relative error and average relative error indicators to assess the predictive models used in this article. Results show that the improved algorithm of wavelet neural network prediction model prediction of sunny to cloudy, cloudy to overcast, moderate rain, the accuracy should be high, cloudy, cloudy to overcast, moderate rain samples using improved wavelet neural network algorithm model can predict the approximate reaction PV output trends, still has high reference value for the photovoltaic power generation system with the power system to develop power generation plan, to prove the correctness and effective of this thesis based on improved wavelet neural network PV output prediction model. The same time, this thesis use ADRC control technology that used to control in photovoltaic DC-DC. Displayed by the simulation results show that ADRC control technology can improve the output characteristics of PV in the external conditions change, PV maximum power output and improve the efficiency of the photovoltaic power generation.更多还原