基于神经网络的光伏功率短期预测方法的研究

发布时间：2018-05-31 12:57

本文选题：光伏功率预测 + 灰色关联分析　；参考：《东北电力大学》2017年硕士论文

【摘要】：太阳能发电受到日照,季节变化及天气波动等气候条件的影响,使得发电系统的输出功率具有间断性、周期性和不确定性的特点。光伏发电功率预测技术作为光伏电站建站必备的技术条件,关系到电站并网及电网调度的准确性和合理性,若能准确掌握短期内光伏电站的输出功率,可大大降低电站并网风险,提高电网运行的安全性和稳定性。本文在对气象因素如何影响光伏发电功率输出以及功率预测技术基础进行了简要介绍之后,将光伏发电功率短期预测问题分为超短期和短期两部分。针对超短期输出功率的预测,提出一种基于气象因素的相似日选取方法:利用光伏发电系统的历史气象信息建立气象特征向量,通过计算灰色关联度寻找到预测日的相似历史日。然后使用相似日历史数据和小波神经网络(WNN,Wavelet Neural Network)构建一种光伏发电功率的超短期预测模型,通过使用某光伏发电系统的历史数据进行建模,对所选两类不同天气类型的预测日的出力情况进行逐时刻预测,预测结果显示模型预测效果较好,尤其对于理想晴天条件下预测的更加精确。针对短期输出功率的预测,提出一种基于思维进化算法(MEA,Mind Evolutionary Algorithm)优化BP神经网络的光伏功率短期预测模型,通过我国青海省锡铁山装机量100MW的光伏电站为期一年的历史运行数据进行建模,按照季节划分为四个预测单元分别对预测模型进行训练和电站出力预测,通过与电站的实际出力情况和电站所配备的预测系统短期预测值的比较分析,由BP算法和MEA-BP算法所构建的模型均达到了一定的预测精度,其中MEA-BP模型有效的降低了BP网络模型的预测误差。最后将相似日与神经网络结合,建立一个基于相似日和神经网络的光伏功率短期预测模型:通过设置两组对照实验:一组使用相似历史日的数据来训练网络并进行预测(实验组),一组使用相邻历史日数据来训练网络并进行预测(对照组),对比实验的结果显示实验组的预测效果更为准确。经过反复预测实验,验证了课题所提出的预测模型能够对有效的预测光伏发电系统的输出功率,预测结果也表明基于神经网络的预测模型在一定程度上能够满足实际应用需求,在光伏电站建站时对功率预测技术的设计具有一定的参考价值。
[Abstract]:Solar power generation is affected by the weather conditions such as sunshine, seasonal variation and weather fluctuation, which makes the output power of the power generation system have the characteristics of discontinuity, periodicity and uncertainty. As a necessary technical condition for the construction of photovoltaic power station, PV generation power prediction technology is related to the accuracy and rationality of grid connection and grid dispatching. If the output power of photovoltaic power station can be accurately grasped in the short term, It can greatly reduce the risk of grid connection and improve the security and stability of power grid operation. After a brief introduction of how meteorological factors affect the output of photovoltaic power generation and the technical basis of power prediction, the short-term prediction of photovoltaic power generation is divided into two parts: ultra short term and short term. According to the prediction of ultra-short-term output power, a similar day selection method based on meteorological factors is proposed: the meteorological characteristic vector is established by using the historical meteorological information of photovoltaic power generation system. The similar historical days of the predicted days are found by calculating the grey correlation degree. Then, using the similar daily historical data and wavelet neural network (WNNN) to construct an ultra-short-term prediction model of photovoltaic power generation, using the historical data of a photovoltaic power generation system to model. The prediction results show that the forecasting effect of the model is better, especially for the ideal sunny weather conditions. In order to predict the short-term output power, a short-term PV power prediction model based on the thinking evolutionary algorithm (MEA ind Evolutionary algorithm) is proposed to optimize the BP neural network. Based on the historical operation data of 100MW photovoltaic power station in Xitieshan of Qinghai Province for one year, the forecasting model is divided into four forecasting units according to the seasons, and the forecasting model is trained and the power generation of the power station is predicted. By comparing with the actual force of the power station and the short-term prediction value of the forecasting system equipped with the power station, the model constructed by BP algorithm and MEA-BP algorithm has achieved a certain prediction accuracy. MEA-BP model can effectively reduce the prediction error of BP network model. Finally, the similar days are combined with neural networks. Establish a short-term photovoltaic power prediction model based on similar days and neural networks: train and predict the network by setting up two groups of controlled experiments: one group uses data from similar historical days to train the network and make prediction (experimental group, group using adjacent data) Historical daily data to train the network and predict (control group, the results of comparative experiments show that the experimental group is more accurate prediction results. After repeated prediction experiments, it is verified that the proposed prediction model can effectively predict the output power of photovoltaic power generation system. The prediction results also show that the prediction model based on neural network can meet the practical application needs to a certain extent. It has certain reference value for the design of power prediction technology in the construction of photovoltaic power station.
【学位授予单位】：东北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;TM615

【参考文献】