| 摘要: |
| 为了丰富贵州西部光伏电站的功率预测思路与方法，利用2019—2021年贵州西部5个光伏电站发电功率、气象要素、卫星反演辐照度、地面观测辐照度数据，分析光伏功率的时间演变和与卫星反演辐照度、地面观测辐照度的相关性。利用前80%样本数据为测试集，剩余数据为预测模拟模型的检验集，利用机器学习中的BP、GRNN神经网络算法和测试集分别对5个电站建立光伏功率预测模拟模型。利用检验集和光伏电站的检验方法对各个模型进行效果验证，并对比不同算法不同站点间的预测模拟效果。结果表明，BP、GRNN算法在5个光伏电站的功率预测模拟中平均日准确率在90%左右，标准化均方误差在0.07~0.12，且FY-4A反演辐照度参与建立的光伏功率预测模拟模型较地面观测辐照度参与建立的模型效果更佳，能够为光伏功率预测提供一种参考方案和思路。 |
| 关键词: 光伏功率；机器学习；神经网络 |
| DOI： |
| 投稿时间：2023-01-09 |
| 基金项目:贵州省气象局科研业务登记项目（黔气科登〔2022〕10-03号）：基于机器学习的贵州西部光伏功率预测方法研究 |
|
| Research on Simulation Method for Photovoltaic Power Prediction Based on Machine Learning |
| HE Dongpo,WAN Chao,YANG Fan,CHEN Yang,PENG Yuxiang |
| (Guizhou Meteorological Observatory, Guiyang 550002 ,China;Qiannan Bouyei and Miao Autonomous Prefecture Meteorological Bureau of Guizhou,Duyun 558000 ,China;Guizhou Weather Modification Office,Guiyang 550081 ,China) |
| Abstract: |
| In order to enrich the power prediction ideas and methods of photovoltaic power stations in western Guizhou, this paper collected the data of power generation, meteorological elements, satellite retrieved irradiance and ground observed irradiance of five photovoltaic stations in western Guizhou from 2019 to 2021, and analyzed the time evolution of photovoltaic power and its correlation with satellite retrieved irradiance and ground observed irradiance. The first 80% sample data is used as the test set, and the remaining data is the test set of the Predictive simulation model. Using BP, GRNN neural network algorithm and test set in machine learning, photovoltaic power Predictive simulation model are established for five power stations respectively. Use the test set and the test method of photovoltaic power station to verify the effect of each model, and compare the prediction effect between different algorithms and different sites. The results show that the average daily accuracy of BP and GRNN algorithms in power prediction simulation of five photovoltaic power stations is about 90%, and the standardized mean square error is 0.07~0.12. Moreover, the PV power prediction simulation model established with FY-4A inversion irradiance is better than the model established with ground observation irradiance, which can provide a reference scheme and idea for PV power prediction. |
| Key words: photovoltaic power； machine learning； neural network |
