基于改进型BP神经网络的光伏功率预测  

Photovoltaic Power Prediction Combined with Popular Learning and Improved BP Neural Network

作　　者：王云艳[1] 罗帅 王子健 WANG Yun-yan;LUO Shuai;WANG Zi-jian(College of Electrical and Electronic Engineering,Hubei University of Technology,Wuhan Hubei 430068,China)

机构地区：[1]湖北工业大学电气与电子工程学院,湖北武汉430068

出　　处：《计算机仿真》2022年第11期153-157,共5页Computer Simulation

基　　金：国家自然科学基金(41601394);湖北工业大学博士启动基金(BSQD2016010)。

摘　　要：光伏发电技术作为新时代主要能源供给手段被广泛应用,针对光伏发电数据采集不完整、筛选不合理、数据挖掘不全面等造成的预测精度低的问题,提出了一种结合流形学习算法和改进型BP神经网络算法。首先用灰度关联算法排除3项与光伏功率关联度最差的因素,然后结合流形学习网络将数据空间进行降维,最后采用模拟端对端结构的改进型BP神经网络来训练获得影响因子,保存的算法模型可以对未来几年的数据进行预测。实验结果表明,相比其它传统的算法,平均绝对百分比误差(MAPE)下降了10%左右,均方根误差(RMSE)下降了0.2kw左右,一定程度导航提高而光伏功率的预测精度。研究证明了算法具有较高的准确性、鲁棒性、模型具有较强的可迁移性。Photovoltaic power generation technology is widely used as the main means of energy supply in the new era.For photovoltaic power generation data collection is not complete,screening is not reasonable,such as incomplete data mining caused the problem of low accuracy,this paper proposes a combination of popular learning algorithm and improved BP neural network algorithm.First,the gray correlation algorithm was used to eliminate the three factors with the worst correlation with photovoltaic power,and then the manifold learning network was used to reduce the dimension of the data space.Finally,the improved BP neural network simulating the end-to-end structure was used to train to obtain the influence factors.The saved algorithm model can forecast data in the next few years.The experimental results show that compared with other traditional algorithms,the mean absolute percentage error(MAPE)is reduced by about 10%,and the root mean square error(RMSE)is reduced by about 0.2 kw,which improved the accuracy of photo voltaic power prediction to some extent.The experimental results show that this algorithm has high accuracy,robustness and strong portability of the model.

关 键 词：流行学习 光伏功率 神经网络 数据预测 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]