基于考虑误差修正的非线性自适应权重组合模型的光伏发电功率预测  

Photovoltaic Power Prediction Based on Nonlinear Adaptive Weight Combination Model Considering Error Correction

作　　者：陈德余 张玮[1] 王辉[2] CHEN Deyu;ZHANG Wei;WANG Hui(School of Information and Automation Engineering,Qilu University of Technology(Shandong Academy of Sciences),Jinan 250353,Shandong,China;School of Electrical Engineering,Shandong University,Jinan 250061,Shandong,China)

机构地区：[1]齐鲁工业大学(山东省科学院)信息与自动化学院,山东济南250353 [2]山东大学电气工程学院,山东济南250061

出　　处：《济南大学学报（自然科学版）》2024年第2期250-256,共7页Journal of University of Jinan(Science and Technology)

基　　金：国家重点研发计划项目(2018YFE0208400)。

摘　　要：为了提高光伏电站光伏发电功率预测精度,解决极限梯度提升模型、长短期记忆模型2种传统单一模型及传统组合模型极限梯度提升-长短期记忆模型的光伏发电功率预测结果滞后、预测效果易突变、预测误差较大、线性拟合性较差等不足,基于极限梯度提升算法、长短期记忆算法和线性自适应权重,提出一种考虑误差修正的非线性自适应权重极限梯度提升-长短期记忆模型进行光伏发电功率预测;分别使用极限梯度提升算法和长短期记忆算法训练得到2种单一模型,将2种单一模型的初步预测值和真实值组成新的训练数据集,利用神经网络算法训练所提出的模型,对2种单一模型的初步预测值分配自适应权重系数,并根据训练时所提出模型的预测值大小分段统计预测误差的分布,预测时根据所提出模型的预测值在预测结果的基础上累加误差均值从而进行误差修正,进一步提高所提出模型的预测精度;利用Python语言分别对所提出的模型、传统组合模型和2种传统单一模型在晴天、阴天和雨天的光伏发电功率预测性能进行仿真。结果表明:与极限梯度提升-长短期记忆模型、极限梯度提升模型、长短期记忆模型相比,所提出模型的均方根误差分别减小28.57%、 39.39%、 49.79%,平均绝对误差分别减小44.25%、 53.33%、 64.8%,决定系数分别增大1.43%、 2.38%、 3.34%,所提出的模型更有效地减小了传统单一模型的光伏发电功率预测误差,优化了传统组合模型的权重系数;3种天气条件下所提出模型的光伏发电功率预测误差相对最小且稳健性最强,验证了所提出模型的有效性。To improve accuracy of photovoltaic power prediction in photovoltaic power plants and solve shortcomings of two traditional single models of extreme gradient boosting model and long short term memory model as well as traditional combined model of extreme gradient boosting-long short term memory model,such as delayed results of photovoltaic power prediction,easy to change the prediction effect,large prediction error,and poor linear fitting,on the basis of extreme gradient boosting algorithm,long short term memory algorithm,and linear adaptive weight,a nonlinear adaptive weight extreme gradient boosting-long short term memory model considering error correction was proposed for photovoltaic power prediction.Preliminary predicted values and real values of the two single models which were trained by using extreme gradient boosting algorithm and long short term memory algorithm respectively were composed into a new training data set.Initial predicted values of the two single models were assigned adaptive weight coefficients through the proposed model trained by using neural network algorithm.Moreover,the distribution of prediction errors was calculated piecewise according to the predicted value of the proposed model in the training,and the mean of errors was accumulated according to the predicted value of the proposed model on the basis of the predicted result for error correction,so as to further improve prediction accuracy of the proposed model.Photovoltaic power prediction performances of the proposed model,traditional combined model and two traditional single models on sunny,cloudy,and rainy days were simulated by using Python language.The results show that compared with extreme gradient boosting-long short term memory model,extreme gradient boosting model,and long short term memory model,the square root mean error of the proposed model is reduced by 28.57%,39.39%and 49.79%,the mean absolute error is reduced by 44.25%,53.33%and 64.8%,and the coefficient of determination is increased by 1.43%,2.38%and 3.34%,respecti

关 键 词：光伏发电 功率预测 自适应权重 误差修正 极限梯度提升算法 长短期记忆算法 

分 类 号：TM711[电气工程—电力系统及自动化]