基于DBN耦合模型的太阳辐照度模拟

Solar irradiance simulation based on DBN coupling model

在低碳经济的发展趋势下,太阳辐射的精确模拟对于光伏发电行业至关重要。影响太阳辐射的因素十分复杂,其与许多要素之间都存在复杂的非线性关系,这使得太阳辐射的模拟变得相对复杂。目前已有研究表明,机器学习模型能够很好地模拟太阳辐射,可挖掘出太阳辐射和各种影响因素之间的数学关系。文中基于支持向量回归(SVR)、随机森林(RF)、贝叶斯岭回归(BR)、梯度提升树(GBDT)、BP神经网络(BP)等五种机器学习模型,采用2020年1月—2021年4月位于河南的4个光伏发电站的真实卫星数据及地面观测数据进行机器学习建模以及地面太阳辐射模拟。由于传统单一模型的模拟精度较低,为提升模拟精度,通过引入DBN深度信念网络对五种模型模拟的结果进行二次建模,最终得到4个站点的高精度太阳辐射模拟值。研究结果表明,DBN耦合模型能够有效提取不同机器学习模型模拟结果的特征,其模拟效果和鲁棒性显著优于单一机器学习模型。

Under the development trend of low-carbon economy,the accurate simulation of solar radiation is important for the photovoltaic power generation industry.There are many complex factors that influence solar radiation.The nonlinear relationships between solar radiation and these factors make the simulation of solar radiation relatively complex.Many researches have shown that machine learning method can well simulate the solar radiation,in which the mathematic relation between solar radiation and various influence factors has been dug out.Based on five machine learning models including support vector regression(SVR),random forest(RF),Bayesian ridge regression(BR),gradient boosting decision tree(GBDT)and BP neural network,in this research,the real satellite data and ground observation data of four photovoltaic power plants located in Henan from January 2020 to April 2021 are adopted to carry out machine learning modeling and ground solar irradiance simulation.The deep belief network(DBN)is introduced to carried out secondary modeling of the simulation results of the five models to improve simulation accuracy because the traditional single model has low simulation accuracy.Finally,the high-precision solar radiation simulation values of the four plants are obtained.The simulation results show that the DBN coupling model can effectively extract the characteristics of the simulation results of different machine learning models,and its simulation effect and robustness are significantly better than of the single machine learning model.