Solar Irradiance Prediction Method Based On Ensemble Learning And Convolutional Neural Network

The current energy shortage and environmental pollution are becoming increasingly serious;the utilization of solar irradiance has drawn the attention of researchers in various fields.Solar energy is widely used in the photothermic field and optoelectronics,and which is considered as the best alternative energy.Due to seasonal factors such as seasons,climate,cloud density and other climatic factors,the stability of solar irradiation and its application are restricted.Therefore,short-term predictions of solar irradiance with high precision are extremely critical.In this paper,ensemble learning and convolutional neural network have been applied to the study of solar irradiance prediction method,American Meteorological Society 2013-2014 Solar Energy Prediction Contest dataset has been introduced to establish experiments.Ensemble learning based on the Bayesian model combination is proposed to improve prediction reliability of solar irradiance.Firstly,K-means clustering and K-fold cross validation are introduced to generate multiple training subsets so as to increase the diversity of base learners and ensure uniform sampling.Secondly,the random forests are defined as base learners to establish an ensemble learning prediction model,each training subset is used to train the corresponding individual base learner.Then,to make up for the defects of base learners,the Bayesian model combination algorithm is introduced to formulate the combination strategy according to the prediction performance of the individual base learner on the verification set.The accuracy and reliability of the proposed ensemble learning method in the solar irradiance prediction are verified by contrast experiments,which can accurately and reliably predict solar irradiance in different meteorological conditions.To analyze the correlation of meteorological data at different sampling times and the correlation of meteorological data at different stations,as well as their relationships with solar irradiance.In this paper,according to the framework of the standard convolutional neural network,a novel convolutional neural network structure is established and utilized for solar irradiance prediction.To alleviate the imperfect prediction performance of the novel convolutional neural network caused by improper selection of hyper parameters,the chaos hybrid algorithm is applied to optimize the structural parameters and learning parameters of the novel convolutional neural network.To verify the performance of the proposed method,firstly,the simulation test of benchmark function has verified its ability to fit nonlinear functions,and it also shows that the hybrid optimization is conducive to further improve the accuracy.Then,the novel network analyzes the correlation of meteorological data and predicts solar irradiance.Four other machine learning methods are introduced for comparative analysis,and the prediction error of the winner in this solar prediction competition is defined as a benchmark.The advantages,accuracy and reliability of the proposed solar energy prediction method are fully demonstrated.