Automatic hourly solar forecasting using machine learning models. (May 2019)

• Record Type: Journal Article
• Authors: Yagli, Gokhan Mert; Yang, Dazhi; Srinivasan, Dipti
• Is Part Of: Renewable & sustainable energy reviews. Volume 105(2019)
• Page Start: 487
• Page End: 498
• Publication Date: 2019-05
• Subjects: ANN Artificial Neural Network -- ANNavg Averaged ANN -- ANNbr Bayesian regularized ANN -- ANNpca ANN with feature extraction -- ANNqr QR using ANN -- BON Bondville, Illinois -- caret Classification and Regression Training -- CIT Conditional Inference Trees -- CSI Clear-sky Index -- CSpers Clear-sky Adjusted Persistence (Smart Persistence) -- CUB Cubist -- CV Cross-validation -- DRA Desert Rock, Nevada -- ENET Elastic Net -- ERT Extremely Randomized Trees -- EVTREE Tree Models using Genetic Algorithms -- FPK Fort Peck, Montana -- FS Forecast Skill -- GB Gradient Boosting -- GBlin Boosted Linear Model -- GBsm Boosted Smoothing Spline -- GBst Stochastic GB -- GCM Goodwin Creek, Mississippi -- GCV Generalized Cross Validation -- GHI Global Horizontal Irradiance -- GLM Generalized Linear Model -- GLMbayes Bayesian GLM -- GLMboos Boosted GLM -- GLMnb Negative Binomial GLM -- GLMNET Lasso and Elastic-Net Regularized GLM -- GLMstepAIC GLM using AIC-minimized feature selection -- GP Gaussian Process -- GPlin GP with Linear Kernel -- GPpoly GP with Polynomial Kernel -- GPrad GP with Radial Basis Function Kernel -- ICR Independent Component Regression -- kNN k-Nearest Neighbors -- LARS Least Angle Regression -- LASSO Least Absolute Shrinkage and Selection Operator -- LASSObayes Bayesian LASSO -- LASSOqr QR with LASSO -- LASSOrel Relaxed LASSO -- LR Linear Regression -- MARS Multivariate Adaptive Regression Splines -- MARSbag Bagged MARS -- MARSgcv MARS using GCV -- MARSgcvBag Bagged MARS using GCV -- ML Machine Learning -- MLP Multilayer Perceptron -- MLPdecay MLP using Decay -- MLPdecayml Multilayer MLP using Decay -- MLPml Multilayer MLP -- MLPmon Monotonic MLP -- nMBE normalized Mean Bias Error -- nRMSE normalized Root-Mean-Square Error -- NH Node Harvest -- NNLS Non-Negative Least Squares -- NREL National Renewable Energy Laboratory -- NSRDB The National Solar Radiation Database -- NWP Numerical Weather Prediction -- NWS National Weather Service -- partDSA Partitioning Using Deletion, Substitution, and Addition Moves -- PCA Principal Component Analysis -- PCR Principal Component Regression -- Pers Persistence Model -- PLR Penalized Linear Regression -- PLS Partial Least Squares -- PPR Projection Pursuit Regression -- PSF Pattern Sequence-based Forecasting -- PSM Physical Solar Model -- PSU Penn. State Univ., Pennsylvania -- PV Photovoltaic -- QR Quantile Regression -- QRpen Penalized QR -- RF Random Forest -- RFqr QR with RF -- RR Ridge Regression -- RRbayes Bayesian Ridge Regression -- SOM Self-organizing Map -- SPLS Sparse Partial Least Squares -- SURFRAD Surface Radiation Budget -- SVRL2 L2 Regularized SVR -- SVR Support Vector Regression -- SVRlin SVR with linear kernel -- SVRpoly SVR with polynomial kernel -- SVRrad SVR with radial basis function kernel -- SVRradc SVRrad with parameter C tuning -- SVRradsig SVRrad with parameter sig tuning -- SXF Sioux Falls, South Dakota -- TBL Table Mountain, Boulder, Colorado -- xGB Extreme Gradient Boosting -- xGBlin xGB with Linear Model -- xGBtree xGB with Tree Model; Automatic machine learning, Solar forecasting, R caret package; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.rser.2019.02.006
• Languages: English
• ISSNs: 1364-0321

Abstract

Abstract: Owing to its recent advance, machine learning has spawned a large collection of solar forecasting works. In particular, machine learning is currently one of the most popular approaches for hourly solar forecasting. Nevertheless, there is evidently a myth on forecast accuracy—virtually all research papers claim superiority over others. Apparently, the "best" model can only be selected with hindsight, i.e., after empirical evaluation. For any new forecasting project, it is irrational for solar forecasters to bet on a single model from the start. In this article, the hourly forecasting performance of 68 machine learning algorithms is evaluated for 3 sky conditions, 7 locations, and 5 climate zones in the continental United States. To ensure a fair comparison, no hybrid model is considered, and only off-the-shelf implementations of these algorithms are used. Moreover, all models are trained using the automatic tuning algorithm available in the Rcaret package. It is found that tree-based methods consistently perform well in terms of two-year overall results, however, they rarely stand out during daily evaluation. Although no universal model can be found, some preferred ones for each sky and climate condition are advised. Abstract : Highlights: Hourly solar forecasting is performed using 68 machine learning models. Models are evaluated at 7 locations in 5 climate zones for 2 years. Tree-based methods consistently perform well in terms of 2-year average metrics. Daily best model cannot be identified, regime-switching approach is advised.