Toward better integration of solar energy in transportation systems: machine learning benchmarks for day-ahead photovoltaic power forecasting

Zhaoyao Bao, Zhe Sun, Yishuo Jiang, Chi Xie, Lijun Sun and H. Oliver Gao

Transportation Research Part A: Policy and Practice, 2026, vol. 210, issue C

Abstract: The integration of solar photovoltaic (PV) electricity into transportation systems presents both opportunities and challenges. This study first provides a concise review of the literature on PV power forecasting and the integration of solar energy into transportation. Despite rapid PV deployment and growing interest in electric fleets, many transportation studies still rely on realized PV data or crude capacity-factor proxies and seldom incorporate day-ahead PV power forecasting. At the same time, much of the PV forecasting literature focuses on introducing a single, increasingly complex neural network model and comparing it primarily against other neural networks, offering limited guidance on method selection for practical applications. This paper benchmarks machine learning models for day-ahead PV power forecasting to better support daily electricity demand management in both the operation and planning of transportation systems. Using a unified feature set and controlled ablation and sensitivity experiments, we evaluate eight widely adopted forecasting methods, including linear regression, tree-based ensembles, and deep neural networks, across 100 PV plants in New York State and further validate the main benchmarking conclusions on 9 additional PV plants from 9 representative U.S. states. Based on over 30,000 evaluations, we systematically characterize forecasting performance across models, input sets, and configurations in terms of both prediction error and computing time, providing practice guidance on model and configuration selection under constraints on input availability and computational resources in transportation applications. We find that: (1) LightGBM achieves the best Root Mean Square Error (RMSE, 6.07 ± 1.11), while the linear regression baseline is competitive (7.27 ± 1.67) with the lowest computing time; (2) Input composition, particularly how realized PV, historical weather, and numerical weather prediction information are combined strongly affects forecasting performance; and (3) The impacts of model complexity, lookback window length, dataset shuffling and extending, training dataset scale, weather feature selection, and seasonal and hourly variation are systematically analyzed and exhibit distinct patterns across methods. A decision flowchart on model selection and configuration for practitioners is provided. All datasets and code are openly released to foster transparent comparison and accelerate the practical adoption of solar PV power forecasting in daily energy management and transportation planning.

Keywords: Transportation; Solar power; Forecasting; Photovoltaic; Machine learning; Benchmark (search for similar items in EconPapers)
Date: 2026
References: Add references at CitEc
Citations:

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0965856426001813
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:transa:v:210:y:2026:i:c:s0965856426001813

Ordering information: This journal article can be ordered from
http://www.elsevier.com/wps/find/supportfaq.cws_home/regional
https://shop.elsevie ... _01_ooc_1&version=01

DOI: 10.1016/j.tra.2026.105040

Access Statistics for this article

Transportation Research Part A: Policy and Practice is currently edited by John (J.M.) Rose

More articles in Transportation Research Part A: Policy and Practice from Elsevier
Bibliographic data for series maintained by Catherine Liu ().