{"title":"A tiered NARX model for forecasting day-ahead energy production in distributed solar PV systems","authors":"","doi":"10.1016/j.clet.2024.100831","DOIUrl":null,"url":null,"abstract":"<div><div>This study presents a hierarchical forecasting approach for day-ahead energy production in distributed solar Photovoltaic (PV) systems using a tiered Nonlinear Autoregressive Exogenous (NARX) model. The methodology was applied to 52 PV systems installed at The University of Jordan, covering three prediction scales: fleet-wide, zone-specific, and site-specific. The model incorporated weather data, including solar irradiation, temperature, and humidity, to forecast the next day's energy production. Based on a new metric called the <span><math><mrow><mi>O</mi><mi>v</mi><mi>e</mi><mi>r</mi><mi>a</mi><mi>l</mi><mi>l</mi><mspace></mspace><mi>M</mi><mi>e</mi><mi>t</mi><mi>r</mi><mi>i</mi><mi>c</mi></mrow></math></span>, fleet-wide predictions outperform the zone-specific and site-specific averages by 3.21% and 5.35%, respectively. Normalized Root Mean Square Errors (<span><math><mrow><mi>n</mi><mi>R</mi><mi>M</mi><mi>S</mi><mi>E</mi></mrow></math></span>) for fleet-wide, zone-specific, and site-specific predictions are 0.148, 0.141, and 0.137, respectively. The Correlation Coefficient (<span><math><mrow><mi>R</mi></mrow></math></span>) is above 80% for all prediction scales, with the accuracy constrained by the model's difficulty in adapting to abrupt weather changes, leading to overestimation. The model performs best when weather patterns and PV generation are consistent with previous days. This demonstrates that adapting models to the characteristics of each scale significantly improves forecast accuracy, enabling more effective macro-level planning and micro-level operational decisions.</div></div>","PeriodicalId":34618,"journal":{"name":"Cleaner Engineering and Technology","volume":null,"pages":null},"PeriodicalIF":5.3000,"publicationDate":"2024-10-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Cleaner Engineering and Technology","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2666790824001113","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ENGINEERING, ENVIRONMENTAL","Score":null,"Total":0}
引用次数: 0
Abstract
This study presents a hierarchical forecasting approach for day-ahead energy production in distributed solar Photovoltaic (PV) systems using a tiered Nonlinear Autoregressive Exogenous (NARX) model. The methodology was applied to 52 PV systems installed at The University of Jordan, covering three prediction scales: fleet-wide, zone-specific, and site-specific. The model incorporated weather data, including solar irradiation, temperature, and humidity, to forecast the next day's energy production. Based on a new metric called the , fleet-wide predictions outperform the zone-specific and site-specific averages by 3.21% and 5.35%, respectively. Normalized Root Mean Square Errors () for fleet-wide, zone-specific, and site-specific predictions are 0.148, 0.141, and 0.137, respectively. The Correlation Coefficient () is above 80% for all prediction scales, with the accuracy constrained by the model's difficulty in adapting to abrupt weather changes, leading to overestimation. The model performs best when weather patterns and PV generation are consistent with previous days. This demonstrates that adapting models to the characteristics of each scale significantly improves forecast accuracy, enabling more effective macro-level planning and micro-level operational decisions.