An artificial neural network to assess the impact of neighbouring photovoltaic systems in power forecasting in Utrecht, the Netherlands

A. G R Vaz; B. Elsinga; W. G J H M van Sark; M. C. Brito

doi:10.1016/j.renene.2015.06.061

An artificial neural network to assess the impact of neighbouring photovoltaic systems in power forecasting in Utrecht, the Netherlands

A. G R Vaz, B. Elsinga, W. G J H M van Sark, M. C. Brito^*

^*Corresponding author for this work

Research output: Contribution to journal › Article › Academic › peer-review

Abstract

In order to perform predictions of a photovoltaic (PV) system power production, a neural network architecture system using the Nonlinear Autoregressive with eXogenous inputs (NARX) model is implemented using not only local meteorological data but also measurements of neighbouring PV systems as inputs. Input configurations are compared to assess the effects of the different inputs. The added value of the information of the neighbouring PV systems has demonstrated to further improve the accuracy of predictions for both winter and summer seasons. Additionally, forecasts up to 1 month are tested and compared with a persistence model. Normalized root mean square errors (nRMSE) ranged between 9% and 25%, with the NARX model clearly outperforming the persistence model for forecast horizons greater than 15min.

Original language	English
Pages (from-to)	631-641
Number of pages	11
Journal	Renewable Energy
Volume	85
DOIs	https://doi.org/10.1016/j.renene.2015.06.061
Publication status	Published - 1 Jan 2016

Keywords

Artificial neural network
Forecasting
NARX model
Photovoltaics
Time series
valorisation

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1016/j.renene.2015.06.061

1-s2.0-S0960148115300847-mainFinal published version, 2.59 MB

Cite this

@article{529a9d004d754f21b9945abe67f0bb7f,

title = "An artificial neural network to assess the impact of neighbouring photovoltaic systems in power forecasting in Utrecht, the Netherlands",

abstract = "In order to perform predictions of a photovoltaic (PV) system power production, a neural network architecture system using the Nonlinear Autoregressive with eXogenous inputs (NARX) model is implemented using not only local meteorological data but also measurements of neighbouring PV systems as inputs. Input configurations are compared to assess the effects of the different inputs. The added value of the information of the neighbouring PV systems has demonstrated to further improve the accuracy of predictions for both winter and summer seasons. Additionally, forecasts up to 1 month are tested and compared with a persistence model. Normalized root mean square errors (nRMSE) ranged between 9% and 25%, with the NARX model clearly outperforming the persistence model for forecast horizons greater than 15min.",

keywords = "Artificial neural network, Forecasting, NARX model, Photovoltaics, Time series, valorisation",

author = "Vaz, {A. G R} and B. Elsinga and {van Sark}, {W. G J H M} and Brito, {M. C.}",

year = "2016",

month = jan,

day = "1",

doi = "10.1016/j.renene.2015.06.061",

language = "English",

volume = "85",

pages = "631--641",

journal = "Renewable Energy",

issn = "0960-1481",

publisher = "Pergamon Press Ltd.",

}

TY - JOUR

T1 - An artificial neural network to assess the impact of neighbouring photovoltaic systems in power forecasting in Utrecht, the Netherlands

AU - Vaz, A. G R

AU - Elsinga, B.

AU - van Sark, W. G J H M

AU - Brito, M. C.

PY - 2016/1/1

Y1 - 2016/1/1

N2 - In order to perform predictions of a photovoltaic (PV) system power production, a neural network architecture system using the Nonlinear Autoregressive with eXogenous inputs (NARX) model is implemented using not only local meteorological data but also measurements of neighbouring PV systems as inputs. Input configurations are compared to assess the effects of the different inputs. The added value of the information of the neighbouring PV systems has demonstrated to further improve the accuracy of predictions for both winter and summer seasons. Additionally, forecasts up to 1 month are tested and compared with a persistence model. Normalized root mean square errors (nRMSE) ranged between 9% and 25%, with the NARX model clearly outperforming the persistence model for forecast horizons greater than 15min.

AB - In order to perform predictions of a photovoltaic (PV) system power production, a neural network architecture system using the Nonlinear Autoregressive with eXogenous inputs (NARX) model is implemented using not only local meteorological data but also measurements of neighbouring PV systems as inputs. Input configurations are compared to assess the effects of the different inputs. The added value of the information of the neighbouring PV systems has demonstrated to further improve the accuracy of predictions for both winter and summer seasons. Additionally, forecasts up to 1 month are tested and compared with a persistence model. Normalized root mean square errors (nRMSE) ranged between 9% and 25%, with the NARX model clearly outperforming the persistence model for forecast horizons greater than 15min.

KW - Artificial neural network

KW - Forecasting

KW - NARX model

KW - Photovoltaics

KW - Time series

KW - valorisation

UR - http://www.scopus.com/inward/record.url?scp=84937158999&partnerID=8YFLogxK

U2 - 10.1016/j.renene.2015.06.061

DO - 10.1016/j.renene.2015.06.061

M3 - Article

AN - SCOPUS:84937158999

SN - 0960-1481

VL - 85

SP - 631

EP - 641

JO - Renewable Energy

JF - Renewable Energy

ER -

An artificial neural network to assess the impact of neighbouring photovoltaic systems in power forecasting in Utrecht, the Netherlands

Abstract

Keywords

UN SDGs

Access to Document

Other files and links

Fingerprint

Cite this