Location-aware green energy availability forecasting for multiple time frames in smart buildings The case of Estonia

2025-05-02 0 0 1.98MB 33 页 10玖币

侵权投诉

Location-aware green energy availability forecasting for multiple

time frames in smart buildings: The case of Estonia

Mehdi Hatamiana,1, Bivas Panigrahib, Chinmaya Kumar Dehuryc,∗

aIndependent Researcher, Tehran, Iran

bDepartment of Refrigeration, Air Conditioning & Energy Engineering, National Chin-Yi University of

Technology, Taichung 41170, Taiwan

cMobile & cloud Lab, Institute of Computer Science, University of Tartu, Tartu 50090, Estonia

Abstract

Renewable Energies (RE) have gained more attention in recent years since they oﬀer

clean and sustainable energy. One of the major sustainable development goals (SDG-7) set

by the United Nations (UN) is to achieve aﬀordable and clean energy for everyone. Among

the world’s all renewable resources, solar energy is considered as the most abundant and can

certainly fulﬁll the target of SDGs. Solar energy is converted into electrical energy through

Photovoltaic (PV) panels with no greenhouse gas emissions. However, power generated

by PV panels is highly dependent on solar radiation received at a particular location over

a given time period. Therefore, it is challenging to forecast the amount of PV output

power. Predicting the output power of PV systems is essential since several public/private

institutes generate such green energy, and need to maintain the balance between demand

and supply. This research aims to forecast PV system output power based on weather and

derived features using diﬀerent machine learning models. The objective is to obtain the best-

ﬁtting model to precisely predict output power by inspecting the data. Moreover, diﬀerent

performance metrics are used to compare and evaluate the accuracy under diﬀerent machine

learning models such as random forest, XGBoost, KNN, etc.

Keywords: Solar Panel, Smart Building, Green Energy Prediction, Machine Learning, PV

Output Power Prediction

Preprint submitted to Solar Energy July 2022

arXiv:2210.01619v1 [cs.LG] 4 Oct 2022

1. Introduction

To achieve sustainable development and growth, United Nations (UN) has set the blueprint

for 17 sustainability development goals (SDG-17). One of the major goals (SDG-7) is to

provide clean and aﬀordable energy to the population [1]. Hence, the generation of Re-

newable Energy (RE) is strongly encouraged and supported by technological advancements

and government policies for viable energy management in the future [2, 3]. Therefore a

sustainable alternative and energy management strategy would be to maximize the usage

of energy produced by the Photovoltaic (PV) system [4]. Renewable energies have recently

gained more attention since they oﬀer clean and sustainable energy. Among the world’s

renewable resources, solar energy is the most abundant one meaning the energy is from an

unlimited source that is not depleted by usage. Solar energy is converted into electrical

energy through PV panels with no greenhouse gas emissions. Predicting the energy produc-

tion by the PV system is essential since many companies generate energy, and they need

to maintain electricity production and demand in balance. Moreover, an eﬃcient way to

convince investors to invest in solar energy is to provide them a time frame for a proﬁt

from their investment. However, predicting the output power generated by PV systems is a

cumbersome task since they are highly dependent on how much solar radiation they receive,

the condition of weather, the position of the PV panel, and the amount of time PV panels

are exposed to sunlight, and many more [5]. Solar radiation is crucial for PV systems, and

the output power of the PV system is determined by total solar irradiance on a particular

day. A unit of irradiance is deﬁned by the total output of the solar source falling on a unit

area. However, solar irradiance is aﬀected by various factors, including weather, location,

time, etc.

Therefore, a direct relationship between energy produced by a PV system and local

weather conditions exists that varies during the day as the amount of solar irradiance changes

∗Corresponding author

Email addresses: mehdi.hatamian86@gmail.com (Mehdi Hatamian ), bivas@ncut.edu.tw (Bivas

Panigrahi), chinmaya.dehury@ut.ee (Chinmaya Kumar Dehury)

1Mr. Mehdi was with the Institute of Computer Science, University of Tartu as a Master student.

Currently, he is an independent researcher.

[6]. Furthermore, predicting the amount of electricity generated by a PV system is crucial

to calculating the size of the system, system load measurements, and return on investment

(ROI) [7, 8]. Such varied parameters play a major role in making the energy production

prediction more complex [9].

Diﬀerent methods have been employed in the literature to predict the output power

of PV systems. Data-driven and model-based are commonly used methods for predicting

green energy generation [10]. While model-based methods rely on analytical equations by

leveraging meteorological weather data [11], the data-driven models utilize machine learning

techniques to predict the output power of the PV system. However, to meet the demand for

modern PV systems and a sustainable energy management strategy, the existing prediction

approaches are not suﬃcient enough.

This paper aims to predict the output power of PV systems using state-of-art machine

learning models, including Extreme Gradient Boosting (XGboost), Random Forest (RF), K-

Nearest Neighbors (KNN), Support Vector Regression (SVR), and Multi-Layer Perceptron

(MLP). We have investigated the eﬀect of meteorological data on predicted output power to

ﬁnd an optimal set of input features. Moreover, the overall impact of three derived features,

including “Hours,” “Month,” and “Prior Output Power,” is analyzed. The list of acronyms

is presented in Table 1.

1.1. Motivations and Goals

High penetration of PV systems is oﬀered as an alternative to energy production methods

due to its economic beneﬁts and sustainable clean energy. However, the stability of the PV

systems might be threatening without an accurate prediction of the PV energy production.

The energy production of the PV systems is dependent on meteorological data. Therefore to

maintain the stability of the PV system, the uncertainty of output power predictions must be

addressed by accurate forecasting or prediction tools. The state-of-the-art prediction model

relies mainly on the historical performance of the PV panels without taking advantage of

the cloud coverage and other weather-related information. To meet the rising demand for

a futuristic green energy eco-system, a prediction model needs to consider the location and

weather. It is observed from the state-of-the-art literature survey that a prediction model

may not work in all geographical regions. Further, a single model may not give the desired

accuracy throughout the year for all the sessions, which is one of the major motivational

situations for this work.

The selection of ML models is another motivational scenario behind this proposal. In-

stead of relying on a single ML model, it is necessary to investigate and compare the perfor-

mance and accuracy of diﬀerent ML models, such as SVR, RF, and XGBoost, in diﬀerent

situations. The performance needs to be monitored based on various metrics, such as Mean

Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), Root Mean Square Error

(RMSE), and R-squared. These aforementioned research challenges motivate us to present a

location-based green energy availability in a smart building located in Tartu city of Estonia.

1.2. Contributions

The main contributions of this research study to the ﬁeld of predicting PV output power

can be summarised as follows:

•The research challenge of predicting the PV energy output is investigated with the

real datasets and recent research results.

•A real historical dataset of 1 yr duration from the solar panels installed on the roof-

top of the university building and nearby weather station in Estonia are collected and

pre-processed.

•Transformation is introduced as an eﬃcient way to normalize the dependent variables

to alter the skewness of the data and remove or lessen the impact of seasonality and

trend in our data.

•Z-score, Pearson correlation, and permutation-based feature importance are proposed

to be applied to the input features as a feature scaling method.

•Five popular ML models: KNN, XGBoost, MLP, SVR, and RF, are implemented, and

the performance results are compared based on MAPE, MAE, R2, and RMSE.

•The importance of the previous output power on the prediction model is analyzed.

The rest of this research paper is structured as follows. In Section 2, an overview of

general research methods is presented, followed by the methodology described in Section

3. Section 4 contains data processing, including target transformation, outlier handling,

and feature selection. In Section 5, the details about the implementation of each model are

given. The importance of prior output power is given in Section 6. Moreover, in Section 7,

results are compared and discussed in detail for each ML model, followed by the concluding

remarks and future works in Section 8.

Table 1. List of Acronyms

Acronyms Description

ANN Artiﬁcial Neural Network

AR Adaptive Recursive Linear

CV Cross Validation

FFBP Feed-forward Back Propagation

GBRT Gradient Boosted Regression Trees

GHI Global Horizontal Irradiance

GRNN General Regression Neural Network

IQR Interquartile Range

KNN K-Nearest Neighbour

MAE Mean Absolute Error

MAPE Mean Absolute Percentage Error

ML Machine Learning

MLP Multilayer Perceptron

NWP Numerical Weather Prediction

PV Photovoltaic

RBF Radial Basis Function

RE Renewable Energy

RF Random Forest

RMSE Root Mean Square Error

ROI Return on Investment

RT Regression Trees

SGD Stochastic Gradient Descent

STD Standard Deviation

SVR Support Vector Regression

XGBoost Extreme Gradient Boosting

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

Location-awaregreenenergyavailabilityforecastingformultipletimeframesinsmartbuildings:ThecaseofEstoniaMehdiHatamiana,1,BivasPanigrahib,ChinmayaKumarDehuryc,aIndependentResearcher,Tehran,IranbDepartmentofRefrigeration,AirConditioning&EnergyEngineering,NationalChin-YiUniversityofTechnology,Taichung41...

展开>> 收起<<

Location-aware green energy availability forecasting for multiple time frames in smart buildings The case of Estonia.pdf

共33页,预览5页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Location-aware green energy availability forecasting for multiple time frames in smart buildings The case of Estonia

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: