Wind energy potential estimation using neural network and SVR approaches

IF 0.7 Q3 ENGINEERING, MULTIDISCIPLINARY

Engineering Review Pub Date : 2022-01-01 DOI:10.30765/er.1632

A. A. Salami, Pierre Akuété Agbessi, A. Ajavon, Seibou Boureima

引用次数: 0

Abstract

The distribution of wind speed and the optimal assessment of wind energy potential are very important factors when selecting a suitable site for a wind power plant. In wind farm design projects for the supply of electrical energy, designers use the Weibull distribution law to analyse the characteristics and variations of wind speed in order to evaluate the wind potential. In our study we used two approaches, namely, the Multilayer Perceptron (MLP) approach and the Support Vector Machine (SVR) approach to determine a distribution law of wind speeds and to optimally evaluate the wind potential. These two approaches were compared to two well-known numerical methods which are the Justus Empirical Method (EMJ) and the Maximum Likelihood Method (MLM). The results show that the neural network approach produces a better fit of the distribution curve with an Root Mean Square Error (RMSE) of 0.00005016 at Lomé, 0.000040289 at Cotonou site and a more interesting estimate of the wind potential. After that SVR show a better result too with an RMSE of 0.0095618 at the Lomé site and 0.0053549 at the Cotonou site.

查看原文本刊更多论文

基于神经网络和SVR方法的风能潜力估计

风速分布和风能潜力的优化评价是风电场选址的重要因素。在用于供电的风电场设计项目中，设计者利用威布尔分布规律分析风速的特征和变化，以评估风势。在我们的研究中，我们使用了两种方法，即多层感知器(MLP)方法和支持向量机(SVR)方法来确定风速的分布规律，并对风势进行最优评估。将这两种方法与两种著名的数值方法Justus Empirical Method (EMJ)和Maximum Likelihood Method (MLM)进行比较。结果表明，神经网络方法能较好地拟合分布曲线，lomoise站点的均方根误差(RMSE)为0.00005016,Cotonou站点的均方根误差(RMSE)为0.000040289，对风势的估计更有趣。在此之后，SVR也显示出更好的结果，在lomoise站点的RMSE为0.0095618，在Cotonou站点的RMSE为0.0053549。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Engineering Review ENGINEERING, MULTIDISCIPLINARY-

CiteScore

1.00

自引率

0.00%

发文量

期刊介绍： Engineering Review is an international journal designed to foster the exchange of ideas and transfer of knowledge between scientists and engineers involved in various engineering sciences that deal with investigations related to design, materials, technology, maintenance and manufacturing processes. It is not limited to the specific details of science and engineering but is instead devoted to a very wide range of subfields in the engineering sciences. It provides an appropriate resort for publishing the papers covering prior applications – based on the research topics comprising the entire engineering spectrum. Topics of particular interest thus include: mechanical engineering, naval architecture and marine engineering, fundamental engineering sciences, electrical engineering, computer sciences and civil engineering. Manuscripts addressing other issues may also be considered if they relate to engineering oriented subjects. The contributions, which may be analytical, numerical or experimental, should be of significance to the progress of mentioned topics. Papers that are merely illustrations of established principles or procedures generally will not be accepted. Occasionally, the magazine is ready to publish high-quality-selected papers from the conference after being renovated, expanded and written in accordance with the rules of the magazine. The high standard of excellence for any of published papers will be ensured by peer-review procedure. The journal takes into consideration only original scientific papers.