Comparative analysis of Artificial Neural Networks with conventional methods for extrapolation of wind speed at an elevated height

5th International Symposium on Hydrogen Energy, Renewable Energy and Materials

能源学;材料科学

Bhavsar, Dhrumin^1 ; Bhatt, Mohil^1 ; Choksi, Mehul^1 ; Nagababu, Garlapati^1

Department of Mechanical Engineering, PDPU, Gandhinagar, Gujarat

382007, India^1

关键词: Atmospheric factors;    Comparative analysis;    Conventional methods;    Levenberg Marquardt backpropagation;    Number of datum;    Root mean square errors;    Root-means-square errors;    Wind directions;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/605/1/012011/pdf
DOI  :  10.1088/1757-899X/605/1/012011

学科分类：材料科学（综合）

来源: IOP

【 摘 要 】

Due to rapidly increasing pollution, it becomes necessary to substitute fossil fuels, and as wind energy is available quite easily and in abundance, researches are carried out in this area. These facts make it imperative to know about the variables and the problems involved behind it. The wind speed is a random variable, and it depends on atmospheric factors like pressure, relative humidity, wind dispersion & wind direction. This paper introduces the method to effectively predict wind speed by making use of the Levenberg-Marquardt backpropagation algorithm in artificial neural network (ANN) and by conventional means like power law & log law in MATLAB. Data from the Gulf of Khambhat, Gujarat provided by LIDAR for a period of 8 months, was used to prepare valid data set to train the neural network and to build a model to predict wind speed. After obtaining a histogram of the predicted values by log law, power law and ANN, it was seen that wind speed values obtained by ANN were quite close to actual values than the values obtained through the other methods. A comparison in terms of root means square error and percentage of the number of data indicates that developed neural network gives less root mean square error and a higher percentage of data whose absolute error lie between-0.2 and 0.2.